xref: /titanic_44/usr/src/uts/sun4u/opl/io/drmach.c (revision 672986541be54a7a471bb088e60780c37e371d7e)
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 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 
29 #include <sys/debug.h>
30 #include <sys/types.h>
31 #include <sys/varargs.h>
32 #include <sys/errno.h>
33 #include <sys/cred.h>
34 #include <sys/dditypes.h>
35 #include <sys/devops.h>
36 #include <sys/modctl.h>
37 #include <sys/poll.h>
38 #include <sys/conf.h>
39 #include <sys/ddi.h>
40 #include <sys/sunddi.h>
41 #include <sys/sunndi.h>
42 #include <sys/ndi_impldefs.h>
43 #include <sys/stat.h>
44 #include <sys/kmem.h>
45 #include <sys/vmem.h>
46 #include <sys/opl_olympus_regs.h>
47 #include <sys/cpuvar.h>
48 #include <sys/cpupart.h>
49 #include <sys/mem_config.h>
50 #include <sys/ddi_impldefs.h>
51 #include <sys/systm.h>
52 #include <sys/machsystm.h>
53 #include <sys/autoconf.h>
54 #include <sys/cmn_err.h>
55 #include <sys/sysmacros.h>
56 #include <sys/x_call.h>
57 #include <sys/promif.h>
58 #include <sys/prom_plat.h>
59 #include <sys/membar.h>
60 #include <vm/seg_kmem.h>
61 #include <sys/mem_cage.h>
62 #include <sys/stack.h>
63 #include <sys/archsystm.h>
64 #include <vm/hat_sfmmu.h>
65 #include <sys/pte.h>
66 #include <sys/mmu.h>
67 #include <sys/cpu_module.h>
68 #include <sys/obpdefs.h>
69 #include <sys/note.h>
70 #include <sys/ontrap.h>
71 #include <sys/cpu_sgnblk_defs.h>
72 #include <sys/opl.h>
73 
74 
75 #include <sys/promimpl.h>
76 #include <sys/prom_plat.h>
77 #include <sys/kobj.h>
78 
79 #include <sys/sysevent.h>
80 #include <sys/sysevent/dr.h>
81 #include <sys/sysevent/eventdefs.h>
82 
83 #include <sys/drmach.h>
84 #include <sys/dr_util.h>
85 
86 #include <sys/fcode.h>
87 #include <sys/opl_cfg.h>
88 
89 extern void		bcopy32_il(uint64_t, uint64_t);
90 extern void		flush_cache_il(void);
91 extern void		drmach_sleep_il(void);
92 
93 typedef struct {
94 	struct drmach_node	*node;
95 	void			*data;
96 } drmach_node_walk_args_t;
97 
98 typedef struct drmach_node {
99 	void		*here;
100 
101 	pnode_t		(*get_dnode)(struct drmach_node *node);
102 	int		(*walk)(struct drmach_node *node, void *data,
103 				int (*cb)(drmach_node_walk_args_t *args));
104 	dev_info_t	*(*n_getdip)(struct drmach_node *node);
105 	int		(*n_getproplen)(struct drmach_node *node, char *name,
106 				int *len);
107 	int		(*n_getprop)(struct drmach_node *node, char *name,
108 				void *buf, int len);
109 	int		(*get_parent)(struct drmach_node *node,
110 				struct drmach_node *pnode);
111 } drmach_node_t;
112 
113 typedef struct {
114 	int		 min_index;
115 	int		 max_index;
116 	int		 arr_sz;
117 	drmachid_t	*arr;
118 } drmach_array_t;
119 
120 typedef struct {
121 	void		*isa;
122 
123 	void		(*dispose)(drmachid_t);
124 	sbd_error_t	*(*release)(drmachid_t);
125 	sbd_error_t	*(*status)(drmachid_t, drmach_status_t *);
126 
127 	char		 name[MAXNAMELEN];
128 } drmach_common_t;
129 
130 typedef	struct {
131 	uint32_t	core_present;
132 	uint32_t	core_hotadded;
133 	uint32_t	core_started;
134 } drmach_cmp_t;
135 
136 typedef struct {
137 	drmach_common_t	 cm;
138 	int		 bnum;
139 	int		 assigned;
140 	int		 powered;
141 	int		 connected;
142 	int		 cond;
143 	drmach_node_t	*tree;
144 	drmach_array_t	*devices;
145 	int		boot_board;	/* if board exists on bootup */
146 	drmach_cmp_t	cores[OPL_MAX_COREID_PER_BOARD];
147 } drmach_board_t;
148 
149 typedef struct {
150 	drmach_common_t	 cm;
151 	drmach_board_t	*bp;
152 	int		 unum;
153 	int		portid;
154 	int		 busy;
155 	int		 powered;
156 	const char	*type;
157 	drmach_node_t	*node;
158 } drmach_device_t;
159 
160 typedef struct drmach_cpu {
161 	drmach_device_t  dev;
162 	processorid_t    cpuid;
163 	int		sb;
164 	int		chipid;
165 	int		coreid;
166 	int		strandid;
167 	int		status;
168 #define	OPL_CPU_HOTADDED	1
169 } drmach_cpu_t;
170 
171 typedef struct drmach_mem {
172 	drmach_device_t  dev;
173 	uint64_t	slice_base;
174 	uint64_t	slice_size;
175 	uint64_t	base_pa;	/* lowest installed memory base */
176 	uint64_t	nbytes;		/* size of installed memory */
177 	struct memlist *memlist;
178 } drmach_mem_t;
179 
180 typedef struct drmach_io {
181 	drmach_device_t  dev;
182 	int	channel;
183 	int	leaf;
184 } drmach_io_t;
185 
186 typedef struct drmach_domain_info {
187 	uint32_t	floating;
188 	int		allow_dr;
189 } drmach_domain_info_t;
190 
191 drmach_domain_info_t drmach_domain;
192 
193 typedef struct {
194 	int		 flags;
195 	drmach_device_t	*dp;
196 	sbd_error_t	*err;
197 	dev_info_t	*dip;
198 } drmach_config_args_t;
199 
200 typedef struct {
201 	drmach_board_t	*obj;
202 	int		 ndevs;
203 	void		*a;
204 	sbd_error_t	*(*found)(void *a, const char *, int, drmachid_t);
205 	sbd_error_t	*err;
206 } drmach_board_cb_data_t;
207 
208 static drmach_array_t	*drmach_boards;
209 
210 static sbd_error_t	*drmach_device_new(drmach_node_t *,
211 				drmach_board_t *, int, drmachid_t *);
212 static sbd_error_t	*drmach_cpu_new(drmach_device_t *, drmachid_t *);
213 static sbd_error_t	*drmach_mem_new(drmach_device_t *, drmachid_t *);
214 static sbd_error_t	*drmach_io_new(drmach_device_t *, drmachid_t *);
215 
216 static dev_info_t	*drmach_node_ddi_get_dip(drmach_node_t *np);
217 static int		 drmach_node_ddi_get_prop(drmach_node_t *np,
218 				char *name, void *buf, int len);
219 static int		 drmach_node_ddi_get_proplen(drmach_node_t *np,
220 				char *name, int *len);
221 
222 static int 		drmach_get_portid(drmach_node_t *);
223 static	sbd_error_t	*drmach_i_status(drmachid_t, drmach_status_t *);
224 static int		opl_check_dr_status();
225 static void		drmach_io_dispose(drmachid_t);
226 static sbd_error_t	*drmach_io_release(drmachid_t);
227 static sbd_error_t	*drmach_io_status(drmachid_t, drmach_status_t *);
228 static int 		drmach_init(void);
229 static void 		drmach_fini(void);
230 static void		drmach_swap_pa(drmach_mem_t *, drmach_mem_t *);
231 static drmach_board_t	*drmach_get_board_by_bnum(int);
232 
233 /* options for the second argument in drmach_add_remove_cpu() */
234 #define	HOTADD_CPU	1
235 #define	HOTREMOVE_CPU	2
236 
237 #define	ON_BOARD_CORE_NUM(x)	(((uint_t)(x) / OPL_MAX_STRANDID_PER_CORE) & \
238 	(OPL_MAX_COREID_PER_BOARD - 1))
239 
240 extern struct cpu	*SIGBCPU;
241 
242 static int		drmach_name2type_idx(char *);
243 static drmach_board_t	*drmach_board_new(int, int);
244 
245 #ifdef DEBUG
246 
247 #define	DRMACH_PR		if (drmach_debug) printf
248 int drmach_debug = 1;		 /* set to non-zero to enable debug messages */
249 #else
250 
251 #define	DRMACH_PR		_NOTE(CONSTANTCONDITION) if (0) printf
252 #endif /* DEBUG */
253 
254 
255 #define	DRMACH_OBJ(id)		((drmach_common_t *)id)
256 
257 #define	DRMACH_NULL_ID(id)	((id) == 0)
258 
259 #define	DRMACH_IS_BOARD_ID(id)	\
260 	((id != 0) &&		\
261 	(DRMACH_OBJ(id)->isa == (void *)drmach_board_new))
262 
263 #define	DRMACH_IS_CPU_ID(id)	\
264 	((id != 0) &&		\
265 	(DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new))
266 
267 #define	DRMACH_IS_MEM_ID(id)	\
268 	((id != 0) &&		\
269 	(DRMACH_OBJ(id)->isa == (void *)drmach_mem_new))
270 
271 #define	DRMACH_IS_IO_ID(id)	\
272 	((id != 0) &&		\
273 	(DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
274 
275 #define	DRMACH_IS_DEVICE_ID(id)					\
276 	((id != 0) &&						\
277 	(DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new ||	\
278 	    DRMACH_OBJ(id)->isa == (void *)drmach_mem_new ||	\
279 	    DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
280 
281 #define	DRMACH_IS_ID(id)					\
282 	((id != 0) &&						\
283 	(DRMACH_OBJ(id)->isa == (void *)drmach_board_new ||	\
284 	    DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new ||	\
285 	    DRMACH_OBJ(id)->isa == (void *)drmach_mem_new ||	\
286 	    DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
287 
288 #define	DRMACH_INTERNAL_ERROR() \
289 	drerr_new(1, EOPL_INTERNAL, drmach_ie_fmt, __LINE__)
290 
291 static char		*drmach_ie_fmt = "drmach.c %d";
292 
293 static struct {
294 	const char	*name;
295 	const char	*type;
296 	sbd_error_t	*(*new)(drmach_device_t *, drmachid_t *);
297 } drmach_name2type[] = {
298 	{ "cpu",	DRMACH_DEVTYPE_CPU,		drmach_cpu_new },
299 	{ "pseudo-mc",	DRMACH_DEVTYPE_MEM,		drmach_mem_new },
300 	{ "pci",	DRMACH_DEVTYPE_PCI,		drmach_io_new  },
301 };
302 
303 /* utility */
304 #define	MBYTE	(1048576ull)
305 
306 /*
307  * drmach autoconfiguration data structures and interfaces
308  */
309 
310 extern struct mod_ops mod_miscops;
311 
312 static struct modlmisc modlmisc = {
313 	&mod_miscops,
314 	"OPL DR 1.1"
315 };
316 
317 static struct modlinkage modlinkage = {
318 	MODREV_1,
319 	(void *)&modlmisc,
320 	NULL
321 };
322 
323 static krwlock_t drmach_boards_rwlock;
324 
325 typedef const char	*fn_t;
326 
327 int
328 _init(void)
329 {
330 	int err;
331 
332 	if ((err = drmach_init()) != 0) {
333 		return (err);
334 	}
335 
336 	if ((err = mod_install(&modlinkage)) != 0) {
337 		drmach_fini();
338 	}
339 
340 	return (err);
341 }
342 
343 int
344 _fini(void)
345 {
346 	int	err;
347 
348 	if ((err = mod_remove(&modlinkage)) == 0)
349 		drmach_fini();
350 
351 	return (err);
352 }
353 
354 int
355 _info(struct modinfo *modinfop)
356 {
357 	return (mod_info(&modlinkage, modinfop));
358 }
359 
360 struct drmach_mc_lookup {
361 	int	bnum;
362 	drmach_board_t	*bp;
363 	dev_info_t *dip;	/* rv - set if found */
364 };
365 
366 #define	_ptob64(p) ((uint64_t)(p) << PAGESHIFT)
367 #define	_b64top(b) ((pgcnt_t)((b) >> PAGESHIFT))
368 
369 static int
370 drmach_setup_mc_info(dev_info_t *dip, drmach_mem_t *mp)
371 {
372 	uint64_t	memory_ranges[128];
373 	int len;
374 	struct memlist	*ml;
375 	int rv;
376 	hwd_sb_t *hwd;
377 	hwd_memory_t *pm;
378 
379 	len = sizeof (memory_ranges);
380 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
381 		DDI_PROP_DONTPASS, "sb-mem-ranges",
382 	    (caddr_t)&memory_ranges[0], &len) != DDI_PROP_SUCCESS) {
383 		mp->slice_base = 0;
384 		mp->slice_size = 0;
385 		return (-1);
386 	}
387 	mp->slice_base = memory_ranges[0];
388 	mp->slice_size = memory_ranges[1];
389 
390 	if (!mp->dev.bp->boot_board) {
391 		int i;
392 
393 		rv = opl_read_hwd(mp->dev.bp->bnum, NULL,  NULL, NULL, &hwd);
394 
395 		if (rv != 0) {
396 			return (-1);
397 		}
398 
399 		ml = NULL;
400 		pm = &hwd->sb_cmu.cmu_memory;
401 		for (i = 0; i < HWD_MAX_MEM_CHUNKS; i++) {
402 			if (pm->mem_chunks[i].chnk_size > 0) {
403 				ml = memlist_add_span(ml,
404 					pm->mem_chunks[i].chnk_start_address,
405 					pm->mem_chunks[i].chnk_size);
406 			}
407 		}
408 	} else {
409 		/*
410 		 * we intersect phys_install to get base_pa.
411 		 * This only works at bootup time.
412 		 */
413 
414 		memlist_read_lock();
415 		ml = memlist_dup(phys_install);
416 		memlist_read_unlock();
417 
418 		ml = memlist_del_span(ml, 0ull, mp->slice_base);
419 		if (ml) {
420 			uint64_t basepa, endpa;
421 			endpa = _ptob64(physmax + 1);
422 
423 			basepa = mp->slice_base + mp->slice_size;
424 
425 			ml = memlist_del_span(ml, basepa, endpa - basepa);
426 		}
427 	}
428 
429 	if (ml) {
430 		uint64_t nbytes = 0;
431 		struct memlist *p;
432 		for (p = ml; p; p = p->next) {
433 			nbytes += p->size;
434 		}
435 		if ((mp->nbytes = nbytes) > 0)
436 			mp->base_pa = ml->address;
437 		else
438 			mp->base_pa = 0;
439 		mp->memlist = ml;
440 	} else {
441 		mp->base_pa = 0;
442 		mp->nbytes = 0;
443 	}
444 	return (0);
445 }
446 
447 
448 struct drmach_hotcpu {
449 	drmach_board_t *bp;
450 	int	bnum;
451 	int	core_id;
452 	int 	rv;
453 	int	option;
454 };
455 
456 static int
457 drmach_cpu_cb(dev_info_t *dip, void *arg)
458 {
459 	struct drmach_hotcpu *p = (struct drmach_hotcpu *)arg;
460 	char name[OBP_MAXDRVNAME];
461 	int len = OBP_MAXDRVNAME;
462 	int bnum, core_id, strand_id;
463 	drmach_board_t *bp;
464 
465 	if (dip == ddi_root_node()) {
466 		return (DDI_WALK_CONTINUE);
467 	}
468 
469 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
470 	    DDI_PROP_DONTPASS, "name",
471 	    (caddr_t)name, &len) != DDI_PROP_SUCCESS) {
472 		return (DDI_WALK_PRUNECHILD);
473 	}
474 
475 	/* only cmp has board number */
476 	bnum = -1;
477 	len = sizeof (bnum);
478 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
479 	    DDI_PROP_DONTPASS, OBP_BOARDNUM,
480 	    (caddr_t)&bnum, &len) != DDI_PROP_SUCCESS) {
481 		bnum = -1;
482 	}
483 
484 	if (strcmp(name, "cmp") == 0) {
485 		if (bnum != p->bnum)
486 			return (DDI_WALK_PRUNECHILD);
487 		return (DDI_WALK_CONTINUE);
488 	}
489 	/* we have already pruned all unwanted cores and cpu's above */
490 	if (strcmp(name, "core") == 0) {
491 		return (DDI_WALK_CONTINUE);
492 	}
493 	if (strcmp(name, "cpu") == 0) {
494 		processorid_t cpuid;
495 		len = sizeof (cpuid);
496 		if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
497 		    DDI_PROP_DONTPASS, "cpuid",
498 		    (caddr_t)&cpuid, &len) != DDI_PROP_SUCCESS) {
499 			p->rv = -1;
500 			return (DDI_WALK_TERMINATE);
501 		}
502 
503 		core_id = p->core_id;
504 
505 		bnum = LSB_ID(cpuid);
506 
507 		if (ON_BOARD_CORE_NUM(cpuid) != core_id)
508 			return (DDI_WALK_CONTINUE);
509 
510 		bp = p->bp;
511 		ASSERT(bnum == bp->bnum);
512 
513 		if (p->option == HOTADD_CPU) {
514 			if (prom_hotaddcpu(cpuid) != 0) {
515 				p->rv = -1;
516 				return (DDI_WALK_TERMINATE);
517 			}
518 			strand_id = STRAND_ID(cpuid);
519 			bp->cores[core_id].core_hotadded |= (1 << strand_id);
520 		} else if (p->option == HOTREMOVE_CPU) {
521 			if (prom_hotremovecpu(cpuid) != 0) {
522 				p->rv = -1;
523 				return (DDI_WALK_TERMINATE);
524 			}
525 			strand_id = STRAND_ID(cpuid);
526 			bp->cores[core_id].core_hotadded &= ~(1 << strand_id);
527 		}
528 		return (DDI_WALK_CONTINUE);
529 	}
530 
531 	return (DDI_WALK_PRUNECHILD);
532 }
533 
534 
535 static int
536 drmach_add_remove_cpu(int bnum, int core_id, int option)
537 {
538 	struct drmach_hotcpu arg;
539 	drmach_board_t *bp;
540 
541 	bp = drmach_get_board_by_bnum(bnum);
542 	ASSERT(bp);
543 
544 	arg.bp = bp;
545 	arg.bnum = bnum;
546 	arg.core_id = core_id;
547 	arg.rv = 0;
548 	arg.option = option;
549 	ddi_walk_devs(ddi_root_node(), drmach_cpu_cb, (void *)&arg);
550 	return (arg.rv);
551 }
552 
553 struct drmach_setup_core_arg {
554 	drmach_board_t *bp;
555 };
556 
557 static int
558 drmach_setup_core_cb(dev_info_t *dip, void *arg)
559 {
560 	struct drmach_setup_core_arg *p = (struct drmach_setup_core_arg *)arg;
561 	char name[OBP_MAXDRVNAME];
562 	int len = OBP_MAXDRVNAME;
563 	int bnum;
564 	int core_id, strand_id;
565 
566 	if (dip == ddi_root_node()) {
567 		return (DDI_WALK_CONTINUE);
568 	}
569 
570 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
571 	    DDI_PROP_DONTPASS, "name",
572 	    (caddr_t)name, &len) != DDI_PROP_SUCCESS) {
573 		return (DDI_WALK_PRUNECHILD);
574 	}
575 
576 	/* only cmp has board number */
577 	bnum = -1;
578 	len = sizeof (bnum);
579 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
580 	    DDI_PROP_DONTPASS, OBP_BOARDNUM,
581 	    (caddr_t)&bnum, &len) != DDI_PROP_SUCCESS) {
582 		bnum = -1;
583 	}
584 
585 	if (strcmp(name, "cmp") == 0) {
586 		if (bnum != p->bp->bnum)
587 			return (DDI_WALK_PRUNECHILD);
588 		return (DDI_WALK_CONTINUE);
589 	}
590 	/* we have already pruned all unwanted cores and cpu's above */
591 	if (strcmp(name, "core") == 0) {
592 		return (DDI_WALK_CONTINUE);
593 	}
594 	if (strcmp(name, "cpu") == 0) {
595 		processorid_t cpuid;
596 		len = sizeof (cpuid);
597 		if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
598 		    DDI_PROP_DONTPASS, "cpuid",
599 		    (caddr_t)&cpuid, &len) != DDI_PROP_SUCCESS) {
600 			return (DDI_WALK_TERMINATE);
601 		}
602 		bnum = LSB_ID(cpuid);
603 		ASSERT(bnum == p->bp->bnum);
604 		core_id = ON_BOARD_CORE_NUM(cpuid);
605 		strand_id = STRAND_ID(cpuid);
606 		p->bp->cores[core_id].core_present |= (1 << strand_id);
607 		return (DDI_WALK_CONTINUE);
608 	}
609 
610 	return (DDI_WALK_PRUNECHILD);
611 }
612 
613 
614 static void
615 drmach_setup_core_info(drmach_board_t *obj)
616 {
617 	struct drmach_setup_core_arg arg;
618 	int i;
619 
620 	for (i = 0; i < OPL_MAX_COREID_PER_BOARD; i++) {
621 		obj->cores[i].core_present = 0;
622 		obj->cores[i].core_hotadded = 0;
623 		obj->cores[i].core_started = 0;
624 	}
625 	arg.bp = obj;
626 	ddi_walk_devs(ddi_root_node(), drmach_setup_core_cb, (void *)&arg);
627 
628 	for (i = 0; i < OPL_MAX_COREID_PER_BOARD; i++) {
629 		if (obj->boot_board) {
630 			obj->cores[i].core_hotadded =
631 				obj->cores[i].core_started =
632 				obj->cores[i].core_present;
633 		}
634 	}
635 }
636 
637 /*
638  * drmach_node_* routines serve the purpose of separating the
639  * rest of the code from the device tree and OBP.  This is necessary
640  * because of In-Kernel-Probing.  Devices probed after stod, are probed
641  * by the in-kernel-prober, not OBP.  These devices, therefore, do not
642  * have dnode ids.
643  */
644 
645 typedef struct {
646 	drmach_node_walk_args_t	*nwargs;
647 	int 			(*cb)(drmach_node_walk_args_t *args);
648 	int			err;
649 } drmach_node_ddi_walk_args_t;
650 
651 static int
652 drmach_node_ddi_walk_cb(dev_info_t *dip, void *arg)
653 {
654 	drmach_node_ddi_walk_args_t	*nargs;
655 
656 	nargs = (drmach_node_ddi_walk_args_t *)arg;
657 
658 	/*
659 	 * dip doesn't have to be held here as we are called
660 	 * from ddi_walk_devs() which holds the dip.
661 	 */
662 	nargs->nwargs->node->here = (void *)dip;
663 
664 	nargs->err = nargs->cb(nargs->nwargs);
665 
666 
667 	/*
668 	 * Set "here" to NULL so that unheld dip is not accessible
669 	 * outside ddi_walk_devs()
670 	 */
671 	nargs->nwargs->node->here = NULL;
672 
673 	if (nargs->err)
674 		return (DDI_WALK_TERMINATE);
675 	else
676 		return (DDI_WALK_CONTINUE);
677 }
678 
679 static int
680 drmach_node_ddi_walk(drmach_node_t *np, void *data,
681 		int (*cb)(drmach_node_walk_args_t *args))
682 {
683 	drmach_node_walk_args_t		args;
684 	drmach_node_ddi_walk_args_t	nargs;
685 
686 
687 	/* initialized args structure for callback */
688 	args.node = np;
689 	args.data = data;
690 
691 	nargs.nwargs = &args;
692 	nargs.cb = cb;
693 	nargs.err = 0;
694 
695 	/*
696 	 * Root node doesn't have to be held in any way.
697 	 */
698 	ddi_walk_devs(ddi_root_node(), drmach_node_ddi_walk_cb,
699 		(void *)&nargs);
700 
701 	return (nargs.err);
702 }
703 
704 static int
705 drmach_node_ddi_get_parent(drmach_node_t *np, drmach_node_t *pp)
706 {
707 	dev_info_t	*ndip;
708 	static char	*fn = "drmach_node_ddi_get_parent";
709 
710 	ndip = np->n_getdip(np);
711 	if (ndip == NULL) {
712 		cmn_err(CE_WARN, "%s: NULL dip", fn);
713 		return (-1);
714 	}
715 
716 	bcopy(np, pp, sizeof (drmach_node_t));
717 
718 	pp->here = (void *)ddi_get_parent(ndip);
719 	if (pp->here == NULL) {
720 		cmn_err(CE_WARN, "%s: NULL parent dip", fn);
721 		return (-1);
722 	}
723 
724 	return (0);
725 }
726 
727 /*ARGSUSED*/
728 static pnode_t
729 drmach_node_ddi_get_dnode(drmach_node_t *np)
730 {
731 	return ((pnode_t)NULL);
732 }
733 
734 static drmach_node_t *
735 drmach_node_new(void)
736 {
737 	drmach_node_t *np;
738 
739 	np = kmem_zalloc(sizeof (drmach_node_t), KM_SLEEP);
740 
741 	np->get_dnode = drmach_node_ddi_get_dnode;
742 	np->walk = drmach_node_ddi_walk;
743 	np->n_getdip = drmach_node_ddi_get_dip;
744 	np->n_getproplen = drmach_node_ddi_get_proplen;
745 	np->n_getprop = drmach_node_ddi_get_prop;
746 	np->get_parent = drmach_node_ddi_get_parent;
747 
748 	return (np);
749 }
750 
751 static void
752 drmach_node_dispose(drmach_node_t *np)
753 {
754 	kmem_free(np, sizeof (*np));
755 }
756 
757 static dev_info_t *
758 drmach_node_ddi_get_dip(drmach_node_t *np)
759 {
760 	return ((dev_info_t *)np->here);
761 }
762 
763 static int
764 drmach_node_walk(drmach_node_t *np, void *param,
765 		int (*cb)(drmach_node_walk_args_t *args))
766 {
767 	return (np->walk(np, param, cb));
768 }
769 
770 static int
771 drmach_node_ddi_get_prop(drmach_node_t *np, char *name, void *buf, int len)
772 {
773 	int		rv = 0;
774 	dev_info_t	*ndip;
775 	static char	*fn = "drmach_node_ddi_get_prop";
776 
777 
778 	ndip = np->n_getdip(np);
779 	if (ndip == NULL) {
780 		cmn_err(CE_WARN, "%s: NULL dip", fn);
781 		rv = -1;
782 	} else if (ddi_getlongprop_buf(DDI_DEV_T_ANY, ndip,
783 	    DDI_PROP_DONTPASS, name,
784 	    (caddr_t)buf, &len) != DDI_PROP_SUCCESS) {
785 		rv = -1;
786 	}
787 
788 	return (rv);
789 }
790 
791 static int
792 drmach_node_ddi_get_proplen(drmach_node_t *np, char *name, int *len)
793 {
794 	int		rv = 0;
795 	dev_info_t	*ndip;
796 
797 	ndip = np->n_getdip(np);
798 	if (ndip == NULL) {
799 		rv = -1;
800 	} else if (ddi_getproplen(DDI_DEV_T_ANY, ndip, DDI_PROP_DONTPASS,
801 		name, len) != DDI_PROP_SUCCESS) {
802 		rv = -1;
803 	}
804 
805 	return (rv);
806 }
807 
808 static drmachid_t
809 drmach_node_dup(drmach_node_t *np)
810 {
811 	drmach_node_t *dup;
812 
813 	dup = drmach_node_new();
814 	dup->here = np->here;
815 	dup->get_dnode = np->get_dnode;
816 	dup->walk = np->walk;
817 	dup->n_getdip = np->n_getdip;
818 	dup->n_getproplen = np->n_getproplen;
819 	dup->n_getprop = np->n_getprop;
820 	dup->get_parent = np->get_parent;
821 
822 	return (dup);
823 }
824 
825 /*
826  * drmach_array provides convenient array construction, access,
827  * bounds checking and array destruction logic.
828  */
829 
830 static drmach_array_t *
831 drmach_array_new(int min_index, int max_index)
832 {
833 	drmach_array_t *arr;
834 
835 	arr = kmem_zalloc(sizeof (drmach_array_t), KM_SLEEP);
836 
837 	arr->arr_sz = (max_index - min_index + 1) * sizeof (void *);
838 	if (arr->arr_sz > 0) {
839 		arr->min_index = min_index;
840 		arr->max_index = max_index;
841 
842 		arr->arr = kmem_zalloc(arr->arr_sz, KM_SLEEP);
843 		return (arr);
844 	} else {
845 		kmem_free(arr, sizeof (*arr));
846 		return (0);
847 	}
848 }
849 
850 static int
851 drmach_array_set(drmach_array_t *arr, int idx, drmachid_t val)
852 {
853 	if (idx < arr->min_index || idx > arr->max_index)
854 		return (-1);
855 	else {
856 		arr->arr[idx - arr->min_index] = val;
857 		return (0);
858 	}
859 	/*NOTREACHED*/
860 }
861 
862 static int
863 drmach_array_get(drmach_array_t *arr, int idx, drmachid_t *val)
864 {
865 	if (idx < arr->min_index || idx > arr->max_index)
866 		return (-1);
867 	else {
868 		*val = arr->arr[idx - arr->min_index];
869 		return (0);
870 	}
871 	/*NOTREACHED*/
872 }
873 
874 static int
875 drmach_array_first(drmach_array_t *arr, int *idx, drmachid_t *val)
876 {
877 	int rv;
878 
879 	*idx = arr->min_index;
880 	while ((rv = drmach_array_get(arr, *idx, val)) == 0 && *val == NULL)
881 		*idx += 1;
882 
883 	return (rv);
884 }
885 
886 static int
887 drmach_array_next(drmach_array_t *arr, int *idx, drmachid_t *val)
888 {
889 	int rv;
890 
891 	*idx += 1;
892 	while ((rv = drmach_array_get(arr, *idx, val)) == 0 && *val == NULL)
893 		*idx += 1;
894 
895 	return (rv);
896 }
897 
898 static void
899 drmach_array_dispose(drmach_array_t *arr, void (*disposer)(drmachid_t))
900 {
901 	drmachid_t	val;
902 	int		idx;
903 	int		rv;
904 
905 	rv = drmach_array_first(arr, &idx, &val);
906 	while (rv == 0) {
907 		(*disposer)(val);
908 		rv = drmach_array_next(arr, &idx, &val);
909 	}
910 
911 	kmem_free(arr->arr, arr->arr_sz);
912 	kmem_free(arr, sizeof (*arr));
913 }
914 
915 static drmach_board_t *
916 drmach_get_board_by_bnum(int bnum)
917 {
918 	drmachid_t id;
919 
920 	if (drmach_array_get(drmach_boards, bnum, &id) == 0)
921 		return ((drmach_board_t *)id);
922 	else
923 		return (NULL);
924 }
925 
926 static pnode_t
927 drmach_node_get_dnode(drmach_node_t *np)
928 {
929 	return (np->get_dnode(np));
930 }
931 
932 /*ARGSUSED*/
933 sbd_error_t *
934 drmach_configure(drmachid_t id, int flags)
935 {
936 	drmach_device_t		*dp;
937 	sbd_error_t		*err = NULL;
938 	dev_info_t		*rdip;
939 	dev_info_t		*fdip = NULL;
940 
941 	if (DRMACH_IS_CPU_ID(id)) {
942 		return (NULL);
943 	}
944 	if (!DRMACH_IS_DEVICE_ID(id))
945 		return (drerr_new(0, EOPL_INAPPROP, NULL));
946 	dp = id;
947 	rdip = dp->node->n_getdip(dp->node);
948 
949 	ASSERT(rdip);
950 
951 	ASSERT(e_ddi_branch_held(rdip));
952 
953 	if (e_ddi_branch_configure(rdip, &fdip, 0) != 0) {
954 		char *path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
955 		dev_info_t *dip = (fdip != NULL) ? fdip : rdip;
956 
957 		(void) ddi_pathname(dip, path);
958 		err = drerr_new(1,  EOPL_DRVFAIL, path);
959 
960 		kmem_free(path, MAXPATHLEN);
961 
962 		/* If non-NULL, fdip is returned held and must be released */
963 		if (fdip != NULL)
964 			ddi_release_devi(fdip);
965 	}
966 
967 	return (err);
968 }
969 
970 
971 static sbd_error_t *
972 drmach_device_new(drmach_node_t *node,
973 	drmach_board_t *bp, int portid, drmachid_t *idp)
974 {
975 	int		 i;
976 	int		 rv;
977 	drmach_device_t	proto;
978 	sbd_error_t	*err;
979 	char		 name[OBP_MAXDRVNAME];
980 
981 	rv = node->n_getprop(node, "name", name, OBP_MAXDRVNAME);
982 	if (rv) {
983 		/* every node is expected to have a name */
984 		err = drerr_new(1, EOPL_GETPROP,
985 			"device node %s: property %s",
986 			ddi_node_name(node->n_getdip(node)), "name");
987 		return (err);
988 	}
989 
990 	/*
991 	 * The node currently being examined is not listed in the name2type[]
992 	 * array.  In this case, the node is no interest to drmach.  Both
993 	 * dp and err are initialized here to yield nothing (no device or
994 	 * error structure) for this case.
995 	 */
996 	i = drmach_name2type_idx(name);
997 
998 
999 	if (i < 0) {
1000 		*idp = (drmachid_t)0;
1001 		return (NULL);
1002 	}
1003 
1004 	/* device specific new function will set unum */
1005 
1006 	bzero(&proto, sizeof (proto));
1007 	proto.type = drmach_name2type[i].type;
1008 	proto.bp = bp;
1009 	proto.node = node;
1010 	proto.portid = portid;
1011 
1012 	return (drmach_name2type[i].new(&proto, idp));
1013 }
1014 
1015 static void
1016 drmach_device_dispose(drmachid_t id)
1017 {
1018 	drmach_device_t *self = id;
1019 
1020 	self->cm.dispose(id);
1021 }
1022 
1023 
1024 static drmach_board_t *
1025 drmach_board_new(int bnum, int boot_board)
1026 {
1027 	static sbd_error_t *drmach_board_release(drmachid_t);
1028 	static sbd_error_t *drmach_board_status(drmachid_t, drmach_status_t *);
1029 
1030 	drmach_board_t	*bp;
1031 
1032 	bp = kmem_zalloc(sizeof (drmach_board_t), KM_SLEEP);
1033 
1034 	bp->cm.isa = (void *)drmach_board_new;
1035 	bp->cm.release = drmach_board_release;
1036 	bp->cm.status = drmach_board_status;
1037 
1038 	(void) drmach_board_name(bnum, bp->cm.name, sizeof (bp->cm.name));
1039 
1040 	bp->bnum = bnum;
1041 	bp->devices = NULL;
1042 	bp->connected = boot_board;
1043 	bp->tree = drmach_node_new();
1044 	bp->assigned = boot_board;
1045 	bp->powered = boot_board;
1046 	bp->boot_board = boot_board;
1047 
1048 	/*
1049 	 * If this is not bootup initialization, we have to wait till
1050 	 * IKP sets up the device nodes in drmach_board_connect().
1051 	 */
1052 	if (boot_board)
1053 		drmach_setup_core_info(bp);
1054 
1055 	drmach_array_set(drmach_boards, bnum, bp);
1056 	return (bp);
1057 }
1058 
1059 static void
1060 drmach_board_dispose(drmachid_t id)
1061 {
1062 	drmach_board_t *bp;
1063 
1064 	ASSERT(DRMACH_IS_BOARD_ID(id));
1065 	bp = id;
1066 
1067 	if (bp->tree)
1068 		drmach_node_dispose(bp->tree);
1069 
1070 	if (bp->devices)
1071 		drmach_array_dispose(bp->devices, drmach_device_dispose);
1072 
1073 	kmem_free(bp, sizeof (*bp));
1074 }
1075 
1076 static sbd_error_t *
1077 drmach_board_status(drmachid_t id, drmach_status_t *stat)
1078 {
1079 	sbd_error_t	*err = NULL;
1080 	drmach_board_t	*bp;
1081 
1082 	if (!DRMACH_IS_BOARD_ID(id))
1083 		return (drerr_new(0, EOPL_INAPPROP, NULL));
1084 	bp = id;
1085 
1086 	stat->assigned = bp->assigned;
1087 	stat->powered = bp->powered;
1088 	stat->busy = 0;			/* assume not busy */
1089 	stat->configured = 0;		/* assume not configured */
1090 	stat->empty = 0;
1091 	stat->cond = bp->cond = SBD_COND_OK;
1092 	strncpy(stat->type, "System Brd", sizeof (stat->type));
1093 	stat->info[0] = '\0';
1094 
1095 	if (bp->devices) {
1096 		int		 rv;
1097 		int		 d_idx;
1098 		drmachid_t	 d_id;
1099 
1100 		rv = drmach_array_first(bp->devices, &d_idx, &d_id);
1101 		while (rv == 0) {
1102 			drmach_status_t	d_stat;
1103 
1104 			err = drmach_i_status(d_id, &d_stat);
1105 			if (err)
1106 				break;
1107 
1108 			stat->busy |= d_stat.busy;
1109 			stat->configured |= d_stat.configured;
1110 
1111 			rv = drmach_array_next(bp->devices, &d_idx, &d_id);
1112 		}
1113 	}
1114 
1115 	return (err);
1116 }
1117 
1118 int
1119 drmach_board_is_floating(drmachid_t id)
1120 {
1121 	drmach_board_t *bp;
1122 
1123 	if (!DRMACH_IS_BOARD_ID(id))
1124 		return (0);
1125 
1126 	bp = (drmach_board_t *)id;
1127 
1128 	return ((drmach_domain.floating & (1 << bp->bnum)) ? 1 : 0);
1129 }
1130 
1131 static int
1132 drmach_init(void)
1133 {
1134 	dev_info_t	*rdip;
1135 	int		i, rv, len;
1136 	int		*floating;
1137 
1138 	rw_init(&drmach_boards_rwlock, NULL, RW_DEFAULT, NULL);
1139 
1140 	drmach_boards = drmach_array_new(0, MAX_BOARDS - 1);
1141 
1142 	rdip = ddi_root_node();
1143 
1144 	if (ddi_getproplen(DDI_DEV_T_ANY, rdip, DDI_PROP_DONTPASS,
1145 		"floating-boards", &len) != DDI_PROP_SUCCESS) {
1146 		cmn_err(CE_WARN, "Cannot get floating-boards proplen\n");
1147 	} else {
1148 		floating = (int *)kmem_alloc(len, KM_SLEEP);
1149 		rv = ddi_prop_op(DDI_DEV_T_ANY, rdip,
1150 			PROP_LEN_AND_VAL_BUF, DDI_PROP_DONTPASS,
1151 			"floating-boards", (caddr_t)floating, &len);
1152 		if (rv != DDI_PROP_SUCCESS) {
1153 			cmn_err(CE_WARN, "Cannot get floating-boards prop\n");
1154 		} else {
1155 			drmach_domain.floating = 0;
1156 			for (i = 0; i < len / sizeof (int); i++) {
1157 				drmach_domain.floating |= (1 << floating[i]);
1158 			}
1159 		}
1160 		kmem_free(floating, len);
1161 	}
1162 	drmach_domain.allow_dr = opl_check_dr_status();
1163 
1164 	rdip = ddi_get_child(ddi_root_node());
1165 	do {
1166 		int		 bnum;
1167 		drmachid_t	 id;
1168 
1169 		bnum = -1;
1170 		bnum = ddi_getprop(DDI_DEV_T_ANY, rdip,
1171 			DDI_PROP_DONTPASS, OBP_BOARDNUM, -1);
1172 		if (bnum == -1)
1173 			continue;
1174 
1175 		if (drmach_array_get(drmach_boards, bnum, &id) == -1) {
1176 			cmn_err(CE_WARN, "Device node 0x%p has"
1177 				" invalid property value, %s=%d",
1178 					rdip, OBP_BOARDNUM, bnum);
1179 			goto error;
1180 		} else if (id == NULL) {
1181 			(void) drmach_board_new(bnum, 1);
1182 		}
1183 	} while ((rdip = ddi_get_next_sibling(rdip)) != NULL);
1184 
1185 	opl_hold_devtree();
1186 
1187 	/*
1188 	 * Initialize the IKP feature.
1189 	 *
1190 	 * This can be done only after DR has acquired a hold on all the
1191 	 * device nodes that are interesting to IKP.
1192 	 */
1193 	if (opl_init_cfg() != 0) {
1194 		cmn_err(CE_WARN, "DR - IKP initialization failed");
1195 
1196 		opl_release_devtree();
1197 
1198 		goto error;
1199 	}
1200 
1201 	return (0);
1202 error:
1203 	drmach_array_dispose(drmach_boards, drmach_board_dispose);
1204 	rw_destroy(&drmach_boards_rwlock);
1205 	return (ENXIO);
1206 }
1207 
1208 static void
1209 drmach_fini(void)
1210 {
1211 	rw_enter(&drmach_boards_rwlock, RW_WRITER);
1212 	drmach_array_dispose(drmach_boards, drmach_board_dispose);
1213 	drmach_boards = NULL;
1214 	rw_exit(&drmach_boards_rwlock);
1215 
1216 	/*
1217 	 * Walk immediate children of the root devinfo node
1218 	 * releasing holds acquired on branches in drmach_init()
1219 	 */
1220 
1221 	opl_release_devtree();
1222 
1223 	rw_destroy(&drmach_boards_rwlock);
1224 }
1225 
1226 /*
1227  *	Each system board contains 2 Oberon PCI bridge and
1228  *	1 CMUCH.
1229  *	Each oberon has 2 channels.
1230  *	Each channel has 2 pci-ex leaf.
1231  *	Each CMUCH has 1 pci bus.
1232  *
1233  *
1234  *	Device Path:
1235  *	/pci@<portid>,reg
1236  *
1237  *	where
1238  *	portid[10] = 0
1239  *	portid[9:0] = LLEAF_ID[9:0] of the Oberon Channel
1240  *
1241  *	LLEAF_ID[9:8] = 0
1242  *	LLEAF_ID[8:4] = LSB_ID[4:0]
1243  *	LLEAF_ID[3:1] = IO Channel#[2:0] (0,1,2,3 for Oberon)
1244  *			channel 4 is pcicmu
1245  *	LLEAF_ID[0] = PCI Leaf Number (0 for leaf-A, 1 for leaf-B)
1246  *
1247  *	Properties:
1248  *	name = pci
1249  *	device_type = "pciex"
1250  *	board# = LSBID
1251  *	reg = int32 * 2, Oberon CSR space of the leaf and the UBC space
1252  *	portid = Jupiter Bus Device ID ((LSB_ID << 3)|pciport#)
1253  */
1254 
1255 static sbd_error_t *
1256 drmach_io_new(drmach_device_t *proto, drmachid_t *idp)
1257 {
1258 	drmach_io_t	*ip;
1259 
1260 	int		 portid;
1261 
1262 	portid = proto->portid;
1263 	ASSERT(portid != -1);
1264 	proto->unum = portid & (MAX_IO_UNITS_PER_BOARD - 1);
1265 
1266 	ip = kmem_zalloc(sizeof (drmach_io_t), KM_SLEEP);
1267 	bcopy(proto, &ip->dev, sizeof (ip->dev));
1268 	ip->dev.node = drmach_node_dup(proto->node);
1269 	ip->dev.cm.isa = (void *)drmach_io_new;
1270 	ip->dev.cm.dispose = drmach_io_dispose;
1271 	ip->dev.cm.release = drmach_io_release;
1272 	ip->dev.cm.status = drmach_io_status;
1273 	ip->channel = (portid >> 1) & 0x7;
1274 	ip->leaf = (portid & 0x1);
1275 
1276 	snprintf(ip->dev.cm.name, sizeof (ip->dev.cm.name), "%s%d",
1277 		ip->dev.type, ip->dev.unum);
1278 
1279 	*idp = (drmachid_t)ip;
1280 	return (NULL);
1281 }
1282 
1283 
1284 static void
1285 drmach_io_dispose(drmachid_t id)
1286 {
1287 	drmach_io_t *self;
1288 
1289 	ASSERT(DRMACH_IS_IO_ID(id));
1290 
1291 	self = id;
1292 	if (self->dev.node)
1293 		drmach_node_dispose(self->dev.node);
1294 
1295 	kmem_free(self, sizeof (*self));
1296 }
1297 
1298 /*ARGSUSED*/
1299 sbd_error_t *
1300 drmach_pre_op(int cmd, drmachid_t id, drmach_opts_t *opts)
1301 {
1302 	drmach_board_t	*bp = (drmach_board_t *)id;
1303 	sbd_error_t	*err = NULL;
1304 
1305 	/* allow status and ncm operations to always succeed */
1306 	if ((cmd == SBD_CMD_STATUS) || (cmd == SBD_CMD_GETNCM)) {
1307 		return (NULL);
1308 	}
1309 
1310 	/* check all other commands for the required option string */
1311 
1312 	if ((opts->size > 0) && (opts->copts != NULL)) {
1313 
1314 		DRMACH_PR("platform options: %s\n", opts->copts);
1315 
1316 		if (strstr(opts->copts, "opldr") == NULL) {
1317 			err = drerr_new(1, EOPL_SUPPORT, NULL);
1318 		}
1319 	} else {
1320 		err = drerr_new(1, EOPL_SUPPORT, NULL);
1321 	}
1322 
1323 	if (!err && id && DRMACH_IS_BOARD_ID(id)) {
1324 		switch (cmd) {
1325 			case SBD_CMD_TEST:
1326 			case SBD_CMD_STATUS:
1327 			case SBD_CMD_GETNCM:
1328 				break;
1329 			case SBD_CMD_CONNECT:
1330 				if (bp->connected)
1331 					err = drerr_new(0, ESBD_STATE, NULL);
1332 				else if (!drmach_domain.allow_dr)
1333 					err = drerr_new(1, EOPL_SUPPORT,
1334 						NULL);
1335 				break;
1336 			case SBD_CMD_DISCONNECT:
1337 				if (!bp->connected)
1338 					err = drerr_new(0, ESBD_STATE, NULL);
1339 				else if (!drmach_domain.allow_dr)
1340 					err = drerr_new(1, EOPL_SUPPORT,
1341 						NULL);
1342 				break;
1343 			default:
1344 				if (!drmach_domain.allow_dr)
1345 					err = drerr_new(1, EOPL_SUPPORT,
1346 						NULL);
1347 				break;
1348 
1349 		}
1350 	}
1351 
1352 	return (err);
1353 }
1354 
1355 /*ARGSUSED*/
1356 sbd_error_t *
1357 drmach_post_op(int cmd, drmachid_t id, drmach_opts_t *opts)
1358 {
1359 	return (NULL);
1360 }
1361 
1362 sbd_error_t *
1363 drmach_board_assign(int bnum, drmachid_t *id)
1364 {
1365 	sbd_error_t	*err = NULL;
1366 
1367 	rw_enter(&drmach_boards_rwlock, RW_WRITER);
1368 
1369 	if (drmach_array_get(drmach_boards, bnum, id) == -1) {
1370 		err = drerr_new(1, EOPL_BNUM, "%d", bnum);
1371 	} else {
1372 		drmach_board_t	*bp;
1373 
1374 		if (*id)
1375 			rw_downgrade(&drmach_boards_rwlock);
1376 
1377 		bp = *id;
1378 		if (!(*id))
1379 			bp = *id  =
1380 				(drmachid_t)drmach_board_new(bnum, 0);
1381 		bp->assigned = 1;
1382 	}
1383 
1384 	rw_exit(&drmach_boards_rwlock);
1385 
1386 	return (err);
1387 }
1388 
1389 /*ARGSUSED*/
1390 sbd_error_t *
1391 drmach_board_connect(drmachid_t id, drmach_opts_t *opts)
1392 {
1393 	drmach_board_t	*obj = (drmach_board_t *)id;
1394 
1395 	if (!DRMACH_IS_BOARD_ID(id))
1396 		return (drerr_new(0, EOPL_INAPPROP, NULL));
1397 
1398 	if (opl_probe_sb(obj->bnum) != 0)
1399 		return (drerr_new(1, EOPL_PROBE, NULL));
1400 
1401 	(void) prom_attach_notice(obj->bnum);
1402 
1403 	drmach_setup_core_info(obj);
1404 
1405 	obj->connected = 1;
1406 
1407 	return (NULL);
1408 }
1409 
1410 static int drmach_cache_flush_flag[NCPU];
1411 
1412 /*ARGSUSED*/
1413 static void
1414 drmach_flush_cache(uint64_t id, uint64_t dummy)
1415 {
1416 	extern void cpu_flush_ecache(void);
1417 
1418 	cpu_flush_ecache();
1419 	drmach_cache_flush_flag[id] = 0;
1420 }
1421 
1422 static void
1423 drmach_flush_all()
1424 {
1425 	cpuset_t	xc_cpuset;
1426 	int		i;
1427 
1428 	xc_cpuset = cpu_ready_set;
1429 	for (i = 0; i < NCPU; i++) {
1430 		if (CPU_IN_SET(xc_cpuset, i)) {
1431 			drmach_cache_flush_flag[i] = 1;
1432 			xc_one(i, drmach_flush_cache, i, 0);
1433 			while (drmach_cache_flush_flag[i]) {
1434 				DELAY(1000);
1435 			}
1436 		}
1437 	}
1438 }
1439 
1440 static int
1441 drmach_disconnect_cpus(drmach_board_t *bp)
1442 {
1443 	int i, bnum;
1444 
1445 	bnum = bp->bnum;
1446 
1447 	for (i = 0; i < OPL_MAX_COREID_PER_BOARD; i++) {
1448 	    if (bp->cores[i].core_present) {
1449 		if (bp->cores[i].core_started)
1450 		    return (-1);
1451 		if (bp->cores[i].core_hotadded) {
1452 		    if (drmach_add_remove_cpu(bnum, i, HOTREMOVE_CPU)) {
1453 			cmn_err(CE_WARN,
1454 			    "Failed to remove CMP %d on board %d\n",
1455 			    i, bnum);
1456 			return (-1);
1457 		    }
1458 		}
1459 	    }
1460 	}
1461 	return (0);
1462 }
1463 
1464 /*ARGSUSED*/
1465 sbd_error_t *
1466 drmach_board_disconnect(drmachid_t id, drmach_opts_t *opts)
1467 {
1468 	drmach_board_t *obj;
1469 	int rv = 0;
1470 	sbd_error_t		*err = NULL;
1471 
1472 	if (DRMACH_NULL_ID(id))
1473 		return (NULL);
1474 
1475 	if (!DRMACH_IS_BOARD_ID(id))
1476 		return (drerr_new(0, EOPL_INAPPROP, NULL));
1477 
1478 	obj = (drmach_board_t *)id;
1479 
1480 	if (drmach_disconnect_cpus(obj)) {
1481 		err = drerr_new(1, EOPL_DEPROBE, obj->cm.name);
1482 		return (err);
1483 	}
1484 
1485 	rv = opl_unprobe_sb(obj->bnum);
1486 
1487 	if (rv == 0) {
1488 		prom_detach_notice(obj->bnum);
1489 		obj->connected = 0;
1490 
1491 	} else
1492 		err = drerr_new(1, EOPL_DEPROBE, obj->cm.name);
1493 
1494 	return (err);
1495 }
1496 
1497 static int
1498 drmach_get_portid(drmach_node_t *np)
1499 {
1500 	int		portid;
1501 	char		type[OBP_MAXPROPNAME];
1502 
1503 	if (np->n_getprop(np, "portid", &portid, sizeof (portid)) == 0)
1504 		return (portid);
1505 
1506 	/*
1507 	 * Get the device_type property to see if we should
1508 	 * continue processing this node.
1509 	 */
1510 	if (np->n_getprop(np, "device_type", &type, sizeof (type)) != 0)
1511 		return (-1);
1512 
1513 	if (strcmp(type, OPL_CPU_NODE) == 0) {
1514 		/*
1515 		 * We return cpuid because it has no portid
1516 		 */
1517 		if (np->n_getprop(np, "cpuid", &portid, sizeof (portid)) == 0)
1518 			return (portid);
1519 	}
1520 
1521 	return (-1);
1522 }
1523 
1524 /*
1525  * This is a helper function to determine if a given
1526  * node should be considered for a dr operation according
1527  * to predefined dr type nodes and the node's name.
1528  * Formal Parameter : The name of a device node.
1529  * Return Value: -1, name does not map to a valid dr type.
1530  *		 A value greater or equal to 0, name is a valid dr type.
1531  */
1532 static int
1533 drmach_name2type_idx(char *name)
1534 {
1535 	int 	index, ntypes;
1536 
1537 	if (name == NULL)
1538 		return (-1);
1539 
1540 	/*
1541 	 * Determine how many possible types are currently supported
1542 	 * for dr.
1543 	 */
1544 	ntypes = sizeof (drmach_name2type) / sizeof (drmach_name2type[0]);
1545 
1546 	/* Determine if the node's name correspond to a predefined type. */
1547 	for (index = 0; index < ntypes; index++) {
1548 		if (strcmp(drmach_name2type[index].name, name) == 0)
1549 			/* The node is an allowed type for dr. */
1550 			return (index);
1551 	}
1552 
1553 	/*
1554 	 * If the name of the node does not map to any of the
1555 	 * types in the array drmach_name2type then the node is not of
1556 	 * interest to dr.
1557 	 */
1558 	return (-1);
1559 }
1560 
1561 /*
1562  * there is some complication on OPL:
1563  * - pseudo-mc nodes do not have portid property
1564  * - portid[9:5] of cmp node is LSB #, portid[7:3] of pci is LSB#
1565  * - cmp has board#
1566  * - core and cpu nodes do not have portid and board# properties
1567  * starcat uses portid to derive the board# but that does not work
1568  * for us.  starfire reads board# property to filter the devices.
1569  * That does not work either.  So for these specific device,
1570  * we use specific hard coded methods to get the board# -
1571  * cpu: LSB# = CPUID[9:5]
1572  */
1573 
1574 static int
1575 drmach_board_find_devices_cb(drmach_node_walk_args_t *args)
1576 {
1577 	drmach_node_t			*node = args->node;
1578 	drmach_board_cb_data_t		*data = args->data;
1579 	drmach_board_t			*obj = data->obj;
1580 
1581 	int		rv, portid;
1582 	int		bnum;
1583 	drmachid_t	id;
1584 	drmach_device_t	*device;
1585 	char name[OBP_MAXDRVNAME];
1586 
1587 	portid = drmach_get_portid(node);
1588 	/*
1589 	 * core, cpu and pseudo-mc do not have portid
1590 	 * we use cpuid as the portid of the cpu node
1591 	 * for pseudo-mc, we do not use portid info.
1592 	 */
1593 
1594 	rv = node->n_getprop(node, "name", name, OBP_MAXDRVNAME);
1595 	if (rv)
1596 		return (0);
1597 
1598 
1599 	rv = node->n_getprop(node, OBP_BOARDNUM, &bnum, sizeof (bnum));
1600 
1601 	if (rv) {
1602 		/*
1603 		 * cpu does not have board# property.  We use
1604 		 * CPUID[9:5]
1605 		 */
1606 		if (strcmp("cpu", name) == 0) {
1607 			bnum = (portid >> 5) & 0x1f;
1608 		} else
1609 			return (0);
1610 	}
1611 
1612 
1613 	if (bnum != obj->bnum)
1614 		return (0);
1615 
1616 	if (drmach_name2type_idx(name) < 0) {
1617 		return (0);
1618 	}
1619 
1620 	/*
1621 	 * Create a device data structure from this node data.
1622 	 * The call may yield nothing if the node is not of interest
1623 	 * to drmach.
1624 	 */
1625 	data->err = drmach_device_new(node, obj, portid, &id);
1626 	if (data->err)
1627 		return (-1);
1628 	else if (!id) {
1629 		/*
1630 		 * drmach_device_new examined the node we passed in
1631 		 * and determined that it was one not of interest to
1632 		 * drmach.  So, it is skipped.
1633 		 */
1634 		return (0);
1635 	}
1636 
1637 	rv = drmach_array_set(obj->devices, data->ndevs++, id);
1638 	if (rv) {
1639 		data->err = DRMACH_INTERNAL_ERROR();
1640 		return (-1);
1641 	}
1642 	device = id;
1643 
1644 	data->err = (*data->found)(data->a, device->type, device->unum, id);
1645 	return (data->err == NULL ? 0 : -1);
1646 }
1647 
1648 sbd_error_t *
1649 drmach_board_find_devices(drmachid_t id, void *a,
1650 	sbd_error_t *(*found)(void *a, const char *, int, drmachid_t))
1651 {
1652 	drmach_board_t		*bp = (drmach_board_t *)id;
1653 	sbd_error_t		*err;
1654 	int			 max_devices;
1655 	int			 rv;
1656 	drmach_board_cb_data_t	data;
1657 
1658 
1659 	if (!DRMACH_IS_BOARD_ID(id))
1660 		return (drerr_new(0, EOPL_INAPPROP, NULL));
1661 
1662 	max_devices  = MAX_CPU_UNITS_PER_BOARD;
1663 	max_devices += MAX_MEM_UNITS_PER_BOARD;
1664 	max_devices += MAX_IO_UNITS_PER_BOARD;
1665 
1666 	bp->devices = drmach_array_new(0, max_devices);
1667 
1668 	if (bp->tree == NULL)
1669 		bp->tree = drmach_node_new();
1670 
1671 	data.obj = bp;
1672 	data.ndevs = 0;
1673 	data.found = found;
1674 	data.a = a;
1675 	data.err = NULL;
1676 
1677 	rv = drmach_node_walk(bp->tree, &data, drmach_board_find_devices_cb);
1678 	if (rv == 0)
1679 		err = NULL;
1680 	else {
1681 		drmach_array_dispose(bp->devices, drmach_device_dispose);
1682 		bp->devices = NULL;
1683 
1684 		if (data.err)
1685 			err = data.err;
1686 		else
1687 			err = DRMACH_INTERNAL_ERROR();
1688 	}
1689 
1690 	return (err);
1691 }
1692 
1693 int
1694 drmach_board_lookup(int bnum, drmachid_t *id)
1695 {
1696 	int	rv = 0;
1697 
1698 	rw_enter(&drmach_boards_rwlock, RW_READER);
1699 	if (drmach_array_get(drmach_boards, bnum, id)) {
1700 		*id = 0;
1701 		rv = -1;
1702 	}
1703 	rw_exit(&drmach_boards_rwlock);
1704 	return (rv);
1705 }
1706 
1707 sbd_error_t *
1708 drmach_board_name(int bnum, char *buf, int buflen)
1709 {
1710 	snprintf(buf, buflen, "SB%d", bnum);
1711 	return (NULL);
1712 }
1713 
1714 sbd_error_t *
1715 drmach_board_poweroff(drmachid_t id)
1716 {
1717 	drmach_board_t	*bp;
1718 	sbd_error_t	*err;
1719 	drmach_status_t	 stat;
1720 
1721 	if (DRMACH_NULL_ID(id))
1722 		return (NULL);
1723 
1724 	if (!DRMACH_IS_BOARD_ID(id))
1725 		return (drerr_new(0, EOPL_INAPPROP, NULL));
1726 	bp = id;
1727 
1728 	err = drmach_board_status(id, &stat);
1729 
1730 	if (!err) {
1731 		if (stat.configured || stat.busy)
1732 			err = drerr_new(0, EOPL_CONFIGBUSY, bp->cm.name);
1733 		else {
1734 			bp->powered = 0;
1735 		}
1736 	}
1737 	return (err);
1738 }
1739 
1740 sbd_error_t *
1741 drmach_board_poweron(drmachid_t id)
1742 {
1743 	drmach_board_t	*bp;
1744 
1745 	if (!DRMACH_IS_BOARD_ID(id))
1746 		return (drerr_new(0, EOPL_INAPPROP, NULL));
1747 	bp = id;
1748 
1749 	bp->powered = 1;
1750 
1751 	return (NULL);
1752 }
1753 
1754 static sbd_error_t *
1755 drmach_board_release(drmachid_t id)
1756 {
1757 	if (!DRMACH_IS_BOARD_ID(id))
1758 		return (drerr_new(0, EOPL_INAPPROP, NULL));
1759 	return (NULL);
1760 }
1761 
1762 /*ARGSUSED*/
1763 sbd_error_t *
1764 drmach_board_test(drmachid_t id, drmach_opts_t *opts, int force)
1765 {
1766 	return (NULL);
1767 }
1768 
1769 sbd_error_t *
1770 drmach_board_unassign(drmachid_t id)
1771 {
1772 	drmach_board_t	*bp;
1773 	sbd_error_t	*err;
1774 	drmach_status_t	 stat;
1775 
1776 	if (DRMACH_NULL_ID(id))
1777 		return (NULL);
1778 
1779 	if (!DRMACH_IS_BOARD_ID(id)) {
1780 		return (drerr_new(0, EOPL_INAPPROP, NULL));
1781 	}
1782 	bp = id;
1783 
1784 	rw_enter(&drmach_boards_rwlock, RW_WRITER);
1785 
1786 	err = drmach_board_status(id, &stat);
1787 	if (err) {
1788 		rw_exit(&drmach_boards_rwlock);
1789 		return (err);
1790 	}
1791 	if (stat.configured || stat.busy) {
1792 		err = drerr_new(0, EOPL_CONFIGBUSY, bp->cm.name);
1793 	} else {
1794 		if (drmach_array_set(drmach_boards, bp->bnum, 0) != 0)
1795 			err = DRMACH_INTERNAL_ERROR();
1796 		else
1797 			drmach_board_dispose(bp);
1798 	}
1799 	rw_exit(&drmach_boards_rwlock);
1800 	return (err);
1801 }
1802 
1803 /*
1804  * We have to do more on OPL - e.g. set up sram tte, read cpuid, strand id,
1805  * implementation #, etc
1806  */
1807 
1808 static sbd_error_t *
1809 drmach_cpu_new(drmach_device_t *proto, drmachid_t *idp)
1810 {
1811 	static void drmach_cpu_dispose(drmachid_t);
1812 	static sbd_error_t *drmach_cpu_release(drmachid_t);
1813 	static sbd_error_t *drmach_cpu_status(drmachid_t, drmach_status_t *);
1814 
1815 	int		 portid;
1816 	drmach_cpu_t	*cp = NULL;
1817 
1818 	/* portid is CPUID of the node */
1819 	portid = proto->portid;
1820 	ASSERT(portid != -1);
1821 
1822 	/* unum = (CMP/CHIP ID) + (ON_BOARD_CORE_NUM * MAX_CMPID_PER_BOARD) */
1823 	proto->unum = ((portid/OPL_MAX_CPUID_PER_CMP) &
1824 		(OPL_MAX_CMPID_PER_BOARD - 1)) +
1825 		((portid & (OPL_MAX_CPUID_PER_CMP - 1)) *
1826 		(OPL_MAX_CMPID_PER_BOARD));
1827 
1828 	cp = kmem_zalloc(sizeof (drmach_cpu_t), KM_SLEEP);
1829 	bcopy(proto, &cp->dev, sizeof (cp->dev));
1830 	cp->dev.node = drmach_node_dup(proto->node);
1831 	cp->dev.cm.isa = (void *)drmach_cpu_new;
1832 	cp->dev.cm.dispose = drmach_cpu_dispose;
1833 	cp->dev.cm.release = drmach_cpu_release;
1834 	cp->dev.cm.status = drmach_cpu_status;
1835 
1836 	snprintf(cp->dev.cm.name, sizeof (cp->dev.cm.name), "%s%d",
1837 		cp->dev.type, cp->dev.unum);
1838 
1839 /*
1840  *	CPU ID representation
1841  *	CPUID[9:5] = SB#
1842  *	CPUID[4:3] = Chip#
1843  *	CPUID[2:1] = Core# (Only 2 core for OPL)
1844  *	CPUID[0:0] = Strand#
1845  */
1846 
1847 /*
1848  *	reg property of the strand contains strand ID
1849  *	reg property of the parent node contains core ID
1850  *	We should use them.
1851  */
1852 	cp->cpuid = portid;
1853 	cp->sb = (portid >> 5) & 0x1f;
1854 	cp->chipid = (portid >> 3) & 0x3;
1855 	cp->coreid = (portid >> 1) & 0x3;
1856 	cp->strandid = portid & 0x1;
1857 
1858 	*idp = (drmachid_t)cp;
1859 	return (NULL);
1860 }
1861 
1862 
1863 static void
1864 drmach_cpu_dispose(drmachid_t id)
1865 {
1866 	drmach_cpu_t	*self;
1867 
1868 	ASSERT(DRMACH_IS_CPU_ID(id));
1869 
1870 	self = id;
1871 	if (self->dev.node)
1872 		drmach_node_dispose(self->dev.node);
1873 
1874 	kmem_free(self, sizeof (*self));
1875 }
1876 
1877 static int
1878 drmach_cpu_start(struct cpu *cp)
1879 {
1880 	int		cpuid = cp->cpu_id;
1881 	extern int	restart_other_cpu(int);
1882 
1883 	ASSERT(MUTEX_HELD(&cpu_lock));
1884 	ASSERT(cpunodes[cpuid].nodeid != (pnode_t)0);
1885 
1886 	cp->cpu_flags &= ~CPU_POWEROFF;
1887 
1888 	/*
1889 	 * NOTE: restart_other_cpu pauses cpus during the
1890 	 *	 slave cpu start.  This helps to quiesce the
1891 	 *	 bus traffic a bit which makes the tick sync
1892 	 *	 routine in the prom more robust.
1893 	 */
1894 	DRMACH_PR("COLD START for cpu (%d)\n", cpuid);
1895 
1896 	restart_other_cpu(cpuid);
1897 
1898 	return (0);
1899 }
1900 
1901 static sbd_error_t *
1902 drmach_cpu_release(drmachid_t id)
1903 {
1904 	if (!DRMACH_IS_CPU_ID(id))
1905 		return (drerr_new(0, EOPL_INAPPROP, NULL));
1906 
1907 	return (NULL);
1908 }
1909 
1910 static sbd_error_t *
1911 drmach_cpu_status(drmachid_t id, drmach_status_t *stat)
1912 {
1913 	drmach_cpu_t *cp;
1914 	drmach_device_t *dp;
1915 
1916 	ASSERT(DRMACH_IS_CPU_ID(id));
1917 	cp = (drmach_cpu_t *)id;
1918 	dp = &cp->dev;
1919 
1920 	stat->assigned = dp->bp->assigned;
1921 	stat->powered = dp->bp->powered;
1922 	mutex_enter(&cpu_lock);
1923 	stat->configured = (cpu_get(cp->cpuid) != NULL);
1924 	mutex_exit(&cpu_lock);
1925 	stat->busy = dp->busy;
1926 	strncpy(stat->type, dp->type, sizeof (stat->type));
1927 	stat->info[0] = '\0';
1928 
1929 	return (NULL);
1930 }
1931 
1932 sbd_error_t *
1933 drmach_cpu_disconnect(drmachid_t id)
1934 {
1935 
1936 	if (!DRMACH_IS_CPU_ID(id))
1937 		return (drerr_new(0, EOPL_INAPPROP, NULL));
1938 
1939 	return (NULL);
1940 }
1941 
1942 sbd_error_t *
1943 drmach_cpu_get_id(drmachid_t id, processorid_t *cpuid)
1944 {
1945 	drmach_cpu_t *cpu;
1946 
1947 	if (!DRMACH_IS_CPU_ID(id))
1948 		return (drerr_new(0, EOPL_INAPPROP, NULL));
1949 	cpu = (drmach_cpu_t *)id;
1950 
1951 	/* get from cpu directly on OPL */
1952 	*cpuid = cpu->cpuid;
1953 	return (NULL);
1954 }
1955 
1956 sbd_error_t *
1957 drmach_cpu_get_impl(drmachid_t id, int *ip)
1958 {
1959 	drmach_device_t *cpu;
1960 	drmach_node_t	*np;
1961 	drmach_node_t	pp;
1962 	int		impl;
1963 	char		type[OBP_MAXPROPNAME];
1964 
1965 	if (!DRMACH_IS_CPU_ID(id))
1966 		return (drerr_new(0, EOPL_INAPPROP, NULL));
1967 
1968 	cpu = id;
1969 	np = cpu->node;
1970 
1971 	if (np->get_parent(np, &pp) != 0) {
1972 		return (DRMACH_INTERNAL_ERROR());
1973 	}
1974 
1975 	/* the parent should be core */
1976 
1977 	if (pp.n_getprop(&pp, "device_type", &type, sizeof (type)) != 0) {
1978 		return (drerr_new(0, EOPL_GETPROP, NULL));
1979 	}
1980 
1981 	if (strcmp(type, OPL_CORE_NODE) == 0) {
1982 		if (pp.n_getprop(&pp, "implementation#",
1983 			&impl, sizeof (impl)) != 0) {
1984 			return (drerr_new(0, EOPL_GETPROP, NULL));
1985 		}
1986 	} else {
1987 		return (DRMACH_INTERNAL_ERROR());
1988 	}
1989 
1990 	*ip = impl;
1991 
1992 	return (NULL);
1993 }
1994 
1995 sbd_error_t *
1996 drmach_get_dip(drmachid_t id, dev_info_t **dip)
1997 {
1998 	drmach_device_t	*dp;
1999 
2000 	if (!DRMACH_IS_DEVICE_ID(id))
2001 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2002 	dp = id;
2003 
2004 	*dip = dp->node->n_getdip(dp->node);
2005 	return (NULL);
2006 }
2007 
2008 sbd_error_t *
2009 drmach_io_is_attached(drmachid_t id, int *yes)
2010 {
2011 	drmach_device_t *dp;
2012 	dev_info_t	*dip;
2013 	int		state;
2014 
2015 	if (!DRMACH_IS_IO_ID(id))
2016 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2017 	dp = id;
2018 
2019 	dip = dp->node->n_getdip(dp->node);
2020 	if (dip == NULL) {
2021 		*yes = 0;
2022 		return (NULL);
2023 	}
2024 
2025 	state = ddi_get_devstate(dip);
2026 	*yes = ((i_ddi_node_state(dip) >= DS_ATTACHED) ||
2027 	    (state == DDI_DEVSTATE_UP));
2028 
2029 	return (NULL);
2030 }
2031 
2032 struct drmach_io_cb {
2033 	char	*name;	/* name of the node */
2034 	int	(*func)(dev_info_t *);
2035 	int	rv;
2036 	dev_info_t *dip;
2037 };
2038 
2039 #define	DRMACH_IO_POST_ATTACH	0
2040 #define	DRMACH_IO_PRE_RELEASE	1
2041 
2042 static int
2043 drmach_io_cb_check(dev_info_t *dip, void *arg)
2044 {
2045 	struct drmach_io_cb *p = (struct drmach_io_cb *)arg;
2046 	char name[OBP_MAXDRVNAME];
2047 	int len = OBP_MAXDRVNAME;
2048 
2049 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
2050 		DDI_PROP_DONTPASS, "name",
2051 	    (caddr_t)name, &len) != DDI_PROP_SUCCESS) {
2052 		return (DDI_WALK_PRUNECHILD);
2053 	}
2054 
2055 	if (strcmp(name, p->name) == 0) {
2056 		ndi_hold_devi(dip);
2057 		p->dip = dip;
2058 		return (DDI_WALK_TERMINATE);
2059 	}
2060 
2061 	return (DDI_WALK_CONTINUE);
2062 }
2063 
2064 
2065 static int
2066 drmach_console_ops(drmachid_t *id, int state)
2067 {
2068 	drmach_io_t *obj = (drmach_io_t *)id;
2069 	struct drmach_io_cb arg;
2070 	int (*msudetp)(dev_info_t *);
2071 	int (*msuattp)(dev_info_t *);
2072 	dev_info_t *dip, *pdip;
2073 	int circ;
2074 
2075 	/* 4 is pcicmu channel */
2076 	if (obj->channel != 4)
2077 		return (0);
2078 
2079 	arg.name = "serial";
2080 	arg.func = NULL;
2081 	if (state == DRMACH_IO_PRE_RELEASE) {
2082 		msudetp = (int (*)(dev_info_t *))
2083 		    modgetsymvalue("oplmsu_dr_detach", 0);
2084 		if (msudetp != NULL)
2085 			arg.func = msudetp;
2086 	} else if (state == DRMACH_IO_POST_ATTACH) {
2087 		msuattp = (int (*)(dev_info_t *))
2088 		    modgetsymvalue("oplmsu_dr_attach", 0);
2089 		if (msuattp != NULL)
2090 			arg.func = msuattp;
2091 	} else {
2092 		return (0);
2093 	}
2094 
2095 	if (arg.func == NULL) {
2096 		return (0);
2097 	}
2098 
2099 	arg.rv = 0;
2100 	arg.dip = NULL;
2101 
2102 	dip = obj->dev.node->n_getdip(obj->dev.node);
2103 	if (pdip = ddi_get_parent(dip)) {
2104 		ndi_hold_devi(pdip);
2105 		ndi_devi_enter(pdip, &circ);
2106 	} else {
2107 		/* this cannot happen unless something bad happens */
2108 		return (-1);
2109 	}
2110 
2111 	ddi_walk_devs(dip, drmach_io_cb_check, (void *)&arg);
2112 
2113 	ndi_devi_exit(pdip, circ);
2114 	ndi_rele_devi(pdip);
2115 
2116 	if (arg.dip) {
2117 		arg.rv = (*arg.func)(arg.dip);
2118 		ndi_rele_devi(arg.dip);
2119 	} else {
2120 		arg.rv = -1;
2121 	}
2122 
2123 	return (arg.rv);
2124 }
2125 
2126 sbd_error_t *
2127 drmach_io_pre_release(drmachid_t id)
2128 {
2129 	int rv;
2130 
2131 	if (!DRMACH_IS_IO_ID(id))
2132 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2133 
2134 	rv = drmach_console_ops(id, DRMACH_IO_PRE_RELEASE);
2135 
2136 	if (rv != 0)
2137 		cmn_err(CE_WARN, "IO callback failed in pre-release\n");
2138 
2139 	return (NULL);
2140 }
2141 
2142 static sbd_error_t *
2143 drmach_io_release(drmachid_t id)
2144 {
2145 	if (!DRMACH_IS_IO_ID(id))
2146 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2147 	return (NULL);
2148 }
2149 
2150 sbd_error_t *
2151 drmach_io_unrelease(drmachid_t id)
2152 {
2153 	if (!DRMACH_IS_IO_ID(id))
2154 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2155 	return (NULL);
2156 }
2157 
2158 /*ARGSUSED*/
2159 sbd_error_t *
2160 drmach_io_post_release(drmachid_t id)
2161 {
2162 	return (NULL);
2163 }
2164 
2165 /*ARGSUSED*/
2166 sbd_error_t *
2167 drmach_io_post_attach(drmachid_t id)
2168 {
2169 	int rv;
2170 
2171 	if (!DRMACH_IS_IO_ID(id))
2172 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2173 
2174 	rv = drmach_console_ops(id, DRMACH_IO_POST_ATTACH);
2175 
2176 	if (rv != 0)
2177 		cmn_err(CE_WARN, "IO callback failed in post-attach\n");
2178 
2179 	return (0);
2180 }
2181 
2182 static sbd_error_t *
2183 drmach_io_status(drmachid_t id, drmach_status_t *stat)
2184 {
2185 	drmach_device_t *dp;
2186 	sbd_error_t	*err;
2187 	int		 configured;
2188 
2189 	ASSERT(DRMACH_IS_IO_ID(id));
2190 	dp = id;
2191 
2192 	err = drmach_io_is_attached(id, &configured);
2193 	if (err)
2194 		return (err);
2195 
2196 	stat->assigned = dp->bp->assigned;
2197 	stat->powered = dp->bp->powered;
2198 	stat->configured = (configured != 0);
2199 	stat->busy = dp->busy;
2200 	strncpy(stat->type, dp->type, sizeof (stat->type));
2201 	stat->info[0] = '\0';
2202 
2203 	return (NULL);
2204 }
2205 
2206 static sbd_error_t *
2207 drmach_mem_new(drmach_device_t *proto, drmachid_t *idp)
2208 {
2209 	static void drmach_mem_dispose(drmachid_t);
2210 	static sbd_error_t *drmach_mem_release(drmachid_t);
2211 	static sbd_error_t *drmach_mem_status(drmachid_t, drmach_status_t *);
2212 	dev_info_t *dip;
2213 	int rv;
2214 
2215 	drmach_mem_t	*mp;
2216 
2217 	rv = 0;
2218 
2219 	if ((proto->node->n_getproplen(proto->node, "mc-addr", &rv) < 0) ||
2220 		(rv <= 0)) {
2221 		*idp = (drmachid_t)0;
2222 		return (NULL);
2223 	}
2224 
2225 	mp = kmem_zalloc(sizeof (drmach_mem_t), KM_SLEEP);
2226 	proto->unum = 0;
2227 
2228 	bcopy(proto, &mp->dev, sizeof (mp->dev));
2229 	mp->dev.node = drmach_node_dup(proto->node);
2230 	mp->dev.cm.isa = (void *)drmach_mem_new;
2231 	mp->dev.cm.dispose = drmach_mem_dispose;
2232 	mp->dev.cm.release = drmach_mem_release;
2233 	mp->dev.cm.status = drmach_mem_status;
2234 
2235 	snprintf(mp->dev.cm.name,
2236 		sizeof (mp->dev.cm.name), "%s", mp->dev.type);
2237 
2238 	dip = mp->dev.node->n_getdip(mp->dev.node);
2239 	if (drmach_setup_mc_info(dip, mp) != 0) {
2240 		return (drerr_new(1, EOPL_MC_SETUP, NULL));
2241 	}
2242 
2243 	/* make sure we do not create memoryless nodes */
2244 	if (mp->nbytes == 0) {
2245 		*idp = (drmachid_t)NULL;
2246 		kmem_free(mp, sizeof (drmach_mem_t));
2247 	} else
2248 		*idp = (drmachid_t)mp;
2249 
2250 	return (NULL);
2251 }
2252 
2253 static void
2254 drmach_mem_dispose(drmachid_t id)
2255 {
2256 	drmach_mem_t *mp;
2257 
2258 	ASSERT(DRMACH_IS_MEM_ID(id));
2259 
2260 
2261 	mp = id;
2262 
2263 	if (mp->dev.node)
2264 		drmach_node_dispose(mp->dev.node);
2265 
2266 	if (mp->memlist) {
2267 		memlist_delete(mp->memlist);
2268 		mp->memlist = NULL;
2269 	}
2270 
2271 	kmem_free(mp, sizeof (*mp));
2272 }
2273 
2274 sbd_error_t *
2275 drmach_mem_add_span(drmachid_t id, uint64_t basepa, uint64_t size)
2276 {
2277 	pfn_t		basepfn = (pfn_t)(basepa >> PAGESHIFT);
2278 	pgcnt_t		npages = (pgcnt_t)(size >> PAGESHIFT);
2279 	int		rv;
2280 
2281 	ASSERT(size != 0);
2282 
2283 	if (!DRMACH_IS_MEM_ID(id))
2284 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2285 
2286 	rv = kcage_range_add(basepfn, npages, KCAGE_DOWN);
2287 	if (rv == ENOMEM) {
2288 		cmn_err(CE_WARN, "%ld megabytes not available to kernel cage",
2289 			(size == 0 ? 0 : size / MBYTE));
2290 	} else if (rv != 0) {
2291 		/* catch this in debug kernels */
2292 		ASSERT(0);
2293 
2294 		cmn_err(CE_WARN, "unexpected kcage_range_add"
2295 			" return value %d", rv);
2296 	}
2297 
2298 	if (rv) {
2299 		return (DRMACH_INTERNAL_ERROR());
2300 	}
2301 	else
2302 		return (NULL);
2303 }
2304 
2305 sbd_error_t *
2306 drmach_mem_del_span(drmachid_t id, uint64_t basepa, uint64_t size)
2307 {
2308 	pfn_t		basepfn = (pfn_t)(basepa >> PAGESHIFT);
2309 	pgcnt_t		npages = (pgcnt_t)(size >> PAGESHIFT);
2310 	int		rv;
2311 
2312 	if (!DRMACH_IS_MEM_ID(id))
2313 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2314 
2315 	if (size > 0) {
2316 		rv = kcage_range_delete_post_mem_del(basepfn, npages);
2317 		if (rv != 0) {
2318 			cmn_err(CE_WARN,
2319 			    "unexpected kcage_range_delete_post_mem_del"
2320 			    " return value %d", rv);
2321 			return (DRMACH_INTERNAL_ERROR());
2322 		}
2323 	}
2324 
2325 	return (NULL);
2326 }
2327 
2328 sbd_error_t *
2329 drmach_mem_disable(drmachid_t id)
2330 {
2331 	if (!DRMACH_IS_MEM_ID(id))
2332 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2333 	else {
2334 		drmach_flush_all();
2335 		return (NULL);
2336 	}
2337 }
2338 
2339 sbd_error_t *
2340 drmach_mem_enable(drmachid_t id)
2341 {
2342 	if (!DRMACH_IS_MEM_ID(id))
2343 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2344 	else
2345 		return (NULL);
2346 }
2347 
2348 sbd_error_t *
2349 drmach_mem_get_info(drmachid_t id, drmach_mem_info_t *mem)
2350 {
2351 	drmach_mem_t *mp;
2352 
2353 	if (!DRMACH_IS_MEM_ID(id))
2354 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2355 
2356 	mp = (drmach_mem_t *)id;
2357 
2358 	/*
2359 	 * This is only used by dr to round up/down the memory
2360 	 * for copying. Our unit of memory isolation is 64 MB.
2361 	 */
2362 
2363 	mem->mi_alignment_mask = (64 * 1024 * 1024 - 1);
2364 	mem->mi_basepa = mp->base_pa;
2365 	mem->mi_size = mp->nbytes;
2366 	mem->mi_slice_size = mp->slice_size;
2367 
2368 	return (NULL);
2369 }
2370 
2371 sbd_error_t *
2372 drmach_mem_get_base_physaddr(drmachid_t id, uint64_t *pa)
2373 {
2374 	drmach_mem_t *mp;
2375 
2376 	if (!DRMACH_IS_MEM_ID(id))
2377 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2378 
2379 	mp = (drmach_mem_t *)id;
2380 
2381 	*pa = mp->base_pa;
2382 	return (NULL);
2383 }
2384 
2385 sbd_error_t *
2386 drmach_mem_get_memlist(drmachid_t id, struct memlist **ml)
2387 {
2388 	drmach_mem_t	*mem;
2389 #ifdef	DEBUG
2390 	int		rv;
2391 #endif
2392 	struct memlist	*mlist;
2393 
2394 	if (!DRMACH_IS_MEM_ID(id))
2395 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2396 
2397 	mem = (drmach_mem_t *)id;
2398 	mlist = memlist_dup(mem->memlist);
2399 
2400 #ifdef DEBUG
2401 	/*
2402 	 * Make sure the incoming memlist doesn't already
2403 	 * intersect with what's present in the system (phys_install).
2404 	 */
2405 	memlist_read_lock();
2406 	rv = memlist_intersect(phys_install, mlist);
2407 	memlist_read_unlock();
2408 	if (rv) {
2409 		DRMACH_PR("Derived memlist intersects"
2410 			" with phys_install\n");
2411 		memlist_dump(mlist);
2412 
2413 		DRMACH_PR("phys_install memlist:\n");
2414 		memlist_dump(phys_install);
2415 
2416 		memlist_delete(mlist);
2417 		return (DRMACH_INTERNAL_ERROR());
2418 	}
2419 
2420 	DRMACH_PR("Derived memlist:");
2421 	memlist_dump(mlist);
2422 #endif
2423 	*ml = mlist;
2424 
2425 	return (NULL);
2426 }
2427 
2428 sbd_error_t *
2429 drmach_mem_get_slice_size(drmachid_t id, uint64_t *bytes)
2430 {
2431 	drmach_mem_t	*mem;
2432 
2433 	if (!DRMACH_IS_MEM_ID(id))
2434 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2435 
2436 	mem = (drmach_mem_t *)id;
2437 
2438 	*bytes = mem->slice_size;
2439 
2440 	return (NULL);
2441 }
2442 
2443 
2444 /* ARGSUSED */
2445 processorid_t
2446 drmach_mem_cpu_affinity(drmachid_t id)
2447 {
2448 	return (CPU_CURRENT);
2449 }
2450 
2451 static sbd_error_t *
2452 drmach_mem_release(drmachid_t id)
2453 {
2454 	if (!DRMACH_IS_MEM_ID(id))
2455 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2456 	return (NULL);
2457 }
2458 
2459 static sbd_error_t *
2460 drmach_mem_status(drmachid_t id, drmach_status_t *stat)
2461 {
2462 	drmach_mem_t *dp;
2463 	uint64_t	 pa, slice_size;
2464 	struct memlist	*ml;
2465 
2466 	ASSERT(DRMACH_IS_MEM_ID(id));
2467 	dp = id;
2468 
2469 	/* get starting physical address of target memory */
2470 	pa = dp->base_pa;
2471 
2472 	/* round down to slice boundary */
2473 	slice_size = dp->slice_size;
2474 	pa &= ~(slice_size - 1);
2475 
2476 	/* stop at first span that is in slice */
2477 	memlist_read_lock();
2478 	for (ml = phys_install; ml; ml = ml->next)
2479 		if (ml->address >= pa && ml->address < pa + slice_size)
2480 			break;
2481 	memlist_read_unlock();
2482 
2483 	stat->assigned = dp->dev.bp->assigned;
2484 	stat->powered = dp->dev.bp->powered;
2485 	stat->configured = (ml != NULL);
2486 	stat->busy = dp->dev.busy;
2487 	strncpy(stat->type, dp->dev.type, sizeof (stat->type));
2488 	stat->info[0] = '\0';
2489 
2490 	return (NULL);
2491 }
2492 
2493 
2494 sbd_error_t *
2495 drmach_board_deprobe(drmachid_t id)
2496 {
2497 	drmach_board_t	*bp;
2498 
2499 	if (!DRMACH_IS_BOARD_ID(id))
2500 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2501 
2502 	bp = id;
2503 
2504 	cmn_err(CE_CONT, "DR: detach board %d\n", bp->bnum);
2505 
2506 	if (bp->tree) {
2507 		drmach_node_dispose(bp->tree);
2508 		bp->tree = NULL;
2509 	}
2510 	if (bp->devices) {
2511 		drmach_array_dispose(bp->devices, drmach_device_dispose);
2512 		bp->devices = NULL;
2513 	}
2514 
2515 	bp->boot_board = 0;
2516 
2517 	return (NULL);
2518 }
2519 
2520 /*ARGSUSED*/
2521 static sbd_error_t *
2522 drmach_pt_ikprobe(drmachid_t id, drmach_opts_t *opts)
2523 {
2524 	drmach_board_t		*bp = (drmach_board_t *)id;
2525 	sbd_error_t		*err = NULL;
2526 	int	rv;
2527 
2528 	if (!DRMACH_IS_BOARD_ID(id))
2529 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2530 
2531 	DRMACH_PR("calling opl_probe_board for bnum=%d\n", bp->bnum);
2532 	rv = opl_probe_sb(bp->bnum);
2533 	if (rv != 0) {
2534 		err = drerr_new(1, EOPL_PROBE, bp->cm.name);
2535 		return (err);
2536 	}
2537 	return (err);
2538 }
2539 
2540 /*ARGSUSED*/
2541 static sbd_error_t *
2542 drmach_pt_ikdeprobe(drmachid_t id, drmach_opts_t *opts)
2543 {
2544 	drmach_board_t	*bp;
2545 	sbd_error_t	*err = NULL;
2546 	int	rv;
2547 
2548 	if (!DRMACH_IS_BOARD_ID(id))
2549 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2550 	bp = (drmach_board_t *)id;
2551 
2552 	cmn_err(CE_CONT, "DR: in-kernel unprobe board %d\n", bp->bnum);
2553 
2554 	rv = opl_unprobe_sb(bp->bnum);
2555 	if (rv != 0) {
2556 		err = drerr_new(1, EOPL_DEPROBE, bp->cm.name);
2557 	}
2558 
2559 	return (err);
2560 }
2561 
2562 
2563 /*ARGSUSED*/
2564 sbd_error_t *
2565 drmach_pt_readmem(drmachid_t id, drmach_opts_t *opts)
2566 {
2567 	struct memlist	*ml;
2568 	uint64_t	src_pa;
2569 	uint64_t	dst_pa;
2570 	uint64_t	dst;
2571 
2572 	dst_pa = va_to_pa(&dst);
2573 
2574 	memlist_read_lock();
2575 	for (ml = phys_install; ml; ml = ml->next) {
2576 		uint64_t	nbytes;
2577 
2578 		src_pa = ml->address;
2579 		nbytes = ml->size;
2580 
2581 		while (nbytes != 0ull) {
2582 
2583 			/* copy 32 bytes at arc_pa to dst_pa */
2584 			bcopy32_il(src_pa, dst_pa);
2585 
2586 			/* increment by 32 bytes */
2587 			src_pa += (4 * sizeof (uint64_t));
2588 
2589 			/* decrement by 32 bytes */
2590 			nbytes -= (4 * sizeof (uint64_t));
2591 		}
2592 	}
2593 	memlist_read_unlock();
2594 
2595 	return (NULL);
2596 }
2597 
2598 static struct {
2599 	const char	*name;
2600 	sbd_error_t	*(*handler)(drmachid_t id, drmach_opts_t *opts);
2601 } drmach_pt_arr[] = {
2602 	{ "readmem",		drmach_pt_readmem		},
2603 	{ "ikprobe",	drmach_pt_ikprobe	},
2604 	{ "ikdeprobe",	drmach_pt_ikdeprobe	},
2605 
2606 	/* the following line must always be last */
2607 	{ NULL,			NULL				}
2608 };
2609 
2610 /*ARGSUSED*/
2611 sbd_error_t *
2612 drmach_passthru(drmachid_t id, drmach_opts_t *opts)
2613 {
2614 	int		i;
2615 	sbd_error_t	*err;
2616 
2617 	i = 0;
2618 	while (drmach_pt_arr[i].name != NULL) {
2619 		int len = strlen(drmach_pt_arr[i].name);
2620 
2621 		if (strncmp(drmach_pt_arr[i].name, opts->copts, len) == 0)
2622 			break;
2623 
2624 		i += 1;
2625 	}
2626 
2627 	if (drmach_pt_arr[i].name == NULL)
2628 		err = drerr_new(0, EOPL_UNKPTCMD, opts->copts);
2629 	else
2630 		err = (*drmach_pt_arr[i].handler)(id, opts);
2631 
2632 	return (err);
2633 }
2634 
2635 sbd_error_t *
2636 drmach_release(drmachid_t id)
2637 {
2638 	drmach_common_t *cp;
2639 
2640 	if (!DRMACH_IS_DEVICE_ID(id))
2641 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2642 	cp = id;
2643 
2644 	return (cp->release(id));
2645 }
2646 
2647 sbd_error_t *
2648 drmach_status(drmachid_t id, drmach_status_t *stat)
2649 {
2650 	drmach_common_t *cp;
2651 	sbd_error_t	*err;
2652 
2653 	rw_enter(&drmach_boards_rwlock, RW_READER);
2654 
2655 	if (!DRMACH_IS_ID(id)) {
2656 		rw_exit(&drmach_boards_rwlock);
2657 		return (drerr_new(0, EOPL_NOTID, NULL));
2658 	}
2659 	cp = (drmach_common_t *)id;
2660 	err = cp->status(id, stat);
2661 
2662 	rw_exit(&drmach_boards_rwlock);
2663 
2664 	return (err);
2665 }
2666 
2667 static sbd_error_t *
2668 drmach_i_status(drmachid_t id, drmach_status_t *stat)
2669 {
2670 	drmach_common_t *cp;
2671 
2672 	if (!DRMACH_IS_ID(id))
2673 		return (drerr_new(0, EOPL_NOTID, NULL));
2674 	cp = id;
2675 
2676 	return (cp->status(id, stat));
2677 }
2678 
2679 /*ARGSUSED*/
2680 sbd_error_t *
2681 drmach_unconfigure(drmachid_t id, int flags)
2682 {
2683 	drmach_device_t *dp;
2684 	dev_info_t	*rdip, *fdip = NULL;
2685 	char name[OBP_MAXDRVNAME];
2686 	int rv;
2687 
2688 	if (DRMACH_IS_CPU_ID(id))
2689 		return (NULL);
2690 
2691 	if (!DRMACH_IS_DEVICE_ID(id))
2692 		return (drerr_new(0, EOPL_INAPPROP, NULL));
2693 
2694 	dp = id;
2695 
2696 	rdip = dp->node->n_getdip(dp->node);
2697 
2698 	ASSERT(rdip);
2699 
2700 	rv = dp->node->n_getprop(dp->node, "name", name, OBP_MAXDRVNAME);
2701 
2702 	if (rv)
2703 		return (NULL);
2704 
2705 	/*
2706 	 * Note: FORCE flag is no longer necessary under devfs
2707 	 */
2708 
2709 	ASSERT(e_ddi_branch_held(rdip));
2710 	if (e_ddi_branch_unconfigure(rdip, &fdip, 0)) {
2711 		sbd_error_t	*err;
2712 		char		*path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2713 
2714 		/*
2715 		 * If non-NULL, fdip is returned held and must be released.
2716 		 */
2717 		if (fdip != NULL) {
2718 			(void) ddi_pathname(fdip, path);
2719 			ndi_rele_devi(fdip);
2720 		} else {
2721 			(void) ddi_pathname(rdip, path);
2722 		}
2723 
2724 		err = drerr_new(1, EOPL_DRVFAIL, path);
2725 
2726 		kmem_free(path, MAXPATHLEN);
2727 
2728 		return (err);
2729 	}
2730 
2731 	return (NULL);
2732 }
2733 
2734 
2735 int
2736 drmach_cpu_poweron(struct cpu *cp)
2737 {
2738 	int bnum, cpuid, onb_core_num, strand_id;
2739 	drmach_board_t *bp;
2740 
2741 	DRMACH_PR("drmach_cpu_poweron: starting cpuid %d\n", cp->cpu_id);
2742 
2743 	cpuid = cp->cpu_id;
2744 	bnum = LSB_ID(cpuid);
2745 	onb_core_num = ON_BOARD_CORE_NUM(cpuid);
2746 	strand_id = STRAND_ID(cpuid);
2747 	bp = drmach_get_board_by_bnum(bnum);
2748 
2749 	ASSERT(bp);
2750 	if (bp->cores[onb_core_num].core_hotadded == 0) {
2751 		if (drmach_add_remove_cpu(bnum, onb_core_num,
2752 			HOTADD_CPU) != 0) {
2753 			cmn_err(CE_WARN, "Failed to add CMP %d on board %d\n",
2754 				onb_core_num, bnum);
2755 			return (EIO);
2756 		}
2757 	}
2758 
2759 	ASSERT(MUTEX_HELD(&cpu_lock));
2760 
2761 	if (drmach_cpu_start(cp) != 0) {
2762 		if (bp->cores[onb_core_num].core_started == 0) {
2763 			/*
2764 			 * we must undo the hotadd or no one will do that
2765 			 * If this fails, we will do this again in
2766 			 * drmach_board_disconnect.
2767 			 */
2768 			if (drmach_add_remove_cpu(bnum, onb_core_num,
2769 				HOTREMOVE_CPU) != 0) {
2770 				cmn_err(CE_WARN, "Failed to remove CMP %d "
2771 					"on board %d\n",
2772 					onb_core_num, bnum);
2773 			}
2774 		}
2775 		return (EBUSY);
2776 	} else {
2777 		bp->cores[onb_core_num].core_started |= (1 << strand_id);
2778 		return (0);
2779 	}
2780 }
2781 
2782 int
2783 drmach_cpu_poweroff(struct cpu *cp)
2784 {
2785 	int 		rv = 0;
2786 	processorid_t	cpuid = cp->cpu_id;
2787 
2788 	DRMACH_PR("drmach_cpu_poweroff: stopping cpuid %d\n", cp->cpu_id);
2789 
2790 	ASSERT(MUTEX_HELD(&cpu_lock));
2791 
2792 	/*
2793 	 * Capture all CPUs (except for detaching proc) to prevent
2794 	 * crosscalls to the detaching proc until it has cleared its
2795 	 * bit in cpu_ready_set.
2796 	 *
2797 	 * The CPU's remain paused and the prom_mutex is known to be free.
2798 	 * This prevents the x-trap victim from blocking when doing prom
2799 	 * IEEE-1275 calls at a high PIL level.
2800 	 */
2801 
2802 	promsafe_pause_cpus();
2803 
2804 	/*
2805 	 * Quiesce interrupts on the target CPU. We do this by setting
2806 	 * the CPU 'not ready'- (i.e. removing the CPU from cpu_ready_set) to
2807 	 * prevent it from receiving cross calls and cross traps.
2808 	 * This prevents the processor from receiving any new soft interrupts.
2809 	 */
2810 	mp_cpu_quiesce(cp);
2811 
2812 	rv = prom_stopcpu_bycpuid(cpuid);
2813 	if (rv == 0)
2814 		cp->cpu_flags = CPU_OFFLINE | CPU_QUIESCED | CPU_POWEROFF;
2815 
2816 	start_cpus();
2817 
2818 	if (rv == 0) {
2819 		int bnum, onb_core_num, strand_id;
2820 		drmach_board_t *bp;
2821 
2822 		CPU_SIGNATURE(OS_SIG, SIGST_DETACHED, SIGSUBST_NULL, cpuid);
2823 
2824 		bnum = LSB_ID(cpuid);
2825 		onb_core_num = ON_BOARD_CORE_NUM(cpuid);
2826 		strand_id = STRAND_ID(cpuid);
2827 		bp = drmach_get_board_by_bnum(bnum);
2828 		ASSERT(bp);
2829 
2830 		bp->cores[onb_core_num].core_started &= ~(1 << strand_id);
2831 		if (bp->cores[onb_core_num].core_started == 0) {
2832 			if (drmach_add_remove_cpu(bnum, onb_core_num,
2833 				HOTREMOVE_CPU) != 0) {
2834 				cmn_err(CE_WARN,
2835 					"Failed to remove CMP %d LSB %d\n",
2836 					onb_core_num, bnum);
2837 				return (EIO);
2838 			}
2839 		}
2840 	}
2841 
2842 	return (rv);
2843 }
2844 
2845 /*ARGSUSED*/
2846 int
2847 drmach_verify_sr(dev_info_t *dip, int sflag)
2848 {
2849 	return (0);
2850 }
2851 
2852 void
2853 drmach_suspend_last(void)
2854 {
2855 }
2856 
2857 void
2858 drmach_resume_first(void)
2859 {
2860 }
2861 
2862 /*
2863  * Log a DR sysevent.
2864  * Return value: 0 success, non-zero failure.
2865  */
2866 int
2867 drmach_log_sysevent(int board, char *hint, int flag, int verbose)
2868 {
2869 	sysevent_t			*ev;
2870 	sysevent_id_t			eid;
2871 	int				rv, km_flag;
2872 	sysevent_value_t		evnt_val;
2873 	sysevent_attr_list_t		*evnt_attr_list = NULL;
2874 	char				attach_pnt[MAXNAMELEN];
2875 
2876 	km_flag = (flag == SE_SLEEP) ? KM_SLEEP : KM_NOSLEEP;
2877 	attach_pnt[0] = '\0';
2878 	if (drmach_board_name(board, attach_pnt, MAXNAMELEN)) {
2879 		rv = -1;
2880 		goto logexit;
2881 	}
2882 	if (verbose) {
2883 		DRMACH_PR("drmach_log_sysevent: %s %s, flag: %d, verbose: %d\n",
2884 			attach_pnt, hint, flag, verbose);
2885 	}
2886 
2887 	if ((ev = sysevent_alloc(EC_DR, ESC_DR_AP_STATE_CHANGE,
2888 		SUNW_KERN_PUB"dr", km_flag)) == NULL) {
2889 		rv = -2;
2890 		goto logexit;
2891 	}
2892 	evnt_val.value_type = SE_DATA_TYPE_STRING;
2893 	evnt_val.value.sv_string = attach_pnt;
2894 	if ((rv = sysevent_add_attr(&evnt_attr_list, DR_AP_ID,
2895 		&evnt_val, km_flag)) != 0)
2896 		goto logexit;
2897 
2898 	evnt_val.value_type = SE_DATA_TYPE_STRING;
2899 	evnt_val.value.sv_string = hint;
2900 	if ((rv = sysevent_add_attr(&evnt_attr_list, DR_HINT,
2901 		&evnt_val, km_flag)) != 0) {
2902 		sysevent_free_attr(evnt_attr_list);
2903 		goto logexit;
2904 	}
2905 
2906 	(void) sysevent_attach_attributes(ev, evnt_attr_list);
2907 
2908 	/*
2909 	 * Log the event but do not sleep waiting for its
2910 	 * delivery. This provides insulation from syseventd.
2911 	 */
2912 	rv = log_sysevent(ev, SE_NOSLEEP, &eid);
2913 
2914 logexit:
2915 	if (ev)
2916 		sysevent_free(ev);
2917 	if ((rv != 0) && verbose)
2918 		cmn_err(CE_WARN,
2919 			"drmach_log_sysevent failed (rv %d) for %s  %s\n",
2920 			rv, attach_pnt, hint);
2921 
2922 	return (rv);
2923 }
2924 
2925 #define	OPL_DR_STATUS_PROP "dr-status"
2926 
2927 static int
2928 opl_check_dr_status()
2929 {
2930 	pnode_t	node;
2931 	int	rtn, len;
2932 	char	*str;
2933 
2934 	node = prom_rootnode();
2935 	if (node == OBP_BADNODE) {
2936 		return (1);
2937 	}
2938 
2939 	len = prom_getproplen(node, OPL_DR_STATUS_PROP);
2940 	if (len == -1) {
2941 		/*
2942 		 * dr-status doesn't exist when DR is activated and
2943 		 * any warning messages aren't needed.
2944 		 */
2945 		return (1);
2946 	}
2947 
2948 	str = (char *)kmem_zalloc(len+1, KM_SLEEP);
2949 	rtn = prom_getprop(node, OPL_DR_STATUS_PROP, str);
2950 	kmem_free(str, len + 1);
2951 	if (rtn == -1) {
2952 		return (1);
2953 	} else {
2954 		return (0);
2955 	}
2956 }
2957 
2958 /* we are allocating memlist from TLB locked pages to avoid tlbmisses */
2959 
2960 static struct memlist *
2961 drmach_memlist_add_span(drmach_copy_rename_program_t *p,
2962 	struct memlist *mlist, uint64_t base, uint64_t len)
2963 {
2964 	struct memlist	*ml, *tl, *nl;
2965 
2966 	if (len == 0ull)
2967 		return (NULL);
2968 
2969 	if (mlist == NULL) {
2970 		mlist = p->free_mlist;
2971 		if (mlist == NULL)
2972 			return (NULL);
2973 		p->free_mlist = mlist->next;
2974 		mlist->address = base;
2975 		mlist->size = len;
2976 		mlist->next = mlist->prev = NULL;
2977 
2978 		return (mlist);
2979 	}
2980 
2981 	for (tl = ml = mlist; ml; tl = ml, ml = ml->next) {
2982 		if (base < ml->address) {
2983 			if ((base + len) < ml->address) {
2984 				nl = p->free_mlist;
2985 				if (nl == NULL)
2986 					return (NULL);
2987 				p->free_mlist = nl->next;
2988 				nl->address = base;
2989 				nl->size = len;
2990 				nl->next = ml;
2991 				if ((nl->prev = ml->prev) != NULL)
2992 					nl->prev->next = nl;
2993 				ml->prev = nl;
2994 				if (mlist == ml)
2995 					mlist = nl;
2996 			} else {
2997 				ml->size = MAX((base + len),
2998 					(ml->address + ml->size)) -
2999 					base;
3000 				ml->address = base;
3001 			}
3002 			break;
3003 
3004 		} else if (base <= (ml->address + ml->size)) {
3005 			ml->size = MAX((base + len),
3006 				(ml->address + ml->size)) -
3007 				MIN(ml->address, base);
3008 			ml->address = MIN(ml->address, base);
3009 			break;
3010 		}
3011 	}
3012 	if (ml == NULL) {
3013 		nl = p->free_mlist;
3014 		if (nl == NULL)
3015 			return (NULL);
3016 		p->free_mlist = nl->next;
3017 		nl->address = base;
3018 		nl->size = len;
3019 		nl->next = NULL;
3020 		nl->prev = tl;
3021 		tl->next = nl;
3022 	}
3023 
3024 	return (mlist);
3025 }
3026 
3027 /*
3028  * The routine performs the necessary memory COPY and MC adr SWITCH.
3029  * Both operations MUST be at the same "level" so that the stack is
3030  * maintained correctly between the copy and switch.  The switch
3031  * portion implements a caching mechanism to guarantee the code text
3032  * is cached prior to execution.  This is to guard against possible
3033  * memory access while the MC adr's are being modified.
3034  *
3035  * IMPORTANT: The _drmach_copy_rename_end() function must immediately
3036  * follow drmach_copy_rename_prog__relocatable() so that the correct
3037  * "length" of the drmach_copy_rename_prog__relocatable can be
3038  * calculated.  This routine MUST be a LEAF function, i.e. it can
3039  * make NO function calls, primarily for two reasons:
3040  *
3041  *	1. We must keep the stack consistent across the "switch".
3042  *	2. Function calls are compiled to relative offsets, and
3043  *	   we execute this function we'll be executing it from
3044  *	   a copied version in a different area of memory, thus
3045  *	   the relative offsets will be bogus.
3046  *
3047  * Moreover, it must have the "__relocatable" suffix to inform DTrace
3048  * providers (and anything else, for that matter) that this
3049  * function's text is manually relocated elsewhere before it is
3050  * executed.  That is, it cannot be safely instrumented with any
3051  * methodology that is PC-relative.
3052  */
3053 
3054 /*
3055  * We multiply this to system_clock_frequency so we
3056  * are setting a delay of fmem_timeout second for
3057  * the rename command.
3058  *
3059  * FMEM command itself should complete within 15 sec.
3060  * We add 2 more sec to be conservative.
3061  *
3062  * Note that there is also a SCF BUSY bit checking
3063  * in drmach_asm.s right before FMEM command is
3064  * issued.  XSCF sets the SCF BUSY bit when the
3065  * other domain on the same PSB reboots and it
3066  * will not be able to service the FMEM command
3067  * within 15 sec.   After setting the SCF BUSY
3068  * bit, XSCF will wait a while before servicing
3069  * other reboot command so there is no race
3070  * condition.
3071  */
3072 
3073 static int	fmem_timeout = 17;
3074 
3075 /*
3076  *	The empirical data on some OPL system shows that
3077  *	we can copy 250 MB per second.  We set it to
3078  * 	80 MB to be conservative.  In normal case,
3079  *	this timeout does not affect anything.
3080  */
3081 
3082 static int	min_copy_size_per_sec = 80 * 1024 * 1024;
3083 
3084 /*
3085  *	This is the timeout value for the xcall synchronization
3086  *	to get all the CPU ready to do the parallel copying.
3087  *	Even on a fully loaded system, 10 sec. should be long
3088  *	enough.
3089  */
3090 
3091 static int	cpu_xcall_delay = 10;
3092 int drmach_disable_mcopy = 0;
3093 
3094 /*
3095  * The following delay loop executes sleep instruction to yield the
3096  * CPU to other strands.  If this is not done, some strand will tie
3097  * up the CPU in busy loops while the other strand cannot do useful
3098  * work.  The copy procedure will take a much longer time without this.
3099  */
3100 #define	DR_DELAY_IL(ms, freq)					\
3101 	{							\
3102 		uint64_t start;					\
3103 		uint64_t nstick;				\
3104 		volatile uint64_t now;				\
3105 		nstick = ((uint64_t)ms * freq)/1000;		\
3106 		start = drmach_get_stick_il();			\
3107 		now = start;					\
3108 		while ((now - start) <= nstick) {		\
3109 			drmach_sleep_il();			\
3110 			now = drmach_get_stick_il();		\
3111 		}						\
3112 	}
3113 
3114 static int
3115 drmach_copy_rename_prog__relocatable(drmach_copy_rename_program_t *prog,
3116 	int cpuid)
3117 {
3118 	struct memlist		*ml;
3119 	register int		rtn;
3120 	int			i;
3121 	register uint64_t	curr, limit;
3122 	extern uint64_t		drmach_get_stick_il();
3123 	extern void		membar_sync_il();
3124 	extern void		flush_instr_mem_il(void*);
3125 	extern void		flush_windows_il(void);
3126 	uint64_t		copy_start;
3127 
3128 	/*
3129 	 * flush_windows is moved here to make sure all
3130 	 * registers used in the callers are flushed to
3131 	 * memory before the copy.
3132 	 *
3133 	 * If flush_windows() is called too early in the
3134 	 * calling function, the compiler might put some
3135 	 * data in the local registers after flush_windows().
3136 	 * After FMA, if there is any fill trap, the registers
3137 	 * will contain stale data.
3138 	 */
3139 
3140 	flush_windows_il();
3141 
3142 	prog->critical->stat[cpuid] = FMEM_LOOP_COPY_READY;
3143 	membar_sync_il();
3144 
3145 	if (prog->data->cpuid == cpuid) {
3146 		limit = drmach_get_stick_il();
3147 		limit += cpu_xcall_delay * system_clock_freq;
3148 		for (i = 0; i < NCPU; i++) {
3149 			if (CPU_IN_SET(prog->data->cpu_slave_set, i)) {
3150 			/* wait for all CPU's to be ready */
3151 			    for (;;) {
3152 				if (prog->critical->stat[i] ==
3153 					FMEM_LOOP_COPY_READY) {
3154 					break;
3155 				}
3156 				DR_DELAY_IL(1, prog->data->stick_freq);
3157 			    }
3158 			    curr = drmach_get_stick_il();
3159 			    if (curr > limit) {
3160 				prog->data->fmem_status.error =
3161 					EOPL_FMEM_XC_TIMEOUT;
3162 				return (EOPL_FMEM_XC_TIMEOUT);
3163 			    }
3164 			}
3165 		}
3166 		prog->data->fmem_status.stat = FMEM_LOOP_COPY_READY;
3167 		membar_sync_il();
3168 		copy_start = drmach_get_stick_il();
3169 	} else {
3170 		for (;;) {
3171 			if (prog->data->fmem_status.stat ==
3172 				FMEM_LOOP_COPY_READY) {
3173 				break;
3174 			}
3175 			if (prog->data->fmem_status.error) {
3176 				prog->data->error[cpuid] =
3177 					EOPL_FMEM_TERMINATE;
3178 				return (EOPL_FMEM_TERMINATE);
3179 			}
3180 			DR_DELAY_IL(1, prog->data->stick_freq);
3181 		}
3182 	}
3183 
3184 	/*
3185 	 * DO COPY.
3186 	 */
3187 	if (CPU_IN_SET(prog->data->cpu_copy_set, cpuid)) {
3188 	    for (ml = prog->data->cpu_ml[cpuid]; ml; ml = ml->next) {
3189 		uint64_t	s_pa, t_pa;
3190 		uint64_t	nbytes;
3191 
3192 		s_pa = prog->data->s_copybasepa + ml->address;
3193 		t_pa = prog->data->t_copybasepa + ml->address;
3194 		nbytes = ml->size;
3195 
3196 		while (nbytes != 0ull) {
3197 			/* If the master has detected error, we just bail out */
3198 			if (prog->data->fmem_status.error != ESBD_NOERROR) {
3199 				prog->data->error[cpuid] =
3200 					EOPL_FMEM_TERMINATE;
3201 				return (EOPL_FMEM_TERMINATE);
3202 			}
3203 			/*
3204 			 * This copy does NOT use an ASI
3205 			 * that avoids the Ecache, therefore
3206 			 * the dst_pa addresses may remain
3207 			 * in our Ecache after the dst_pa
3208 			 * has been removed from the system.
3209 			 * A subsequent write-back to memory
3210 			 * will cause an ARB-stop because the
3211 			 * physical address no longer exists
3212 			 * in the system. Therefore we must
3213 			 * flush out local Ecache after we
3214 			 * finish the copy.
3215 			 */
3216 
3217 			/* copy 32 bytes at src_pa to dst_pa */
3218 			bcopy32_il(s_pa, t_pa);
3219 
3220 			/* increment the counter to signal that we are alive */
3221 			prog->stat->nbytes[cpuid] += 32;
3222 
3223 			/* increment by 32 bytes */
3224 			s_pa += (4 * sizeof (uint64_t));
3225 			t_pa += (4 * sizeof (uint64_t));
3226 
3227 			/* decrement by 32 bytes */
3228 			nbytes -= (4 * sizeof (uint64_t));
3229 		}
3230 	    }
3231 	    prog->critical->stat[cpuid] = FMEM_LOOP_COPY_DONE;
3232 	    membar_sync_il();
3233 	}
3234 
3235 	/*
3236 	 * Since bcopy32_il() does NOT use an ASI to bypass
3237 	 * the Ecache, we need to flush our Ecache after
3238 	 * the copy is complete.
3239 	 */
3240 	flush_cache_il();
3241 
3242 	/*
3243 	 * drmach_fmem_exec_script()
3244 	 */
3245 	if (prog->data->cpuid == cpuid) {
3246 		uint64_t	last, now;
3247 
3248 		limit = copy_start + prog->data->copy_delay;
3249 		for (i = 0; i < NCPU; i++) {
3250 			if (CPU_IN_SET(prog->data->cpu_slave_set, i)) {
3251 			    for (;;) {
3252 				/* we get FMEM_LOOP_FMEM_READY in normal case */
3253 				if (prog->critical->stat[i] ==
3254 					FMEM_LOOP_FMEM_READY) {
3255 					break;
3256 				}
3257 				/* got error traps */
3258 				if (prog->data->error[i] ==
3259 					EOPL_FMEM_COPY_ERROR) {
3260 					prog->data->fmem_status.error =
3261 						EOPL_FMEM_COPY_ERROR;
3262 					return (EOPL_FMEM_COPY_ERROR);
3263 				}
3264 				/* if we have not reached limit, wait more */
3265 				curr = drmach_get_stick_il();
3266 				if (curr <= limit)
3267 					continue;
3268 
3269 				prog->data->slowest_cpuid = i;
3270 				prog->data->copy_wait_time =
3271 					curr - copy_start;
3272 
3273 				/* now check if slave is alive */
3274 				last = prog->stat->nbytes[i];
3275 
3276 				DR_DELAY_IL(1, prog->data->stick_freq);
3277 
3278 				now = prog->stat->nbytes[i];
3279 				if (now <= last) {
3280 					/* no progress, perhaps just finished */
3281 					DR_DELAY_IL(1, prog->data->stick_freq);
3282 					if (prog->critical->stat[i] ==
3283 						FMEM_LOOP_FMEM_READY)
3284 						break;
3285 					/* copy error */
3286 					if (prog->data->error[i] ==
3287 						EOPL_FMEM_COPY_ERROR) {
3288 						prog->data->fmem_status.error =
3289 							EOPL_FMEM_COPY_ERROR;
3290 						return (EOPL_FMEM_COPY_ERROR);
3291 					}
3292 					prog->data->fmem_status.error =
3293 					    EOPL_FMEM_COPY_TIMEOUT;
3294 					return (EOPL_FMEM_COPY_TIMEOUT);
3295 				}
3296 			    }
3297 			}
3298 		}
3299 
3300 		prog->critical->stat[cpuid] = FMEM_LOOP_FMEM_READY;
3301 		prog->data->fmem_status.stat  = FMEM_LOOP_FMEM_READY;
3302 
3303 		membar_sync_il();
3304 		flush_instr_mem_il((void*) (prog->critical));
3305 		/*
3306 		 * drmach_fmem_exec_script()
3307 		 */
3308 		rtn = prog->critical->fmem((void *)prog->critical, PAGESIZE);
3309 		return (rtn);
3310 	} else {
3311 		flush_instr_mem_il((void*) (prog->critical));
3312 		/*
3313 		 * drmach_fmem_loop_script()
3314 		 */
3315 		rtn = prog->critical->loop((void *)(prog->critical),
3316 			PAGESIZE, (void *)&(prog->critical->stat[cpuid]));
3317 		prog->data->error[cpuid] = rtn;
3318 		/* slave thread does not care the rv */
3319 		return (0);
3320 	}
3321 }
3322 
3323 static void
3324 drmach_copy_rename_end(void)
3325 {
3326 	/*
3327 	 * IMPORTANT:	This function's location MUST be located immediately
3328 	 *		following drmach_copy_rename_prog__relocatable to
3329 	 *		accurately estimate its size.  Note that this assumes
3330 	 *		the compiler keeps these functions in the order in
3331 	 *		which they appear :-o
3332 	 */
3333 }
3334 
3335 
3336 static void
3337 drmach_setup_memlist(drmach_copy_rename_program_t *p)
3338 {
3339 	struct memlist *ml;
3340 	caddr_t buf;
3341 	int nbytes, s;
3342 
3343 	nbytes = PAGESIZE;
3344 	s = roundup(sizeof (struct memlist), sizeof (void *));
3345 	p->free_mlist = NULL;
3346 	buf = p->memlist_buffer;
3347 	while (nbytes >= sizeof (struct memlist)) {
3348 		ml = (struct memlist *)buf;
3349 		ml->next = p->free_mlist;
3350 		p->free_mlist = ml;
3351 		buf += s;
3352 		nbytes -= s;
3353 	}
3354 }
3355 
3356 static void
3357 drmach_lock_critical(caddr_t va, caddr_t new_va)
3358 {
3359 	tte_t tte;
3360 	int i;
3361 
3362 	kpreempt_disable();
3363 
3364 	for (i = 0; i < DRMACH_FMEM_LOCKED_PAGES; i++) {
3365 		vtag_flushpage(new_va, (uint64_t)ksfmmup);
3366 		sfmmu_memtte(&tte, va_to_pfn(va),
3367 			PROC_DATA|HAT_NOSYNC, TTE8K);
3368 		tte.tte_intlo |= TTE_LCK_INT;
3369 		sfmmu_dtlb_ld_kva(new_va, &tte);
3370 		sfmmu_itlb_ld_kva(new_va, &tte);
3371 		va += PAGESIZE;
3372 		new_va += PAGESIZE;
3373 	}
3374 }
3375 
3376 static void
3377 drmach_unlock_critical(caddr_t va)
3378 {
3379 	int i;
3380 
3381 	for (i = 0; i < DRMACH_FMEM_LOCKED_PAGES; i++) {
3382 		vtag_flushpage(va, (uint64_t)ksfmmup);
3383 		va += PAGESIZE;
3384 	}
3385 
3386 	kpreempt_enable();
3387 }
3388 
3389 sbd_error_t *
3390 drmach_copy_rename_init(drmachid_t t_id, drmachid_t s_id,
3391 	struct memlist *c_ml, drmachid_t *pgm_id)
3392 {
3393 	drmach_mem_t	*s_mem;
3394 	drmach_mem_t	*t_mem;
3395 	struct memlist	*x_ml;
3396 	uint64_t	s_copybasepa, t_copybasepa;
3397 	uint_t		len;
3398 	caddr_t		bp, wp;
3399 	int			s_bd, t_bd, cpuid, active_cpus, i;
3400 	uint64_t		c_addr;
3401 	size_t			c_size, copy_sz, sz;
3402 	extern void		drmach_fmem_loop_script();
3403 	extern void		drmach_fmem_loop_script_rtn();
3404 	extern int		drmach_fmem_exec_script();
3405 	extern void		drmach_fmem_exec_script_end();
3406 	sbd_error_t	*err;
3407 	drmach_copy_rename_program_t *prog = NULL;
3408 	drmach_copy_rename_program_t *prog_kmem = NULL;
3409 	void		(*mc_suspend)(void);
3410 	void		(*mc_resume)(void);
3411 	int		(*scf_fmem_start)(int, int);
3412 	int		(*scf_fmem_end)(void);
3413 	int		(*scf_fmem_cancel)(void);
3414 	uint64_t	(*scf_get_base_addr)(void);
3415 
3416 	if (!DRMACH_IS_MEM_ID(s_id))
3417 		return (drerr_new(0, EOPL_INAPPROP, NULL));
3418 	if (!DRMACH_IS_MEM_ID(t_id))
3419 		return (drerr_new(0, EOPL_INAPPROP, NULL));
3420 
3421 	for (i = 0; i < NCPU; i++) {
3422 		int lsb_id, onb_core_num, strand_id;
3423 		drmach_board_t *bp;
3424 
3425 		/*
3426 		 * this kind of CPU will spin in cache
3427 		 */
3428 		if (CPU_IN_SET(cpu_ready_set, i))
3429 			continue;
3430 
3431 		/*
3432 		 * Now check for any inactive CPU's that
3433 		 * have been hotadded.  This can only occur in
3434 		 * error condition in drmach_cpu_poweron().
3435 		 */
3436 		lsb_id = LSB_ID(i);
3437 		onb_core_num = ON_BOARD_CORE_NUM(i);
3438 		strand_id = STRAND_ID(i);
3439 		bp = drmach_get_board_by_bnum(lsb_id);
3440 		if (bp == NULL)
3441 			continue;
3442 		if (bp->cores[onb_core_num].core_hotadded &
3443 		    (1 << strand_id)) {
3444 		    if (!(bp->cores[onb_core_num].core_started &
3445 			(1 << strand_id))) {
3446 			return (drerr_new(1, EOPL_CPU_STATE, NULL));
3447 		    }
3448 		}
3449 	}
3450 
3451 	mc_suspend = (void (*)(void))
3452 	    modgetsymvalue("opl_mc_suspend", 0);
3453 	mc_resume = (void (*)(void))
3454 	    modgetsymvalue("opl_mc_resume", 0);
3455 
3456 	if (mc_suspend == NULL || mc_resume == NULL) {
3457 		return (drerr_new(1, EOPL_MC_OPL, NULL));
3458 	}
3459 
3460 	scf_fmem_start = (int (*)(int, int))
3461 	    modgetsymvalue("scf_fmem_start", 0);
3462 	if (scf_fmem_start == NULL) {
3463 		return (drerr_new(1, EOPL_SCF_FMEM, NULL));
3464 	}
3465 	scf_fmem_end = (int (*)(void))
3466 	    modgetsymvalue("scf_fmem_end", 0);
3467 	if (scf_fmem_end == NULL) {
3468 		return (drerr_new(1, EOPL_SCF_FMEM, NULL));
3469 	}
3470 	scf_fmem_cancel = (int (*)(void))
3471 	    modgetsymvalue("scf_fmem_cancel", 0);
3472 	if (scf_fmem_cancel == NULL) {
3473 		return (drerr_new(1, EOPL_SCF_FMEM, NULL));
3474 	}
3475 	scf_get_base_addr = (uint64_t (*)(void))
3476 	    modgetsymvalue("scf_get_base_addr", 0);
3477 	if (scf_get_base_addr == NULL) {
3478 		return (drerr_new(1, EOPL_SCF_FMEM, NULL));
3479 	}
3480 	s_mem = s_id;
3481 	t_mem = t_id;
3482 
3483 	s_bd = s_mem->dev.bp->bnum;
3484 	t_bd = t_mem->dev.bp->bnum;
3485 
3486 	/* calculate source and target base pa */
3487 
3488 	s_copybasepa = s_mem->slice_base;
3489 	t_copybasepa = t_mem->slice_base;
3490 
3491 	/* adjust copy memlist addresses to be relative to copy base pa */
3492 	x_ml = c_ml;
3493 	while (x_ml != NULL) {
3494 		x_ml->address -= s_copybasepa;
3495 		x_ml = x_ml->next;
3496 	}
3497 
3498 	/*
3499 	 * bp will be page aligned, since we're calling
3500 	 * kmem_zalloc() with an exact multiple of PAGESIZE.
3501 	 */
3502 
3503 	prog_kmem = (drmach_copy_rename_program_t *)kmem_zalloc(
3504 		DRMACH_FMEM_LOCKED_PAGES * PAGESIZE, KM_SLEEP);
3505 
3506 	prog_kmem->prog = prog_kmem;
3507 
3508 	/*
3509 	 * To avoid MTLB hit, we allocate a new VM space and remap
3510 	 * the kmem_alloc buffer to that address.  This solves
3511 	 * 2 problems we found:
3512 	 * - the kmem_alloc buffer can be just a chunk inside
3513 	 *   a much larger, e.g. 4MB buffer and MTLB will occur
3514 	 *   if there are both a 4MB and a 8K TLB mapping to
3515 	 *   the same VA range.
3516 	 * - the kmem mapping got dropped into the TLB by other
3517 	 *   strands, unintentionally.
3518 	 * Note that the pointers like data, critical, memlist_buffer,
3519 	 * and stat inside the copy rename structure are mapped to this
3520 	 * alternate VM space so we must make sure we lock the TLB mapping
3521 	 * whenever we access data pointed to by these pointers.
3522 	 */
3523 
3524 	prog = prog_kmem->locked_prog = vmem_alloc(heap_arena,
3525 		DRMACH_FMEM_LOCKED_PAGES * PAGESIZE, VM_SLEEP);
3526 	wp = bp = (caddr_t)prog;
3527 
3528 	/* Now remap prog_kmem to prog */
3529 	drmach_lock_critical((caddr_t)prog_kmem, (caddr_t)prog);
3530 
3531 	/* All pointers in prog are based on the alternate mapping */
3532 	prog->data = (drmach_copy_rename_data_t *)roundup(((uint64_t)prog +
3533 		sizeof (drmach_copy_rename_program_t)), sizeof (void *));
3534 
3535 	ASSERT(((uint64_t)prog->data + sizeof (drmach_copy_rename_data_t))
3536 		<= ((uint64_t)prog + PAGESIZE));
3537 
3538 	prog->critical = (drmach_copy_rename_critical_t *)
3539 		(wp + DRMACH_FMEM_CRITICAL_PAGE * PAGESIZE);
3540 
3541 	prog->memlist_buffer = (caddr_t)(wp +
3542 		DRMACH_FMEM_MLIST_PAGE * PAGESIZE);
3543 
3544 	prog->stat = (drmach_cr_stat_t *)(wp +
3545 		DRMACH_FMEM_STAT_PAGE * PAGESIZE);
3546 
3547 	/* LINTED */
3548 	ASSERT(sizeof (drmach_cr_stat_t)
3549 		<= ((DRMACH_FMEM_LOCKED_PAGES - DRMACH_FMEM_STAT_PAGE)
3550 		* PAGESIZE));
3551 
3552 	prog->critical->scf_reg_base = (uint64_t)-1;
3553 	prog->critical->scf_td[0] = (s_bd & 0xff);
3554 	prog->critical->scf_td[1] = (t_bd & 0xff);
3555 	for (i = 2; i < 15; i++) {
3556 		prog->critical->scf_td[i]   = 0;
3557 	}
3558 	prog->critical->scf_td[15] = ((0xaa + s_bd + t_bd) & 0xff);
3559 
3560 	bp = (caddr_t)prog->critical;
3561 	len = sizeof (drmach_copy_rename_critical_t);
3562 	wp = (caddr_t)roundup((uint64_t)bp + len, sizeof (void *));
3563 
3564 	len = (uint_t)((ulong_t)drmach_copy_rename_end -
3565 		(ulong_t)drmach_copy_rename_prog__relocatable);
3566 
3567 	/*
3568 	 * We always leave 1K nop's to prevent the processor from
3569 	 * speculative execution that causes memory access
3570 	 */
3571 	wp = wp + len + 1024;
3572 
3573 	len = (uint_t)((ulong_t)drmach_fmem_exec_script_end -
3574 		(ulong_t)drmach_fmem_exec_script);
3575 	/* this is the entry point of the loop script */
3576 	wp = wp + len + 1024;
3577 
3578 	len = (uint_t)((ulong_t)drmach_fmem_exec_script -
3579 		(ulong_t)drmach_fmem_loop_script);
3580 	wp = wp + len + 1024;
3581 
3582 	/* now we make sure there is 1K extra */
3583 
3584 	if ((wp - bp) > PAGESIZE) {
3585 		err = drerr_new(1, EOPL_FMEM_SETUP, NULL);
3586 		goto out;
3587 	}
3588 
3589 	bp = (caddr_t)prog->critical;
3590 	len = sizeof (drmach_copy_rename_critical_t);
3591 	wp = (caddr_t)roundup((uint64_t)bp + len, sizeof (void *));
3592 
3593 	prog->critical->run = (int (*)())(wp);
3594 	len = (uint_t)((ulong_t)drmach_copy_rename_end -
3595 		(ulong_t)drmach_copy_rename_prog__relocatable);
3596 
3597 	bcopy((caddr_t)drmach_copy_rename_prog__relocatable, wp, len);
3598 
3599 	wp = (caddr_t)roundup((uint64_t)wp + len, 1024);
3600 
3601 	prog->critical->fmem = (int (*)())(wp);
3602 	len = (int)((ulong_t)drmach_fmem_exec_script_end -
3603 		(ulong_t)drmach_fmem_exec_script);
3604 	bcopy((caddr_t)drmach_fmem_exec_script, wp, len);
3605 
3606 	len = (int)((ulong_t)drmach_fmem_exec_script_end -
3607 		(ulong_t)drmach_fmem_exec_script);
3608 	wp = (caddr_t)roundup((uint64_t)wp + len, 1024);
3609 
3610 	prog->critical->loop = (int (*)())(wp);
3611 	len = (int)((ulong_t)drmach_fmem_exec_script -
3612 		(ulong_t)drmach_fmem_loop_script);
3613 	bcopy((caddr_t)drmach_fmem_loop_script, (void *)wp, len);
3614 	len = (int)((ulong_t)drmach_fmem_loop_script_rtn-
3615 		(ulong_t)drmach_fmem_loop_script);
3616 	prog->critical->loop_rtn = (void (*)()) (wp+len);
3617 
3618 	prog->data->fmem_status.error = ESBD_NOERROR;
3619 
3620 	/* now we are committed, call SCF, soft suspend mac patrol */
3621 	if ((*scf_fmem_start)(s_bd, t_bd)) {
3622 		err = drerr_new(1, EOPL_SCF_FMEM_START, NULL);
3623 		goto out;
3624 	}
3625 	prog->data->scf_fmem_end = scf_fmem_end;
3626 	prog->data->scf_fmem_cancel = scf_fmem_cancel;
3627 	prog->data->scf_get_base_addr = scf_get_base_addr;
3628 	prog->data->fmem_status.op |= OPL_FMEM_SCF_START;
3629 
3630 	/* soft suspend mac patrol */
3631 	(*mc_suspend)();
3632 	prog->data->fmem_status.op |= OPL_FMEM_MC_SUSPEND;
3633 	prog->data->mc_resume = mc_resume;
3634 
3635 	prog->critical->inst_loop_ret  =
3636 		*(uint64_t *)(prog->critical->loop_rtn);
3637 
3638 	/*
3639 	 * 0x30800000 is op code "ba,a	+0"
3640 	 */
3641 
3642 	*(uint_t *)(prog->critical->loop_rtn) = (uint_t)(0x30800000);
3643 
3644 	/*
3645 	 * set the value of SCF FMEM TIMEOUT
3646 	 */
3647 	prog->critical->delay = fmem_timeout * system_clock_freq;
3648 
3649 	prog->data->s_mem = (drmachid_t)s_mem;
3650 	prog->data->t_mem = (drmachid_t)t_mem;
3651 
3652 	cpuid = CPU->cpu_id;
3653 	prog->data->cpuid = cpuid;
3654 	prog->data->cpu_ready_set = cpu_ready_set;
3655 	prog->data->cpu_slave_set = cpu_ready_set;
3656 	prog->data->slowest_cpuid = (processorid_t)-1;
3657 	prog->data->copy_wait_time = 0;
3658 	CPUSET_DEL(prog->data->cpu_slave_set, cpuid);
3659 
3660 	for (i = 0; i < NCPU; i++) {
3661 		prog->data->cpu_ml[i] = NULL;
3662 	}
3663 
3664 	active_cpus = 0;
3665 	if (drmach_disable_mcopy) {
3666 		active_cpus = 1;
3667 		CPUSET_ADD(prog->data->cpu_copy_set, cpuid);
3668 	} else {
3669 		for (i = 0; i < NCPU; i++) {
3670 			if (CPU_IN_SET(cpu_ready_set, i) &&
3671 				CPU_ACTIVE(cpu[i])) {
3672 				CPUSET_ADD(prog->data->cpu_copy_set, i);
3673 				active_cpus++;
3674 			}
3675 		}
3676 	}
3677 
3678 	drmach_setup_memlist(prog);
3679 
3680 	x_ml = c_ml;
3681 	sz = 0;
3682 	while (x_ml != NULL) {
3683 		sz += x_ml->size;
3684 		x_ml = x_ml->next;
3685 	}
3686 
3687 	copy_sz = sz/active_cpus;
3688 	copy_sz = roundup(copy_sz, MMU_PAGESIZE4M);
3689 
3690 	while (sz > copy_sz*active_cpus) {
3691 		copy_sz += MMU_PAGESIZE4M;
3692 	}
3693 
3694 	prog->data->stick_freq = system_clock_freq;
3695 	prog->data->copy_delay = ((copy_sz / min_copy_size_per_sec) + 2) *
3696 		system_clock_freq;
3697 
3698 	x_ml = c_ml;
3699 	c_addr = x_ml->address;
3700 	c_size = x_ml->size;
3701 
3702 	for (i = 0; i < NCPU; i++) {
3703 		prog->stat->nbytes[i] = 0;
3704 		if (!CPU_IN_SET(prog->data->cpu_copy_set, i)) {
3705 			continue;
3706 		}
3707 		sz = copy_sz;
3708 
3709 		while (sz) {
3710 			if (c_size > sz) {
3711 				prog->data->cpu_ml[i] =
3712 					drmach_memlist_add_span(prog,
3713 					prog->data->cpu_ml[i],
3714 					c_addr, sz);
3715 				c_addr += sz;
3716 				c_size -= sz;
3717 				break;
3718 			} else {
3719 				sz -= c_size;
3720 				prog->data->cpu_ml[i] = drmach_memlist_add_span(
3721 					prog, prog->data->cpu_ml[i],
3722 						c_addr, c_size);
3723 				x_ml = x_ml->next;
3724 				if (x_ml != NULL) {
3725 					c_addr = x_ml->address;
3726 					c_size = x_ml->size;
3727 				} else {
3728 					goto end;
3729 				}
3730 			}
3731 		}
3732 	}
3733 end:
3734 	prog->data->s_copybasepa = s_copybasepa;
3735 	prog->data->t_copybasepa = t_copybasepa;
3736 	prog->data->c_ml = c_ml;
3737 	*pgm_id = prog_kmem;
3738 
3739 	/* Unmap the alternate space.  It will have to be remapped again */
3740 	drmach_unlock_critical((caddr_t)prog);
3741 	return (NULL);
3742 out:
3743 	if (prog != NULL) {
3744 		drmach_unlock_critical((caddr_t)prog);
3745 		vmem_free(heap_arena, prog,
3746 			DRMACH_FMEM_LOCKED_PAGES * PAGESIZE);
3747 	}
3748 	if (prog_kmem != NULL) {
3749 		kmem_free(prog_kmem, DRMACH_FMEM_LOCKED_PAGES * PAGESIZE);
3750 	}
3751 	return (err);
3752 }
3753 
3754 sbd_error_t *
3755 drmach_copy_rename_fini(drmachid_t id)
3756 {
3757 	drmach_copy_rename_program_t	*prog = id;
3758 	sbd_error_t			*err = NULL;
3759 	int				rv;
3760 	uint_t				fmem_error;
3761 
3762 	/*
3763 	 * Note that we have to delay calling SCF to find out the
3764 	 * status of the FMEM operation here because SCF cannot
3765 	 * respond while it is suspended.
3766 	 * This create a small window when we are sure about the
3767 	 * base address of the system board.
3768 	 * If there is any call to mc-opl to get memory unum,
3769 	 * mc-opl will return UNKNOWN as the unum.
3770 	 */
3771 
3772 	/*
3773 	 * we have to remap again because all the pointer like data,
3774 	 * critical in prog are based on the alternate vmem space.
3775 	 */
3776 	(void) drmach_lock_critical((caddr_t)prog, (caddr_t)prog->locked_prog);
3777 
3778 	if (prog->data->c_ml != NULL)
3779 		memlist_delete(prog->data->c_ml);
3780 
3781 	if ((prog->data->fmem_status.op &
3782 		(OPL_FMEM_SCF_START| OPL_FMEM_MC_SUSPEND)) !=
3783 		(OPL_FMEM_SCF_START | OPL_FMEM_MC_SUSPEND)) {
3784 		cmn_err(CE_PANIC, "drmach_copy_rename_fini: "
3785 			"invalid op code %x\n",
3786 				prog->data->fmem_status.op);
3787 	}
3788 
3789 	fmem_error = prog->data->fmem_status.error;
3790 	if (fmem_error != ESBD_NOERROR) {
3791 		err = drerr_new(1, fmem_error, NULL);
3792 	}
3793 
3794 	/* possible ops are SCF_START, MC_SUSPEND */
3795 	if (prog->critical->fmem_issued) {
3796 		if (fmem_error != ESBD_NOERROR) {
3797 		    cmn_err(CE_PANIC, "Irrecoverable FMEM error %d\n",
3798 			fmem_error);
3799 		}
3800 		rv = (*prog->data->scf_fmem_end)();
3801 		if (rv) {
3802 			cmn_err(CE_PANIC, "scf_fmem_end() failed rv=%d", rv);
3803 		}
3804 		/*
3805 		 * If we get here, rename is successful.
3806 		 * Do all the copy rename post processing.
3807 		 */
3808 		drmach_swap_pa((drmach_mem_t *)prog->data->s_mem,
3809 			(drmach_mem_t *)prog->data->t_mem);
3810 	} else {
3811 		rv = (*prog->data->scf_fmem_cancel)();
3812 		if (rv) {
3813 		    cmn_err(CE_WARN, "scf_fmem_cancel() failed rv=0x%x", rv);
3814 		    if (!err)
3815 			err = drerr_new(1, EOPL_SCF_FMEM_CANCEL,
3816 			    "scf_fmem_cancel() failed. rv = 0x%x", rv);
3817 		}
3818 	}
3819 	/* soft resume mac patrol */
3820 	(*prog->data->mc_resume)();
3821 
3822 	drmach_unlock_critical((caddr_t)prog->locked_prog);
3823 
3824 	vmem_free(heap_arena, prog->locked_prog,
3825 		DRMACH_FMEM_LOCKED_PAGES * PAGESIZE);
3826 	kmem_free(prog, DRMACH_FMEM_LOCKED_PAGES * PAGESIZE);
3827 	return (err);
3828 }
3829 
3830 /*ARGSUSED*/
3831 static void
3832 drmach_copy_rename_slave(struct regs *rp, drmachid_t id)
3833 {
3834 	drmach_copy_rename_program_t	*prog =
3835 		(drmach_copy_rename_program_t *)id;
3836 	register int			cpuid;
3837 	extern void			drmach_flush();
3838 	extern void			membar_sync_il();
3839 	extern void			drmach_flush_icache();
3840 	on_trap_data_t			otd;
3841 
3842 	cpuid = CPU->cpu_id;
3843 
3844 	if (on_trap(&otd, OT_DATA_EC)) {
3845 		no_trap();
3846 		prog->data->error[cpuid] = EOPL_FMEM_COPY_ERROR;
3847 		prog->critical->stat[cpuid] = FMEM_LOOP_EXIT;
3848 		drmach_flush_icache();
3849 		membar_sync_il();
3850 		return;
3851 	}
3852 
3853 
3854 	/*
3855 	 * jmp drmach_copy_rename_prog().
3856 	 */
3857 
3858 	drmach_flush(prog->critical, PAGESIZE);
3859 	(void) prog->critical->run(prog, cpuid);
3860 	drmach_flush_icache();
3861 
3862 	no_trap();
3863 
3864 	prog->critical->stat[cpuid] = FMEM_LOOP_EXIT;
3865 
3866 	membar_sync_il();
3867 }
3868 
3869 static void
3870 drmach_swap_pa(drmach_mem_t *s_mem, drmach_mem_t *t_mem)
3871 {
3872 	uint64_t s_base, t_base;
3873 	drmach_board_t *s_board, *t_board;
3874 	struct memlist *ml;
3875 
3876 	s_board = s_mem->dev.bp;
3877 	t_board = t_mem->dev.bp;
3878 	if (s_board == NULL || t_board == NULL) {
3879 		cmn_err(CE_PANIC, "Cannot locate source or target board\n");
3880 		return;
3881 	}
3882 	s_base = s_mem->slice_base;
3883 	t_base = t_mem->slice_base;
3884 
3885 	s_mem->slice_base = t_base;
3886 	s_mem->base_pa = (s_mem->base_pa - s_base) + t_base;
3887 
3888 	for (ml = s_mem->memlist; ml; ml = ml->next) {
3889 		ml->address = ml->address - s_base + t_base;
3890 	}
3891 
3892 	t_mem->slice_base = s_base;
3893 	t_mem->base_pa = (t_mem->base_pa - t_base) + s_base;
3894 
3895 	for (ml = t_mem->memlist; ml; ml = ml->next) {
3896 		ml->address = ml->address - t_base + s_base;
3897 	}
3898 
3899 	/*
3900 	 * IKP has to update the sb-mem-ranges for mac patrol driver
3901 	 * when it resumes, it will re-read the sb-mem-range property
3902 	 * to get the new base address
3903 	 */
3904 	if (oplcfg_pa_swap(s_board->bnum, t_board->bnum) != 0)
3905 		cmn_err(CE_PANIC, "Could not update device nodes\n");
3906 }
3907 
3908 void
3909 drmach_copy_rename(drmachid_t id)
3910 {
3911 	drmach_copy_rename_program_t	*prog_kmem = id;
3912 	drmach_copy_rename_program_t	*prog;
3913 	cpuset_t	cpuset;
3914 	int		cpuid;
3915 	uint64_t	inst;
3916 	register int	rtn;
3917 	extern int	in_sync;
3918 	int		old_in_sync;
3919 	extern void	drmach_sys_trap();
3920 	extern void	drmach_flush();
3921 	extern void	drmach_flush_icache();
3922 	extern uint64_t	patch_inst(uint64_t *, uint64_t);
3923 	on_trap_data_t	otd;
3924 
3925 
3926 	prog = prog_kmem->locked_prog;
3927 
3928 
3929 	/*
3930 	 * We must immediately drop in the TLB because all pointers
3931 	 * are based on the alternate vmem space.
3932 	 */
3933 
3934 	(void) drmach_lock_critical((caddr_t)prog_kmem, (caddr_t)prog);
3935 
3936 	/*
3937 	 * we call scf to get the base address here becuase if scf
3938 	 * has not been suspended yet, the active path can be changing and
3939 	 * sometimes it is not even mapped.  We call the interface when
3940 	 * the OS has been quiesced.
3941 	 */
3942 	prog->critical->scf_reg_base = (*prog->data->scf_get_base_addr)();
3943 
3944 	if (prog->critical->scf_reg_base == (uint64_t)-1 ||
3945 		prog->critical->scf_reg_base == NULL) {
3946 		prog->data->fmem_status.error = EOPL_FMEM_SCF_ERR;
3947 		drmach_unlock_critical((caddr_t)prog);
3948 		return;
3949 	}
3950 
3951 	cpuset = prog->data->cpu_ready_set;
3952 
3953 	for (cpuid = 0; cpuid < NCPU; cpuid++) {
3954 		if (CPU_IN_SET(cpuset, cpuid)) {
3955 			prog->critical->stat[cpuid] = FMEM_LOOP_START;
3956 			prog->data->error[cpuid] = ESBD_NOERROR;
3957 		}
3958 	}
3959 
3960 	old_in_sync = in_sync;
3961 	in_sync = 1;
3962 	cpuid = CPU->cpu_id;
3963 
3964 	CPUSET_DEL(cpuset, cpuid);
3965 
3966 	for (cpuid = 0; cpuid < NCPU; cpuid++) {
3967 		if (CPU_IN_SET(cpuset, cpuid)) {
3968 			xc_one(cpuid, (xcfunc_t *)drmach_lock_critical,
3969 				(uint64_t)prog_kmem, (uint64_t)prog);
3970 		}
3971 	}
3972 
3973 	cpuid = CPU->cpu_id;
3974 
3975 	xt_some(cpuset, (xcfunc_t *)drmach_sys_trap,
3976 				(uint64_t)drmach_copy_rename_slave,
3977 				(uint64_t)prog);
3978 	xt_sync(cpuset);
3979 
3980 	if (on_trap(&otd, OT_DATA_EC)) {
3981 		rtn = EOPL_FMEM_COPY_ERROR;
3982 		drmach_flush_icache();
3983 		goto done;
3984 	}
3985 
3986 	/*
3987 	 * jmp drmach_copy_rename_prog().
3988 	 */
3989 
3990 	drmach_flush(prog->critical, PAGESIZE);
3991 	rtn = prog->critical->run(prog, cpuid);
3992 	drmach_flush_icache();
3993 
3994 
3995 done:
3996 	no_trap();
3997 	if (rtn == EOPL_FMEM_HW_ERROR) {
3998 		kpreempt_enable();
3999 		prom_panic("URGENT_ERROR_TRAP is "
4000 			"detected during FMEM.\n");
4001 	}
4002 
4003 	/*
4004 	 * In normal case, all slave CPU's are still spinning in
4005 	 * the assembly code.  The master has to patch the instruction
4006 	 * to get them out.
4007 	 * In error case, e.g. COPY_ERROR, some slave CPU's might
4008 	 * have aborted and already returned and sset LOOP_EXIT status.
4009 	 * Some CPU might still be copying.
4010 	 * In any case, some delay is necessary to give them
4011 	 * enough time to set the LOOP_EXIT status.
4012 	 */
4013 
4014 	for (;;) {
4015 		inst = patch_inst((uint64_t *)prog->critical->loop_rtn,
4016 			prog->critical->inst_loop_ret);
4017 		if (prog->critical->inst_loop_ret == inst) {
4018 			break;
4019 		}
4020 	}
4021 
4022 	for (cpuid = 0; cpuid < NCPU; cpuid++) {
4023 		uint64_t	last, now;
4024 		if (!CPU_IN_SET(cpuset, cpuid)) {
4025 			continue;
4026 		}
4027 		last = prog->stat->nbytes[cpuid];
4028 		/*
4029 		 * Wait for all CPU to exit.
4030 		 * However we do not want an infinite loop
4031 		 * so we detect hangup situation here.
4032 		 * If the slave CPU is still copying data,
4033 		 * we will continue to wait.
4034 		 * In error cases, the master has already set
4035 		 * fmem_status.error to abort the copying.
4036 		 * 1 m.s delay for them to abort copying and
4037 		 * return to drmach_copy_rename_slave to set
4038 		 * FMEM_LOOP_EXIT status should be enough.
4039 		 */
4040 		for (;;) {
4041 			if (prog->critical->stat[cpuid] == FMEM_LOOP_EXIT)
4042 				break;
4043 			drmach_sleep_il();
4044 			drv_usecwait(1000);
4045 			now = prog->stat->nbytes[cpuid];
4046 			if (now <= last) {
4047 			    drv_usecwait(1000);
4048 			    if (prog->critical->stat[cpuid] == FMEM_LOOP_EXIT)
4049 				break;
4050 			    cmn_err(CE_PANIC,
4051 				"CPU %d hang during Copy Rename", cpuid);
4052 			}
4053 			last = now;
4054 		}
4055 		if (prog->data->error[cpuid] == EOPL_FMEM_HW_ERROR) {
4056 			prom_panic("URGENT_ERROR_TRAP is "
4057 				"detected during FMEM.\n");
4058 		}
4059 	}
4060 
4061 	/*
4062 	 * This must be done after all strands have exit.
4063 	 * Removing the TLB entry will affect both strands
4064 	 * in the same core.
4065 	 */
4066 
4067 	for (cpuid = 0; cpuid < NCPU; cpuid++) {
4068 		if (CPU_IN_SET(cpuset, cpuid)) {
4069 			xc_one(cpuid, (xcfunc_t *)drmach_unlock_critical,
4070 				(uint64_t)prog, 0);
4071 		}
4072 	}
4073 
4074 	in_sync = old_in_sync;
4075 
4076 	/*
4077 	 * we should unlock before the following lock to keep the kpreempt
4078 	 * count correct.
4079 	 */
4080 	(void) drmach_unlock_critical((caddr_t)prog);
4081 
4082 	/*
4083 	 * we must remap again.  TLB might have been removed in above xcall.
4084 	 */
4085 
4086 	(void) drmach_lock_critical((caddr_t)prog_kmem, (caddr_t)prog);
4087 
4088 	if (prog->data->fmem_status.error == ESBD_NOERROR)
4089 		prog->data->fmem_status.error = rtn;
4090 
4091 	if (prog->data->copy_wait_time > 0) {
4092 		DRMACH_PR("Unexpected long wait time %ld seconds "
4093 			"during copy rename on CPU %d\n",
4094 			prog->data->copy_wait_time/prog->data->stick_freq,
4095 			prog->data->slowest_cpuid);
4096 	}
4097 	drmach_unlock_critical((caddr_t)prog);
4098 }
4099