xref: /titanic_44/usr/src/uts/sun4u/starfire/io/drmach.c (revision 8eea8e29cc4374d1ee24c25a07f45af132db3499)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/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/processor.h>
47 #include <sys/spitregs.h>
48 #include <sys/cpuvar.h>
49 #include <sys/cpupart.h>
50 #include <sys/mem_config.h>
51 #include <sys/ddi_impldefs.h>
52 #include <sys/systm.h>
53 #include <sys/machsystm.h>
54 #include <sys/autoconf.h>
55 #include <sys/cmn_err.h>
56 #include <sys/sysmacros.h>
57 #include <sys/x_call.h>
58 #include <sys/promif.h>
59 #include <sys/prom_plat.h>
60 #include <sys/membar.h>
61 #include <vm/seg_kmem.h>
62 #include <sys/mem_cage.h>
63 #include <sys/stack.h>
64 #include <sys/archsystm.h>
65 #include <vm/hat_sfmmu.h>
66 #include <sys/pte.h>
67 #include <sys/mmu.h>
68 #include <sys/cpu_module.h>
69 #include <sys/obpdefs.h>
70 #include <sys/note.h>
71 
72 #include <sys/starfire.h>	/* plat_max_... decls */
73 #include <sys/cvc.h>
74 #include <sys/cpu_sgnblk_defs.h>
75 #include <sys/drmach.h>
76 #include <sys/dr_util.h>
77 #include <sys/pda.h>
78 
79 #include <sys/sysevent.h>
80 #include <sys/sysevent/dr.h>
81 #include <sys/sysevent/eventdefs.h>
82 
83 
84 extern void		bcopy32_il(uint64_t, uint64_t);
85 extern void		flush_ecache_il(
86 				uint64_t physaddr, int size, int linesz);
87 extern uint_t		ldphysio_il(uint64_t physaddr);
88 extern void		stphysio_il(uint64_t physaddr, uint_t value);
89 
90 extern uint64_t		mc_get_mem_alignment(void);
91 extern uint64_t		mc_get_asr_addr(dnode_t);
92 extern uint64_t		mc_get_idle_addr(dnode_t);
93 extern uint64_t		mc_get_alignment_mask(dnode_t);
94 extern int		mc_read_asr(dnode_t, uint_t *);
95 extern int		mc_write_asr(dnode_t, uint_t);
96 extern uint64_t		mc_asr_to_pa(uint_t);
97 extern uint_t		mc_pa_to_asr(uint_t, uint64_t);
98 
99 extern int		pc_madr_add(int, int, int, int);
100 
101 typedef struct {
102 	struct drmach_node	*node;
103 	void			*data;
104 } drmach_node_walk_args_t;
105 
106 typedef struct drmach_node {
107 	void		*here;
108 
109 	dnode_t		 (*get_dnode)(struct drmach_node *node);
110 	int		 (*walk)(struct drmach_node *node, void *data,
111 				int (*cb)(drmach_node_walk_args_t *args));
112 } drmach_node_t;
113 
114 typedef struct {
115 	int		 min_index;
116 	int		 max_index;
117 	int		 arr_sz;
118 	drmachid_t	*arr;
119 } drmach_array_t;
120 
121 typedef struct {
122 	void		*isa;
123 
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 	drmach_common_t	 cm;
132 	int		 bnum;
133 	int		 assigned;
134 	int		 powered;
135 	int		 connect_cpuid;
136 	int		 cond;
137 	drmach_node_t	*tree;
138 	drmach_array_t	*devices;
139 } drmach_board_t;
140 
141 typedef struct {
142 	drmach_common_t	 cm;
143 	drmach_board_t	*bp;
144 	int		 unum;
145 	int		 busy;
146 	int		 powered;
147 	const char	*type;
148 	drmach_node_t	*node;
149 } drmach_device_t;
150 
151 typedef struct {
152 	int		 flags;
153 	drmach_device_t	*dp;
154 	sbd_error_t	*err;
155 	dev_info_t	*dip;
156 } drmach_config_args_t;
157 
158 typedef struct {
159 	uint64_t	 idle_addr;
160 	drmach_device_t	*mem;
161 } drmach_mc_idle_script_t;
162 
163 typedef struct {
164 	uint64_t	masr_addr;
165 	uint_t		masr;
166 	uint_t		_filler;
167 } drmach_rename_script_t;
168 
169 typedef struct {
170 	void		(*run)(void *arg);
171 	caddr_t		data;
172 	pda_handle_t	*ph;
173 	struct memlist	*c_ml;
174 	uint64_t	s_copybasepa;
175 	uint64_t	t_copybasepa;
176 	drmach_device_t	*restless_mc;	/* diagnostic output */
177 } drmach_copy_rename_program_t;
178 
179 typedef enum {
180 	DO_IDLE,
181 	DO_UNIDLE,
182 	DO_PAUSE,
183 	DO_UNPAUSE
184 } drmach_iopc_op_t;
185 
186 typedef struct {
187 	drmach_board_t	*obj;
188 	int		 ndevs;
189 	void		*a;
190 	sbd_error_t	*(*found)(void *a, const char *, int, drmachid_t);
191 	sbd_error_t	*err;
192 } drmach_board_cb_data_t;
193 
194 static caddr_t		 drmach_shutdown_va;
195 
196 static int		 drmach_initialized;
197 static drmach_array_t	*drmach_boards;
198 
199 static int		 drmach_cpu_delay = 100;
200 static int		 drmach_cpu_ntries = 50000;
201 
202 volatile uchar_t	*drmach_xt_mb;
203 
204 /*
205  * Do not change the drmach_shutdown_mbox structure without
206  * considering the drmach_shutdown_asm assembly language code.
207  */
208 struct drmach_shutdown_mbox {
209 	uint64_t	estack;
210 	uint64_t	flushaddr;
211 	int		size;
212 	int		linesize;
213 	uint64_t	physaddr;
214 };
215 struct drmach_shutdown_mbox	*drmach_shutdown_asm_mbox;
216 static sbd_error_t	*drmach_device_new(drmach_node_t *,
217 				drmach_board_t *, drmach_device_t **);
218 static sbd_error_t	*drmach_cpu_new(drmach_device_t *);
219 static sbd_error_t	*drmach_mem_new(drmach_device_t *);
220 static sbd_error_t	*drmach_io_new(drmach_device_t *);
221 
222 extern struct cpu	*SIGBCPU;
223 
224 #ifdef DEBUG
225 
226 #define	DRMACH_PR		if (drmach_debug) printf
227 int drmach_debug = 0;		 /* set to non-zero to enable debug messages */
228 #else
229 
230 #define	DRMACH_PR		_NOTE(CONSTANTCONDITION) if (0) printf
231 #endif /* DEBUG */
232 
233 #define	DRMACH_OBJ(id)		((drmach_common_t *)id)
234 
235 #define	DRMACH_IS_BOARD_ID(id)	\
236 	((id != 0) &&		\
237 	(DRMACH_OBJ(id)->isa == (void *)drmach_board_new))
238 
239 #define	DRMACH_IS_CPU_ID(id)	\
240 	((id != 0) &&		\
241 	(DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new))
242 
243 #define	DRMACH_IS_MEM_ID(id)	\
244 	((id != 0) &&		\
245 	(DRMACH_OBJ(id)->isa == (void *)drmach_mem_new))
246 
247 #define	DRMACH_IS_IO_ID(id)	\
248 	((id != 0) &&		\
249 	(DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
250 
251 #define	DRMACH_IS_DEVICE_ID(id)					\
252 	((id != 0) &&						\
253 	(DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new ||	\
254 	    DRMACH_OBJ(id)->isa == (void *)drmach_mem_new ||	\
255 	    DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
256 
257 #define	DRMACH_IS_ID(id)					\
258 	((id != 0) &&						\
259 	(DRMACH_OBJ(id)->isa == (void *)drmach_board_new ||	\
260 	    DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new ||	\
261 	    DRMACH_OBJ(id)->isa == (void *)drmach_mem_new ||	\
262 	    DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
263 
264 #define	DRMACH_CPUID2BNUM(cpuid) \
265 	((cpuid) / MAX_CPU_UNITS_PER_BOARD)
266 
267 #define	DRMACH_INTERNAL_ERROR() \
268 	drerr_new(1, ESTF_INTERNAL, drmach_ie_fmt, __LINE__)
269 static char		*drmach_ie_fmt = "drmach.c %d";
270 
271 static struct {
272 	const char	 *name;
273 	const char	 *type;
274 	sbd_error_t	 *(*new)(drmach_device_t *);
275 } name2type[] = {
276 	{ "SUNW,UltraSPARC",	DRMACH_DEVTYPE_CPU,  drmach_cpu_new },
277 	{ "mem-unit",		DRMACH_DEVTYPE_MEM,  drmach_mem_new },
278 	{ "pci",		DRMACH_DEVTYPE_PCI,  drmach_io_new  },
279 	{ "sbus",		DRMACH_DEVTYPE_SBUS, drmach_io_new  },
280 };
281 
282 /* node types to cleanup when a board is unconfigured */
283 #define	MISC_COUNTER_TIMER_DEVNAME	"counter-timer"
284 #define	MISC_PERF_COUNTER_DEVNAME	"perf-counter"
285 
286 /* utility */
287 #define	MBYTE	(1048576ull)
288 
289 /*
290  * This is necessary because the CPU support needs
291  * to call cvc_assign_iocpu.
292  */
293 #ifndef lint
294 static char _depends_on[] = "drv/cvc";
295 #endif  /* lint */
296 
297 /*
298  * drmach autoconfiguration data structures and interfaces
299  */
300 
301 extern struct mod_ops mod_miscops;
302 
303 static struct modlmisc modlmisc = {
304 	&mod_miscops,
305 	"Sun Enterprise 10000 DR %I%"
306 };
307 
308 static struct modlinkage modlinkage = {
309 	MODREV_1,
310 	(void *)&modlmisc,
311 	NULL
312 };
313 
314 static kmutex_t drmach_i_lock;
315 
316 int
317 _init(void)
318 {
319 	int err;
320 
321 	/* check that we have the correct version of obp */
322 	if (prom_test("SUNW,UE10000,add-brd") != 0) {
323 
324 		cmn_err(CE_WARN, "!OBP/SSP upgrade is required to enable "
325 		    "DR Functionality");
326 
327 		return (-1);
328 	}
329 
330 	mutex_init(&drmach_i_lock, NULL, MUTEX_DRIVER, NULL);
331 
332 	drmach_xt_mb = (uchar_t *)vmem_alloc(static_alloc_arena,
333 	    NCPU * sizeof (uchar_t), VM_SLEEP);
334 	drmach_shutdown_asm_mbox = (struct drmach_shutdown_mbox *)
335 	    vmem_alloc(static_alloc_arena, sizeof (struct drmach_shutdown_mbox),
336 	    VM_SLEEP);
337 
338 	if ((err = mod_install(&modlinkage)) != 0) {
339 		mutex_destroy(&drmach_i_lock);
340 		vmem_free(static_alloc_arena, (void *)drmach_xt_mb,
341 		    NCPU * sizeof (uchar_t));
342 		vmem_free(static_alloc_arena, (void *)drmach_shutdown_asm_mbox,
343 		    sizeof (struct drmach_shutdown_mbox));
344 	}
345 
346 	return (err);
347 }
348 
349 int
350 _fini(void)
351 {
352 	static int drmach_fini(void);
353 
354 	if (drmach_fini())
355 		return (DDI_FAILURE);
356 	else
357 		return (mod_remove(&modlinkage));
358 }
359 
360 int
361 _info(struct modinfo *modinfop)
362 {
363 	return (mod_info(&modlinkage, modinfop));
364 }
365 
366 static dnode_t
367 drmach_node_obp_get_dnode(drmach_node_t *np)
368 {
369 	return ((dnode_t)np->here);
370 }
371 
372 static int
373 drmach_node_obp_walk(drmach_node_t *np, void *data,
374 		int (*cb)(drmach_node_walk_args_t *args))
375 {
376 	dnode_t			nodeid;
377 	int			rv;
378 	drmach_node_walk_args_t	args;
379 
380 	/* initialized args structure for callback */
381 	args.node = np;
382 	args.data = data;
383 
384 	nodeid = prom_childnode(prom_rootnode());
385 
386 	/* save our new position with in the tree */
387 	np->here = (void *)nodeid;
388 
389 	rv = 0;
390 	while (nodeid != OBP_NONODE) {
391 		rv = (*cb)(&args);
392 		if (rv)
393 			break;
394 
395 		nodeid = prom_nextnode(nodeid);
396 
397 		/* save our new position with in the tree */
398 		np->here = (void *)nodeid;
399 	}
400 
401 	return (rv);
402 }
403 
404 static drmach_node_t *
405 drmach_node_new(void)
406 {
407 	drmach_node_t *np;
408 
409 	np = kmem_zalloc(sizeof (drmach_node_t), KM_SLEEP);
410 
411 	np->get_dnode = drmach_node_obp_get_dnode;
412 	np->walk = drmach_node_obp_walk;
413 
414 	return (np);
415 }
416 
417 static void
418 drmach_node_dispose(drmach_node_t *np)
419 {
420 	kmem_free(np, sizeof (*np));
421 }
422 
423 static dev_info_t *
424 drmach_node_get_dip(drmach_node_t *np)
425 {
426 	dnode_t nodeid;
427 
428 	nodeid = np->get_dnode(np);
429 	if (nodeid == OBP_NONODE)
430 		return (NULL);
431 	else {
432 		dev_info_t *dip;
433 
434 		/* The root node doesn't have to be held */
435 		dip = e_ddi_nodeid_to_dip(nodeid);
436 		if (dip) {
437 			/*
438 			 * Branch rooted at dip is already held, so release
439 			 * hold acquired in e_ddi_nodeid_to_dip()
440 			 */
441 			ddi_release_devi(dip);
442 			ASSERT(e_ddi_branch_held(dip));
443 		}
444 
445 		return (dip);
446 	}
447 	/*NOTREACHED*/
448 }
449 
450 static dnode_t
451 drmach_node_get_dnode(drmach_node_t *np)
452 {
453 	return (np->get_dnode(np));
454 }
455 
456 static int
457 drmach_node_walk(drmach_node_t *np, void *param,
458 		int (*cb)(drmach_node_walk_args_t *args))
459 {
460 	return (np->walk(np, param, cb));
461 }
462 
463 static int
464 drmach_node_get_prop(drmach_node_t *np, char *name, void *buf)
465 {
466 	dnode_t	nodeid;
467 	int	rv;
468 
469 	nodeid = np->get_dnode(np);
470 	if (nodeid == OBP_NONODE)
471 		rv = -1;
472 	else if (prom_getproplen(nodeid, (caddr_t)name) < 0)
473 		rv = -1;
474 	else {
475 		(void) prom_getprop(nodeid, (caddr_t)name, (caddr_t)buf);
476 		rv = 0;
477 	}
478 
479 	return (rv);
480 }
481 
482 static int
483 drmach_node_get_proplen(drmach_node_t *np, char *name, int *len)
484 {
485 	dnode_t	 nodeid;
486 	int	 rv;
487 
488 	nodeid = np->get_dnode(np);
489 	if (nodeid == OBP_NONODE)
490 		rv = -1;
491 	else {
492 		*len = prom_getproplen(nodeid, (caddr_t)name);
493 		rv = (*len < 0 ? -1 : 0);
494 	}
495 
496 	return (rv);
497 }
498 
499 static drmachid_t
500 drmach_node_dup(drmach_node_t *np)
501 {
502 	drmach_node_t *dup;
503 
504 	dup = drmach_node_new();
505 	dup->here = np->here;
506 
507 	return (dup);
508 }
509 
510 /*
511  * drmach_array provides convenient array construction, access,
512  * bounds checking and array destruction logic.
513  */
514 
515 static drmach_array_t *
516 drmach_array_new(int min_index, int max_index)
517 {
518 	drmach_array_t *arr;
519 
520 	arr = kmem_zalloc(sizeof (drmach_array_t), KM_SLEEP);
521 
522 	arr->arr_sz = (max_index - min_index + 1) * sizeof (void *);
523 	if (arr->arr_sz > 0) {
524 		arr->min_index = min_index;
525 		arr->max_index = max_index;
526 
527 		arr->arr = kmem_zalloc(arr->arr_sz, KM_SLEEP);
528 		return (arr);
529 	} else {
530 		kmem_free(arr, sizeof (*arr));
531 		return (0);
532 	}
533 }
534 
535 static int
536 drmach_array_set(drmach_array_t *arr, int idx, drmachid_t val)
537 {
538 	if (idx < arr->min_index || idx > arr->max_index)
539 		return (-1);
540 	else {
541 		arr->arr[idx - arr->min_index] = val;
542 		return (0);
543 	}
544 	/*NOTREACHED*/
545 }
546 
547 static int
548 drmach_array_get(drmach_array_t *arr, int idx, drmachid_t *val)
549 {
550 	if (idx < arr->min_index || idx > arr->max_index)
551 		return (-1);
552 	else {
553 		*val = arr->arr[idx - arr->min_index];
554 		return (0);
555 	}
556 	/*NOTREACHED*/
557 }
558 
559 static int
560 drmach_array_first(drmach_array_t *arr, int *idx, drmachid_t *val)
561 {
562 	int rv;
563 
564 	*idx = arr->min_index;
565 	while ((rv = drmach_array_get(arr, *idx, val)) == 0 && *val == NULL)
566 		*idx += 1;
567 
568 	return (rv);
569 }
570 
571 static int
572 drmach_array_next(drmach_array_t *arr, int *idx, drmachid_t *val)
573 {
574 	int rv;
575 
576 	*idx += 1;
577 	while ((rv = drmach_array_get(arr, *idx, val)) == 0 && *val == NULL)
578 		*idx += 1;
579 
580 	return (rv);
581 }
582 
583 static void
584 drmach_array_dispose(drmach_array_t *arr, void (*disposer)(drmachid_t))
585 {
586 	drmachid_t	val;
587 	int		idx;
588 	int		rv;
589 
590 	rv = drmach_array_first(arr, &idx, &val);
591 	while (rv == 0) {
592 		(*disposer)(val);
593 		rv = drmach_array_next(arr, &idx, &val);
594 	}
595 
596 	kmem_free(arr->arr, arr->arr_sz);
597 	kmem_free(arr, sizeof (*arr));
598 }
599 
600 /*ARGSUSED*/
601 static int
602 drmach_prom_select(dnode_t nodeid, void *arg, uint_t flags)
603 {
604 	int			rprop[64];
605 	dnode_t			saved;
606 	drmach_config_args_t	*ap = (drmach_config_args_t *)arg;
607 	drmach_device_t		*dp = ap->dp;
608 	sbd_error_t		*err;
609 
610 	saved = drmach_node_get_dnode(dp->node);
611 
612 	if (nodeid != saved)
613 		return (DDI_FAILURE);
614 
615 	if (saved == OBP_NONODE) {
616 		err = DRMACH_INTERNAL_ERROR();
617 		DRERR_SET_C(&ap->err, &err);
618 		return (DDI_FAILURE);
619 	}
620 
621 	if (prom_getprop(nodeid, OBP_REG, (caddr_t)rprop) <= 0) {
622 		return (DDI_FAILURE);
623 	}
624 
625 	return (DDI_SUCCESS);
626 }
627 
628 /*ARGSUSED*/
629 static void
630 drmach_branch_callback(dev_info_t *rdip, void *arg, uint_t flags)
631 {
632 	drmach_config_args_t	*ap = (drmach_config_args_t *)arg;
633 
634 	ASSERT(ap->dip == NULL);
635 
636 	ap->dip = rdip;
637 }
638 
639 sbd_error_t *
640 drmach_configure(drmachid_t id, int flags)
641 {
642 	drmach_device_t		*dp;
643 	sbd_error_t		*err;
644 	drmach_config_args_t	ca;
645 	devi_branch_t		b = {0};
646 	dev_info_t		*fdip = NULL;
647 
648 	if (!DRMACH_IS_DEVICE_ID(id))
649 		return (drerr_new(0, ESTF_INAPPROP, NULL));
650 	dp = id;
651 
652 	ca.dp = dp;
653 	ca.flags = flags;
654 	ca.err = NULL;		/* will be set if error detected */
655 	ca.dip = NULL;
656 
657 	b.arg = &ca;
658 	b.type = DEVI_BRANCH_PROM;
659 	b.create.prom_branch_select = drmach_prom_select;
660 	b.devi_branch_callback = drmach_branch_callback;
661 
662 	if (e_ddi_branch_create(ddi_root_node(), &b, &fdip,
663 	    DEVI_BRANCH_CHILD | DEVI_BRANCH_CONFIGURE) != 0) {
664 		char *path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
665 
666 		/*
667 		 * If non-NULL, fdip is returned held and must be released.
668 		 */
669 		if (fdip != NULL) {
670 			(void) ddi_pathname(fdip, path);
671 			ddi_release_devi(fdip);
672 		} else if (ca.dip != NULL) {
673 			/* safe to call ddi_pathname as dip already held */
674 			(void) ddi_pathname(ca.dip, path);
675 		} else {
676 			(void) strcpy(path, "<none>");
677 		}
678 
679 		err = drerr_new(1, ESTF_DRVFAIL, path);
680 		DRERR_SET_C(&ca.err, &err);
681 		kmem_free(path, MAXPATHLEN);
682 	}
683 
684 	return (ca.err);
685 }
686 
687 static sbd_error_t *
688 drmach_device_new(drmach_node_t *node,
689 	drmach_board_t *bp, drmach_device_t **dpp)
690 {
691 	int		 i;
692 	int		 rv;
693 	drmach_device_t	*dp;
694 	sbd_error_t	*err;
695 	char		 name[OBP_MAXDRVNAME];
696 
697 	rv = drmach_node_get_prop(node, OBP_NAME, name);
698 	if (rv) {
699 		/* every node is expected to have a name */
700 		err = drerr_new(1, ESTF_GETPROP,
701 			"PROM Node 0x%x: property %s",
702 			(uint_t)node->get_dnode(node), OBP_NAME);
703 
704 		return (err);
705 	}
706 
707 	/*
708 	 * The node currently being examined is not listed in the name2type[]
709 	 * array.  In this case, the node is no interest to drmach.  Both
710 	 * dp and err are initialized here to yield nothing (no device or
711 	 * error structure) for this case.
712 	 */
713 	for (i = 0; i < sizeof (name2type) / sizeof (name2type[0]); i++)
714 		if (strcmp(name2type[i].name, name) == 0)
715 			break;
716 
717 	if (i < sizeof (name2type) / sizeof (name2type[0])) {
718 		dp = kmem_zalloc(sizeof (drmach_device_t), KM_SLEEP);
719 
720 		dp->bp = bp;
721 		dp->unum = -1;
722 		dp->node = drmach_node_dup(node);
723 		dp->type = name2type[i].type;
724 
725 		err = (name2type[i].new)(dp);
726 		if (err) {
727 			drmach_node_dispose(node);
728 			kmem_free(dp, sizeof (*dp));
729 			dp = NULL;
730 		}
731 
732 		*dpp = dp;
733 		return (err);
734 	}
735 
736 	/*
737 	 * The node currently being examined is not listed in the name2type[]
738 	 * array.  In this case, the node is no interest to drmach.  Both
739 	 * dp and err are initialized here to yield nothing (no device or
740 	 * error structure) for this case.
741 	 */
742 	*dpp = NULL;
743 	return (NULL);
744 }
745 
746 static void
747 drmach_device_dispose(drmachid_t id)
748 {
749 	drmach_device_t *self = id;
750 
751 	if (self->node)
752 		drmach_node_dispose(self->node);
753 
754 	kmem_free(self, sizeof (*self));
755 }
756 
757 static sbd_error_t *
758 drmach_device_get_prop(drmach_device_t *dp, char *name, void *buf)
759 {
760 	sbd_error_t	*err = NULL;
761 	int		 rv;
762 
763 	rv = drmach_node_get_prop(dp->node, name, buf);
764 	if (rv) {
765 		err = drerr_new(1, ESTF_GETPROP,
766 			"%s::%s: property %s",
767 			dp->bp->cm.name, dp->cm.name, name);
768 	}
769 
770 	return (err);
771 }
772 
773 static sbd_error_t *
774 drmach_device_get_proplen(drmach_device_t *dp, char *name, int *len)
775 {
776 	sbd_error_t	*err = NULL;
777 	int		 rv;
778 
779 	rv = drmach_node_get_proplen(dp->node, name, len);
780 	if (rv) {
781 		err = drerr_new(1, ESTF_GETPROPLEN,
782 			"%s::%s: property %s",
783 			dp->bp->cm.name, dp->cm.name, name);
784 	}
785 
786 	return (err);
787 }
788 
789 static drmach_board_t *
790 drmach_board_new(int bnum)
791 {
792 	static sbd_error_t *drmach_board_release(drmachid_t);
793 	static sbd_error_t *drmach_board_status(drmachid_t, drmach_status_t *);
794 
795 	drmach_board_t	*bp;
796 
797 	bp = kmem_zalloc(sizeof (drmach_board_t), KM_SLEEP);
798 
799 	bp->cm.isa = (void *)drmach_board_new;
800 	bp->cm.release = drmach_board_release;
801 	bp->cm.status = drmach_board_status;
802 
803 	(void) drmach_board_name(bnum, bp->cm.name, sizeof (bp->cm.name));
804 
805 	bp->bnum = bnum;
806 	bp->devices = NULL;
807 	bp->connect_cpuid = -1;
808 	bp->tree = drmach_node_new();
809 	bp->assigned = !drmach_initialized;
810 	bp->powered = !drmach_initialized;
811 
812 	drmach_array_set(drmach_boards, bnum, bp);
813 	return (bp);
814 }
815 
816 static void
817 drmach_board_dispose(drmachid_t id)
818 {
819 	drmach_board_t *bp;
820 
821 	ASSERT(DRMACH_IS_BOARD_ID(id));
822 	bp = id;
823 
824 	if (bp->tree)
825 		drmach_node_dispose(bp->tree);
826 
827 	if (bp->devices)
828 		drmach_array_dispose(bp->devices, drmach_device_dispose);
829 
830 	kmem_free(bp, sizeof (*bp));
831 }
832 
833 static sbd_error_t *
834 drmach_board_status(drmachid_t id, drmach_status_t *stat)
835 {
836 	sbd_error_t	*err = NULL;
837 	drmach_board_t	*bp;
838 
839 	if (!DRMACH_IS_BOARD_ID(id))
840 		return (drerr_new(0, ESTF_INAPPROP, NULL));
841 	bp = id;
842 
843 	stat->assigned = bp->assigned;
844 	stat->powered = bp->powered;
845 	stat->busy = 0;			/* assume not busy */
846 	stat->configured = 0;		/* assume not configured */
847 	stat->empty = 0;
848 	stat->cond = bp->cond = SBD_COND_OK;
849 	strncpy(stat->type, "System Brd", sizeof (stat->type));
850 	stat->info[0] = '\0';
851 
852 	if (bp->devices) {
853 		int		 rv;
854 		int		 d_idx;
855 		drmachid_t	 d_id;
856 
857 		rv = drmach_array_first(bp->devices, &d_idx, &d_id);
858 		while (rv == 0) {
859 			drmach_status_t	d_stat;
860 
861 			err = drmach_status(d_id, &d_stat);
862 			if (err)
863 				break;
864 
865 			stat->busy |= d_stat.busy;
866 			stat->configured |= d_stat.configured;
867 
868 			rv = drmach_array_next(bp->devices, &d_idx, &d_id);
869 		}
870 	}
871 
872 	return (err);
873 }
874 
875 /* a simple routine to reduce redundancy of this common logic */
876 static pda_handle_t
877 drmach_pda_open(void)
878 {
879 	pda_handle_t ph;
880 
881 	ph = pda_open();
882 	if (ph == NULL) {
883 		/* catch in debug kernels */
884 		ASSERT(0);
885 		cmn_err(CE_WARN, "pda_open failed");
886 	}
887 
888 	return (ph);
889 }
890 
891 #ifdef DEBUG
892 int drmach_init_break = 0;
893 #endif
894 
895 static int
896 hold_rele_branch(dev_info_t *rdip, void *arg)
897 {
898 	int	i;
899 	int	*holdp = (int *)arg;
900 	char	*name = ddi_node_name(rdip);
901 
902 	/*
903 	 * For Starfire, we must be children of the root devinfo node
904 	 */
905 	ASSERT(ddi_get_parent(rdip) == ddi_root_node());
906 
907 	for (i = 0; i < sizeof (name2type) / sizeof (name2type[0]); i++)
908 		if (strcmp(name2type[i].name, name) == 0)
909 			break;
910 
911 	if (i == sizeof (name2type) / sizeof (name2type[0])) {
912 		/* Not of interest to us */
913 		return (DDI_WALK_PRUNECHILD);
914 	}
915 
916 	if (*holdp) {
917 		ASSERT(!e_ddi_branch_held(rdip));
918 		e_ddi_branch_hold(rdip);
919 	} else {
920 		ASSERT(e_ddi_branch_held(rdip));
921 		e_ddi_branch_rele(rdip);
922 	}
923 
924 	return (DDI_WALK_PRUNECHILD);
925 }
926 
927 static int
928 drmach_init(void)
929 {
930 	dnode_t		nodeid;
931 	dev_info_t	*rdip;
932 	int		hold, circ;
933 
934 #ifdef DEBUG
935 	if (drmach_init_break)
936 		debug_enter("drmach_init: drmach_init_break set\n");
937 #endif
938 	mutex_enter(&drmach_i_lock);
939 	if (drmach_initialized) {
940 		mutex_exit(&drmach_i_lock);
941 		return (0);
942 	}
943 
944 	drmach_boards = drmach_array_new(0, MAX_BOARDS - 1);
945 
946 	nodeid = prom_childnode(prom_rootnode());
947 	do {
948 		int		 bnum;
949 		drmachid_t	 id;
950 
951 		bnum = -1;
952 		(void) prom_getprop(nodeid, OBP_BOARDNUM, (caddr_t)&bnum);
953 		if (bnum == -1)
954 			continue;
955 
956 		if (drmach_array_get(drmach_boards, bnum, &id) == -1) {
957 			cmn_err(CE_WARN, "OBP node 0x%x has"
958 				" invalid property value, %s=%d",
959 				nodeid, OBP_BOARDNUM, bnum);
960 
961 			/* clean up */
962 			drmach_array_dispose(
963 				drmach_boards, drmach_board_dispose);
964 
965 			mutex_exit(&drmach_i_lock);
966 			return (-1);
967 		} else if (id == NULL)
968 			(void) drmach_board_new(bnum);
969 	} while ((nodeid = prom_nextnode(nodeid)) != OBP_NONODE);
970 
971 	drmach_shutdown_va = vmem_alloc(heap_arena, PAGESIZE, VM_SLEEP);
972 
973 	/*
974 	 * Walk immediate children of devinfo root node and hold
975 	 * all devinfo branches of interest.
976 	 */
977 	hold = 1;
978 	rdip = ddi_root_node();
979 
980 	ndi_devi_enter(rdip, &circ);
981 	ddi_walk_devs(ddi_get_child(rdip), hold_rele_branch, &hold);
982 	ndi_devi_exit(rdip, circ);
983 
984 	drmach_initialized = 1;
985 
986 	mutex_exit(&drmach_i_lock);
987 
988 	return (0);
989 }
990 
991 static int
992 drmach_fini(void)
993 {
994 	dev_info_t	*rdip;
995 	int		hold, circ;
996 
997 	if (drmach_initialized) {
998 		int		busy = 0;
999 		int		rv;
1000 		int		idx;
1001 		drmachid_t	id;
1002 
1003 		ASSERT(drmach_boards != NULL);
1004 
1005 		rv = drmach_array_first(drmach_boards, &idx, &id);
1006 		while (rv == 0) {
1007 			sbd_error_t	*err;
1008 			drmach_status_t stat;
1009 
1010 			err = drmach_board_status(id, &stat);
1011 			if (err) {
1012 				/* catch in debug kernels */
1013 				ASSERT(0);
1014 				sbd_err_clear(&err);
1015 				busy = 1;
1016 			} else
1017 				busy |= stat.busy;
1018 
1019 			rv = drmach_array_next(drmach_boards, &idx, &id);
1020 		}
1021 
1022 		if (busy)
1023 			return (-1);
1024 
1025 		drmach_array_dispose(drmach_boards, drmach_board_dispose);
1026 		drmach_boards = NULL;
1027 
1028 		vmem_free(heap_arena, drmach_shutdown_va, PAGESIZE);
1029 
1030 		/*
1031 		 * Walk immediate children of the root devinfo node
1032 		 * releasing holds acquired on branches in drmach_init()
1033 		 */
1034 		hold = 0;
1035 		rdip = ddi_root_node();
1036 
1037 		ndi_devi_enter(rdip, &circ);
1038 		ddi_walk_devs(ddi_get_child(rdip), hold_rele_branch, &hold);
1039 		ndi_devi_exit(rdip, circ);
1040 
1041 		mutex_destroy(&drmach_i_lock);
1042 
1043 		drmach_initialized = 0;
1044 	}
1045 	if (drmach_xt_mb != NULL) {
1046 		vmem_free(static_alloc_arena, (void *)drmach_xt_mb,
1047 		    NCPU * sizeof (uchar_t));
1048 	}
1049 	if (drmach_shutdown_asm_mbox != NULL) {
1050 		vmem_free(static_alloc_arena, (void *)drmach_shutdown_asm_mbox,
1051 		    sizeof (struct drmach_shutdown_mbox));
1052 	}
1053 	return (0);
1054 }
1055 
1056 static sbd_error_t *
1057 drmach_get_mc_asr_addr(drmachid_t id, uint64_t *pa)
1058 {
1059 	drmach_device_t	*dp;
1060 	dnode_t		nodeid;
1061 	uint64_t	addr;
1062 
1063 	if (!DRMACH_IS_MEM_ID(id))
1064 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1065 	dp = id;
1066 
1067 	nodeid = drmach_node_get_dnode(dp->node);
1068 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1069 		return (DRMACH_INTERNAL_ERROR());
1070 
1071 	addr = mc_get_asr_addr(nodeid);
1072 	if (addr == (uint64_t)-1)
1073 		return (DRMACH_INTERNAL_ERROR());
1074 
1075 	*pa = addr;
1076 	return (NULL);
1077 }
1078 
1079 static sbd_error_t *
1080 drmach_get_mc_idle_addr(drmachid_t id, uint64_t *pa)
1081 {
1082 	drmach_device_t	*dp;
1083 	dnode_t		nodeid;
1084 	uint64_t	addr;
1085 
1086 	if (!DRMACH_IS_MEM_ID(id))
1087 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1088 	dp = id;
1089 
1090 	nodeid = drmach_node_get_dnode(dp->node);
1091 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1092 		return (DRMACH_INTERNAL_ERROR());
1093 
1094 	addr = mc_get_idle_addr(nodeid);
1095 	if (addr == (uint64_t)-1)
1096 		return (DRMACH_INTERNAL_ERROR());
1097 
1098 	*pa = addr;
1099 	return (NULL);
1100 }
1101 
1102 static sbd_error_t *
1103 drmach_read_mc_asr(drmachid_t id, uint_t *mcregp)
1104 {
1105 	drmach_device_t	*dp;
1106 	dnode_t		 nodeid;
1107 	sbd_error_t	*err;
1108 
1109 	if (!DRMACH_IS_MEM_ID(id))
1110 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1111 	dp = id;
1112 
1113 	nodeid = drmach_node_get_dnode(dp->node);
1114 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1115 		err = DRMACH_INTERNAL_ERROR();
1116 	else if (mc_read_asr(nodeid, mcregp) == -1)
1117 		err = DRMACH_INTERNAL_ERROR();
1118 	else
1119 		err = NULL;
1120 
1121 	return (err);
1122 }
1123 
1124 static sbd_error_t *
1125 drmach_write_mc_asr(drmachid_t id, uint_t mcreg)
1126 {
1127 	drmach_device_t	*dp;
1128 	dnode_t		 nodeid;
1129 	sbd_error_t	*err;
1130 
1131 	if (!DRMACH_IS_MEM_ID(id))
1132 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1133 	dp = id;
1134 
1135 	nodeid = drmach_node_get_dnode(dp->node);
1136 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1137 		err = DRMACH_INTERNAL_ERROR();
1138 	else if (mc_write_asr(nodeid, mcreg) == -1)
1139 		err = DRMACH_INTERNAL_ERROR();
1140 	else
1141 		err = NULL;
1142 
1143 	return (err);
1144 }
1145 
1146 static struct memlist *
1147 memlist_add_span(struct memlist *mlist, uint64_t base, uint64_t len)
1148 {
1149 	struct memlist	*ml, *tl, *nl;
1150 
1151 	if (len == 0ull)
1152 		return (NULL);
1153 
1154 	if (mlist == NULL) {
1155 		mlist = GETSTRUCT(struct memlist, 1);
1156 		mlist->address = base;
1157 		mlist->size = len;
1158 		mlist->next = mlist->prev = NULL;
1159 
1160 		return (mlist);
1161 	}
1162 
1163 	for (tl = ml = mlist; ml; tl = ml, ml = ml->next) {
1164 		if (base < ml->address) {
1165 			if ((base + len) < ml->address) {
1166 				nl = GETSTRUCT(struct memlist, 1);
1167 				nl->address = base;
1168 				nl->size = len;
1169 				nl->next = ml;
1170 				if ((nl->prev = ml->prev) != NULL)
1171 					nl->prev->next = nl;
1172 				ml->prev = nl;
1173 				if (mlist == ml)
1174 					mlist = nl;
1175 			} else {
1176 				ml->size = MAX((base + len),
1177 						(ml->address + ml->size)) -
1178 						base;
1179 				ml->address = base;
1180 			}
1181 			break;
1182 
1183 		} else if (base <= (ml->address + ml->size)) {
1184 			ml->size = MAX((base + len),
1185 					(ml->address + ml->size)) -
1186 					MIN(ml->address, base);
1187 			ml->address = MIN(ml->address, base);
1188 			break;
1189 		}
1190 	}
1191 	if (ml == NULL) {
1192 		nl = GETSTRUCT(struct memlist, 1);
1193 		nl->address = base;
1194 		nl->size = len;
1195 		nl->next = NULL;
1196 		nl->prev = tl;
1197 		tl->next = nl;
1198 	}
1199 
1200 	memlist_coalesce(mlist);
1201 
1202 	return (mlist);
1203 }
1204 
1205 static sbd_error_t *
1206 drmach_prep_rename_script(drmach_device_t *s_mem, drmach_device_t *t_mem,
1207 	uint64_t t_slice_offset, caddr_t buf, int buflen)
1208 {
1209 	int			i, b, m;
1210 	drmach_mc_idle_script_t	*isp;
1211 	drmach_rename_script_t	*rsp;
1212 	int			s_bd, t_bd;
1213 	uint_t			s_masr, t_masr;
1214 	uint64_t		s_new_basepa, t_new_basepa;
1215 	int			b_idx, rv;
1216 	sbd_error_t		*err;
1217 	drmachid_t		 b_id;
1218 	drmach_board_t		*brd;
1219 
1220 #ifdef DEBUG
1221 	/*
1222 	 * Starfire CPU/MEM/IO boards have only one MC per board.
1223 	 * This function has been coded with that fact in mind.
1224 	 */
1225 	ASSERT(MAX_MEM_UNITS_PER_BOARD == 1);
1226 
1227 	/*
1228 	 * calculate the maximum space that could be consumed,
1229 	 * then verify the available buffer space is adequate.
1230 	 */
1231 	m  = sizeof (drmach_mc_idle_script_t *) * 2; /* two MCs */
1232 	b  = sizeof (drmach_rename_script_t *) * 3 * MAX_CPU_UNITS_PER_BOARD;
1233 	b += sizeof (drmach_rename_script_t *) * 3 * MAX_IO_UNITS_PER_BOARD;
1234 	b *= MAX_BOARDS;
1235 	b += sizeof (drmach_rename_script_t *) * 3;
1236 	b += sizeof (drmach_rename_script_t *) * 1;
1237 	ASSERT(m + b < buflen);
1238 #endif
1239 
1240 	/*
1241 	 * construct an array of MC idle register addresses of
1242 	 * both MCs.  The array is zero terminated -- as expected
1243 	 * by drmach_copy_rename_prog__relocatable().
1244 	 */
1245 	isp = (drmach_mc_idle_script_t *)buf;
1246 
1247 	/* source mc */
1248 	err = drmach_get_mc_idle_addr(s_mem, &isp->idle_addr);
1249 	if (err)
1250 		return (err);
1251 	isp->mem = s_mem;
1252 	isp += 1;
1253 
1254 	/* target mc */
1255 	err = drmach_get_mc_idle_addr(t_mem, &isp->idle_addr);
1256 	if (err)
1257 		return (err);
1258 	isp->mem = t_mem;
1259 	isp += 1;
1260 
1261 	/* terminator */
1262 	isp->idle_addr = 0;
1263 	isp->mem = NULL;
1264 	isp += 1;
1265 
1266 	/* fetch source mc asr register value */
1267 	err = drmach_read_mc_asr(s_mem, &s_masr);
1268 	if (err)
1269 		return (err);
1270 	else if (s_masr & STARFIRE_MC_INTERLEAVE_MASK) {
1271 		return (drerr_new(1, ESTF_INTERBOARD, "%s::%s",
1272 				s_mem->bp->cm.name, s_mem->cm.name));
1273 	}
1274 
1275 	/* fetch target mc asr register value */
1276 	err = drmach_read_mc_asr(t_mem, &t_masr);
1277 	if (err)
1278 		return (err);
1279 	else if (t_masr & STARFIRE_MC_INTERLEAVE_MASK) {
1280 		return (drerr_new(1, ESTF_INTERBOARD, "%s::%s",
1281 				t_mem->bp->cm.name, t_mem->cm.name));
1282 	}
1283 
1284 	/* get new source base pa from target's masr */
1285 	s_new_basepa = mc_asr_to_pa(t_masr);
1286 
1287 	/*
1288 	 * remove any existing slice offset to realign
1289 	 * memory with board's slice boundary
1290 	 */
1291 	s_new_basepa &= ~ (mc_get_mem_alignment() - 1);
1292 
1293 	/* get new target base pa from source's masr */
1294 	t_new_basepa  = mc_asr_to_pa(s_masr);
1295 
1296 	/* remove any existing slice offset, then apply new offset */
1297 	t_new_basepa &= ~ (mc_get_mem_alignment() - 1);
1298 	t_new_basepa += t_slice_offset;
1299 
1300 	/* encode new base pa into s_masr.  turn off mem present bit */
1301 	s_masr  = mc_pa_to_asr(s_masr, s_new_basepa);
1302 	s_masr &= ~STARFIRE_MC_MEM_PRESENT_MASK;
1303 
1304 	/* encode new base pa into t_masr.  turn on mem present bit */
1305 	t_masr  = mc_pa_to_asr(t_masr, t_new_basepa);
1306 	t_masr |= STARFIRE_MC_MEM_PRESENT_MASK;
1307 
1308 	/*
1309 	 * Step 0:	Mark source memory as not present.
1310 	 */
1311 	m = 0;
1312 	rsp = (drmach_rename_script_t *)isp;
1313 	err = drmach_get_mc_asr_addr(s_mem, &rsp[m].masr_addr);
1314 	if (err)
1315 		return (err);
1316 	rsp[m].masr = s_masr;
1317 	m++;
1318 
1319 	/*
1320 	 * Step 1:	Write source base address to target MC
1321 	 *		with present bit off.
1322 	 */
1323 	err = drmach_get_mc_asr_addr(t_mem, &rsp[m].masr_addr);
1324 	if (err)
1325 		return (err);
1326 	rsp[m].masr = t_masr & ~STARFIRE_MC_MEM_PRESENT_MASK;
1327 	m++;
1328 
1329 	/*
1330 	 * Step 2:	Now rewrite target reg with present bit on.
1331 	 */
1332 	rsp[m].masr_addr = rsp[m-1].masr_addr;
1333 	rsp[m].masr = t_masr;
1334 	m++;
1335 
1336 	s_bd = s_mem->bp->bnum;
1337 	t_bd = t_mem->bp->bnum;
1338 
1339 	DRMACH_PR("preparing script for CPU and IO units:\n");
1340 
1341 	rv = drmach_array_first(drmach_boards, &b_idx, &b_id);
1342 	if (rv) {
1343 		/* catch this in debug kernels */
1344 		ASSERT(0);
1345 		return (DRMACH_INTERNAL_ERROR());
1346 	}
1347 
1348 	do {
1349 		int			 d_idx;
1350 		drmachid_t		 d_id;
1351 		drmach_device_t		*device;
1352 
1353 		ASSERT(DRMACH_IS_BOARD_ID(b_id));
1354 		brd = b_id;
1355 		b = brd->bnum;
1356 
1357 		/*
1358 		 * Step 3:	Update PC MADR tables for CPUs.
1359 		 */
1360 		rv = drmach_array_first(brd->devices, &d_idx, &d_id);
1361 		if (rv) {
1362 			/* must mean no devices on this board */
1363 			break;
1364 		}
1365 
1366 		DRMACH_PR("\t%s\n", brd->cm.name);
1367 
1368 		do {
1369 			ASSERT(DRMACH_IS_DEVICE_ID(d_id));
1370 
1371 			if (!DRMACH_IS_CPU_ID(d_id))
1372 				continue;
1373 
1374 			device = d_id;
1375 			i = device->unum;
1376 
1377 			DRMACH_PR("\t\t%s\n", device->cm.name);
1378 
1379 			/*
1380 			 * Disabled detaching mem node.
1381 			 */
1382 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, s_bd, i);
1383 			rsp[m].masr = s_masr;
1384 			m++;
1385 			/*
1386 			 * Always write masr with present bit
1387 			 * off and then again with it on.
1388 			 */
1389 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, t_bd, i);
1390 			rsp[m].masr = t_masr & ~STARFIRE_MC_MEM_PRESENT_MASK;
1391 			m++;
1392 			rsp[m].masr_addr = rsp[m-1].masr_addr;
1393 			rsp[m].masr = t_masr;
1394 			m++;
1395 
1396 		} while (drmach_array_next(brd->devices, &d_idx, &d_id) == 0);
1397 
1398 		/*
1399 		 * Step 4:	Update PC MADR tables for IOs.
1400 		 */
1401 		rv = drmach_array_first(brd->devices, &d_idx, &d_id);
1402 		/* this worked for previous loop, must work here too */
1403 		ASSERT(rv == 0);
1404 
1405 		do {
1406 			ASSERT(DRMACH_IS_DEVICE_ID(d_id));
1407 
1408 			if (!DRMACH_IS_IO_ID(d_id))
1409 				continue;
1410 
1411 			device = d_id;
1412 			i = device->unum;
1413 
1414 			DRMACH_PR("\t\t%s\n", device->cm.name);
1415 
1416 			/*
1417 			 * Disabled detaching mem node.
1418 			 */
1419 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, s_bd, i+4);
1420 			rsp[m].masr = s_masr;
1421 			m++;
1422 			/*
1423 			 * Always write masr with present bit
1424 			 * off and then again with it on.
1425 			 */
1426 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, t_bd, i+4);
1427 			rsp[m].masr = t_masr & ~STARFIRE_MC_MEM_PRESENT_MASK;
1428 			m++;
1429 			rsp[m].masr_addr = rsp[m-1].masr_addr;
1430 			rsp[m].masr = t_masr;
1431 			m++;
1432 
1433 		} while (drmach_array_next(brd->devices, &d_idx, &d_id) == 0);
1434 	} while (drmach_array_next(drmach_boards, &b_idx, &b_id) == 0);
1435 
1436 	/*
1437 	 * Zero masr_addr value indicates the END.
1438 	 */
1439 	rsp[m].masr_addr = 0ull;
1440 	rsp[m].masr = 0;
1441 	DRMACH_PR("number of steps in rename script = %d\n", m);
1442 	m++;
1443 
1444 	/* paranoia */
1445 	ASSERT((caddr_t)&rsp[m] <= buf + buflen);
1446 
1447 #ifdef DEBUG
1448 	{
1449 		int	j;
1450 
1451 		DRMACH_PR("mc idle register address list:");
1452 		isp = (drmach_mc_idle_script_t *)buf;
1453 		DRMACH_PR("source mc idle addr 0x%llx, mem id %p",
1454 			isp[0].idle_addr, isp[0].mem);
1455 		DRMACH_PR("target mc idle addr 0x%llx, mem id %p",
1456 			isp[1].idle_addr, isp[1].mem);
1457 		ASSERT(isp[2].idle_addr == 0);
1458 
1459 		DRMACH_PR("copy-rename script:");
1460 		for (j = 0; j < m; j++) {
1461 			DRMACH_PR("0x%llx = 0x%08x",
1462 				rsp[j].masr_addr, rsp[j].masr);
1463 		}
1464 
1465 		DELAY(1000000);
1466 	}
1467 #endif
1468 
1469 	/* return number of bytes consumed */
1470 	b = (caddr_t)&rsp[m] - buf;
1471 	DRMACH_PR("total number of bytes consumed is %d\n", b);
1472 	ASSERT(b <= buflen);
1473 
1474 #ifdef lint
1475 	buflen = buflen;
1476 #endif
1477 
1478 	return (NULL);
1479 }
1480 
1481 /*
1482  * The routine performs the necessary memory COPY and MC adr SWITCH.
1483  * Both operations MUST be at the same "level" so that the stack is
1484  * maintained correctly between the copy and switch.  The switch
1485  * portion implements a caching mechanism to guarantee the code text
1486  * is cached prior to execution.  This is to guard against possible
1487  * memory access while the MC adr's are being modified.
1488  *
1489  * IMPORTANT: The _drmach_copy_rename_end() function must immediately
1490  * follow drmach_copy_rename_prog__relocatable() so that the correct
1491  * "length" of the drmach_copy_rename_prog__relocatable can be
1492  * calculated.  This routine MUST be a LEAF function, i.e. it can
1493  * make NO function calls, primarily for two reasons:
1494  *
1495  *	1. We must keep the stack consistent across the "switch".
1496  *	2. Function calls are compiled to relative offsets, and
1497  *	   we execute this function we'll be executing it from
1498  *	   a copied version in a different area of memory, thus
1499  *	   the relative offsets will be bogus.
1500  *
1501  * Moreover, it must have the "__relocatable" suffix to inform DTrace
1502  * providers (and anything else, for that matter) that this
1503  * function's text is manually relocated elsewhere before it is
1504  * executed.  That is, it cannot be safely instrumented with any
1505  * methodology that is PC-relative.
1506  */
1507 static void
1508 drmach_copy_rename_prog__relocatable(drmach_copy_rename_program_t *prog)
1509 {
1510 	extern void drmach_exec_script_il(drmach_rename_script_t *rsp);
1511 
1512 	drmach_mc_idle_script_t		*isp;
1513 	struct memlist			*ml;
1514 	int				csize;
1515 	int				lnsize;
1516 	uint64_t			caddr;
1517 
1518 	isp = (drmach_mc_idle_script_t *)prog->data;
1519 
1520 	caddr = ecache_flushaddr;
1521 	csize = (cpunodes[CPU->cpu_id].ecache_size << 1);
1522 	lnsize = cpunodes[CPU->cpu_id].ecache_linesize;
1523 
1524 	/*
1525 	 * DO COPY.
1526 	 */
1527 	for (ml = prog->c_ml; ml; ml = ml->next) {
1528 		uint64_t	s_pa, t_pa;
1529 		uint64_t	nbytes;
1530 
1531 		s_pa = prog->s_copybasepa + ml->address;
1532 		t_pa = prog->t_copybasepa + ml->address;
1533 		nbytes = ml->size;
1534 
1535 		while (nbytes != 0ull) {
1536 			/*
1537 			 * This copy does NOT use an ASI
1538 			 * that avoids the Ecache, therefore
1539 			 * the dst_pa addresses may remain
1540 			 * in our Ecache after the dst_pa
1541 			 * has been removed from the system.
1542 			 * A subsequent write-back to memory
1543 			 * will cause an ARB-stop because the
1544 			 * physical address no longer exists
1545 			 * in the system. Therefore we must
1546 			 * flush out local Ecache after we
1547 			 * finish the copy.
1548 			 */
1549 
1550 			/* copy 32 bytes at src_pa to dst_pa */
1551 			bcopy32_il(s_pa, t_pa);
1552 
1553 			/* increment by 32 bytes */
1554 			s_pa += (4 * sizeof (uint64_t));
1555 			t_pa += (4 * sizeof (uint64_t));
1556 
1557 			/* decrement by 32 bytes */
1558 			nbytes -= (4 * sizeof (uint64_t));
1559 		}
1560 	}
1561 
1562 	/*
1563 	 * Since bcopy32_il() does NOT use an ASI to bypass
1564 	 * the Ecache, we need to flush our Ecache after
1565 	 * the copy is complete.
1566 	 */
1567 	flush_ecache_il(caddr, csize, lnsize);		/* inline version */
1568 
1569 	/*
1570 	 * Wait for MCs to go idle.
1571 	 */
1572 	do {
1573 		register int	t = 10;
1574 		register uint_t	v;
1575 
1576 		/* loop t cycles waiting for each mc to indicate it's idle */
1577 		do {
1578 			v = ldphysio_il(isp->idle_addr)
1579 				& STARFIRE_MC_IDLE_MASK;
1580 
1581 		} while (v != STARFIRE_MC_IDLE_MASK && t-- > 0);
1582 
1583 		/* bailout if timedout */
1584 		if (t <= 0) {
1585 			prog->restless_mc = isp->mem;
1586 			return;
1587 		}
1588 
1589 		isp += 1;
1590 
1591 		/* stop if terminating zero has been reached */
1592 	} while (isp->idle_addr != 0);
1593 
1594 	/* advance passed terminating zero */
1595 	isp += 1;
1596 
1597 	/*
1598 	 * The following inline assembly routine caches
1599 	 * the rename script and then caches the code that
1600 	 * will do the rename.  This is necessary
1601 	 * so that we don't have any memory references during
1602 	 * the reprogramming.  We accomplish this by first
1603 	 * jumping through the code to guarantee it's cached
1604 	 * before we actually execute it.
1605 	 */
1606 	drmach_exec_script_il((drmach_rename_script_t *)isp);
1607 }
1608 
1609 static void
1610 drmach_copy_rename_end(void)
1611 {
1612 	/*
1613 	 * IMPORTANT:	This function's location MUST be located immediately
1614 	 *		following drmach_copy_rename_prog__relocatable to
1615 	 *		accurately estimate its size.  Note that this assumes
1616 	 *		the compiler keeps these functions in the order in
1617 	 *		which they appear :-o
1618 	 */
1619 }
1620 
1621 sbd_error_t *
1622 drmach_copy_rename_init(drmachid_t t_id, uint64_t t_slice_offset,
1623 	drmachid_t s_id, struct memlist *c_ml, drmachid_t *pgm_id)
1624 {
1625 	drmach_device_t	*s_mem;
1626 	drmach_device_t	*t_mem;
1627 	struct memlist	*x_ml;
1628 	uint64_t	off_mask, s_copybasepa, t_copybasepa, t_basepa;
1629 	int		len;
1630 	caddr_t		bp, wp;
1631 	pda_handle_t	ph;
1632 	sbd_error_t	*err;
1633 	drmach_copy_rename_program_t *prog;
1634 
1635 	if (!DRMACH_IS_MEM_ID(s_id))
1636 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1637 	if (!DRMACH_IS_MEM_ID(t_id))
1638 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1639 	s_mem = s_id;
1640 	t_mem = t_id;
1641 
1642 	/* get starting physical address of target memory */
1643 	err = drmach_mem_get_base_physaddr(t_id, &t_basepa);
1644 	if (err)
1645 		return (err);
1646 
1647 	/* calculate slice offset mask from slice size */
1648 	off_mask = mc_get_mem_alignment() - 1;
1649 
1650 	/* calculate source and target base pa */
1651 	s_copybasepa = c_ml->address;
1652 	t_copybasepa = t_basepa + ((c_ml->address & off_mask) - t_slice_offset);
1653 
1654 	/* paranoia */
1655 	ASSERT((c_ml->address & off_mask) >= t_slice_offset);
1656 
1657 	/* adjust copy memlist addresses to be relative to copy base pa */
1658 	x_ml = c_ml;
1659 	while (x_ml != NULL) {
1660 		x_ml->address -= s_copybasepa;
1661 		x_ml = x_ml->next;
1662 	}
1663 
1664 #ifdef DEBUG
1665 	{
1666 	uint64_t s_basepa, s_size, t_size;
1667 
1668 	x_ml = c_ml;
1669 	while (x_ml->next != NULL)
1670 		x_ml = x_ml->next;
1671 
1672 	DRMACH_PR("source copy span: base pa 0x%llx, end pa 0x%llx\n",
1673 		s_copybasepa,
1674 		s_copybasepa + x_ml->address + x_ml->size);
1675 
1676 	DRMACH_PR("target copy span: base pa 0x%llx, end pa 0x%llx\n",
1677 		t_copybasepa,
1678 		t_copybasepa + x_ml->address + x_ml->size);
1679 
1680 	DRMACH_PR("copy memlist (relative to copy base pa):\n");
1681 	MEMLIST_DUMP(c_ml);
1682 
1683 	err = drmach_mem_get_base_physaddr(s_id, &s_basepa);
1684 	ASSERT(err == NULL);
1685 
1686 	err = drmach_mem_get_size(s_id, &s_size);
1687 	ASSERT(err == NULL);
1688 
1689 	err = drmach_mem_get_size(t_id, &t_size);
1690 	ASSERT(err == NULL);
1691 
1692 	DRMACH_PR("current source base pa 0x%llx, size 0x%llx\n",
1693 		s_basepa, s_size);
1694 	DRMACH_PR("current target base pa 0x%llx, size 0x%llx\n",
1695 		t_basepa, t_size);
1696 
1697 	ASSERT(s_copybasepa + x_ml->address + x_ml->size <= s_basepa + s_size);
1698 	ASSERT(t_copybasepa + x_ml->address + x_ml->size <= t_basepa + t_size);
1699 	}
1700 #endif
1701 
1702 	ph = drmach_pda_open();
1703 	if (ph == NULL)
1704 		return (DRMACH_INTERNAL_ERROR());
1705 
1706 	/*
1707 	 * bp will be page aligned, since we're calling
1708 	 * kmem_zalloc() with an exact multiple of PAGESIZE.
1709 	 */
1710 	wp = bp = kmem_zalloc(PAGESIZE, KM_SLEEP);
1711 
1712 	/* allocate space for copy rename struct */
1713 	len = sizeof (drmach_copy_rename_program_t);
1714 	DRMACH_PR("prog = 0x%p, header len %d\n", wp, len);
1715 	prog = (drmach_copy_rename_program_t *)wp;
1716 	wp += (len + ecache_alignsize - 1) & ~ (ecache_alignsize - 1);
1717 
1718 	/*
1719 	 * Copy the code for the copy-rename routine into
1720 	 * a page aligned piece of memory.  We do this to guarantee
1721 	 * that we're executing within the same page and thus reduce
1722 	 * the possibility of cache collisions between different
1723 	 * pages.
1724 	 */
1725 	len = (int)((ulong_t)drmach_copy_rename_end -
1726 		    (ulong_t)drmach_copy_rename_prog__relocatable);
1727 	ASSERT(wp + len < bp + PAGESIZE);
1728 	bcopy((caddr_t)drmach_copy_rename_prog__relocatable, wp, len);
1729 
1730 	DRMACH_PR("copy-rename function 0x%p, len %d\n", wp, len);
1731 	prog->run = (void (*)())wp;
1732 	wp += (len + ecache_alignsize - 1) & ~ (ecache_alignsize - 1);
1733 
1734 	/*
1735 	 * Prepare data page that will contain script of
1736 	 * operations to perform during copy-rename.
1737 	 * Allocate temporary buffer to hold script.
1738 	 */
1739 	err = drmach_prep_rename_script(s_mem, t_mem, t_slice_offset,
1740 		wp, PAGESIZE - (wp - bp));
1741 	if (err) {
1742 		(void) drmach_copy_rename_fini(prog);
1743 		return (err);
1744 	}
1745 
1746 	DRMACH_PR("copy-rename script 0x%p, len %d\n", wp, len);
1747 	prog->data = wp;
1748 	wp += (len + ecache_alignsize - 1) & ~ (ecache_alignsize - 1);
1749 
1750 	prog->ph = ph;
1751 	prog->s_copybasepa = s_copybasepa;
1752 	prog->t_copybasepa = t_copybasepa;
1753 	prog->c_ml = c_ml;
1754 	*pgm_id = prog;
1755 
1756 	return (NULL);
1757 }
1758 
1759 sbd_error_t *
1760 drmach_copy_rename_fini(drmachid_t id)
1761 {
1762 	drmach_copy_rename_program_t	*prog = id;
1763 	sbd_error_t			*err = NULL;
1764 
1765 	if (prog->c_ml != NULL)
1766 		memlist_delete(prog->c_ml);
1767 
1768 	if (prog->ph != NULL)
1769 		pda_close(prog->ph);
1770 
1771 	if (prog->restless_mc != 0) {
1772 		cmn_err(CE_WARN, "MC did not idle; OBP Node 0x%x",
1773 			(uint_t)drmach_node_get_dnode(prog->restless_mc->node));
1774 
1775 		err = DRMACH_INTERNAL_ERROR();
1776 	}
1777 
1778 	kmem_free(prog, PAGESIZE);
1779 
1780 	return (err);
1781 }
1782 
1783 static sbd_error_t *
1784 drmach_io_new(drmach_device_t *dp)
1785 {
1786 	static sbd_error_t *drmach_io_release(drmachid_t);
1787 	static sbd_error_t *drmach_io_status(drmachid_t, drmach_status_t *);
1788 
1789 	sbd_error_t	*err;
1790 	int		 portid;
1791 
1792 	err = drmach_device_get_prop(dp, "upa-portid", &portid);
1793 	if (err == NULL) {
1794 		ASSERT(portid & 0x40);
1795 		dp->unum = portid & 1;
1796 	}
1797 
1798 	dp->cm.isa = (void *)drmach_io_new;
1799 	dp->cm.release = drmach_io_release;
1800 	dp->cm.status = drmach_io_status;
1801 
1802 	snprintf(dp->cm.name, sizeof (dp->cm.name), "%s%d", dp->type, dp->unum);
1803 
1804 	return (err);
1805 }
1806 
1807 static void
1808 drmach_iopc_op(pda_handle_t ph, drmach_iopc_op_t op)
1809 {
1810 	register int b;
1811 
1812 	for (b = 0; b < MAX_BOARDS; b++) {
1813 		int		p;
1814 		ushort_t	bda_ioc;
1815 		board_desc_t	*bdesc;
1816 
1817 		if (pda_board_present(ph, b) == 0)
1818 			continue;
1819 
1820 		bdesc = (board_desc_t *)pda_get_board_info(ph, b);
1821 		/*
1822 		 * Update PCs for IOCs.
1823 		 */
1824 		bda_ioc = bdesc->bda_ioc;
1825 		for (p = 0; p < MAX_IOCS; p++) {
1826 			u_longlong_t	idle_addr;
1827 			uchar_t		value;
1828 
1829 			if (BDA_NBL(bda_ioc, p) != BDAN_GOOD)
1830 				continue;
1831 
1832 			idle_addr = STARFIRE_BB_PC_ADDR(b, p, 1);
1833 
1834 			switch (op) {
1835 			case DO_PAUSE:
1836 				value = STARFIRE_BB_PC_PAUSE(p);
1837 				break;
1838 
1839 			case DO_IDLE:
1840 				value = STARFIRE_BB_PC_IDLE(p);
1841 				break;
1842 
1843 			case DO_UNPAUSE:
1844 				value = ldbphysio(idle_addr);
1845 				value &= ~STARFIRE_BB_PC_PAUSE(p);
1846 				break;
1847 
1848 			case DO_UNIDLE:
1849 				value = ldbphysio(idle_addr);
1850 				value &= ~STARFIRE_BB_PC_IDLE(p);
1851 				break;
1852 
1853 			default:
1854 				cmn_err(CE_PANIC,
1855 					"drmach_iopc_op: unknown op (%d)",
1856 					(int)op);
1857 				/*NOTREACHED*/
1858 			}
1859 			stbphysio(idle_addr, value);
1860 		}
1861 	}
1862 }
1863 
1864 void
1865 drmach_copy_rename(drmachid_t id)
1866 {
1867 	drmach_copy_rename_program_t	*prog = id;
1868 	uint64_t			neer;
1869 
1870 	/*
1871 	 * UPA IDLE
1872 	 * Protocol = PAUSE -> IDLE -> UNPAUSE
1873 	 * In reality since we only "idle" the IOPCs it's sufficient
1874 	 * to just issue the IDLE operation since (in theory) all IOPCs
1875 	 * in the field are PC6.  However, we'll be robust and do the
1876 	 * proper workaround protocol so that we never have to worry!
1877 	 */
1878 	drmach_iopc_op(prog->ph, DO_PAUSE);
1879 	drmach_iopc_op(prog->ph, DO_IDLE);
1880 	DELAY(100);
1881 	drmach_iopc_op(prog->ph, DO_UNPAUSE);
1882 	DELAY(100);
1883 
1884 	/* disable CE reporting */
1885 	neer = get_error_enable();
1886 	set_error_enable(neer & ~EER_CEEN);
1887 
1888 	/* run the copy/rename program */
1889 	prog->run(prog);
1890 
1891 	/* enable CE reporting */
1892 	set_error_enable(neer);
1893 
1894 	/*
1895 	 * UPA UNIDLE
1896 	 * Protocol = UNIDLE
1897 	 */
1898 	drmach_iopc_op(prog->ph, DO_UNIDLE);
1899 	DELAY(100);
1900 }
1901 
1902 /*
1903  * The counter-timer and perf-counter nodes are not being cleaned
1904  * up after a board that was present at start of day is detached.
1905  * If the board has become unconfigured with this operation, walk
1906  * the prom tree and find all counter-timer and perf-counter nodes
1907  * that have the same board number as the board that was just
1908  * unconfigured and remove them.
1909  */
1910 static sbd_error_t *
1911 drmach_remove_counter_nodes(drmachid_t id)
1912 {
1913 	int		num;
1914 	char		name[OBP_MAXDRVNAME];
1915 	dnode_t		child;
1916 	dev_info_t	*dip;
1917 	sbd_error_t	*err;
1918 	drmach_status_t	stat;
1919 	drmach_board_t	*bp;
1920 
1921 	if (!DRMACH_IS_BOARD_ID(id)) {
1922 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1923 	}
1924 
1925 	if ((err = drmach_board_status(id, &stat)) != NULL) {
1926 		return (err);
1927 	}
1928 
1929 	/*
1930 	 * Only clean up the counter-timer and perf-counter
1931 	 * nodes when the entire board is unconfigured.
1932 	 */
1933 	if (stat.configured) {
1934 		return (NULL);
1935 	}
1936 
1937 	bp = (drmach_board_t *)id;
1938 
1939 	err = NULL;
1940 
1941 	for (child = prom_childnode(prom_rootnode()); child != OBP_NONODE;
1942 	    child = prom_nextnode(child)) {
1943 
1944 		if (prom_getprop(child, OBP_BOARDNUM, (caddr_t)&num) == -1) {
1945 			continue;
1946 		}
1947 
1948 		if (bp->bnum != num) {
1949 			continue;
1950 		}
1951 
1952 		if (prom_getprop(child, OBP_NAME, (caddr_t)name) == -1) {
1953 			continue;
1954 		}
1955 
1956 		if (strncmp(name, MISC_COUNTER_TIMER_DEVNAME, OBP_MAXDRVNAME) &&
1957 		    strncmp(name, MISC_PERF_COUNTER_DEVNAME, OBP_MAXDRVNAME)) {
1958 				continue;
1959 		}
1960 
1961 		/* Root node doesn't have to be held */
1962 		dip = e_ddi_nodeid_to_dip(child);
1963 
1964 		/*
1965 		 * If the node is only in the OBP tree, then
1966 		 * we don't have to remove it.
1967 		 */
1968 		if (dip) {
1969 			dev_info_t *fdip = NULL;
1970 
1971 			DRMACH_PR("removing %s devinfo node\n", name);
1972 
1973 			e_ddi_branch_hold(dip);
1974 			ddi_release_devi(dip); /* held in e_ddi_nodeid_to_dip */
1975 
1976 			if (e_ddi_branch_destroy(dip, &fdip, 0)) {
1977 				char *path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1978 
1979 				/*
1980 				 * If non-NULL, fdip is held and must be
1981 				 * released.
1982 				 */
1983 				if (fdip != NULL) {
1984 					(void) ddi_pathname(fdip, path);
1985 					ddi_release_devi(fdip);
1986 				} else {
1987 					(void) ddi_pathname(dip, path);
1988 				}
1989 
1990 				err = drerr_new(1, ESTF_DRVFAIL, path);
1991 				kmem_free(path, MAXPATHLEN);
1992 				e_ddi_branch_rele(dip);
1993 				break;
1994 			}
1995 		}
1996 	}
1997 
1998 	return (err);
1999 }
2000 
2001 /*ARGSUSED*/
2002 sbd_error_t *
2003 drmach_pre_op(int cmd, drmachid_t id, drmach_opts_t *opts)
2004 {
2005 	/* allow status and ncm operations to always succeed */
2006 	if ((cmd == SBD_CMD_STATUS) || (cmd == SBD_CMD_GETNCM)) {
2007 		return (NULL);
2008 	}
2009 
2010 	/* check all other commands for the required option string */
2011 	if ((opts->size > 0) && (opts->copts != NULL)) {
2012 
2013 		DRMACH_PR("platform options: %s\n", opts->copts);
2014 
2015 		if (strstr(opts->copts, "xfdr") != NULL) {
2016 			return (NULL);
2017 		}
2018 	}
2019 
2020 	return (drerr_new(0, ESTF_SUPPORT, NULL));
2021 }
2022 
2023 /*ARGSUSED*/
2024 sbd_error_t *
2025 drmach_post_op(int cmd, drmachid_t id, drmach_opts_t *opts)
2026 {
2027 	sbd_error_t	*err = NULL;
2028 
2029 	switch (cmd) {
2030 	case SBD_CMD_UNCONFIGURE:
2031 
2032 		err = drmach_remove_counter_nodes(id);
2033 		break;
2034 
2035 	case SBD_CMD_CONFIGURE:
2036 	case SBD_CMD_DISCONNECT:
2037 	case SBD_CMD_CONNECT:
2038 	case SBD_CMD_GETNCM:
2039 	case SBD_CMD_STATUS:
2040 		break;
2041 
2042 	default:
2043 		break;
2044 	}
2045 
2046 	return (err);
2047 }
2048 
2049 sbd_error_t *
2050 drmach_board_assign(int bnum, drmachid_t *id)
2051 {
2052 	sbd_error_t	*err;
2053 
2054 	if (!drmach_initialized && drmach_init() == -1) {
2055 		err = DRMACH_INTERNAL_ERROR();
2056 	} else if (drmach_array_get(drmach_boards, bnum, id) == -1) {
2057 		err = drerr_new(1, ESTF_BNUM, "%d", bnum);
2058 	} else if (*id != NULL) {
2059 		err = NULL;
2060 	} else {
2061 		drmach_board_t	*bp;
2062 
2063 		*id  = (drmachid_t)drmach_board_new(bnum);
2064 		bp = *id;
2065 		bp->assigned = 1;
2066 		err = NULL;
2067 	}
2068 
2069 	return (err);
2070 }
2071 
2072 static int
2073 drmach_attach_board(void *arg)
2074 {
2075 	drmach_board_t	*obj = (drmach_board_t *)arg;
2076 	cpuset_t	cset;
2077 	int		retval;
2078 
2079 	/*
2080 	 * OBP disables traps during the board probe.
2081 	 * So, in order to prevent cross-call/cross-trap timeouts,
2082 	 * and thus panics, we effectively block anybody from
2083 	 * issuing xc's/xt's by doing a promsafe_xc_attention.
2084 	 * In the previous version of Starfire DR (2.6), a timeout
2085 	 * suspension mechanism was implemented in the send-mondo
2086 	 * assembly.  That mechanism is unnecessary with the
2087 	 * existence of xc_attention/xc_dismissed.
2088 	 */
2089 	cset = cpu_ready_set;
2090 	promsafe_xc_attention(cset);
2091 
2092 	retval = prom_starfire_add_brd(obj->connect_cpuid);
2093 
2094 	xc_dismissed(cset);
2095 
2096 	return (retval);
2097 }
2098 
2099 sbd_error_t *
2100 drmach_board_connect(drmachid_t id, drmach_opts_t *opts)
2101 {
2102 	drmach_board_t	*obj = (drmach_board_t *)id;
2103 	int		retval;
2104 	sbd_error_t	*err;
2105 	char		*cptr, *copts;
2106 
2107 	if (!DRMACH_IS_BOARD_ID(id))
2108 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2109 
2110 	if (opts->size > 0)
2111 		copts = opts->copts;
2112 
2113 	if ((cptr = strstr(copts, "cpuid=")) != NULL) {
2114 		int cpuid;
2115 
2116 		cptr += strlen("cpuid=");
2117 		cpuid = stoi(&cptr);
2118 
2119 		if (DRMACH_CPUID2BNUM(cpuid) == obj->bnum) {
2120 			obj->connect_cpuid = cpuid;
2121 			obj->assigned = 1;
2122 		} else
2123 			return (drerr_new(1, ESTF_SETCPUVAL, "%d", cpuid));
2124 	} else {
2125 		/* cpuid was not specified */
2126 		obj->connect_cpuid = -1;
2127 	}
2128 
2129 	if (obj->connect_cpuid == -1) {
2130 		err =  drerr_new(1, ESTF_NOCPUID, obj->cm.name);
2131 		return (err);
2132 	}
2133 
2134 	cmn_err(CE_CONT, "DRMACH: PROM attach %s CPU %d\n",
2135 		obj->cm.name, obj->connect_cpuid);
2136 
2137 	retval = prom_tree_update(drmach_attach_board, obj);
2138 
2139 	if (retval == 0)
2140 		err = NULL;
2141 	else {
2142 		cmn_err(CE_WARN, "prom error: prom_starfire_add_brd(%d) "
2143 			"returned %d", obj->connect_cpuid, retval);
2144 
2145 		err = drerr_new(1, ESTF_PROBE, obj->cm.name);
2146 	}
2147 
2148 	obj->connect_cpuid = -1;
2149 
2150 	return (err);
2151 }
2152 
2153 /*ARGSUSED*/
2154 sbd_error_t *
2155 drmach_board_disconnect(drmachid_t id, drmach_opts_t *opts)
2156 {
2157 	return (NULL);
2158 }
2159 
2160 static int
2161 drmach_board_find_devices_cb(drmach_node_walk_args_t *args)
2162 {
2163 	drmach_node_t			*node = args->node;
2164 	drmach_board_cb_data_t		*data = args->data;
2165 	drmach_board_t			*obj = data->obj;
2166 
2167 	int		 rv;
2168 	int		 bnum;
2169 	drmach_device_t	*device;
2170 
2171 	rv = drmach_node_get_prop(node, OBP_BOARDNUM, &bnum);
2172 	if (rv) {
2173 		/*
2174 		 * if the node does not have a board# property, then
2175 		 * by that information alone it is known that drmach
2176 		 * is not interested in it.
2177 		 */
2178 		return (0);
2179 	} else if (bnum != obj->bnum)
2180 		return (0);
2181 
2182 	/*
2183 	 * Create a device data structure from this node data.
2184 	 * The call may yield nothing if the node is not of interest
2185 	 * to drmach.
2186 	 */
2187 	data->err = drmach_device_new(node, obj, &device);
2188 	if (data->err)
2189 		return (-1);
2190 	else if (device == NULL) {
2191 		/*
2192 		 * drmach_device_new examined the node we passed in
2193 		 * and determined that it was one not of interest to
2194 		 * drmach.  So, it is skipped.
2195 		 */
2196 		return (0);
2197 	}
2198 
2199 	rv = drmach_array_set(obj->devices, data->ndevs++, device);
2200 	if (rv) {
2201 		drmach_device_dispose(device);
2202 		data->err = DRMACH_INTERNAL_ERROR();
2203 		return (-1);
2204 	}
2205 
2206 	data->err = (*data->found)(data->a, device->type, device->unum, device);
2207 	return (data->err == NULL ? 0 : -1);
2208 }
2209 
2210 sbd_error_t *
2211 drmach_board_find_devices(drmachid_t id, void *a,
2212 	sbd_error_t *(*found)(void *a, const char *, int, drmachid_t))
2213 {
2214 	extern int		 plat_max_cpu_units_per_board();
2215 	extern int		 plat_max_mem_units_per_board();
2216 	extern int		 plat_max_io_units_per_board();
2217 
2218 	drmach_board_t		*obj = (drmach_board_t *)id;
2219 	sbd_error_t		*err;
2220 	int			 max_devices;
2221 	int			 rv;
2222 	drmach_board_cb_data_t	data;
2223 
2224 	max_devices  = plat_max_cpu_units_per_board();
2225 	max_devices += plat_max_mem_units_per_board();
2226 	max_devices += plat_max_io_units_per_board();
2227 
2228 	obj->devices = drmach_array_new(0, max_devices);
2229 
2230 	data.obj = obj;
2231 	data.ndevs = 0;
2232 	data.found = found;
2233 	data.a = a;
2234 	data.err = NULL;
2235 
2236 	rv = drmach_node_walk(obj->tree, &data, drmach_board_find_devices_cb);
2237 	if (rv == 0)
2238 		err = NULL;
2239 	else {
2240 		drmach_array_dispose(obj->devices, drmach_device_dispose);
2241 		obj->devices = NULL;
2242 
2243 		if (data.err)
2244 			err = data.err;
2245 		else
2246 			err = DRMACH_INTERNAL_ERROR();
2247 	}
2248 
2249 	return (err);
2250 }
2251 
2252 int
2253 drmach_board_lookup(int bnum, drmachid_t *id)
2254 {
2255 	int	rv = 0;
2256 
2257 	if (!drmach_initialized && drmach_init() == -1) {
2258 		*id = 0;
2259 		rv = -1;
2260 	} else if (drmach_array_get(drmach_boards, bnum, id)) {
2261 		*id = 0;
2262 		rv = -1;
2263 	}
2264 	return (rv);
2265 }
2266 
2267 sbd_error_t *
2268 drmach_board_name(int bnum, char *buf, int buflen)
2269 {
2270 	snprintf(buf, buflen, "SB%d", bnum);
2271 	return (NULL);
2272 }
2273 
2274 sbd_error_t *
2275 drmach_board_poweroff(drmachid_t id)
2276 {
2277 	drmach_board_t	*bp;
2278 	sbd_error_t	*err;
2279 	drmach_status_t	 stat;
2280 
2281 	if (!DRMACH_IS_BOARD_ID(id))
2282 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2283 	bp = id;
2284 
2285 	err = drmach_board_status(id, &stat);
2286 	if (err)
2287 		return (err);
2288 	else if (stat.configured || stat.busy)
2289 		return (drerr_new(0, ESTF_CONFIGBUSY, bp->cm.name));
2290 	else {
2291 		/* board power off is essentially a noop for Starfire */
2292 		bp->powered = 0;
2293 		return (NULL);
2294 	}
2295 	/*NOTREACHED*/
2296 }
2297 
2298 sbd_error_t *
2299 drmach_board_poweron(drmachid_t id)
2300 {
2301 	drmach_board_t	*bp;
2302 
2303 	if (!DRMACH_IS_BOARD_ID(id))
2304 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2305 	bp = id;
2306 
2307 	/* board power on is essentially a noop for Starfire */
2308 	bp->powered = 1;
2309 
2310 	return (NULL);
2311 }
2312 
2313 static sbd_error_t *
2314 drmach_board_release(drmachid_t id)
2315 {
2316 	if (!DRMACH_IS_BOARD_ID(id))
2317 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2318 	return (NULL);
2319 }
2320 
2321 /*ARGSUSED*/
2322 sbd_error_t *
2323 drmach_board_test(drmachid_t id, drmach_opts_t *opts, int force)
2324 {
2325 	return (NULL);
2326 }
2327 
2328 sbd_error_t *
2329 drmach_board_unassign(drmachid_t id)
2330 {
2331 	drmach_board_t	*bp;
2332 	sbd_error_t	*err;
2333 	drmach_status_t	 stat;
2334 
2335 	if (!DRMACH_IS_BOARD_ID(id))
2336 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2337 	bp = id;
2338 
2339 	err = drmach_board_status(id, &stat);
2340 	if (err)
2341 		return (err);
2342 	else if (stat.configured || stat.busy)
2343 		return (drerr_new(0, ESTF_CONFIGBUSY, bp->cm.name));
2344 	else if (drmach_array_set(drmach_boards, bp->bnum, 0) != 0)
2345 		return (DRMACH_INTERNAL_ERROR());
2346 	else {
2347 		drmach_board_dispose(bp);
2348 		return (NULL);
2349 	}
2350 	/*NOTREACHED*/
2351 }
2352 
2353 static sbd_error_t *
2354 drmach_cpu_new(drmach_device_t *dp)
2355 {
2356 	static sbd_error_t *drmach_cpu_release(drmachid_t);
2357 	static sbd_error_t *drmach_cpu_status(drmachid_t, drmach_status_t *);
2358 
2359 	sbd_error_t	*err;
2360 	int		 portid;
2361 
2362 	err = drmach_device_get_prop(dp, "upa-portid", &portid);
2363 	if (err == NULL)
2364 		dp->unum = portid & 3;
2365 
2366 	dp->cm.isa = (void *)drmach_cpu_new;
2367 	dp->cm.release = drmach_cpu_release;
2368 	dp->cm.status = drmach_cpu_status;
2369 
2370 	snprintf(dp->cm.name, sizeof (dp->cm.name), "%s%d", dp->type, dp->unum);
2371 
2372 	return (err);
2373 }
2374 
2375 /*
2376  * drmach_cpu_obp_detach()
2377  *  This requires two steps, first, we must put the cpuid into the OBP
2378  *  idle loop (Idle in Program) state.  Then we call OBP to place the CPU
2379  *  into the "Detached" state, which does any special processing to
2380  *  actually detach the cpu, such as flushing ecache, and also ensures
2381  *  that a subsequent breakpoint won't restart the cpu (if it was just in
2382  *  Idle in Program state).
2383  */
2384 static void
2385 drmach_cpu_obp_detach(int cpuid)
2386 {
2387 	/*
2388 	 * Cpu may not be under OBP's control. Eg, if cpu exited to download
2389 	 * helper on a prior attach.
2390 	 */
2391 	if (CPU_SGN_EXISTS(cpuid) &&
2392 			!SGN_CPU_IS_OS(cpuid) &&
2393 			!SGN_CPU_IS_OBP(cpuid)) {
2394 		cmn_err(CE_WARN,
2395 			"unexpected signature (0x%x) for cpu %d",
2396 			get_cpu_sgn(cpuid), cpuid);
2397 	}
2398 
2399 	/*
2400 	 * Now we place the CPU into the "Detached" idle loop in OBP.
2401 	 * This is so that the CPU won't be restarted if we break into
2402 	 * OBP with a breakpoint or BREAK key from the console, and also
2403 	 * if we need to do any special processing, such as flushing the
2404 	 * cpu's ecache, disabling interrupts (by turning of the ET bit in
2405 	 * the PSR) and/or spinning in BBSRAM rather than global memory.
2406 	 */
2407 	DRMACH_PR("prom_starfire_rm_cpu(%d)\n", cpuid);
2408 	prom_starfire_rm_cpu(cpuid);
2409 }
2410 
2411 /*
2412  * drmach_cpu_obp_is_detached() returns TRUE if the cpu sigblock signature state
2413  * is SIGBST_DETACHED; otherwise it returns FALSE. This routine should only
2414  * be called after we have asked OBP to detach the CPU. It should NOT be
2415  * called as a check during any other flow.
2416  */
2417 static int
2418 drmach_cpu_obp_is_detached(int cpuid)
2419 {
2420 	if (!CPU_SGN_EXISTS(cpuid) ||
2421 		(SGN_CPU_IS_OS(cpuid) && SGN_CPU_STATE_IS_DETACHED(cpuid)))
2422 		return (1);
2423 	else
2424 		return (0);
2425 }
2426 
2427 static int
2428 drmach_cpu_start(struct cpu *cp)
2429 {
2430 	int		cpuid = cp->cpu_id;
2431 	int		ntries = drmach_cpu_ntries;
2432 	extern void	restart_other_cpu(int);
2433 
2434 	ASSERT(MUTEX_HELD(&cpu_lock));
2435 	ASSERT(cpunodes[cpuid].nodeid != (dnode_t)0);
2436 
2437 	cp->cpu_flags &= ~CPU_POWEROFF;
2438 
2439 	/*
2440 	 * NOTE: restart_other_cpu pauses cpus during the
2441 	 *	 slave cpu start.  This helps to quiesce the
2442 	 *	 bus traffic a bit which makes the tick sync
2443 	 *	 routine in the prom more robust.
2444 	 */
2445 	DRMACH_PR("COLD START for cpu (%d)\n", cpuid);
2446 
2447 	prom_starfire_add_cpu(cpuid);
2448 
2449 	restart_other_cpu(cpuid);
2450 
2451 	/*
2452 	 * Wait for the cpu to reach its idle thread before
2453 	 * we zap him with a request to blow away the mappings
2454 	 * he (might) have for the drmach_shutdown_asm code
2455 	 * he may have executed on unconfigure.
2456 	 */
2457 	while ((cp->cpu_thread != cp->cpu_idle_thread) && (ntries > 0)) {
2458 		DELAY(drmach_cpu_delay);
2459 		ntries--;
2460 	}
2461 
2462 	DRMACH_PR("waited %d out of %d loops for cpu %d\n",
2463 		drmach_cpu_ntries - ntries, drmach_cpu_ntries, cpuid);
2464 
2465 	xt_one(cpuid, vtag_flushpage_tl1,
2466 		(uint64_t)drmach_shutdown_va, (uint64_t)KCONTEXT);
2467 
2468 	return (0);
2469 }
2470 
2471 /*
2472  * A detaching CPU is xcalled with an xtrap to drmach_cpu_stop_self() after
2473  * it has been offlined. The function of this routine is to get the cpu
2474  * spinning in a safe place. The requirement is that the system will not
2475  * reference anything on the detaching board (memory and i/o is detached
2476  * elsewhere) and that the CPU not reference anything on any other board
2477  * in the system.  This isolation is required during and after the writes
2478  * to the domain masks to remove the board from the domain.
2479  *
2480  * To accomplish this isolation the following is done:
2481  *	1) Create a locked mapping to a location in BBSRAM where
2482  *	   the cpu will execute.
2483  *	2) Copy the target function (drmach_shutdown_asm) in which
2484  *	   the cpu will execute into BBSRAM.
2485  *	3) Jump into function with BBSRAM.
2486  *	   Function will:
2487  *	   3.1) Flush its Ecache (displacement).
2488  *	   3.2) Flush its Dcache with HW mechanism.
2489  *	   3.3) Flush its Icache with HW mechanism.
2490  *	   3.4) Flush all valid and _unlocked_ D-TLB entries.
2491  *	   3.5) Flush all valid and _unlocked_ I-TLB entries.
2492  *	   3.6) Clear xt_mb to signal completion. Note: cache line is
2493  *		recovered by drmach_cpu_poweroff().
2494  *	4) Jump into a tight loop.
2495  */
2496 #define	DRMACH_BBSRAM_OFFSET	0x1000
2497 
2498 static void
2499 drmach_cpu_stop_self(void)
2500 {
2501 	int		cpuid = (int)CPU->cpu_id;
2502 	tte_t		tte;
2503 	volatile uint_t	*src, *dst;
2504 	uint_t		funclen;
2505 	uint64_t	bbsram_pa, bbsram_offset;
2506 	uint_t		bbsram_pfn;
2507 	uint64_t	bbsram_addr;
2508 	void		(*bbsram_func)(uint64_t);
2509 	extern void	drmach_shutdown_asm(uint64_t);
2510 	extern void	drmach_shutdown_asm_end(void);
2511 
2512 	funclen = (uint_t)drmach_shutdown_asm_end - (uint_t)drmach_shutdown_asm;
2513 	ASSERT(funclen <= MMU_PAGESIZE);
2514 	/*
2515 	 * We'll start from the 0th's base.
2516 	 */
2517 	bbsram_pa = STARFIRE_UPAID2UPS(cpuid) | STARFIRE_PSI_BASE;
2518 	bbsram_offset = bbsram_pa | 0xfe0ULL;
2519 	bbsram_pa += ldphysio(bbsram_offset) + DRMACH_BBSRAM_OFFSET;
2520 
2521 	bbsram_pfn = (uint_t)(bbsram_pa >> MMU_PAGESHIFT);
2522 
2523 	bbsram_addr = (uint64_t)drmach_shutdown_va;
2524 	drmach_shutdown_asm_mbox->estack = bbsram_addr + (uint64_t)funclen;
2525 
2526 	tte.tte_inthi = TTE_VALID_INT | TTE_SZ_INT(TTE8K) |
2527 			TTE_PFN_INTHI(bbsram_pfn);
2528 	tte.tte_intlo = TTE_PFN_INTLO(bbsram_pfn) |
2529 			TTE_HWWR_INT | TTE_PRIV_INT | TTE_LCK_INT;
2530 	sfmmu_dtlb_ld(drmach_shutdown_va, KCONTEXT, &tte);	/* load dtlb */
2531 	sfmmu_itlb_ld(drmach_shutdown_va, KCONTEXT, &tte);	/* load itlb */
2532 
2533 	for (src = (uint_t *)drmach_shutdown_asm, dst = (uint_t *)bbsram_addr;
2534 		src < (uint_t *)drmach_shutdown_asm_end; src++, dst++)
2535 		*dst = *src;
2536 
2537 	bbsram_func = (void (*)())bbsram_addr;
2538 	drmach_shutdown_asm_mbox->flushaddr = ecache_flushaddr;
2539 	drmach_shutdown_asm_mbox->size = (cpunodes[cpuid].ecache_size << 1);
2540 	drmach_shutdown_asm_mbox->linesize = cpunodes[cpuid].ecache_linesize;
2541 	drmach_shutdown_asm_mbox->physaddr
2542 				    = va_to_pa((void *)&drmach_xt_mb[cpuid]);
2543 
2544 	/*
2545 	 * Signal to drmach_cpu_poweroff() is via drmach_xt_mb cleared
2546 	 * by asm code
2547 	 */
2548 
2549 	(*bbsram_func)(va_to_pa((void *)drmach_shutdown_asm_mbox));
2550 }
2551 
2552 static void
2553 drmach_cpu_shutdown_self(void)
2554 {
2555 	cpu_t		*cp = CPU;
2556 	int		cpuid = cp->cpu_id;
2557 	extern void	flush_windows(void);
2558 
2559 	flush_windows();
2560 
2561 	(void) spl8();
2562 
2563 	ASSERT(cp->cpu_intr_actv == 0);
2564 	ASSERT(cp->cpu_thread == cp->cpu_idle_thread);
2565 
2566 	cp->cpu_flags = CPU_OFFLINE | CPU_QUIESCED | CPU_POWEROFF;
2567 
2568 	drmach_cpu_stop_self();
2569 
2570 	cmn_err(CE_PANIC, "CPU %d FAILED TO SHUTDOWN", cpuid);
2571 }
2572 
2573 /* a helper routine to keep the math in one place */
2574 static processorid_t
2575 drmach_cpu_calc_id(drmach_device_t *dp)
2576 {
2577 	return (dp->bp->bnum * MAX_CPU_UNITS_PER_BOARD + dp->unum);
2578 }
2579 
2580 /*
2581  * Move bootproc (SIGBCPU) to another cpu.  If dst_cpu is NULL, a
2582  * destination cpu is chosen from the set of cpus not located on the
2583  * same board as the current bootproc cpu.
2584  */
2585 static sbd_error_t *
2586 drmach_cpu_juggle_bootproc(drmach_device_t *dst_cpu)
2587 {
2588 	processorid_t	 cpuid;
2589 	struct cpu	*cp;
2590 	sbd_error_t	*err;
2591 	int		 rv;
2592 
2593 	ASSERT(MUTEX_HELD(&cpu_lock));
2594 
2595 	/* dst_cpu is NULL when target cpu is unspecified. So, pick one. */
2596 	if (dst_cpu == NULL) {
2597 		int avoid_board = DRMACH_CPUID2BNUM(SIGBCPU->cpu_id);
2598 		int max_cpuid = MAX_BOARDS * MAX_CPU_UNITS_PER_BOARD;
2599 
2600 		for (cpuid = 0; cpuid < max_cpuid; cpuid++)
2601 			if (DRMACH_CPUID2BNUM(cpuid) != avoid_board) {
2602 				cp = cpu_get(cpuid);
2603 				if (cp != NULL && cpu_is_online(cp))
2604 					break;
2605 			}
2606 
2607 		if (cpuid == max_cpuid) {
2608 			err = drerr_new(1, ESTF_JUGGLE, NULL);
2609 			return (err);
2610 		}
2611 
2612 		/* else, cp points to the selected target cpu */
2613 	} else {
2614 		cpuid = drmach_cpu_calc_id(dst_cpu);
2615 
2616 		if ((cp = cpu_get(cpuid)) == NULL) {
2617 			err = drerr_new(1, ESTF_NODEV, "%s::%s",
2618 				dst_cpu->bp->cm.name, dst_cpu->cm.name);
2619 			return (err);
2620 		}
2621 
2622 		if (cpuid == SIGBCPU->cpu_id) {
2623 			cmn_err(CE_WARN,
2624 				"SIGBCPU(%d) same as new selection(%d)",
2625 				SIGBCPU->cpu_id, cpuid);
2626 
2627 			/* technically not an error, but a no-op */
2628 			return (NULL);
2629 		}
2630 	}
2631 
2632 	cmn_err(CE_NOTE, "?relocating SIGBCPU from %d to %d",
2633 		SIGBCPU->cpu_id, cpuid);
2634 
2635 	DRMACH_PR("moving SIGBCPU to CPU %d\n", cpuid);
2636 
2637 	/*
2638 	 * Tell OBP to initialize cvc-offset field of new CPU0
2639 	 * so that it's in sync with OBP and cvc_server
2640 	 */
2641 	prom_starfire_init_console(cpuid);
2642 
2643 	/*
2644 	 * Assign cvc to new cpu0's bbsram for I/O.  This has to be
2645 	 * done BEFORE cpu0 is moved via obp, since this logic
2646 	 * will cause obp_helper to switch to a different bbsram for
2647 	 * cvc I/O.  We don't want cvc writing to a buffer from which
2648 	 * nobody will pick up the data!
2649 	 */
2650 	cvc_assign_iocpu(cpuid);
2651 
2652 	rv = prom_starfire_move_cpu0(cpuid);
2653 
2654 	if (rv == 0) {
2655 		SIGBCPU = cp;
2656 
2657 		DRMACH_PR("successfully juggled to CPU %d\n", cpuid);
2658 		return (NULL);
2659 	} else {
2660 		DRMACH_PR("prom error: prom_starfire_move_cpu0(%d) "
2661 			"returned %d\n", cpuid, rv);
2662 
2663 		/*
2664 		 * The move failed, hopefully obp_helper is still back
2665 		 * at the old bootproc.  Move cvc back there.
2666 		 */
2667 		cvc_assign_iocpu(SIGBCPU->cpu_id);
2668 
2669 
2670 		err = drerr_new(1, ESTF_MOVESIGB, "CPU %d", cpuid);
2671 		return (err);
2672 	}
2673 	/*NOTREACHED*/
2674 }
2675 
2676 static sbd_error_t *
2677 drmach_cpu_release(drmachid_t id)
2678 {
2679 	drmach_device_t	*dp;
2680 	processorid_t	 cpuid;
2681 	struct cpu	*cp;
2682 	sbd_error_t	*err;
2683 
2684 	if (!DRMACH_IS_CPU_ID(id))
2685 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2686 	dp = id;
2687 	cpuid = drmach_cpu_calc_id(dp);
2688 
2689 	ASSERT(MUTEX_HELD(&cpu_lock));
2690 
2691 	cp = cpu_get(cpuid);
2692 	if (cp == NULL)
2693 		err = DRMACH_INTERNAL_ERROR();
2694 	else if (SIGBCPU->cpu_id == cp->cpu_id)
2695 		err = drmach_cpu_juggle_bootproc(NULL);
2696 	else
2697 		err = NULL;
2698 
2699 	return (err);
2700 }
2701 
2702 static sbd_error_t *
2703 drmach_cpu_status(drmachid_t id, drmach_status_t *stat)
2704 {
2705 	drmach_device_t *dp;
2706 
2707 	ASSERT(DRMACH_IS_CPU_ID(id));
2708 	dp = id;
2709 
2710 	stat->assigned = dp->bp->assigned;
2711 	stat->powered = dp->bp->powered;
2712 	mutex_enter(&cpu_lock);
2713 	stat->configured = (cpu_get(drmach_cpu_calc_id(dp)) != NULL);
2714 	mutex_exit(&cpu_lock);
2715 	stat->busy = dp->busy;
2716 	strncpy(stat->type, dp->type, sizeof (stat->type));
2717 	stat->info[0] = '\0';
2718 
2719 	return (NULL);
2720 }
2721 
2722 sbd_error_t *
2723 drmach_cpu_disconnect(drmachid_t id)
2724 {
2725 	drmach_device_t	*cpu;
2726 	int		 cpuid;
2727 	int		 ntries;
2728 	int		 p;
2729 	u_longlong_t	 pc_addr;
2730 	uchar_t		 rvalue;
2731 
2732 	if (!DRMACH_IS_CPU_ID(id))
2733 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2734 	cpu = id;
2735 
2736 	cpuid = drmach_cpu_calc_id(cpu);
2737 	if (SIGBCPU->cpu_id == cpuid) {
2738 		/* this cpu is SIGBCPU, can't disconnect */
2739 		return (drerr_new(1, ESTF_HASSIGB, "%s::%s",
2740 				cpu->bp->cm.name, cpu->cm.name));
2741 	}
2742 
2743 	/*
2744 	 * Make sure SIGBST_DETACHED is set before
2745 	 * mapping out the sig block.
2746 	 */
2747 	ntries = drmach_cpu_ntries;
2748 	while (!drmach_cpu_obp_is_detached(cpuid) && ntries) {
2749 		DELAY(drmach_cpu_delay);
2750 		ntries--;
2751 	}
2752 	if (!drmach_cpu_obp_is_detached(cpuid)) {
2753 		cmn_err(CE_WARN, "failed to mark cpu %d detached in sigblock",
2754 			cpuid);
2755 	}
2756 
2757 	/* map out signature block */
2758 	if (CPU_SGN_EXISTS(cpuid)) {
2759 		CPU_SGN_MAPOUT(cpuid);
2760 	}
2761 
2762 	/*
2763 	 * We now PC IDLE the processor to guarantee we
2764 	 * stop any transactions from coming from it.
2765 	 */
2766 	p = cpu->unum & 1;
2767 	pc_addr = STARFIRE_BB_PC_ADDR(cpu->bp->bnum, cpu->unum, 0);
2768 
2769 	DRMACH_PR("PC idle cpu %d (addr = 0x%llx, port = %d, p = %d)",
2770 		drmach_cpu_calc_id(cpu), pc_addr, cpu->unum, p);
2771 
2772 	rvalue = ldbphysio(pc_addr);
2773 	rvalue |= STARFIRE_BB_PC_IDLE(p);
2774 	stbphysio(pc_addr, rvalue);
2775 	DELAY(50000);
2776 
2777 	return (NULL);
2778 }
2779 
2780 sbd_error_t *
2781 drmach_cpu_get_id(drmachid_t id, processorid_t *cpuid)
2782 {
2783 	drmach_device_t *cpu;
2784 
2785 	if (!DRMACH_IS_CPU_ID(id))
2786 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2787 	cpu = id;
2788 
2789 	*cpuid = drmach_cpu_calc_id(cpu);
2790 	return (NULL);
2791 }
2792 
2793 sbd_error_t *
2794 drmach_cpu_get_impl(drmachid_t id, int *ip)
2795 {
2796 	drmach_device_t *cpu;
2797 	int		impl;
2798 
2799 	if (!DRMACH_IS_CPU_ID(id))
2800 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2801 
2802 	cpu = id;
2803 
2804 	if (drmach_node_get_prop(cpu->node, "implementation#", &impl) == -1) {
2805 		return (DRMACH_INTERNAL_ERROR());
2806 	}
2807 
2808 	*ip = impl;
2809 
2810 	return (NULL);
2811 }
2812 
2813 void
2814 drmach_cpu_flush_ecache_sync(void)
2815 {
2816 	ASSERT(curthread->t_bound_cpu == CPU);
2817 
2818 	/*
2819 	 * Now let's flush our ecache thereby removing all references
2820 	 * to the target (detaching) memory from all ecache's in
2821 	 * system.
2822 	 */
2823 	cpu_flush_ecache();
2824 
2825 	/*
2826 	 * Delay 100 usec out of paranoia to insure everything
2827 	 * (hardware queues) has drained before we start reprogramming
2828 	 * the hardware.
2829 	 */
2830 	DELAY(100);
2831 }
2832 
2833 sbd_error_t *
2834 drmach_get_dip(drmachid_t id, dev_info_t **dip)
2835 {
2836 	drmach_device_t	*dp;
2837 
2838 	if (!DRMACH_IS_DEVICE_ID(id))
2839 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2840 	dp = id;
2841 
2842 	*dip = drmach_node_get_dip(dp->node);
2843 	return (NULL);
2844 }
2845 
2846 sbd_error_t *
2847 drmach_io_is_attached(drmachid_t id, int *yes)
2848 {
2849 	drmach_device_t *dp;
2850 	dev_info_t	*dip;
2851 	int		state;
2852 
2853 	if (!DRMACH_IS_IO_ID(id))
2854 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2855 	dp = id;
2856 
2857 	dip = drmach_node_get_dip(dp->node);
2858 	if (dip == NULL) {
2859 		*yes = 0;
2860 		return (NULL);
2861 	}
2862 
2863 	state = ddi_get_devstate(dip);
2864 	*yes = ((i_ddi_node_state(dip) >= DS_ATTACHED) ||
2865 	    (state == DDI_DEVSTATE_UP));
2866 
2867 	return (NULL);
2868 }
2869 
2870 sbd_error_t *
2871 drmach_io_pre_release(drmachid_t id)
2872 {
2873 	if (!DRMACH_IS_IO_ID(id))
2874 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2875 	return (NULL);
2876 }
2877 
2878 static sbd_error_t *
2879 drmach_io_release(drmachid_t id)
2880 {
2881 	if (!DRMACH_IS_IO_ID(id))
2882 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2883 	return (NULL);
2884 }
2885 
2886 sbd_error_t *
2887 drmach_io_unrelease(drmachid_t id)
2888 {
2889 	if (!DRMACH_IS_IO_ID(id))
2890 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2891 	return (NULL);
2892 }
2893 
2894 /*ARGSUSED*/
2895 sbd_error_t *
2896 drmach_io_post_release(drmachid_t id)
2897 {
2898 	return (NULL);
2899 }
2900 
2901 /*ARGSUSED*/
2902 sbd_error_t *
2903 drmach_io_post_attach(drmachid_t id)
2904 {
2905 	return (NULL);
2906 }
2907 
2908 static sbd_error_t *
2909 drmach_io_status(drmachid_t id, drmach_status_t *stat)
2910 {
2911 	drmach_device_t *dp;
2912 	sbd_error_t	*err;
2913 	int		 configured;
2914 
2915 	ASSERT(DRMACH_IS_IO_ID(id));
2916 	dp = id;
2917 
2918 	err = drmach_io_is_attached(id, &configured);
2919 	if (err)
2920 		return (err);
2921 
2922 	stat->assigned = dp->bp->assigned;
2923 	stat->powered = dp->bp->powered;
2924 	stat->configured = (configured != 0);
2925 	stat->busy = dp->busy;
2926 	strncpy(stat->type, dp->type, sizeof (stat->type));
2927 	stat->info[0] = '\0';
2928 
2929 	return (NULL);
2930 }
2931 
2932 static sbd_error_t *
2933 drmach_mem_new(drmach_device_t *dp)
2934 {
2935 	static sbd_error_t *drmach_mem_release(drmachid_t);
2936 	static sbd_error_t *drmach_mem_status(drmachid_t, drmach_status_t *);
2937 
2938 	dp->unum = 0;
2939 	dp->cm.isa = (void *)drmach_mem_new;
2940 	dp->cm.release = drmach_mem_release;
2941 	dp->cm.status = drmach_mem_status;
2942 
2943 	snprintf(dp->cm.name, sizeof (dp->cm.name), "%s", dp->type);
2944 
2945 	return (NULL);
2946 }
2947 
2948 sbd_error_t *
2949 drmach_mem_add_span(drmachid_t id, uint64_t basepa, uint64_t size)
2950 {
2951 	pfn_t		basepfn = (pfn_t)(basepa >> PAGESHIFT);
2952 	pgcnt_t		npages = (pgcnt_t)(size >> PAGESHIFT);
2953 	pda_handle_t	ph;
2954 	int		rv;
2955 
2956 	ASSERT(size != 0);
2957 
2958 	if (!DRMACH_IS_MEM_ID(id))
2959 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2960 
2961 	kcage_range_lock();
2962 	rv = kcage_range_add(basepfn, npages, 1);
2963 	kcage_range_unlock();
2964 	if (rv == ENOMEM) {
2965 		cmn_err(CE_WARN, "%lld megabytes not available to kernel cage",
2966 			(size == 0 ? 0 : size / MBYTE));
2967 	} else if (rv != 0) {
2968 		/* catch this in debug kernels */
2969 		ASSERT(0);
2970 
2971 		cmn_err(CE_WARN, "unexpected kcage_range_add"
2972 			" return value %d", rv);
2973 	}
2974 
2975 	/*
2976 	 * Update the PDA (post2obp) structure with the
2977 	 * range of the newly added memory.
2978 	 */
2979 	ph = drmach_pda_open();
2980 	if (ph != NULL) {
2981 		pda_mem_add_span(ph, basepa, size);
2982 		pda_close(ph);
2983 	}
2984 
2985 	return (NULL);
2986 }
2987 
2988 sbd_error_t *
2989 drmach_mem_del_span(drmachid_t id, uint64_t basepa, uint64_t size)
2990 {
2991 	drmach_device_t	*mem = id;
2992 	pfn_t		basepfn = (pfn_t)(basepa >> PAGESHIFT);
2993 	pgcnt_t		npages = (pgcnt_t)(size >> PAGESHIFT);
2994 	uint_t		mcreg;
2995 	sbd_error_t	*err;
2996 	pda_handle_t	ph;
2997 	int		rv;
2998 
2999 	err = drmach_read_mc_asr(id, &mcreg);
3000 	if (err)
3001 		return (err);
3002 	else if (mcreg & STARFIRE_MC_INTERLEAVE_MASK) {
3003 		return (drerr_new(1, ESTF_INTERBOARD, "%s::%s",
3004 				mem->bp->cm.name, mem->cm.name));
3005 	}
3006 
3007 	if (size > 0) {
3008 		kcage_range_lock();
3009 		rv = kcage_range_delete_post_mem_del(basepfn, npages);
3010 		kcage_range_unlock();
3011 		if (rv != 0) {
3012 			cmn_err(CE_WARN,
3013 			    "unexpected kcage_range_delete_post_mem_del"
3014 			    " return value %d", rv);
3015 			return (DRMACH_INTERNAL_ERROR());
3016 		}
3017 	}
3018 
3019 	/*
3020 	 * Update the PDA (post2obp) structure with the
3021 	 * range of removed memory.
3022 	 */
3023 	ph = drmach_pda_open();
3024 	if (ph != NULL) {
3025 		if (size > 0)
3026 			pda_mem_del_span(ph, basepa, size);
3027 
3028 		/* update PDA to board's new mc register settings */
3029 		pda_mem_sync(ph, mem->bp->bnum, 0);
3030 
3031 		pda_close(ph);
3032 	}
3033 
3034 	return (NULL);
3035 }
3036 
3037 /* support routine for enable and disable */
3038 static sbd_error_t *
3039 drmach_mem_update_interconnect(drmachid_t id, uint_t mcreg)
3040 {
3041 	drmach_device_t	*dp;
3042 	pda_handle_t	 ph;
3043 	int		 b;
3044 
3045 	if (!DRMACH_IS_MEM_ID(id))
3046 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3047 	dp = id;
3048 
3049 	ph = drmach_pda_open();
3050 	if (ph == NULL)
3051 		return (DRMACH_INTERNAL_ERROR());
3052 
3053 	for (b = 0; b < MAX_BOARDS; b++) {
3054 		int		p;
3055 		int		rv;
3056 		ushort_t	bda_proc, bda_ioc;
3057 		board_desc_t	*bdesc;
3058 
3059 		if (pda_board_present(ph, b) == 0)
3060 			continue;
3061 
3062 		bdesc = (board_desc_t *)pda_get_board_info(ph, b);
3063 
3064 		/*
3065 		 * Update PCs for CPUs.
3066 		 */
3067 
3068 		/* make sure definition in platmod is in sync with pda */
3069 		ASSERT(MAX_PROCMODS == MAX_CPU_UNITS_PER_BOARD);
3070 
3071 		bda_proc = bdesc->bda_proc;
3072 		for (p = 0; p < MAX_PROCMODS; p++) {
3073 			if (BDA_NBL(bda_proc, p) != BDAN_GOOD)
3074 				continue;
3075 
3076 			rv = pc_madr_add(b, dp->bp->bnum, p, mcreg);
3077 			if (rv) {
3078 				pda_close(ph);
3079 				return (DRMACH_INTERNAL_ERROR());
3080 			}
3081 		}
3082 
3083 		/*
3084 		 * Update PCs for IOCs.
3085 		 */
3086 
3087 		/* make sure definition in platmod is in sync with pda */
3088 		ASSERT(MAX_IOCS == MAX_IO_UNITS_PER_BOARD);
3089 
3090 		bda_ioc = bdesc->bda_ioc;
3091 		for (p = 0; p < MAX_IOCS; p++) {
3092 			if (BDA_NBL(bda_ioc, p) != BDAN_GOOD)
3093 				continue;
3094 
3095 			rv = pc_madr_add(b, dp->bp->bnum, p + 4, mcreg);
3096 			if (rv) {
3097 				pda_close(ph);
3098 				return (DRMACH_INTERNAL_ERROR());
3099 			}
3100 		}
3101 	}
3102 
3103 	pda_close(ph);
3104 	return (NULL);
3105 }
3106 
3107 sbd_error_t *
3108 drmach_mem_disable(drmachid_t id)
3109 {
3110 	sbd_error_t	*err;
3111 	uint_t		 mcreg;
3112 
3113 	err = drmach_read_mc_asr(id, &mcreg);
3114 	if (err == NULL) {
3115 		ASSERT(mcreg & STARFIRE_MC_MEM_PRESENT_MASK);
3116 
3117 		/* Turn off presence bit. */
3118 		mcreg &= ~STARFIRE_MC_MEM_PRESENT_MASK;
3119 
3120 		err = drmach_mem_update_interconnect(id, mcreg);
3121 		if (err == NULL)
3122 			err = drmach_write_mc_asr(id, mcreg);
3123 	}
3124 
3125 	return (err);
3126 }
3127 
3128 sbd_error_t *
3129 drmach_mem_enable(drmachid_t id)
3130 {
3131 	sbd_error_t	*err;
3132 	uint_t		 mcreg;
3133 
3134 	err = drmach_read_mc_asr(id, &mcreg);
3135 	if (err == NULL) {
3136 		mcreg |= STARFIRE_MC_MEM_PRESENT_MASK;
3137 
3138 		err = drmach_write_mc_asr(id, mcreg);
3139 		if (err == NULL)
3140 			err = drmach_mem_update_interconnect(id, mcreg);
3141 	}
3142 
3143 	return (err);
3144 }
3145 
3146 sbd_error_t *
3147 drmach_mem_get_alignment(drmachid_t id, uint64_t *mask)
3148 {
3149 	drmach_device_t	*mem;
3150 	sbd_error_t	*err;
3151 	dnode_t		 nodeid;
3152 
3153 	if (!DRMACH_IS_MEM_ID(id))
3154 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3155 	mem = id;
3156 
3157 	nodeid = drmach_node_get_dnode(mem->node);
3158 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
3159 		err = DRMACH_INTERNAL_ERROR();
3160 	else {
3161 		uint64_t size;
3162 
3163 		size = mc_get_alignment_mask(nodeid);
3164 		if (size == (uint64_t)-1)
3165 			err = DRMACH_INTERNAL_ERROR();
3166 		else {
3167 			*mask = size - 1;
3168 			err = NULL;
3169 		}
3170 	}
3171 
3172 	return (err);
3173 }
3174 
3175 sbd_error_t *
3176 drmach_mem_get_base_physaddr(drmachid_t id, uint64_t *pa)
3177 {
3178 	sbd_error_t	*err;
3179 	uint_t		 mcreg;
3180 
3181 	err = drmach_read_mc_asr(id, &mcreg);
3182 	if (err == NULL)
3183 		*pa = mc_asr_to_pa(mcreg);
3184 
3185 	return (err);
3186 }
3187 
3188 /*
3189  * Use of this routine after copy/rename will yield incorrect results,
3190  * because the OBP MEMAVAIL property will not correctly reflect the
3191  * programming of the MCs.
3192  */
3193 sbd_error_t *
3194 drmach_mem_get_memlist(drmachid_t id, struct memlist **ml)
3195 {
3196 	drmach_device_t	*mem;
3197 	int		rv, i, rlen, rblks;
3198 	sbd_error_t	*err;
3199 	struct memlist	*mlist;
3200 	struct sf_memunit_regspec *rlist;
3201 
3202 	if (!DRMACH_IS_MEM_ID(id))
3203 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3204 	mem = id;
3205 
3206 	err = drmach_device_get_proplen(mem, "dr-available", &rlen);
3207 	if (err)
3208 		return (err);
3209 
3210 	rlist = kmem_zalloc(rlen, KM_SLEEP);
3211 
3212 	err = drmach_device_get_prop(mem, "dr-available", rlist);
3213 	if (err) {
3214 		kmem_free(rlist, rlen);
3215 		return (err);
3216 	}
3217 
3218 	mlist = NULL;
3219 	rblks = rlen / sizeof (struct sf_memunit_regspec);
3220 	for (i = 0; i < rblks; i++) {
3221 		uint64_t	addr, size;
3222 
3223 		addr  = (uint64_t)rlist[i].regspec_addr_hi << 32;
3224 		addr |= (uint64_t)rlist[i].regspec_addr_lo;
3225 		size  = (uint64_t)rlist[i].regspec_size_hi << 32;
3226 		size |= (uint64_t)rlist[i].regspec_size_lo;
3227 
3228 		mlist = memlist_add_span(mlist, addr, size);
3229 	}
3230 
3231 	kmem_free(rlist, rlen);
3232 
3233 	/*
3234 	 * Make sure the incoming memlist doesn't already
3235 	 * intersect with what's present in the system (phys_install).
3236 	 */
3237 	memlist_read_lock();
3238 	rv = memlist_intersect(phys_install, mlist);
3239 	memlist_read_unlock();
3240 	if (rv) {
3241 #ifdef DEBUG
3242 		DRMACH_PR("OBP derived memlist intersects"
3243 			" with phys_install\n");
3244 		memlist_dump(mlist);
3245 
3246 		DRMACH_PR("phys_install memlist:\n");
3247 		memlist_dump(phys_install);
3248 #endif
3249 
3250 		memlist_delete(mlist);
3251 		return (DRMACH_INTERNAL_ERROR());
3252 	}
3253 
3254 #ifdef DEBUG
3255 	DRMACH_PR("OBP derived memlist:");
3256 	memlist_dump(mlist);
3257 #endif
3258 
3259 	*ml = mlist;
3260 	return (NULL);
3261 }
3262 
3263 sbd_error_t *
3264 drmach_mem_get_size(drmachid_t id, uint64_t *bytes)
3265 {
3266 	drmach_device_t	*mem;
3267 	pda_handle_t	ph;
3268 	pgcnt_t		npages;
3269 
3270 	if (!DRMACH_IS_MEM_ID(id))
3271 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3272 	mem = id;
3273 
3274 	ph = drmach_pda_open();
3275 	if (ph == NULL)
3276 		return (DRMACH_INTERNAL_ERROR());
3277 
3278 	npages = pda_get_mem_size(ph, mem->bp->bnum);
3279 	*bytes = (uint64_t)npages << PAGESHIFT;
3280 
3281 	pda_close(ph);
3282 	return (NULL);
3283 }
3284 
3285 sbd_error_t *
3286 drmach_mem_get_slice_size(drmachid_t id, uint64_t *bytes)
3287 {
3288 	if (!DRMACH_IS_MEM_ID(id))
3289 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3290 
3291 	*bytes = mc_get_mem_alignment();
3292 	return (NULL);
3293 }
3294 
3295 /* field debugging tool */
3296 processorid_t drmach_mem_cpu_affinity_nail = 0;
3297 
3298 processorid_t
3299 drmach_mem_cpu_affinity(drmachid_t id)
3300 {
3301 	drmach_device_t	*mp;
3302 	drmach_board_t	*bp;
3303 	processorid_t	 cpuid;
3304 
3305 	if (!DRMACH_IS_MEM_ID(id))
3306 		return (CPU_CURRENT);
3307 
3308 	if (drmach_mem_cpu_affinity_nail) {
3309 		cpuid = drmach_mem_cpu_affinity_nail;
3310 
3311 		if (cpuid < 0 || cpuid > NCPU)
3312 			return (CPU_CURRENT);
3313 
3314 		mutex_enter(&cpu_lock);
3315 		if (cpu[cpuid] == NULL || !CPU_ACTIVE(cpu[cpuid]))
3316 			cpuid = CPU_CURRENT;
3317 		mutex_exit(&cpu_lock);
3318 
3319 		return (cpuid);
3320 	}
3321 
3322 	/* try to choose a proc on the target board */
3323 	mp = id;
3324 	bp = mp->bp;
3325 	if (bp->devices) {
3326 		int		rv;
3327 		int		d_idx;
3328 		drmachid_t	d_id;
3329 
3330 		rv = drmach_array_first(bp->devices, &d_idx, &d_id);
3331 		while (rv == 0) {
3332 			if (DRMACH_IS_CPU_ID(d_id)) {
3333 				cpuid = drmach_cpu_calc_id(d_id);
3334 
3335 				mutex_enter(&cpu_lock);
3336 				if (cpu[cpuid] && CPU_ACTIVE(cpu[cpuid])) {
3337 					mutex_exit(&cpu_lock);
3338 					DRMACH_PR("drmach_mem_cpu_affinity: "
3339 					    "selected cpuid=%d\n", cpuid);
3340 					return (cpuid);
3341 				} else {
3342 					mutex_exit(&cpu_lock);
3343 				}
3344 			}
3345 
3346 			rv = drmach_array_next(bp->devices, &d_idx, &d_id);
3347 		}
3348 	}
3349 
3350 	/* otherwise, this proc, wherever it is */
3351 	DRMACH_PR("drmach_mem_cpu_affinity: using default CPU_CURRENT\n");
3352 
3353 	return (CPU_CURRENT);
3354 }
3355 
3356 static sbd_error_t *
3357 drmach_mem_release(drmachid_t id)
3358 {
3359 	if (!DRMACH_IS_MEM_ID(id))
3360 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3361 	return (NULL);
3362 }
3363 
3364 static sbd_error_t *
3365 drmach_mem_status(drmachid_t id, drmach_status_t *stat)
3366 {
3367 	drmach_device_t *dp;
3368 	sbd_error_t	*err;
3369 	uint64_t	 pa, slice_size;
3370 	struct memlist	*ml;
3371 
3372 	ASSERT(DRMACH_IS_MEM_ID(id));
3373 	dp = id;
3374 
3375 	/* get starting physical address of target memory */
3376 	err = drmach_mem_get_base_physaddr(id, &pa);
3377 	if (err)
3378 		return (err);
3379 
3380 	/* round down to slice boundary */
3381 	slice_size = mc_get_mem_alignment();
3382 	pa &= ~ (slice_size - 1);
3383 
3384 	/* stop at first span that is in slice */
3385 	memlist_read_lock();
3386 	for (ml = phys_install; ml; ml = ml->next)
3387 		if (ml->address >= pa && ml->address < pa + slice_size)
3388 			break;
3389 	memlist_read_unlock();
3390 
3391 	stat->assigned = dp->bp->assigned;
3392 	stat->powered = dp->bp->powered;
3393 	stat->configured = (ml != NULL);
3394 	stat->busy = dp->busy;
3395 	strncpy(stat->type, dp->type, sizeof (stat->type));
3396 	stat->info[0] = '\0';
3397 
3398 	return (NULL);
3399 }
3400 
3401 static int
3402 drmach_detach_board(void *arg)
3403 {
3404 	cpuset_t	cset;
3405 	int		retval;
3406 	drmach_board_t	*bp = (drmach_board_t *)arg;
3407 
3408 	cset = cpu_ready_set;
3409 	promsafe_xc_attention(cset);
3410 
3411 	retval = prom_starfire_rm_brd(bp->bnum);
3412 
3413 	xc_dismissed(cset);
3414 
3415 	return (retval);
3416 }
3417 
3418 sbd_error_t *
3419 drmach_board_deprobe(drmachid_t id)
3420 {
3421 	drmach_board_t	*bp;
3422 	int		 retval;
3423 
3424 	if (!DRMACH_IS_BOARD_ID(id))
3425 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3426 	bp = id;
3427 
3428 	cmn_err(CE_CONT, "DR: PROM detach board %d\n", bp->bnum);
3429 
3430 	retval = prom_tree_update(drmach_detach_board, bp);
3431 
3432 	if (retval == 0)
3433 		return (NULL);
3434 	else {
3435 		cmn_err(CE_WARN, "prom error: prom_starfire_rm_brd(%d) "
3436 			"returned %d", bp->bnum, retval);
3437 		return (drerr_new(1, ESTF_DEPROBE, "%s", bp->cm.name));
3438 	}
3439 }
3440 
3441 /*ARGSUSED*/
3442 static sbd_error_t *
3443 drmach_pt_juggle_bootproc(drmachid_t id, drmach_opts_t *opts)
3444 {
3445 	drmach_device_t	*cpu;
3446 	sbd_error_t	*err;
3447 
3448 	if (!DRMACH_IS_CPU_ID(id))
3449 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3450 	cpu = id;
3451 
3452 	mutex_enter(&cpu_lock);
3453 
3454 	err = drmach_cpu_juggle_bootproc(cpu);
3455 
3456 	mutex_exit(&cpu_lock);
3457 
3458 	return (err);
3459 }
3460 
3461 /*ARGSUSED*/
3462 static sbd_error_t *
3463 drmach_pt_dump_pdainfo(drmachid_t id, drmach_opts_t *opts)
3464 {
3465 	drmach_board_t	*bp;
3466 	int		board;
3467 	int		i;
3468 	pda_handle_t	ph;
3469 	board_desc_t	*bdesc;
3470 
3471 	if (!DRMACH_IS_BOARD_ID(id))
3472 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3473 	bp = id;
3474 	board = bp->bnum;
3475 
3476 	ph = drmach_pda_open();
3477 	if (ph == NULL)
3478 		return (DRMACH_INTERNAL_ERROR());
3479 
3480 	if (pda_board_present(ph, board) == 0) {
3481 		cmn_err(CE_CONT, "board %d is MISSING\n", board);
3482 		pda_close(ph);
3483 		return (DRMACH_INTERNAL_ERROR());
3484 	}
3485 
3486 	cmn_err(CE_CONT, "board %d is PRESENT\n", board);
3487 
3488 	bdesc = (board_desc_t *)pda_get_board_info(ph, board);
3489 	if (bdesc == NULL) {
3490 		cmn_err(CE_CONT,
3491 			"no board descriptor found for board %d\n",
3492 			board);
3493 		pda_close(ph);
3494 		return (DRMACH_INTERNAL_ERROR());
3495 	}
3496 
3497 	/* make sure definition in platmod is in sync with pda */
3498 	ASSERT(MAX_PROCMODS == MAX_CPU_UNITS_PER_BOARD);
3499 
3500 	for (i = 0; i < MAX_PROCMODS; i++) {
3501 		if (BDA_NBL(bdesc->bda_proc, i) == BDAN_GOOD)
3502 			cmn_err(CE_CONT,
3503 				"proc %d.%d PRESENT\n", board, i);
3504 		else
3505 			cmn_err(CE_CONT,
3506 				"proc %d.%d MISSING\n", board, i);
3507 	}
3508 
3509 	for (i = 0; i < MAX_MGROUPS; i++) {
3510 		if (BDA_NBL(bdesc->bda_mgroup, i) == BDAN_GOOD)
3511 			cmn_err(CE_CONT,
3512 				"mgroup %d.%d PRESENT\n", board, i);
3513 		else
3514 			cmn_err(CE_CONT,
3515 				"mgroup %d.%d MISSING\n", board, i);
3516 	}
3517 
3518 	/* make sure definition in platmod is in sync with pda */
3519 	ASSERT(MAX_IOCS == MAX_IO_UNITS_PER_BOARD);
3520 
3521 	for (i = 0; i < MAX_IOCS; i++) {
3522 		int	s;
3523 
3524 		if (BDA_NBL(bdesc->bda_ioc, i) == BDAN_GOOD) {
3525 			cmn_err(CE_CONT,
3526 				"ioc %d.%d PRESENT\n", board, i);
3527 			for (s = 0; s < MAX_SLOTS_PER_IOC; s++) {
3528 				if (BDA_NBL(bdesc->bda_ios[i], s) != BDAN_GOOD)
3529 					continue;
3530 				cmn_err(CE_CONT,
3531 					"..scard %d.%d.%d PRESENT\n",
3532 					board, i, s);
3533 			}
3534 		} else {
3535 			cmn_err(CE_CONT,
3536 				"ioc %d.%d MISSING\n",
3537 				board, i);
3538 		}
3539 	}
3540 
3541 	cmn_err(CE_CONT,
3542 		"board %d memsize = %d pages\n",
3543 		board, pda_get_mem_size(ph, board));
3544 
3545 	pda_close(ph);
3546 
3547 	return (NULL);
3548 }
3549 
3550 /*ARGSUSED*/
3551 sbd_error_t *
3552 drmach_pt_readmem(drmachid_t id, drmach_opts_t *opts)
3553 {
3554 	struct memlist	*ml;
3555 	uint64_t	src_pa;
3556 	uint64_t	dst_pa;
3557 	uint64_t	dst;
3558 
3559 	dst_pa = va_to_pa(&dst);
3560 
3561 	memlist_read_lock();
3562 	for (ml = phys_install; ml; ml = ml->next) {
3563 		uint64_t	nbytes;
3564 
3565 		src_pa = ml->address;
3566 		nbytes = ml->size;
3567 
3568 		while (nbytes != 0ull) {
3569 
3570 			/* copy 32 bytes at arc_pa to dst_pa */
3571 			bcopy32_il(src_pa, dst_pa);
3572 
3573 			/* increment by 32 bytes */
3574 			src_pa += (4 * sizeof (uint64_t));
3575 
3576 			/* decrement by 32 bytes */
3577 			nbytes -= (4 * sizeof (uint64_t));
3578 		}
3579 	}
3580 	memlist_read_unlock();
3581 
3582 	return (NULL);
3583 }
3584 
3585 static struct {
3586 	const char	*name;
3587 	sbd_error_t	*(*handler)(drmachid_t id, drmach_opts_t *opts);
3588 } drmach_pt_arr[] = {
3589 	{ "juggle",		drmach_pt_juggle_bootproc	},
3590 	{ "pda",		drmach_pt_dump_pdainfo		},
3591 	{ "readmem",		drmach_pt_readmem		},
3592 
3593 	/* the following line must always be last */
3594 	{ NULL,			NULL				}
3595 };
3596 
3597 /*ARGSUSED*/
3598 sbd_error_t *
3599 drmach_passthru(drmachid_t id, drmach_opts_t *opts)
3600 {
3601 	int		i;
3602 	sbd_error_t	*err;
3603 
3604 	i = 0;
3605 	while (drmach_pt_arr[i].name != NULL) {
3606 		int len = strlen(drmach_pt_arr[i].name);
3607 
3608 		if (strncmp(drmach_pt_arr[i].name, opts->copts, len) == 0)
3609 			break;
3610 
3611 		i += 1;
3612 	}
3613 
3614 	if (drmach_pt_arr[i].name == NULL)
3615 		err = drerr_new(0, ESTF_UNKPTCMD, opts->copts);
3616 	else
3617 		err = (*drmach_pt_arr[i].handler)(id, opts);
3618 
3619 	return (err);
3620 }
3621 
3622 sbd_error_t *
3623 drmach_release(drmachid_t id)
3624 {
3625 	drmach_common_t *cp;
3626 	if (!DRMACH_IS_DEVICE_ID(id))
3627 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3628 	cp = id;
3629 
3630 	return (cp->release(id));
3631 }
3632 
3633 sbd_error_t *
3634 drmach_status(drmachid_t id, drmach_status_t *stat)
3635 {
3636 	drmach_common_t *cp;
3637 
3638 	if (!DRMACH_IS_ID(id))
3639 		return (drerr_new(0, ESTF_NOTID, NULL));
3640 	cp = id;
3641 
3642 	return (cp->status(id, stat));
3643 }
3644 
3645 sbd_error_t *
3646 drmach_unconfigure(drmachid_t id, int flags)
3647 {
3648 	drmach_device_t	*dp;
3649 	dnode_t		 nodeid;
3650 	dev_info_t	*dip, *fdip = NULL;
3651 	uint_t 		ddi_flags;
3652 
3653 	if (!DRMACH_IS_DEVICE_ID(id))
3654 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3655 
3656 	dp = id;
3657 
3658 	nodeid = drmach_node_get_dnode(dp->node);
3659 	if (nodeid == OBP_NONODE)
3660 		return (DRMACH_INTERNAL_ERROR());
3661 
3662 	dip = e_ddi_nodeid_to_dip(nodeid);
3663 	if (dip == NULL)
3664 		return (NULL);
3665 
3666 	/*
3667 	 * Branch already held, so hold acquired in
3668 	 * e_ddi_nodeid_to_dip() can be released
3669 	 */
3670 	ddi_release_devi(dip);
3671 
3672 	ddi_flags = 0;
3673 
3674 	if (flags & DRMACH_DEVI_REMOVE)
3675 		ddi_flags |= DEVI_BRANCH_DESTROY | DEVI_BRANCH_EVENT;
3676 
3677 	/*
3678 	 * Force flag is no longer necessary. See starcat/io/drmach.c
3679 	 * for details.
3680 	 */
3681 	ASSERT(e_ddi_branch_held(dip));
3682 	if (e_ddi_branch_unconfigure(dip, &fdip, ddi_flags)) {
3683 		sbd_error_t	*err;
3684 		char		*path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3685 
3686 		/*
3687 		 * If non-NULL, fdip is returned held and must be released.
3688 		 */
3689 		if (fdip != NULL) {
3690 			(void) ddi_pathname(fdip, path);
3691 			ndi_rele_devi(fdip);
3692 		} else {
3693 			(void) ddi_pathname(dip, path);
3694 		}
3695 
3696 		err = drerr_new(1, ESTF_DRVFAIL, path);
3697 
3698 		kmem_free(path, MAXPATHLEN);
3699 
3700 		return (err);
3701 	}
3702 
3703 	return (NULL);
3704 }
3705 
3706 /*
3707  * drmach interfaces to legacy Starfire platmod logic
3708  * linkage via runtime symbol look up, called from plat_cpu_power*
3709  */
3710 
3711 /*
3712  * Start up a cpu.  It is possible that we're attempting to restart
3713  * the cpu after an UNCONFIGURE in which case the cpu will be
3714  * spinning in its cache.  So, all we have to do is wakeup him up.
3715  * Under normal circumstances the cpu will be coming from a previous
3716  * CONNECT and thus will be spinning in OBP.  In both cases, the
3717  * startup sequence is the same.
3718  */
3719 int
3720 drmach_cpu_poweron(struct cpu *cp)
3721 {
3722 	DRMACH_PR("drmach_cpu_poweron: starting cpuid %d\n", cp->cpu_id);
3723 
3724 	ASSERT(MUTEX_HELD(&cpu_lock));
3725 
3726 	if (drmach_cpu_start(cp) != 0)
3727 		return (EBUSY);
3728 	else
3729 		return (0);
3730 }
3731 
3732 int
3733 drmach_cpu_poweroff(struct cpu *cp)
3734 {
3735 	int		ntries, cnt;
3736 	processorid_t	cpuid = cp->cpu_id;
3737 	void		drmach_cpu_shutdown_self(void);
3738 
3739 	DRMACH_PR("drmach_cpu_poweroff: stopping cpuid %d\n", cp->cpu_id);
3740 
3741 	ASSERT(MUTEX_HELD(&cpu_lock));
3742 
3743 	/*
3744 	 * Capture all CPUs (except for detaching proc) to prevent
3745 	 * crosscalls to the detaching proc until it has cleared its
3746 	 * bit in cpu_ready_set.
3747 	 *
3748 	 * The CPU's remain paused and the prom_mutex is known to be free.
3749 	 * This prevents the x-trap victim from blocking when doing prom
3750 	 * IEEE-1275 calls at a high PIL level.
3751 	 */
3752 	promsafe_pause_cpus();
3753 
3754 	/*
3755 	 * Quiesce interrupts on the target CPU. We do this by setting
3756 	 * the CPU 'not ready'- (i.e. removing the CPU from cpu_ready_set) to
3757 	 * prevent it from receiving cross calls and cross traps.
3758 	 * This prevents the processor from receiving any new soft interrupts.
3759 	 */
3760 	mp_cpu_quiesce(cp);
3761 
3762 	/* setup xt_mb, will be cleared by drmach_shutdown_asm when ready */
3763 	drmach_xt_mb[cpuid] = 0x80;
3764 
3765 	xt_one_unchecked(cpuid, (xcfunc_t *)idle_stop_xcall,
3766 		(uint64_t)drmach_cpu_shutdown_self, NULL);
3767 
3768 	ntries = drmach_cpu_ntries;
3769 	cnt = 0;
3770 	while (drmach_xt_mb[cpuid] && ntries) {
3771 		DELAY(drmach_cpu_delay);
3772 		ntries--;
3773 		cnt++;
3774 	}
3775 
3776 	drmach_xt_mb[cpuid] = 0;	/* steal the cache line back */
3777 
3778 	start_cpus();
3779 
3780 	DRMACH_PR("waited %d out of %d tries for "
3781 		"drmach_cpu_shutdown_self on cpu%d",
3782 		drmach_cpu_ntries - ntries, drmach_cpu_ntries, cp->cpu_id);
3783 
3784 	drmach_cpu_obp_detach(cpuid);
3785 
3786 	CPU_SIGNATURE(OS_SIG, SIGST_DETACHED, SIGSUBST_NULL, cpuid);
3787 
3788 	return (0);
3789 }
3790 
3791 /*ARGSUSED*/
3792 int
3793 drmach_verify_sr(dev_info_t *dip, int sflag)
3794 {
3795 	return (0);
3796 }
3797 
3798 void
3799 drmach_suspend_last(void)
3800 {
3801 }
3802 
3803 void
3804 drmach_resume_first(void)
3805 {
3806 }
3807 
3808 /*
3809  * Log a DR sysevent.
3810  * Return value: 0 success, non-zero failure.
3811  */
3812 int
3813 drmach_log_sysevent(int board, char *hint, int flag, int verbose)
3814 {
3815 	sysevent_t			*ev;
3816 	sysevent_id_t			eid;
3817 	int				rv, km_flag;
3818 	sysevent_value_t		evnt_val;
3819 	sysevent_attr_list_t		*evnt_attr_list = NULL;
3820 	char				attach_pnt[MAXNAMELEN];
3821 
3822 	km_flag = (flag == SE_SLEEP) ? KM_SLEEP : KM_NOSLEEP;
3823 	attach_pnt[0] = '\0';
3824 	if (drmach_board_name(board, attach_pnt, MAXNAMELEN)) {
3825 		rv = -1;
3826 		goto logexit;
3827 	}
3828 	if (verbose)
3829 		DRMACH_PR("drmach_log_sysevent: %s %s, flag: %d, verbose: %d\n",
3830 			    attach_pnt, hint, flag, verbose);
3831 
3832 	if ((ev = sysevent_alloc(EC_DR, ESC_DR_AP_STATE_CHANGE,
3833 				    SUNW_KERN_PUB"dr", km_flag)) == NULL) {
3834 		rv = -2;
3835 		goto logexit;
3836 	}
3837 	evnt_val.value_type = SE_DATA_TYPE_STRING;
3838 	evnt_val.value.sv_string = attach_pnt;
3839 	if ((rv = sysevent_add_attr(&evnt_attr_list, DR_AP_ID,
3840 				    &evnt_val, km_flag)) != 0)
3841 		goto logexit;
3842 
3843 	evnt_val.value_type = SE_DATA_TYPE_STRING;
3844 	evnt_val.value.sv_string = hint;
3845 	if ((rv = sysevent_add_attr(&evnt_attr_list, DR_HINT,
3846 				    &evnt_val, km_flag)) != 0) {
3847 		sysevent_free_attr(evnt_attr_list);
3848 		goto logexit;
3849 	}
3850 
3851 	(void) sysevent_attach_attributes(ev, evnt_attr_list);
3852 
3853 	/*
3854 	 * Log the event but do not sleep waiting for its
3855 	 * delivery. This provides insulation from syseventd.
3856 	 */
3857 	rv = log_sysevent(ev, SE_NOSLEEP, &eid);
3858 
3859 logexit:
3860 	if (ev)
3861 		sysevent_free(ev);
3862 	if ((rv != 0) && verbose)
3863 		cmn_err(CE_WARN,
3864 			    "drmach_log_sysevent failed (rv %d) for %s  %s\n",
3865 			    rv, attach_pnt, hint);
3866 
3867 	return (rv);
3868 }
3869 
3870 /*ARGSUSED*/
3871 int
3872 drmach_allow_memrange_modify(drmachid_t id)
3873 {
3874 	return (1);	/* TRUE */
3875 }
3876