xref: /titanic_51/usr/src/uts/sun4u/starfire/io/drmach.c (revision d2ec6b54cfaa9a3bf7fc816dc44fabab46f9ecc4)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/debug.h>
29 #include <sys/types.h>
30 #include <sys/varargs.h>
31 #include <sys/errno.h>
32 #include <sys/cred.h>
33 #include <sys/dditypes.h>
34 #include <sys/devops.h>
35 #include <sys/modctl.h>
36 #include <sys/poll.h>
37 #include <sys/conf.h>
38 #include <sys/ddi.h>
39 #include <sys/sunddi.h>
40 #include <sys/sunndi.h>
41 #include <sys/ndi_impldefs.h>
42 #include <sys/stat.h>
43 #include <sys/kmem.h>
44 #include <sys/vmem.h>
45 #include <sys/processor.h>
46 #include <sys/spitregs.h>
47 #include <sys/cpuvar.h>
48 #include <sys/cpupart.h>
49 #include <sys/mem_config.h>
50 #include <sys/ddi_impldefs.h>
51 #include <sys/systm.h>
52 #include <sys/machsystm.h>
53 #include <sys/autoconf.h>
54 #include <sys/cmn_err.h>
55 #include <sys/sysmacros.h>
56 #include <sys/x_call.h>
57 #include <sys/promif.h>
58 #include <sys/prom_plat.h>
59 #include <sys/membar.h>
60 #include <vm/seg_kmem.h>
61 #include <sys/mem_cage.h>
62 #include <sys/stack.h>
63 #include <sys/archsystm.h>
64 #include <vm/hat_sfmmu.h>
65 #include <sys/pte.h>
66 #include <sys/mmu.h>
67 #include <sys/cpu_module.h>
68 #include <sys/obpdefs.h>
69 #include <sys/note.h>
70 
71 #include <sys/starfire.h>	/* plat_max_... decls */
72 #include <sys/cvc.h>
73 #include <sys/cpu_sgnblk_defs.h>
74 #include <sys/drmach.h>
75 #include <sys/dr_util.h>
76 #include <sys/pda.h>
77 
78 #include <sys/sysevent.h>
79 #include <sys/sysevent/dr.h>
80 #include <sys/sysevent/eventdefs.h>
81 
82 
83 extern void		bcopy32_il(uint64_t, uint64_t);
84 extern void		flush_ecache_il(
85 				uint64_t physaddr, int size, int linesz);
86 extern uint_t		ldphysio_il(uint64_t physaddr);
87 extern void		stphysio_il(uint64_t physaddr, uint_t value);
88 
89 extern uint64_t		mc_get_mem_alignment(void);
90 extern uint64_t		mc_get_asr_addr(pnode_t);
91 extern uint64_t		mc_get_idle_addr(pnode_t);
92 extern uint64_t		mc_get_alignment_mask(pnode_t);
93 extern int		mc_read_asr(pnode_t, uint_t *);
94 extern int		mc_write_asr(pnode_t, uint_t);
95 extern uint64_t		mc_asr_to_pa(uint_t);
96 extern uint_t		mc_pa_to_asr(uint_t, uint64_t);
97 
98 extern int		pc_madr_add(int, int, int, int);
99 
100 typedef struct {
101 	struct drmach_node	*node;
102 	void			*data;
103 } drmach_node_walk_args_t;
104 
105 typedef struct drmach_node {
106 	void		*here;
107 
108 	pnode_t		 (*get_dnode)(struct drmach_node *node);
109 	int		 (*walk)(struct drmach_node *node, void *data,
110 				int (*cb)(drmach_node_walk_args_t *args));
111 } drmach_node_t;
112 
113 typedef struct {
114 	int		 min_index;
115 	int		 max_index;
116 	int		 arr_sz;
117 	drmachid_t	*arr;
118 } drmach_array_t;
119 
120 typedef struct {
121 	void		*isa;
122 
123 	sbd_error_t	*(*release)(drmachid_t);
124 	sbd_error_t	*(*status)(drmachid_t, drmach_status_t *);
125 
126 	char		 name[MAXNAMELEN];
127 } drmach_common_t;
128 
129 typedef struct {
130 	drmach_common_t	 cm;
131 	int		 bnum;
132 	int		 assigned;
133 	int		 powered;
134 	int		 connect_cpuid;
135 	int		 cond;
136 	drmach_node_t	*tree;
137 	drmach_array_t	*devices;
138 } drmach_board_t;
139 
140 typedef struct {
141 	drmach_common_t	 cm;
142 	drmach_board_t	*bp;
143 	int		 unum;
144 	int		 busy;
145 	int		 powered;
146 	const char	*type;
147 	drmach_node_t	*node;
148 } drmach_device_t;
149 
150 typedef struct {
151 	int		 flags;
152 	drmach_device_t	*dp;
153 	sbd_error_t	*err;
154 	dev_info_t	*dip;
155 } drmach_config_args_t;
156 
157 typedef struct {
158 	uint64_t	 idle_addr;
159 	drmach_device_t	*mem;
160 } drmach_mc_idle_script_t;
161 
162 typedef struct {
163 	uint64_t	masr_addr;
164 	uint_t		masr;
165 	uint_t		_filler;
166 } drmach_rename_script_t;
167 
168 typedef struct {
169 	void		(*run)(void *arg);
170 	caddr_t		data;
171 	pda_handle_t	*ph;
172 	struct memlist	*c_ml;
173 	uint64_t	s_copybasepa;
174 	uint64_t	t_copybasepa;
175 	drmach_device_t	*restless_mc;	/* diagnostic output */
176 } drmach_copy_rename_program_t;
177 
178 typedef enum {
179 	DO_IDLE,
180 	DO_UNIDLE,
181 	DO_PAUSE,
182 	DO_UNPAUSE
183 } drmach_iopc_op_t;
184 
185 typedef struct {
186 	drmach_board_t	*obj;
187 	int		 ndevs;
188 	void		*a;
189 	sbd_error_t	*(*found)(void *a, const char *, int, drmachid_t);
190 	sbd_error_t	*err;
191 } drmach_board_cb_data_t;
192 
193 static caddr_t		 drmach_shutdown_va;
194 
195 static int		 drmach_initialized;
196 static drmach_array_t	*drmach_boards;
197 
198 static int		 drmach_cpu_delay = 100;
199 static int		 drmach_cpu_ntries = 50000;
200 
201 volatile uchar_t	*drmach_xt_mb;
202 
203 /*
204  * Do not change the drmach_shutdown_mbox structure without
205  * considering the drmach_shutdown_asm assembly language code.
206  */
207 struct drmach_shutdown_mbox {
208 	uint64_t	estack;
209 	uint64_t	flushaddr;
210 	int		size;
211 	int		linesize;
212 	uint64_t	physaddr;
213 };
214 struct drmach_shutdown_mbox	*drmach_shutdown_asm_mbox;
215 static sbd_error_t	*drmach_device_new(drmach_node_t *,
216 				drmach_board_t *, drmach_device_t **);
217 static sbd_error_t	*drmach_cpu_new(drmach_device_t *);
218 static sbd_error_t	*drmach_mem_new(drmach_device_t *);
219 static sbd_error_t	*drmach_io_new(drmach_device_t *);
220 
221 extern struct cpu	*SIGBCPU;
222 
223 #ifdef DEBUG
224 
225 #define	DRMACH_PR		if (drmach_debug) printf
226 int drmach_debug = 0;		 /* set to non-zero to enable debug messages */
227 #else
228 
229 #define	DRMACH_PR		_NOTE(CONSTANTCONDITION) if (0) printf
230 #endif /* DEBUG */
231 
232 #define	DRMACH_OBJ(id)		((drmach_common_t *)id)
233 
234 #define	DRMACH_IS_BOARD_ID(id)	\
235 	((id != 0) &&		\
236 	(DRMACH_OBJ(id)->isa == (void *)drmach_board_new))
237 
238 #define	DRMACH_IS_CPU_ID(id)	\
239 	((id != 0) &&		\
240 	(DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new))
241 
242 #define	DRMACH_IS_MEM_ID(id)	\
243 	((id != 0) &&		\
244 	(DRMACH_OBJ(id)->isa == (void *)drmach_mem_new))
245 
246 #define	DRMACH_IS_IO_ID(id)	\
247 	((id != 0) &&		\
248 	(DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
249 
250 #define	DRMACH_IS_DEVICE_ID(id)					\
251 	((id != 0) &&						\
252 	(DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new ||	\
253 	    DRMACH_OBJ(id)->isa == (void *)drmach_mem_new ||	\
254 	    DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
255 
256 #define	DRMACH_IS_ID(id)					\
257 	((id != 0) &&						\
258 	(DRMACH_OBJ(id)->isa == (void *)drmach_board_new ||	\
259 	    DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new ||	\
260 	    DRMACH_OBJ(id)->isa == (void *)drmach_mem_new ||	\
261 	    DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
262 
263 #define	DRMACH_CPUID2BNUM(cpuid) \
264 	((cpuid) / MAX_CPU_UNITS_PER_BOARD)
265 
266 #define	DRMACH_INTERNAL_ERROR() \
267 	drerr_new(1, ESTF_INTERNAL, drmach_ie_fmt, __LINE__)
268 static char		*drmach_ie_fmt = "drmach.c %d";
269 
270 static struct {
271 	const char	 *name;
272 	const char	 *type;
273 	sbd_error_t	 *(*new)(drmach_device_t *);
274 } name2type[] = {
275 	{ "SUNW,UltraSPARC",	DRMACH_DEVTYPE_CPU,  drmach_cpu_new },
276 	{ "mem-unit",		DRMACH_DEVTYPE_MEM,  drmach_mem_new },
277 	{ "pci",		DRMACH_DEVTYPE_PCI,  drmach_io_new  },
278 	{ "sbus",		DRMACH_DEVTYPE_SBUS, drmach_io_new  },
279 };
280 
281 /* node types to cleanup when a board is unconfigured */
282 #define	MISC_COUNTER_TIMER_DEVNAME	"counter-timer"
283 #define	MISC_PERF_COUNTER_DEVNAME	"perf-counter"
284 
285 /* utility */
286 #define	MBYTE	(1048576ull)
287 
288 /*
289  * This is necessary because the CPU support needs
290  * to call cvc_assign_iocpu.
291  */
292 #ifndef lint
293 char _depends_on[] = "drv/cvc";
294 #endif  /* lint */
295 
296 /*
297  * drmach autoconfiguration data structures and interfaces
298  */
299 
300 extern struct mod_ops mod_miscops;
301 
302 static struct modlmisc modlmisc = {
303 	&mod_miscops,
304 	"Sun Enterprise 10000 DR %I%"
305 };
306 
307 static struct modlinkage modlinkage = {
308 	MODREV_1,
309 	(void *)&modlmisc,
310 	NULL
311 };
312 
313 static kmutex_t drmach_i_lock;
314 
315 int
316 _init(void)
317 {
318 	int err;
319 
320 	/* check that we have the correct version of obp */
321 	if (prom_test("SUNW,UE10000,add-brd") != 0) {
322 
323 		cmn_err(CE_WARN, "!OBP/SSP upgrade is required to enable "
324 		    "DR Functionality");
325 
326 		return (-1);
327 	}
328 
329 	mutex_init(&drmach_i_lock, NULL, MUTEX_DRIVER, NULL);
330 
331 	drmach_xt_mb = (uchar_t *)vmem_alloc(static_alloc_arena,
332 	    NCPU * sizeof (uchar_t), VM_SLEEP);
333 	drmach_shutdown_asm_mbox = (struct drmach_shutdown_mbox *)
334 	    vmem_alloc(static_alloc_arena, sizeof (struct drmach_shutdown_mbox),
335 	    VM_SLEEP);
336 
337 	if ((err = mod_install(&modlinkage)) != 0) {
338 		mutex_destroy(&drmach_i_lock);
339 		vmem_free(static_alloc_arena, (void *)drmach_xt_mb,
340 		    NCPU * sizeof (uchar_t));
341 		vmem_free(static_alloc_arena, (void *)drmach_shutdown_asm_mbox,
342 		    sizeof (struct drmach_shutdown_mbox));
343 	}
344 
345 	return (err);
346 }
347 
348 int
349 _fini(void)
350 {
351 	static int drmach_fini(void);
352 
353 	if (drmach_fini())
354 		return (DDI_FAILURE);
355 	else
356 		return (mod_remove(&modlinkage));
357 }
358 
359 int
360 _info(struct modinfo *modinfop)
361 {
362 	return (mod_info(&modlinkage, modinfop));
363 }
364 
365 static pnode_t
366 drmach_node_obp_get_dnode(drmach_node_t *np)
367 {
368 	return ((pnode_t)(uintptr_t)np->here);
369 }
370 
371 static int
372 drmach_node_obp_walk(drmach_node_t *np, void *data,
373 		int (*cb)(drmach_node_walk_args_t *args))
374 {
375 	pnode_t			nodeid;
376 	int			rv;
377 	drmach_node_walk_args_t	args;
378 
379 	/* initialized args structure for callback */
380 	args.node = np;
381 	args.data = data;
382 
383 	nodeid = prom_childnode(prom_rootnode());
384 
385 	/* save our new position with in the tree */
386 	np->here = (void *)(uintptr_t)nodeid;
387 
388 	rv = 0;
389 	while (nodeid != OBP_NONODE) {
390 		rv = (*cb)(&args);
391 		if (rv)
392 			break;
393 
394 		nodeid = prom_nextnode(nodeid);
395 
396 		/* save our new position with in the tree */
397 		np->here = (void *)(uintptr_t)nodeid;
398 	}
399 
400 	return (rv);
401 }
402 
403 static drmach_node_t *
404 drmach_node_new(void)
405 {
406 	drmach_node_t *np;
407 
408 	np = kmem_zalloc(sizeof (drmach_node_t), KM_SLEEP);
409 
410 	np->get_dnode = drmach_node_obp_get_dnode;
411 	np->walk = drmach_node_obp_walk;
412 
413 	return (np);
414 }
415 
416 static void
417 drmach_node_dispose(drmach_node_t *np)
418 {
419 	kmem_free(np, sizeof (*np));
420 }
421 
422 static dev_info_t *
423 drmach_node_get_dip(drmach_node_t *np)
424 {
425 	pnode_t nodeid;
426 
427 	nodeid = np->get_dnode(np);
428 	if (nodeid == OBP_NONODE)
429 		return (NULL);
430 	else {
431 		dev_info_t *dip;
432 
433 		/* The root node doesn't have to be held */
434 		dip = e_ddi_nodeid_to_dip(nodeid);
435 		if (dip) {
436 			/*
437 			 * Branch rooted at dip is already held, so release
438 			 * hold acquired in e_ddi_nodeid_to_dip()
439 			 */
440 			ddi_release_devi(dip);
441 			ASSERT(e_ddi_branch_held(dip));
442 		}
443 
444 		return (dip);
445 	}
446 	/*NOTREACHED*/
447 }
448 
449 static pnode_t
450 drmach_node_get_dnode(drmach_node_t *np)
451 {
452 	return (np->get_dnode(np));
453 }
454 
455 static int
456 drmach_node_walk(drmach_node_t *np, void *param,
457 		int (*cb)(drmach_node_walk_args_t *args))
458 {
459 	return (np->walk(np, param, cb));
460 }
461 
462 static int
463 drmach_node_get_prop(drmach_node_t *np, char *name, void *buf)
464 {
465 	pnode_t	nodeid;
466 	int	rv;
467 
468 	nodeid = np->get_dnode(np);
469 	if (nodeid == OBP_NONODE)
470 		rv = -1;
471 	else if (prom_getproplen(nodeid, (caddr_t)name) < 0)
472 		rv = -1;
473 	else {
474 		(void) prom_getprop(nodeid, (caddr_t)name, (caddr_t)buf);
475 		rv = 0;
476 	}
477 
478 	return (rv);
479 }
480 
481 static int
482 drmach_node_get_proplen(drmach_node_t *np, char *name, int *len)
483 {
484 	pnode_t	 nodeid;
485 	int	 rv;
486 
487 	nodeid = np->get_dnode(np);
488 	if (nodeid == OBP_NONODE)
489 		rv = -1;
490 	else {
491 		*len = prom_getproplen(nodeid, (caddr_t)name);
492 		rv = (*len < 0 ? -1 : 0);
493 	}
494 
495 	return (rv);
496 }
497 
498 static drmachid_t
499 drmach_node_dup(drmach_node_t *np)
500 {
501 	drmach_node_t *dup;
502 
503 	dup = drmach_node_new();
504 	dup->here = np->here;
505 
506 	return (dup);
507 }
508 
509 /*
510  * drmach_array provides convenient array construction, access,
511  * bounds checking and array destruction logic.
512  */
513 
514 static drmach_array_t *
515 drmach_array_new(int min_index, int max_index)
516 {
517 	drmach_array_t *arr;
518 
519 	arr = kmem_zalloc(sizeof (drmach_array_t), KM_SLEEP);
520 
521 	arr->arr_sz = (max_index - min_index + 1) * sizeof (void *);
522 	if (arr->arr_sz > 0) {
523 		arr->min_index = min_index;
524 		arr->max_index = max_index;
525 
526 		arr->arr = kmem_zalloc(arr->arr_sz, KM_SLEEP);
527 		return (arr);
528 	} else {
529 		kmem_free(arr, sizeof (*arr));
530 		return (0);
531 	}
532 }
533 
534 static int
535 drmach_array_set(drmach_array_t *arr, int idx, drmachid_t val)
536 {
537 	if (idx < arr->min_index || idx > arr->max_index)
538 		return (-1);
539 	else {
540 		arr->arr[idx - arr->min_index] = val;
541 		return (0);
542 	}
543 	/*NOTREACHED*/
544 }
545 
546 static int
547 drmach_array_get(drmach_array_t *arr, int idx, drmachid_t *val)
548 {
549 	if (idx < arr->min_index || idx > arr->max_index)
550 		return (-1);
551 	else {
552 		*val = arr->arr[idx - arr->min_index];
553 		return (0);
554 	}
555 	/*NOTREACHED*/
556 }
557 
558 static int
559 drmach_array_first(drmach_array_t *arr, int *idx, drmachid_t *val)
560 {
561 	int rv;
562 
563 	*idx = arr->min_index;
564 	while ((rv = drmach_array_get(arr, *idx, val)) == 0 && *val == NULL)
565 		*idx += 1;
566 
567 	return (rv);
568 }
569 
570 static int
571 drmach_array_next(drmach_array_t *arr, int *idx, drmachid_t *val)
572 {
573 	int rv;
574 
575 	*idx += 1;
576 	while ((rv = drmach_array_get(arr, *idx, val)) == 0 && *val == NULL)
577 		*idx += 1;
578 
579 	return (rv);
580 }
581 
582 static void
583 drmach_array_dispose(drmach_array_t *arr, void (*disposer)(drmachid_t))
584 {
585 	drmachid_t	val;
586 	int		idx;
587 	int		rv;
588 
589 	rv = drmach_array_first(arr, &idx, &val);
590 	while (rv == 0) {
591 		(*disposer)(val);
592 		rv = drmach_array_next(arr, &idx, &val);
593 	}
594 
595 	kmem_free(arr->arr, arr->arr_sz);
596 	kmem_free(arr, sizeof (*arr));
597 }
598 
599 /*ARGSUSED*/
600 static int
601 drmach_prom_select(pnode_t nodeid, void *arg, uint_t flags)
602 {
603 	int			rprop[64];
604 	pnode_t			saved;
605 	drmach_config_args_t	*ap = (drmach_config_args_t *)arg;
606 	drmach_device_t		*dp = ap->dp;
607 	sbd_error_t		*err;
608 
609 	saved = drmach_node_get_dnode(dp->node);
610 
611 	if (nodeid != saved)
612 		return (DDI_FAILURE);
613 
614 	if (saved == OBP_NONODE) {
615 		err = DRMACH_INTERNAL_ERROR();
616 		DRERR_SET_C(&ap->err, &err);
617 		return (DDI_FAILURE);
618 	}
619 
620 	if (prom_getprop(nodeid, OBP_REG, (caddr_t)rprop) <= 0) {
621 		return (DDI_FAILURE);
622 	}
623 
624 	return (DDI_SUCCESS);
625 }
626 
627 /*ARGSUSED*/
628 static void
629 drmach_branch_callback(dev_info_t *rdip, void *arg, uint_t flags)
630 {
631 	drmach_config_args_t	*ap = (drmach_config_args_t *)arg;
632 
633 	ASSERT(ap->dip == NULL);
634 
635 	ap->dip = rdip;
636 }
637 
638 sbd_error_t *
639 drmach_configure(drmachid_t id, int flags)
640 {
641 	drmach_device_t		*dp;
642 	sbd_error_t		*err;
643 	drmach_config_args_t	ca;
644 	devi_branch_t		b = {0};
645 	dev_info_t		*fdip = NULL;
646 
647 	if (!DRMACH_IS_DEVICE_ID(id))
648 		return (drerr_new(0, ESTF_INAPPROP, NULL));
649 	dp = id;
650 
651 	ca.dp = dp;
652 	ca.flags = flags;
653 	ca.err = NULL;		/* will be set if error detected */
654 	ca.dip = NULL;
655 
656 	b.arg = &ca;
657 	b.type = DEVI_BRANCH_PROM;
658 	b.create.prom_branch_select = drmach_prom_select;
659 	b.devi_branch_callback = drmach_branch_callback;
660 
661 	if (e_ddi_branch_create(ddi_root_node(), &b, &fdip,
662 	    DEVI_BRANCH_CHILD | DEVI_BRANCH_CONFIGURE) != 0) {
663 		char *path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
664 
665 		/*
666 		 * If non-NULL, fdip is returned held and must be released.
667 		 */
668 		if (fdip != NULL) {
669 			(void) ddi_pathname(fdip, path);
670 			ddi_release_devi(fdip);
671 		} else if (ca.dip != NULL) {
672 			/* safe to call ddi_pathname as dip already held */
673 			(void) ddi_pathname(ca.dip, path);
674 		} else {
675 			(void) strcpy(path, "<none>");
676 		}
677 
678 		err = drerr_new(1, ESTF_DRVFAIL, path);
679 		DRERR_SET_C(&ca.err, &err);
680 		kmem_free(path, MAXPATHLEN);
681 	}
682 
683 	return (ca.err);
684 }
685 
686 static sbd_error_t *
687 drmach_device_new(drmach_node_t *node,
688 	drmach_board_t *bp, drmach_device_t **dpp)
689 {
690 	int		 i;
691 	int		 rv;
692 	drmach_device_t	*dp;
693 	sbd_error_t	*err;
694 	char		 name[OBP_MAXDRVNAME];
695 
696 	rv = drmach_node_get_prop(node, OBP_NAME, name);
697 	if (rv) {
698 		/* every node is expected to have a name */
699 		err = drerr_new(1, ESTF_GETPROP,
700 			"PROM Node 0x%x: property %s",
701 			(uint_t)node->get_dnode(node), OBP_NAME);
702 
703 		return (err);
704 	}
705 
706 	/*
707 	 * The node currently being examined is not listed in the name2type[]
708 	 * array.  In this case, the node is no interest to drmach.  Both
709 	 * dp and err are initialized here to yield nothing (no device or
710 	 * error structure) for this case.
711 	 */
712 	for (i = 0; i < sizeof (name2type) / sizeof (name2type[0]); i++)
713 		if (strcmp(name2type[i].name, name) == 0)
714 			break;
715 
716 	if (i < sizeof (name2type) / sizeof (name2type[0])) {
717 		dp = kmem_zalloc(sizeof (drmach_device_t), KM_SLEEP);
718 
719 		dp->bp = bp;
720 		dp->unum = -1;
721 		dp->node = drmach_node_dup(node);
722 		dp->type = name2type[i].type;
723 
724 		err = (name2type[i].new)(dp);
725 		if (err) {
726 			drmach_node_dispose(node);
727 			kmem_free(dp, sizeof (*dp));
728 			dp = NULL;
729 		}
730 
731 		*dpp = dp;
732 		return (err);
733 	}
734 
735 	/*
736 	 * The node currently being examined is not listed in the name2type[]
737 	 * array.  In this case, the node is no interest to drmach.  Both
738 	 * dp and err are initialized here to yield nothing (no device or
739 	 * error structure) for this case.
740 	 */
741 	*dpp = NULL;
742 	return (NULL);
743 }
744 
745 static void
746 drmach_device_dispose(drmachid_t id)
747 {
748 	drmach_device_t *self = id;
749 
750 	if (self->node)
751 		drmach_node_dispose(self->node);
752 
753 	kmem_free(self, sizeof (*self));
754 }
755 
756 static sbd_error_t *
757 drmach_device_get_prop(drmach_device_t *dp, char *name, void *buf)
758 {
759 	sbd_error_t	*err = NULL;
760 	int		 rv;
761 
762 	rv = drmach_node_get_prop(dp->node, name, buf);
763 	if (rv) {
764 		err = drerr_new(1, ESTF_GETPROP,
765 			"%s::%s: property %s",
766 			dp->bp->cm.name, dp->cm.name, name);
767 	}
768 
769 	return (err);
770 }
771 
772 static sbd_error_t *
773 drmach_device_get_proplen(drmach_device_t *dp, char *name, int *len)
774 {
775 	sbd_error_t	*err = NULL;
776 	int		 rv;
777 
778 	rv = drmach_node_get_proplen(dp->node, name, len);
779 	if (rv) {
780 		err = drerr_new(1, ESTF_GETPROPLEN,
781 			"%s::%s: property %s",
782 			dp->bp->cm.name, dp->cm.name, name);
783 	}
784 
785 	return (err);
786 }
787 
788 static drmach_board_t *
789 drmach_board_new(int bnum)
790 {
791 	static sbd_error_t *drmach_board_release(drmachid_t);
792 	static sbd_error_t *drmach_board_status(drmachid_t, drmach_status_t *);
793 
794 	drmach_board_t	*bp;
795 
796 	bp = kmem_zalloc(sizeof (drmach_board_t), KM_SLEEP);
797 
798 	bp->cm.isa = (void *)drmach_board_new;
799 	bp->cm.release = drmach_board_release;
800 	bp->cm.status = drmach_board_status;
801 
802 	(void) drmach_board_name(bnum, bp->cm.name, sizeof (bp->cm.name));
803 
804 	bp->bnum = bnum;
805 	bp->devices = NULL;
806 	bp->connect_cpuid = -1;
807 	bp->tree = drmach_node_new();
808 	bp->assigned = !drmach_initialized;
809 	bp->powered = !drmach_initialized;
810 
811 	drmach_array_set(drmach_boards, bnum, bp);
812 	return (bp);
813 }
814 
815 static void
816 drmach_board_dispose(drmachid_t id)
817 {
818 	drmach_board_t *bp;
819 
820 	ASSERT(DRMACH_IS_BOARD_ID(id));
821 	bp = id;
822 
823 	if (bp->tree)
824 		drmach_node_dispose(bp->tree);
825 
826 	if (bp->devices)
827 		drmach_array_dispose(bp->devices, drmach_device_dispose);
828 
829 	kmem_free(bp, sizeof (*bp));
830 }
831 
832 static sbd_error_t *
833 drmach_board_status(drmachid_t id, drmach_status_t *stat)
834 {
835 	sbd_error_t	*err = NULL;
836 	drmach_board_t	*bp;
837 
838 	if (!DRMACH_IS_BOARD_ID(id))
839 		return (drerr_new(0, ESTF_INAPPROP, NULL));
840 	bp = id;
841 
842 	stat->assigned = bp->assigned;
843 	stat->powered = bp->powered;
844 	stat->busy = 0;			/* assume not busy */
845 	stat->configured = 0;		/* assume not configured */
846 	stat->empty = 0;
847 	stat->cond = bp->cond = SBD_COND_OK;
848 	strncpy(stat->type, "System Brd", sizeof (stat->type));
849 	stat->info[0] = '\0';
850 
851 	if (bp->devices) {
852 		int		 rv;
853 		int		 d_idx;
854 		drmachid_t	 d_id;
855 
856 		rv = drmach_array_first(bp->devices, &d_idx, &d_id);
857 		while (rv == 0) {
858 			drmach_status_t	d_stat;
859 
860 			err = drmach_status(d_id, &d_stat);
861 			if (err)
862 				break;
863 
864 			stat->busy |= d_stat.busy;
865 			stat->configured |= d_stat.configured;
866 
867 			rv = drmach_array_next(bp->devices, &d_idx, &d_id);
868 		}
869 	}
870 
871 	return (err);
872 }
873 
874 /* a simple routine to reduce redundancy of this common logic */
875 static pda_handle_t
876 drmach_pda_open(void)
877 {
878 	pda_handle_t ph;
879 
880 	ph = pda_open();
881 	if (ph == NULL) {
882 		/* catch in debug kernels */
883 		ASSERT(0);
884 		cmn_err(CE_WARN, "pda_open failed");
885 	}
886 
887 	return (ph);
888 }
889 
890 #ifdef DEBUG
891 int drmach_init_break = 0;
892 #endif
893 
894 static int
895 hold_rele_branch(dev_info_t *rdip, void *arg)
896 {
897 	int	i;
898 	int	*holdp = (int *)arg;
899 	char	*name = ddi_node_name(rdip);
900 
901 	/*
902 	 * For Starfire, we must be children of the root devinfo node
903 	 */
904 	ASSERT(ddi_get_parent(rdip) == ddi_root_node());
905 
906 	for (i = 0; i < sizeof (name2type) / sizeof (name2type[0]); i++)
907 		if (strcmp(name2type[i].name, name) == 0)
908 			break;
909 
910 	if (i == sizeof (name2type) / sizeof (name2type[0])) {
911 		/* Not of interest to us */
912 		return (DDI_WALK_PRUNECHILD);
913 	}
914 
915 	if (*holdp) {
916 		ASSERT(!e_ddi_branch_held(rdip));
917 		e_ddi_branch_hold(rdip);
918 	} else {
919 		ASSERT(e_ddi_branch_held(rdip));
920 		e_ddi_branch_rele(rdip);
921 	}
922 
923 	return (DDI_WALK_PRUNECHILD);
924 }
925 
926 static int
927 drmach_init(void)
928 {
929 	pnode_t		nodeid;
930 	dev_info_t	*rdip;
931 	int		hold, circ;
932 
933 #ifdef DEBUG
934 	if (drmach_init_break)
935 		debug_enter("drmach_init: drmach_init_break set\n");
936 #endif
937 	mutex_enter(&drmach_i_lock);
938 	if (drmach_initialized) {
939 		mutex_exit(&drmach_i_lock);
940 		return (0);
941 	}
942 
943 	drmach_boards = drmach_array_new(0, MAX_BOARDS - 1);
944 
945 	nodeid = prom_childnode(prom_rootnode());
946 	do {
947 		int		 bnum;
948 		drmachid_t	 id;
949 
950 		bnum = -1;
951 		(void) prom_getprop(nodeid, OBP_BOARDNUM, (caddr_t)&bnum);
952 		if (bnum == -1)
953 			continue;
954 
955 		if (drmach_array_get(drmach_boards, bnum, &id) == -1) {
956 			cmn_err(CE_WARN, "OBP node 0x%x has"
957 				" invalid property value, %s=%d",
958 				nodeid, OBP_BOARDNUM, bnum);
959 
960 			/* clean up */
961 			drmach_array_dispose(
962 				drmach_boards, drmach_board_dispose);
963 
964 			mutex_exit(&drmach_i_lock);
965 			return (-1);
966 		} else if (id == NULL)
967 			(void) drmach_board_new(bnum);
968 	} while ((nodeid = prom_nextnode(nodeid)) != OBP_NONODE);
969 
970 	drmach_shutdown_va = vmem_alloc(heap_arena, PAGESIZE, VM_SLEEP);
971 
972 	/*
973 	 * Walk immediate children of devinfo root node and hold
974 	 * all devinfo branches of interest.
975 	 */
976 	hold = 1;
977 	rdip = ddi_root_node();
978 
979 	ndi_devi_enter(rdip, &circ);
980 	ddi_walk_devs(ddi_get_child(rdip), hold_rele_branch, &hold);
981 	ndi_devi_exit(rdip, circ);
982 
983 	drmach_initialized = 1;
984 
985 	mutex_exit(&drmach_i_lock);
986 
987 	return (0);
988 }
989 
990 static int
991 drmach_fini(void)
992 {
993 	dev_info_t	*rdip;
994 	int		hold, circ;
995 
996 	if (drmach_initialized) {
997 		int		busy = 0;
998 		int		rv;
999 		int		idx;
1000 		drmachid_t	id;
1001 
1002 		ASSERT(drmach_boards != NULL);
1003 
1004 		rv = drmach_array_first(drmach_boards, &idx, &id);
1005 		while (rv == 0) {
1006 			sbd_error_t	*err;
1007 			drmach_status_t stat;
1008 
1009 			err = drmach_board_status(id, &stat);
1010 			if (err) {
1011 				/* catch in debug kernels */
1012 				ASSERT(0);
1013 				sbd_err_clear(&err);
1014 				busy = 1;
1015 			} else
1016 				busy |= stat.busy;
1017 
1018 			rv = drmach_array_next(drmach_boards, &idx, &id);
1019 		}
1020 
1021 		if (busy)
1022 			return (-1);
1023 
1024 		drmach_array_dispose(drmach_boards, drmach_board_dispose);
1025 		drmach_boards = NULL;
1026 
1027 		vmem_free(heap_arena, drmach_shutdown_va, PAGESIZE);
1028 
1029 		/*
1030 		 * Walk immediate children of the root devinfo node
1031 		 * releasing holds acquired on branches in drmach_init()
1032 		 */
1033 		hold = 0;
1034 		rdip = ddi_root_node();
1035 
1036 		ndi_devi_enter(rdip, &circ);
1037 		ddi_walk_devs(ddi_get_child(rdip), hold_rele_branch, &hold);
1038 		ndi_devi_exit(rdip, circ);
1039 
1040 		mutex_destroy(&drmach_i_lock);
1041 
1042 		drmach_initialized = 0;
1043 	}
1044 	if (drmach_xt_mb != NULL) {
1045 		vmem_free(static_alloc_arena, (void *)drmach_xt_mb,
1046 		    NCPU * sizeof (uchar_t));
1047 	}
1048 	if (drmach_shutdown_asm_mbox != NULL) {
1049 		vmem_free(static_alloc_arena, (void *)drmach_shutdown_asm_mbox,
1050 		    sizeof (struct drmach_shutdown_mbox));
1051 	}
1052 	return (0);
1053 }
1054 
1055 static sbd_error_t *
1056 drmach_get_mc_asr_addr(drmachid_t id, uint64_t *pa)
1057 {
1058 	drmach_device_t	*dp;
1059 	pnode_t		nodeid;
1060 	uint64_t	addr;
1061 
1062 	if (!DRMACH_IS_MEM_ID(id))
1063 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1064 	dp = id;
1065 
1066 	nodeid = drmach_node_get_dnode(dp->node);
1067 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1068 		return (DRMACH_INTERNAL_ERROR());
1069 
1070 	addr = mc_get_asr_addr(nodeid);
1071 	if (addr == (uint64_t)-1)
1072 		return (DRMACH_INTERNAL_ERROR());
1073 
1074 	*pa = addr;
1075 	return (NULL);
1076 }
1077 
1078 static sbd_error_t *
1079 drmach_get_mc_idle_addr(drmachid_t id, uint64_t *pa)
1080 {
1081 	drmach_device_t	*dp;
1082 	pnode_t		nodeid;
1083 	uint64_t	addr;
1084 
1085 	if (!DRMACH_IS_MEM_ID(id))
1086 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1087 	dp = id;
1088 
1089 	nodeid = drmach_node_get_dnode(dp->node);
1090 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1091 		return (DRMACH_INTERNAL_ERROR());
1092 
1093 	addr = mc_get_idle_addr(nodeid);
1094 	if (addr == (uint64_t)-1)
1095 		return (DRMACH_INTERNAL_ERROR());
1096 
1097 	*pa = addr;
1098 	return (NULL);
1099 }
1100 
1101 static sbd_error_t *
1102 drmach_read_mc_asr(drmachid_t id, uint_t *mcregp)
1103 {
1104 	drmach_device_t	*dp;
1105 	pnode_t		 nodeid;
1106 	sbd_error_t	*err;
1107 
1108 	if (!DRMACH_IS_MEM_ID(id))
1109 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1110 	dp = id;
1111 
1112 	nodeid = drmach_node_get_dnode(dp->node);
1113 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1114 		err = DRMACH_INTERNAL_ERROR();
1115 	else if (mc_read_asr(nodeid, mcregp) == -1)
1116 		err = DRMACH_INTERNAL_ERROR();
1117 	else
1118 		err = NULL;
1119 
1120 	return (err);
1121 }
1122 
1123 static sbd_error_t *
1124 drmach_write_mc_asr(drmachid_t id, uint_t mcreg)
1125 {
1126 	drmach_device_t	*dp;
1127 	pnode_t		 nodeid;
1128 	sbd_error_t	*err;
1129 
1130 	if (!DRMACH_IS_MEM_ID(id))
1131 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1132 	dp = id;
1133 
1134 	nodeid = drmach_node_get_dnode(dp->node);
1135 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1136 		err = DRMACH_INTERNAL_ERROR();
1137 	else if (mc_write_asr(nodeid, mcreg) == -1)
1138 		err = DRMACH_INTERNAL_ERROR();
1139 	else
1140 		err = NULL;
1141 
1142 	return (err);
1143 }
1144 
1145 static sbd_error_t *
1146 drmach_prep_rename_script(drmach_device_t *s_mem, drmach_device_t *t_mem,
1147 	uint64_t t_slice_offset, caddr_t buf, int buflen)
1148 {
1149 	int			i, b, m;
1150 	drmach_mc_idle_script_t	*isp;
1151 	drmach_rename_script_t	*rsp;
1152 	int			s_bd, t_bd;
1153 	uint_t			s_masr, t_masr;
1154 	uint64_t		s_new_basepa, t_new_basepa;
1155 	int			b_idx, rv;
1156 	sbd_error_t		*err;
1157 	drmachid_t		 b_id;
1158 	drmach_board_t		*brd;
1159 
1160 #ifdef DEBUG
1161 	/*
1162 	 * Starfire CPU/MEM/IO boards have only one MC per board.
1163 	 * This function has been coded with that fact in mind.
1164 	 */
1165 	ASSERT(MAX_MEM_UNITS_PER_BOARD == 1);
1166 
1167 	/*
1168 	 * calculate the maximum space that could be consumed,
1169 	 * then verify the available buffer space is adequate.
1170 	 */
1171 	m  = sizeof (drmach_mc_idle_script_t *) * 2; /* two MCs */
1172 	b  = sizeof (drmach_rename_script_t *) * 3 * MAX_CPU_UNITS_PER_BOARD;
1173 	b += sizeof (drmach_rename_script_t *) * 3 * MAX_IO_UNITS_PER_BOARD;
1174 	b *= MAX_BOARDS;
1175 	b += sizeof (drmach_rename_script_t *) * 3;
1176 	b += sizeof (drmach_rename_script_t *) * 1;
1177 	ASSERT(m + b < buflen);
1178 #endif
1179 
1180 	/*
1181 	 * construct an array of MC idle register addresses of
1182 	 * both MCs.  The array is zero terminated -- as expected
1183 	 * by drmach_copy_rename_prog__relocatable().
1184 	 */
1185 	isp = (drmach_mc_idle_script_t *)buf;
1186 
1187 	/* source mc */
1188 	err = drmach_get_mc_idle_addr(s_mem, &isp->idle_addr);
1189 	if (err)
1190 		return (err);
1191 	isp->mem = s_mem;
1192 	isp += 1;
1193 
1194 	/* target mc */
1195 	err = drmach_get_mc_idle_addr(t_mem, &isp->idle_addr);
1196 	if (err)
1197 		return (err);
1198 	isp->mem = t_mem;
1199 	isp += 1;
1200 
1201 	/* terminator */
1202 	isp->idle_addr = 0;
1203 	isp->mem = NULL;
1204 	isp += 1;
1205 
1206 	/* fetch source mc asr register value */
1207 	err = drmach_read_mc_asr(s_mem, &s_masr);
1208 	if (err)
1209 		return (err);
1210 	else if (s_masr & STARFIRE_MC_INTERLEAVE_MASK) {
1211 		return (drerr_new(1, ESTF_INTERBOARD, "%s::%s",
1212 				s_mem->bp->cm.name, s_mem->cm.name));
1213 	}
1214 
1215 	/* fetch target mc asr register value */
1216 	err = drmach_read_mc_asr(t_mem, &t_masr);
1217 	if (err)
1218 		return (err);
1219 	else if (t_masr & STARFIRE_MC_INTERLEAVE_MASK) {
1220 		return (drerr_new(1, ESTF_INTERBOARD, "%s::%s",
1221 				t_mem->bp->cm.name, t_mem->cm.name));
1222 	}
1223 
1224 	/* get new source base pa from target's masr */
1225 	s_new_basepa = mc_asr_to_pa(t_masr);
1226 
1227 	/*
1228 	 * remove any existing slice offset to realign
1229 	 * memory with board's slice boundary
1230 	 */
1231 	s_new_basepa &= ~ (mc_get_mem_alignment() - 1);
1232 
1233 	/* get new target base pa from source's masr */
1234 	t_new_basepa  = mc_asr_to_pa(s_masr);
1235 
1236 	/* remove any existing slice offset, then apply new offset */
1237 	t_new_basepa &= ~ (mc_get_mem_alignment() - 1);
1238 	t_new_basepa += t_slice_offset;
1239 
1240 	/* encode new base pa into s_masr.  turn off mem present bit */
1241 	s_masr  = mc_pa_to_asr(s_masr, s_new_basepa);
1242 	s_masr &= ~STARFIRE_MC_MEM_PRESENT_MASK;
1243 
1244 	/* encode new base pa into t_masr.  turn on mem present bit */
1245 	t_masr  = mc_pa_to_asr(t_masr, t_new_basepa);
1246 	t_masr |= STARFIRE_MC_MEM_PRESENT_MASK;
1247 
1248 	/*
1249 	 * Step 0:	Mark source memory as not present.
1250 	 */
1251 	m = 0;
1252 	rsp = (drmach_rename_script_t *)isp;
1253 	err = drmach_get_mc_asr_addr(s_mem, &rsp[m].masr_addr);
1254 	if (err)
1255 		return (err);
1256 	rsp[m].masr = s_masr;
1257 	m++;
1258 
1259 	/*
1260 	 * Step 1:	Write source base address to target MC
1261 	 *		with present bit off.
1262 	 */
1263 	err = drmach_get_mc_asr_addr(t_mem, &rsp[m].masr_addr);
1264 	if (err)
1265 		return (err);
1266 	rsp[m].masr = t_masr & ~STARFIRE_MC_MEM_PRESENT_MASK;
1267 	m++;
1268 
1269 	/*
1270 	 * Step 2:	Now rewrite target reg with present bit on.
1271 	 */
1272 	rsp[m].masr_addr = rsp[m-1].masr_addr;
1273 	rsp[m].masr = t_masr;
1274 	m++;
1275 
1276 	s_bd = s_mem->bp->bnum;
1277 	t_bd = t_mem->bp->bnum;
1278 
1279 	DRMACH_PR("preparing script for CPU and IO units:\n");
1280 
1281 	rv = drmach_array_first(drmach_boards, &b_idx, &b_id);
1282 	if (rv) {
1283 		/* catch this in debug kernels */
1284 		ASSERT(0);
1285 		return (DRMACH_INTERNAL_ERROR());
1286 	}
1287 
1288 	do {
1289 		int			 d_idx;
1290 		drmachid_t		 d_id;
1291 		drmach_device_t		*device;
1292 
1293 		ASSERT(DRMACH_IS_BOARD_ID(b_id));
1294 		brd = b_id;
1295 		b = brd->bnum;
1296 
1297 		/*
1298 		 * Step 3:	Update PC MADR tables for CPUs.
1299 		 */
1300 		rv = drmach_array_first(brd->devices, &d_idx, &d_id);
1301 		if (rv) {
1302 			/* must mean no devices on this board */
1303 			break;
1304 		}
1305 
1306 		DRMACH_PR("\t%s\n", brd->cm.name);
1307 
1308 		do {
1309 			ASSERT(DRMACH_IS_DEVICE_ID(d_id));
1310 
1311 			if (!DRMACH_IS_CPU_ID(d_id))
1312 				continue;
1313 
1314 			device = d_id;
1315 			i = device->unum;
1316 
1317 			DRMACH_PR("\t\t%s\n", device->cm.name);
1318 
1319 			/*
1320 			 * Disabled detaching mem node.
1321 			 */
1322 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, s_bd, i);
1323 			rsp[m].masr = s_masr;
1324 			m++;
1325 			/*
1326 			 * Always write masr with present bit
1327 			 * off and then again with it on.
1328 			 */
1329 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, t_bd, i);
1330 			rsp[m].masr = t_masr & ~STARFIRE_MC_MEM_PRESENT_MASK;
1331 			m++;
1332 			rsp[m].masr_addr = rsp[m-1].masr_addr;
1333 			rsp[m].masr = t_masr;
1334 			m++;
1335 
1336 		} while (drmach_array_next(brd->devices, &d_idx, &d_id) == 0);
1337 
1338 		/*
1339 		 * Step 4:	Update PC MADR tables for IOs.
1340 		 */
1341 		rv = drmach_array_first(brd->devices, &d_idx, &d_id);
1342 		/* this worked for previous loop, must work here too */
1343 		ASSERT(rv == 0);
1344 
1345 		do {
1346 			ASSERT(DRMACH_IS_DEVICE_ID(d_id));
1347 
1348 			if (!DRMACH_IS_IO_ID(d_id))
1349 				continue;
1350 
1351 			device = d_id;
1352 			i = device->unum;
1353 
1354 			DRMACH_PR("\t\t%s\n", device->cm.name);
1355 
1356 			/*
1357 			 * Disabled detaching mem node.
1358 			 */
1359 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, s_bd, i+4);
1360 			rsp[m].masr = s_masr;
1361 			m++;
1362 			/*
1363 			 * Always write masr with present bit
1364 			 * off and then again with it on.
1365 			 */
1366 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, t_bd, i+4);
1367 			rsp[m].masr = t_masr & ~STARFIRE_MC_MEM_PRESENT_MASK;
1368 			m++;
1369 			rsp[m].masr_addr = rsp[m-1].masr_addr;
1370 			rsp[m].masr = t_masr;
1371 			m++;
1372 
1373 		} while (drmach_array_next(brd->devices, &d_idx, &d_id) == 0);
1374 	} while (drmach_array_next(drmach_boards, &b_idx, &b_id) == 0);
1375 
1376 	/*
1377 	 * Zero masr_addr value indicates the END.
1378 	 */
1379 	rsp[m].masr_addr = 0ull;
1380 	rsp[m].masr = 0;
1381 	DRMACH_PR("number of steps in rename script = %d\n", m);
1382 	m++;
1383 
1384 	/* paranoia */
1385 	ASSERT((caddr_t)&rsp[m] <= buf + buflen);
1386 
1387 #ifdef DEBUG
1388 	{
1389 		int	j;
1390 
1391 		DRMACH_PR("mc idle register address list:");
1392 		isp = (drmach_mc_idle_script_t *)buf;
1393 		DRMACH_PR("source mc idle addr 0x%lx, mem id %p",
1394 			isp[0].idle_addr, isp[0].mem);
1395 		DRMACH_PR("target mc idle addr 0x%lx, mem id %p",
1396 			isp[1].idle_addr, isp[1].mem);
1397 		ASSERT(isp[2].idle_addr == 0);
1398 
1399 		DRMACH_PR("copy-rename script:");
1400 		for (j = 0; j < m; j++) {
1401 			DRMACH_PR("0x%lx = 0x%08x",
1402 				rsp[j].masr_addr, rsp[j].masr);
1403 		}
1404 
1405 		DELAY(1000000);
1406 	}
1407 #endif
1408 
1409 	/* return number of bytes consumed */
1410 	b = (caddr_t)&rsp[m] - buf;
1411 	DRMACH_PR("total number of bytes consumed is %d\n", b);
1412 	ASSERT(b <= buflen);
1413 
1414 #ifdef lint
1415 	buflen = buflen;
1416 #endif
1417 
1418 	return (NULL);
1419 }
1420 
1421 /*
1422  * The routine performs the necessary memory COPY and MC adr SWITCH.
1423  * Both operations MUST be at the same "level" so that the stack is
1424  * maintained correctly between the copy and switch.  The switch
1425  * portion implements a caching mechanism to guarantee the code text
1426  * is cached prior to execution.  This is to guard against possible
1427  * memory access while the MC adr's are being modified.
1428  *
1429  * IMPORTANT: The _drmach_copy_rename_end() function must immediately
1430  * follow drmach_copy_rename_prog__relocatable() so that the correct
1431  * "length" of the drmach_copy_rename_prog__relocatable can be
1432  * calculated.  This routine MUST be a LEAF function, i.e. it can
1433  * make NO function calls, primarily for two reasons:
1434  *
1435  *	1. We must keep the stack consistent across the "switch".
1436  *	2. Function calls are compiled to relative offsets, and
1437  *	   we execute this function we'll be executing it from
1438  *	   a copied version in a different area of memory, thus
1439  *	   the relative offsets will be bogus.
1440  *
1441  * Moreover, it must have the "__relocatable" suffix to inform DTrace
1442  * providers (and anything else, for that matter) that this
1443  * function's text is manually relocated elsewhere before it is
1444  * executed.  That is, it cannot be safely instrumented with any
1445  * methodology that is PC-relative.
1446  */
1447 static void
1448 drmach_copy_rename_prog__relocatable(drmach_copy_rename_program_t *prog)
1449 {
1450 	extern void drmach_exec_script_il(drmach_rename_script_t *rsp);
1451 
1452 	drmach_mc_idle_script_t		*isp;
1453 	struct memlist			*ml;
1454 	int				csize;
1455 	int				lnsize;
1456 	uint64_t			caddr;
1457 
1458 	isp = (drmach_mc_idle_script_t *)prog->data;
1459 
1460 	caddr = ecache_flushaddr;
1461 	csize = (cpunodes[CPU->cpu_id].ecache_size << 1);
1462 	lnsize = cpunodes[CPU->cpu_id].ecache_linesize;
1463 
1464 	/*
1465 	 * DO COPY.
1466 	 */
1467 	for (ml = prog->c_ml; ml; ml = ml->next) {
1468 		uint64_t	s_pa, t_pa;
1469 		uint64_t	nbytes;
1470 
1471 		s_pa = prog->s_copybasepa + ml->address;
1472 		t_pa = prog->t_copybasepa + ml->address;
1473 		nbytes = ml->size;
1474 
1475 		while (nbytes != 0ull) {
1476 			/*
1477 			 * This copy does NOT use an ASI
1478 			 * that avoids the Ecache, therefore
1479 			 * the dst_pa addresses may remain
1480 			 * in our Ecache after the dst_pa
1481 			 * has been removed from the system.
1482 			 * A subsequent write-back to memory
1483 			 * will cause an ARB-stop because the
1484 			 * physical address no longer exists
1485 			 * in the system. Therefore we must
1486 			 * flush out local Ecache after we
1487 			 * finish the copy.
1488 			 */
1489 
1490 			/* copy 32 bytes at src_pa to dst_pa */
1491 			bcopy32_il(s_pa, t_pa);
1492 
1493 			/* increment by 32 bytes */
1494 			s_pa += (4 * sizeof (uint64_t));
1495 			t_pa += (4 * sizeof (uint64_t));
1496 
1497 			/* decrement by 32 bytes */
1498 			nbytes -= (4 * sizeof (uint64_t));
1499 		}
1500 	}
1501 
1502 	/*
1503 	 * Since bcopy32_il() does NOT use an ASI to bypass
1504 	 * the Ecache, we need to flush our Ecache after
1505 	 * the copy is complete.
1506 	 */
1507 	flush_ecache_il(caddr, csize, lnsize);		/* inline version */
1508 
1509 	/*
1510 	 * Wait for MCs to go idle.
1511 	 */
1512 	do {
1513 		register int	t = 10;
1514 		register uint_t	v;
1515 
1516 		/* loop t cycles waiting for each mc to indicate it's idle */
1517 		do {
1518 			v = ldphysio_il(isp->idle_addr)
1519 				& STARFIRE_MC_IDLE_MASK;
1520 
1521 		} while (v != STARFIRE_MC_IDLE_MASK && t-- > 0);
1522 
1523 		/* bailout if timedout */
1524 		if (t <= 0) {
1525 			prog->restless_mc = isp->mem;
1526 			return;
1527 		}
1528 
1529 		isp += 1;
1530 
1531 		/* stop if terminating zero has been reached */
1532 	} while (isp->idle_addr != 0);
1533 
1534 	/* advance passed terminating zero */
1535 	isp += 1;
1536 
1537 	/*
1538 	 * The following inline assembly routine caches
1539 	 * the rename script and then caches the code that
1540 	 * will do the rename.  This is necessary
1541 	 * so that we don't have any memory references during
1542 	 * the reprogramming.  We accomplish this by first
1543 	 * jumping through the code to guarantee it's cached
1544 	 * before we actually execute it.
1545 	 */
1546 	drmach_exec_script_il((drmach_rename_script_t *)isp);
1547 }
1548 
1549 static void
1550 drmach_copy_rename_end(void)
1551 {
1552 	/*
1553 	 * IMPORTANT:	This function's location MUST be located immediately
1554 	 *		following drmach_copy_rename_prog__relocatable to
1555 	 *		accurately estimate its size.  Note that this assumes
1556 	 *		the compiler keeps these functions in the order in
1557 	 *		which they appear :-o
1558 	 */
1559 }
1560 
1561 sbd_error_t *
1562 drmach_copy_rename_init(drmachid_t t_id, uint64_t t_slice_offset,
1563 	drmachid_t s_id, struct memlist *c_ml, drmachid_t *pgm_id)
1564 {
1565 	drmach_device_t	*s_mem;
1566 	drmach_device_t	*t_mem;
1567 	struct memlist	*x_ml;
1568 	uint64_t	off_mask, s_copybasepa, t_copybasepa, t_basepa;
1569 	int		len;
1570 	caddr_t		bp, wp;
1571 	pda_handle_t	ph;
1572 	sbd_error_t	*err;
1573 	drmach_copy_rename_program_t *prog;
1574 
1575 	if (!DRMACH_IS_MEM_ID(s_id))
1576 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1577 	if (!DRMACH_IS_MEM_ID(t_id))
1578 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1579 	s_mem = s_id;
1580 	t_mem = t_id;
1581 
1582 	/* get starting physical address of target memory */
1583 	err = drmach_mem_get_base_physaddr(t_id, &t_basepa);
1584 	if (err)
1585 		return (err);
1586 
1587 	/* calculate slice offset mask from slice size */
1588 	off_mask = mc_get_mem_alignment() - 1;
1589 
1590 	/* calculate source and target base pa */
1591 	s_copybasepa = c_ml->address;
1592 	t_copybasepa = t_basepa + ((c_ml->address & off_mask) - t_slice_offset);
1593 
1594 	/* paranoia */
1595 	ASSERT((c_ml->address & off_mask) >= t_slice_offset);
1596 
1597 	/* adjust copy memlist addresses to be relative to copy base pa */
1598 	x_ml = c_ml;
1599 	while (x_ml != NULL) {
1600 		x_ml->address -= s_copybasepa;
1601 		x_ml = x_ml->next;
1602 	}
1603 
1604 #ifdef DEBUG
1605 	{
1606 	uint64_t s_basepa, s_size, t_size;
1607 
1608 	x_ml = c_ml;
1609 	while (x_ml->next != NULL)
1610 		x_ml = x_ml->next;
1611 
1612 	DRMACH_PR("source copy span: base pa 0x%lx, end pa 0x%lx\n",
1613 		s_copybasepa,
1614 		s_copybasepa + x_ml->address + x_ml->size);
1615 
1616 	DRMACH_PR("target copy span: base pa 0x%lx, end pa 0x%lx\n",
1617 		t_copybasepa,
1618 		t_copybasepa + x_ml->address + x_ml->size);
1619 
1620 	DRMACH_PR("copy memlist (relative to copy base pa):\n");
1621 	MEMLIST_DUMP(c_ml);
1622 
1623 	err = drmach_mem_get_base_physaddr(s_id, &s_basepa);
1624 	ASSERT(err == NULL);
1625 
1626 	err = drmach_mem_get_size(s_id, &s_size);
1627 	ASSERT(err == NULL);
1628 
1629 	err = drmach_mem_get_size(t_id, &t_size);
1630 	ASSERT(err == NULL);
1631 
1632 	DRMACH_PR("current source base pa 0x%lx, size 0x%lx\n",
1633 		s_basepa, s_size);
1634 	DRMACH_PR("current target base pa 0x%lx, size 0x%lx\n",
1635 		t_basepa, t_size);
1636 
1637 	ASSERT(s_copybasepa + x_ml->address + x_ml->size <= s_basepa + s_size);
1638 	ASSERT(t_copybasepa + x_ml->address + x_ml->size <= t_basepa + t_size);
1639 	}
1640 #endif
1641 
1642 	ph = drmach_pda_open();
1643 	if (ph == NULL)
1644 		return (DRMACH_INTERNAL_ERROR());
1645 
1646 	/*
1647 	 * bp will be page aligned, since we're calling
1648 	 * kmem_zalloc() with an exact multiple of PAGESIZE.
1649 	 */
1650 	wp = bp = kmem_zalloc(PAGESIZE, KM_SLEEP);
1651 
1652 	/* allocate space for copy rename struct */
1653 	len = sizeof (drmach_copy_rename_program_t);
1654 	DRMACH_PR("prog = 0x%p, header len %d\n", wp, len);
1655 	prog = (drmach_copy_rename_program_t *)wp;
1656 	wp += (len + ecache_alignsize - 1) & ~ (ecache_alignsize - 1);
1657 
1658 	/*
1659 	 * Copy the code for the copy-rename routine into
1660 	 * a page aligned piece of memory.  We do this to guarantee
1661 	 * that we're executing within the same page and thus reduce
1662 	 * the possibility of cache collisions between different
1663 	 * pages.
1664 	 */
1665 	len = (int)((ulong_t)drmach_copy_rename_end -
1666 		    (ulong_t)drmach_copy_rename_prog__relocatable);
1667 	ASSERT(wp + len < bp + PAGESIZE);
1668 	bcopy((caddr_t)drmach_copy_rename_prog__relocatable, wp, len);
1669 
1670 	DRMACH_PR("copy-rename function 0x%p, len %d\n", wp, len);
1671 	prog->run = (void (*)())wp;
1672 	wp += (len + ecache_alignsize - 1) & ~ (ecache_alignsize - 1);
1673 
1674 	/*
1675 	 * Prepare data page that will contain script of
1676 	 * operations to perform during copy-rename.
1677 	 * Allocate temporary buffer to hold script.
1678 	 */
1679 	err = drmach_prep_rename_script(s_mem, t_mem, t_slice_offset,
1680 		wp, PAGESIZE - (wp - bp));
1681 	if (err) {
1682 		(void) drmach_copy_rename_fini(prog);
1683 		return (err);
1684 	}
1685 
1686 	DRMACH_PR("copy-rename script 0x%p, len %d\n", wp, len);
1687 	prog->data = wp;
1688 	wp += (len + ecache_alignsize - 1) & ~ (ecache_alignsize - 1);
1689 
1690 	prog->ph = ph;
1691 	prog->s_copybasepa = s_copybasepa;
1692 	prog->t_copybasepa = t_copybasepa;
1693 	prog->c_ml = c_ml;
1694 	*pgm_id = prog;
1695 
1696 	return (NULL);
1697 }
1698 
1699 sbd_error_t *
1700 drmach_copy_rename_fini(drmachid_t id)
1701 {
1702 	drmach_copy_rename_program_t	*prog = id;
1703 	sbd_error_t			*err = NULL;
1704 
1705 	if (prog->c_ml != NULL)
1706 		memlist_delete(prog->c_ml);
1707 
1708 	if (prog->ph != NULL)
1709 		pda_close(prog->ph);
1710 
1711 	if (prog->restless_mc != 0) {
1712 		cmn_err(CE_WARN, "MC did not idle; OBP Node 0x%x",
1713 			(uint_t)drmach_node_get_dnode(prog->restless_mc->node));
1714 
1715 		err = DRMACH_INTERNAL_ERROR();
1716 	}
1717 
1718 	kmem_free(prog, PAGESIZE);
1719 
1720 	return (err);
1721 }
1722 
1723 static sbd_error_t *
1724 drmach_io_new(drmach_device_t *dp)
1725 {
1726 	static sbd_error_t *drmach_io_release(drmachid_t);
1727 	static sbd_error_t *drmach_io_status(drmachid_t, drmach_status_t *);
1728 
1729 	sbd_error_t	*err;
1730 	int		 portid;
1731 
1732 	err = drmach_device_get_prop(dp, "upa-portid", &portid);
1733 	if (err == NULL) {
1734 		ASSERT(portid & 0x40);
1735 		dp->unum = portid & 1;
1736 	}
1737 
1738 	dp->cm.isa = (void *)drmach_io_new;
1739 	dp->cm.release = drmach_io_release;
1740 	dp->cm.status = drmach_io_status;
1741 
1742 	snprintf(dp->cm.name, sizeof (dp->cm.name), "%s%d", dp->type, dp->unum);
1743 
1744 	return (err);
1745 }
1746 
1747 static void
1748 drmach_iopc_op(pda_handle_t ph, drmach_iopc_op_t op)
1749 {
1750 	register int b;
1751 
1752 	for (b = 0; b < MAX_BOARDS; b++) {
1753 		int		p;
1754 		ushort_t	bda_ioc;
1755 		board_desc_t	*bdesc;
1756 
1757 		if (pda_board_present(ph, b) == 0)
1758 			continue;
1759 
1760 		bdesc = (board_desc_t *)pda_get_board_info(ph, b);
1761 		/*
1762 		 * Update PCs for IOCs.
1763 		 */
1764 		bda_ioc = bdesc->bda_ioc;
1765 		for (p = 0; p < MAX_IOCS; p++) {
1766 			u_longlong_t	idle_addr;
1767 			uchar_t		value;
1768 
1769 			if (BDA_NBL(bda_ioc, p) != BDAN_GOOD)
1770 				continue;
1771 
1772 			idle_addr = STARFIRE_BB_PC_ADDR(b, p, 1);
1773 
1774 			switch (op) {
1775 			case DO_PAUSE:
1776 				value = STARFIRE_BB_PC_PAUSE(p);
1777 				break;
1778 
1779 			case DO_IDLE:
1780 				value = STARFIRE_BB_PC_IDLE(p);
1781 				break;
1782 
1783 			case DO_UNPAUSE:
1784 				value = ldbphysio(idle_addr);
1785 				value &= ~STARFIRE_BB_PC_PAUSE(p);
1786 				break;
1787 
1788 			case DO_UNIDLE:
1789 				value = ldbphysio(idle_addr);
1790 				value &= ~STARFIRE_BB_PC_IDLE(p);
1791 				break;
1792 
1793 			default:
1794 				cmn_err(CE_PANIC,
1795 					"drmach_iopc_op: unknown op (%d)",
1796 					(int)op);
1797 				/*NOTREACHED*/
1798 			}
1799 			stbphysio(idle_addr, value);
1800 		}
1801 	}
1802 }
1803 
1804 void
1805 drmach_copy_rename(drmachid_t id)
1806 {
1807 	drmach_copy_rename_program_t	*prog = id;
1808 	uint64_t			neer;
1809 
1810 	/*
1811 	 * UPA IDLE
1812 	 * Protocol = PAUSE -> IDLE -> UNPAUSE
1813 	 * In reality since we only "idle" the IOPCs it's sufficient
1814 	 * to just issue the IDLE operation since (in theory) all IOPCs
1815 	 * in the field are PC6.  However, we'll be robust and do the
1816 	 * proper workaround protocol so that we never have to worry!
1817 	 */
1818 	drmach_iopc_op(prog->ph, DO_PAUSE);
1819 	drmach_iopc_op(prog->ph, DO_IDLE);
1820 	DELAY(100);
1821 	drmach_iopc_op(prog->ph, DO_UNPAUSE);
1822 	DELAY(100);
1823 
1824 	/* disable CE reporting */
1825 	neer = get_error_enable();
1826 	set_error_enable(neer & ~EER_CEEN);
1827 
1828 	/* run the copy/rename program */
1829 	prog->run(prog);
1830 
1831 	/* enable CE reporting */
1832 	set_error_enable(neer);
1833 
1834 	/*
1835 	 * UPA UNIDLE
1836 	 * Protocol = UNIDLE
1837 	 */
1838 	drmach_iopc_op(prog->ph, DO_UNIDLE);
1839 	DELAY(100);
1840 }
1841 
1842 /*
1843  * The counter-timer and perf-counter nodes are not being cleaned
1844  * up after a board that was present at start of day is detached.
1845  * If the board has become unconfigured with this operation, walk
1846  * the prom tree and find all counter-timer and perf-counter nodes
1847  * that have the same board number as the board that was just
1848  * unconfigured and remove them.
1849  */
1850 static sbd_error_t *
1851 drmach_remove_counter_nodes(drmachid_t id)
1852 {
1853 	int		num;
1854 	char		name[OBP_MAXDRVNAME];
1855 	pnode_t		child;
1856 	dev_info_t	*dip;
1857 	sbd_error_t	*err;
1858 	drmach_status_t	stat;
1859 	drmach_board_t	*bp;
1860 
1861 	if (!DRMACH_IS_BOARD_ID(id)) {
1862 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1863 	}
1864 
1865 	if ((err = drmach_board_status(id, &stat)) != NULL) {
1866 		return (err);
1867 	}
1868 
1869 	/*
1870 	 * Only clean up the counter-timer and perf-counter
1871 	 * nodes when the entire board is unconfigured.
1872 	 */
1873 	if (stat.configured) {
1874 		return (NULL);
1875 	}
1876 
1877 	bp = (drmach_board_t *)id;
1878 
1879 	err = NULL;
1880 
1881 	for (child = prom_childnode(prom_rootnode()); child != OBP_NONODE;
1882 	    child = prom_nextnode(child)) {
1883 
1884 		if (prom_getprop(child, OBP_BOARDNUM, (caddr_t)&num) == -1) {
1885 			continue;
1886 		}
1887 
1888 		if (bp->bnum != num) {
1889 			continue;
1890 		}
1891 
1892 		if (prom_getprop(child, OBP_NAME, (caddr_t)name) == -1) {
1893 			continue;
1894 		}
1895 
1896 		if (strncmp(name, MISC_COUNTER_TIMER_DEVNAME, OBP_MAXDRVNAME) &&
1897 		    strncmp(name, MISC_PERF_COUNTER_DEVNAME, OBP_MAXDRVNAME)) {
1898 				continue;
1899 		}
1900 
1901 		/* Root node doesn't have to be held */
1902 		dip = e_ddi_nodeid_to_dip(child);
1903 
1904 		/*
1905 		 * If the node is only in the OBP tree, then
1906 		 * we don't have to remove it.
1907 		 */
1908 		if (dip) {
1909 			dev_info_t *fdip = NULL;
1910 
1911 			DRMACH_PR("removing %s devinfo node\n", name);
1912 
1913 			e_ddi_branch_hold(dip);
1914 			ddi_release_devi(dip); /* held in e_ddi_nodeid_to_dip */
1915 
1916 			if (e_ddi_branch_destroy(dip, &fdip, 0)) {
1917 				char *path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1918 
1919 				/*
1920 				 * If non-NULL, fdip is held and must be
1921 				 * released.
1922 				 */
1923 				if (fdip != NULL) {
1924 					(void) ddi_pathname(fdip, path);
1925 					ddi_release_devi(fdip);
1926 				} else {
1927 					(void) ddi_pathname(dip, path);
1928 				}
1929 
1930 				err = drerr_new(1, ESTF_DRVFAIL, path);
1931 				kmem_free(path, MAXPATHLEN);
1932 				e_ddi_branch_rele(dip);
1933 				break;
1934 			}
1935 		}
1936 	}
1937 
1938 	return (err);
1939 }
1940 
1941 /*ARGSUSED*/
1942 sbd_error_t *
1943 drmach_pre_op(int cmd, drmachid_t id, drmach_opts_t *opts)
1944 {
1945 	/* allow status and ncm operations to always succeed */
1946 	if ((cmd == SBD_CMD_STATUS) || (cmd == SBD_CMD_GETNCM)) {
1947 		return (NULL);
1948 	}
1949 
1950 	/* check all other commands for the required option string */
1951 	if ((opts->size > 0) && (opts->copts != NULL)) {
1952 
1953 		DRMACH_PR("platform options: %s\n", opts->copts);
1954 
1955 		if (strstr(opts->copts, "xfdr") != NULL) {
1956 			return (NULL);
1957 		}
1958 	}
1959 
1960 	return (drerr_new(0, ESTF_SUPPORT, NULL));
1961 }
1962 
1963 /*ARGSUSED*/
1964 sbd_error_t *
1965 drmach_post_op(int cmd, drmachid_t id, drmach_opts_t *opts)
1966 {
1967 	sbd_error_t	*err = NULL;
1968 
1969 	switch (cmd) {
1970 	case SBD_CMD_UNCONFIGURE:
1971 
1972 		err = drmach_remove_counter_nodes(id);
1973 		break;
1974 
1975 	case SBD_CMD_CONFIGURE:
1976 	case SBD_CMD_DISCONNECT:
1977 	case SBD_CMD_CONNECT:
1978 	case SBD_CMD_GETNCM:
1979 	case SBD_CMD_STATUS:
1980 		break;
1981 
1982 	default:
1983 		break;
1984 	}
1985 
1986 	return (err);
1987 }
1988 
1989 sbd_error_t *
1990 drmach_board_assign(int bnum, drmachid_t *id)
1991 {
1992 	sbd_error_t	*err;
1993 
1994 	if (!drmach_initialized && drmach_init() == -1) {
1995 		err = DRMACH_INTERNAL_ERROR();
1996 	} else if (drmach_array_get(drmach_boards, bnum, id) == -1) {
1997 		err = drerr_new(1, ESTF_BNUM, "%d", bnum);
1998 	} else if (*id != NULL) {
1999 		err = NULL;
2000 	} else {
2001 		drmach_board_t	*bp;
2002 
2003 		*id  = (drmachid_t)drmach_board_new(bnum);
2004 		bp = *id;
2005 		bp->assigned = 1;
2006 		err = NULL;
2007 	}
2008 
2009 	return (err);
2010 }
2011 
2012 static int
2013 drmach_attach_board(void *arg)
2014 {
2015 	drmach_board_t	*obj = (drmach_board_t *)arg;
2016 	cpuset_t	cset;
2017 	int		retval;
2018 
2019 	/*
2020 	 * OBP disables traps during the board probe.
2021 	 * So, in order to prevent cross-call/cross-trap timeouts,
2022 	 * and thus panics, we effectively block anybody from
2023 	 * issuing xc's/xt's by doing a promsafe_xc_attention.
2024 	 * In the previous version of Starfire DR (2.6), a timeout
2025 	 * suspension mechanism was implemented in the send-mondo
2026 	 * assembly.  That mechanism is unnecessary with the
2027 	 * existence of xc_attention/xc_dismissed.
2028 	 */
2029 	cset = cpu_ready_set;
2030 	promsafe_xc_attention(cset);
2031 
2032 	retval = prom_starfire_add_brd(obj->connect_cpuid);
2033 
2034 	xc_dismissed(cset);
2035 
2036 	return (retval);
2037 }
2038 
2039 sbd_error_t *
2040 drmach_board_connect(drmachid_t id, drmach_opts_t *opts)
2041 {
2042 	drmach_board_t	*obj = (drmach_board_t *)id;
2043 	int		retval;
2044 	sbd_error_t	*err;
2045 	char		*cptr, *copts;
2046 
2047 	if (!DRMACH_IS_BOARD_ID(id))
2048 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2049 
2050 	if (opts->size > 0)
2051 		copts = opts->copts;
2052 
2053 	if ((cptr = strstr(copts, "cpuid=")) != NULL) {
2054 		int cpuid;
2055 
2056 		cptr += strlen("cpuid=");
2057 		cpuid = stoi(&cptr);
2058 
2059 		if (DRMACH_CPUID2BNUM(cpuid) == obj->bnum) {
2060 			obj->connect_cpuid = cpuid;
2061 			obj->assigned = 1;
2062 		} else
2063 			return (drerr_new(1, ESTF_SETCPUVAL, "%d", cpuid));
2064 	} else {
2065 		/* cpuid was not specified */
2066 		obj->connect_cpuid = -1;
2067 	}
2068 
2069 	if (obj->connect_cpuid == -1) {
2070 		err =  drerr_new(1, ESTF_NOCPUID, obj->cm.name);
2071 		return (err);
2072 	}
2073 
2074 	cmn_err(CE_CONT, "DRMACH: PROM attach %s CPU %d\n",
2075 		obj->cm.name, obj->connect_cpuid);
2076 
2077 	retval = prom_tree_update(drmach_attach_board, obj);
2078 
2079 	if (retval == 0)
2080 		err = NULL;
2081 	else {
2082 		cmn_err(CE_WARN, "prom error: prom_starfire_add_brd(%d) "
2083 			"returned %d", obj->connect_cpuid, retval);
2084 
2085 		err = drerr_new(1, ESTF_PROBE, obj->cm.name);
2086 	}
2087 
2088 	obj->connect_cpuid = -1;
2089 
2090 	return (err);
2091 }
2092 
2093 /*ARGSUSED*/
2094 sbd_error_t *
2095 drmach_board_disconnect(drmachid_t id, drmach_opts_t *opts)
2096 {
2097 	drmach_board_t		*bp;
2098 	int			rv;
2099 	int			d_idx;	/* device index */
2100 	drmachid_t		d_id;	/* device ID */
2101 	sbd_error_t		*err;
2102 
2103 	if (!DRMACH_IS_BOARD_ID(id))
2104 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2105 
2106 	bp = id;
2107 
2108 	/*
2109 	 * We need to make sure all of the board's device nodes
2110 	 * have been removed from the Solaris device tree before
2111 	 * continuing with the disconnect. Otherwise, we could
2112 	 * disconnect the board and remove the OBP device tree
2113 	 * nodes with Solaris device tree nodes remaining.
2114 	 *
2115 	 * On Starfire, Solaris device tree nodes are deleted
2116 	 * during unconfigure by drmach_unconfigure(). It's
2117 	 * necessary to do this here because drmach_unconfigure()
2118 	 * failures are not handled during unconfigure.
2119 	 */
2120 	if (bp->devices) {
2121 		rv = drmach_array_first(bp->devices, &d_idx, &d_id);
2122 		while (rv == 0) {
2123 			err = drmach_unconfigure(d_id, DRMACH_DEVI_REMOVE);
2124 			if (err)
2125 				return (err);
2126 
2127 			rv = drmach_array_next(bp->devices, &d_idx, &d_id);
2128 		}
2129 	}
2130 
2131 	/*
2132 	 * Starfire board Solaris device tree counter nodes,
2133 	 * which are only present on start-of-day boards, are
2134 	 * removed in the dr_post_op() code flow after the
2135 	 * board is unconfigured. We call the counter node
2136 	 * removal function here because unconfigure errors
2137 	 * can cause the dr_post_op() function to be skipped
2138 	 * after an unconfigure operation even though all of
2139 	 * the board's devices have been transitioned to the
2140 	 * unconfigured state.
2141 	 */
2142 	err = drmach_remove_counter_nodes(id);
2143 	if (err)
2144 		return (err);
2145 
2146 	return (NULL);
2147 }
2148 
2149 static int
2150 drmach_board_find_devices_cb(drmach_node_walk_args_t *args)
2151 {
2152 	drmach_node_t			*node = args->node;
2153 	drmach_board_cb_data_t		*data = args->data;
2154 	drmach_board_t			*obj = data->obj;
2155 
2156 	int		 rv;
2157 	int		 bnum;
2158 	drmach_device_t	*device;
2159 
2160 	rv = drmach_node_get_prop(node, OBP_BOARDNUM, &bnum);
2161 	if (rv) {
2162 		/*
2163 		 * if the node does not have a board# property, then
2164 		 * by that information alone it is known that drmach
2165 		 * is not interested in it.
2166 		 */
2167 		return (0);
2168 	} else if (bnum != obj->bnum)
2169 		return (0);
2170 
2171 	/*
2172 	 * Create a device data structure from this node data.
2173 	 * The call may yield nothing if the node is not of interest
2174 	 * to drmach.
2175 	 */
2176 	data->err = drmach_device_new(node, obj, &device);
2177 	if (data->err)
2178 		return (-1);
2179 	else if (device == NULL) {
2180 		/*
2181 		 * drmach_device_new examined the node we passed in
2182 		 * and determined that it was one not of interest to
2183 		 * drmach.  So, it is skipped.
2184 		 */
2185 		return (0);
2186 	}
2187 
2188 	rv = drmach_array_set(obj->devices, data->ndevs++, device);
2189 	if (rv) {
2190 		drmach_device_dispose(device);
2191 		data->err = DRMACH_INTERNAL_ERROR();
2192 		return (-1);
2193 	}
2194 
2195 	data->err = (*data->found)(data->a, device->type, device->unum, device);
2196 	return (data->err == NULL ? 0 : -1);
2197 }
2198 
2199 sbd_error_t *
2200 drmach_board_find_devices(drmachid_t id, void *a,
2201 	sbd_error_t *(*found)(void *a, const char *, int, drmachid_t))
2202 {
2203 	extern int		 plat_max_cpu_units_per_board();
2204 	extern int		 plat_max_mem_units_per_board();
2205 	extern int		 plat_max_io_units_per_board();
2206 
2207 	drmach_board_t		*obj = (drmach_board_t *)id;
2208 	sbd_error_t		*err;
2209 	int			 max_devices;
2210 	int			 rv;
2211 	drmach_board_cb_data_t	data;
2212 
2213 	max_devices  = plat_max_cpu_units_per_board();
2214 	max_devices += plat_max_mem_units_per_board();
2215 	max_devices += plat_max_io_units_per_board();
2216 
2217 	obj->devices = drmach_array_new(0, max_devices);
2218 
2219 	data.obj = obj;
2220 	data.ndevs = 0;
2221 	data.found = found;
2222 	data.a = a;
2223 	data.err = NULL;
2224 
2225 	rv = drmach_node_walk(obj->tree, &data, drmach_board_find_devices_cb);
2226 	if (rv == 0)
2227 		err = NULL;
2228 	else {
2229 		drmach_array_dispose(obj->devices, drmach_device_dispose);
2230 		obj->devices = NULL;
2231 
2232 		if (data.err)
2233 			err = data.err;
2234 		else
2235 			err = DRMACH_INTERNAL_ERROR();
2236 	}
2237 
2238 	return (err);
2239 }
2240 
2241 int
2242 drmach_board_lookup(int bnum, drmachid_t *id)
2243 {
2244 	int	rv = 0;
2245 
2246 	if (!drmach_initialized && drmach_init() == -1) {
2247 		*id = 0;
2248 		rv = -1;
2249 	} else if (drmach_array_get(drmach_boards, bnum, id)) {
2250 		*id = 0;
2251 		rv = -1;
2252 	}
2253 	return (rv);
2254 }
2255 
2256 sbd_error_t *
2257 drmach_board_name(int bnum, char *buf, int buflen)
2258 {
2259 	snprintf(buf, buflen, "SB%d", bnum);
2260 	return (NULL);
2261 }
2262 
2263 sbd_error_t *
2264 drmach_board_poweroff(drmachid_t id)
2265 {
2266 	drmach_board_t	*bp;
2267 	sbd_error_t	*err;
2268 	drmach_status_t	 stat;
2269 
2270 	if (!DRMACH_IS_BOARD_ID(id))
2271 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2272 	bp = id;
2273 
2274 	err = drmach_board_status(id, &stat);
2275 	if (err)
2276 		return (err);
2277 	else if (stat.configured || stat.busy)
2278 		return (drerr_new(0, ESTF_CONFIGBUSY, bp->cm.name));
2279 	else {
2280 		/* board power off is essentially a noop for Starfire */
2281 		bp->powered = 0;
2282 		return (NULL);
2283 	}
2284 	/*NOTREACHED*/
2285 }
2286 
2287 sbd_error_t *
2288 drmach_board_poweron(drmachid_t id)
2289 {
2290 	drmach_board_t	*bp;
2291 
2292 	if (!DRMACH_IS_BOARD_ID(id))
2293 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2294 	bp = id;
2295 
2296 	/* board power on is essentially a noop for Starfire */
2297 	bp->powered = 1;
2298 
2299 	return (NULL);
2300 }
2301 
2302 static sbd_error_t *
2303 drmach_board_release(drmachid_t id)
2304 {
2305 	if (!DRMACH_IS_BOARD_ID(id))
2306 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2307 	return (NULL);
2308 }
2309 
2310 /*ARGSUSED*/
2311 sbd_error_t *
2312 drmach_board_test(drmachid_t id, drmach_opts_t *opts, int force)
2313 {
2314 	return (NULL);
2315 }
2316 
2317 sbd_error_t *
2318 drmach_board_unassign(drmachid_t id)
2319 {
2320 	drmach_board_t	*bp;
2321 	sbd_error_t	*err;
2322 	drmach_status_t	 stat;
2323 
2324 	if (!DRMACH_IS_BOARD_ID(id))
2325 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2326 	bp = id;
2327 
2328 	err = drmach_board_status(id, &stat);
2329 	if (err)
2330 		return (err);
2331 	else if (stat.configured || stat.busy)
2332 		return (drerr_new(0, ESTF_CONFIGBUSY, bp->cm.name));
2333 	else if (drmach_array_set(drmach_boards, bp->bnum, 0) != 0)
2334 		return (DRMACH_INTERNAL_ERROR());
2335 	else {
2336 		drmach_board_dispose(bp);
2337 		return (NULL);
2338 	}
2339 	/*NOTREACHED*/
2340 }
2341 
2342 static sbd_error_t *
2343 drmach_cpu_new(drmach_device_t *dp)
2344 {
2345 	static sbd_error_t *drmach_cpu_release(drmachid_t);
2346 	static sbd_error_t *drmach_cpu_status(drmachid_t, drmach_status_t *);
2347 
2348 	sbd_error_t	*err;
2349 	int		 portid;
2350 
2351 	err = drmach_device_get_prop(dp, "upa-portid", &portid);
2352 	if (err == NULL)
2353 		dp->unum = portid & 3;
2354 
2355 	dp->cm.isa = (void *)drmach_cpu_new;
2356 	dp->cm.release = drmach_cpu_release;
2357 	dp->cm.status = drmach_cpu_status;
2358 
2359 	snprintf(dp->cm.name, sizeof (dp->cm.name), "%s%d", dp->type, dp->unum);
2360 
2361 	return (err);
2362 }
2363 
2364 /*
2365  * drmach_cpu_obp_detach()
2366  *  This requires two steps, first, we must put the cpuid into the OBP
2367  *  idle loop (Idle in Program) state.  Then we call OBP to place the CPU
2368  *  into the "Detached" state, which does any special processing to
2369  *  actually detach the cpu, such as flushing ecache, and also ensures
2370  *  that a subsequent breakpoint won't restart the cpu (if it was just in
2371  *  Idle in Program state).
2372  */
2373 static void
2374 drmach_cpu_obp_detach(int cpuid)
2375 {
2376 	/*
2377 	 * Cpu may not be under OBP's control. Eg, if cpu exited to download
2378 	 * helper on a prior attach.
2379 	 */
2380 	if (CPU_SGN_EXISTS(cpuid) &&
2381 			!SGN_CPU_IS_OS(cpuid) &&
2382 			!SGN_CPU_IS_OBP(cpuid)) {
2383 		cmn_err(CE_WARN,
2384 			"unexpected signature (0x%x) for cpu %d",
2385 			get_cpu_sgn(cpuid), cpuid);
2386 	}
2387 
2388 	/*
2389 	 * Now we place the CPU into the "Detached" idle loop in OBP.
2390 	 * This is so that the CPU won't be restarted if we break into
2391 	 * OBP with a breakpoint or BREAK key from the console, and also
2392 	 * if we need to do any special processing, such as flushing the
2393 	 * cpu's ecache, disabling interrupts (by turning of the ET bit in
2394 	 * the PSR) and/or spinning in BBSRAM rather than global memory.
2395 	 */
2396 	DRMACH_PR("prom_starfire_rm_cpu(%d)\n", cpuid);
2397 	prom_starfire_rm_cpu(cpuid);
2398 }
2399 
2400 /*
2401  * drmach_cpu_obp_is_detached() returns TRUE if the cpu sigblock signature state
2402  * is SIGBST_DETACHED; otherwise it returns FALSE. This routine should only
2403  * be called after we have asked OBP to detach the CPU. It should NOT be
2404  * called as a check during any other flow.
2405  */
2406 static int
2407 drmach_cpu_obp_is_detached(int cpuid)
2408 {
2409 	if (!CPU_SGN_EXISTS(cpuid) ||
2410 		(SGN_CPU_IS_OS(cpuid) && SGN_CPU_STATE_IS_DETACHED(cpuid)))
2411 		return (1);
2412 	else
2413 		return (0);
2414 }
2415 
2416 static int
2417 drmach_cpu_start(struct cpu *cp)
2418 {
2419 	int		cpuid = cp->cpu_id;
2420 	int		ntries = drmach_cpu_ntries;
2421 	extern void	restart_other_cpu(int);
2422 
2423 	ASSERT(MUTEX_HELD(&cpu_lock));
2424 	ASSERT(cpunodes[cpuid].nodeid != (pnode_t)0);
2425 
2426 	cp->cpu_flags &= ~CPU_POWEROFF;
2427 
2428 	/*
2429 	 * NOTE: restart_other_cpu pauses cpus during the
2430 	 *	 slave cpu start.  This helps to quiesce the
2431 	 *	 bus traffic a bit which makes the tick sync
2432 	 *	 routine in the prom more robust.
2433 	 */
2434 	DRMACH_PR("COLD START for cpu (%d)\n", cpuid);
2435 
2436 	prom_starfire_add_cpu(cpuid);
2437 
2438 	restart_other_cpu(cpuid);
2439 
2440 	/*
2441 	 * Wait for the cpu to reach its idle thread before
2442 	 * we zap him with a request to blow away the mappings
2443 	 * he (might) have for the drmach_shutdown_asm code
2444 	 * he may have executed on unconfigure.
2445 	 */
2446 	while ((cp->cpu_thread != cp->cpu_idle_thread) && (ntries > 0)) {
2447 		DELAY(drmach_cpu_delay);
2448 		ntries--;
2449 	}
2450 
2451 	DRMACH_PR("waited %d out of %d loops for cpu %d\n",
2452 		drmach_cpu_ntries - ntries, drmach_cpu_ntries, cpuid);
2453 
2454 	xt_one(cpuid, vtag_flushpage_tl1,
2455 		(uint64_t)drmach_shutdown_va, (uint64_t)KCONTEXT);
2456 
2457 	return (0);
2458 }
2459 
2460 /*
2461  * A detaching CPU is xcalled with an xtrap to drmach_cpu_stop_self() after
2462  * it has been offlined. The function of this routine is to get the cpu
2463  * spinning in a safe place. The requirement is that the system will not
2464  * reference anything on the detaching board (memory and i/o is detached
2465  * elsewhere) and that the CPU not reference anything on any other board
2466  * in the system.  This isolation is required during and after the writes
2467  * to the domain masks to remove the board from the domain.
2468  *
2469  * To accomplish this isolation the following is done:
2470  *	1) Create a locked mapping to a location in BBSRAM where
2471  *	   the cpu will execute.
2472  *	2) Copy the target function (drmach_shutdown_asm) in which
2473  *	   the cpu will execute into BBSRAM.
2474  *	3) Jump into function with BBSRAM.
2475  *	   Function will:
2476  *	   3.1) Flush its Ecache (displacement).
2477  *	   3.2) Flush its Dcache with HW mechanism.
2478  *	   3.3) Flush its Icache with HW mechanism.
2479  *	   3.4) Flush all valid and _unlocked_ D-TLB entries.
2480  *	   3.5) Flush all valid and _unlocked_ I-TLB entries.
2481  *	   3.6) Clear xt_mb to signal completion. Note: cache line is
2482  *		recovered by drmach_cpu_poweroff().
2483  *	4) Jump into a tight loop.
2484  */
2485 #define	DRMACH_BBSRAM_OFFSET	0x1000
2486 
2487 static void
2488 drmach_cpu_stop_self(void)
2489 {
2490 	int		cpuid = (int)CPU->cpu_id;
2491 	tte_t		tte;
2492 	volatile uint_t	*src, *dst;
2493 	uint_t		funclen;
2494 	uint64_t	bbsram_pa, bbsram_offset;
2495 	uint_t		bbsram_pfn;
2496 	uint64_t	bbsram_addr;
2497 	void		(*bbsram_func)(uint64_t);
2498 	extern void	drmach_shutdown_asm(uint64_t);
2499 	extern void	drmach_shutdown_asm_end(void);
2500 
2501 	funclen = (uint_t)drmach_shutdown_asm_end - (uint_t)drmach_shutdown_asm;
2502 	ASSERT(funclen <= MMU_PAGESIZE);
2503 	/*
2504 	 * We'll start from the 0th's base.
2505 	 */
2506 	bbsram_pa = STARFIRE_UPAID2UPS(cpuid) | STARFIRE_PSI_BASE;
2507 	bbsram_offset = bbsram_pa | 0xfe0ULL;
2508 	bbsram_pa += ldphysio(bbsram_offset) + DRMACH_BBSRAM_OFFSET;
2509 
2510 	bbsram_pfn = (uint_t)(bbsram_pa >> MMU_PAGESHIFT);
2511 
2512 	bbsram_addr = (uint64_t)drmach_shutdown_va;
2513 	drmach_shutdown_asm_mbox->estack = bbsram_addr + (uint64_t)funclen;
2514 
2515 	tte.tte_inthi = TTE_VALID_INT | TTE_SZ_INT(TTE8K) |
2516 			TTE_PFN_INTHI(bbsram_pfn);
2517 	tte.tte_intlo = TTE_PFN_INTLO(bbsram_pfn) |
2518 			TTE_HWWR_INT | TTE_PRIV_INT | TTE_LCK_INT;
2519 	sfmmu_dtlb_ld(drmach_shutdown_va, KCONTEXT, &tte);	/* load dtlb */
2520 	sfmmu_itlb_ld(drmach_shutdown_va, KCONTEXT, &tte);	/* load itlb */
2521 
2522 	for (src = (uint_t *)drmach_shutdown_asm, dst = (uint_t *)bbsram_addr;
2523 		src < (uint_t *)drmach_shutdown_asm_end; src++, dst++)
2524 		*dst = *src;
2525 
2526 	bbsram_func = (void (*)())bbsram_addr;
2527 	drmach_shutdown_asm_mbox->flushaddr = ecache_flushaddr;
2528 	drmach_shutdown_asm_mbox->size = (cpunodes[cpuid].ecache_size << 1);
2529 	drmach_shutdown_asm_mbox->linesize = cpunodes[cpuid].ecache_linesize;
2530 	drmach_shutdown_asm_mbox->physaddr
2531 				    = va_to_pa((void *)&drmach_xt_mb[cpuid]);
2532 
2533 	/*
2534 	 * Signal to drmach_cpu_poweroff() is via drmach_xt_mb cleared
2535 	 * by asm code
2536 	 */
2537 
2538 	(*bbsram_func)(va_to_pa((void *)drmach_shutdown_asm_mbox));
2539 }
2540 
2541 static void
2542 drmach_cpu_shutdown_self(void)
2543 {
2544 	cpu_t		*cp = CPU;
2545 	int		cpuid = cp->cpu_id;
2546 	extern void	flush_windows(void);
2547 
2548 	flush_windows();
2549 
2550 	(void) spl8();
2551 
2552 	ASSERT(cp->cpu_intr_actv == 0);
2553 	ASSERT(cp->cpu_thread == cp->cpu_idle_thread ||
2554 	    cp->cpu_thread == cp->cpu_startup_thread);
2555 
2556 	cp->cpu_flags = CPU_OFFLINE | CPU_QUIESCED | CPU_POWEROFF;
2557 
2558 	drmach_cpu_stop_self();
2559 
2560 	cmn_err(CE_PANIC, "CPU %d FAILED TO SHUTDOWN", cpuid);
2561 }
2562 
2563 /* a helper routine to keep the math in one place */
2564 static processorid_t
2565 drmach_cpu_calc_id(drmach_device_t *dp)
2566 {
2567 	return (dp->bp->bnum * MAX_CPU_UNITS_PER_BOARD + dp->unum);
2568 }
2569 
2570 /*
2571  * Move bootproc (SIGBCPU) to another cpu.  If dst_cpu is NULL, a
2572  * destination cpu is chosen from the set of cpus not located on the
2573  * same board as the current bootproc cpu.
2574  */
2575 static sbd_error_t *
2576 drmach_cpu_juggle_bootproc(drmach_device_t *dst_cpu)
2577 {
2578 	processorid_t	 cpuid;
2579 	struct cpu	*cp;
2580 	sbd_error_t	*err;
2581 	int		 rv;
2582 
2583 	ASSERT(MUTEX_HELD(&cpu_lock));
2584 
2585 	/* dst_cpu is NULL when target cpu is unspecified. So, pick one. */
2586 	if (dst_cpu == NULL) {
2587 		int avoid_board = DRMACH_CPUID2BNUM(SIGBCPU->cpu_id);
2588 		int max_cpuid = MAX_BOARDS * MAX_CPU_UNITS_PER_BOARD;
2589 
2590 		for (cpuid = 0; cpuid < max_cpuid; cpuid++)
2591 			if (DRMACH_CPUID2BNUM(cpuid) != avoid_board) {
2592 				cp = cpu_get(cpuid);
2593 				if (cp != NULL && cpu_is_online(cp))
2594 					break;
2595 			}
2596 
2597 		if (cpuid == max_cpuid) {
2598 			err = drerr_new(1, ESTF_JUGGLE, NULL);
2599 			return (err);
2600 		}
2601 
2602 		/* else, cp points to the selected target cpu */
2603 	} else {
2604 		cpuid = drmach_cpu_calc_id(dst_cpu);
2605 
2606 		if ((cp = cpu_get(cpuid)) == NULL) {
2607 			err = drerr_new(1, ESTF_NODEV, "%s::%s",
2608 				dst_cpu->bp->cm.name, dst_cpu->cm.name);
2609 			return (err);
2610 		}
2611 
2612 		if (cpuid == SIGBCPU->cpu_id) {
2613 			cmn_err(CE_WARN,
2614 				"SIGBCPU(%d) same as new selection(%d)",
2615 				SIGBCPU->cpu_id, cpuid);
2616 
2617 			/* technically not an error, but a no-op */
2618 			return (NULL);
2619 		}
2620 	}
2621 
2622 	cmn_err(CE_NOTE, "?relocating SIGBCPU from %d to %d",
2623 		SIGBCPU->cpu_id, cpuid);
2624 
2625 	DRMACH_PR("moving SIGBCPU to CPU %d\n", cpuid);
2626 
2627 	/*
2628 	 * Tell OBP to initialize cvc-offset field of new CPU0
2629 	 * so that it's in sync with OBP and cvc_server
2630 	 */
2631 	prom_starfire_init_console(cpuid);
2632 
2633 	/*
2634 	 * Assign cvc to new cpu0's bbsram for I/O.  This has to be
2635 	 * done BEFORE cpu0 is moved via obp, since this logic
2636 	 * will cause obp_helper to switch to a different bbsram for
2637 	 * cvc I/O.  We don't want cvc writing to a buffer from which
2638 	 * nobody will pick up the data!
2639 	 */
2640 	cvc_assign_iocpu(cpuid);
2641 
2642 	rv = prom_starfire_move_cpu0(cpuid);
2643 
2644 	if (rv == 0) {
2645 		SIGBCPU = cp;
2646 
2647 		DRMACH_PR("successfully juggled to CPU %d\n", cpuid);
2648 		return (NULL);
2649 	} else {
2650 		DRMACH_PR("prom error: prom_starfire_move_cpu0(%d) "
2651 			"returned %d\n", cpuid, rv);
2652 
2653 		/*
2654 		 * The move failed, hopefully obp_helper is still back
2655 		 * at the old bootproc.  Move cvc back there.
2656 		 */
2657 		cvc_assign_iocpu(SIGBCPU->cpu_id);
2658 
2659 
2660 		err = drerr_new(1, ESTF_MOVESIGB, "CPU %d", cpuid);
2661 		return (err);
2662 	}
2663 	/*NOTREACHED*/
2664 }
2665 
2666 static sbd_error_t *
2667 drmach_cpu_release(drmachid_t id)
2668 {
2669 	drmach_device_t	*dp;
2670 	processorid_t	 cpuid;
2671 	struct cpu	*cp;
2672 	sbd_error_t	*err;
2673 
2674 	if (!DRMACH_IS_CPU_ID(id))
2675 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2676 	dp = id;
2677 	cpuid = drmach_cpu_calc_id(dp);
2678 
2679 	ASSERT(MUTEX_HELD(&cpu_lock));
2680 
2681 	cp = cpu_get(cpuid);
2682 	if (cp == NULL)
2683 		err = DRMACH_INTERNAL_ERROR();
2684 	else if (SIGBCPU->cpu_id == cp->cpu_id)
2685 		err = drmach_cpu_juggle_bootproc(NULL);
2686 	else
2687 		err = NULL;
2688 
2689 	return (err);
2690 }
2691 
2692 static sbd_error_t *
2693 drmach_cpu_status(drmachid_t id, drmach_status_t *stat)
2694 {
2695 	drmach_device_t *dp;
2696 
2697 	ASSERT(DRMACH_IS_CPU_ID(id));
2698 	dp = id;
2699 
2700 	stat->assigned = dp->bp->assigned;
2701 	stat->powered = dp->bp->powered;
2702 	mutex_enter(&cpu_lock);
2703 	stat->configured = (cpu_get(drmach_cpu_calc_id(dp)) != NULL);
2704 	mutex_exit(&cpu_lock);
2705 	stat->busy = dp->busy;
2706 	strncpy(stat->type, dp->type, sizeof (stat->type));
2707 	stat->info[0] = '\0';
2708 
2709 	return (NULL);
2710 }
2711 
2712 sbd_error_t *
2713 drmach_cpu_disconnect(drmachid_t id)
2714 {
2715 	drmach_device_t	*cpu;
2716 	int		 cpuid;
2717 	int		 ntries;
2718 	int		 p;
2719 	u_longlong_t	 pc_addr;
2720 	uchar_t		 rvalue;
2721 
2722 	if (!DRMACH_IS_CPU_ID(id))
2723 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2724 	cpu = id;
2725 
2726 	cpuid = drmach_cpu_calc_id(cpu);
2727 	if (SIGBCPU->cpu_id == cpuid) {
2728 		/* this cpu is SIGBCPU, can't disconnect */
2729 		return (drerr_new(1, ESTF_HASSIGB, "%s::%s",
2730 				cpu->bp->cm.name, cpu->cm.name));
2731 	}
2732 
2733 	/*
2734 	 * Make sure SIGBST_DETACHED is set before
2735 	 * mapping out the sig block.
2736 	 */
2737 	ntries = drmach_cpu_ntries;
2738 	while (!drmach_cpu_obp_is_detached(cpuid) && ntries) {
2739 		DELAY(drmach_cpu_delay);
2740 		ntries--;
2741 	}
2742 	if (!drmach_cpu_obp_is_detached(cpuid)) {
2743 		cmn_err(CE_WARN, "failed to mark cpu %d detached in sigblock",
2744 			cpuid);
2745 	}
2746 
2747 	/* map out signature block */
2748 	if (CPU_SGN_EXISTS(cpuid)) {
2749 		CPU_SGN_MAPOUT(cpuid);
2750 	}
2751 
2752 	/*
2753 	 * We now PC IDLE the processor to guarantee we
2754 	 * stop any transactions from coming from it.
2755 	 */
2756 	p = cpu->unum & 1;
2757 	pc_addr = STARFIRE_BB_PC_ADDR(cpu->bp->bnum, cpu->unum, 0);
2758 
2759 	DRMACH_PR("PC idle cpu %d (addr = 0x%llx, port = %d, p = %d)",
2760 		drmach_cpu_calc_id(cpu), pc_addr, cpu->unum, p);
2761 
2762 	rvalue = ldbphysio(pc_addr);
2763 	rvalue |= STARFIRE_BB_PC_IDLE(p);
2764 	stbphysio(pc_addr, rvalue);
2765 	DELAY(50000);
2766 
2767 	return (NULL);
2768 }
2769 
2770 sbd_error_t *
2771 drmach_cpu_get_id(drmachid_t id, processorid_t *cpuid)
2772 {
2773 	drmach_device_t *cpu;
2774 
2775 	if (!DRMACH_IS_CPU_ID(id))
2776 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2777 	cpu = id;
2778 
2779 	*cpuid = drmach_cpu_calc_id(cpu);
2780 	return (NULL);
2781 }
2782 
2783 sbd_error_t *
2784 drmach_cpu_get_impl(drmachid_t id, int *ip)
2785 {
2786 	drmach_device_t *cpu;
2787 	int		impl;
2788 
2789 	if (!DRMACH_IS_CPU_ID(id))
2790 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2791 
2792 	cpu = id;
2793 
2794 	if (drmach_node_get_prop(cpu->node, "implementation#", &impl) == -1) {
2795 		return (DRMACH_INTERNAL_ERROR());
2796 	}
2797 
2798 	*ip = impl;
2799 
2800 	return (NULL);
2801 }
2802 
2803 void
2804 drmach_cpu_flush_ecache_sync(void)
2805 {
2806 	ASSERT(curthread->t_bound_cpu == CPU);
2807 
2808 	/*
2809 	 * Now let's flush our ecache thereby removing all references
2810 	 * to the target (detaching) memory from all ecache's in
2811 	 * system.
2812 	 */
2813 	cpu_flush_ecache();
2814 
2815 	/*
2816 	 * Delay 100 usec out of paranoia to insure everything
2817 	 * (hardware queues) has drained before we start reprogramming
2818 	 * the hardware.
2819 	 */
2820 	DELAY(100);
2821 }
2822 
2823 sbd_error_t *
2824 drmach_get_dip(drmachid_t id, dev_info_t **dip)
2825 {
2826 	drmach_device_t	*dp;
2827 
2828 	if (!DRMACH_IS_DEVICE_ID(id))
2829 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2830 	dp = id;
2831 
2832 	*dip = drmach_node_get_dip(dp->node);
2833 	return (NULL);
2834 }
2835 
2836 sbd_error_t *
2837 drmach_io_is_attached(drmachid_t id, int *yes)
2838 {
2839 	drmach_device_t *dp;
2840 	dev_info_t	*dip;
2841 	int		state;
2842 
2843 	if (!DRMACH_IS_IO_ID(id))
2844 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2845 	dp = id;
2846 
2847 	dip = drmach_node_get_dip(dp->node);
2848 	if (dip == NULL) {
2849 		*yes = 0;
2850 		return (NULL);
2851 	}
2852 
2853 	state = ddi_get_devstate(dip);
2854 	*yes = (i_ddi_devi_attached(dip) || (state == DDI_DEVSTATE_UP));
2855 
2856 	return (NULL);
2857 }
2858 
2859 sbd_error_t *
2860 drmach_io_pre_release(drmachid_t id)
2861 {
2862 	if (!DRMACH_IS_IO_ID(id))
2863 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2864 	return (NULL);
2865 }
2866 
2867 static sbd_error_t *
2868 drmach_io_release(drmachid_t id)
2869 {
2870 	if (!DRMACH_IS_IO_ID(id))
2871 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2872 	return (NULL);
2873 }
2874 
2875 sbd_error_t *
2876 drmach_io_unrelease(drmachid_t id)
2877 {
2878 	if (!DRMACH_IS_IO_ID(id))
2879 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2880 	return (NULL);
2881 }
2882 
2883 /*ARGSUSED*/
2884 sbd_error_t *
2885 drmach_io_post_release(drmachid_t id)
2886 {
2887 	return (NULL);
2888 }
2889 
2890 /*ARGSUSED*/
2891 sbd_error_t *
2892 drmach_io_post_attach(drmachid_t id)
2893 {
2894 	return (NULL);
2895 }
2896 
2897 static sbd_error_t *
2898 drmach_io_status(drmachid_t id, drmach_status_t *stat)
2899 {
2900 	drmach_device_t *dp;
2901 	sbd_error_t	*err;
2902 	int		 configured;
2903 
2904 	ASSERT(DRMACH_IS_IO_ID(id));
2905 	dp = id;
2906 
2907 	err = drmach_io_is_attached(id, &configured);
2908 	if (err)
2909 		return (err);
2910 
2911 	stat->assigned = dp->bp->assigned;
2912 	stat->powered = dp->bp->powered;
2913 	stat->configured = (configured != 0);
2914 	stat->busy = dp->busy;
2915 	strncpy(stat->type, dp->type, sizeof (stat->type));
2916 	stat->info[0] = '\0';
2917 
2918 	return (NULL);
2919 }
2920 
2921 static sbd_error_t *
2922 drmach_mem_new(drmach_device_t *dp)
2923 {
2924 	static sbd_error_t *drmach_mem_release(drmachid_t);
2925 	static sbd_error_t *drmach_mem_status(drmachid_t, drmach_status_t *);
2926 
2927 	dp->unum = 0;
2928 	dp->cm.isa = (void *)drmach_mem_new;
2929 	dp->cm.release = drmach_mem_release;
2930 	dp->cm.status = drmach_mem_status;
2931 
2932 	snprintf(dp->cm.name, sizeof (dp->cm.name), "%s", dp->type);
2933 
2934 	return (NULL);
2935 }
2936 
2937 sbd_error_t *
2938 drmach_mem_add_span(drmachid_t id, uint64_t basepa, uint64_t size)
2939 {
2940 	pfn_t		basepfn = (pfn_t)(basepa >> PAGESHIFT);
2941 	pgcnt_t		npages = (pgcnt_t)(size >> PAGESHIFT);
2942 	pda_handle_t	ph;
2943 	int		rv;
2944 
2945 	ASSERT(size != 0);
2946 
2947 	if (!DRMACH_IS_MEM_ID(id))
2948 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2949 
2950 	kcage_range_lock();
2951 	rv = kcage_range_add(basepfn, npages, 1);
2952 	kcage_range_unlock();
2953 	if (rv == ENOMEM) {
2954 		cmn_err(CE_WARN, "%ld megabytes not available to kernel cage",
2955 			(size == 0 ? 0 : size / MBYTE));
2956 	} else if (rv != 0) {
2957 		/* catch this in debug kernels */
2958 		ASSERT(0);
2959 
2960 		cmn_err(CE_WARN, "unexpected kcage_range_add"
2961 			" return value %d", rv);
2962 	}
2963 
2964 	/*
2965 	 * Update the PDA (post2obp) structure with the
2966 	 * range of the newly added memory.
2967 	 */
2968 	ph = drmach_pda_open();
2969 	if (ph != NULL) {
2970 		pda_mem_add_span(ph, basepa, size);
2971 		pda_close(ph);
2972 	}
2973 
2974 	return (NULL);
2975 }
2976 
2977 sbd_error_t *
2978 drmach_mem_del_span(drmachid_t id, uint64_t basepa, uint64_t size)
2979 {
2980 	drmach_device_t	*mem = id;
2981 	pfn_t		basepfn = (pfn_t)(basepa >> PAGESHIFT);
2982 	pgcnt_t		npages = (pgcnt_t)(size >> PAGESHIFT);
2983 	uint_t		mcreg;
2984 	sbd_error_t	*err;
2985 	pda_handle_t	ph;
2986 	int		rv;
2987 
2988 	err = drmach_read_mc_asr(id, &mcreg);
2989 	if (err)
2990 		return (err);
2991 	else if (mcreg & STARFIRE_MC_INTERLEAVE_MASK) {
2992 		return (drerr_new(1, ESTF_INTERBOARD, "%s::%s",
2993 				mem->bp->cm.name, mem->cm.name));
2994 	}
2995 
2996 	if (size > 0) {
2997 		kcage_range_lock();
2998 		rv = kcage_range_delete_post_mem_del(basepfn, npages);
2999 		kcage_range_unlock();
3000 		if (rv != 0) {
3001 			cmn_err(CE_WARN,
3002 			    "unexpected kcage_range_delete_post_mem_del"
3003 			    " return value %d", rv);
3004 			return (DRMACH_INTERNAL_ERROR());
3005 		}
3006 	}
3007 
3008 	/*
3009 	 * Update the PDA (post2obp) structure with the
3010 	 * range of removed memory.
3011 	 */
3012 	ph = drmach_pda_open();
3013 	if (ph != NULL) {
3014 		if (size > 0)
3015 			pda_mem_del_span(ph, basepa, size);
3016 
3017 		/* update PDA to board's new mc register settings */
3018 		pda_mem_sync(ph, mem->bp->bnum, 0);
3019 
3020 		pda_close(ph);
3021 	}
3022 
3023 	return (NULL);
3024 }
3025 
3026 /* support routine for enable and disable */
3027 static sbd_error_t *
3028 drmach_mem_update_interconnect(drmachid_t id, uint_t mcreg)
3029 {
3030 	drmach_device_t	*dp;
3031 	pda_handle_t	 ph;
3032 	int		 b;
3033 
3034 	if (!DRMACH_IS_MEM_ID(id))
3035 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3036 	dp = id;
3037 
3038 	ph = drmach_pda_open();
3039 	if (ph == NULL)
3040 		return (DRMACH_INTERNAL_ERROR());
3041 
3042 	for (b = 0; b < MAX_BOARDS; b++) {
3043 		int		p;
3044 		int		rv;
3045 		ushort_t	bda_proc, bda_ioc;
3046 		board_desc_t	*bdesc;
3047 
3048 		if (pda_board_present(ph, b) == 0)
3049 			continue;
3050 
3051 		bdesc = (board_desc_t *)pda_get_board_info(ph, b);
3052 
3053 		/*
3054 		 * Update PCs for CPUs.
3055 		 */
3056 
3057 		/* make sure definition in platmod is in sync with pda */
3058 		ASSERT(MAX_PROCMODS == MAX_CPU_UNITS_PER_BOARD);
3059 
3060 		bda_proc = bdesc->bda_proc;
3061 		for (p = 0; p < MAX_PROCMODS; p++) {
3062 			if (BDA_NBL(bda_proc, p) != BDAN_GOOD)
3063 				continue;
3064 
3065 			rv = pc_madr_add(b, dp->bp->bnum, p, mcreg);
3066 			if (rv) {
3067 				pda_close(ph);
3068 				return (DRMACH_INTERNAL_ERROR());
3069 			}
3070 		}
3071 
3072 		/*
3073 		 * Update PCs for IOCs.
3074 		 */
3075 
3076 		/* make sure definition in platmod is in sync with pda */
3077 		ASSERT(MAX_IOCS == MAX_IO_UNITS_PER_BOARD);
3078 
3079 		bda_ioc = bdesc->bda_ioc;
3080 		for (p = 0; p < MAX_IOCS; p++) {
3081 			if (BDA_NBL(bda_ioc, p) != BDAN_GOOD)
3082 				continue;
3083 
3084 			rv = pc_madr_add(b, dp->bp->bnum, p + 4, mcreg);
3085 			if (rv) {
3086 				pda_close(ph);
3087 				return (DRMACH_INTERNAL_ERROR());
3088 			}
3089 		}
3090 	}
3091 
3092 	pda_close(ph);
3093 	return (NULL);
3094 }
3095 
3096 sbd_error_t *
3097 drmach_mem_disable(drmachid_t id)
3098 {
3099 	sbd_error_t	*err;
3100 	uint_t		 mcreg;
3101 
3102 	err = drmach_read_mc_asr(id, &mcreg);
3103 	if (err == NULL) {
3104 		ASSERT(mcreg & STARFIRE_MC_MEM_PRESENT_MASK);
3105 
3106 		/* Turn off presence bit. */
3107 		mcreg &= ~STARFIRE_MC_MEM_PRESENT_MASK;
3108 
3109 		err = drmach_mem_update_interconnect(id, mcreg);
3110 		if (err == NULL)
3111 			err = drmach_write_mc_asr(id, mcreg);
3112 	}
3113 
3114 	return (err);
3115 }
3116 
3117 sbd_error_t *
3118 drmach_mem_enable(drmachid_t id)
3119 {
3120 	sbd_error_t	*err;
3121 	uint_t		 mcreg;
3122 
3123 	err = drmach_read_mc_asr(id, &mcreg);
3124 	if (err == NULL) {
3125 		mcreg |= STARFIRE_MC_MEM_PRESENT_MASK;
3126 
3127 		err = drmach_write_mc_asr(id, mcreg);
3128 		if (err == NULL)
3129 			err = drmach_mem_update_interconnect(id, mcreg);
3130 	}
3131 
3132 	return (err);
3133 }
3134 
3135 sbd_error_t *
3136 drmach_mem_get_alignment(drmachid_t id, uint64_t *mask)
3137 {
3138 	drmach_device_t	*mem;
3139 	sbd_error_t	*err;
3140 	pnode_t		 nodeid;
3141 
3142 	if (!DRMACH_IS_MEM_ID(id))
3143 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3144 	mem = id;
3145 
3146 	nodeid = drmach_node_get_dnode(mem->node);
3147 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
3148 		err = DRMACH_INTERNAL_ERROR();
3149 	else {
3150 		uint64_t size;
3151 
3152 		size = mc_get_alignment_mask(nodeid);
3153 		if (size == (uint64_t)-1)
3154 			err = DRMACH_INTERNAL_ERROR();
3155 		else {
3156 			*mask = size - 1;
3157 			err = NULL;
3158 		}
3159 	}
3160 
3161 	return (err);
3162 }
3163 
3164 sbd_error_t *
3165 drmach_mem_get_base_physaddr(drmachid_t id, uint64_t *pa)
3166 {
3167 	sbd_error_t	*err;
3168 	uint_t		 mcreg;
3169 
3170 	err = drmach_read_mc_asr(id, &mcreg);
3171 	if (err == NULL)
3172 		*pa = mc_asr_to_pa(mcreg);
3173 
3174 	return (err);
3175 }
3176 
3177 /*
3178  * Use of this routine after copy/rename will yield incorrect results,
3179  * because the OBP MEMAVAIL property will not correctly reflect the
3180  * programming of the MCs.
3181  */
3182 sbd_error_t *
3183 drmach_mem_get_memlist(drmachid_t id, struct memlist **ml)
3184 {
3185 	drmach_device_t	*mem;
3186 	int		rv, i, rlen, rblks;
3187 	sbd_error_t	*err;
3188 	struct memlist	*mlist;
3189 	struct sf_memunit_regspec *rlist;
3190 
3191 	if (!DRMACH_IS_MEM_ID(id))
3192 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3193 	mem = id;
3194 
3195 	err = drmach_device_get_proplen(mem, "dr-available", &rlen);
3196 	if (err)
3197 		return (err);
3198 
3199 	rlist = kmem_zalloc(rlen, KM_SLEEP);
3200 
3201 	err = drmach_device_get_prop(mem, "dr-available", rlist);
3202 	if (err) {
3203 		kmem_free(rlist, rlen);
3204 		return (err);
3205 	}
3206 
3207 	mlist = NULL;
3208 	rblks = rlen / sizeof (struct sf_memunit_regspec);
3209 	for (i = 0; i < rblks; i++) {
3210 		uint64_t	addr, size;
3211 
3212 		addr  = (uint64_t)rlist[i].regspec_addr_hi << 32;
3213 		addr |= (uint64_t)rlist[i].regspec_addr_lo;
3214 		size  = (uint64_t)rlist[i].regspec_size_hi << 32;
3215 		size |= (uint64_t)rlist[i].regspec_size_lo;
3216 
3217 		mlist = memlist_add_span(mlist, addr, size);
3218 	}
3219 
3220 	kmem_free(rlist, rlen);
3221 
3222 	/*
3223 	 * Make sure the incoming memlist doesn't already
3224 	 * intersect with what's present in the system (phys_install).
3225 	 */
3226 	memlist_read_lock();
3227 	rv = memlist_intersect(phys_install, mlist);
3228 	memlist_read_unlock();
3229 	if (rv) {
3230 #ifdef DEBUG
3231 		DRMACH_PR("OBP derived memlist intersects"
3232 			" with phys_install\n");
3233 		memlist_dump(mlist);
3234 
3235 		DRMACH_PR("phys_install memlist:\n");
3236 		memlist_dump(phys_install);
3237 #endif
3238 
3239 		memlist_delete(mlist);
3240 		return (DRMACH_INTERNAL_ERROR());
3241 	}
3242 
3243 #ifdef DEBUG
3244 	DRMACH_PR("OBP derived memlist:");
3245 	memlist_dump(mlist);
3246 #endif
3247 
3248 	*ml = mlist;
3249 	return (NULL);
3250 }
3251 
3252 sbd_error_t *
3253 drmach_mem_get_size(drmachid_t id, uint64_t *bytes)
3254 {
3255 	drmach_device_t	*mem;
3256 	pda_handle_t	ph;
3257 	pgcnt_t		npages;
3258 
3259 	if (!DRMACH_IS_MEM_ID(id))
3260 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3261 	mem = id;
3262 
3263 	ph = drmach_pda_open();
3264 	if (ph == NULL)
3265 		return (DRMACH_INTERNAL_ERROR());
3266 
3267 	npages = pda_get_mem_size(ph, mem->bp->bnum);
3268 	*bytes = (uint64_t)npages << PAGESHIFT;
3269 
3270 	pda_close(ph);
3271 	return (NULL);
3272 }
3273 
3274 sbd_error_t *
3275 drmach_mem_get_slice_size(drmachid_t id, uint64_t *bytes)
3276 {
3277 	if (!DRMACH_IS_MEM_ID(id))
3278 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3279 
3280 	*bytes = mc_get_mem_alignment();
3281 	return (NULL);
3282 }
3283 
3284 /* field debugging tool */
3285 processorid_t drmach_mem_cpu_affinity_nail = 0;
3286 
3287 processorid_t
3288 drmach_mem_cpu_affinity(drmachid_t id)
3289 {
3290 	drmach_device_t	*mp;
3291 	drmach_board_t	*bp;
3292 	processorid_t	 cpuid;
3293 
3294 	if (!DRMACH_IS_MEM_ID(id))
3295 		return (CPU_CURRENT);
3296 
3297 	if (drmach_mem_cpu_affinity_nail) {
3298 		cpuid = drmach_mem_cpu_affinity_nail;
3299 
3300 		if (cpuid < 0 || cpuid > NCPU)
3301 			return (CPU_CURRENT);
3302 
3303 		mutex_enter(&cpu_lock);
3304 		if (cpu[cpuid] == NULL || !CPU_ACTIVE(cpu[cpuid]))
3305 			cpuid = CPU_CURRENT;
3306 		mutex_exit(&cpu_lock);
3307 
3308 		return (cpuid);
3309 	}
3310 
3311 	/* try to choose a proc on the target board */
3312 	mp = id;
3313 	bp = mp->bp;
3314 	if (bp->devices) {
3315 		int		rv;
3316 		int		d_idx;
3317 		drmachid_t	d_id;
3318 
3319 		rv = drmach_array_first(bp->devices, &d_idx, &d_id);
3320 		while (rv == 0) {
3321 			if (DRMACH_IS_CPU_ID(d_id)) {
3322 				cpuid = drmach_cpu_calc_id(d_id);
3323 
3324 				mutex_enter(&cpu_lock);
3325 				if (cpu[cpuid] && CPU_ACTIVE(cpu[cpuid])) {
3326 					mutex_exit(&cpu_lock);
3327 					DRMACH_PR("drmach_mem_cpu_affinity: "
3328 					    "selected cpuid=%d\n", cpuid);
3329 					return (cpuid);
3330 				} else {
3331 					mutex_exit(&cpu_lock);
3332 				}
3333 			}
3334 
3335 			rv = drmach_array_next(bp->devices, &d_idx, &d_id);
3336 		}
3337 	}
3338 
3339 	/* otherwise, this proc, wherever it is */
3340 	DRMACH_PR("drmach_mem_cpu_affinity: using default CPU_CURRENT\n");
3341 
3342 	return (CPU_CURRENT);
3343 }
3344 
3345 static sbd_error_t *
3346 drmach_mem_release(drmachid_t id)
3347 {
3348 	if (!DRMACH_IS_MEM_ID(id))
3349 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3350 	return (NULL);
3351 }
3352 
3353 static sbd_error_t *
3354 drmach_mem_status(drmachid_t id, drmach_status_t *stat)
3355 {
3356 	drmach_device_t *dp;
3357 	sbd_error_t	*err;
3358 	uint64_t	 pa, slice_size;
3359 	struct memlist	*ml;
3360 
3361 	ASSERT(DRMACH_IS_MEM_ID(id));
3362 	dp = id;
3363 
3364 	/* get starting physical address of target memory */
3365 	err = drmach_mem_get_base_physaddr(id, &pa);
3366 	if (err)
3367 		return (err);
3368 
3369 	/* round down to slice boundary */
3370 	slice_size = mc_get_mem_alignment();
3371 	pa &= ~ (slice_size - 1);
3372 
3373 	/* stop at first span that is in slice */
3374 	memlist_read_lock();
3375 	for (ml = phys_install; ml; ml = ml->next)
3376 		if (ml->address >= pa && ml->address < pa + slice_size)
3377 			break;
3378 	memlist_read_unlock();
3379 
3380 	stat->assigned = dp->bp->assigned;
3381 	stat->powered = dp->bp->powered;
3382 	stat->configured = (ml != NULL);
3383 	stat->busy = dp->busy;
3384 	strncpy(stat->type, dp->type, sizeof (stat->type));
3385 	stat->info[0] = '\0';
3386 
3387 	return (NULL);
3388 }
3389 
3390 static int
3391 drmach_detach_board(void *arg)
3392 {
3393 	cpuset_t	cset;
3394 	int		retval;
3395 	drmach_board_t	*bp = (drmach_board_t *)arg;
3396 
3397 	cset = cpu_ready_set;
3398 	promsafe_xc_attention(cset);
3399 
3400 	retval = prom_starfire_rm_brd(bp->bnum);
3401 
3402 	xc_dismissed(cset);
3403 
3404 	return (retval);
3405 }
3406 
3407 sbd_error_t *
3408 drmach_board_deprobe(drmachid_t id)
3409 {
3410 	drmach_board_t	*bp;
3411 	int		 retval;
3412 
3413 	if (!DRMACH_IS_BOARD_ID(id))
3414 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3415 	bp = id;
3416 
3417 	cmn_err(CE_CONT, "DR: PROM detach board %d\n", bp->bnum);
3418 
3419 	retval = prom_tree_update(drmach_detach_board, bp);
3420 
3421 	if (retval == 0)
3422 		return (NULL);
3423 	else {
3424 		cmn_err(CE_WARN, "prom error: prom_starfire_rm_brd(%d) "
3425 			"returned %d", bp->bnum, retval);
3426 		return (drerr_new(1, ESTF_DEPROBE, "%s", bp->cm.name));
3427 	}
3428 }
3429 
3430 /*ARGSUSED*/
3431 static sbd_error_t *
3432 drmach_pt_juggle_bootproc(drmachid_t id, drmach_opts_t *opts)
3433 {
3434 	drmach_device_t	*cpu;
3435 	sbd_error_t	*err;
3436 
3437 	if (!DRMACH_IS_CPU_ID(id))
3438 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3439 	cpu = id;
3440 
3441 	mutex_enter(&cpu_lock);
3442 
3443 	err = drmach_cpu_juggle_bootproc(cpu);
3444 
3445 	mutex_exit(&cpu_lock);
3446 
3447 	return (err);
3448 }
3449 
3450 /*ARGSUSED*/
3451 static sbd_error_t *
3452 drmach_pt_dump_pdainfo(drmachid_t id, drmach_opts_t *opts)
3453 {
3454 	drmach_board_t	*bp;
3455 	int		board;
3456 	int		i;
3457 	pda_handle_t	ph;
3458 	board_desc_t	*bdesc;
3459 
3460 	if (!DRMACH_IS_BOARD_ID(id))
3461 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3462 	bp = id;
3463 	board = bp->bnum;
3464 
3465 	ph = drmach_pda_open();
3466 	if (ph == NULL)
3467 		return (DRMACH_INTERNAL_ERROR());
3468 
3469 	if (pda_board_present(ph, board) == 0) {
3470 		cmn_err(CE_CONT, "board %d is MISSING\n", board);
3471 		pda_close(ph);
3472 		return (DRMACH_INTERNAL_ERROR());
3473 	}
3474 
3475 	cmn_err(CE_CONT, "board %d is PRESENT\n", board);
3476 
3477 	bdesc = (board_desc_t *)pda_get_board_info(ph, board);
3478 	if (bdesc == NULL) {
3479 		cmn_err(CE_CONT,
3480 			"no board descriptor found for board %d\n",
3481 			board);
3482 		pda_close(ph);
3483 		return (DRMACH_INTERNAL_ERROR());
3484 	}
3485 
3486 	/* make sure definition in platmod is in sync with pda */
3487 	ASSERT(MAX_PROCMODS == MAX_CPU_UNITS_PER_BOARD);
3488 
3489 	for (i = 0; i < MAX_PROCMODS; i++) {
3490 		if (BDA_NBL(bdesc->bda_proc, i) == BDAN_GOOD)
3491 			cmn_err(CE_CONT,
3492 				"proc %d.%d PRESENT\n", board, i);
3493 		else
3494 			cmn_err(CE_CONT,
3495 				"proc %d.%d MISSING\n", board, i);
3496 	}
3497 
3498 	for (i = 0; i < MAX_MGROUPS; i++) {
3499 		if (BDA_NBL(bdesc->bda_mgroup, i) == BDAN_GOOD)
3500 			cmn_err(CE_CONT,
3501 				"mgroup %d.%d PRESENT\n", board, i);
3502 		else
3503 			cmn_err(CE_CONT,
3504 				"mgroup %d.%d MISSING\n", board, i);
3505 	}
3506 
3507 	/* make sure definition in platmod is in sync with pda */
3508 	ASSERT(MAX_IOCS == MAX_IO_UNITS_PER_BOARD);
3509 
3510 	for (i = 0; i < MAX_IOCS; i++) {
3511 		int	s;
3512 
3513 		if (BDA_NBL(bdesc->bda_ioc, i) == BDAN_GOOD) {
3514 			cmn_err(CE_CONT,
3515 				"ioc %d.%d PRESENT\n", board, i);
3516 			for (s = 0; s < MAX_SLOTS_PER_IOC; s++) {
3517 				if (BDA_NBL(bdesc->bda_ios[i], s) != BDAN_GOOD)
3518 					continue;
3519 				cmn_err(CE_CONT,
3520 					"..scard %d.%d.%d PRESENT\n",
3521 					board, i, s);
3522 			}
3523 		} else {
3524 			cmn_err(CE_CONT,
3525 				"ioc %d.%d MISSING\n",
3526 				board, i);
3527 		}
3528 	}
3529 
3530 	cmn_err(CE_CONT,
3531 		"board %d memsize = %d pages\n",
3532 		board, pda_get_mem_size(ph, board));
3533 
3534 	pda_close(ph);
3535 
3536 	return (NULL);
3537 }
3538 
3539 /*ARGSUSED*/
3540 sbd_error_t *
3541 drmach_pt_readmem(drmachid_t id, drmach_opts_t *opts)
3542 {
3543 	struct memlist	*ml;
3544 	uint64_t	src_pa;
3545 	uint64_t	dst_pa;
3546 	uint64_t	dst;
3547 
3548 	dst_pa = va_to_pa(&dst);
3549 
3550 	memlist_read_lock();
3551 	for (ml = phys_install; ml; ml = ml->next) {
3552 		uint64_t	nbytes;
3553 
3554 		src_pa = ml->address;
3555 		nbytes = ml->size;
3556 
3557 		while (nbytes != 0ull) {
3558 
3559 			/* copy 32 bytes at arc_pa to dst_pa */
3560 			bcopy32_il(src_pa, dst_pa);
3561 
3562 			/* increment by 32 bytes */
3563 			src_pa += (4 * sizeof (uint64_t));
3564 
3565 			/* decrement by 32 bytes */
3566 			nbytes -= (4 * sizeof (uint64_t));
3567 		}
3568 	}
3569 	memlist_read_unlock();
3570 
3571 	return (NULL);
3572 }
3573 
3574 static struct {
3575 	const char	*name;
3576 	sbd_error_t	*(*handler)(drmachid_t id, drmach_opts_t *opts);
3577 } drmach_pt_arr[] = {
3578 	{ "juggle",		drmach_pt_juggle_bootproc	},
3579 	{ "pda",		drmach_pt_dump_pdainfo		},
3580 	{ "readmem",		drmach_pt_readmem		},
3581 
3582 	/* the following line must always be last */
3583 	{ NULL,			NULL				}
3584 };
3585 
3586 /*ARGSUSED*/
3587 sbd_error_t *
3588 drmach_passthru(drmachid_t id, drmach_opts_t *opts)
3589 {
3590 	int		i;
3591 	sbd_error_t	*err;
3592 
3593 	i = 0;
3594 	while (drmach_pt_arr[i].name != NULL) {
3595 		int len = strlen(drmach_pt_arr[i].name);
3596 
3597 		if (strncmp(drmach_pt_arr[i].name, opts->copts, len) == 0)
3598 			break;
3599 
3600 		i += 1;
3601 	}
3602 
3603 	if (drmach_pt_arr[i].name == NULL)
3604 		err = drerr_new(0, ESTF_UNKPTCMD, opts->copts);
3605 	else
3606 		err = (*drmach_pt_arr[i].handler)(id, opts);
3607 
3608 	return (err);
3609 }
3610 
3611 sbd_error_t *
3612 drmach_release(drmachid_t id)
3613 {
3614 	drmach_common_t *cp;
3615 	if (!DRMACH_IS_DEVICE_ID(id))
3616 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3617 	cp = id;
3618 
3619 	return (cp->release(id));
3620 }
3621 
3622 sbd_error_t *
3623 drmach_status(drmachid_t id, drmach_status_t *stat)
3624 {
3625 	drmach_common_t *cp;
3626 
3627 	if (!DRMACH_IS_ID(id))
3628 		return (drerr_new(0, ESTF_NOTID, NULL));
3629 	cp = id;
3630 
3631 	return (cp->status(id, stat));
3632 }
3633 
3634 sbd_error_t *
3635 drmach_unconfigure(drmachid_t id, int flags)
3636 {
3637 	drmach_device_t	*dp;
3638 	pnode_t		 nodeid;
3639 	dev_info_t	*dip, *fdip = NULL;
3640 
3641 	if (!DRMACH_IS_DEVICE_ID(id))
3642 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3643 
3644 	dp = id;
3645 
3646 	nodeid = drmach_node_get_dnode(dp->node);
3647 	if (nodeid == OBP_NONODE)
3648 		return (DRMACH_INTERNAL_ERROR());
3649 
3650 	dip = e_ddi_nodeid_to_dip(nodeid);
3651 	if (dip == NULL)
3652 		return (NULL);
3653 
3654 	/*
3655 	 * Branch already held, so hold acquired in
3656 	 * e_ddi_nodeid_to_dip() can be released
3657 	 */
3658 	ddi_release_devi(dip);
3659 
3660 	if (flags & DEVI_BRANCH_DESTROY)
3661 		flags |= DEVI_BRANCH_EVENT;
3662 
3663 	/*
3664 	 * Force flag is no longer necessary. See starcat/io/drmach.c
3665 	 * for details.
3666 	 */
3667 	ASSERT(e_ddi_branch_held(dip));
3668 	if (e_ddi_branch_unconfigure(dip, &fdip, flags)) {
3669 		sbd_error_t	*err;
3670 		char		*path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3671 
3672 		/*
3673 		 * If non-NULL, fdip is returned held and must be released.
3674 		 */
3675 		if (fdip != NULL) {
3676 			(void) ddi_pathname(fdip, path);
3677 			ndi_rele_devi(fdip);
3678 		} else {
3679 			(void) ddi_pathname(dip, path);
3680 		}
3681 
3682 		err = drerr_new(1, ESTF_DRVFAIL, path);
3683 
3684 		kmem_free(path, MAXPATHLEN);
3685 
3686 		return (err);
3687 	}
3688 
3689 	return (NULL);
3690 }
3691 
3692 /*
3693  * drmach interfaces to legacy Starfire platmod logic
3694  * linkage via runtime symbol look up, called from plat_cpu_power*
3695  */
3696 
3697 /*
3698  * Start up a cpu.  It is possible that we're attempting to restart
3699  * the cpu after an UNCONFIGURE in which case the cpu will be
3700  * spinning in its cache.  So, all we have to do is wakeup him up.
3701  * Under normal circumstances the cpu will be coming from a previous
3702  * CONNECT and thus will be spinning in OBP.  In both cases, the
3703  * startup sequence is the same.
3704  */
3705 int
3706 drmach_cpu_poweron(struct cpu *cp)
3707 {
3708 	DRMACH_PR("drmach_cpu_poweron: starting cpuid %d\n", cp->cpu_id);
3709 
3710 	ASSERT(MUTEX_HELD(&cpu_lock));
3711 
3712 	if (drmach_cpu_start(cp) != 0)
3713 		return (EBUSY);
3714 	else
3715 		return (0);
3716 }
3717 
3718 int
3719 drmach_cpu_poweroff(struct cpu *cp)
3720 {
3721 	int		ntries, cnt;
3722 	processorid_t	cpuid = cp->cpu_id;
3723 	void		drmach_cpu_shutdown_self(void);
3724 
3725 	DRMACH_PR("drmach_cpu_poweroff: stopping cpuid %d\n", cp->cpu_id);
3726 
3727 	ASSERT(MUTEX_HELD(&cpu_lock));
3728 
3729 	/*
3730 	 * Capture all CPUs (except for detaching proc) to prevent
3731 	 * crosscalls to the detaching proc until it has cleared its
3732 	 * bit in cpu_ready_set.
3733 	 *
3734 	 * The CPU's remain paused and the prom_mutex is known to be free.
3735 	 * This prevents the x-trap victim from blocking when doing prom
3736 	 * IEEE-1275 calls at a high PIL level.
3737 	 */
3738 	promsafe_pause_cpus();
3739 
3740 	/*
3741 	 * Quiesce interrupts on the target CPU. We do this by setting
3742 	 * the CPU 'not ready'- (i.e. removing the CPU from cpu_ready_set) to
3743 	 * prevent it from receiving cross calls and cross traps.
3744 	 * This prevents the processor from receiving any new soft interrupts.
3745 	 */
3746 	mp_cpu_quiesce(cp);
3747 
3748 	/* setup xt_mb, will be cleared by drmach_shutdown_asm when ready */
3749 	drmach_xt_mb[cpuid] = 0x80;
3750 
3751 	xt_one_unchecked(cpuid, (xcfunc_t *)idle_stop_xcall,
3752 		(uint64_t)drmach_cpu_shutdown_self, NULL);
3753 
3754 	ntries = drmach_cpu_ntries;
3755 	cnt = 0;
3756 	while (drmach_xt_mb[cpuid] && ntries) {
3757 		DELAY(drmach_cpu_delay);
3758 		ntries--;
3759 		cnt++;
3760 	}
3761 
3762 	drmach_xt_mb[cpuid] = 0;	/* steal the cache line back */
3763 
3764 	start_cpus();
3765 
3766 	DRMACH_PR("waited %d out of %d tries for "
3767 		"drmach_cpu_shutdown_self on cpu%d",
3768 		drmach_cpu_ntries - ntries, drmach_cpu_ntries, cp->cpu_id);
3769 
3770 	drmach_cpu_obp_detach(cpuid);
3771 
3772 	CPU_SIGNATURE(OS_SIG, SIGST_DETACHED, SIGSUBST_NULL, cpuid);
3773 
3774 	return (0);
3775 }
3776 
3777 /*ARGSUSED*/
3778 int
3779 drmach_verify_sr(dev_info_t *dip, int sflag)
3780 {
3781 	return (0);
3782 }
3783 
3784 void
3785 drmach_suspend_last(void)
3786 {
3787 }
3788 
3789 void
3790 drmach_resume_first(void)
3791 {
3792 }
3793 
3794 /*
3795  * Log a DR sysevent.
3796  * Return value: 0 success, non-zero failure.
3797  */
3798 int
3799 drmach_log_sysevent(int board, char *hint, int flag, int verbose)
3800 {
3801 	sysevent_t			*ev;
3802 	sysevent_id_t			eid;
3803 	int				rv, km_flag;
3804 	sysevent_value_t		evnt_val;
3805 	sysevent_attr_list_t		*evnt_attr_list = NULL;
3806 	char				attach_pnt[MAXNAMELEN];
3807 
3808 	km_flag = (flag == SE_SLEEP) ? KM_SLEEP : KM_NOSLEEP;
3809 	attach_pnt[0] = '\0';
3810 	if (drmach_board_name(board, attach_pnt, MAXNAMELEN)) {
3811 		rv = -1;
3812 		goto logexit;
3813 	}
3814 	if (verbose)
3815 		DRMACH_PR("drmach_log_sysevent: %s %s, flag: %d, verbose: %d\n",
3816 			    attach_pnt, hint, flag, verbose);
3817 
3818 	if ((ev = sysevent_alloc(EC_DR, ESC_DR_AP_STATE_CHANGE,
3819 				    SUNW_KERN_PUB"dr", km_flag)) == NULL) {
3820 		rv = -2;
3821 		goto logexit;
3822 	}
3823 	evnt_val.value_type = SE_DATA_TYPE_STRING;
3824 	evnt_val.value.sv_string = attach_pnt;
3825 	if ((rv = sysevent_add_attr(&evnt_attr_list, DR_AP_ID,
3826 				    &evnt_val, km_flag)) != 0)
3827 		goto logexit;
3828 
3829 	evnt_val.value_type = SE_DATA_TYPE_STRING;
3830 	evnt_val.value.sv_string = hint;
3831 	if ((rv = sysevent_add_attr(&evnt_attr_list, DR_HINT,
3832 				    &evnt_val, km_flag)) != 0) {
3833 		sysevent_free_attr(evnt_attr_list);
3834 		goto logexit;
3835 	}
3836 
3837 	(void) sysevent_attach_attributes(ev, evnt_attr_list);
3838 
3839 	/*
3840 	 * Log the event but do not sleep waiting for its
3841 	 * delivery. This provides insulation from syseventd.
3842 	 */
3843 	rv = log_sysevent(ev, SE_NOSLEEP, &eid);
3844 
3845 logexit:
3846 	if (ev)
3847 		sysevent_free(ev);
3848 	if ((rv != 0) && verbose)
3849 		cmn_err(CE_WARN,
3850 			    "drmach_log_sysevent failed (rv %d) for %s  %s\n",
3851 			    rv, attach_pnt, hint);
3852 
3853 	return (rv);
3854 }
3855 
3856 /*ARGSUSED*/
3857 int
3858 drmach_allow_memrange_modify(drmachid_t id)
3859 {
3860 	return (1);	/* TRUE */
3861 }
3862