xref: /titanic_51/usr/src/uts/sun4u/starfire/io/drmach.c (revision 8b4cf83723ce3e17ebe029b21f549390e849b03f)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #include <sys/debug.h>
28 #include <sys/types.h>
29 #include <sys/varargs.h>
30 #include <sys/errno.h>
31 #include <sys/cred.h>
32 #include <sys/dditypes.h>
33 #include <sys/devops.h>
34 #include <sys/modctl.h>
35 #include <sys/poll.h>
36 #include <sys/conf.h>
37 #include <sys/ddi.h>
38 #include <sys/sunddi.h>
39 #include <sys/sunndi.h>
40 #include <sys/ndi_impldefs.h>
41 #include <sys/stat.h>
42 #include <sys/kmem.h>
43 #include <sys/vmem.h>
44 #include <sys/processor.h>
45 #include <sys/spitregs.h>
46 #include <sys/cpuvar.h>
47 #include <sys/cpupart.h>
48 #include <sys/mem_config.h>
49 #include <sys/ddi_impldefs.h>
50 #include <sys/systm.h>
51 #include <sys/machsystm.h>
52 #include <sys/autoconf.h>
53 #include <sys/cmn_err.h>
54 #include <sys/sysmacros.h>
55 #include <sys/x_call.h>
56 #include <sys/promif.h>
57 #include <sys/prom_plat.h>
58 #include <sys/membar.h>
59 #include <vm/seg_kmem.h>
60 #include <sys/mem_cage.h>
61 #include <sys/stack.h>
62 #include <sys/archsystm.h>
63 #include <vm/hat_sfmmu.h>
64 #include <sys/pte.h>
65 #include <sys/mmu.h>
66 #include <sys/cpu_module.h>
67 #include <sys/obpdefs.h>
68 #include <sys/note.h>
69 
70 #include <sys/starfire.h>	/* plat_max_... decls */
71 #include <sys/cvc.h>
72 #include <sys/cpu_sgnblk_defs.h>
73 #include <sys/drmach.h>
74 #include <sys/dr_util.h>
75 #include <sys/pda.h>
76 
77 #include <sys/sysevent.h>
78 #include <sys/sysevent/dr.h>
79 #include <sys/sysevent/eventdefs.h>
80 
81 
82 extern void		bcopy32_il(uint64_t, uint64_t);
83 extern void		flush_ecache_il(
84 				uint64_t physaddr, int size, int linesz);
85 extern uint_t		ldphysio_il(uint64_t physaddr);
86 extern void		stphysio_il(uint64_t physaddr, uint_t value);
87 
88 extern uint64_t		mc_get_mem_alignment(void);
89 extern uint64_t		mc_get_asr_addr(pnode_t);
90 extern uint64_t		mc_get_idle_addr(pnode_t);
91 extern uint64_t		mc_get_alignment_mask(pnode_t);
92 extern int		mc_read_asr(pnode_t, uint_t *);
93 extern int		mc_write_asr(pnode_t, uint_t);
94 extern uint64_t		mc_asr_to_pa(uint_t);
95 extern uint_t		mc_pa_to_asr(uint_t, uint64_t);
96 
97 extern int		pc_madr_add(int, int, int, int);
98 
99 typedef struct {
100 	struct drmach_node	*node;
101 	void			*data;
102 } drmach_node_walk_args_t;
103 
104 typedef struct drmach_node {
105 	void		*here;
106 
107 	pnode_t		 (*get_dnode)(struct drmach_node *node);
108 	int		 (*walk)(struct drmach_node *node, void *data,
109 				int (*cb)(drmach_node_walk_args_t *args));
110 } drmach_node_t;
111 
112 typedef struct {
113 	int		 min_index;
114 	int		 max_index;
115 	int		 arr_sz;
116 	drmachid_t	*arr;
117 } drmach_array_t;
118 
119 typedef struct {
120 	void		*isa;
121 
122 	sbd_error_t	*(*release)(drmachid_t);
123 	sbd_error_t	*(*status)(drmachid_t, drmach_status_t *);
124 
125 	char		 name[MAXNAMELEN];
126 } drmach_common_t;
127 
128 typedef struct {
129 	drmach_common_t	 cm;
130 	int		 bnum;
131 	int		 assigned;
132 	int		 powered;
133 	int		 connect_cpuid;
134 	int		 cond;
135 	drmach_node_t	*tree;
136 	drmach_array_t	*devices;
137 } drmach_board_t;
138 
139 typedef struct {
140 	drmach_common_t	 cm;
141 	drmach_board_t	*bp;
142 	int		 unum;
143 	int		 busy;
144 	int		 powered;
145 	const char	*type;
146 	drmach_node_t	*node;
147 } drmach_device_t;
148 
149 typedef struct {
150 	int		 flags;
151 	drmach_device_t	*dp;
152 	sbd_error_t	*err;
153 	dev_info_t	*dip;
154 } drmach_config_args_t;
155 
156 typedef struct {
157 	uint64_t	 idle_addr;
158 	drmach_device_t	*mem;
159 } drmach_mc_idle_script_t;
160 
161 typedef struct {
162 	uint64_t	masr_addr;
163 	uint_t		masr;
164 	uint_t		_filler;
165 } drmach_rename_script_t;
166 
167 typedef struct {
168 	void		(*run)(void *arg);
169 	caddr_t		data;
170 	pda_handle_t	*ph;
171 	struct memlist	*c_ml;
172 	uint64_t	s_copybasepa;
173 	uint64_t	t_copybasepa;
174 	drmach_device_t	*restless_mc;	/* diagnostic output */
175 } drmach_copy_rename_program_t;
176 
177 typedef enum {
178 	DO_IDLE,
179 	DO_UNIDLE,
180 	DO_PAUSE,
181 	DO_UNPAUSE
182 } drmach_iopc_op_t;
183 
184 typedef struct {
185 	drmach_board_t	*obj;
186 	int		 ndevs;
187 	void		*a;
188 	sbd_error_t	*(*found)(void *a, const char *, int, drmachid_t);
189 	sbd_error_t	*err;
190 } drmach_board_cb_data_t;
191 
192 static caddr_t		 drmach_shutdown_va;
193 
194 static int		 drmach_initialized;
195 static drmach_array_t	*drmach_boards;
196 
197 static int		 drmach_cpu_delay = 100;
198 static int		 drmach_cpu_ntries = 50000;
199 
200 volatile uchar_t	*drmach_xt_mb;
201 
202 /*
203  * Do not change the drmach_shutdown_mbox structure without
204  * considering the drmach_shutdown_asm assembly language code.
205  */
206 struct drmach_shutdown_mbox {
207 	uint64_t	estack;
208 	uint64_t	flushaddr;
209 	int		size;
210 	int		linesize;
211 	uint64_t	physaddr;
212 };
213 struct drmach_shutdown_mbox	*drmach_shutdown_asm_mbox;
214 static sbd_error_t	*drmach_device_new(drmach_node_t *,
215 				drmach_board_t *, drmach_device_t **);
216 static sbd_error_t	*drmach_cpu_new(drmach_device_t *);
217 static sbd_error_t	*drmach_mem_new(drmach_device_t *);
218 static sbd_error_t	*drmach_io_new(drmach_device_t *);
219 
220 extern struct cpu	*SIGBCPU;
221 
222 #ifdef DEBUG
223 
224 #define	DRMACH_PR		if (drmach_debug) printf
225 int drmach_debug = 0;		 /* set to non-zero to enable debug messages */
226 #else
227 
228 #define	DRMACH_PR		_NOTE(CONSTANTCONDITION) if (0) printf
229 #endif /* DEBUG */
230 
231 #define	DRMACH_OBJ(id)		((drmach_common_t *)id)
232 
233 #define	DRMACH_IS_BOARD_ID(id)	\
234 	((id != 0) &&		\
235 	(DRMACH_OBJ(id)->isa == (void *)drmach_board_new))
236 
237 #define	DRMACH_IS_CPU_ID(id)	\
238 	((id != 0) &&		\
239 	(DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new))
240 
241 #define	DRMACH_IS_MEM_ID(id)	\
242 	((id != 0) &&		\
243 	(DRMACH_OBJ(id)->isa == (void *)drmach_mem_new))
244 
245 #define	DRMACH_IS_IO_ID(id)	\
246 	((id != 0) &&		\
247 	(DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
248 
249 #define	DRMACH_IS_DEVICE_ID(id)					\
250 	((id != 0) &&						\
251 	(DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new ||	\
252 	    DRMACH_OBJ(id)->isa == (void *)drmach_mem_new ||	\
253 	    DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
254 
255 #define	DRMACH_IS_ID(id)					\
256 	((id != 0) &&						\
257 	(DRMACH_OBJ(id)->isa == (void *)drmach_board_new ||	\
258 	    DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new ||	\
259 	    DRMACH_OBJ(id)->isa == (void *)drmach_mem_new ||	\
260 	    DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
261 
262 #define	DRMACH_CPUID2BNUM(cpuid) \
263 	((cpuid) / MAX_CPU_UNITS_PER_BOARD)
264 
265 #define	DRMACH_INTERNAL_ERROR() \
266 	drerr_new(1, ESTF_INTERNAL, drmach_ie_fmt, __LINE__)
267 static char		*drmach_ie_fmt = "drmach.c %d";
268 
269 static struct {
270 	const char	 *name;
271 	const char	 *type;
272 	sbd_error_t	 *(*new)(drmach_device_t *);
273 } name2type[] = {
274 	{ "SUNW,UltraSPARC",	DRMACH_DEVTYPE_CPU,  drmach_cpu_new },
275 	{ "mem-unit",		DRMACH_DEVTYPE_MEM,  drmach_mem_new },
276 	{ "pci",		DRMACH_DEVTYPE_PCI,  drmach_io_new  },
277 	{ "sbus",		DRMACH_DEVTYPE_SBUS, drmach_io_new  },
278 };
279 
280 /* node types to cleanup when a board is unconfigured */
281 #define	MISC_COUNTER_TIMER_DEVNAME	"counter-timer"
282 #define	MISC_PERF_COUNTER_DEVNAME	"perf-counter"
283 
284 /* utility */
285 #define	MBYTE	(1048576ull)
286 
287 /*
288  * This is necessary because the CPU support needs
289  * to call cvc_assign_iocpu.
290  */
291 #ifndef lint
292 char _depends_on[] = "drv/cvc";
293 #endif  /* lint */
294 
295 /*
296  * drmach autoconfiguration data structures and interfaces
297  */
298 
299 extern struct mod_ops mod_miscops;
300 
301 static struct modlmisc modlmisc = {
302 	&mod_miscops,
303 	"Sun Enterprise 10000 DR"
304 };
305 
306 static struct modlinkage modlinkage = {
307 	MODREV_1,
308 	(void *)&modlmisc,
309 	NULL
310 };
311 
312 static kmutex_t drmach_i_lock;
313 
314 int
315 _init(void)
316 {
317 	int err;
318 
319 	/* check that we have the correct version of obp */
320 	if (prom_test("SUNW,UE10000,add-brd") != 0) {
321 
322 		cmn_err(CE_WARN, "!OBP/SSP upgrade is required to enable "
323 		    "DR Functionality");
324 
325 		return (-1);
326 	}
327 
328 	mutex_init(&drmach_i_lock, NULL, MUTEX_DRIVER, NULL);
329 
330 	drmach_xt_mb = (uchar_t *)vmem_alloc(static_alloc_arena,
331 	    NCPU * sizeof (uchar_t), VM_SLEEP);
332 	drmach_shutdown_asm_mbox = (struct drmach_shutdown_mbox *)
333 	    vmem_alloc(static_alloc_arena, sizeof (struct drmach_shutdown_mbox),
334 	    VM_SLEEP);
335 
336 	if ((err = mod_install(&modlinkage)) != 0) {
337 		mutex_destroy(&drmach_i_lock);
338 		vmem_free(static_alloc_arena, (void *)drmach_xt_mb,
339 		    NCPU * sizeof (uchar_t));
340 		vmem_free(static_alloc_arena, (void *)drmach_shutdown_asm_mbox,
341 		    sizeof (struct drmach_shutdown_mbox));
342 	}
343 
344 	return (err);
345 }
346 
347 int
348 _fini(void)
349 {
350 	static int drmach_fini(void);
351 
352 	if (drmach_fini())
353 		return (DDI_FAILURE);
354 	else
355 		return (mod_remove(&modlinkage));
356 }
357 
358 int
359 _info(struct modinfo *modinfop)
360 {
361 	return (mod_info(&modlinkage, modinfop));
362 }
363 
364 static pnode_t
365 drmach_node_obp_get_dnode(drmach_node_t *np)
366 {
367 	return ((pnode_t)(uintptr_t)np->here);
368 }
369 
370 static int
371 drmach_node_obp_walk(drmach_node_t *np, void *data,
372 		int (*cb)(drmach_node_walk_args_t *args))
373 {
374 	pnode_t			nodeid;
375 	int			rv;
376 	drmach_node_walk_args_t	args;
377 
378 	/* initialized args structure for callback */
379 	args.node = np;
380 	args.data = data;
381 
382 	nodeid = prom_childnode(prom_rootnode());
383 
384 	/* save our new position with in the tree */
385 	np->here = (void *)(uintptr_t)nodeid;
386 
387 	rv = 0;
388 	while (nodeid != OBP_NONODE) {
389 		rv = (*cb)(&args);
390 		if (rv)
391 			break;
392 
393 		nodeid = prom_nextnode(nodeid);
394 
395 		/* save our new position with in the tree */
396 		np->here = (void *)(uintptr_t)nodeid;
397 	}
398 
399 	return (rv);
400 }
401 
402 static drmach_node_t *
403 drmach_node_new(void)
404 {
405 	drmach_node_t *np;
406 
407 	np = kmem_zalloc(sizeof (drmach_node_t), KM_SLEEP);
408 
409 	np->get_dnode = drmach_node_obp_get_dnode;
410 	np->walk = drmach_node_obp_walk;
411 
412 	return (np);
413 }
414 
415 static void
416 drmach_node_dispose(drmach_node_t *np)
417 {
418 	kmem_free(np, sizeof (*np));
419 }
420 
421 static dev_info_t *
422 drmach_node_get_dip(drmach_node_t *np)
423 {
424 	pnode_t nodeid;
425 
426 	nodeid = np->get_dnode(np);
427 	if (nodeid == OBP_NONODE)
428 		return (NULL);
429 	else {
430 		dev_info_t *dip;
431 
432 		/* The root node doesn't have to be held */
433 		dip = e_ddi_nodeid_to_dip(nodeid);
434 		if (dip) {
435 			/*
436 			 * Branch rooted at dip is already held, so release
437 			 * hold acquired in e_ddi_nodeid_to_dip()
438 			 */
439 			ddi_release_devi(dip);
440 			ASSERT(e_ddi_branch_held(dip));
441 		}
442 
443 		return (dip);
444 	}
445 	/*NOTREACHED*/
446 }
447 
448 static pnode_t
449 drmach_node_get_dnode(drmach_node_t *np)
450 {
451 	return (np->get_dnode(np));
452 }
453 
454 static int
455 drmach_node_walk(drmach_node_t *np, void *param,
456 		int (*cb)(drmach_node_walk_args_t *args))
457 {
458 	return (np->walk(np, param, cb));
459 }
460 
461 static int
462 drmach_node_get_prop(drmach_node_t *np, char *name, void *buf)
463 {
464 	pnode_t	nodeid;
465 	int	rv;
466 
467 	nodeid = np->get_dnode(np);
468 	if (nodeid == OBP_NONODE)
469 		rv = -1;
470 	else if (prom_getproplen(nodeid, (caddr_t)name) < 0)
471 		rv = -1;
472 	else {
473 		(void) prom_getprop(nodeid, (caddr_t)name, (caddr_t)buf);
474 		rv = 0;
475 	}
476 
477 	return (rv);
478 }
479 
480 static int
481 drmach_node_get_proplen(drmach_node_t *np, char *name, int *len)
482 {
483 	pnode_t	 nodeid;
484 	int	 rv;
485 
486 	nodeid = np->get_dnode(np);
487 	if (nodeid == OBP_NONODE)
488 		rv = -1;
489 	else {
490 		*len = prom_getproplen(nodeid, (caddr_t)name);
491 		rv = (*len < 0 ? -1 : 0);
492 	}
493 
494 	return (rv);
495 }
496 
497 static drmachid_t
498 drmach_node_dup(drmach_node_t *np)
499 {
500 	drmach_node_t *dup;
501 
502 	dup = drmach_node_new();
503 	dup->here = np->here;
504 
505 	return (dup);
506 }
507 
508 /*
509  * drmach_array provides convenient array construction, access,
510  * bounds checking and array destruction logic.
511  */
512 
513 static drmach_array_t *
514 drmach_array_new(int min_index, int max_index)
515 {
516 	drmach_array_t *arr;
517 
518 	arr = kmem_zalloc(sizeof (drmach_array_t), KM_SLEEP);
519 
520 	arr->arr_sz = (max_index - min_index + 1) * sizeof (void *);
521 	if (arr->arr_sz > 0) {
522 		arr->min_index = min_index;
523 		arr->max_index = max_index;
524 
525 		arr->arr = kmem_zalloc(arr->arr_sz, KM_SLEEP);
526 		return (arr);
527 	} else {
528 		kmem_free(arr, sizeof (*arr));
529 		return (0);
530 	}
531 }
532 
533 static int
534 drmach_array_set(drmach_array_t *arr, int idx, drmachid_t val)
535 {
536 	if (idx < arr->min_index || idx > arr->max_index)
537 		return (-1);
538 	else {
539 		arr->arr[idx - arr->min_index] = val;
540 		return (0);
541 	}
542 	/*NOTREACHED*/
543 }
544 
545 static int
546 drmach_array_get(drmach_array_t *arr, int idx, drmachid_t *val)
547 {
548 	if (idx < arr->min_index || idx > arr->max_index)
549 		return (-1);
550 	else {
551 		*val = arr->arr[idx - arr->min_index];
552 		return (0);
553 	}
554 	/*NOTREACHED*/
555 }
556 
557 static int
558 drmach_array_first(drmach_array_t *arr, int *idx, drmachid_t *val)
559 {
560 	int rv;
561 
562 	*idx = arr->min_index;
563 	while ((rv = drmach_array_get(arr, *idx, val)) == 0 && *val == NULL)
564 		*idx += 1;
565 
566 	return (rv);
567 }
568 
569 static int
570 drmach_array_next(drmach_array_t *arr, int *idx, drmachid_t *val)
571 {
572 	int rv;
573 
574 	*idx += 1;
575 	while ((rv = drmach_array_get(arr, *idx, val)) == 0 && *val == NULL)
576 		*idx += 1;
577 
578 	return (rv);
579 }
580 
581 static void
582 drmach_array_dispose(drmach_array_t *arr, void (*disposer)(drmachid_t))
583 {
584 	drmachid_t	val;
585 	int		idx;
586 	int		rv;
587 
588 	rv = drmach_array_first(arr, &idx, &val);
589 	while (rv == 0) {
590 		(*disposer)(val);
591 		rv = drmach_array_next(arr, &idx, &val);
592 	}
593 
594 	kmem_free(arr->arr, arr->arr_sz);
595 	kmem_free(arr, sizeof (*arr));
596 }
597 
598 /*ARGSUSED*/
599 static int
600 drmach_prom_select(pnode_t nodeid, void *arg, uint_t flags)
601 {
602 	int			rprop[64];
603 	pnode_t			saved;
604 	drmach_config_args_t	*ap = (drmach_config_args_t *)arg;
605 	drmach_device_t		*dp = ap->dp;
606 	sbd_error_t		*err;
607 
608 	saved = drmach_node_get_dnode(dp->node);
609 
610 	if (nodeid != saved)
611 		return (DDI_FAILURE);
612 
613 	if (saved == OBP_NONODE) {
614 		err = DRMACH_INTERNAL_ERROR();
615 		DRERR_SET_C(&ap->err, &err);
616 		return (DDI_FAILURE);
617 	}
618 
619 	if (prom_getprop(nodeid, OBP_REG, (caddr_t)rprop) <= 0) {
620 		return (DDI_FAILURE);
621 	}
622 
623 	return (DDI_SUCCESS);
624 }
625 
626 /*ARGSUSED*/
627 static void
628 drmach_branch_callback(dev_info_t *rdip, void *arg, uint_t flags)
629 {
630 	drmach_config_args_t	*ap = (drmach_config_args_t *)arg;
631 
632 	ASSERT(ap->dip == NULL);
633 
634 	ap->dip = rdip;
635 }
636 
637 sbd_error_t *
638 drmach_configure(drmachid_t id, int flags)
639 {
640 	drmach_device_t		*dp;
641 	sbd_error_t		*err;
642 	drmach_config_args_t	ca;
643 	devi_branch_t		b = {0};
644 	dev_info_t		*fdip = NULL;
645 
646 	if (!DRMACH_IS_DEVICE_ID(id))
647 		return (drerr_new(0, ESTF_INAPPROP, NULL));
648 	dp = id;
649 
650 	ca.dp = dp;
651 	ca.flags = flags;
652 	ca.err = NULL;		/* will be set if error detected */
653 	ca.dip = NULL;
654 
655 	b.arg = &ca;
656 	b.type = DEVI_BRANCH_PROM;
657 	b.create.prom_branch_select = drmach_prom_select;
658 	b.devi_branch_callback = drmach_branch_callback;
659 
660 	if (e_ddi_branch_create(ddi_root_node(), &b, &fdip,
661 	    DEVI_BRANCH_CHILD | DEVI_BRANCH_CONFIGURE) != 0) {
662 		char *path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
663 
664 		/*
665 		 * If non-NULL, fdip is returned held and must be released.
666 		 */
667 		if (fdip != NULL) {
668 			(void) ddi_pathname(fdip, path);
669 			ddi_release_devi(fdip);
670 		} else if (ca.dip != NULL) {
671 			/* safe to call ddi_pathname as dip already held */
672 			(void) ddi_pathname(ca.dip, path);
673 		} else {
674 			(void) strcpy(path, "<none>");
675 		}
676 
677 		err = drerr_new(1, ESTF_DRVFAIL, path);
678 		DRERR_SET_C(&ca.err, &err);
679 		kmem_free(path, MAXPATHLEN);
680 	}
681 
682 	return (ca.err);
683 }
684 
685 static sbd_error_t *
686 drmach_device_new(drmach_node_t *node,
687 	drmach_board_t *bp, drmach_device_t **dpp)
688 {
689 	int		 i;
690 	int		 rv;
691 	drmach_device_t	*dp;
692 	sbd_error_t	*err;
693 	char		 name[OBP_MAXDRVNAME];
694 
695 	rv = drmach_node_get_prop(node, OBP_NAME, name);
696 	if (rv) {
697 		/* every node is expected to have a name */
698 		err = drerr_new(1, ESTF_GETPROP,
699 			"PROM Node 0x%x: property %s",
700 			(uint_t)node->get_dnode(node), OBP_NAME);
701 
702 		return (err);
703 	}
704 
705 	/*
706 	 * The node currently being examined is not listed in the name2type[]
707 	 * array.  In this case, the node is no interest to drmach.  Both
708 	 * dp and err are initialized here to yield nothing (no device or
709 	 * error structure) for this case.
710 	 */
711 	for (i = 0; i < sizeof (name2type) / sizeof (name2type[0]); i++)
712 		if (strcmp(name2type[i].name, name) == 0)
713 			break;
714 
715 	if (i < sizeof (name2type) / sizeof (name2type[0])) {
716 		dp = kmem_zalloc(sizeof (drmach_device_t), KM_SLEEP);
717 
718 		dp->bp = bp;
719 		dp->unum = -1;
720 		dp->node = drmach_node_dup(node);
721 		dp->type = name2type[i].type;
722 
723 		err = (name2type[i].new)(dp);
724 		if (err) {
725 			drmach_node_dispose(node);
726 			kmem_free(dp, sizeof (*dp));
727 			dp = NULL;
728 		}
729 
730 		*dpp = dp;
731 		return (err);
732 	}
733 
734 	/*
735 	 * The node currently being examined is not listed in the name2type[]
736 	 * array.  In this case, the node is no interest to drmach.  Both
737 	 * dp and err are initialized here to yield nothing (no device or
738 	 * error structure) for this case.
739 	 */
740 	*dpp = NULL;
741 	return (NULL);
742 }
743 
744 static void
745 drmach_device_dispose(drmachid_t id)
746 {
747 	drmach_device_t *self = id;
748 
749 	if (self->node)
750 		drmach_node_dispose(self->node);
751 
752 	kmem_free(self, sizeof (*self));
753 }
754 
755 static sbd_error_t *
756 drmach_device_get_prop(drmach_device_t *dp, char *name, void *buf)
757 {
758 	sbd_error_t	*err = NULL;
759 	int		 rv;
760 
761 	rv = drmach_node_get_prop(dp->node, name, buf);
762 	if (rv) {
763 		err = drerr_new(1, ESTF_GETPROP,
764 			"%s::%s: property %s",
765 			dp->bp->cm.name, dp->cm.name, name);
766 	}
767 
768 	return (err);
769 }
770 
771 static sbd_error_t *
772 drmach_device_get_proplen(drmach_device_t *dp, char *name, int *len)
773 {
774 	sbd_error_t	*err = NULL;
775 	int		 rv;
776 
777 	rv = drmach_node_get_proplen(dp->node, name, len);
778 	if (rv) {
779 		err = drerr_new(1, ESTF_GETPROPLEN,
780 			"%s::%s: property %s",
781 			dp->bp->cm.name, dp->cm.name, name);
782 	}
783 
784 	return (err);
785 }
786 
787 static drmach_board_t *
788 drmach_board_new(int bnum)
789 {
790 	static sbd_error_t *drmach_board_release(drmachid_t);
791 	static sbd_error_t *drmach_board_status(drmachid_t, drmach_status_t *);
792 
793 	drmach_board_t	*bp;
794 
795 	bp = kmem_zalloc(sizeof (drmach_board_t), KM_SLEEP);
796 
797 	bp->cm.isa = (void *)drmach_board_new;
798 	bp->cm.release = drmach_board_release;
799 	bp->cm.status = drmach_board_status;
800 
801 	(void) drmach_board_name(bnum, bp->cm.name, sizeof (bp->cm.name));
802 
803 	bp->bnum = bnum;
804 	bp->devices = NULL;
805 	bp->connect_cpuid = -1;
806 	bp->tree = drmach_node_new();
807 	bp->assigned = !drmach_initialized;
808 	bp->powered = !drmach_initialized;
809 
810 	(void) drmach_array_set(drmach_boards, bnum, bp);
811 	return (bp);
812 }
813 
814 static void
815 drmach_board_dispose(drmachid_t id)
816 {
817 	drmach_board_t *bp;
818 
819 	ASSERT(DRMACH_IS_BOARD_ID(id));
820 	bp = id;
821 
822 	if (bp->tree)
823 		drmach_node_dispose(bp->tree);
824 
825 	if (bp->devices)
826 		drmach_array_dispose(bp->devices, drmach_device_dispose);
827 
828 	kmem_free(bp, sizeof (*bp));
829 }
830 
831 static sbd_error_t *
832 drmach_board_status(drmachid_t id, drmach_status_t *stat)
833 {
834 	sbd_error_t	*err = NULL;
835 	drmach_board_t	*bp;
836 
837 	if (!DRMACH_IS_BOARD_ID(id))
838 		return (drerr_new(0, ESTF_INAPPROP, NULL));
839 	bp = id;
840 
841 	stat->assigned = bp->assigned;
842 	stat->powered = bp->powered;
843 	stat->busy = 0;			/* assume not busy */
844 	stat->configured = 0;		/* assume not configured */
845 	stat->empty = 0;
846 	stat->cond = bp->cond = SBD_COND_OK;
847 	(void) strncpy(stat->type, "System Brd", sizeof (stat->type));
848 	stat->info[0] = '\0';
849 
850 	if (bp->devices) {
851 		int		 rv;
852 		int		 d_idx;
853 		drmachid_t	 d_id;
854 
855 		rv = drmach_array_first(bp->devices, &d_idx, &d_id);
856 		while (rv == 0) {
857 			drmach_status_t	d_stat;
858 
859 			err = drmach_status(d_id, &d_stat);
860 			if (err)
861 				break;
862 
863 			stat->busy |= d_stat.busy;
864 			stat->configured |= d_stat.configured;
865 
866 			rv = drmach_array_next(bp->devices, &d_idx, &d_id);
867 		}
868 	}
869 
870 	return (err);
871 }
872 
873 /* a simple routine to reduce redundancy of this common logic */
874 static pda_handle_t
875 drmach_pda_open(void)
876 {
877 	pda_handle_t ph;
878 
879 	ph = pda_open();
880 	if (ph == NULL) {
881 		/* catch in debug kernels */
882 		ASSERT(0);
883 		cmn_err(CE_WARN, "pda_open failed");
884 	}
885 
886 	return (ph);
887 }
888 
889 #ifdef DEBUG
890 int drmach_init_break = 0;
891 #endif
892 
893 static int
894 hold_rele_branch(dev_info_t *rdip, void *arg)
895 {
896 	int	i;
897 	int	*holdp = (int *)arg;
898 	char	*name = ddi_node_name(rdip);
899 
900 	/*
901 	 * For Starfire, we must be children of the root devinfo node
902 	 */
903 	ASSERT(ddi_get_parent(rdip) == ddi_root_node());
904 
905 	for (i = 0; i < sizeof (name2type) / sizeof (name2type[0]); i++)
906 		if (strcmp(name2type[i].name, name) == 0)
907 			break;
908 
909 	if (i == sizeof (name2type) / sizeof (name2type[0])) {
910 		/* Not of interest to us */
911 		return (DDI_WALK_PRUNECHILD);
912 	}
913 
914 	if (*holdp) {
915 		ASSERT(!e_ddi_branch_held(rdip));
916 		e_ddi_branch_hold(rdip);
917 	} else {
918 		ASSERT(e_ddi_branch_held(rdip));
919 		e_ddi_branch_rele(rdip);
920 	}
921 
922 	return (DDI_WALK_PRUNECHILD);
923 }
924 
925 static int
926 drmach_init(void)
927 {
928 	pnode_t		nodeid;
929 	dev_info_t	*rdip;
930 	int		hold, circ;
931 
932 #ifdef DEBUG
933 	if (drmach_init_break)
934 		debug_enter("drmach_init: drmach_init_break set\n");
935 #endif
936 	mutex_enter(&drmach_i_lock);
937 	if (drmach_initialized) {
938 		mutex_exit(&drmach_i_lock);
939 		return (0);
940 	}
941 
942 	drmach_boards = drmach_array_new(0, MAX_BOARDS - 1);
943 
944 	nodeid = prom_childnode(prom_rootnode());
945 	do {
946 		int		 bnum;
947 		drmachid_t	 id;
948 
949 		bnum = -1;
950 		(void) prom_getprop(nodeid, OBP_BOARDNUM, (caddr_t)&bnum);
951 		if (bnum == -1)
952 			continue;
953 
954 		if (drmach_array_get(drmach_boards, bnum, &id) == -1) {
955 			cmn_err(CE_WARN, "OBP node 0x%x has"
956 				" invalid property value, %s=%d",
957 				nodeid, OBP_BOARDNUM, bnum);
958 
959 			/* clean up */
960 			drmach_array_dispose(
961 				drmach_boards, drmach_board_dispose);
962 
963 			mutex_exit(&drmach_i_lock);
964 			return (-1);
965 		} else if (id == NULL)
966 			(void) drmach_board_new(bnum);
967 	} while ((nodeid = prom_nextnode(nodeid)) != OBP_NONODE);
968 
969 	drmach_shutdown_va = vmem_alloc(heap_arena, PAGESIZE, VM_SLEEP);
970 
971 	/*
972 	 * Walk immediate children of devinfo root node and hold
973 	 * all devinfo branches of interest.
974 	 */
975 	hold = 1;
976 	rdip = ddi_root_node();
977 
978 	ndi_devi_enter(rdip, &circ);
979 	ddi_walk_devs(ddi_get_child(rdip), hold_rele_branch, &hold);
980 	ndi_devi_exit(rdip, circ);
981 
982 	drmach_initialized = 1;
983 
984 	mutex_exit(&drmach_i_lock);
985 
986 	return (0);
987 }
988 
989 static int
990 drmach_fini(void)
991 {
992 	dev_info_t	*rdip;
993 	int		hold, circ;
994 
995 	if (drmach_initialized) {
996 		int		busy = 0;
997 		int		rv;
998 		int		idx;
999 		drmachid_t	id;
1000 
1001 		ASSERT(drmach_boards != NULL);
1002 
1003 		rv = drmach_array_first(drmach_boards, &idx, &id);
1004 		while (rv == 0) {
1005 			sbd_error_t	*err;
1006 			drmach_status_t stat;
1007 
1008 			err = drmach_board_status(id, &stat);
1009 			if (err) {
1010 				/* catch in debug kernels */
1011 				ASSERT(0);
1012 				sbd_err_clear(&err);
1013 				busy = 1;
1014 			} else
1015 				busy |= stat.busy;
1016 
1017 			rv = drmach_array_next(drmach_boards, &idx, &id);
1018 		}
1019 
1020 		if (busy)
1021 			return (-1);
1022 
1023 		drmach_array_dispose(drmach_boards, drmach_board_dispose);
1024 		drmach_boards = NULL;
1025 
1026 		vmem_free(heap_arena, drmach_shutdown_va, PAGESIZE);
1027 
1028 		/*
1029 		 * Walk immediate children of the root devinfo node
1030 		 * releasing holds acquired on branches in drmach_init()
1031 		 */
1032 		hold = 0;
1033 		rdip = ddi_root_node();
1034 
1035 		ndi_devi_enter(rdip, &circ);
1036 		ddi_walk_devs(ddi_get_child(rdip), hold_rele_branch, &hold);
1037 		ndi_devi_exit(rdip, circ);
1038 
1039 		mutex_destroy(&drmach_i_lock);
1040 
1041 		drmach_initialized = 0;
1042 	}
1043 	if (drmach_xt_mb != NULL) {
1044 		vmem_free(static_alloc_arena, (void *)drmach_xt_mb,
1045 		    NCPU * sizeof (uchar_t));
1046 	}
1047 	if (drmach_shutdown_asm_mbox != NULL) {
1048 		vmem_free(static_alloc_arena, (void *)drmach_shutdown_asm_mbox,
1049 		    sizeof (struct drmach_shutdown_mbox));
1050 	}
1051 	return (0);
1052 }
1053 
1054 static sbd_error_t *
1055 drmach_get_mc_asr_addr(drmachid_t id, uint64_t *pa)
1056 {
1057 	drmach_device_t	*dp;
1058 	pnode_t		nodeid;
1059 	uint64_t	addr;
1060 
1061 	if (!DRMACH_IS_MEM_ID(id))
1062 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1063 	dp = id;
1064 
1065 	nodeid = drmach_node_get_dnode(dp->node);
1066 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1067 		return (DRMACH_INTERNAL_ERROR());
1068 
1069 	addr = mc_get_asr_addr(nodeid);
1070 	if (addr == (uint64_t)-1)
1071 		return (DRMACH_INTERNAL_ERROR());
1072 
1073 	*pa = addr;
1074 	return (NULL);
1075 }
1076 
1077 static sbd_error_t *
1078 drmach_get_mc_idle_addr(drmachid_t id, uint64_t *pa)
1079 {
1080 	drmach_device_t	*dp;
1081 	pnode_t		nodeid;
1082 	uint64_t	addr;
1083 
1084 	if (!DRMACH_IS_MEM_ID(id))
1085 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1086 	dp = id;
1087 
1088 	nodeid = drmach_node_get_dnode(dp->node);
1089 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1090 		return (DRMACH_INTERNAL_ERROR());
1091 
1092 	addr = mc_get_idle_addr(nodeid);
1093 	if (addr == (uint64_t)-1)
1094 		return (DRMACH_INTERNAL_ERROR());
1095 
1096 	*pa = addr;
1097 	return (NULL);
1098 }
1099 
1100 static sbd_error_t *
1101 drmach_read_mc_asr(drmachid_t id, uint_t *mcregp)
1102 {
1103 	drmach_device_t	*dp;
1104 	pnode_t		 nodeid;
1105 	sbd_error_t	*err;
1106 
1107 	if (!DRMACH_IS_MEM_ID(id))
1108 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1109 	dp = id;
1110 
1111 	nodeid = drmach_node_get_dnode(dp->node);
1112 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1113 		err = DRMACH_INTERNAL_ERROR();
1114 	else if (mc_read_asr(nodeid, mcregp) == -1)
1115 		err = DRMACH_INTERNAL_ERROR();
1116 	else
1117 		err = NULL;
1118 
1119 	return (err);
1120 }
1121 
1122 static sbd_error_t *
1123 drmach_write_mc_asr(drmachid_t id, uint_t mcreg)
1124 {
1125 	drmach_device_t	*dp;
1126 	pnode_t		 nodeid;
1127 	sbd_error_t	*err;
1128 
1129 	if (!DRMACH_IS_MEM_ID(id))
1130 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1131 	dp = id;
1132 
1133 	nodeid = drmach_node_get_dnode(dp->node);
1134 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1135 		err = DRMACH_INTERNAL_ERROR();
1136 	else if (mc_write_asr(nodeid, mcreg) == -1)
1137 		err = DRMACH_INTERNAL_ERROR();
1138 	else
1139 		err = NULL;
1140 
1141 	return (err);
1142 }
1143 
1144 static sbd_error_t *
1145 drmach_prep_rename_script(drmach_device_t *s_mem, drmach_device_t *t_mem,
1146 	uint64_t t_slice_offset, caddr_t buf, int buflen)
1147 {
1148 	int			i, b, m;
1149 	drmach_mc_idle_script_t	*isp;
1150 	drmach_rename_script_t	*rsp;
1151 	int			s_bd, t_bd;
1152 	uint_t			s_masr, t_masr;
1153 	uint64_t		s_new_basepa, t_new_basepa;
1154 	int			b_idx, rv;
1155 	sbd_error_t		*err;
1156 	drmachid_t		 b_id;
1157 	drmach_board_t		*brd;
1158 
1159 #ifdef DEBUG
1160 	/*
1161 	 * Starfire CPU/MEM/IO boards have only one MC per board.
1162 	 * This function has been coded with that fact in mind.
1163 	 */
1164 	ASSERT(MAX_MEM_UNITS_PER_BOARD == 1);
1165 
1166 	/*
1167 	 * calculate the maximum space that could be consumed,
1168 	 * then verify the available buffer space is adequate.
1169 	 */
1170 	m  = sizeof (drmach_mc_idle_script_t *) * 2; /* two MCs */
1171 	b  = sizeof (drmach_rename_script_t *) * 3 * MAX_CPU_UNITS_PER_BOARD;
1172 	b += sizeof (drmach_rename_script_t *) * 3 * MAX_IO_UNITS_PER_BOARD;
1173 	b *= MAX_BOARDS;
1174 	b += sizeof (drmach_rename_script_t *) * 3;
1175 	b += sizeof (drmach_rename_script_t *) * 1;
1176 	ASSERT(m + b < buflen);
1177 #endif
1178 
1179 	/*
1180 	 * construct an array of MC idle register addresses of
1181 	 * both MCs.  The array is zero terminated -- as expected
1182 	 * by drmach_copy_rename_prog__relocatable().
1183 	 */
1184 	isp = (drmach_mc_idle_script_t *)buf;
1185 
1186 	/* source mc */
1187 	err = drmach_get_mc_idle_addr(s_mem, &isp->idle_addr);
1188 	if (err)
1189 		return (err);
1190 	isp->mem = s_mem;
1191 	isp += 1;
1192 
1193 	/* target mc */
1194 	err = drmach_get_mc_idle_addr(t_mem, &isp->idle_addr);
1195 	if (err)
1196 		return (err);
1197 	isp->mem = t_mem;
1198 	isp += 1;
1199 
1200 	/* terminator */
1201 	isp->idle_addr = 0;
1202 	isp->mem = NULL;
1203 	isp += 1;
1204 
1205 	/* fetch source mc asr register value */
1206 	err = drmach_read_mc_asr(s_mem, &s_masr);
1207 	if (err)
1208 		return (err);
1209 	else if (s_masr & STARFIRE_MC_INTERLEAVE_MASK) {
1210 		return (drerr_new(1, ESTF_INTERBOARD, "%s::%s",
1211 				s_mem->bp->cm.name, s_mem->cm.name));
1212 	}
1213 
1214 	/* fetch target mc asr register value */
1215 	err = drmach_read_mc_asr(t_mem, &t_masr);
1216 	if (err)
1217 		return (err);
1218 	else if (t_masr & STARFIRE_MC_INTERLEAVE_MASK) {
1219 		return (drerr_new(1, ESTF_INTERBOARD, "%s::%s",
1220 				t_mem->bp->cm.name, t_mem->cm.name));
1221 	}
1222 
1223 	/* get new source base pa from target's masr */
1224 	s_new_basepa = mc_asr_to_pa(t_masr);
1225 
1226 	/*
1227 	 * remove any existing slice offset to realign
1228 	 * memory with board's slice boundary
1229 	 */
1230 	s_new_basepa &= ~ (mc_get_mem_alignment() - 1);
1231 
1232 	/* get new target base pa from source's masr */
1233 	t_new_basepa  = mc_asr_to_pa(s_masr);
1234 
1235 	/* remove any existing slice offset, then apply new offset */
1236 	t_new_basepa &= ~ (mc_get_mem_alignment() - 1);
1237 	t_new_basepa += t_slice_offset;
1238 
1239 	/* encode new base pa into s_masr.  turn off mem present bit */
1240 	s_masr  = mc_pa_to_asr(s_masr, s_new_basepa);
1241 	s_masr &= ~STARFIRE_MC_MEM_PRESENT_MASK;
1242 
1243 	/* encode new base pa into t_masr.  turn on mem present bit */
1244 	t_masr  = mc_pa_to_asr(t_masr, t_new_basepa);
1245 	t_masr |= STARFIRE_MC_MEM_PRESENT_MASK;
1246 
1247 	/*
1248 	 * Step 0:	Mark source memory as not present.
1249 	 */
1250 	m = 0;
1251 	rsp = (drmach_rename_script_t *)isp;
1252 	err = drmach_get_mc_asr_addr(s_mem, &rsp[m].masr_addr);
1253 	if (err)
1254 		return (err);
1255 	rsp[m].masr = s_masr;
1256 	m++;
1257 
1258 	/*
1259 	 * Step 1:	Write source base address to target MC
1260 	 *		with present bit off.
1261 	 */
1262 	err = drmach_get_mc_asr_addr(t_mem, &rsp[m].masr_addr);
1263 	if (err)
1264 		return (err);
1265 	rsp[m].masr = t_masr & ~STARFIRE_MC_MEM_PRESENT_MASK;
1266 	m++;
1267 
1268 	/*
1269 	 * Step 2:	Now rewrite target reg with present bit on.
1270 	 */
1271 	rsp[m].masr_addr = rsp[m-1].masr_addr;
1272 	rsp[m].masr = t_masr;
1273 	m++;
1274 
1275 	s_bd = s_mem->bp->bnum;
1276 	t_bd = t_mem->bp->bnum;
1277 
1278 	DRMACH_PR("preparing script for CPU and IO units:\n");
1279 
1280 	rv = drmach_array_first(drmach_boards, &b_idx, &b_id);
1281 	if (rv) {
1282 		/* catch this in debug kernels */
1283 		ASSERT(0);
1284 		return (DRMACH_INTERNAL_ERROR());
1285 	}
1286 
1287 	do {
1288 		int			 d_idx;
1289 		drmachid_t		 d_id;
1290 		drmach_device_t		*device;
1291 
1292 		ASSERT(DRMACH_IS_BOARD_ID(b_id));
1293 		brd = b_id;
1294 		b = brd->bnum;
1295 
1296 		/*
1297 		 * Step 3:	Update PC MADR tables for CPUs.
1298 		 */
1299 		if (brd->devices == NULL) {
1300 			/* devices not initialized */
1301 			continue;
1302 		}
1303 
1304 		rv = drmach_array_first(brd->devices, &d_idx, &d_id);
1305 		if (rv) {
1306 			/* must mean no devices on this board */
1307 			break;
1308 		}
1309 
1310 		DRMACH_PR("\t%s\n", brd->cm.name);
1311 
1312 		do {
1313 			ASSERT(DRMACH_IS_DEVICE_ID(d_id));
1314 
1315 			if (!DRMACH_IS_CPU_ID(d_id))
1316 				continue;
1317 
1318 			device = d_id;
1319 			i = device->unum;
1320 
1321 			DRMACH_PR("\t\t%s\n", device->cm.name);
1322 
1323 			/*
1324 			 * Disabled detaching mem node.
1325 			 */
1326 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, s_bd, i);
1327 			rsp[m].masr = s_masr;
1328 			m++;
1329 			/*
1330 			 * Always write masr with present bit
1331 			 * off and then again with it on.
1332 			 */
1333 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, t_bd, i);
1334 			rsp[m].masr = t_masr & ~STARFIRE_MC_MEM_PRESENT_MASK;
1335 			m++;
1336 			rsp[m].masr_addr = rsp[m-1].masr_addr;
1337 			rsp[m].masr = t_masr;
1338 			m++;
1339 
1340 		} while (drmach_array_next(brd->devices, &d_idx, &d_id) == 0);
1341 
1342 		/*
1343 		 * Step 4:	Update PC MADR tables for IOs.
1344 		 */
1345 		rv = drmach_array_first(brd->devices, &d_idx, &d_id);
1346 		/* this worked for previous loop, must work here too */
1347 		ASSERT(rv == 0);
1348 
1349 		do {
1350 			ASSERT(DRMACH_IS_DEVICE_ID(d_id));
1351 
1352 			if (!DRMACH_IS_IO_ID(d_id))
1353 				continue;
1354 
1355 			device = d_id;
1356 			i = device->unum;
1357 
1358 			DRMACH_PR("\t\t%s\n", device->cm.name);
1359 
1360 			/*
1361 			 * Disabled detaching mem node.
1362 			 */
1363 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, s_bd, i+4);
1364 			rsp[m].masr = s_masr;
1365 			m++;
1366 			/*
1367 			 * Always write masr with present bit
1368 			 * off and then again with it on.
1369 			 */
1370 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, t_bd, i+4);
1371 			rsp[m].masr = t_masr & ~STARFIRE_MC_MEM_PRESENT_MASK;
1372 			m++;
1373 			rsp[m].masr_addr = rsp[m-1].masr_addr;
1374 			rsp[m].masr = t_masr;
1375 			m++;
1376 
1377 		} while (drmach_array_next(brd->devices, &d_idx, &d_id) == 0);
1378 	} while (drmach_array_next(drmach_boards, &b_idx, &b_id) == 0);
1379 
1380 	/*
1381 	 * Zero masr_addr value indicates the END.
1382 	 */
1383 	rsp[m].masr_addr = 0ull;
1384 	rsp[m].masr = 0;
1385 	DRMACH_PR("number of steps in rename script = %d\n", m);
1386 	m++;
1387 
1388 	/* paranoia */
1389 	ASSERT((caddr_t)&rsp[m] <= buf + buflen);
1390 
1391 #ifdef DEBUG
1392 	{
1393 		int	j;
1394 
1395 		DRMACH_PR("mc idle register address list:");
1396 		isp = (drmach_mc_idle_script_t *)buf;
1397 		DRMACH_PR("source mc idle addr 0x%lx, mem id %p",
1398 			isp[0].idle_addr, (void *)isp[0].mem);
1399 		DRMACH_PR("target mc idle addr 0x%lx, mem id %p",
1400 			isp[1].idle_addr, (void *)isp[1].mem);
1401 		ASSERT(isp[2].idle_addr == 0);
1402 
1403 		DRMACH_PR("copy-rename script:");
1404 		for (j = 0; j < m; j++) {
1405 			DRMACH_PR("0x%lx = 0x%08x",
1406 				rsp[j].masr_addr, rsp[j].masr);
1407 		}
1408 
1409 		DELAY(1000000);
1410 	}
1411 #endif
1412 
1413 	/* return number of bytes consumed */
1414 	b = (caddr_t)&rsp[m] - buf;
1415 	DRMACH_PR("total number of bytes consumed is %d\n", b);
1416 	ASSERT(b <= buflen);
1417 
1418 #ifdef lint
1419 	buflen = buflen;
1420 #endif
1421 
1422 	return (NULL);
1423 }
1424 
1425 /*
1426  * The routine performs the necessary memory COPY and MC adr SWITCH.
1427  * Both operations MUST be at the same "level" so that the stack is
1428  * maintained correctly between the copy and switch.  The switch
1429  * portion implements a caching mechanism to guarantee the code text
1430  * is cached prior to execution.  This is to guard against possible
1431  * memory access while the MC adr's are being modified.
1432  *
1433  * IMPORTANT: The _drmach_copy_rename_end() function must immediately
1434  * follow drmach_copy_rename_prog__relocatable() so that the correct
1435  * "length" of the drmach_copy_rename_prog__relocatable can be
1436  * calculated.  This routine MUST be a LEAF function, i.e. it can
1437  * make NO function calls, primarily for two reasons:
1438  *
1439  *	1. We must keep the stack consistent across the "switch".
1440  *	2. Function calls are compiled to relative offsets, and
1441  *	   we execute this function we'll be executing it from
1442  *	   a copied version in a different area of memory, thus
1443  *	   the relative offsets will be bogus.
1444  *
1445  * Moreover, it must have the "__relocatable" suffix to inform DTrace
1446  * providers (and anything else, for that matter) that this
1447  * function's text is manually relocated elsewhere before it is
1448  * executed.  That is, it cannot be safely instrumented with any
1449  * methodology that is PC-relative.
1450  */
1451 static void
1452 drmach_copy_rename_prog__relocatable(drmach_copy_rename_program_t *prog)
1453 {
1454 	extern void drmach_exec_script_il(drmach_rename_script_t *rsp);
1455 
1456 	drmach_mc_idle_script_t		*isp;
1457 	struct memlist			*ml;
1458 	int				csize;
1459 	int				lnsize;
1460 	uint64_t			caddr;
1461 
1462 	isp = (drmach_mc_idle_script_t *)prog->data;
1463 
1464 	caddr = ecache_flushaddr;
1465 	csize = (cpunodes[CPU->cpu_id].ecache_size << 1);
1466 	lnsize = cpunodes[CPU->cpu_id].ecache_linesize;
1467 
1468 	/*
1469 	 * DO COPY.
1470 	 */
1471 	for (ml = prog->c_ml; ml; ml = ml->next) {
1472 		uint64_t	s_pa, t_pa;
1473 		uint64_t	nbytes;
1474 
1475 		s_pa = prog->s_copybasepa + ml->address;
1476 		t_pa = prog->t_copybasepa + ml->address;
1477 		nbytes = ml->size;
1478 
1479 		while (nbytes != 0ull) {
1480 			/*
1481 			 * This copy does NOT use an ASI
1482 			 * that avoids the Ecache, therefore
1483 			 * the dst_pa addresses may remain
1484 			 * in our Ecache after the dst_pa
1485 			 * has been removed from the system.
1486 			 * A subsequent write-back to memory
1487 			 * will cause an ARB-stop because the
1488 			 * physical address no longer exists
1489 			 * in the system. Therefore we must
1490 			 * flush out local Ecache after we
1491 			 * finish the copy.
1492 			 */
1493 
1494 			/* copy 32 bytes at src_pa to dst_pa */
1495 			bcopy32_il(s_pa, t_pa);
1496 
1497 			/* increment by 32 bytes */
1498 			s_pa += (4 * sizeof (uint64_t));
1499 			t_pa += (4 * sizeof (uint64_t));
1500 
1501 			/* decrement by 32 bytes */
1502 			nbytes -= (4 * sizeof (uint64_t));
1503 		}
1504 	}
1505 
1506 	/*
1507 	 * Since bcopy32_il() does NOT use an ASI to bypass
1508 	 * the Ecache, we need to flush our Ecache after
1509 	 * the copy is complete.
1510 	 */
1511 	flush_ecache_il(caddr, csize, lnsize);		/* inline version */
1512 
1513 	/*
1514 	 * Wait for MCs to go idle.
1515 	 */
1516 	do {
1517 		register int	t = 10;
1518 		register uint_t	v;
1519 
1520 		/* loop t cycles waiting for each mc to indicate it's idle */
1521 		do {
1522 			v = ldphysio_il(isp->idle_addr)
1523 				& STARFIRE_MC_IDLE_MASK;
1524 
1525 		} while (v != STARFIRE_MC_IDLE_MASK && t-- > 0);
1526 
1527 		/* bailout if timedout */
1528 		if (t <= 0) {
1529 			prog->restless_mc = isp->mem;
1530 			return;
1531 		}
1532 
1533 		isp += 1;
1534 
1535 		/* stop if terminating zero has been reached */
1536 	} while (isp->idle_addr != 0);
1537 
1538 	/* advance passed terminating zero */
1539 	isp += 1;
1540 
1541 	/*
1542 	 * The following inline assembly routine caches
1543 	 * the rename script and then caches the code that
1544 	 * will do the rename.  This is necessary
1545 	 * so that we don't have any memory references during
1546 	 * the reprogramming.  We accomplish this by first
1547 	 * jumping through the code to guarantee it's cached
1548 	 * before we actually execute it.
1549 	 */
1550 	drmach_exec_script_il((drmach_rename_script_t *)isp);
1551 }
1552 
1553 static void
1554 drmach_copy_rename_end(void)
1555 {
1556 	/*
1557 	 * IMPORTANT:	This function's location MUST be located immediately
1558 	 *		following drmach_copy_rename_prog__relocatable to
1559 	 *		accurately estimate its size.  Note that this assumes
1560 	 *		the compiler keeps these functions in the order in
1561 	 *		which they appear :-o
1562 	 */
1563 }
1564 
1565 sbd_error_t *
1566 drmach_copy_rename_init(drmachid_t t_id, uint64_t t_slice_offset,
1567 	drmachid_t s_id, struct memlist *c_ml, drmachid_t *pgm_id)
1568 {
1569 	drmach_device_t	*s_mem;
1570 	drmach_device_t	*t_mem;
1571 	struct memlist	*x_ml;
1572 	uint64_t	off_mask, s_copybasepa, t_copybasepa, t_basepa;
1573 	int		len;
1574 	caddr_t		bp, wp;
1575 	pda_handle_t	ph;
1576 	sbd_error_t	*err;
1577 	drmach_copy_rename_program_t *prog;
1578 
1579 	if (!DRMACH_IS_MEM_ID(s_id))
1580 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1581 	if (!DRMACH_IS_MEM_ID(t_id))
1582 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1583 	s_mem = s_id;
1584 	t_mem = t_id;
1585 
1586 	/* get starting physical address of target memory */
1587 	err = drmach_mem_get_base_physaddr(t_id, &t_basepa);
1588 	if (err)
1589 		return (err);
1590 
1591 	/* calculate slice offset mask from slice size */
1592 	off_mask = mc_get_mem_alignment() - 1;
1593 
1594 	/* calculate source and target base pa */
1595 	s_copybasepa = c_ml->address;
1596 	t_copybasepa = t_basepa + ((c_ml->address & off_mask) - t_slice_offset);
1597 
1598 	/* paranoia */
1599 	ASSERT((c_ml->address & off_mask) >= t_slice_offset);
1600 
1601 	/* adjust copy memlist addresses to be relative to copy base pa */
1602 	x_ml = c_ml;
1603 	while (x_ml != NULL) {
1604 		x_ml->address -= s_copybasepa;
1605 		x_ml = x_ml->next;
1606 	}
1607 
1608 #ifdef DEBUG
1609 	{
1610 	uint64_t s_basepa, s_size, t_size;
1611 
1612 	x_ml = c_ml;
1613 	while (x_ml->next != NULL)
1614 		x_ml = x_ml->next;
1615 
1616 	DRMACH_PR("source copy span: base pa 0x%lx, end pa 0x%lx\n",
1617 		s_copybasepa,
1618 		s_copybasepa + x_ml->address + x_ml->size);
1619 
1620 	DRMACH_PR("target copy span: base pa 0x%lx, end pa 0x%lx\n",
1621 		t_copybasepa,
1622 		t_copybasepa + x_ml->address + x_ml->size);
1623 
1624 	DRMACH_PR("copy memlist (relative to copy base pa):\n");
1625 	MEMLIST_DUMP(c_ml);
1626 
1627 	err = drmach_mem_get_base_physaddr(s_id, &s_basepa);
1628 	ASSERT(err == NULL);
1629 
1630 	err = drmach_mem_get_size(s_id, &s_size);
1631 	ASSERT(err == NULL);
1632 
1633 	err = drmach_mem_get_size(t_id, &t_size);
1634 	ASSERT(err == NULL);
1635 
1636 	DRMACH_PR("current source base pa 0x%lx, size 0x%lx\n",
1637 		s_basepa, s_size);
1638 	DRMACH_PR("current target base pa 0x%lx, size 0x%lx\n",
1639 		t_basepa, t_size);
1640 
1641 	ASSERT(s_copybasepa + x_ml->address + x_ml->size <= s_basepa + s_size);
1642 	ASSERT(t_copybasepa + x_ml->address + x_ml->size <= t_basepa + t_size);
1643 	}
1644 #endif
1645 
1646 	ph = drmach_pda_open();
1647 	if (ph == NULL)
1648 		return (DRMACH_INTERNAL_ERROR());
1649 
1650 	/*
1651 	 * bp will be page aligned, since we're calling
1652 	 * kmem_zalloc() with an exact multiple of PAGESIZE.
1653 	 */
1654 	wp = bp = kmem_zalloc(PAGESIZE, KM_SLEEP);
1655 
1656 	/* allocate space for copy rename struct */
1657 	len = sizeof (drmach_copy_rename_program_t);
1658 	DRMACH_PR("prog = 0x%p, header len %d\n", (void *)wp, len);
1659 	prog = (drmach_copy_rename_program_t *)wp;
1660 	wp += (len + ecache_alignsize - 1) & ~ (ecache_alignsize - 1);
1661 
1662 	/*
1663 	 * Copy the code for the copy-rename routine into
1664 	 * a page aligned piece of memory.  We do this to guarantee
1665 	 * that we're executing within the same page and thus reduce
1666 	 * the possibility of cache collisions between different
1667 	 * pages.
1668 	 */
1669 	len = (int)((ulong_t)drmach_copy_rename_end -
1670 		    (ulong_t)drmach_copy_rename_prog__relocatable);
1671 	ASSERT(wp + len < bp + PAGESIZE);
1672 	bcopy((caddr_t)drmach_copy_rename_prog__relocatable, wp, len);
1673 
1674 	DRMACH_PR("copy-rename function 0x%p, len %d\n", (void *)wp, len);
1675 	prog->run = (void (*)())wp;
1676 	wp += (len + ecache_alignsize - 1) & ~ (ecache_alignsize - 1);
1677 
1678 	/*
1679 	 * Prepare data page that will contain script of
1680 	 * operations to perform during copy-rename.
1681 	 * Allocate temporary buffer to hold script.
1682 	 */
1683 	err = drmach_prep_rename_script(s_mem, t_mem, t_slice_offset,
1684 		wp, PAGESIZE - (wp - bp));
1685 	if (err) {
1686 		(void) drmach_copy_rename_fini(prog);
1687 		return (err);
1688 	}
1689 
1690 	DRMACH_PR("copy-rename script 0x%p, len %d\n", (void *)wp, len);
1691 	prog->data = wp;
1692 	wp += (len + ecache_alignsize - 1) & ~ (ecache_alignsize - 1);
1693 
1694 	prog->ph = ph;
1695 	prog->s_copybasepa = s_copybasepa;
1696 	prog->t_copybasepa = t_copybasepa;
1697 	prog->c_ml = c_ml;
1698 	*pgm_id = prog;
1699 
1700 	return (NULL);
1701 }
1702 
1703 sbd_error_t *
1704 drmach_copy_rename_fini(drmachid_t id)
1705 {
1706 	drmach_copy_rename_program_t	*prog = id;
1707 	sbd_error_t			*err = NULL;
1708 
1709 	if (prog->c_ml != NULL)
1710 		memlist_delete(prog->c_ml);
1711 
1712 	if (prog->ph != NULL)
1713 		pda_close(prog->ph);
1714 
1715 	if (prog->restless_mc != 0) {
1716 		cmn_err(CE_WARN, "MC did not idle; OBP Node 0x%x",
1717 			(uint_t)drmach_node_get_dnode(prog->restless_mc->node));
1718 
1719 		err = DRMACH_INTERNAL_ERROR();
1720 	}
1721 
1722 	kmem_free(prog, PAGESIZE);
1723 
1724 	return (err);
1725 }
1726 
1727 static sbd_error_t *
1728 drmach_io_new(drmach_device_t *dp)
1729 {
1730 	static sbd_error_t *drmach_io_release(drmachid_t);
1731 	static sbd_error_t *drmach_io_status(drmachid_t, drmach_status_t *);
1732 
1733 	sbd_error_t	*err;
1734 	int		 portid;
1735 
1736 	err = drmach_device_get_prop(dp, "upa-portid", &portid);
1737 	if (err == NULL) {
1738 		ASSERT(portid & 0x40);
1739 		dp->unum = portid & 1;
1740 	}
1741 
1742 	dp->cm.isa = (void *)drmach_io_new;
1743 	dp->cm.release = drmach_io_release;
1744 	dp->cm.status = drmach_io_status;
1745 
1746 	(void) snprintf(dp->cm.name, sizeof (dp->cm.name), "%s%d", dp->type,
1747 	    dp->unum);
1748 
1749 	return (err);
1750 }
1751 
1752 static void
1753 drmach_iopc_op(pda_handle_t ph, drmach_iopc_op_t op)
1754 {
1755 	register int b;
1756 
1757 	for (b = 0; b < MAX_BOARDS; b++) {
1758 		int		p;
1759 		ushort_t	bda_ioc;
1760 		board_desc_t	*bdesc;
1761 
1762 		if (pda_board_present(ph, b) == 0)
1763 			continue;
1764 
1765 		bdesc = (board_desc_t *)pda_get_board_info(ph, b);
1766 		/*
1767 		 * Update PCs for IOCs.
1768 		 */
1769 		bda_ioc = bdesc->bda_ioc;
1770 		for (p = 0; p < MAX_IOCS; p++) {
1771 			u_longlong_t	idle_addr;
1772 			uchar_t		value;
1773 
1774 			if (BDA_NBL(bda_ioc, p) != BDAN_GOOD)
1775 				continue;
1776 
1777 			idle_addr = STARFIRE_BB_PC_ADDR(b, p, 1);
1778 
1779 			switch (op) {
1780 			case DO_PAUSE:
1781 				value = STARFIRE_BB_PC_PAUSE(p);
1782 				break;
1783 
1784 			case DO_IDLE:
1785 				value = STARFIRE_BB_PC_IDLE(p);
1786 				break;
1787 
1788 			case DO_UNPAUSE:
1789 				value = ldbphysio(idle_addr);
1790 				value &= ~STARFIRE_BB_PC_PAUSE(p);
1791 				break;
1792 
1793 			case DO_UNIDLE:
1794 				value = ldbphysio(idle_addr);
1795 				value &= ~STARFIRE_BB_PC_IDLE(p);
1796 				break;
1797 
1798 			default:
1799 				cmn_err(CE_PANIC,
1800 					"drmach_iopc_op: unknown op (%d)",
1801 					(int)op);
1802 				/*NOTREACHED*/
1803 			}
1804 			stbphysio(idle_addr, value);
1805 		}
1806 	}
1807 }
1808 
1809 void
1810 drmach_copy_rename(drmachid_t id)
1811 {
1812 	drmach_copy_rename_program_t	*prog = id;
1813 	uint64_t			neer;
1814 
1815 	/*
1816 	 * UPA IDLE
1817 	 * Protocol = PAUSE -> IDLE -> UNPAUSE
1818 	 * In reality since we only "idle" the IOPCs it's sufficient
1819 	 * to just issue the IDLE operation since (in theory) all IOPCs
1820 	 * in the field are PC6.  However, we'll be robust and do the
1821 	 * proper workaround protocol so that we never have to worry!
1822 	 */
1823 	drmach_iopc_op(prog->ph, DO_PAUSE);
1824 	drmach_iopc_op(prog->ph, DO_IDLE);
1825 	DELAY(100);
1826 	drmach_iopc_op(prog->ph, DO_UNPAUSE);
1827 	DELAY(100);
1828 
1829 	/* disable CE reporting */
1830 	neer = get_error_enable();
1831 	set_error_enable(neer & ~EER_CEEN);
1832 
1833 	/* run the copy/rename program */
1834 	prog->run(prog);
1835 
1836 	/* enable CE reporting */
1837 	set_error_enable(neer);
1838 
1839 	/*
1840 	 * UPA UNIDLE
1841 	 * Protocol = UNIDLE
1842 	 */
1843 	drmach_iopc_op(prog->ph, DO_UNIDLE);
1844 	DELAY(100);
1845 }
1846 
1847 /*
1848  * The counter-timer and perf-counter nodes are not being cleaned
1849  * up after a board that was present at start of day is detached.
1850  * If the board has become unconfigured with this operation, walk
1851  * the prom tree and find all counter-timer and perf-counter nodes
1852  * that have the same board number as the board that was just
1853  * unconfigured and remove them.
1854  */
1855 static sbd_error_t *
1856 drmach_remove_counter_nodes(drmachid_t id)
1857 {
1858 	int		num;
1859 	char		name[OBP_MAXDRVNAME];
1860 	pnode_t		child;
1861 	dev_info_t	*dip;
1862 	sbd_error_t	*err;
1863 	drmach_status_t	stat;
1864 	drmach_board_t	*bp;
1865 
1866 	if (!DRMACH_IS_BOARD_ID(id)) {
1867 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1868 	}
1869 
1870 	if ((err = drmach_board_status(id, &stat)) != NULL) {
1871 		return (err);
1872 	}
1873 
1874 	/*
1875 	 * Only clean up the counter-timer and perf-counter
1876 	 * nodes when the entire board is unconfigured.
1877 	 */
1878 	if (stat.configured) {
1879 		return (NULL);
1880 	}
1881 
1882 	bp = (drmach_board_t *)id;
1883 
1884 	err = NULL;
1885 
1886 	for (child = prom_childnode(prom_rootnode()); child != OBP_NONODE;
1887 	    child = prom_nextnode(child)) {
1888 
1889 		if (prom_getprop(child, OBP_BOARDNUM, (caddr_t)&num) == -1) {
1890 			continue;
1891 		}
1892 
1893 		if (bp->bnum != num) {
1894 			continue;
1895 		}
1896 
1897 		if (prom_getprop(child, OBP_NAME, (caddr_t)name) == -1) {
1898 			continue;
1899 		}
1900 
1901 		if (strncmp(name, MISC_COUNTER_TIMER_DEVNAME, OBP_MAXDRVNAME) &&
1902 		    strncmp(name, MISC_PERF_COUNTER_DEVNAME, OBP_MAXDRVNAME)) {
1903 				continue;
1904 		}
1905 
1906 		/* Root node doesn't have to be held */
1907 		dip = e_ddi_nodeid_to_dip(child);
1908 
1909 		/*
1910 		 * If the node is only in the OBP tree, then
1911 		 * we don't have to remove it.
1912 		 */
1913 		if (dip) {
1914 			dev_info_t *fdip = NULL;
1915 
1916 			DRMACH_PR("removing %s devinfo node\n", name);
1917 
1918 			e_ddi_branch_hold(dip);
1919 			ddi_release_devi(dip); /* held in e_ddi_nodeid_to_dip */
1920 
1921 			if (e_ddi_branch_destroy(dip, &fdip, 0)) {
1922 				char *path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1923 
1924 				/*
1925 				 * If non-NULL, fdip is held and must be
1926 				 * released.
1927 				 */
1928 				if (fdip != NULL) {
1929 					(void) ddi_pathname(fdip, path);
1930 					ddi_release_devi(fdip);
1931 				} else {
1932 					(void) ddi_pathname(dip, path);
1933 				}
1934 
1935 				err = drerr_new(1, ESTF_DRVFAIL, path);
1936 				kmem_free(path, MAXPATHLEN);
1937 				e_ddi_branch_rele(dip);
1938 				break;
1939 			}
1940 		}
1941 	}
1942 
1943 	return (err);
1944 }
1945 
1946 /*ARGSUSED*/
1947 sbd_error_t *
1948 drmach_pre_op(int cmd, drmachid_t id, drmach_opts_t *opts)
1949 {
1950 	/* allow status and ncm operations to always succeed */
1951 	if ((cmd == SBD_CMD_STATUS) || (cmd == SBD_CMD_GETNCM)) {
1952 		return (NULL);
1953 	}
1954 
1955 	/* check all other commands for the required option string */
1956 	if ((opts->size > 0) && (opts->copts != NULL)) {
1957 
1958 		DRMACH_PR("platform options: %s\n", opts->copts);
1959 
1960 		if (strstr(opts->copts, "xfdr") != NULL) {
1961 			return (NULL);
1962 		}
1963 	}
1964 
1965 	return (drerr_new(0, ESTF_SUPPORT, NULL));
1966 }
1967 
1968 /*ARGSUSED*/
1969 sbd_error_t *
1970 drmach_post_op(int cmd, drmachid_t id, drmach_opts_t *opts)
1971 {
1972 	sbd_error_t	*err = NULL;
1973 
1974 	switch (cmd) {
1975 	case SBD_CMD_UNCONFIGURE:
1976 
1977 		err = drmach_remove_counter_nodes(id);
1978 		break;
1979 
1980 	case SBD_CMD_CONFIGURE:
1981 	case SBD_CMD_DISCONNECT:
1982 	case SBD_CMD_CONNECT:
1983 	case SBD_CMD_GETNCM:
1984 	case SBD_CMD_STATUS:
1985 		break;
1986 
1987 	default:
1988 		break;
1989 	}
1990 
1991 	return (err);
1992 }
1993 
1994 sbd_error_t *
1995 drmach_board_assign(int bnum, drmachid_t *id)
1996 {
1997 	sbd_error_t	*err;
1998 
1999 	if (!drmach_initialized && drmach_init() == -1) {
2000 		err = DRMACH_INTERNAL_ERROR();
2001 	} else if (drmach_array_get(drmach_boards, bnum, id) == -1) {
2002 		err = drerr_new(1, ESTF_BNUM, "%d", bnum);
2003 	} else if (*id != NULL) {
2004 		err = NULL;
2005 	} else {
2006 		drmach_board_t	*bp;
2007 
2008 		*id  = (drmachid_t)drmach_board_new(bnum);
2009 		bp = *id;
2010 		bp->assigned = 1;
2011 		err = NULL;
2012 	}
2013 
2014 	return (err);
2015 }
2016 
2017 static int
2018 drmach_attach_board(void *arg)
2019 {
2020 	drmach_board_t	*obj = (drmach_board_t *)arg;
2021 	cpuset_t	cset;
2022 	int		retval;
2023 
2024 	/*
2025 	 * OBP disables traps during the board probe.
2026 	 * So, in order to prevent cross-call/cross-trap timeouts,
2027 	 * and thus panics, we effectively block anybody from
2028 	 * issuing xc's/xt's by doing a promsafe_xc_attention.
2029 	 * In the previous version of Starfire DR (2.6), a timeout
2030 	 * suspension mechanism was implemented in the send-mondo
2031 	 * assembly.  That mechanism is unnecessary with the
2032 	 * existence of xc_attention/xc_dismissed.
2033 	 */
2034 	cset = cpu_ready_set;
2035 	promsafe_xc_attention(cset);
2036 
2037 	retval = prom_starfire_add_brd(obj->connect_cpuid);
2038 
2039 	xc_dismissed(cset);
2040 
2041 	return (retval);
2042 }
2043 
2044 sbd_error_t *
2045 drmach_board_connect(drmachid_t id, drmach_opts_t *opts)
2046 {
2047 	drmach_board_t	*obj = (drmach_board_t *)id;
2048 	int		retval;
2049 	sbd_error_t	*err;
2050 	char		*cptr, *copts;
2051 
2052 	if (!DRMACH_IS_BOARD_ID(id))
2053 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2054 
2055 	if (opts->size > 0)
2056 		copts = opts->copts;
2057 
2058 	if ((cptr = strstr(copts, "cpuid=")) != NULL) {
2059 		int cpuid;
2060 
2061 		cptr += strlen("cpuid=");
2062 		cpuid = stoi(&cptr);
2063 
2064 		if (DRMACH_CPUID2BNUM(cpuid) == obj->bnum) {
2065 			obj->connect_cpuid = cpuid;
2066 			obj->assigned = 1;
2067 		} else
2068 			return (drerr_new(1, ESTF_SETCPUVAL, "%d", cpuid));
2069 	} else {
2070 		/* cpuid was not specified */
2071 		obj->connect_cpuid = -1;
2072 	}
2073 
2074 	if (obj->connect_cpuid == -1) {
2075 		err =  drerr_new(1, ESTF_NOCPUID, obj->cm.name);
2076 		return (err);
2077 	}
2078 
2079 	cmn_err(CE_CONT, "DRMACH: PROM attach %s CPU %d\n",
2080 		obj->cm.name, obj->connect_cpuid);
2081 
2082 	retval = prom_tree_update(drmach_attach_board, obj);
2083 
2084 	if (retval == 0)
2085 		err = NULL;
2086 	else {
2087 		cmn_err(CE_WARN, "prom error: prom_starfire_add_brd(%d) "
2088 			"returned %d", obj->connect_cpuid, retval);
2089 
2090 		err = drerr_new(1, ESTF_PROBE, obj->cm.name);
2091 	}
2092 
2093 	obj->connect_cpuid = -1;
2094 
2095 	return (err);
2096 }
2097 
2098 /*ARGSUSED*/
2099 sbd_error_t *
2100 drmach_board_disconnect(drmachid_t id, drmach_opts_t *opts)
2101 {
2102 	drmach_board_t		*bp;
2103 	int			rv;
2104 	int			d_idx;	/* device index */
2105 	drmachid_t		d_id;	/* device ID */
2106 	sbd_error_t		*err;
2107 
2108 	if (!DRMACH_IS_BOARD_ID(id))
2109 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2110 
2111 	bp = id;
2112 
2113 	/*
2114 	 * We need to make sure all of the board's device nodes
2115 	 * have been removed from the Solaris device tree before
2116 	 * continuing with the disconnect. Otherwise, we could
2117 	 * disconnect the board and remove the OBP device tree
2118 	 * nodes with Solaris device tree nodes remaining.
2119 	 *
2120 	 * On Starfire, Solaris device tree nodes are deleted
2121 	 * during unconfigure by drmach_unconfigure(). It's
2122 	 * necessary to do this here because drmach_unconfigure()
2123 	 * failures are not handled during unconfigure.
2124 	 */
2125 	if (bp->devices) {
2126 		rv = drmach_array_first(bp->devices, &d_idx, &d_id);
2127 		while (rv == 0) {
2128 			err = drmach_unconfigure(d_id, DRMACH_DEVI_REMOVE);
2129 			if (err)
2130 				return (err);
2131 
2132 			rv = drmach_array_next(bp->devices, &d_idx, &d_id);
2133 		}
2134 	}
2135 
2136 	/*
2137 	 * Starfire board Solaris device tree counter nodes,
2138 	 * which are only present on start-of-day boards, are
2139 	 * removed in the dr_post_op() code flow after the
2140 	 * board is unconfigured. We call the counter node
2141 	 * removal function here because unconfigure errors
2142 	 * can cause the dr_post_op() function to be skipped
2143 	 * after an unconfigure operation even though all of
2144 	 * the board's devices have been transitioned to the
2145 	 * unconfigured state.
2146 	 */
2147 	err = drmach_remove_counter_nodes(id);
2148 	if (err)
2149 		return (err);
2150 
2151 	return (NULL);
2152 }
2153 
2154 static int
2155 drmach_board_find_devices_cb(drmach_node_walk_args_t *args)
2156 {
2157 	drmach_node_t			*node = args->node;
2158 	drmach_board_cb_data_t		*data = args->data;
2159 	drmach_board_t			*obj = data->obj;
2160 
2161 	int		 rv;
2162 	int		 bnum;
2163 	drmach_device_t	*device;
2164 
2165 	rv = drmach_node_get_prop(node, OBP_BOARDNUM, &bnum);
2166 	if (rv) {
2167 		/*
2168 		 * if the node does not have a board# property, then
2169 		 * by that information alone it is known that drmach
2170 		 * is not interested in it.
2171 		 */
2172 		return (0);
2173 	} else if (bnum != obj->bnum)
2174 		return (0);
2175 
2176 	/*
2177 	 * Create a device data structure from this node data.
2178 	 * The call may yield nothing if the node is not of interest
2179 	 * to drmach.
2180 	 */
2181 	data->err = drmach_device_new(node, obj, &device);
2182 	if (data->err)
2183 		return (-1);
2184 	else if (device == NULL) {
2185 		/*
2186 		 * drmach_device_new examined the node we passed in
2187 		 * and determined that it was one not of interest to
2188 		 * drmach.  So, it is skipped.
2189 		 */
2190 		return (0);
2191 	}
2192 
2193 	rv = drmach_array_set(obj->devices, data->ndevs++, device);
2194 	if (rv) {
2195 		drmach_device_dispose(device);
2196 		data->err = DRMACH_INTERNAL_ERROR();
2197 		return (-1);
2198 	}
2199 
2200 	data->err = (*data->found)(data->a, device->type, device->unum, device);
2201 	return (data->err == NULL ? 0 : -1);
2202 }
2203 
2204 sbd_error_t *
2205 drmach_board_find_devices(drmachid_t id, void *a,
2206 	sbd_error_t *(*found)(void *a, const char *, int, drmachid_t))
2207 {
2208 	extern int		 plat_max_cpu_units_per_board();
2209 	extern int		 plat_max_mem_units_per_board();
2210 	extern int		 plat_max_io_units_per_board();
2211 
2212 	drmach_board_t		*obj = (drmach_board_t *)id;
2213 	sbd_error_t		*err;
2214 	int			 max_devices;
2215 	int			 rv;
2216 	drmach_board_cb_data_t	data;
2217 
2218 	max_devices  = plat_max_cpu_units_per_board();
2219 	max_devices += plat_max_mem_units_per_board();
2220 	max_devices += plat_max_io_units_per_board();
2221 
2222 	obj->devices = drmach_array_new(0, max_devices);
2223 
2224 	data.obj = obj;
2225 	data.ndevs = 0;
2226 	data.found = found;
2227 	data.a = a;
2228 	data.err = NULL;
2229 
2230 	rv = drmach_node_walk(obj->tree, &data, drmach_board_find_devices_cb);
2231 	if (rv == 0)
2232 		err = NULL;
2233 	else {
2234 		drmach_array_dispose(obj->devices, drmach_device_dispose);
2235 		obj->devices = NULL;
2236 
2237 		if (data.err)
2238 			err = data.err;
2239 		else
2240 			err = DRMACH_INTERNAL_ERROR();
2241 	}
2242 
2243 	return (err);
2244 }
2245 
2246 int
2247 drmach_board_lookup(int bnum, drmachid_t *id)
2248 {
2249 	int	rv = 0;
2250 
2251 	if (!drmach_initialized && drmach_init() == -1) {
2252 		*id = 0;
2253 		rv = -1;
2254 	} else if (drmach_array_get(drmach_boards, bnum, id)) {
2255 		*id = 0;
2256 		rv = -1;
2257 	}
2258 	return (rv);
2259 }
2260 
2261 sbd_error_t *
2262 drmach_board_name(int bnum, char *buf, int buflen)
2263 {
2264 	(void) snprintf(buf, buflen, "SB%d", bnum);
2265 	return (NULL);
2266 }
2267 
2268 sbd_error_t *
2269 drmach_board_poweroff(drmachid_t id)
2270 {
2271 	drmach_board_t	*bp;
2272 	sbd_error_t	*err;
2273 	drmach_status_t	 stat;
2274 
2275 	if (!DRMACH_IS_BOARD_ID(id))
2276 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2277 	bp = id;
2278 
2279 	err = drmach_board_status(id, &stat);
2280 	if (err)
2281 		return (err);
2282 	else if (stat.configured || stat.busy)
2283 		return (drerr_new(0, ESTF_CONFIGBUSY, bp->cm.name));
2284 	else {
2285 		/* board power off is essentially a noop for Starfire */
2286 		bp->powered = 0;
2287 		return (NULL);
2288 	}
2289 	/*NOTREACHED*/
2290 }
2291 
2292 sbd_error_t *
2293 drmach_board_poweron(drmachid_t id)
2294 {
2295 	drmach_board_t	*bp;
2296 
2297 	if (!DRMACH_IS_BOARD_ID(id))
2298 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2299 	bp = id;
2300 
2301 	/* board power on is essentially a noop for Starfire */
2302 	bp->powered = 1;
2303 
2304 	return (NULL);
2305 }
2306 
2307 static sbd_error_t *
2308 drmach_board_release(drmachid_t id)
2309 {
2310 	if (!DRMACH_IS_BOARD_ID(id))
2311 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2312 	return (NULL);
2313 }
2314 
2315 /*ARGSUSED*/
2316 sbd_error_t *
2317 drmach_board_test(drmachid_t id, drmach_opts_t *opts, int force)
2318 {
2319 	return (NULL);
2320 }
2321 
2322 sbd_error_t *
2323 drmach_board_unassign(drmachid_t id)
2324 {
2325 	drmach_board_t	*bp;
2326 	sbd_error_t	*err;
2327 	drmach_status_t	 stat;
2328 
2329 	if (!DRMACH_IS_BOARD_ID(id))
2330 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2331 	bp = id;
2332 
2333 	err = drmach_board_status(id, &stat);
2334 	if (err)
2335 		return (err);
2336 	else if (stat.configured || stat.busy)
2337 		return (drerr_new(0, ESTF_CONFIGBUSY, bp->cm.name));
2338 	else if (drmach_array_set(drmach_boards, bp->bnum, 0) != 0)
2339 		return (DRMACH_INTERNAL_ERROR());
2340 	else {
2341 		drmach_board_dispose(bp);
2342 		return (NULL);
2343 	}
2344 	/*NOTREACHED*/
2345 }
2346 
2347 static sbd_error_t *
2348 drmach_cpu_new(drmach_device_t *dp)
2349 {
2350 	static sbd_error_t *drmach_cpu_release(drmachid_t);
2351 	static sbd_error_t *drmach_cpu_status(drmachid_t, drmach_status_t *);
2352 
2353 	sbd_error_t	*err;
2354 	int		 portid;
2355 
2356 	err = drmach_device_get_prop(dp, "upa-portid", &portid);
2357 	if (err == NULL)
2358 		dp->unum = portid & 3;
2359 
2360 	dp->cm.isa = (void *)drmach_cpu_new;
2361 	dp->cm.release = drmach_cpu_release;
2362 	dp->cm.status = drmach_cpu_status;
2363 
2364 	(void) snprintf(dp->cm.name, sizeof (dp->cm.name), "%s%d", dp->type,
2365 	    dp->unum);
2366 
2367 	return (err);
2368 }
2369 
2370 /*
2371  * drmach_cpu_obp_detach()
2372  *  This requires two steps, first, we must put the cpuid into the OBP
2373  *  idle loop (Idle in Program) state.  Then we call OBP to place the CPU
2374  *  into the "Detached" state, which does any special processing to
2375  *  actually detach the cpu, such as flushing ecache, and also ensures
2376  *  that a subsequent breakpoint won't restart the cpu (if it was just in
2377  *  Idle in Program state).
2378  */
2379 static void
2380 drmach_cpu_obp_detach(int cpuid)
2381 {
2382 	/*
2383 	 * Cpu may not be under OBP's control. Eg, if cpu exited to download
2384 	 * helper on a prior attach.
2385 	 */
2386 	if (CPU_SGN_EXISTS(cpuid) &&
2387 			!SGN_CPU_IS_OS(cpuid) &&
2388 			!SGN_CPU_IS_OBP(cpuid)) {
2389 		cmn_err(CE_WARN,
2390 			"unexpected signature (0x%x) for cpu %d",
2391 			get_cpu_sgn(cpuid), cpuid);
2392 	}
2393 
2394 	/*
2395 	 * Now we place the CPU into the "Detached" idle loop in OBP.
2396 	 * This is so that the CPU won't be restarted if we break into
2397 	 * OBP with a breakpoint or BREAK key from the console, and also
2398 	 * if we need to do any special processing, such as flushing the
2399 	 * cpu's ecache, disabling interrupts (by turning of the ET bit in
2400 	 * the PSR) and/or spinning in BBSRAM rather than global memory.
2401 	 */
2402 	DRMACH_PR("prom_starfire_rm_cpu(%d)\n", cpuid);
2403 	prom_starfire_rm_cpu(cpuid);
2404 }
2405 
2406 /*
2407  * drmach_cpu_obp_is_detached() returns TRUE if the cpu sigblock signature state
2408  * is SIGBST_DETACHED; otherwise it returns FALSE. This routine should only
2409  * be called after we have asked OBP to detach the CPU. It should NOT be
2410  * called as a check during any other flow.
2411  */
2412 static int
2413 drmach_cpu_obp_is_detached(int cpuid)
2414 {
2415 	if (!CPU_SGN_EXISTS(cpuid) ||
2416 		(SGN_CPU_IS_OS(cpuid) && SGN_CPU_STATE_IS_DETACHED(cpuid)))
2417 		return (1);
2418 	else
2419 		return (0);
2420 }
2421 
2422 static int
2423 drmach_cpu_start(struct cpu *cp)
2424 {
2425 	int		cpuid = cp->cpu_id;
2426 	int		ntries = drmach_cpu_ntries;
2427 	extern void	restart_other_cpu(int);
2428 
2429 	ASSERT(MUTEX_HELD(&cpu_lock));
2430 	ASSERT(cpunodes[cpuid].nodeid != (pnode_t)0);
2431 
2432 	cp->cpu_flags &= ~CPU_POWEROFF;
2433 
2434 	/*
2435 	 * NOTE: restart_other_cpu pauses cpus during the
2436 	 *	 slave cpu start.  This helps to quiesce the
2437 	 *	 bus traffic a bit which makes the tick sync
2438 	 *	 routine in the prom more robust.
2439 	 */
2440 	DRMACH_PR("COLD START for cpu (%d)\n", cpuid);
2441 
2442 	prom_starfire_add_cpu(cpuid);
2443 
2444 	restart_other_cpu(cpuid);
2445 
2446 	/*
2447 	 * Wait for the cpu to reach its idle thread before
2448 	 * we zap him with a request to blow away the mappings
2449 	 * he (might) have for the drmach_shutdown_asm code
2450 	 * he may have executed on unconfigure.
2451 	 */
2452 	while ((cp->cpu_thread != cp->cpu_idle_thread) && (ntries > 0)) {
2453 		DELAY(drmach_cpu_delay);
2454 		ntries--;
2455 	}
2456 
2457 	DRMACH_PR("waited %d out of %d loops for cpu %d\n",
2458 		drmach_cpu_ntries - ntries, drmach_cpu_ntries, cpuid);
2459 
2460 	xt_one(cpuid, vtag_flushpage_tl1,
2461 		(uint64_t)drmach_shutdown_va, (uint64_t)ksfmmup);
2462 
2463 	return (0);
2464 }
2465 
2466 /*
2467  * A detaching CPU is xcalled with an xtrap to drmach_cpu_stop_self() after
2468  * it has been offlined. The function of this routine is to get the cpu
2469  * spinning in a safe place. The requirement is that the system will not
2470  * reference anything on the detaching board (memory and i/o is detached
2471  * elsewhere) and that the CPU not reference anything on any other board
2472  * in the system.  This isolation is required during and after the writes
2473  * to the domain masks to remove the board from the domain.
2474  *
2475  * To accomplish this isolation the following is done:
2476  *	1) Create a locked mapping to a location in BBSRAM where
2477  *	   the cpu will execute.
2478  *	2) Copy the target function (drmach_shutdown_asm) in which
2479  *	   the cpu will execute into BBSRAM.
2480  *	3) Jump into function with BBSRAM.
2481  *	   Function will:
2482  *	   3.1) Flush its Ecache (displacement).
2483  *	   3.2) Flush its Dcache with HW mechanism.
2484  *	   3.3) Flush its Icache with HW mechanism.
2485  *	   3.4) Flush all valid and _unlocked_ D-TLB entries.
2486  *	   3.5) Flush all valid and _unlocked_ I-TLB entries.
2487  *	   3.6) Clear xt_mb to signal completion. Note: cache line is
2488  *		recovered by drmach_cpu_poweroff().
2489  *	4) Jump into a tight loop.
2490  */
2491 #define	DRMACH_BBSRAM_OFFSET	0x1000
2492 
2493 static void
2494 drmach_cpu_stop_self(void)
2495 {
2496 	int		cpuid = (int)CPU->cpu_id;
2497 	tte_t		tte;
2498 	volatile uint_t	*src, *dst;
2499 	uint_t		funclen;
2500 	uint64_t	bbsram_pa, bbsram_offset;
2501 	uint_t		bbsram_pfn;
2502 	uint64_t	bbsram_addr;
2503 	void		(*bbsram_func)(uint64_t);
2504 	extern void	drmach_shutdown_asm(uint64_t);
2505 	extern void	drmach_shutdown_asm_end(void);
2506 
2507 	funclen = (uint_t)drmach_shutdown_asm_end - (uint_t)drmach_shutdown_asm;
2508 	ASSERT(funclen <= MMU_PAGESIZE);
2509 	/*
2510 	 * We'll start from the 0th's base.
2511 	 */
2512 	bbsram_pa = STARFIRE_UPAID2UPS(cpuid) | STARFIRE_PSI_BASE;
2513 	bbsram_offset = bbsram_pa | 0xfe0ULL;
2514 	bbsram_pa += ldphysio(bbsram_offset) + DRMACH_BBSRAM_OFFSET;
2515 
2516 	bbsram_pfn = (uint_t)(bbsram_pa >> MMU_PAGESHIFT);
2517 
2518 	bbsram_addr = (uint64_t)drmach_shutdown_va;
2519 	drmach_shutdown_asm_mbox->estack = bbsram_addr + (uint64_t)funclen;
2520 
2521 	tte.tte_inthi = TTE_VALID_INT | TTE_SZ_INT(TTE8K) |
2522 			TTE_PFN_INTHI(bbsram_pfn);
2523 	tte.tte_intlo = TTE_PFN_INTLO(bbsram_pfn) |
2524 			TTE_HWWR_INT | TTE_PRIV_INT | TTE_LCK_INT;
2525 	sfmmu_dtlb_ld_kva(drmach_shutdown_va, &tte);	/* load dtlb */
2526 	sfmmu_itlb_ld_kva(drmach_shutdown_va, &tte);	/* load itlb */
2527 
2528 	for (src = (uint_t *)drmach_shutdown_asm, dst = (uint_t *)bbsram_addr;
2529 		src < (uint_t *)drmach_shutdown_asm_end; src++, dst++)
2530 		*dst = *src;
2531 
2532 	bbsram_func = (void (*)())bbsram_addr;
2533 	drmach_shutdown_asm_mbox->flushaddr = ecache_flushaddr;
2534 	drmach_shutdown_asm_mbox->size = (cpunodes[cpuid].ecache_size << 1);
2535 	drmach_shutdown_asm_mbox->linesize = cpunodes[cpuid].ecache_linesize;
2536 	drmach_shutdown_asm_mbox->physaddr
2537 				    = va_to_pa((void *)&drmach_xt_mb[cpuid]);
2538 
2539 	/*
2540 	 * Signal to drmach_cpu_poweroff() is via drmach_xt_mb cleared
2541 	 * by asm code
2542 	 */
2543 
2544 	(*bbsram_func)(va_to_pa((void *)drmach_shutdown_asm_mbox));
2545 }
2546 
2547 static void
2548 drmach_cpu_shutdown_self(void)
2549 {
2550 	cpu_t		*cp = CPU;
2551 	int		cpuid = cp->cpu_id;
2552 	extern void	flush_windows(void);
2553 
2554 	flush_windows();
2555 
2556 	(void) spl8();
2557 
2558 	ASSERT(cp->cpu_intr_actv == 0);
2559 	ASSERT(cp->cpu_thread == cp->cpu_idle_thread ||
2560 	    cp->cpu_thread == cp->cpu_startup_thread);
2561 
2562 	cp->cpu_flags = CPU_OFFLINE | CPU_QUIESCED | CPU_POWEROFF;
2563 
2564 	drmach_cpu_stop_self();
2565 
2566 	cmn_err(CE_PANIC, "CPU %d FAILED TO SHUTDOWN", cpuid);
2567 }
2568 
2569 /* a helper routine to keep the math in one place */
2570 static processorid_t
2571 drmach_cpu_calc_id(drmach_device_t *dp)
2572 {
2573 	return (dp->bp->bnum * MAX_CPU_UNITS_PER_BOARD + dp->unum);
2574 }
2575 
2576 /*
2577  * Move bootproc (SIGBCPU) to another cpu.  If dst_cpu is NULL, a
2578  * destination cpu is chosen from the set of cpus not located on the
2579  * same board as the current bootproc cpu.
2580  */
2581 static sbd_error_t *
2582 drmach_cpu_juggle_bootproc(drmach_device_t *dst_cpu)
2583 {
2584 	processorid_t	 cpuid;
2585 	struct cpu	*cp;
2586 	sbd_error_t	*err;
2587 	int		 rv;
2588 
2589 	ASSERT(MUTEX_HELD(&cpu_lock));
2590 
2591 	/* dst_cpu is NULL when target cpu is unspecified. So, pick one. */
2592 	if (dst_cpu == NULL) {
2593 		int avoid_board = DRMACH_CPUID2BNUM(SIGBCPU->cpu_id);
2594 		int max_cpuid = MAX_BOARDS * MAX_CPU_UNITS_PER_BOARD;
2595 
2596 		for (cpuid = 0; cpuid < max_cpuid; cpuid++)
2597 			if (DRMACH_CPUID2BNUM(cpuid) != avoid_board) {
2598 				cp = cpu_get(cpuid);
2599 				if (cp != NULL && cpu_is_online(cp))
2600 					break;
2601 			}
2602 
2603 		if (cpuid == max_cpuid) {
2604 			err = drerr_new(1, ESTF_JUGGLE, NULL);
2605 			return (err);
2606 		}
2607 
2608 		/* else, cp points to the selected target cpu */
2609 	} else {
2610 		cpuid = drmach_cpu_calc_id(dst_cpu);
2611 
2612 		if ((cp = cpu_get(cpuid)) == NULL) {
2613 			err = drerr_new(1, ESTF_NODEV, "%s::%s",
2614 				dst_cpu->bp->cm.name, dst_cpu->cm.name);
2615 			return (err);
2616 		}
2617 
2618 		if (cpuid == SIGBCPU->cpu_id) {
2619 			cmn_err(CE_WARN,
2620 				"SIGBCPU(%d) same as new selection(%d)",
2621 				SIGBCPU->cpu_id, cpuid);
2622 
2623 			/* technically not an error, but a no-op */
2624 			return (NULL);
2625 		}
2626 	}
2627 
2628 	cmn_err(CE_NOTE, "?relocating SIGBCPU from %d to %d",
2629 		SIGBCPU->cpu_id, cpuid);
2630 
2631 	DRMACH_PR("moving SIGBCPU to CPU %d\n", cpuid);
2632 
2633 	/*
2634 	 * Tell OBP to initialize cvc-offset field of new CPU0
2635 	 * so that it's in sync with OBP and cvc_server
2636 	 */
2637 	prom_starfire_init_console(cpuid);
2638 
2639 	/*
2640 	 * Assign cvc to new cpu0's bbsram for I/O.  This has to be
2641 	 * done BEFORE cpu0 is moved via obp, since this logic
2642 	 * will cause obp_helper to switch to a different bbsram for
2643 	 * cvc I/O.  We don't want cvc writing to a buffer from which
2644 	 * nobody will pick up the data!
2645 	 */
2646 	cvc_assign_iocpu(cpuid);
2647 
2648 	rv = prom_starfire_move_cpu0(cpuid);
2649 
2650 	if (rv == 0) {
2651 		SIGBCPU = cp;
2652 
2653 		DRMACH_PR("successfully juggled to CPU %d\n", cpuid);
2654 		return (NULL);
2655 	} else {
2656 		DRMACH_PR("prom error: prom_starfire_move_cpu0(%d) "
2657 			"returned %d\n", cpuid, rv);
2658 
2659 		/*
2660 		 * The move failed, hopefully obp_helper is still back
2661 		 * at the old bootproc.  Move cvc back there.
2662 		 */
2663 		cvc_assign_iocpu(SIGBCPU->cpu_id);
2664 
2665 
2666 		err = drerr_new(1, ESTF_MOVESIGB, "CPU %d", cpuid);
2667 		return (err);
2668 	}
2669 	/*NOTREACHED*/
2670 }
2671 
2672 static sbd_error_t *
2673 drmach_cpu_release(drmachid_t id)
2674 {
2675 	drmach_device_t	*dp;
2676 	processorid_t	 cpuid;
2677 	struct cpu	*cp;
2678 	sbd_error_t	*err;
2679 
2680 	if (!DRMACH_IS_CPU_ID(id))
2681 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2682 	dp = id;
2683 	cpuid = drmach_cpu_calc_id(dp);
2684 
2685 	ASSERT(MUTEX_HELD(&cpu_lock));
2686 
2687 	cp = cpu_get(cpuid);
2688 	if (cp == NULL)
2689 		err = DRMACH_INTERNAL_ERROR();
2690 	else if (SIGBCPU->cpu_id == cp->cpu_id)
2691 		err = drmach_cpu_juggle_bootproc(NULL);
2692 	else
2693 		err = NULL;
2694 
2695 	return (err);
2696 }
2697 
2698 static sbd_error_t *
2699 drmach_cpu_status(drmachid_t id, drmach_status_t *stat)
2700 {
2701 	drmach_device_t *dp;
2702 
2703 	ASSERT(DRMACH_IS_CPU_ID(id));
2704 	dp = id;
2705 
2706 	stat->assigned = dp->bp->assigned;
2707 	stat->powered = dp->bp->powered;
2708 	mutex_enter(&cpu_lock);
2709 	stat->configured = (cpu_get(drmach_cpu_calc_id(dp)) != NULL);
2710 	mutex_exit(&cpu_lock);
2711 	stat->busy = dp->busy;
2712 	(void) strncpy(stat->type, dp->type, sizeof (stat->type));
2713 	stat->info[0] = '\0';
2714 
2715 	return (NULL);
2716 }
2717 
2718 sbd_error_t *
2719 drmach_cpu_disconnect(drmachid_t id)
2720 {
2721 	drmach_device_t	*cpu;
2722 	int		 cpuid;
2723 	int		 ntries;
2724 	int		 p;
2725 	u_longlong_t	 pc_addr;
2726 	uchar_t		 rvalue;
2727 
2728 	if (!DRMACH_IS_CPU_ID(id))
2729 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2730 	cpu = id;
2731 
2732 	cpuid = drmach_cpu_calc_id(cpu);
2733 	if (SIGBCPU->cpu_id == cpuid) {
2734 		/* this cpu is SIGBCPU, can't disconnect */
2735 		return (drerr_new(1, ESTF_HASSIGB, "%s::%s",
2736 				cpu->bp->cm.name, cpu->cm.name));
2737 	}
2738 
2739 	/*
2740 	 * Make sure SIGBST_DETACHED is set before
2741 	 * mapping out the sig block.
2742 	 */
2743 	ntries = drmach_cpu_ntries;
2744 	while (!drmach_cpu_obp_is_detached(cpuid) && ntries) {
2745 		DELAY(drmach_cpu_delay);
2746 		ntries--;
2747 	}
2748 	if (!drmach_cpu_obp_is_detached(cpuid)) {
2749 		cmn_err(CE_WARN, "failed to mark cpu %d detached in sigblock",
2750 			cpuid);
2751 	}
2752 
2753 	/* map out signature block */
2754 	if (CPU_SGN_EXISTS(cpuid)) {
2755 		CPU_SGN_MAPOUT(cpuid);
2756 	}
2757 
2758 	/*
2759 	 * We now PC IDLE the processor to guarantee we
2760 	 * stop any transactions from coming from it.
2761 	 */
2762 	p = cpu->unum & 1;
2763 	pc_addr = STARFIRE_BB_PC_ADDR(cpu->bp->bnum, cpu->unum, 0);
2764 
2765 	DRMACH_PR("PC idle cpu %d (addr = 0x%llx, port = %d, p = %d)",
2766 		drmach_cpu_calc_id(cpu), pc_addr, cpu->unum, p);
2767 
2768 	rvalue = ldbphysio(pc_addr);
2769 	rvalue |= STARFIRE_BB_PC_IDLE(p);
2770 	stbphysio(pc_addr, rvalue);
2771 	DELAY(50000);
2772 
2773 	return (NULL);
2774 }
2775 
2776 sbd_error_t *
2777 drmach_cpu_get_id(drmachid_t id, processorid_t *cpuid)
2778 {
2779 	drmach_device_t *cpu;
2780 
2781 	if (!DRMACH_IS_CPU_ID(id))
2782 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2783 	cpu = id;
2784 
2785 	*cpuid = drmach_cpu_calc_id(cpu);
2786 	return (NULL);
2787 }
2788 
2789 sbd_error_t *
2790 drmach_cpu_get_impl(drmachid_t id, int *ip)
2791 {
2792 	drmach_device_t *cpu;
2793 	int		impl;
2794 
2795 	if (!DRMACH_IS_CPU_ID(id))
2796 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2797 
2798 	cpu = id;
2799 
2800 	if (drmach_node_get_prop(cpu->node, "implementation#", &impl) == -1) {
2801 		return (DRMACH_INTERNAL_ERROR());
2802 	}
2803 
2804 	*ip = impl;
2805 
2806 	return (NULL);
2807 }
2808 
2809 void
2810 drmach_cpu_flush_ecache_sync(void)
2811 {
2812 	ASSERT(curthread->t_bound_cpu == CPU);
2813 
2814 	/*
2815 	 * Now let's flush our ecache thereby removing all references
2816 	 * to the target (detaching) memory from all ecache's in
2817 	 * system.
2818 	 */
2819 	cpu_flush_ecache();
2820 
2821 	/*
2822 	 * Delay 100 usec out of paranoia to insure everything
2823 	 * (hardware queues) has drained before we start reprogramming
2824 	 * the hardware.
2825 	 */
2826 	DELAY(100);
2827 }
2828 
2829 sbd_error_t *
2830 drmach_get_dip(drmachid_t id, dev_info_t **dip)
2831 {
2832 	drmach_device_t	*dp;
2833 
2834 	if (!DRMACH_IS_DEVICE_ID(id))
2835 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2836 	dp = id;
2837 
2838 	*dip = drmach_node_get_dip(dp->node);
2839 	return (NULL);
2840 }
2841 
2842 sbd_error_t *
2843 drmach_io_is_attached(drmachid_t id, int *yes)
2844 {
2845 	drmach_device_t *dp;
2846 	dev_info_t	*dip;
2847 	int		state;
2848 
2849 	if (!DRMACH_IS_IO_ID(id))
2850 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2851 	dp = id;
2852 
2853 	dip = drmach_node_get_dip(dp->node);
2854 	if (dip == NULL) {
2855 		*yes = 0;
2856 		return (NULL);
2857 	}
2858 
2859 	state = ddi_get_devstate(dip);
2860 	*yes = (i_ddi_devi_attached(dip) || (state == DDI_DEVSTATE_UP));
2861 
2862 	return (NULL);
2863 }
2864 
2865 sbd_error_t *
2866 drmach_io_pre_release(drmachid_t id)
2867 {
2868 	if (!DRMACH_IS_IO_ID(id))
2869 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2870 	return (NULL);
2871 }
2872 
2873 static sbd_error_t *
2874 drmach_io_release(drmachid_t id)
2875 {
2876 	if (!DRMACH_IS_IO_ID(id))
2877 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2878 	return (NULL);
2879 }
2880 
2881 sbd_error_t *
2882 drmach_io_unrelease(drmachid_t id)
2883 {
2884 	if (!DRMACH_IS_IO_ID(id))
2885 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2886 	return (NULL);
2887 }
2888 
2889 /*ARGSUSED*/
2890 sbd_error_t *
2891 drmach_io_post_release(drmachid_t id)
2892 {
2893 	return (NULL);
2894 }
2895 
2896 /*ARGSUSED*/
2897 sbd_error_t *
2898 drmach_io_post_attach(drmachid_t id)
2899 {
2900 	return (NULL);
2901 }
2902 
2903 static sbd_error_t *
2904 drmach_io_status(drmachid_t id, drmach_status_t *stat)
2905 {
2906 	drmach_device_t *dp;
2907 	sbd_error_t	*err;
2908 	int		 configured;
2909 
2910 	ASSERT(DRMACH_IS_IO_ID(id));
2911 	dp = id;
2912 
2913 	err = drmach_io_is_attached(id, &configured);
2914 	if (err)
2915 		return (err);
2916 
2917 	stat->assigned = dp->bp->assigned;
2918 	stat->powered = dp->bp->powered;
2919 	stat->configured = (configured != 0);
2920 	stat->busy = dp->busy;
2921 	(void) strncpy(stat->type, dp->type, sizeof (stat->type));
2922 	stat->info[0] = '\0';
2923 
2924 	return (NULL);
2925 }
2926 
2927 static sbd_error_t *
2928 drmach_mem_new(drmach_device_t *dp)
2929 {
2930 	static sbd_error_t *drmach_mem_release(drmachid_t);
2931 	static sbd_error_t *drmach_mem_status(drmachid_t, drmach_status_t *);
2932 
2933 	dp->unum = 0;
2934 	dp->cm.isa = (void *)drmach_mem_new;
2935 	dp->cm.release = drmach_mem_release;
2936 	dp->cm.status = drmach_mem_status;
2937 
2938 	(void) snprintf(dp->cm.name, sizeof (dp->cm.name), "%s", dp->type);
2939 
2940 	return (NULL);
2941 }
2942 
2943 sbd_error_t *
2944 drmach_mem_add_span(drmachid_t id, uint64_t basepa, uint64_t size)
2945 {
2946 	pfn_t		basepfn = (pfn_t)(basepa >> PAGESHIFT);
2947 	pgcnt_t		npages = (pgcnt_t)(size >> PAGESHIFT);
2948 	pda_handle_t	ph;
2949 	int		rv;
2950 
2951 	ASSERT(size != 0);
2952 
2953 	if (!DRMACH_IS_MEM_ID(id))
2954 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2955 
2956 	rv = kcage_range_add(basepfn, npages, KCAGE_DOWN);
2957 	if (rv == ENOMEM) {
2958 		cmn_err(CE_WARN, "%lu megabytes not available to kernel cage",
2959 			(ulong_t)(size == 0 ? 0 : size / MBYTE));
2960 	} else if (rv != 0) {
2961 		/* catch this in debug kernels */
2962 		ASSERT(0);
2963 
2964 		cmn_err(CE_WARN, "unexpected kcage_range_add"
2965 			" return value %d", rv);
2966 	}
2967 
2968 	/*
2969 	 * Update the PDA (post2obp) structure with the
2970 	 * range of the newly added memory.
2971 	 */
2972 	ph = drmach_pda_open();
2973 	if (ph != NULL) {
2974 		pda_mem_add_span(ph, basepa, size);
2975 		pda_close(ph);
2976 	}
2977 
2978 	return (NULL);
2979 }
2980 
2981 sbd_error_t *
2982 drmach_mem_del_span(drmachid_t id, uint64_t basepa, uint64_t size)
2983 {
2984 	drmach_device_t	*mem = id;
2985 	pfn_t		basepfn = (pfn_t)(basepa >> PAGESHIFT);
2986 	pgcnt_t		npages = (pgcnt_t)(size >> PAGESHIFT);
2987 	uint_t		mcreg;
2988 	sbd_error_t	*err;
2989 	pda_handle_t	ph;
2990 	int		rv;
2991 
2992 	err = drmach_read_mc_asr(id, &mcreg);
2993 	if (err)
2994 		return (err);
2995 	else if (mcreg & STARFIRE_MC_INTERLEAVE_MASK) {
2996 		return (drerr_new(1, ESTF_INTERBOARD, "%s::%s",
2997 				mem->bp->cm.name, mem->cm.name));
2998 	}
2999 
3000 	if (size > 0) {
3001 		rv = kcage_range_delete_post_mem_del(basepfn, npages);
3002 		if (rv != 0) {
3003 			cmn_err(CE_WARN,
3004 			    "unexpected kcage_range_delete_post_mem_del"
3005 			    " return value %d", rv);
3006 			return (DRMACH_INTERNAL_ERROR());
3007 		}
3008 	}
3009 
3010 	/*
3011 	 * Update the PDA (post2obp) structure with the
3012 	 * range of removed memory.
3013 	 */
3014 	ph = drmach_pda_open();
3015 	if (ph != NULL) {
3016 		if (size > 0)
3017 			pda_mem_del_span(ph, basepa, size);
3018 
3019 		/* update PDA to board's new mc register settings */
3020 		pda_mem_sync(ph, mem->bp->bnum, 0);
3021 
3022 		pda_close(ph);
3023 	}
3024 
3025 	return (NULL);
3026 }
3027 
3028 /* support routine for enable and disable */
3029 static sbd_error_t *
3030 drmach_mem_update_interconnect(drmachid_t id, uint_t mcreg)
3031 {
3032 	drmach_device_t	*dp;
3033 	pda_handle_t	 ph;
3034 	int		 b;
3035 
3036 	if (!DRMACH_IS_MEM_ID(id))
3037 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3038 	dp = id;
3039 
3040 	ph = drmach_pda_open();
3041 	if (ph == NULL)
3042 		return (DRMACH_INTERNAL_ERROR());
3043 
3044 	for (b = 0; b < MAX_BOARDS; b++) {
3045 		int		p;
3046 		int		rv;
3047 		ushort_t	bda_proc, bda_ioc;
3048 		board_desc_t	*bdesc;
3049 
3050 		if (pda_board_present(ph, b) == 0)
3051 			continue;
3052 
3053 		bdesc = (board_desc_t *)pda_get_board_info(ph, b);
3054 
3055 		/*
3056 		 * Update PCs for CPUs.
3057 		 */
3058 
3059 		/* make sure definition in platmod is in sync with pda */
3060 		ASSERT(MAX_PROCMODS == MAX_CPU_UNITS_PER_BOARD);
3061 
3062 		bda_proc = bdesc->bda_proc;
3063 		for (p = 0; p < MAX_PROCMODS; p++) {
3064 			if (BDA_NBL(bda_proc, p) != BDAN_GOOD)
3065 				continue;
3066 
3067 			rv = pc_madr_add(b, dp->bp->bnum, p, mcreg);
3068 			if (rv) {
3069 				pda_close(ph);
3070 				return (DRMACH_INTERNAL_ERROR());
3071 			}
3072 		}
3073 
3074 		/*
3075 		 * Update PCs for IOCs.
3076 		 */
3077 
3078 		/* make sure definition in platmod is in sync with pda */
3079 		ASSERT(MAX_IOCS == MAX_IO_UNITS_PER_BOARD);
3080 
3081 		bda_ioc = bdesc->bda_ioc;
3082 		for (p = 0; p < MAX_IOCS; p++) {
3083 			if (BDA_NBL(bda_ioc, p) != BDAN_GOOD)
3084 				continue;
3085 
3086 			rv = pc_madr_add(b, dp->bp->bnum, p + 4, mcreg);
3087 			if (rv) {
3088 				pda_close(ph);
3089 				return (DRMACH_INTERNAL_ERROR());
3090 			}
3091 		}
3092 	}
3093 
3094 	pda_close(ph);
3095 	return (NULL);
3096 }
3097 
3098 sbd_error_t *
3099 drmach_mem_disable(drmachid_t id)
3100 {
3101 	sbd_error_t	*err;
3102 	uint_t		 mcreg;
3103 
3104 	err = drmach_read_mc_asr(id, &mcreg);
3105 	if (err == NULL) {
3106 		ASSERT(mcreg & STARFIRE_MC_MEM_PRESENT_MASK);
3107 
3108 		/* Turn off presence bit. */
3109 		mcreg &= ~STARFIRE_MC_MEM_PRESENT_MASK;
3110 
3111 		err = drmach_mem_update_interconnect(id, mcreg);
3112 		if (err == NULL)
3113 			err = drmach_write_mc_asr(id, mcreg);
3114 	}
3115 
3116 	return (err);
3117 }
3118 
3119 sbd_error_t *
3120 drmach_mem_enable(drmachid_t id)
3121 {
3122 	sbd_error_t	*err;
3123 	uint_t		 mcreg;
3124 
3125 	err = drmach_read_mc_asr(id, &mcreg);
3126 	if (err == NULL) {
3127 		mcreg |= STARFIRE_MC_MEM_PRESENT_MASK;
3128 
3129 		err = drmach_write_mc_asr(id, mcreg);
3130 		if (err == NULL)
3131 			err = drmach_mem_update_interconnect(id, mcreg);
3132 	}
3133 
3134 	return (err);
3135 }
3136 
3137 sbd_error_t *
3138 drmach_mem_get_alignment(drmachid_t id, uint64_t *mask)
3139 {
3140 	drmach_device_t	*mem;
3141 	sbd_error_t	*err;
3142 	pnode_t		 nodeid;
3143 
3144 	if (!DRMACH_IS_MEM_ID(id))
3145 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3146 	mem = id;
3147 
3148 	nodeid = drmach_node_get_dnode(mem->node);
3149 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
3150 		err = DRMACH_INTERNAL_ERROR();
3151 	else {
3152 		uint64_t size;
3153 
3154 		size = mc_get_alignment_mask(nodeid);
3155 		if (size == (uint64_t)-1)
3156 			err = DRMACH_INTERNAL_ERROR();
3157 		else {
3158 			*mask = size - 1;
3159 			err = NULL;
3160 		}
3161 	}
3162 
3163 	return (err);
3164 }
3165 
3166 sbd_error_t *
3167 drmach_mem_get_base_physaddr(drmachid_t id, uint64_t *pa)
3168 {
3169 	sbd_error_t	*err;
3170 	uint_t		 mcreg;
3171 
3172 	err = drmach_read_mc_asr(id, &mcreg);
3173 	if (err == NULL)
3174 		*pa = mc_asr_to_pa(mcreg);
3175 
3176 	return (err);
3177 }
3178 
3179 /*
3180  * Use of this routine after copy/rename will yield incorrect results,
3181  * because the OBP MEMAVAIL property will not correctly reflect the
3182  * programming of the MCs.
3183  */
3184 sbd_error_t *
3185 drmach_mem_get_memlist(drmachid_t id, struct memlist **ml)
3186 {
3187 	drmach_device_t	*mem;
3188 	int		rv, i, rlen, rblks;
3189 	sbd_error_t	*err;
3190 	struct memlist	*mlist;
3191 	struct sf_memunit_regspec *rlist;
3192 
3193 	if (!DRMACH_IS_MEM_ID(id))
3194 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3195 	mem = id;
3196 
3197 	err = drmach_device_get_proplen(mem, "dr-available", &rlen);
3198 	if (err)
3199 		return (err);
3200 
3201 	rlist = kmem_zalloc(rlen, KM_SLEEP);
3202 
3203 	err = drmach_device_get_prop(mem, "dr-available", rlist);
3204 	if (err) {
3205 		kmem_free(rlist, rlen);
3206 		return (err);
3207 	}
3208 
3209 	mlist = NULL;
3210 	rblks = rlen / sizeof (struct sf_memunit_regspec);
3211 	for (i = 0; i < rblks; i++) {
3212 		uint64_t	addr, size;
3213 
3214 		addr  = (uint64_t)rlist[i].regspec_addr_hi << 32;
3215 		addr |= (uint64_t)rlist[i].regspec_addr_lo;
3216 		size  = (uint64_t)rlist[i].regspec_size_hi << 32;
3217 		size |= (uint64_t)rlist[i].regspec_size_lo;
3218 
3219 		mlist = memlist_add_span(mlist, addr, size);
3220 	}
3221 
3222 	kmem_free(rlist, rlen);
3223 
3224 	/*
3225 	 * Make sure the incoming memlist doesn't already
3226 	 * intersect with what's present in the system (phys_install).
3227 	 */
3228 	memlist_read_lock();
3229 	rv = memlist_intersect(phys_install, mlist);
3230 	memlist_read_unlock();
3231 	if (rv) {
3232 #ifdef DEBUG
3233 		DRMACH_PR("OBP derived memlist intersects"
3234 			" with phys_install\n");
3235 		memlist_dump(mlist);
3236 
3237 		DRMACH_PR("phys_install memlist:\n");
3238 		memlist_dump(phys_install);
3239 #endif
3240 
3241 		memlist_delete(mlist);
3242 		return (DRMACH_INTERNAL_ERROR());
3243 	}
3244 
3245 #ifdef DEBUG
3246 	DRMACH_PR("OBP derived memlist:");
3247 	memlist_dump(mlist);
3248 #endif
3249 
3250 	*ml = mlist;
3251 	return (NULL);
3252 }
3253 
3254 sbd_error_t *
3255 drmach_mem_get_size(drmachid_t id, uint64_t *bytes)
3256 {
3257 	drmach_device_t	*mem;
3258 	pda_handle_t	ph;
3259 	pgcnt_t		npages;
3260 
3261 	if (!DRMACH_IS_MEM_ID(id))
3262 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3263 	mem = id;
3264 
3265 	ph = drmach_pda_open();
3266 	if (ph == NULL)
3267 		return (DRMACH_INTERNAL_ERROR());
3268 
3269 	npages = pda_get_mem_size(ph, mem->bp->bnum);
3270 	*bytes = (uint64_t)npages << PAGESHIFT;
3271 
3272 	pda_close(ph);
3273 	return (NULL);
3274 }
3275 
3276 sbd_error_t *
3277 drmach_mem_get_slice_size(drmachid_t id, uint64_t *bytes)
3278 {
3279 	if (!DRMACH_IS_MEM_ID(id))
3280 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3281 
3282 	*bytes = mc_get_mem_alignment();
3283 	return (NULL);
3284 }
3285 
3286 /* field debugging tool */
3287 processorid_t drmach_mem_cpu_affinity_nail = 0;
3288 
3289 processorid_t
3290 drmach_mem_cpu_affinity(drmachid_t id)
3291 {
3292 	drmach_device_t	*mp;
3293 	drmach_board_t	*bp;
3294 	processorid_t	 cpuid;
3295 
3296 	if (!DRMACH_IS_MEM_ID(id))
3297 		return (CPU_CURRENT);
3298 
3299 	if (drmach_mem_cpu_affinity_nail) {
3300 		cpuid = drmach_mem_cpu_affinity_nail;
3301 
3302 		if (cpuid < 0 || cpuid > NCPU)
3303 			return (CPU_CURRENT);
3304 
3305 		mutex_enter(&cpu_lock);
3306 		if (cpu[cpuid] == NULL || !CPU_ACTIVE(cpu[cpuid]))
3307 			cpuid = CPU_CURRENT;
3308 		mutex_exit(&cpu_lock);
3309 
3310 		return (cpuid);
3311 	}
3312 
3313 	/* try to choose a proc on the target board */
3314 	mp = id;
3315 	bp = mp->bp;
3316 	if (bp->devices) {
3317 		int		rv;
3318 		int		d_idx;
3319 		drmachid_t	d_id;
3320 
3321 		rv = drmach_array_first(bp->devices, &d_idx, &d_id);
3322 		while (rv == 0) {
3323 			if (DRMACH_IS_CPU_ID(d_id)) {
3324 				cpuid = drmach_cpu_calc_id(d_id);
3325 
3326 				mutex_enter(&cpu_lock);
3327 				if (cpu[cpuid] && CPU_ACTIVE(cpu[cpuid])) {
3328 					mutex_exit(&cpu_lock);
3329 					DRMACH_PR("drmach_mem_cpu_affinity: "
3330 					    "selected cpuid=%d\n", cpuid);
3331 					return (cpuid);
3332 				} else {
3333 					mutex_exit(&cpu_lock);
3334 				}
3335 			}
3336 
3337 			rv = drmach_array_next(bp->devices, &d_idx, &d_id);
3338 		}
3339 	}
3340 
3341 	/* otherwise, this proc, wherever it is */
3342 	DRMACH_PR("drmach_mem_cpu_affinity: using default CPU_CURRENT\n");
3343 
3344 	return (CPU_CURRENT);
3345 }
3346 
3347 static sbd_error_t *
3348 drmach_mem_release(drmachid_t id)
3349 {
3350 	if (!DRMACH_IS_MEM_ID(id))
3351 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3352 	return (NULL);
3353 }
3354 
3355 static sbd_error_t *
3356 drmach_mem_status(drmachid_t id, drmach_status_t *stat)
3357 {
3358 	drmach_device_t *dp;
3359 	sbd_error_t	*err;
3360 	uint64_t	 pa, slice_size;
3361 	struct memlist	*ml;
3362 
3363 	ASSERT(DRMACH_IS_MEM_ID(id));
3364 	dp = id;
3365 
3366 	/* get starting physical address of target memory */
3367 	err = drmach_mem_get_base_physaddr(id, &pa);
3368 	if (err)
3369 		return (err);
3370 
3371 	/* round down to slice boundary */
3372 	slice_size = mc_get_mem_alignment();
3373 	pa &= ~ (slice_size - 1);
3374 
3375 	/* stop at first span that is in slice */
3376 	memlist_read_lock();
3377 	for (ml = phys_install; ml; ml = ml->next)
3378 		if (ml->address >= pa && ml->address < pa + slice_size)
3379 			break;
3380 	memlist_read_unlock();
3381 
3382 	stat->assigned = dp->bp->assigned;
3383 	stat->powered = dp->bp->powered;
3384 	stat->configured = (ml != NULL);
3385 	stat->busy = dp->busy;
3386 	(void) strncpy(stat->type, dp->type, sizeof (stat->type));
3387 	stat->info[0] = '\0';
3388 
3389 	return (NULL);
3390 }
3391 
3392 static int
3393 drmach_detach_board(void *arg)
3394 {
3395 	cpuset_t	cset;
3396 	int		retval;
3397 	drmach_board_t	*bp = (drmach_board_t *)arg;
3398 
3399 	cset = cpu_ready_set;
3400 	promsafe_xc_attention(cset);
3401 
3402 	retval = prom_starfire_rm_brd(bp->bnum);
3403 
3404 	xc_dismissed(cset);
3405 
3406 	return (retval);
3407 }
3408 
3409 sbd_error_t *
3410 drmach_board_deprobe(drmachid_t id)
3411 {
3412 	drmach_board_t	*bp;
3413 	int		 retval;
3414 
3415 	if (!DRMACH_IS_BOARD_ID(id))
3416 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3417 	bp = id;
3418 
3419 	cmn_err(CE_CONT, "DR: PROM detach board %d\n", bp->bnum);
3420 
3421 	retval = prom_tree_update(drmach_detach_board, bp);
3422 
3423 	if (retval == 0)
3424 		return (NULL);
3425 	else {
3426 		cmn_err(CE_WARN, "prom error: prom_starfire_rm_brd(%d) "
3427 			"returned %d", bp->bnum, retval);
3428 		return (drerr_new(1, ESTF_DEPROBE, "%s", bp->cm.name));
3429 	}
3430 }
3431 
3432 /*ARGSUSED*/
3433 static sbd_error_t *
3434 drmach_pt_juggle_bootproc(drmachid_t id, drmach_opts_t *opts)
3435 {
3436 	drmach_device_t	*cpu;
3437 	sbd_error_t	*err;
3438 
3439 	if (!DRMACH_IS_CPU_ID(id))
3440 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3441 	cpu = id;
3442 
3443 	mutex_enter(&cpu_lock);
3444 
3445 	err = drmach_cpu_juggle_bootproc(cpu);
3446 
3447 	mutex_exit(&cpu_lock);
3448 
3449 	return (err);
3450 }
3451 
3452 /*ARGSUSED*/
3453 static sbd_error_t *
3454 drmach_pt_dump_pdainfo(drmachid_t id, drmach_opts_t *opts)
3455 {
3456 	drmach_board_t	*bp;
3457 	int		board;
3458 	int		i;
3459 	pda_handle_t	ph;
3460 	board_desc_t	*bdesc;
3461 
3462 	if (!DRMACH_IS_BOARD_ID(id))
3463 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3464 	bp = id;
3465 	board = bp->bnum;
3466 
3467 	ph = drmach_pda_open();
3468 	if (ph == NULL)
3469 		return (DRMACH_INTERNAL_ERROR());
3470 
3471 	if (pda_board_present(ph, board) == 0) {
3472 		cmn_err(CE_CONT, "board %d is MISSING\n", board);
3473 		pda_close(ph);
3474 		return (DRMACH_INTERNAL_ERROR());
3475 	}
3476 
3477 	cmn_err(CE_CONT, "board %d is PRESENT\n", board);
3478 
3479 	bdesc = (board_desc_t *)pda_get_board_info(ph, board);
3480 	if (bdesc == NULL) {
3481 		cmn_err(CE_CONT,
3482 			"no board descriptor found for board %d\n",
3483 			board);
3484 		pda_close(ph);
3485 		return (DRMACH_INTERNAL_ERROR());
3486 	}
3487 
3488 	/* make sure definition in platmod is in sync with pda */
3489 	ASSERT(MAX_PROCMODS == MAX_CPU_UNITS_PER_BOARD);
3490 
3491 	for (i = 0; i < MAX_PROCMODS; i++) {
3492 		if (BDA_NBL(bdesc->bda_proc, i) == BDAN_GOOD)
3493 			cmn_err(CE_CONT,
3494 				"proc %d.%d PRESENT\n", board, i);
3495 		else
3496 			cmn_err(CE_CONT,
3497 				"proc %d.%d MISSING\n", board, i);
3498 	}
3499 
3500 	for (i = 0; i < MAX_MGROUPS; i++) {
3501 		if (BDA_NBL(bdesc->bda_mgroup, i) == BDAN_GOOD)
3502 			cmn_err(CE_CONT,
3503 				"mgroup %d.%d PRESENT\n", board, i);
3504 		else
3505 			cmn_err(CE_CONT,
3506 				"mgroup %d.%d MISSING\n", board, i);
3507 	}
3508 
3509 	/* make sure definition in platmod is in sync with pda */
3510 	ASSERT(MAX_IOCS == MAX_IO_UNITS_PER_BOARD);
3511 
3512 	for (i = 0; i < MAX_IOCS; i++) {
3513 		int	s;
3514 
3515 		if (BDA_NBL(bdesc->bda_ioc, i) == BDAN_GOOD) {
3516 			cmn_err(CE_CONT,
3517 				"ioc %d.%d PRESENT\n", board, i);
3518 			for (s = 0; s < MAX_SLOTS_PER_IOC; s++) {
3519 				if (BDA_NBL(bdesc->bda_ios[i], s) != BDAN_GOOD)
3520 					continue;
3521 				cmn_err(CE_CONT,
3522 					"..scard %d.%d.%d PRESENT\n",
3523 					board, i, s);
3524 			}
3525 		} else {
3526 			cmn_err(CE_CONT,
3527 				"ioc %d.%d MISSING\n",
3528 				board, i);
3529 		}
3530 	}
3531 
3532 	cmn_err(CE_CONT,
3533 		"board %d memsize = %d pages\n",
3534 		board, pda_get_mem_size(ph, board));
3535 
3536 	pda_close(ph);
3537 
3538 	return (NULL);
3539 }
3540 
3541 /*ARGSUSED*/
3542 sbd_error_t *
3543 drmach_pt_readmem(drmachid_t id, drmach_opts_t *opts)
3544 {
3545 	struct memlist	*ml;
3546 	uint64_t	src_pa;
3547 	uint64_t	dst_pa;
3548 	uint64_t	dst;
3549 
3550 	dst_pa = va_to_pa(&dst);
3551 
3552 	memlist_read_lock();
3553 	for (ml = phys_install; ml; ml = ml->next) {
3554 		uint64_t	nbytes;
3555 
3556 		src_pa = ml->address;
3557 		nbytes = ml->size;
3558 
3559 		while (nbytes != 0ull) {
3560 
3561 			/* copy 32 bytes at arc_pa to dst_pa */
3562 			bcopy32_il(src_pa, dst_pa);
3563 
3564 			/* increment by 32 bytes */
3565 			src_pa += (4 * sizeof (uint64_t));
3566 
3567 			/* decrement by 32 bytes */
3568 			nbytes -= (4 * sizeof (uint64_t));
3569 		}
3570 	}
3571 	memlist_read_unlock();
3572 
3573 	return (NULL);
3574 }
3575 
3576 static struct {
3577 	const char	*name;
3578 	sbd_error_t	*(*handler)(drmachid_t id, drmach_opts_t *opts);
3579 } drmach_pt_arr[] = {
3580 	{ "juggle",		drmach_pt_juggle_bootproc	},
3581 	{ "pda",		drmach_pt_dump_pdainfo		},
3582 	{ "readmem",		drmach_pt_readmem		},
3583 
3584 	/* the following line must always be last */
3585 	{ NULL,			NULL				}
3586 };
3587 
3588 /*ARGSUSED*/
3589 sbd_error_t *
3590 drmach_passthru(drmachid_t id, drmach_opts_t *opts)
3591 {
3592 	int		i;
3593 	sbd_error_t	*err;
3594 
3595 	i = 0;
3596 	while (drmach_pt_arr[i].name != NULL) {
3597 		int len = strlen(drmach_pt_arr[i].name);
3598 
3599 		if (strncmp(drmach_pt_arr[i].name, opts->copts, len) == 0)
3600 			break;
3601 
3602 		i += 1;
3603 	}
3604 
3605 	if (drmach_pt_arr[i].name == NULL)
3606 		err = drerr_new(0, ESTF_UNKPTCMD, opts->copts);
3607 	else
3608 		err = (*drmach_pt_arr[i].handler)(id, opts);
3609 
3610 	return (err);
3611 }
3612 
3613 sbd_error_t *
3614 drmach_release(drmachid_t id)
3615 {
3616 	drmach_common_t *cp;
3617 	if (!DRMACH_IS_DEVICE_ID(id))
3618 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3619 	cp = id;
3620 
3621 	return (cp->release(id));
3622 }
3623 
3624 sbd_error_t *
3625 drmach_status(drmachid_t id, drmach_status_t *stat)
3626 {
3627 	drmach_common_t *cp;
3628 
3629 	if (!DRMACH_IS_ID(id))
3630 		return (drerr_new(0, ESTF_NOTID, NULL));
3631 	cp = id;
3632 
3633 	return (cp->status(id, stat));
3634 }
3635 
3636 sbd_error_t *
3637 drmach_unconfigure(drmachid_t id, int flags)
3638 {
3639 	drmach_device_t	*dp;
3640 	pnode_t		 nodeid;
3641 	dev_info_t	*dip, *fdip = NULL;
3642 
3643 	if (!DRMACH_IS_DEVICE_ID(id))
3644 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3645 
3646 	dp = id;
3647 
3648 	nodeid = drmach_node_get_dnode(dp->node);
3649 	if (nodeid == OBP_NONODE)
3650 		return (DRMACH_INTERNAL_ERROR());
3651 
3652 	dip = e_ddi_nodeid_to_dip(nodeid);
3653 	if (dip == NULL)
3654 		return (NULL);
3655 
3656 	/*
3657 	 * Branch already held, so hold acquired in
3658 	 * e_ddi_nodeid_to_dip() can be released
3659 	 */
3660 	ddi_release_devi(dip);
3661 
3662 	if (flags & DEVI_BRANCH_DESTROY)
3663 		flags |= DEVI_BRANCH_EVENT;
3664 
3665 	/*
3666 	 * Force flag is no longer necessary. See starcat/io/drmach.c
3667 	 * for details.
3668 	 */
3669 	ASSERT(e_ddi_branch_held(dip));
3670 	if (e_ddi_branch_unconfigure(dip, &fdip, flags)) {
3671 		sbd_error_t	*err;
3672 		char		*path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3673 
3674 		/*
3675 		 * If non-NULL, fdip is returned held and must be released.
3676 		 */
3677 		if (fdip != NULL) {
3678 			(void) ddi_pathname(fdip, path);
3679 			ndi_rele_devi(fdip);
3680 		} else {
3681 			(void) ddi_pathname(dip, path);
3682 		}
3683 
3684 		err = drerr_new(1, ESTF_DRVFAIL, path);
3685 
3686 		kmem_free(path, MAXPATHLEN);
3687 
3688 		return (err);
3689 	}
3690 
3691 	return (NULL);
3692 }
3693 
3694 /*
3695  * drmach interfaces to legacy Starfire platmod logic
3696  * linkage via runtime symbol look up, called from plat_cpu_power*
3697  */
3698 
3699 /*
3700  * Start up a cpu.  It is possible that we're attempting to restart
3701  * the cpu after an UNCONFIGURE in which case the cpu will be
3702  * spinning in its cache.  So, all we have to do is wakeup him up.
3703  * Under normal circumstances the cpu will be coming from a previous
3704  * CONNECT and thus will be spinning in OBP.  In both cases, the
3705  * startup sequence is the same.
3706  */
3707 int
3708 drmach_cpu_poweron(struct cpu *cp)
3709 {
3710 	DRMACH_PR("drmach_cpu_poweron: starting cpuid %d\n", cp->cpu_id);
3711 
3712 	ASSERT(MUTEX_HELD(&cpu_lock));
3713 
3714 	if (drmach_cpu_start(cp) != 0)
3715 		return (EBUSY);
3716 	else
3717 		return (0);
3718 }
3719 
3720 int
3721 drmach_cpu_poweroff(struct cpu *cp)
3722 {
3723 	int		ntries, cnt;
3724 	processorid_t	cpuid = cp->cpu_id;
3725 	void		drmach_cpu_shutdown_self(void);
3726 
3727 	DRMACH_PR("drmach_cpu_poweroff: stopping cpuid %d\n", cp->cpu_id);
3728 
3729 	ASSERT(MUTEX_HELD(&cpu_lock));
3730 
3731 	/*
3732 	 * Capture all CPUs (except for detaching proc) to prevent
3733 	 * crosscalls to the detaching proc until it has cleared its
3734 	 * bit in cpu_ready_set.
3735 	 *
3736 	 * The CPU's remain paused and the prom_mutex is known to be free.
3737 	 * This prevents the x-trap victim from blocking when doing prom
3738 	 * IEEE-1275 calls at a high PIL level.
3739 	 */
3740 	promsafe_pause_cpus();
3741 
3742 	/*
3743 	 * Quiesce interrupts on the target CPU. We do this by setting
3744 	 * the CPU 'not ready'- (i.e. removing the CPU from cpu_ready_set) to
3745 	 * prevent it from receiving cross calls and cross traps.
3746 	 * This prevents the processor from receiving any new soft interrupts.
3747 	 */
3748 	mp_cpu_quiesce(cp);
3749 
3750 	/* setup xt_mb, will be cleared by drmach_shutdown_asm when ready */
3751 	drmach_xt_mb[cpuid] = 0x80;
3752 
3753 	xt_one_unchecked(cpuid, (xcfunc_t *)idle_stop_xcall,
3754 		(uint64_t)drmach_cpu_shutdown_self, NULL);
3755 
3756 	ntries = drmach_cpu_ntries;
3757 	cnt = 0;
3758 	while (drmach_xt_mb[cpuid] && ntries) {
3759 		DELAY(drmach_cpu_delay);
3760 		ntries--;
3761 		cnt++;
3762 	}
3763 
3764 	drmach_xt_mb[cpuid] = 0;	/* steal the cache line back */
3765 
3766 	start_cpus();
3767 
3768 	DRMACH_PR("waited %d out of %d tries for "
3769 		"drmach_cpu_shutdown_self on cpu%d",
3770 		drmach_cpu_ntries - ntries, drmach_cpu_ntries, cp->cpu_id);
3771 
3772 	drmach_cpu_obp_detach(cpuid);
3773 
3774 	CPU_SIGNATURE(OS_SIG, SIGST_DETACHED, SIGSUBST_NULL, cpuid);
3775 
3776 	return (0);
3777 }
3778 
3779 /*ARGSUSED*/
3780 int
3781 drmach_verify_sr(dev_info_t *dip, int sflag)
3782 {
3783 	return (0);
3784 }
3785 
3786 void
3787 drmach_suspend_last(void)
3788 {
3789 }
3790 
3791 void
3792 drmach_resume_first(void)
3793 {
3794 }
3795 
3796 /*
3797  * Log a DR sysevent.
3798  * Return value: 0 success, non-zero failure.
3799  */
3800 int
3801 drmach_log_sysevent(int board, char *hint, int flag, int verbose)
3802 {
3803 	sysevent_t			*ev;
3804 	sysevent_id_t			eid;
3805 	int				rv, km_flag;
3806 	sysevent_value_t		evnt_val;
3807 	sysevent_attr_list_t		*evnt_attr_list = NULL;
3808 	char				attach_pnt[MAXNAMELEN];
3809 
3810 	km_flag = (flag == SE_SLEEP) ? KM_SLEEP : KM_NOSLEEP;
3811 	attach_pnt[0] = '\0';
3812 	if (drmach_board_name(board, attach_pnt, MAXNAMELEN)) {
3813 		rv = -1;
3814 		goto logexit;
3815 	}
3816 	if (verbose)
3817 		DRMACH_PR("drmach_log_sysevent: %s %s, flag: %d, verbose: %d\n",
3818 			    attach_pnt, hint, flag, verbose);
3819 
3820 	if ((ev = sysevent_alloc(EC_DR, ESC_DR_AP_STATE_CHANGE,
3821 				    SUNW_KERN_PUB"dr", km_flag)) == NULL) {
3822 		rv = -2;
3823 		goto logexit;
3824 	}
3825 	evnt_val.value_type = SE_DATA_TYPE_STRING;
3826 	evnt_val.value.sv_string = attach_pnt;
3827 	if ((rv = sysevent_add_attr(&evnt_attr_list, DR_AP_ID,
3828 				    &evnt_val, km_flag)) != 0)
3829 		goto logexit;
3830 
3831 	evnt_val.value_type = SE_DATA_TYPE_STRING;
3832 	evnt_val.value.sv_string = hint;
3833 	if ((rv = sysevent_add_attr(&evnt_attr_list, DR_HINT,
3834 				    &evnt_val, km_flag)) != 0) {
3835 		sysevent_free_attr(evnt_attr_list);
3836 		goto logexit;
3837 	}
3838 
3839 	(void) sysevent_attach_attributes(ev, evnt_attr_list);
3840 
3841 	/*
3842 	 * Log the event but do not sleep waiting for its
3843 	 * delivery. This provides insulation from syseventd.
3844 	 */
3845 	rv = log_sysevent(ev, SE_NOSLEEP, &eid);
3846 
3847 logexit:
3848 	if (ev)
3849 		sysevent_free(ev);
3850 	if ((rv != 0) && verbose)
3851 		cmn_err(CE_WARN,
3852 			    "drmach_log_sysevent failed (rv %d) for %s  %s\n",
3853 			    rv, attach_pnt, hint);
3854 
3855 	return (rv);
3856 }
3857 
3858 /*ARGSUSED*/
3859 int
3860 drmach_allow_memrange_modify(drmachid_t id)
3861 {
3862 	return (1);	/* TRUE */
3863 }
3864