xref: /titanic_50/usr/src/uts/sun4u/starfire/io/drmach.c (revision b3700b074e637f8c6991b70754c88a2cfffb246b)
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 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  * Copyright (c) 2011 Bayard G. Bell. All rights reserved.
26  */
27 
28 #include <sys/debug.h>
29 #include <sys/types.h>
30 #include <sys/varargs.h>
31 #include <sys/errno.h>
32 #include <sys/cred.h>
33 #include <sys/dditypes.h>
34 #include <sys/devops.h>
35 #include <sys/modctl.h>
36 #include <sys/poll.h>
37 #include <sys/conf.h>
38 #include <sys/ddi.h>
39 #include <sys/sunddi.h>
40 #include <sys/sunndi.h>
41 #include <sys/ndi_impldefs.h>
42 #include <sys/stat.h>
43 #include <sys/kmem.h>
44 #include <sys/vmem.h>
45 #include <sys/processor.h>
46 #include <sys/spitregs.h>
47 #include <sys/cpuvar.h>
48 #include <sys/cpupart.h>
49 #include <sys/mem_config.h>
50 #include <sys/ddi_impldefs.h>
51 #include <sys/systm.h>
52 #include <sys/machsystm.h>
53 #include <sys/autoconf.h>
54 #include <sys/cmn_err.h>
55 #include <sys/sysmacros.h>
56 #include <sys/x_call.h>
57 #include <sys/promif.h>
58 #include <sys/prom_plat.h>
59 #include <sys/membar.h>
60 #include <vm/seg_kmem.h>
61 #include <sys/mem_cage.h>
62 #include <sys/stack.h>
63 #include <sys/archsystm.h>
64 #include <vm/hat_sfmmu.h>
65 #include <sys/pte.h>
66 #include <sys/mmu.h>
67 #include <sys/cpu_module.h>
68 #include <sys/obpdefs.h>
69 #include <sys/note.h>
70 
71 #include <sys/starfire.h>	/* plat_max_... decls */
72 #include <sys/cvc.h>
73 #include <sys/cpu_sgnblk_defs.h>
74 #include <sys/drmach.h>
75 #include <sys/dr_util.h>
76 #include <sys/pda.h>
77 
78 #include <sys/sysevent.h>
79 #include <sys/sysevent/dr.h>
80 #include <sys/sysevent/eventdefs.h>
81 
82 
83 extern void		bcopy32_il(uint64_t, uint64_t);
84 extern void		flush_ecache_il(
85 				uint64_t physaddr, int size, int linesz);
86 extern uint_t		ldphysio_il(uint64_t physaddr);
87 extern void		stphysio_il(uint64_t physaddr, uint_t value);
88 
89 extern uint64_t		mc_get_mem_alignment(void);
90 extern uint64_t		mc_get_asr_addr(pnode_t);
91 extern uint64_t		mc_get_idle_addr(pnode_t);
92 extern uint64_t		mc_get_alignment_mask(pnode_t);
93 extern int		mc_read_asr(pnode_t, uint_t *);
94 extern int		mc_write_asr(pnode_t, uint_t);
95 extern uint64_t		mc_asr_to_pa(uint_t);
96 extern uint_t		mc_pa_to_asr(uint_t, uint64_t);
97 
98 extern int		pc_madr_add(int, int, int, int);
99 
100 typedef struct {
101 	struct drmach_node	*node;
102 	void			*data;
103 } drmach_node_walk_args_t;
104 
105 typedef struct drmach_node {
106 	void		*here;
107 
108 	pnode_t		 (*get_dnode)(struct drmach_node *node);
109 	int		 (*walk)(struct drmach_node *node, void *data,
110 				int (*cb)(drmach_node_walk_args_t *args));
111 } drmach_node_t;
112 
113 typedef struct {
114 	int		 min_index;
115 	int		 max_index;
116 	int		 arr_sz;
117 	drmachid_t	*arr;
118 } drmach_array_t;
119 
120 typedef struct {
121 	void		*isa;
122 
123 	sbd_error_t	*(*release)(drmachid_t);
124 	sbd_error_t	*(*status)(drmachid_t, drmach_status_t *);
125 
126 	char		 name[MAXNAMELEN];
127 } drmach_common_t;
128 
129 typedef struct {
130 	drmach_common_t	 cm;
131 	int		 bnum;
132 	int		 assigned;
133 	int		 powered;
134 	int		 connect_cpuid;
135 	int		 cond;
136 	drmach_node_t	*tree;
137 	drmach_array_t	*devices;
138 } drmach_board_t;
139 
140 typedef struct {
141 	drmach_common_t	 cm;
142 	drmach_board_t	*bp;
143 	int		 unum;
144 	int		 busy;
145 	int		 powered;
146 	const char	*type;
147 	drmach_node_t	*node;
148 } drmach_device_t;
149 
150 typedef struct {
151 	int		 flags;
152 	drmach_device_t	*dp;
153 	sbd_error_t	*err;
154 	dev_info_t	*dip;
155 } drmach_config_args_t;
156 
157 typedef struct {
158 	uint64_t	 idle_addr;
159 	drmach_device_t	*mem;
160 } drmach_mc_idle_script_t;
161 
162 typedef struct {
163 	uint64_t	masr_addr;
164 	uint_t		masr;
165 	uint_t		_filler;
166 } drmach_rename_script_t;
167 
168 typedef struct {
169 	void		(*run)(void *arg);
170 	caddr_t		data;
171 	pda_handle_t	*ph;
172 	struct memlist	*c_ml;
173 	uint64_t	s_copybasepa;
174 	uint64_t	t_copybasepa;
175 	drmach_device_t	*restless_mc;	/* diagnostic output */
176 } drmach_copy_rename_program_t;
177 
178 typedef enum {
179 	DO_IDLE,
180 	DO_UNIDLE,
181 	DO_PAUSE,
182 	DO_UNPAUSE
183 } drmach_iopc_op_t;
184 
185 typedef struct {
186 	drmach_board_t	*obj;
187 	int		 ndevs;
188 	void		*a;
189 	sbd_error_t	*(*found)(void *a, const char *, int, drmachid_t);
190 	sbd_error_t	*err;
191 } drmach_board_cb_data_t;
192 
193 static caddr_t		 drmach_shutdown_va;
194 
195 static int		 drmach_initialized;
196 static drmach_array_t	*drmach_boards;
197 
198 static int		 drmach_cpu_delay = 100;
199 static int		 drmach_cpu_ntries = 50000;
200 
201 volatile uchar_t	*drmach_xt_mb;
202 
203 /*
204  * Do not change the drmach_shutdown_mbox structure without
205  * considering the drmach_shutdown_asm assembly language code.
206  */
207 struct drmach_shutdown_mbox {
208 	uint64_t	estack;
209 	uint64_t	flushaddr;
210 	int		size;
211 	int		linesize;
212 	uint64_t	physaddr;
213 };
214 struct drmach_shutdown_mbox	*drmach_shutdown_asm_mbox;
215 
216 static int		drmach_fini(void);
217 static sbd_error_t	*drmach_device_new(drmach_node_t *,
218 				drmach_board_t *, drmach_device_t **);
219 static sbd_error_t	*drmach_cpu_new(drmach_device_t *);
220 static sbd_error_t	*drmach_mem_new(drmach_device_t *);
221 static sbd_error_t	*drmach_io_new(drmach_device_t *);
222 static sbd_error_t	*drmach_board_release(drmachid_t);
223 static sbd_error_t	*drmach_board_status(drmachid_t, drmach_status_t *);
224 static sbd_error_t	*drmach_cpu_release(drmachid_t);
225 static sbd_error_t	*drmach_cpu_status(drmachid_t, drmach_status_t *);
226 static sbd_error_t	*drmach_io_release(drmachid_t);
227 static sbd_error_t	*drmach_io_status(drmachid_t, drmach_status_t *);
228 static sbd_error_t	*drmach_mem_release(drmachid_t);
229 static sbd_error_t	*drmach_mem_status(drmachid_t, drmach_status_t *);
230 
231 extern struct cpu	*SIGBCPU;
232 
233 #ifdef DEBUG
234 
235 #define	DRMACH_PR		if (drmach_debug) printf
236 int drmach_debug = 0;		 /* set to non-zero to enable debug messages */
237 #else
238 
239 #define	DRMACH_PR		_NOTE(CONSTANTCONDITION) if (0) printf
240 #endif /* DEBUG */
241 
242 #define	DRMACH_OBJ(id)		((drmach_common_t *)id)
243 
244 #define	DRMACH_IS_BOARD_ID(id)	\
245 	((id != 0) &&		\
246 	(DRMACH_OBJ(id)->isa == (void *)drmach_board_new))
247 
248 #define	DRMACH_IS_CPU_ID(id)	\
249 	((id != 0) &&		\
250 	(DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new))
251 
252 #define	DRMACH_IS_MEM_ID(id)	\
253 	((id != 0) &&		\
254 	(DRMACH_OBJ(id)->isa == (void *)drmach_mem_new))
255 
256 #define	DRMACH_IS_IO_ID(id)	\
257 	((id != 0) &&		\
258 	(DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
259 
260 #define	DRMACH_IS_DEVICE_ID(id)					\
261 	((id != 0) &&						\
262 	(DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new ||	\
263 	    DRMACH_OBJ(id)->isa == (void *)drmach_mem_new ||	\
264 	    DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
265 
266 #define	DRMACH_IS_ID(id)					\
267 	((id != 0) &&						\
268 	(DRMACH_OBJ(id)->isa == (void *)drmach_board_new ||	\
269 	    DRMACH_OBJ(id)->isa == (void *)drmach_cpu_new ||	\
270 	    DRMACH_OBJ(id)->isa == (void *)drmach_mem_new ||	\
271 	    DRMACH_OBJ(id)->isa == (void *)drmach_io_new))
272 
273 #define	DRMACH_CPUID2BNUM(cpuid) \
274 	((cpuid) / MAX_CPU_UNITS_PER_BOARD)
275 
276 #define	DRMACH_INTERNAL_ERROR() \
277 	drerr_new(1, ESTF_INTERNAL, drmach_ie_fmt, __LINE__)
278 static char		*drmach_ie_fmt = "drmach.c %d";
279 
280 static struct {
281 	const char	 *name;
282 	const char	 *type;
283 	sbd_error_t	 *(*new)(drmach_device_t *);
284 } name2type[] = {
285 	{ "SUNW,UltraSPARC",	DRMACH_DEVTYPE_CPU,  drmach_cpu_new },
286 	{ "mem-unit",		DRMACH_DEVTYPE_MEM,  drmach_mem_new },
287 	{ "pci",		DRMACH_DEVTYPE_PCI,  drmach_io_new  },
288 	{ "sbus",		DRMACH_DEVTYPE_SBUS, drmach_io_new  },
289 };
290 
291 /* node types to cleanup when a board is unconfigured */
292 #define	MISC_COUNTER_TIMER_DEVNAME	"counter-timer"
293 #define	MISC_PERF_COUNTER_DEVNAME	"perf-counter"
294 
295 /* utility */
296 #define	MBYTE	(1048576ull)
297 
298 /*
299  * drmach autoconfiguration data structures and interfaces
300  */
301 
302 extern struct mod_ops mod_miscops;
303 
304 static struct modlmisc modlmisc = {
305 	&mod_miscops,
306 	"Sun Enterprise 10000 DR"
307 };
308 
309 static struct modlinkage modlinkage = {
310 	MODREV_1,
311 	(void *)&modlmisc,
312 	NULL
313 };
314 
315 static kmutex_t drmach_i_lock;
316 
317 int
318 _init(void)
319 {
320 	int err;
321 
322 	/* check that we have the correct version of obp */
323 	if (prom_test("SUNW,UE10000,add-brd") != 0) {
324 
325 		cmn_err(CE_WARN, "!OBP/SSP upgrade is required to enable "
326 		    "DR Functionality");
327 
328 		return (-1);
329 	}
330 
331 	mutex_init(&drmach_i_lock, NULL, MUTEX_DRIVER, NULL);
332 
333 	drmach_xt_mb = (uchar_t *)vmem_alloc(static_alloc_arena,
334 	    NCPU * sizeof (uchar_t), VM_SLEEP);
335 	drmach_shutdown_asm_mbox = (struct drmach_shutdown_mbox *)
336 	    vmem_alloc(static_alloc_arena, sizeof (struct drmach_shutdown_mbox),
337 	    VM_SLEEP);
338 
339 	if ((err = mod_install(&modlinkage)) != 0) {
340 		mutex_destroy(&drmach_i_lock);
341 		vmem_free(static_alloc_arena, (void *)drmach_xt_mb,
342 		    NCPU * sizeof (uchar_t));
343 		vmem_free(static_alloc_arena, (void *)drmach_shutdown_asm_mbox,
344 		    sizeof (struct drmach_shutdown_mbox));
345 	}
346 
347 	return (err);
348 }
349 
350 int
351 _fini(void)
352 {
353 	if (drmach_fini())
354 		return (DDI_FAILURE);
355 	else
356 		return (mod_remove(&modlinkage));
357 }
358 
359 int
360 _info(struct modinfo *modinfop)
361 {
362 	return (mod_info(&modlinkage, modinfop));
363 }
364 
365 static pnode_t
366 drmach_node_obp_get_dnode(drmach_node_t *np)
367 {
368 	return ((pnode_t)(uintptr_t)np->here);
369 }
370 
371 static int
372 drmach_node_obp_walk(drmach_node_t *np, void *data,
373 		int (*cb)(drmach_node_walk_args_t *args))
374 {
375 	pnode_t			nodeid;
376 	int			rv;
377 	drmach_node_walk_args_t	args;
378 
379 	/* initialized args structure for callback */
380 	args.node = np;
381 	args.data = data;
382 
383 	nodeid = prom_childnode(prom_rootnode());
384 
385 	/* save our new position with in the tree */
386 	np->here = (void *)(uintptr_t)nodeid;
387 
388 	rv = 0;
389 	while (nodeid != OBP_NONODE) {
390 		rv = (*cb)(&args);
391 		if (rv)
392 			break;
393 
394 		nodeid = prom_nextnode(nodeid);
395 
396 		/* save our new position with in the tree */
397 		np->here = (void *)(uintptr_t)nodeid;
398 	}
399 
400 	return (rv);
401 }
402 
403 static drmach_node_t *
404 drmach_node_new(void)
405 {
406 	drmach_node_t *np;
407 
408 	np = kmem_zalloc(sizeof (drmach_node_t), KM_SLEEP);
409 
410 	np->get_dnode = drmach_node_obp_get_dnode;
411 	np->walk = drmach_node_obp_walk;
412 
413 	return (np);
414 }
415 
416 static void
417 drmach_node_dispose(drmach_node_t *np)
418 {
419 	kmem_free(np, sizeof (*np));
420 }
421 
422 static dev_info_t *
423 drmach_node_get_dip(drmach_node_t *np)
424 {
425 	pnode_t nodeid;
426 
427 	nodeid = np->get_dnode(np);
428 	if (nodeid == OBP_NONODE)
429 		return (NULL);
430 	else {
431 		dev_info_t *dip;
432 
433 		/* The root node doesn't have to be held */
434 		dip = e_ddi_nodeid_to_dip(nodeid);
435 		if (dip) {
436 			/*
437 			 * Branch rooted at dip is already held, so release
438 			 * hold acquired in e_ddi_nodeid_to_dip()
439 			 */
440 			ddi_release_devi(dip);
441 			ASSERT(e_ddi_branch_held(dip));
442 		}
443 
444 		return (dip);
445 	}
446 	/*NOTREACHED*/
447 }
448 
449 static pnode_t
450 drmach_node_get_dnode(drmach_node_t *np)
451 {
452 	return (np->get_dnode(np));
453 }
454 
455 static int
456 drmach_node_walk(drmach_node_t *np, void *param,
457 		int (*cb)(drmach_node_walk_args_t *args))
458 {
459 	return (np->walk(np, param, cb));
460 }
461 
462 static int
463 drmach_node_get_prop(drmach_node_t *np, char *name, void *buf)
464 {
465 	pnode_t	nodeid;
466 	int	rv;
467 
468 	nodeid = np->get_dnode(np);
469 	if (nodeid == OBP_NONODE)
470 		rv = -1;
471 	else if (prom_getproplen(nodeid, (caddr_t)name) < 0)
472 		rv = -1;
473 	else {
474 		(void) prom_getprop(nodeid, (caddr_t)name, (caddr_t)buf);
475 		rv = 0;
476 	}
477 
478 	return (rv);
479 }
480 
481 static int
482 drmach_node_get_proplen(drmach_node_t *np, char *name, int *len)
483 {
484 	pnode_t	 nodeid;
485 	int	 rv;
486 
487 	nodeid = np->get_dnode(np);
488 	if (nodeid == OBP_NONODE)
489 		rv = -1;
490 	else {
491 		*len = prom_getproplen(nodeid, (caddr_t)name);
492 		rv = (*len < 0 ? -1 : 0);
493 	}
494 
495 	return (rv);
496 }
497 
498 static drmachid_t
499 drmach_node_dup(drmach_node_t *np)
500 {
501 	drmach_node_t *dup;
502 
503 	dup = drmach_node_new();
504 	dup->here = np->here;
505 
506 	return (dup);
507 }
508 
509 /*
510  * drmach_array provides convenient array construction, access,
511  * bounds checking and array destruction logic.
512  */
513 
514 static drmach_array_t *
515 drmach_array_new(int min_index, int max_index)
516 {
517 	drmach_array_t *arr;
518 
519 	arr = kmem_zalloc(sizeof (drmach_array_t), KM_SLEEP);
520 
521 	arr->arr_sz = (max_index - min_index + 1) * sizeof (void *);
522 	if (arr->arr_sz > 0) {
523 		arr->min_index = min_index;
524 		arr->max_index = max_index;
525 
526 		arr->arr = kmem_zalloc(arr->arr_sz, KM_SLEEP);
527 		return (arr);
528 	} else {
529 		kmem_free(arr, sizeof (*arr));
530 		return (0);
531 	}
532 }
533 
534 static int
535 drmach_array_set(drmach_array_t *arr, int idx, drmachid_t val)
536 {
537 	if (idx < arr->min_index || idx > arr->max_index)
538 		return (-1);
539 	else {
540 		arr->arr[idx - arr->min_index] = val;
541 		return (0);
542 	}
543 	/*NOTREACHED*/
544 }
545 
546 static int
547 drmach_array_get(drmach_array_t *arr, int idx, drmachid_t *val)
548 {
549 	if (idx < arr->min_index || idx > arr->max_index)
550 		return (-1);
551 	else {
552 		*val = arr->arr[idx - arr->min_index];
553 		return (0);
554 	}
555 	/*NOTREACHED*/
556 }
557 
558 static int
559 drmach_array_first(drmach_array_t *arr, int *idx, drmachid_t *val)
560 {
561 	int rv;
562 
563 	*idx = arr->min_index;
564 	while ((rv = drmach_array_get(arr, *idx, val)) == 0 && *val == NULL)
565 		*idx += 1;
566 
567 	return (rv);
568 }
569 
570 static int
571 drmach_array_next(drmach_array_t *arr, int *idx, drmachid_t *val)
572 {
573 	int rv;
574 
575 	*idx += 1;
576 	while ((rv = drmach_array_get(arr, *idx, val)) == 0 && *val == NULL)
577 		*idx += 1;
578 
579 	return (rv);
580 }
581 
582 static void
583 drmach_array_dispose(drmach_array_t *arr, void (*disposer)(drmachid_t))
584 {
585 	drmachid_t	val;
586 	int		idx;
587 	int		rv;
588 
589 	rv = drmach_array_first(arr, &idx, &val);
590 	while (rv == 0) {
591 		(*disposer)(val);
592 		rv = drmach_array_next(arr, &idx, &val);
593 	}
594 
595 	kmem_free(arr->arr, arr->arr_sz);
596 	kmem_free(arr, sizeof (*arr));
597 }
598 
599 /*ARGSUSED*/
600 static int
601 drmach_prom_select(pnode_t nodeid, void *arg, uint_t flags)
602 {
603 	int			rprop[64];
604 	pnode_t			saved;
605 	drmach_config_args_t	*ap = (drmach_config_args_t *)arg;
606 	drmach_device_t		*dp = ap->dp;
607 	sbd_error_t		*err;
608 
609 	saved = drmach_node_get_dnode(dp->node);
610 
611 	if (nodeid != saved)
612 		return (DDI_FAILURE);
613 
614 	if (saved == OBP_NONODE) {
615 		err = DRMACH_INTERNAL_ERROR();
616 		DRERR_SET_C(&ap->err, &err);
617 		return (DDI_FAILURE);
618 	}
619 
620 	if (prom_getprop(nodeid, OBP_REG, (caddr_t)rprop) <= 0) {
621 		return (DDI_FAILURE);
622 	}
623 
624 	return (DDI_SUCCESS);
625 }
626 
627 /*ARGSUSED*/
628 static void
629 drmach_branch_callback(dev_info_t *rdip, void *arg, uint_t flags)
630 {
631 	drmach_config_args_t	*ap = (drmach_config_args_t *)arg;
632 
633 	ASSERT(ap->dip == NULL);
634 
635 	ap->dip = rdip;
636 }
637 
638 sbd_error_t *
639 drmach_configure(drmachid_t id, int flags)
640 {
641 	drmach_device_t		*dp;
642 	sbd_error_t		*err;
643 	drmach_config_args_t	ca;
644 	devi_branch_t		b = {0};
645 	dev_info_t		*fdip = NULL;
646 
647 	if (!DRMACH_IS_DEVICE_ID(id))
648 		return (drerr_new(0, ESTF_INAPPROP, NULL));
649 	dp = id;
650 
651 	ca.dp = dp;
652 	ca.flags = flags;
653 	ca.err = NULL;		/* will be set if error detected */
654 	ca.dip = NULL;
655 
656 	b.arg = &ca;
657 	b.type = DEVI_BRANCH_PROM;
658 	b.create.prom_branch_select = drmach_prom_select;
659 	b.devi_branch_callback = drmach_branch_callback;
660 
661 	if (e_ddi_branch_create(ddi_root_node(), &b, &fdip,
662 	    DEVI_BRANCH_CHILD | DEVI_BRANCH_CONFIGURE) != 0) {
663 		char *path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
664 
665 		/*
666 		 * If non-NULL, fdip is returned held and must be released.
667 		 */
668 		if (fdip != NULL) {
669 			(void) ddi_pathname(fdip, path);
670 			ddi_release_devi(fdip);
671 		} else if (ca.dip != NULL) {
672 			/* safe to call ddi_pathname as dip already held */
673 			(void) ddi_pathname(ca.dip, path);
674 		} else {
675 			(void) strcpy(path, "<none>");
676 		}
677 
678 		err = drerr_new(1, ESTF_DRVFAIL, path);
679 		DRERR_SET_C(&ca.err, &err);
680 		kmem_free(path, MAXPATHLEN);
681 	}
682 
683 	return (ca.err);
684 }
685 
686 static sbd_error_t *
687 drmach_device_new(drmach_node_t *node,
688 	drmach_board_t *bp, drmach_device_t **dpp)
689 {
690 	int		 i;
691 	int		 rv;
692 	drmach_device_t	*dp;
693 	sbd_error_t	*err;
694 	char		 name[OBP_MAXDRVNAME];
695 
696 	rv = drmach_node_get_prop(node, OBP_NAME, name);
697 	if (rv) {
698 		/* every node is expected to have a name */
699 		err = drerr_new(1, ESTF_GETPROP,
700 		    "PROM Node 0x%x: property %s",
701 		    (uint_t)node->get_dnode(node), OBP_NAME);
702 
703 		return (err);
704 	}
705 
706 	/*
707 	 * The node currently being examined is not listed in the name2type[]
708 	 * array.  In this case, the node is no interest to drmach.  Both
709 	 * dp and err are initialized here to yield nothing (no device or
710 	 * error structure) for this case.
711 	 */
712 	for (i = 0; i < sizeof (name2type) / sizeof (name2type[0]); i++)
713 		if (strcmp(name2type[i].name, name) == 0)
714 			break;
715 
716 	if (i < sizeof (name2type) / sizeof (name2type[0])) {
717 		dp = kmem_zalloc(sizeof (drmach_device_t), KM_SLEEP);
718 
719 		dp->bp = bp;
720 		dp->unum = -1;
721 		dp->node = drmach_node_dup(node);
722 		dp->type = name2type[i].type;
723 
724 		err = (name2type[i].new)(dp);
725 		if (err) {
726 			drmach_node_dispose(node);
727 			kmem_free(dp, sizeof (*dp));
728 			dp = NULL;
729 		}
730 
731 		*dpp = dp;
732 		return (err);
733 	}
734 
735 	/*
736 	 * The node currently being examined is not listed in the name2type[]
737 	 * array.  In this case, the node is no interest to drmach.  Both
738 	 * dp and err are initialized here to yield nothing (no device or
739 	 * error structure) for this case.
740 	 */
741 	*dpp = NULL;
742 	return (NULL);
743 }
744 
745 static void
746 drmach_device_dispose(drmachid_t id)
747 {
748 	drmach_device_t *self = id;
749 
750 	if (self->node)
751 		drmach_node_dispose(self->node);
752 
753 	kmem_free(self, sizeof (*self));
754 }
755 
756 static sbd_error_t *
757 drmach_device_get_prop(drmach_device_t *dp, char *name, void *buf)
758 {
759 	sbd_error_t	*err = NULL;
760 	int		 rv;
761 
762 	rv = drmach_node_get_prop(dp->node, name, buf);
763 	if (rv) {
764 		err = drerr_new(1, ESTF_GETPROP,
765 		    "%s::%s: property %s",
766 		    dp->bp->cm.name, dp->cm.name, name);
767 	}
768 
769 	return (err);
770 }
771 
772 static sbd_error_t *
773 drmach_device_get_proplen(drmach_device_t *dp, char *name, int *len)
774 {
775 	sbd_error_t	*err = NULL;
776 	int		 rv;
777 
778 	rv = drmach_node_get_proplen(dp->node, name, len);
779 	if (rv) {
780 		err = drerr_new(1, ESTF_GETPROPLEN,
781 		    "%s::%s: property %s",
782 		    dp->bp->cm.name, dp->cm.name, name);
783 	}
784 
785 	return (err);
786 }
787 
788 static drmach_board_t *
789 drmach_board_new(int bnum)
790 {
791 	drmach_board_t	*bp;
792 
793 	bp = kmem_zalloc(sizeof (drmach_board_t), KM_SLEEP);
794 
795 	bp->cm.isa = (void *)drmach_board_new;
796 	bp->cm.release = drmach_board_release;
797 	bp->cm.status = drmach_board_status;
798 
799 	(void) drmach_board_name(bnum, bp->cm.name, sizeof (bp->cm.name));
800 
801 	bp->bnum = bnum;
802 	bp->devices = NULL;
803 	bp->connect_cpuid = -1;
804 	bp->tree = drmach_node_new();
805 	bp->assigned = !drmach_initialized;
806 	bp->powered = !drmach_initialized;
807 
808 	(void) drmach_array_set(drmach_boards, bnum, bp);
809 	return (bp);
810 }
811 
812 static void
813 drmach_board_dispose(drmachid_t id)
814 {
815 	drmach_board_t *bp;
816 
817 	ASSERT(DRMACH_IS_BOARD_ID(id));
818 	bp = id;
819 
820 	if (bp->tree)
821 		drmach_node_dispose(bp->tree);
822 
823 	if (bp->devices)
824 		drmach_array_dispose(bp->devices, drmach_device_dispose);
825 
826 	kmem_free(bp, sizeof (*bp));
827 }
828 
829 static sbd_error_t *
830 drmach_board_status(drmachid_t id, drmach_status_t *stat)
831 {
832 	sbd_error_t	*err = NULL;
833 	drmach_board_t	*bp;
834 
835 	if (!DRMACH_IS_BOARD_ID(id))
836 		return (drerr_new(0, ESTF_INAPPROP, NULL));
837 	bp = id;
838 
839 	stat->assigned = bp->assigned;
840 	stat->powered = bp->powered;
841 	stat->busy = 0;			/* assume not busy */
842 	stat->configured = 0;		/* assume not configured */
843 	stat->empty = 0;
844 	stat->cond = bp->cond = SBD_COND_OK;
845 	(void) strncpy(stat->type, "System Brd", sizeof (stat->type));
846 	stat->info[0] = '\0';
847 
848 	if (bp->devices) {
849 		int		 rv;
850 		int		 d_idx;
851 		drmachid_t	 d_id;
852 
853 		rv = drmach_array_first(bp->devices, &d_idx, &d_id);
854 		while (rv == 0) {
855 			drmach_status_t	d_stat;
856 
857 			err = drmach_status(d_id, &d_stat);
858 			if (err)
859 				break;
860 
861 			stat->busy |= d_stat.busy;
862 			stat->configured |= d_stat.configured;
863 
864 			rv = drmach_array_next(bp->devices, &d_idx, &d_id);
865 		}
866 	}
867 
868 	return (err);
869 }
870 
871 /* a simple routine to reduce redundancy of this common logic */
872 static pda_handle_t
873 drmach_pda_open(void)
874 {
875 	pda_handle_t ph;
876 
877 	ph = pda_open();
878 	if (ph == NULL) {
879 		/* catch in debug kernels */
880 		ASSERT(0);
881 		cmn_err(CE_WARN, "pda_open failed");
882 	}
883 
884 	return (ph);
885 }
886 
887 #ifdef DEBUG
888 int drmach_init_break = 0;
889 #endif
890 
891 static int
892 hold_rele_branch(dev_info_t *rdip, void *arg)
893 {
894 	int	i;
895 	int	*holdp = (int *)arg;
896 	char	*name = ddi_node_name(rdip);
897 
898 	/*
899 	 * For Starfire, we must be children of the root devinfo node
900 	 */
901 	ASSERT(ddi_get_parent(rdip) == ddi_root_node());
902 
903 	for (i = 0; i < sizeof (name2type) / sizeof (name2type[0]); i++)
904 		if (strcmp(name2type[i].name, name) == 0)
905 			break;
906 
907 	if (i == sizeof (name2type) / sizeof (name2type[0])) {
908 		/* Not of interest to us */
909 		return (DDI_WALK_PRUNECHILD);
910 	}
911 
912 	if (*holdp) {
913 		ASSERT(!e_ddi_branch_held(rdip));
914 		e_ddi_branch_hold(rdip);
915 	} else {
916 		ASSERT(e_ddi_branch_held(rdip));
917 		e_ddi_branch_rele(rdip);
918 	}
919 
920 	return (DDI_WALK_PRUNECHILD);
921 }
922 
923 static int
924 drmach_init(void)
925 {
926 	pnode_t		nodeid;
927 	dev_info_t	*rdip;
928 	int		hold, circ;
929 
930 #ifdef DEBUG
931 	if (drmach_init_break)
932 		debug_enter("drmach_init: drmach_init_break set\n");
933 #endif
934 	mutex_enter(&drmach_i_lock);
935 	if (drmach_initialized) {
936 		mutex_exit(&drmach_i_lock);
937 		return (0);
938 	}
939 
940 	drmach_boards = drmach_array_new(0, MAX_BOARDS - 1);
941 
942 	nodeid = prom_childnode(prom_rootnode());
943 	do {
944 		int		 bnum;
945 		drmachid_t	 id;
946 
947 		bnum = -1;
948 		(void) prom_getprop(nodeid, OBP_BOARDNUM, (caddr_t)&bnum);
949 		if (bnum == -1)
950 			continue;
951 
952 		if (drmach_array_get(drmach_boards, bnum, &id) == -1) {
953 			cmn_err(CE_WARN, "OBP node 0x%x has"
954 			    " invalid property value, %s=%d",
955 			    nodeid, OBP_BOARDNUM, bnum);
956 
957 			/* clean up */
958 			drmach_array_dispose(
959 			    drmach_boards, drmach_board_dispose);
960 
961 			mutex_exit(&drmach_i_lock);
962 			return (-1);
963 		} else if (id == NULL)
964 			(void) drmach_board_new(bnum);
965 	} while ((nodeid = prom_nextnode(nodeid)) != OBP_NONODE);
966 
967 	drmach_shutdown_va = vmem_alloc(heap_arena, PAGESIZE, VM_SLEEP);
968 
969 	/*
970 	 * Walk immediate children of devinfo root node and hold
971 	 * all devinfo branches of interest.
972 	 */
973 	hold = 1;
974 	rdip = ddi_root_node();
975 
976 	ndi_devi_enter(rdip, &circ);
977 	ddi_walk_devs(ddi_get_child(rdip), hold_rele_branch, &hold);
978 	ndi_devi_exit(rdip, circ);
979 
980 	drmach_initialized = 1;
981 
982 	mutex_exit(&drmach_i_lock);
983 
984 	return (0);
985 }
986 
987 static int
988 drmach_fini(void)
989 {
990 	dev_info_t	*rdip;
991 	int		hold, circ;
992 
993 	if (drmach_initialized) {
994 		int		busy = 0;
995 		int		rv;
996 		int		idx;
997 		drmachid_t	id;
998 
999 		ASSERT(drmach_boards != NULL);
1000 
1001 		rv = drmach_array_first(drmach_boards, &idx, &id);
1002 		while (rv == 0) {
1003 			sbd_error_t	*err;
1004 			drmach_status_t stat;
1005 
1006 			err = drmach_board_status(id, &stat);
1007 			if (err) {
1008 				/* catch in debug kernels */
1009 				ASSERT(0);
1010 				sbd_err_clear(&err);
1011 				busy = 1;
1012 			} else
1013 				busy |= stat.busy;
1014 
1015 			rv = drmach_array_next(drmach_boards, &idx, &id);
1016 		}
1017 
1018 		if (busy)
1019 			return (-1);
1020 
1021 		drmach_array_dispose(drmach_boards, drmach_board_dispose);
1022 		drmach_boards = NULL;
1023 
1024 		vmem_free(heap_arena, drmach_shutdown_va, PAGESIZE);
1025 
1026 		/*
1027 		 * Walk immediate children of the root devinfo node
1028 		 * releasing holds acquired on branches in drmach_init()
1029 		 */
1030 		hold = 0;
1031 		rdip = ddi_root_node();
1032 
1033 		ndi_devi_enter(rdip, &circ);
1034 		ddi_walk_devs(ddi_get_child(rdip), hold_rele_branch, &hold);
1035 		ndi_devi_exit(rdip, circ);
1036 
1037 		mutex_destroy(&drmach_i_lock);
1038 
1039 		drmach_initialized = 0;
1040 	}
1041 	if (drmach_xt_mb != NULL) {
1042 		vmem_free(static_alloc_arena, (void *)drmach_xt_mb,
1043 		    NCPU * sizeof (uchar_t));
1044 	}
1045 	if (drmach_shutdown_asm_mbox != NULL) {
1046 		vmem_free(static_alloc_arena, (void *)drmach_shutdown_asm_mbox,
1047 		    sizeof (struct drmach_shutdown_mbox));
1048 	}
1049 	return (0);
1050 }
1051 
1052 static sbd_error_t *
1053 drmach_get_mc_asr_addr(drmachid_t id, uint64_t *pa)
1054 {
1055 	drmach_device_t	*dp;
1056 	pnode_t		nodeid;
1057 	uint64_t	addr;
1058 
1059 	if (!DRMACH_IS_MEM_ID(id))
1060 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1061 	dp = id;
1062 
1063 	nodeid = drmach_node_get_dnode(dp->node);
1064 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1065 		return (DRMACH_INTERNAL_ERROR());
1066 
1067 	addr = mc_get_asr_addr(nodeid);
1068 	if (addr == (uint64_t)-1)
1069 		return (DRMACH_INTERNAL_ERROR());
1070 
1071 	*pa = addr;
1072 	return (NULL);
1073 }
1074 
1075 static sbd_error_t *
1076 drmach_get_mc_idle_addr(drmachid_t id, uint64_t *pa)
1077 {
1078 	drmach_device_t	*dp;
1079 	pnode_t		nodeid;
1080 	uint64_t	addr;
1081 
1082 	if (!DRMACH_IS_MEM_ID(id))
1083 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1084 	dp = id;
1085 
1086 	nodeid = drmach_node_get_dnode(dp->node);
1087 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1088 		return (DRMACH_INTERNAL_ERROR());
1089 
1090 	addr = mc_get_idle_addr(nodeid);
1091 	if (addr == (uint64_t)-1)
1092 		return (DRMACH_INTERNAL_ERROR());
1093 
1094 	*pa = addr;
1095 	return (NULL);
1096 }
1097 
1098 static sbd_error_t *
1099 drmach_read_mc_asr(drmachid_t id, uint_t *mcregp)
1100 {
1101 	drmach_device_t	*dp;
1102 	pnode_t		 nodeid;
1103 	sbd_error_t	*err;
1104 
1105 	if (!DRMACH_IS_MEM_ID(id))
1106 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1107 	dp = id;
1108 
1109 	nodeid = drmach_node_get_dnode(dp->node);
1110 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1111 		err = DRMACH_INTERNAL_ERROR();
1112 	else if (mc_read_asr(nodeid, mcregp) == -1)
1113 		err = DRMACH_INTERNAL_ERROR();
1114 	else
1115 		err = NULL;
1116 
1117 	return (err);
1118 }
1119 
1120 static sbd_error_t *
1121 drmach_write_mc_asr(drmachid_t id, uint_t mcreg)
1122 {
1123 	drmach_device_t	*dp;
1124 	pnode_t		 nodeid;
1125 	sbd_error_t	*err;
1126 
1127 	if (!DRMACH_IS_MEM_ID(id))
1128 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1129 	dp = id;
1130 
1131 	nodeid = drmach_node_get_dnode(dp->node);
1132 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
1133 		err = DRMACH_INTERNAL_ERROR();
1134 	else if (mc_write_asr(nodeid, mcreg) == -1)
1135 		err = DRMACH_INTERNAL_ERROR();
1136 	else
1137 		err = NULL;
1138 
1139 	return (err);
1140 }
1141 
1142 static sbd_error_t *
1143 drmach_prep_rename_script(drmach_device_t *s_mem, drmach_device_t *t_mem,
1144 	uint64_t t_slice_offset, caddr_t buf, int buflen)
1145 {
1146 	int			i, b, m;
1147 	drmach_mc_idle_script_t	*isp;
1148 	drmach_rename_script_t	*rsp;
1149 	int			s_bd, t_bd;
1150 	uint_t			s_masr, t_masr;
1151 	uint64_t		s_new_basepa, t_new_basepa;
1152 	int			b_idx, rv;
1153 	sbd_error_t		*err;
1154 	drmachid_t		 b_id;
1155 	drmach_board_t		*brd;
1156 
1157 #ifdef DEBUG
1158 	/*
1159 	 * Starfire CPU/MEM/IO boards have only one MC per board.
1160 	 * This function has been coded with that fact in mind.
1161 	 */
1162 	ASSERT(MAX_MEM_UNITS_PER_BOARD == 1);
1163 
1164 	/*
1165 	 * calculate the maximum space that could be consumed,
1166 	 * then verify the available buffer space is adequate.
1167 	 */
1168 	m  = sizeof (drmach_mc_idle_script_t *) * 2; /* two MCs */
1169 	b  = sizeof (drmach_rename_script_t *) * 3 * MAX_CPU_UNITS_PER_BOARD;
1170 	b += sizeof (drmach_rename_script_t *) * 3 * MAX_IO_UNITS_PER_BOARD;
1171 	b *= MAX_BOARDS;
1172 	b += sizeof (drmach_rename_script_t *) * 3;
1173 	b += sizeof (drmach_rename_script_t *) * 1;
1174 	ASSERT(m + b < buflen);
1175 #endif
1176 
1177 	/*
1178 	 * construct an array of MC idle register addresses of
1179 	 * both MCs.  The array is zero terminated -- as expected
1180 	 * by drmach_copy_rename_prog__relocatable().
1181 	 */
1182 	isp = (drmach_mc_idle_script_t *)buf;
1183 
1184 	/* source mc */
1185 	err = drmach_get_mc_idle_addr(s_mem, &isp->idle_addr);
1186 	if (err)
1187 		return (err);
1188 	isp->mem = s_mem;
1189 	isp += 1;
1190 
1191 	/* target mc */
1192 	err = drmach_get_mc_idle_addr(t_mem, &isp->idle_addr);
1193 	if (err)
1194 		return (err);
1195 	isp->mem = t_mem;
1196 	isp += 1;
1197 
1198 	/* terminator */
1199 	isp->idle_addr = 0;
1200 	isp->mem = NULL;
1201 	isp += 1;
1202 
1203 	/* fetch source mc asr register value */
1204 	err = drmach_read_mc_asr(s_mem, &s_masr);
1205 	if (err)
1206 		return (err);
1207 	else if (s_masr & STARFIRE_MC_INTERLEAVE_MASK) {
1208 		return (drerr_new(1, ESTF_INTERBOARD, "%s::%s",
1209 		    s_mem->bp->cm.name, s_mem->cm.name));
1210 	}
1211 
1212 	/* fetch target mc asr register value */
1213 	err = drmach_read_mc_asr(t_mem, &t_masr);
1214 	if (err)
1215 		return (err);
1216 	else if (t_masr & STARFIRE_MC_INTERLEAVE_MASK) {
1217 		return (drerr_new(1, ESTF_INTERBOARD, "%s::%s",
1218 		    t_mem->bp->cm.name, t_mem->cm.name));
1219 	}
1220 
1221 	/* get new source base pa from target's masr */
1222 	s_new_basepa = mc_asr_to_pa(t_masr);
1223 
1224 	/*
1225 	 * remove any existing slice offset to realign
1226 	 * memory with board's slice boundary
1227 	 */
1228 	s_new_basepa &= ~ (mc_get_mem_alignment() - 1);
1229 
1230 	/* get new target base pa from source's masr */
1231 	t_new_basepa  = mc_asr_to_pa(s_masr);
1232 
1233 	/* remove any existing slice offset, then apply new offset */
1234 	t_new_basepa &= ~ (mc_get_mem_alignment() - 1);
1235 	t_new_basepa += t_slice_offset;
1236 
1237 	/* encode new base pa into s_masr.  turn off mem present bit */
1238 	s_masr  = mc_pa_to_asr(s_masr, s_new_basepa);
1239 	s_masr &= ~STARFIRE_MC_MEM_PRESENT_MASK;
1240 
1241 	/* encode new base pa into t_masr.  turn on mem present bit */
1242 	t_masr  = mc_pa_to_asr(t_masr, t_new_basepa);
1243 	t_masr |= STARFIRE_MC_MEM_PRESENT_MASK;
1244 
1245 	/*
1246 	 * Step 0:	Mark source memory as not present.
1247 	 */
1248 	m = 0;
1249 	rsp = (drmach_rename_script_t *)isp;
1250 	err = drmach_get_mc_asr_addr(s_mem, &rsp[m].masr_addr);
1251 	if (err)
1252 		return (err);
1253 	rsp[m].masr = s_masr;
1254 	m++;
1255 
1256 	/*
1257 	 * Step 1:	Write source base address to target MC
1258 	 *		with present bit off.
1259 	 */
1260 	err = drmach_get_mc_asr_addr(t_mem, &rsp[m].masr_addr);
1261 	if (err)
1262 		return (err);
1263 	rsp[m].masr = t_masr & ~STARFIRE_MC_MEM_PRESENT_MASK;
1264 	m++;
1265 
1266 	/*
1267 	 * Step 2:	Now rewrite target reg with present bit on.
1268 	 */
1269 	rsp[m].masr_addr = rsp[m-1].masr_addr;
1270 	rsp[m].masr = t_masr;
1271 	m++;
1272 
1273 	s_bd = s_mem->bp->bnum;
1274 	t_bd = t_mem->bp->bnum;
1275 
1276 	DRMACH_PR("preparing script for CPU and IO units:\n");
1277 
1278 	rv = drmach_array_first(drmach_boards, &b_idx, &b_id);
1279 	if (rv) {
1280 		/* catch this in debug kernels */
1281 		ASSERT(0);
1282 		return (DRMACH_INTERNAL_ERROR());
1283 	}
1284 
1285 	do {
1286 		int			 d_idx;
1287 		drmachid_t		 d_id;
1288 		drmach_device_t		*device;
1289 
1290 		ASSERT(DRMACH_IS_BOARD_ID(b_id));
1291 		brd = b_id;
1292 		b = brd->bnum;
1293 
1294 		/*
1295 		 * Step 3:	Update PC MADR tables for CPUs.
1296 		 */
1297 		if (brd->devices == NULL) {
1298 			/* devices not initialized */
1299 			continue;
1300 		}
1301 
1302 		rv = drmach_array_first(brd->devices, &d_idx, &d_id);
1303 		if (rv) {
1304 			/* must mean no devices on this board */
1305 			break;
1306 		}
1307 
1308 		DRMACH_PR("\t%s\n", brd->cm.name);
1309 
1310 		do {
1311 			ASSERT(DRMACH_IS_DEVICE_ID(d_id));
1312 
1313 			if (!DRMACH_IS_CPU_ID(d_id))
1314 				continue;
1315 
1316 			device = d_id;
1317 			i = device->unum;
1318 
1319 			DRMACH_PR("\t\t%s\n", device->cm.name);
1320 
1321 			/*
1322 			 * Disabled detaching mem node.
1323 			 */
1324 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, s_bd, i);
1325 			rsp[m].masr = s_masr;
1326 			m++;
1327 			/*
1328 			 * Always write masr with present bit
1329 			 * off and then again with it on.
1330 			 */
1331 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, t_bd, i);
1332 			rsp[m].masr = t_masr & ~STARFIRE_MC_MEM_PRESENT_MASK;
1333 			m++;
1334 			rsp[m].masr_addr = rsp[m-1].masr_addr;
1335 			rsp[m].masr = t_masr;
1336 			m++;
1337 
1338 		} while (drmach_array_next(brd->devices, &d_idx, &d_id) == 0);
1339 
1340 		/*
1341 		 * Step 4:	Update PC MADR tables for IOs.
1342 		 */
1343 		rv = drmach_array_first(brd->devices, &d_idx, &d_id);
1344 		/* this worked for previous loop, must work here too */
1345 		ASSERT(rv == 0);
1346 
1347 		do {
1348 			ASSERT(DRMACH_IS_DEVICE_ID(d_id));
1349 
1350 			if (!DRMACH_IS_IO_ID(d_id))
1351 				continue;
1352 
1353 			device = d_id;
1354 			i = device->unum;
1355 
1356 			DRMACH_PR("\t\t%s\n", device->cm.name);
1357 
1358 			/*
1359 			 * Disabled detaching mem node.
1360 			 */
1361 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, s_bd, i+4);
1362 			rsp[m].masr = s_masr;
1363 			m++;
1364 			/*
1365 			 * Always write masr with present bit
1366 			 * off and then again with it on.
1367 			 */
1368 			rsp[m].masr_addr = STARFIRE_PC_MADR_ADDR(b, t_bd, i+4);
1369 			rsp[m].masr = t_masr & ~STARFIRE_MC_MEM_PRESENT_MASK;
1370 			m++;
1371 			rsp[m].masr_addr = rsp[m-1].masr_addr;
1372 			rsp[m].masr = t_masr;
1373 			m++;
1374 
1375 		} while (drmach_array_next(brd->devices, &d_idx, &d_id) == 0);
1376 	} while (drmach_array_next(drmach_boards, &b_idx, &b_id) == 0);
1377 
1378 	/*
1379 	 * Zero masr_addr value indicates the END.
1380 	 */
1381 	rsp[m].masr_addr = 0ull;
1382 	rsp[m].masr = 0;
1383 	DRMACH_PR("number of steps in rename script = %d\n", m);
1384 	m++;
1385 
1386 	/* paranoia */
1387 	ASSERT((caddr_t)&rsp[m] <= buf + buflen);
1388 
1389 #ifdef DEBUG
1390 	{
1391 		int	j;
1392 
1393 		DRMACH_PR("mc idle register address list:");
1394 		isp = (drmach_mc_idle_script_t *)buf;
1395 		DRMACH_PR("source mc idle addr 0x%lx, mem id %p",
1396 		    isp[0].idle_addr, (void *)isp[0].mem);
1397 		DRMACH_PR("target mc idle addr 0x%lx, mem id %p",
1398 		    isp[1].idle_addr, (void *)isp[1].mem);
1399 		ASSERT(isp[2].idle_addr == 0);
1400 
1401 		DRMACH_PR("copy-rename script:");
1402 		for (j = 0; j < m; j++) {
1403 			DRMACH_PR("0x%lx = 0x%08x",
1404 			    rsp[j].masr_addr, rsp[j].masr);
1405 		}
1406 
1407 		DELAY(1000000);
1408 	}
1409 #endif
1410 
1411 	/* return number of bytes consumed */
1412 	b = (caddr_t)&rsp[m] - buf;
1413 	DRMACH_PR("total number of bytes consumed is %d\n", b);
1414 	ASSERT(b <= buflen);
1415 
1416 #ifdef lint
1417 	buflen = buflen;
1418 #endif
1419 
1420 	return (NULL);
1421 }
1422 
1423 /*
1424  * The routine performs the necessary memory COPY and MC adr SWITCH.
1425  * Both operations MUST be at the same "level" so that the stack is
1426  * maintained correctly between the copy and switch.  The switch
1427  * portion implements a caching mechanism to guarantee the code text
1428  * is cached prior to execution.  This is to guard against possible
1429  * memory access while the MC adr's are being modified.
1430  *
1431  * IMPORTANT: The _drmach_copy_rename_end() function must immediately
1432  * follow drmach_copy_rename_prog__relocatable() so that the correct
1433  * "length" of the drmach_copy_rename_prog__relocatable can be
1434  * calculated.  This routine MUST be a LEAF function, i.e. it can
1435  * make NO function calls, primarily for two reasons:
1436  *
1437  *	1. We must keep the stack consistent across the "switch".
1438  *	2. Function calls are compiled to relative offsets, and
1439  *	   we execute this function we'll be executing it from
1440  *	   a copied version in a different area of memory, thus
1441  *	   the relative offsets will be bogus.
1442  *
1443  * Moreover, it must have the "__relocatable" suffix to inform DTrace
1444  * providers (and anything else, for that matter) that this
1445  * function's text is manually relocated elsewhere before it is
1446  * executed.  That is, it cannot be safely instrumented with any
1447  * methodology that is PC-relative.
1448  */
1449 static void
1450 drmach_copy_rename_prog__relocatable(drmach_copy_rename_program_t *prog)
1451 {
1452 	extern void drmach_exec_script_il(drmach_rename_script_t *rsp);
1453 
1454 	drmach_mc_idle_script_t		*isp;
1455 	struct memlist			*ml;
1456 	int				csize;
1457 	int				lnsize;
1458 	uint64_t			caddr;
1459 
1460 	isp = (drmach_mc_idle_script_t *)prog->data;
1461 
1462 	caddr = ecache_flushaddr;
1463 	csize = (cpunodes[CPU->cpu_id].ecache_size << 1);
1464 	lnsize = cpunodes[CPU->cpu_id].ecache_linesize;
1465 
1466 	/*
1467 	 * DO COPY.
1468 	 */
1469 	for (ml = prog->c_ml; ml; ml = ml->ml_next) {
1470 		uint64_t	s_pa, t_pa;
1471 		uint64_t	nbytes;
1472 
1473 		s_pa = prog->s_copybasepa + ml->ml_address;
1474 		t_pa = prog->t_copybasepa + ml->ml_address;
1475 		nbytes = ml->ml_size;
1476 
1477 		while (nbytes != 0ull) {
1478 			/*
1479 			 * This copy does NOT use an ASI
1480 			 * that avoids the Ecache, therefore
1481 			 * the dst_pa addresses may remain
1482 			 * in our Ecache after the dst_pa
1483 			 * has been removed from the system.
1484 			 * A subsequent write-back to memory
1485 			 * will cause an ARB-stop because the
1486 			 * physical address no longer exists
1487 			 * in the system. Therefore we must
1488 			 * flush out local Ecache after we
1489 			 * finish the copy.
1490 			 */
1491 
1492 			/* copy 32 bytes at src_pa to dst_pa */
1493 			bcopy32_il(s_pa, t_pa);
1494 
1495 			/* increment by 32 bytes */
1496 			s_pa += (4 * sizeof (uint64_t));
1497 			t_pa += (4 * sizeof (uint64_t));
1498 
1499 			/* decrement by 32 bytes */
1500 			nbytes -= (4 * sizeof (uint64_t));
1501 		}
1502 	}
1503 
1504 	/*
1505 	 * Since bcopy32_il() does NOT use an ASI to bypass
1506 	 * the Ecache, we need to flush our Ecache after
1507 	 * the copy is complete.
1508 	 */
1509 	flush_ecache_il(caddr, csize, lnsize);		/* inline version */
1510 
1511 	/*
1512 	 * Wait for MCs to go idle.
1513 	 */
1514 	do {
1515 		register int	t = 10;
1516 		register uint_t	v;
1517 
1518 		/* loop t cycles waiting for each mc to indicate it's idle */
1519 		do {
1520 			v = ldphysio_il(isp->idle_addr)
1521 			    & STARFIRE_MC_IDLE_MASK;
1522 
1523 		} while (v != STARFIRE_MC_IDLE_MASK && t-- > 0);
1524 
1525 		/* bailout if timedout */
1526 		if (t <= 0) {
1527 			prog->restless_mc = isp->mem;
1528 			return;
1529 		}
1530 
1531 		isp += 1;
1532 
1533 		/* stop if terminating zero has been reached */
1534 	} while (isp->idle_addr != 0);
1535 
1536 	/* advance passed terminating zero */
1537 	isp += 1;
1538 
1539 	/*
1540 	 * The following inline assembly routine caches
1541 	 * the rename script and then caches the code that
1542 	 * will do the rename.  This is necessary
1543 	 * so that we don't have any memory references during
1544 	 * the reprogramming.  We accomplish this by first
1545 	 * jumping through the code to guarantee it's cached
1546 	 * before we actually execute it.
1547 	 */
1548 	drmach_exec_script_il((drmach_rename_script_t *)isp);
1549 }
1550 
1551 static void
1552 drmach_copy_rename_end(void)
1553 {
1554 	/*
1555 	 * IMPORTANT:	This function's location MUST be located immediately
1556 	 *		following drmach_copy_rename_prog__relocatable to
1557 	 *		accurately estimate its size.  Note that this assumes
1558 	 *		the compiler keeps these functions in the order in
1559 	 *		which they appear :-o
1560 	 */
1561 }
1562 
1563 sbd_error_t *
1564 drmach_copy_rename_init(drmachid_t t_id, uint64_t t_slice_offset,
1565 	drmachid_t s_id, struct memlist *c_ml, drmachid_t *pgm_id)
1566 {
1567 	drmach_device_t	*s_mem;
1568 	drmach_device_t	*t_mem;
1569 	struct memlist	*x_ml;
1570 	uint64_t	off_mask, s_copybasepa, t_copybasepa, t_basepa;
1571 	int		len;
1572 	caddr_t		bp, wp;
1573 	pda_handle_t	ph;
1574 	sbd_error_t	*err;
1575 	drmach_copy_rename_program_t *prog;
1576 
1577 	if (!DRMACH_IS_MEM_ID(s_id))
1578 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1579 	if (!DRMACH_IS_MEM_ID(t_id))
1580 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1581 	s_mem = s_id;
1582 	t_mem = t_id;
1583 
1584 	/* get starting physical address of target memory */
1585 	err = drmach_mem_get_base_physaddr(t_id, &t_basepa);
1586 	if (err)
1587 		return (err);
1588 
1589 	/* calculate slice offset mask from slice size */
1590 	off_mask = mc_get_mem_alignment() - 1;
1591 
1592 	/* calculate source and target base pa */
1593 	s_copybasepa = c_ml->ml_address;
1594 	t_copybasepa =
1595 	    t_basepa + ((c_ml->ml_address & off_mask) - t_slice_offset);
1596 
1597 	/* paranoia */
1598 	ASSERT((c_ml->ml_address & off_mask) >= t_slice_offset);
1599 
1600 	/* adjust copy memlist addresses to be relative to copy base pa */
1601 	x_ml = c_ml;
1602 	while (x_ml != NULL) {
1603 		x_ml->ml_address -= s_copybasepa;
1604 		x_ml = x_ml->ml_next;
1605 	}
1606 
1607 #ifdef DEBUG
1608 	{
1609 	uint64_t s_basepa, s_size, t_size;
1610 
1611 	x_ml = c_ml;
1612 	while (x_ml->ml_next != NULL)
1613 		x_ml = x_ml->ml_next;
1614 
1615 	DRMACH_PR("source copy span: base pa 0x%lx, end pa 0x%lx\n",
1616 	    s_copybasepa,
1617 	    s_copybasepa + x_ml->ml_address + x_ml->ml_size);
1618 
1619 	DRMACH_PR("target copy span: base pa 0x%lx, end pa 0x%lx\n",
1620 	    t_copybasepa,
1621 	    t_copybasepa + x_ml->ml_address + x_ml->ml_size);
1622 
1623 	DRMACH_PR("copy memlist (relative to copy base pa):\n");
1624 	MEMLIST_DUMP(c_ml);
1625 
1626 	err = drmach_mem_get_base_physaddr(s_id, &s_basepa);
1627 	ASSERT(err == NULL);
1628 
1629 	err = drmach_mem_get_size(s_id, &s_size);
1630 	ASSERT(err == NULL);
1631 
1632 	err = drmach_mem_get_size(t_id, &t_size);
1633 	ASSERT(err == NULL);
1634 
1635 	DRMACH_PR("current source base pa 0x%lx, size 0x%lx\n",
1636 	    s_basepa, s_size);
1637 	DRMACH_PR("current target base pa 0x%lx, size 0x%lx\n",
1638 	    t_basepa, t_size);
1639 
1640 	ASSERT(s_copybasepa + x_ml->ml_address + x_ml->ml_size <=
1641 	    s_basepa + s_size);
1642 	ASSERT(t_copybasepa + x_ml->ml_address + x_ml->ml_size <=
1643 	    t_basepa + t_size);
1644 	}
1645 #endif
1646 
1647 	ph = drmach_pda_open();
1648 	if (ph == NULL)
1649 		return (DRMACH_INTERNAL_ERROR());
1650 
1651 	/*
1652 	 * bp will be page aligned, since we're calling
1653 	 * kmem_zalloc() with an exact multiple of PAGESIZE.
1654 	 */
1655 	wp = bp = kmem_zalloc(PAGESIZE, KM_SLEEP);
1656 
1657 	/* allocate space for copy rename struct */
1658 	len = sizeof (drmach_copy_rename_program_t);
1659 	DRMACH_PR("prog = 0x%p, header len %d\n", (void *)wp, len);
1660 	prog = (drmach_copy_rename_program_t *)wp;
1661 	wp += (len + ecache_alignsize - 1) & ~ (ecache_alignsize - 1);
1662 
1663 	/*
1664 	 * Copy the code for the copy-rename routine into
1665 	 * a page aligned piece of memory.  We do this to guarantee
1666 	 * that we're executing within the same page and thus reduce
1667 	 * the possibility of cache collisions between different
1668 	 * pages.
1669 	 */
1670 	len = (int)((ulong_t)drmach_copy_rename_end -
1671 	    (ulong_t)drmach_copy_rename_prog__relocatable);
1672 	ASSERT(wp + len < bp + PAGESIZE);
1673 	bcopy((caddr_t)drmach_copy_rename_prog__relocatable, wp, len);
1674 
1675 	DRMACH_PR("copy-rename function 0x%p, len %d\n", (void *)wp, len);
1676 	prog->run = (void (*)())wp;
1677 	wp += (len + ecache_alignsize - 1) & ~ (ecache_alignsize - 1);
1678 
1679 	/*
1680 	 * Prepare data page that will contain script of
1681 	 * operations to perform during copy-rename.
1682 	 * Allocate temporary buffer to hold script.
1683 	 */
1684 	err = drmach_prep_rename_script(s_mem, t_mem, t_slice_offset,
1685 	    wp, PAGESIZE - (wp - bp));
1686 	if (err) {
1687 		(void) drmach_copy_rename_fini(prog);
1688 		return (err);
1689 	}
1690 
1691 	DRMACH_PR("copy-rename script 0x%p, len %d\n", (void *)wp, len);
1692 	prog->data = wp;
1693 	wp += (len + ecache_alignsize - 1) & ~ (ecache_alignsize - 1);
1694 
1695 	prog->ph = ph;
1696 	prog->s_copybasepa = s_copybasepa;
1697 	prog->t_copybasepa = t_copybasepa;
1698 	prog->c_ml = c_ml;
1699 	*pgm_id = prog;
1700 
1701 	return (NULL);
1702 }
1703 
1704 sbd_error_t *
1705 drmach_copy_rename_fini(drmachid_t id)
1706 {
1707 	drmach_copy_rename_program_t	*prog = id;
1708 	sbd_error_t			*err = NULL;
1709 
1710 	if (prog->c_ml != NULL)
1711 		memlist_delete(prog->c_ml);
1712 
1713 	if (prog->ph != NULL)
1714 		pda_close(prog->ph);
1715 
1716 	if (prog->restless_mc != 0) {
1717 		cmn_err(CE_WARN, "MC did not idle; OBP Node 0x%x",
1718 		    (uint_t)drmach_node_get_dnode(prog->restless_mc->node));
1719 
1720 		err = DRMACH_INTERNAL_ERROR();
1721 	}
1722 
1723 	kmem_free(prog, PAGESIZE);
1724 
1725 	return (err);
1726 }
1727 
1728 static sbd_error_t *
1729 drmach_io_new(drmach_device_t *dp)
1730 {
1731 	sbd_error_t	*err;
1732 	int		 portid;
1733 
1734 	err = drmach_device_get_prop(dp, "upa-portid", &portid);
1735 	if (err == NULL) {
1736 		ASSERT(portid & 0x40);
1737 		dp->unum = portid & 1;
1738 	}
1739 
1740 	dp->cm.isa = (void *)drmach_io_new;
1741 	dp->cm.release = drmach_io_release;
1742 	dp->cm.status = drmach_io_status;
1743 
1744 	(void) snprintf(dp->cm.name, sizeof (dp->cm.name), "%s%d", dp->type,
1745 	    dp->unum);
1746 
1747 	return (err);
1748 }
1749 
1750 static void
1751 drmach_iopc_op(pda_handle_t ph, drmach_iopc_op_t op)
1752 {
1753 	register int b;
1754 
1755 	for (b = 0; b < MAX_BOARDS; b++) {
1756 		int		p;
1757 		ushort_t	bda_ioc;
1758 		board_desc_t	*bdesc;
1759 
1760 		if (pda_board_present(ph, b) == 0)
1761 			continue;
1762 
1763 		bdesc = (board_desc_t *)pda_get_board_info(ph, b);
1764 		/*
1765 		 * Update PCs for IOCs.
1766 		 */
1767 		bda_ioc = bdesc->bda_ioc;
1768 		for (p = 0; p < MAX_IOCS; p++) {
1769 			u_longlong_t	idle_addr;
1770 			uchar_t		value;
1771 
1772 			if (BDA_NBL(bda_ioc, p) != BDAN_GOOD)
1773 				continue;
1774 
1775 			idle_addr = STARFIRE_BB_PC_ADDR(b, p, 1);
1776 
1777 			switch (op) {
1778 			case DO_PAUSE:
1779 				value = STARFIRE_BB_PC_PAUSE(p);
1780 				break;
1781 
1782 			case DO_IDLE:
1783 				value = STARFIRE_BB_PC_IDLE(p);
1784 				break;
1785 
1786 			case DO_UNPAUSE:
1787 				value = ldbphysio(idle_addr);
1788 				value &= ~STARFIRE_BB_PC_PAUSE(p);
1789 				break;
1790 
1791 			case DO_UNIDLE:
1792 				value = ldbphysio(idle_addr);
1793 				value &= ~STARFIRE_BB_PC_IDLE(p);
1794 				break;
1795 
1796 			default:
1797 				cmn_err(CE_PANIC,
1798 				    "drmach_iopc_op: unknown op (%d)",
1799 				    (int)op);
1800 				/*NOTREACHED*/
1801 			}
1802 			stbphysio(idle_addr, value);
1803 		}
1804 	}
1805 }
1806 
1807 void
1808 drmach_copy_rename(drmachid_t id)
1809 {
1810 	drmach_copy_rename_program_t	*prog = id;
1811 	uint64_t			neer;
1812 
1813 	/*
1814 	 * UPA IDLE
1815 	 * Protocol = PAUSE -> IDLE -> UNPAUSE
1816 	 * In reality since we only "idle" the IOPCs it's sufficient
1817 	 * to just issue the IDLE operation since (in theory) all IOPCs
1818 	 * in the field are PC6.  However, we'll be robust and do the
1819 	 * proper workaround protocol so that we never have to worry!
1820 	 */
1821 	drmach_iopc_op(prog->ph, DO_PAUSE);
1822 	drmach_iopc_op(prog->ph, DO_IDLE);
1823 	DELAY(100);
1824 	drmach_iopc_op(prog->ph, DO_UNPAUSE);
1825 	DELAY(100);
1826 
1827 	/* disable CE reporting */
1828 	neer = get_error_enable();
1829 	set_error_enable(neer & ~EER_CEEN);
1830 
1831 	/* run the copy/rename program */
1832 	prog->run(prog);
1833 
1834 	/* enable CE reporting */
1835 	set_error_enable(neer);
1836 
1837 	/*
1838 	 * UPA UNIDLE
1839 	 * Protocol = UNIDLE
1840 	 */
1841 	drmach_iopc_op(prog->ph, DO_UNIDLE);
1842 	DELAY(100);
1843 }
1844 
1845 /*
1846  * The counter-timer and perf-counter nodes are not being cleaned
1847  * up after a board that was present at start of day is detached.
1848  * If the board has become unconfigured with this operation, walk
1849  * the prom tree and find all counter-timer and perf-counter nodes
1850  * that have the same board number as the board that was just
1851  * unconfigured and remove them.
1852  */
1853 static sbd_error_t *
1854 drmach_remove_counter_nodes(drmachid_t id)
1855 {
1856 	int		num;
1857 	char		name[OBP_MAXDRVNAME];
1858 	pnode_t		child;
1859 	dev_info_t	*dip;
1860 	sbd_error_t	*err;
1861 	drmach_status_t	stat;
1862 	drmach_board_t	*bp;
1863 
1864 	if (!DRMACH_IS_BOARD_ID(id)) {
1865 		return (drerr_new(0, ESTF_INAPPROP, NULL));
1866 	}
1867 
1868 	if ((err = drmach_board_status(id, &stat)) != NULL) {
1869 		return (err);
1870 	}
1871 
1872 	/*
1873 	 * Only clean up the counter-timer and perf-counter
1874 	 * nodes when the entire board is unconfigured.
1875 	 */
1876 	if (stat.configured) {
1877 		return (NULL);
1878 	}
1879 
1880 	bp = (drmach_board_t *)id;
1881 
1882 	err = NULL;
1883 
1884 	for (child = prom_childnode(prom_rootnode()); child != OBP_NONODE;
1885 	    child = prom_nextnode(child)) {
1886 
1887 		if (prom_getprop(child, OBP_BOARDNUM, (caddr_t)&num) == -1) {
1888 			continue;
1889 		}
1890 
1891 		if (bp->bnum != num) {
1892 			continue;
1893 		}
1894 
1895 		if (prom_getprop(child, OBP_NAME, (caddr_t)name) == -1) {
1896 			continue;
1897 		}
1898 
1899 		if (strncmp(name, MISC_COUNTER_TIMER_DEVNAME, OBP_MAXDRVNAME) &&
1900 		    strncmp(name, MISC_PERF_COUNTER_DEVNAME, OBP_MAXDRVNAME)) {
1901 				continue;
1902 		}
1903 
1904 		/* Root node doesn't have to be held */
1905 		dip = e_ddi_nodeid_to_dip(child);
1906 
1907 		/*
1908 		 * If the node is only in the OBP tree, then
1909 		 * we don't have to remove it.
1910 		 */
1911 		if (dip) {
1912 			dev_info_t *fdip = NULL;
1913 
1914 			DRMACH_PR("removing %s devinfo node\n", name);
1915 
1916 			e_ddi_branch_hold(dip);
1917 			ddi_release_devi(dip); /* held in e_ddi_nodeid_to_dip */
1918 
1919 			if (e_ddi_branch_destroy(dip, &fdip, 0)) {
1920 				char *path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1921 
1922 				/*
1923 				 * If non-NULL, fdip is held and must be
1924 				 * released.
1925 				 */
1926 				if (fdip != NULL) {
1927 					(void) ddi_pathname(fdip, path);
1928 					ddi_release_devi(fdip);
1929 				} else {
1930 					(void) ddi_pathname(dip, path);
1931 				}
1932 
1933 				err = drerr_new(1, ESTF_DRVFAIL, path);
1934 				kmem_free(path, MAXPATHLEN);
1935 				e_ddi_branch_rele(dip);
1936 				break;
1937 			}
1938 		}
1939 	}
1940 
1941 	return (err);
1942 }
1943 
1944 /*ARGSUSED*/
1945 sbd_error_t *
1946 drmach_pre_op(int cmd, drmachid_t id, drmach_opts_t *opts)
1947 {
1948 	/* allow status and ncm operations to always succeed */
1949 	if ((cmd == SBD_CMD_STATUS) || (cmd == SBD_CMD_GETNCM)) {
1950 		return (NULL);
1951 	}
1952 
1953 	/* check all other commands for the required option string */
1954 	if ((opts->size > 0) && (opts->copts != NULL)) {
1955 
1956 		DRMACH_PR("platform options: %s\n", opts->copts);
1957 
1958 		if (strstr(opts->copts, "xfdr") != NULL) {
1959 			return (NULL);
1960 		}
1961 	}
1962 
1963 	return (drerr_new(0, ESTF_SUPPORT, NULL));
1964 }
1965 
1966 /*ARGSUSED*/
1967 sbd_error_t *
1968 drmach_post_op(int cmd, drmachid_t id, drmach_opts_t *opts)
1969 {
1970 	sbd_error_t	*err = NULL;
1971 
1972 	switch (cmd) {
1973 	case SBD_CMD_UNCONFIGURE:
1974 
1975 		err = drmach_remove_counter_nodes(id);
1976 		break;
1977 
1978 	case SBD_CMD_CONFIGURE:
1979 	case SBD_CMD_DISCONNECT:
1980 	case SBD_CMD_CONNECT:
1981 	case SBD_CMD_GETNCM:
1982 	case SBD_CMD_STATUS:
1983 		break;
1984 
1985 	default:
1986 		break;
1987 	}
1988 
1989 	return (err);
1990 }
1991 
1992 sbd_error_t *
1993 drmach_board_assign(int bnum, drmachid_t *id)
1994 {
1995 	sbd_error_t	*err;
1996 
1997 	if (!drmach_initialized && drmach_init() == -1) {
1998 		err = DRMACH_INTERNAL_ERROR();
1999 	} else if (drmach_array_get(drmach_boards, bnum, id) == -1) {
2000 		err = drerr_new(1, ESTF_BNUM, "%d", bnum);
2001 	} else if (*id != NULL) {
2002 		err = NULL;
2003 	} else {
2004 		drmach_board_t	*bp;
2005 
2006 		*id  = (drmachid_t)drmach_board_new(bnum);
2007 		bp = *id;
2008 		bp->assigned = 1;
2009 		err = NULL;
2010 	}
2011 
2012 	return (err);
2013 }
2014 
2015 static int
2016 drmach_attach_board(void *arg)
2017 {
2018 	drmach_board_t	*obj = (drmach_board_t *)arg;
2019 	cpuset_t	cset;
2020 	int		retval;
2021 
2022 	/*
2023 	 * OBP disables traps during the board probe.
2024 	 * So, in order to prevent cross-call/cross-trap timeouts,
2025 	 * and thus panics, we effectively block anybody from
2026 	 * issuing xc's/xt's by doing a promsafe_xc_attention.
2027 	 * In the previous version of Starfire DR (2.6), a timeout
2028 	 * suspension mechanism was implemented in the send-mondo
2029 	 * assembly.  That mechanism is unnecessary with the
2030 	 * existence of xc_attention/xc_dismissed.
2031 	 */
2032 	cset = cpu_ready_set;
2033 	promsafe_xc_attention(cset);
2034 
2035 	retval = prom_starfire_add_brd(obj->connect_cpuid);
2036 
2037 	xc_dismissed(cset);
2038 
2039 	return (retval);
2040 }
2041 
2042 sbd_error_t *
2043 drmach_board_connect(drmachid_t id, drmach_opts_t *opts)
2044 {
2045 	drmach_board_t	*obj = (drmach_board_t *)id;
2046 	int		retval;
2047 	sbd_error_t	*err;
2048 	char		*cptr, *copts;
2049 
2050 	if (!DRMACH_IS_BOARD_ID(id))
2051 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2052 
2053 	if (opts->size > 0)
2054 		copts = opts->copts;
2055 
2056 	if ((cptr = strstr(copts, "cpuid=")) != NULL) {
2057 		int cpuid;
2058 
2059 		cptr += strlen("cpuid=");
2060 		cpuid = stoi(&cptr);
2061 
2062 		if (DRMACH_CPUID2BNUM(cpuid) == obj->bnum) {
2063 			obj->connect_cpuid = cpuid;
2064 			obj->assigned = 1;
2065 		} else
2066 			return (drerr_new(1, ESTF_SETCPUVAL, "%d", cpuid));
2067 	} else {
2068 		/* cpuid was not specified */
2069 		obj->connect_cpuid = -1;
2070 	}
2071 
2072 	if (obj->connect_cpuid == -1) {
2073 		err =  drerr_new(1, ESTF_NOCPUID, obj->cm.name);
2074 		return (err);
2075 	}
2076 
2077 	cmn_err(CE_CONT, "DRMACH: PROM attach %s CPU %d\n",
2078 	    obj->cm.name, obj->connect_cpuid);
2079 
2080 	retval = prom_tree_update(drmach_attach_board, obj);
2081 
2082 	if (retval == 0)
2083 		err = NULL;
2084 	else {
2085 		cmn_err(CE_WARN, "prom error: prom_starfire_add_brd(%d) "
2086 		    "returned %d", obj->connect_cpuid, retval);
2087 
2088 		err = drerr_new(1, ESTF_PROBE, obj->cm.name);
2089 	}
2090 
2091 	obj->connect_cpuid = -1;
2092 
2093 	return (err);
2094 }
2095 
2096 /*ARGSUSED*/
2097 sbd_error_t *
2098 drmach_board_disconnect(drmachid_t id, drmach_opts_t *opts)
2099 {
2100 	drmach_board_t		*bp;
2101 	int			rv;
2102 	int			d_idx;	/* device index */
2103 	drmachid_t		d_id;	/* device ID */
2104 	sbd_error_t		*err;
2105 
2106 	if (!DRMACH_IS_BOARD_ID(id))
2107 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2108 
2109 	bp = id;
2110 
2111 	/*
2112 	 * We need to make sure all of the board's device nodes
2113 	 * have been removed from the Solaris device tree before
2114 	 * continuing with the disconnect. Otherwise, we could
2115 	 * disconnect the board and remove the OBP device tree
2116 	 * nodes with Solaris device tree nodes remaining.
2117 	 *
2118 	 * On Starfire, Solaris device tree nodes are deleted
2119 	 * during unconfigure by drmach_unconfigure(). It's
2120 	 * necessary to do this here because drmach_unconfigure()
2121 	 * failures are not handled during unconfigure.
2122 	 */
2123 	if (bp->devices) {
2124 		rv = drmach_array_first(bp->devices, &d_idx, &d_id);
2125 		while (rv == 0) {
2126 			err = drmach_unconfigure(d_id, DRMACH_DEVI_REMOVE);
2127 			if (err)
2128 				return (err);
2129 
2130 			rv = drmach_array_next(bp->devices, &d_idx, &d_id);
2131 		}
2132 	}
2133 
2134 	/*
2135 	 * Starfire board Solaris device tree counter nodes,
2136 	 * which are only present on start-of-day boards, are
2137 	 * removed in the dr_post_op() code flow after the
2138 	 * board is unconfigured. We call the counter node
2139 	 * removal function here because unconfigure errors
2140 	 * can cause the dr_post_op() function to be skipped
2141 	 * after an unconfigure operation even though all of
2142 	 * the board's devices have been transitioned to the
2143 	 * unconfigured state.
2144 	 */
2145 	err = drmach_remove_counter_nodes(id);
2146 	if (err)
2147 		return (err);
2148 
2149 	return (NULL);
2150 }
2151 
2152 static int
2153 drmach_board_find_devices_cb(drmach_node_walk_args_t *args)
2154 {
2155 	drmach_node_t			*node = args->node;
2156 	drmach_board_cb_data_t		*data = args->data;
2157 	drmach_board_t			*obj = data->obj;
2158 
2159 	int		 rv;
2160 	int		 bnum;
2161 	drmach_device_t	*device;
2162 
2163 	rv = drmach_node_get_prop(node, OBP_BOARDNUM, &bnum);
2164 	if (rv) {
2165 		/*
2166 		 * if the node does not have a board# property, then
2167 		 * by that information alone it is known that drmach
2168 		 * is not interested in it.
2169 		 */
2170 		return (0);
2171 	} else if (bnum != obj->bnum)
2172 		return (0);
2173 
2174 	/*
2175 	 * Create a device data structure from this node data.
2176 	 * The call may yield nothing if the node is not of interest
2177 	 * to drmach.
2178 	 */
2179 	data->err = drmach_device_new(node, obj, &device);
2180 	if (data->err)
2181 		return (-1);
2182 	else if (device == NULL) {
2183 		/*
2184 		 * drmach_device_new examined the node we passed in
2185 		 * and determined that it was one not of interest to
2186 		 * drmach.  So, it is skipped.
2187 		 */
2188 		return (0);
2189 	}
2190 
2191 	rv = drmach_array_set(obj->devices, data->ndevs++, device);
2192 	if (rv) {
2193 		drmach_device_dispose(device);
2194 		data->err = DRMACH_INTERNAL_ERROR();
2195 		return (-1);
2196 	}
2197 
2198 	data->err = (*data->found)(data->a, device->type, device->unum, device);
2199 	return (data->err == NULL ? 0 : -1);
2200 }
2201 
2202 sbd_error_t *
2203 drmach_board_find_devices(drmachid_t id, void *a,
2204 	sbd_error_t *(*found)(void *a, const char *, int, drmachid_t))
2205 {
2206 	extern int		 plat_max_cpu_units_per_board();
2207 	extern int		 plat_max_mem_units_per_board();
2208 	extern int		 plat_max_io_units_per_board();
2209 
2210 	drmach_board_t		*obj = (drmach_board_t *)id;
2211 	sbd_error_t		*err;
2212 	int			 max_devices;
2213 	int			 rv;
2214 	drmach_board_cb_data_t	data;
2215 
2216 	max_devices  = plat_max_cpu_units_per_board();
2217 	max_devices += plat_max_mem_units_per_board();
2218 	max_devices += plat_max_io_units_per_board();
2219 
2220 	obj->devices = drmach_array_new(0, max_devices);
2221 
2222 	data.obj = obj;
2223 	data.ndevs = 0;
2224 	data.found = found;
2225 	data.a = a;
2226 	data.err = NULL;
2227 
2228 	rv = drmach_node_walk(obj->tree, &data, drmach_board_find_devices_cb);
2229 	if (rv == 0)
2230 		err = NULL;
2231 	else {
2232 		drmach_array_dispose(obj->devices, drmach_device_dispose);
2233 		obj->devices = NULL;
2234 
2235 		if (data.err)
2236 			err = data.err;
2237 		else
2238 			err = DRMACH_INTERNAL_ERROR();
2239 	}
2240 
2241 	return (err);
2242 }
2243 
2244 int
2245 drmach_board_lookup(int bnum, drmachid_t *id)
2246 {
2247 	int	rv = 0;
2248 
2249 	if (!drmach_initialized && drmach_init() == -1) {
2250 		*id = 0;
2251 		rv = -1;
2252 	} else if (drmach_array_get(drmach_boards, bnum, id)) {
2253 		*id = 0;
2254 		rv = -1;
2255 	}
2256 	return (rv);
2257 }
2258 
2259 sbd_error_t *
2260 drmach_board_name(int bnum, char *buf, int buflen)
2261 {
2262 	(void) snprintf(buf, buflen, "SB%d", bnum);
2263 	return (NULL);
2264 }
2265 
2266 sbd_error_t *
2267 drmach_board_poweroff(drmachid_t id)
2268 {
2269 	drmach_board_t	*bp;
2270 	sbd_error_t	*err;
2271 	drmach_status_t	 stat;
2272 
2273 	if (!DRMACH_IS_BOARD_ID(id))
2274 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2275 	bp = id;
2276 
2277 	err = drmach_board_status(id, &stat);
2278 	if (err)
2279 		return (err);
2280 	else if (stat.configured || stat.busy)
2281 		return (drerr_new(0, ESTF_CONFIGBUSY, bp->cm.name));
2282 	else {
2283 		/* board power off is essentially a noop for Starfire */
2284 		bp->powered = 0;
2285 		return (NULL);
2286 	}
2287 	/*NOTREACHED*/
2288 }
2289 
2290 sbd_error_t *
2291 drmach_board_poweron(drmachid_t id)
2292 {
2293 	drmach_board_t	*bp;
2294 
2295 	if (!DRMACH_IS_BOARD_ID(id))
2296 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2297 	bp = id;
2298 
2299 	/* board power on is essentially a noop for Starfire */
2300 	bp->powered = 1;
2301 
2302 	return (NULL);
2303 }
2304 
2305 static sbd_error_t *
2306 drmach_board_release(drmachid_t id)
2307 {
2308 	if (!DRMACH_IS_BOARD_ID(id))
2309 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2310 	return (NULL);
2311 }
2312 
2313 /*ARGSUSED*/
2314 sbd_error_t *
2315 drmach_board_test(drmachid_t id, drmach_opts_t *opts, int force)
2316 {
2317 	return (NULL);
2318 }
2319 
2320 sbd_error_t *
2321 drmach_board_unassign(drmachid_t id)
2322 {
2323 	drmach_board_t	*bp;
2324 	sbd_error_t	*err;
2325 	drmach_status_t	 stat;
2326 
2327 	if (!DRMACH_IS_BOARD_ID(id))
2328 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2329 	bp = id;
2330 
2331 	err = drmach_board_status(id, &stat);
2332 	if (err)
2333 		return (err);
2334 	else if (stat.configured || stat.busy)
2335 		return (drerr_new(0, ESTF_CONFIGBUSY, bp->cm.name));
2336 	else if (drmach_array_set(drmach_boards, bp->bnum, 0) != 0)
2337 		return (DRMACH_INTERNAL_ERROR());
2338 	else {
2339 		drmach_board_dispose(bp);
2340 		return (NULL);
2341 	}
2342 	/*NOTREACHED*/
2343 }
2344 
2345 static sbd_error_t *
2346 drmach_cpu_new(drmach_device_t *dp)
2347 {
2348 	sbd_error_t	*err;
2349 	int		 portid;
2350 
2351 	err = drmach_device_get_prop(dp, "upa-portid", &portid);
2352 	if (err == NULL)
2353 		dp->unum = portid & 3;
2354 
2355 	dp->cm.isa = (void *)drmach_cpu_new;
2356 	dp->cm.release = drmach_cpu_release;
2357 	dp->cm.status = drmach_cpu_status;
2358 
2359 	(void) snprintf(dp->cm.name, sizeof (dp->cm.name), "%s%d", dp->type,
2360 	    dp->unum);
2361 
2362 	return (err);
2363 }
2364 
2365 /*
2366  * drmach_cpu_obp_detach()
2367  *  This requires two steps, first, we must put the cpuid into the OBP
2368  *  idle loop (Idle in Program) state.  Then we call OBP to place the CPU
2369  *  into the "Detached" state, which does any special processing to
2370  *  actually detach the cpu, such as flushing ecache, and also ensures
2371  *  that a subsequent breakpoint won't restart the cpu (if it was just in
2372  *  Idle in Program state).
2373  */
2374 static void
2375 drmach_cpu_obp_detach(int cpuid)
2376 {
2377 	/*
2378 	 * Cpu may not be under OBP's control. Eg, if cpu exited to download
2379 	 * helper on a prior attach.
2380 	 */
2381 	if (CPU_SGN_EXISTS(cpuid) &&
2382 	    !SGN_CPU_IS_OS(cpuid) &&
2383 	    !SGN_CPU_IS_OBP(cpuid)) {
2384 		cmn_err(CE_WARN,
2385 		    "unexpected signature (0x%x) for cpu %d",
2386 		    get_cpu_sgn(cpuid), cpuid);
2387 	}
2388 
2389 	/*
2390 	 * Now we place the CPU into the "Detached" idle loop in OBP.
2391 	 * This is so that the CPU won't be restarted if we break into
2392 	 * OBP with a breakpoint or BREAK key from the console, and also
2393 	 * if we need to do any special processing, such as flushing the
2394 	 * cpu's ecache, disabling interrupts (by turning of the ET bit in
2395 	 * the PSR) and/or spinning in BBSRAM rather than global memory.
2396 	 */
2397 	DRMACH_PR("prom_starfire_rm_cpu(%d)\n", cpuid);
2398 	prom_starfire_rm_cpu(cpuid);
2399 }
2400 
2401 /*
2402  * drmach_cpu_obp_is_detached() returns TRUE if the cpu sigblock signature state
2403  * is SIGBST_DETACHED; otherwise it returns FALSE. This routine should only
2404  * be called after we have asked OBP to detach the CPU. It should NOT be
2405  * called as a check during any other flow.
2406  */
2407 static int
2408 drmach_cpu_obp_is_detached(int cpuid)
2409 {
2410 	if (!CPU_SGN_EXISTS(cpuid) ||
2411 	    (SGN_CPU_IS_OS(cpuid) && SGN_CPU_STATE_IS_DETACHED(cpuid)))
2412 		return (1);
2413 	else
2414 		return (0);
2415 }
2416 
2417 static int
2418 drmach_cpu_start(struct cpu *cp)
2419 {
2420 	int		cpuid = cp->cpu_id;
2421 	int		ntries = drmach_cpu_ntries;
2422 	extern void	restart_other_cpu(int);
2423 
2424 	ASSERT(MUTEX_HELD(&cpu_lock));
2425 	ASSERT(cpunodes[cpuid].nodeid != (pnode_t)0);
2426 
2427 	cp->cpu_flags &= ~CPU_POWEROFF;
2428 
2429 	/*
2430 	 * NOTE: restart_other_cpu pauses cpus during the
2431 	 *	 slave cpu start.  This helps to quiesce the
2432 	 *	 bus traffic a bit which makes the tick sync
2433 	 *	 routine in the prom more robust.
2434 	 */
2435 	DRMACH_PR("COLD START for cpu (%d)\n", cpuid);
2436 
2437 	prom_starfire_add_cpu(cpuid);
2438 
2439 	restart_other_cpu(cpuid);
2440 
2441 	/*
2442 	 * Wait for the cpu to reach its idle thread before
2443 	 * we zap him with a request to blow away the mappings
2444 	 * he (might) have for the drmach_shutdown_asm code
2445 	 * he may have executed on unconfigure.
2446 	 */
2447 	while ((cp->cpu_thread != cp->cpu_idle_thread) && (ntries > 0)) {
2448 		DELAY(drmach_cpu_delay);
2449 		ntries--;
2450 	}
2451 
2452 	DRMACH_PR("waited %d out of %d loops for cpu %d\n",
2453 	    drmach_cpu_ntries - ntries, drmach_cpu_ntries, cpuid);
2454 
2455 	xt_one(cpuid, vtag_flushpage_tl1,
2456 	    (uint64_t)drmach_shutdown_va, (uint64_t)ksfmmup);
2457 
2458 	return (0);
2459 }
2460 
2461 /*
2462  * A detaching CPU is xcalled with an xtrap to drmach_cpu_stop_self() after
2463  * it has been offlined. The function of this routine is to get the cpu
2464  * spinning in a safe place. The requirement is that the system will not
2465  * reference anything on the detaching board (memory and i/o is detached
2466  * elsewhere) and that the CPU not reference anything on any other board
2467  * in the system.  This isolation is required during and after the writes
2468  * to the domain masks to remove the board from the domain.
2469  *
2470  * To accomplish this isolation the following is done:
2471  *	1) Create a locked mapping to a location in BBSRAM where
2472  *	   the cpu will execute.
2473  *	2) Copy the target function (drmach_shutdown_asm) in which
2474  *	   the cpu will execute into BBSRAM.
2475  *	3) Jump into function with BBSRAM.
2476  *	   Function will:
2477  *	   3.1) Flush its Ecache (displacement).
2478  *	   3.2) Flush its Dcache with HW mechanism.
2479  *	   3.3) Flush its Icache with HW mechanism.
2480  *	   3.4) Flush all valid and _unlocked_ D-TLB entries.
2481  *	   3.5) Flush all valid and _unlocked_ I-TLB entries.
2482  *	   3.6) Clear xt_mb to signal completion. Note: cache line is
2483  *		recovered by drmach_cpu_poweroff().
2484  *	4) Jump into a tight loop.
2485  */
2486 #define	DRMACH_BBSRAM_OFFSET	0x1000
2487 
2488 static void
2489 drmach_cpu_stop_self(void)
2490 {
2491 	int		cpuid = (int)CPU->cpu_id;
2492 	tte_t		tte;
2493 	volatile uint_t	*src, *dst;
2494 	size_t		funclen;
2495 	uint64_t	bbsram_pa, bbsram_offset;
2496 	uint_t		bbsram_pfn;
2497 	uint64_t	bbsram_addr;
2498 	void		(*bbsram_func)(uint64_t);
2499 	extern void	drmach_shutdown_asm(uint64_t);
2500 	extern void	drmach_shutdown_asm_end(void);
2501 
2502 	funclen = (uintptr_t)drmach_shutdown_asm_end -
2503 	    (uintptr_t)drmach_shutdown_asm;
2504 	ASSERT(funclen <= MMU_PAGESIZE);
2505 	/*
2506 	 * We'll start from the 0th's base.
2507 	 */
2508 	bbsram_pa = STARFIRE_UPAID2UPS(cpuid) | STARFIRE_PSI_BASE;
2509 	bbsram_offset = bbsram_pa | 0xfe0ULL;
2510 	bbsram_pa += ldphysio(bbsram_offset) + DRMACH_BBSRAM_OFFSET;
2511 
2512 	bbsram_pfn = (uint_t)(bbsram_pa >> MMU_PAGESHIFT);
2513 
2514 	bbsram_addr = (uint64_t)drmach_shutdown_va;
2515 	drmach_shutdown_asm_mbox->estack = bbsram_addr + funclen;
2516 
2517 	tte.tte_inthi = TTE_VALID_INT | TTE_SZ_INT(TTE8K) |
2518 	    TTE_PFN_INTHI(bbsram_pfn);
2519 	tte.tte_intlo = TTE_PFN_INTLO(bbsram_pfn) |
2520 	    TTE_HWWR_INT | TTE_PRIV_INT | TTE_LCK_INT;
2521 	sfmmu_dtlb_ld_kva(drmach_shutdown_va, &tte);	/* load dtlb */
2522 	sfmmu_itlb_ld_kva(drmach_shutdown_va, &tte);	/* load itlb */
2523 
2524 	for (src = (uint_t *)drmach_shutdown_asm, dst = (uint_t *)bbsram_addr;
2525 	    src < (uint_t *)drmach_shutdown_asm_end; src++, dst++)
2526 		*dst = *src;
2527 
2528 	bbsram_func = (void (*)())bbsram_addr;
2529 	drmach_shutdown_asm_mbox->flushaddr = ecache_flushaddr;
2530 	drmach_shutdown_asm_mbox->size = (cpunodes[cpuid].ecache_size << 1);
2531 	drmach_shutdown_asm_mbox->linesize = cpunodes[cpuid].ecache_linesize;
2532 	drmach_shutdown_asm_mbox->physaddr =
2533 	    va_to_pa((void *)&drmach_xt_mb[cpuid]);
2534 
2535 	/*
2536 	 * Signal to drmach_cpu_poweroff() is via drmach_xt_mb cleared
2537 	 * by asm code
2538 	 */
2539 
2540 	(*bbsram_func)(va_to_pa((void *)drmach_shutdown_asm_mbox));
2541 }
2542 
2543 static void
2544 drmach_cpu_shutdown_self(void)
2545 {
2546 	cpu_t		*cp = CPU;
2547 	int		cpuid = cp->cpu_id;
2548 	extern void	flush_windows(void);
2549 
2550 	flush_windows();
2551 
2552 	(void) spl8();
2553 
2554 	ASSERT(cp->cpu_intr_actv == 0);
2555 	ASSERT(cp->cpu_thread == cp->cpu_idle_thread ||
2556 	    cp->cpu_thread == cp->cpu_startup_thread);
2557 
2558 	cp->cpu_flags = CPU_OFFLINE | CPU_QUIESCED | CPU_POWEROFF;
2559 
2560 	drmach_cpu_stop_self();
2561 
2562 	cmn_err(CE_PANIC, "CPU %d FAILED TO SHUTDOWN", cpuid);
2563 }
2564 
2565 /* a helper routine to keep the math in one place */
2566 static processorid_t
2567 drmach_cpu_calc_id(drmach_device_t *dp)
2568 {
2569 	return (dp->bp->bnum * MAX_CPU_UNITS_PER_BOARD + dp->unum);
2570 }
2571 
2572 /*
2573  * Move bootproc (SIGBCPU) to another cpu.  If dst_cpu is NULL, a
2574  * destination cpu is chosen from the set of cpus not located on the
2575  * same board as the current bootproc cpu.
2576  */
2577 static sbd_error_t *
2578 drmach_cpu_juggle_bootproc(drmach_device_t *dst_cpu)
2579 {
2580 	processorid_t	 cpuid;
2581 	struct cpu	*cp;
2582 	sbd_error_t	*err;
2583 	int		 rv;
2584 
2585 	ASSERT(MUTEX_HELD(&cpu_lock));
2586 
2587 	/* dst_cpu is NULL when target cpu is unspecified. So, pick one. */
2588 	if (dst_cpu == NULL) {
2589 		int avoid_board = DRMACH_CPUID2BNUM(SIGBCPU->cpu_id);
2590 		int max_cpuid = MAX_BOARDS * MAX_CPU_UNITS_PER_BOARD;
2591 
2592 		for (cpuid = 0; cpuid < max_cpuid; cpuid++)
2593 			if (DRMACH_CPUID2BNUM(cpuid) != avoid_board) {
2594 				cp = cpu_get(cpuid);
2595 				if (cp != NULL && cpu_is_online(cp))
2596 					break;
2597 			}
2598 
2599 		if (cpuid == max_cpuid) {
2600 			err = drerr_new(1, ESTF_JUGGLE, NULL);
2601 			return (err);
2602 		}
2603 
2604 		/* else, cp points to the selected target cpu */
2605 	} else {
2606 		cpuid = drmach_cpu_calc_id(dst_cpu);
2607 
2608 		if ((cp = cpu_get(cpuid)) == NULL) {
2609 			err = drerr_new(1, ESTF_NODEV, "%s::%s",
2610 			    dst_cpu->bp->cm.name, dst_cpu->cm.name);
2611 			return (err);
2612 		}
2613 
2614 		if (cpuid == SIGBCPU->cpu_id) {
2615 			cmn_err(CE_WARN,
2616 			    "SIGBCPU(%d) same as new selection(%d)",
2617 			    SIGBCPU->cpu_id, cpuid);
2618 
2619 			/* technically not an error, but a no-op */
2620 			return (NULL);
2621 		}
2622 	}
2623 
2624 	cmn_err(CE_NOTE, "?relocating SIGBCPU from %d to %d",
2625 	    SIGBCPU->cpu_id, cpuid);
2626 
2627 	DRMACH_PR("moving SIGBCPU to CPU %d\n", cpuid);
2628 
2629 	/*
2630 	 * Tell OBP to initialize cvc-offset field of new CPU0
2631 	 * so that it's in sync with OBP and cvc_server
2632 	 */
2633 	prom_starfire_init_console(cpuid);
2634 
2635 	/*
2636 	 * Assign cvc to new cpu0's bbsram for I/O.  This has to be
2637 	 * done BEFORE cpu0 is moved via obp, since this logic
2638 	 * will cause obp_helper to switch to a different bbsram for
2639 	 * cvc I/O.  We don't want cvc writing to a buffer from which
2640 	 * nobody will pick up the data!
2641 	 */
2642 	cvc_assign_iocpu(cpuid);
2643 
2644 	rv = prom_starfire_move_cpu0(cpuid);
2645 
2646 	if (rv == 0) {
2647 		SIGBCPU = cp;
2648 
2649 		DRMACH_PR("successfully juggled to CPU %d\n", cpuid);
2650 		return (NULL);
2651 	} else {
2652 		DRMACH_PR("prom error: prom_starfire_move_cpu0(%d) "
2653 		    "returned %d\n", cpuid, rv);
2654 
2655 		/*
2656 		 * The move failed, hopefully obp_helper is still back
2657 		 * at the old bootproc.  Move cvc back there.
2658 		 */
2659 		cvc_assign_iocpu(SIGBCPU->cpu_id);
2660 
2661 
2662 		err = drerr_new(1, ESTF_MOVESIGB, "CPU %d", cpuid);
2663 		return (err);
2664 	}
2665 	/*NOTREACHED*/
2666 }
2667 
2668 static sbd_error_t *
2669 drmach_cpu_release(drmachid_t id)
2670 {
2671 	drmach_device_t	*dp;
2672 	processorid_t	 cpuid;
2673 	struct cpu	*cp;
2674 	sbd_error_t	*err;
2675 
2676 	if (!DRMACH_IS_CPU_ID(id))
2677 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2678 	dp = id;
2679 	cpuid = drmach_cpu_calc_id(dp);
2680 
2681 	ASSERT(MUTEX_HELD(&cpu_lock));
2682 
2683 	cp = cpu_get(cpuid);
2684 	if (cp == NULL)
2685 		err = DRMACH_INTERNAL_ERROR();
2686 	else if (SIGBCPU->cpu_id == cp->cpu_id)
2687 		err = drmach_cpu_juggle_bootproc(NULL);
2688 	else
2689 		err = NULL;
2690 
2691 	return (err);
2692 }
2693 
2694 static sbd_error_t *
2695 drmach_cpu_status(drmachid_t id, drmach_status_t *stat)
2696 {
2697 	drmach_device_t *dp;
2698 
2699 	ASSERT(DRMACH_IS_CPU_ID(id));
2700 	dp = id;
2701 
2702 	stat->assigned = dp->bp->assigned;
2703 	stat->powered = dp->bp->powered;
2704 	mutex_enter(&cpu_lock);
2705 	stat->configured = (cpu_get(drmach_cpu_calc_id(dp)) != NULL);
2706 	mutex_exit(&cpu_lock);
2707 	stat->busy = dp->busy;
2708 	(void) strncpy(stat->type, dp->type, sizeof (stat->type));
2709 	stat->info[0] = '\0';
2710 
2711 	return (NULL);
2712 }
2713 
2714 sbd_error_t *
2715 drmach_cpu_disconnect(drmachid_t id)
2716 {
2717 	drmach_device_t	*cpu;
2718 	int		 cpuid;
2719 	int		 ntries;
2720 	int		 p;
2721 	u_longlong_t	 pc_addr;
2722 	uchar_t		 rvalue;
2723 
2724 	if (!DRMACH_IS_CPU_ID(id))
2725 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2726 	cpu = id;
2727 
2728 	cpuid = drmach_cpu_calc_id(cpu);
2729 	if (SIGBCPU->cpu_id == cpuid) {
2730 		/* this cpu is SIGBCPU, can't disconnect */
2731 		return (drerr_new(1, ESTF_HASSIGB, "%s::%s",
2732 		    cpu->bp->cm.name, cpu->cm.name));
2733 	}
2734 
2735 	/*
2736 	 * Make sure SIGBST_DETACHED is set before
2737 	 * mapping out the sig block.
2738 	 */
2739 	ntries = drmach_cpu_ntries;
2740 	while (!drmach_cpu_obp_is_detached(cpuid) && ntries) {
2741 		DELAY(drmach_cpu_delay);
2742 		ntries--;
2743 	}
2744 	if (!drmach_cpu_obp_is_detached(cpuid)) {
2745 		cmn_err(CE_WARN, "failed to mark cpu %d detached in sigblock",
2746 		    cpuid);
2747 	}
2748 
2749 	/* map out signature block */
2750 	if (CPU_SGN_EXISTS(cpuid)) {
2751 		CPU_SGN_MAPOUT(cpuid);
2752 	}
2753 
2754 	/*
2755 	 * We now PC IDLE the processor to guarantee we
2756 	 * stop any transactions from coming from it.
2757 	 */
2758 	p = cpu->unum & 1;
2759 	pc_addr = STARFIRE_BB_PC_ADDR(cpu->bp->bnum, cpu->unum, 0);
2760 
2761 	DRMACH_PR("PC idle cpu %d (addr = 0x%llx, port = %d, p = %d)",
2762 	    drmach_cpu_calc_id(cpu), pc_addr, cpu->unum, p);
2763 
2764 	rvalue = ldbphysio(pc_addr);
2765 	rvalue |= STARFIRE_BB_PC_IDLE(p);
2766 	stbphysio(pc_addr, rvalue);
2767 	DELAY(50000);
2768 
2769 	return (NULL);
2770 }
2771 
2772 sbd_error_t *
2773 drmach_cpu_get_id(drmachid_t id, processorid_t *cpuid)
2774 {
2775 	drmach_device_t *cpu;
2776 
2777 	if (!DRMACH_IS_CPU_ID(id))
2778 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2779 	cpu = id;
2780 
2781 	*cpuid = drmach_cpu_calc_id(cpu);
2782 	return (NULL);
2783 }
2784 
2785 sbd_error_t *
2786 drmach_cpu_get_impl(drmachid_t id, int *ip)
2787 {
2788 	drmach_device_t *cpu;
2789 	int		impl;
2790 
2791 	if (!DRMACH_IS_CPU_ID(id))
2792 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2793 
2794 	cpu = id;
2795 
2796 	if (drmach_node_get_prop(cpu->node, "implementation#", &impl) == -1) {
2797 		return (DRMACH_INTERNAL_ERROR());
2798 	}
2799 
2800 	*ip = impl;
2801 
2802 	return (NULL);
2803 }
2804 
2805 void
2806 drmach_cpu_flush_ecache_sync(void)
2807 {
2808 	ASSERT(curthread->t_bound_cpu == CPU);
2809 
2810 	/*
2811 	 * Now let's flush our ecache thereby removing all references
2812 	 * to the target (detaching) memory from all ecache's in
2813 	 * system.
2814 	 */
2815 	cpu_flush_ecache();
2816 
2817 	/*
2818 	 * Delay 100 usec out of paranoia to insure everything
2819 	 * (hardware queues) has drained before we start reprogramming
2820 	 * the hardware.
2821 	 */
2822 	DELAY(100);
2823 }
2824 
2825 sbd_error_t *
2826 drmach_get_dip(drmachid_t id, dev_info_t **dip)
2827 {
2828 	drmach_device_t	*dp;
2829 
2830 	if (!DRMACH_IS_DEVICE_ID(id))
2831 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2832 	dp = id;
2833 
2834 	*dip = drmach_node_get_dip(dp->node);
2835 	return (NULL);
2836 }
2837 
2838 sbd_error_t *
2839 drmach_io_is_attached(drmachid_t id, int *yes)
2840 {
2841 	drmach_device_t *dp;
2842 	dev_info_t	*dip;
2843 	int		state;
2844 
2845 	if (!DRMACH_IS_IO_ID(id))
2846 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2847 	dp = id;
2848 
2849 	dip = drmach_node_get_dip(dp->node);
2850 	if (dip == NULL) {
2851 		*yes = 0;
2852 		return (NULL);
2853 	}
2854 
2855 	state = ddi_get_devstate(dip);
2856 	*yes = (i_ddi_devi_attached(dip) || (state == DDI_DEVSTATE_UP));
2857 
2858 	return (NULL);
2859 }
2860 
2861 sbd_error_t *
2862 drmach_io_pre_release(drmachid_t id)
2863 {
2864 	if (!DRMACH_IS_IO_ID(id))
2865 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2866 	return (NULL);
2867 }
2868 
2869 static sbd_error_t *
2870 drmach_io_release(drmachid_t id)
2871 {
2872 	if (!DRMACH_IS_IO_ID(id))
2873 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2874 	return (NULL);
2875 }
2876 
2877 sbd_error_t *
2878 drmach_io_unrelease(drmachid_t id)
2879 {
2880 	if (!DRMACH_IS_IO_ID(id))
2881 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2882 	return (NULL);
2883 }
2884 
2885 /*ARGSUSED*/
2886 sbd_error_t *
2887 drmach_io_post_release(drmachid_t id)
2888 {
2889 	return (NULL);
2890 }
2891 
2892 /*ARGSUSED*/
2893 sbd_error_t *
2894 drmach_io_post_attach(drmachid_t id)
2895 {
2896 	return (NULL);
2897 }
2898 
2899 static sbd_error_t *
2900 drmach_io_status(drmachid_t id, drmach_status_t *stat)
2901 {
2902 	drmach_device_t *dp;
2903 	sbd_error_t	*err;
2904 	int		 configured;
2905 
2906 	ASSERT(DRMACH_IS_IO_ID(id));
2907 	dp = id;
2908 
2909 	err = drmach_io_is_attached(id, &configured);
2910 	if (err)
2911 		return (err);
2912 
2913 	stat->assigned = dp->bp->assigned;
2914 	stat->powered = dp->bp->powered;
2915 	stat->configured = (configured != 0);
2916 	stat->busy = dp->busy;
2917 	(void) strncpy(stat->type, dp->type, sizeof (stat->type));
2918 	stat->info[0] = '\0';
2919 
2920 	return (NULL);
2921 }
2922 
2923 static sbd_error_t *
2924 drmach_mem_new(drmach_device_t *dp)
2925 {
2926 	dp->unum = 0;
2927 	dp->cm.isa = (void *)drmach_mem_new;
2928 	dp->cm.release = drmach_mem_release;
2929 	dp->cm.status = drmach_mem_status;
2930 
2931 	(void) snprintf(dp->cm.name, sizeof (dp->cm.name), "%s", dp->type);
2932 
2933 	return (NULL);
2934 }
2935 
2936 sbd_error_t *
2937 drmach_mem_add_span(drmachid_t id, uint64_t basepa, uint64_t size)
2938 {
2939 	pfn_t		basepfn = (pfn_t)(basepa >> PAGESHIFT);
2940 	pgcnt_t		npages = (pgcnt_t)(size >> PAGESHIFT);
2941 	pda_handle_t	ph;
2942 	int		rv;
2943 
2944 	ASSERT(size != 0);
2945 
2946 	if (!DRMACH_IS_MEM_ID(id))
2947 		return (drerr_new(0, ESTF_INAPPROP, NULL));
2948 
2949 	rv = kcage_range_add(basepfn, npages, KCAGE_DOWN);
2950 	if (rv == ENOMEM) {
2951 		cmn_err(CE_WARN, "%lu megabytes not available to kernel cage",
2952 		    (ulong_t)(size == 0 ? 0 : size / MBYTE));
2953 	} else if (rv != 0) {
2954 		/* catch this in debug kernels */
2955 		ASSERT(0);
2956 
2957 		cmn_err(CE_WARN, "unexpected kcage_range_add"
2958 		    " return value %d", rv);
2959 	}
2960 
2961 	/*
2962 	 * Update the PDA (post2obp) structure with the
2963 	 * range of the newly added memory.
2964 	 */
2965 	ph = drmach_pda_open();
2966 	if (ph != NULL) {
2967 		pda_mem_add_span(ph, basepa, size);
2968 		pda_close(ph);
2969 	}
2970 
2971 	return (NULL);
2972 }
2973 
2974 sbd_error_t *
2975 drmach_mem_del_span(drmachid_t id, uint64_t basepa, uint64_t size)
2976 {
2977 	drmach_device_t	*mem = id;
2978 	pfn_t		basepfn = (pfn_t)(basepa >> PAGESHIFT);
2979 	pgcnt_t		npages = (pgcnt_t)(size >> PAGESHIFT);
2980 	uint_t		mcreg;
2981 	sbd_error_t	*err;
2982 	pda_handle_t	ph;
2983 	int		rv;
2984 
2985 	err = drmach_read_mc_asr(id, &mcreg);
2986 	if (err)
2987 		return (err);
2988 	else if (mcreg & STARFIRE_MC_INTERLEAVE_MASK) {
2989 		return (drerr_new(1, ESTF_INTERBOARD, "%s::%s",
2990 		    mem->bp->cm.name, mem->cm.name));
2991 	}
2992 
2993 	if (size > 0) {
2994 		rv = kcage_range_delete_post_mem_del(basepfn, npages);
2995 		if (rv != 0) {
2996 			cmn_err(CE_WARN,
2997 			    "unexpected kcage_range_delete_post_mem_del"
2998 			    " return value %d", rv);
2999 			return (DRMACH_INTERNAL_ERROR());
3000 		}
3001 	}
3002 
3003 	/*
3004 	 * Update the PDA (post2obp) structure with the
3005 	 * range of removed memory.
3006 	 */
3007 	ph = drmach_pda_open();
3008 	if (ph != NULL) {
3009 		if (size > 0)
3010 			pda_mem_del_span(ph, basepa, size);
3011 
3012 		/* update PDA to board's new mc register settings */
3013 		pda_mem_sync(ph, mem->bp->bnum, 0);
3014 
3015 		pda_close(ph);
3016 	}
3017 
3018 	return (NULL);
3019 }
3020 
3021 /* support routine for enable and disable */
3022 static sbd_error_t *
3023 drmach_mem_update_interconnect(drmachid_t id, uint_t mcreg)
3024 {
3025 	drmach_device_t	*dp;
3026 	pda_handle_t	 ph;
3027 	int		 b;
3028 
3029 	if (!DRMACH_IS_MEM_ID(id))
3030 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3031 	dp = id;
3032 
3033 	ph = drmach_pda_open();
3034 	if (ph == NULL)
3035 		return (DRMACH_INTERNAL_ERROR());
3036 
3037 	for (b = 0; b < MAX_BOARDS; b++) {
3038 		int		p;
3039 		int		rv;
3040 		ushort_t	bda_proc, bda_ioc;
3041 		board_desc_t	*bdesc;
3042 
3043 		if (pda_board_present(ph, b) == 0)
3044 			continue;
3045 
3046 		bdesc = (board_desc_t *)pda_get_board_info(ph, b);
3047 
3048 		/*
3049 		 * Update PCs for CPUs.
3050 		 */
3051 
3052 		/* make sure definition in platmod is in sync with pda */
3053 		ASSERT(MAX_PROCMODS == MAX_CPU_UNITS_PER_BOARD);
3054 
3055 		bda_proc = bdesc->bda_proc;
3056 		for (p = 0; p < MAX_PROCMODS; p++) {
3057 			if (BDA_NBL(bda_proc, p) != BDAN_GOOD)
3058 				continue;
3059 
3060 			rv = pc_madr_add(b, dp->bp->bnum, p, mcreg);
3061 			if (rv) {
3062 				pda_close(ph);
3063 				return (DRMACH_INTERNAL_ERROR());
3064 			}
3065 		}
3066 
3067 		/*
3068 		 * Update PCs for IOCs.
3069 		 */
3070 
3071 		/* make sure definition in platmod is in sync with pda */
3072 		ASSERT(MAX_IOCS == MAX_IO_UNITS_PER_BOARD);
3073 
3074 		bda_ioc = bdesc->bda_ioc;
3075 		for (p = 0; p < MAX_IOCS; p++) {
3076 			if (BDA_NBL(bda_ioc, p) != BDAN_GOOD)
3077 				continue;
3078 
3079 			rv = pc_madr_add(b, dp->bp->bnum, p + 4, mcreg);
3080 			if (rv) {
3081 				pda_close(ph);
3082 				return (DRMACH_INTERNAL_ERROR());
3083 			}
3084 		}
3085 	}
3086 
3087 	pda_close(ph);
3088 	return (NULL);
3089 }
3090 
3091 sbd_error_t *
3092 drmach_mem_disable(drmachid_t id)
3093 {
3094 	sbd_error_t	*err;
3095 	uint_t		 mcreg;
3096 
3097 	err = drmach_read_mc_asr(id, &mcreg);
3098 	if (err == NULL) {
3099 		ASSERT(mcreg & STARFIRE_MC_MEM_PRESENT_MASK);
3100 
3101 		/* Turn off presence bit. */
3102 		mcreg &= ~STARFIRE_MC_MEM_PRESENT_MASK;
3103 
3104 		err = drmach_mem_update_interconnect(id, mcreg);
3105 		if (err == NULL)
3106 			err = drmach_write_mc_asr(id, mcreg);
3107 	}
3108 
3109 	return (err);
3110 }
3111 
3112 sbd_error_t *
3113 drmach_mem_enable(drmachid_t id)
3114 {
3115 	sbd_error_t	*err;
3116 	uint_t		 mcreg;
3117 
3118 	err = drmach_read_mc_asr(id, &mcreg);
3119 	if (err == NULL) {
3120 		mcreg |= STARFIRE_MC_MEM_PRESENT_MASK;
3121 
3122 		err = drmach_write_mc_asr(id, mcreg);
3123 		if (err == NULL)
3124 			err = drmach_mem_update_interconnect(id, mcreg);
3125 	}
3126 
3127 	return (err);
3128 }
3129 
3130 sbd_error_t *
3131 drmach_mem_get_alignment(drmachid_t id, uint64_t *mask)
3132 {
3133 	drmach_device_t	*mem;
3134 	sbd_error_t	*err;
3135 	pnode_t		 nodeid;
3136 
3137 	if (!DRMACH_IS_MEM_ID(id))
3138 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3139 	mem = id;
3140 
3141 	nodeid = drmach_node_get_dnode(mem->node);
3142 	if (nodeid == OBP_NONODE || nodeid == OBP_BADNODE)
3143 		err = DRMACH_INTERNAL_ERROR();
3144 	else {
3145 		uint64_t size;
3146 
3147 		size = mc_get_alignment_mask(nodeid);
3148 		if (size == (uint64_t)-1)
3149 			err = DRMACH_INTERNAL_ERROR();
3150 		else {
3151 			*mask = size - 1;
3152 			err = NULL;
3153 		}
3154 	}
3155 
3156 	return (err);
3157 }
3158 
3159 sbd_error_t *
3160 drmach_mem_get_base_physaddr(drmachid_t id, uint64_t *pa)
3161 {
3162 	sbd_error_t	*err;
3163 	uint_t		 mcreg;
3164 
3165 	err = drmach_read_mc_asr(id, &mcreg);
3166 	if (err == NULL)
3167 		*pa = mc_asr_to_pa(mcreg);
3168 
3169 	return (err);
3170 }
3171 
3172 /*
3173  * Use of this routine after copy/rename will yield incorrect results,
3174  * because the OBP MEMAVAIL property will not correctly reflect the
3175  * programming of the MCs.
3176  */
3177 sbd_error_t *
3178 drmach_mem_get_memlist(drmachid_t id, struct memlist **ml)
3179 {
3180 	drmach_device_t	*mem;
3181 	int		rv, i, rlen, rblks;
3182 	sbd_error_t	*err;
3183 	struct memlist	*mlist;
3184 	struct sf_memunit_regspec *rlist;
3185 
3186 	if (!DRMACH_IS_MEM_ID(id))
3187 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3188 	mem = id;
3189 
3190 	err = drmach_device_get_proplen(mem, "dr-available", &rlen);
3191 	if (err)
3192 		return (err);
3193 
3194 	rlist = kmem_zalloc(rlen, KM_SLEEP);
3195 
3196 	err = drmach_device_get_prop(mem, "dr-available", rlist);
3197 	if (err) {
3198 		kmem_free(rlist, rlen);
3199 		return (err);
3200 	}
3201 
3202 	mlist = NULL;
3203 	rblks = rlen / sizeof (struct sf_memunit_regspec);
3204 	for (i = 0; i < rblks; i++) {
3205 		uint64_t	addr, size;
3206 
3207 		addr  = (uint64_t)rlist[i].regspec_addr_hi << 32;
3208 		addr |= (uint64_t)rlist[i].regspec_addr_lo;
3209 		size  = (uint64_t)rlist[i].regspec_size_hi << 32;
3210 		size |= (uint64_t)rlist[i].regspec_size_lo;
3211 
3212 		mlist = memlist_add_span(mlist, addr, size);
3213 	}
3214 
3215 	kmem_free(rlist, rlen);
3216 
3217 	/*
3218 	 * Make sure the incoming memlist doesn't already
3219 	 * intersect with what's present in the system (phys_install).
3220 	 */
3221 	memlist_read_lock();
3222 	rv = memlist_intersect(phys_install, mlist);
3223 	memlist_read_unlock();
3224 	if (rv) {
3225 #ifdef DEBUG
3226 		DRMACH_PR("OBP derived memlist intersects"
3227 		    " with phys_install\n");
3228 		memlist_dump(mlist);
3229 
3230 		DRMACH_PR("phys_install memlist:\n");
3231 		memlist_dump(phys_install);
3232 #endif
3233 
3234 		memlist_delete(mlist);
3235 		return (DRMACH_INTERNAL_ERROR());
3236 	}
3237 
3238 #ifdef DEBUG
3239 	DRMACH_PR("OBP derived memlist:");
3240 	memlist_dump(mlist);
3241 #endif
3242 
3243 	*ml = mlist;
3244 	return (NULL);
3245 }
3246 
3247 sbd_error_t *
3248 drmach_mem_get_size(drmachid_t id, uint64_t *bytes)
3249 {
3250 	drmach_device_t	*mem;
3251 	pda_handle_t	ph;
3252 	pgcnt_t		npages;
3253 
3254 	if (!DRMACH_IS_MEM_ID(id))
3255 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3256 	mem = id;
3257 
3258 	ph = drmach_pda_open();
3259 	if (ph == NULL)
3260 		return (DRMACH_INTERNAL_ERROR());
3261 
3262 	npages = pda_get_mem_size(ph, mem->bp->bnum);
3263 	*bytes = (uint64_t)npages << PAGESHIFT;
3264 
3265 	pda_close(ph);
3266 	return (NULL);
3267 }
3268 
3269 sbd_error_t *
3270 drmach_mem_get_slice_size(drmachid_t id, uint64_t *bytes)
3271 {
3272 	if (!DRMACH_IS_MEM_ID(id))
3273 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3274 
3275 	*bytes = mc_get_mem_alignment();
3276 	return (NULL);
3277 }
3278 
3279 /* field debugging tool */
3280 processorid_t drmach_mem_cpu_affinity_nail = 0;
3281 
3282 processorid_t
3283 drmach_mem_cpu_affinity(drmachid_t id)
3284 {
3285 	drmach_device_t	*mp;
3286 	drmach_board_t	*bp;
3287 	processorid_t	 cpuid;
3288 
3289 	if (!DRMACH_IS_MEM_ID(id))
3290 		return (CPU_CURRENT);
3291 
3292 	if (drmach_mem_cpu_affinity_nail) {
3293 		cpuid = drmach_mem_cpu_affinity_nail;
3294 
3295 		if (cpuid < 0 || cpuid > NCPU)
3296 			return (CPU_CURRENT);
3297 
3298 		mutex_enter(&cpu_lock);
3299 		if (cpu[cpuid] == NULL || !CPU_ACTIVE(cpu[cpuid]))
3300 			cpuid = CPU_CURRENT;
3301 		mutex_exit(&cpu_lock);
3302 
3303 		return (cpuid);
3304 	}
3305 
3306 	/* try to choose a proc on the target board */
3307 	mp = id;
3308 	bp = mp->bp;
3309 	if (bp->devices) {
3310 		int		rv;
3311 		int		d_idx;
3312 		drmachid_t	d_id;
3313 
3314 		rv = drmach_array_first(bp->devices, &d_idx, &d_id);
3315 		while (rv == 0) {
3316 			if (DRMACH_IS_CPU_ID(d_id)) {
3317 				cpuid = drmach_cpu_calc_id(d_id);
3318 
3319 				mutex_enter(&cpu_lock);
3320 				if (cpu[cpuid] && CPU_ACTIVE(cpu[cpuid])) {
3321 					mutex_exit(&cpu_lock);
3322 					DRMACH_PR("drmach_mem_cpu_affinity: "
3323 					    "selected cpuid=%d\n", cpuid);
3324 					return (cpuid);
3325 				} else {
3326 					mutex_exit(&cpu_lock);
3327 				}
3328 			}
3329 
3330 			rv = drmach_array_next(bp->devices, &d_idx, &d_id);
3331 		}
3332 	}
3333 
3334 	/* otherwise, this proc, wherever it is */
3335 	DRMACH_PR("drmach_mem_cpu_affinity: using default CPU_CURRENT\n");
3336 
3337 	return (CPU_CURRENT);
3338 }
3339 
3340 static sbd_error_t *
3341 drmach_mem_release(drmachid_t id)
3342 {
3343 	if (!DRMACH_IS_MEM_ID(id))
3344 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3345 	return (NULL);
3346 }
3347 
3348 static sbd_error_t *
3349 drmach_mem_status(drmachid_t id, drmach_status_t *stat)
3350 {
3351 	drmach_device_t *dp;
3352 	sbd_error_t	*err;
3353 	uint64_t	 pa, slice_size;
3354 	struct memlist	*ml;
3355 
3356 	ASSERT(DRMACH_IS_MEM_ID(id));
3357 	dp = id;
3358 
3359 	/* get starting physical address of target memory */
3360 	err = drmach_mem_get_base_physaddr(id, &pa);
3361 	if (err)
3362 		return (err);
3363 
3364 	/* round down to slice boundary */
3365 	slice_size = mc_get_mem_alignment();
3366 	pa &= ~ (slice_size - 1);
3367 
3368 	/* stop at first span that is in slice */
3369 	memlist_read_lock();
3370 	for (ml = phys_install; ml; ml = ml->ml_next)
3371 		if (ml->ml_address >= pa && ml->ml_address < pa + slice_size)
3372 			break;
3373 	memlist_read_unlock();
3374 
3375 	stat->assigned = dp->bp->assigned;
3376 	stat->powered = dp->bp->powered;
3377 	stat->configured = (ml != NULL);
3378 	stat->busy = dp->busy;
3379 	(void) strncpy(stat->type, dp->type, sizeof (stat->type));
3380 	stat->info[0] = '\0';
3381 
3382 	return (NULL);
3383 }
3384 
3385 static int
3386 drmach_detach_board(void *arg)
3387 {
3388 	cpuset_t	cset;
3389 	int		retval;
3390 	drmach_board_t	*bp = (drmach_board_t *)arg;
3391 
3392 	cset = cpu_ready_set;
3393 	promsafe_xc_attention(cset);
3394 
3395 	retval = prom_starfire_rm_brd(bp->bnum);
3396 
3397 	xc_dismissed(cset);
3398 
3399 	return (retval);
3400 }
3401 
3402 sbd_error_t *
3403 drmach_board_deprobe(drmachid_t id)
3404 {
3405 	drmach_board_t	*bp;
3406 	int		 retval;
3407 
3408 	if (!DRMACH_IS_BOARD_ID(id))
3409 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3410 	bp = id;
3411 
3412 	cmn_err(CE_CONT, "DR: PROM detach board %d\n", bp->bnum);
3413 
3414 	retval = prom_tree_update(drmach_detach_board, bp);
3415 
3416 	if (retval == 0)
3417 		return (NULL);
3418 	else {
3419 		cmn_err(CE_WARN, "prom error: prom_starfire_rm_brd(%d) "
3420 		    "returned %d", bp->bnum, retval);
3421 		return (drerr_new(1, ESTF_DEPROBE, "%s", bp->cm.name));
3422 	}
3423 }
3424 
3425 /*ARGSUSED*/
3426 static sbd_error_t *
3427 drmach_pt_juggle_bootproc(drmachid_t id, drmach_opts_t *opts)
3428 {
3429 	drmach_device_t	*cpu;
3430 	sbd_error_t	*err;
3431 
3432 	if (!DRMACH_IS_CPU_ID(id))
3433 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3434 	cpu = id;
3435 
3436 	mutex_enter(&cpu_lock);
3437 
3438 	err = drmach_cpu_juggle_bootproc(cpu);
3439 
3440 	mutex_exit(&cpu_lock);
3441 
3442 	return (err);
3443 }
3444 
3445 /*ARGSUSED*/
3446 static sbd_error_t *
3447 drmach_pt_dump_pdainfo(drmachid_t id, drmach_opts_t *opts)
3448 {
3449 	drmach_board_t	*bp;
3450 	int		board;
3451 	int		i;
3452 	pda_handle_t	ph;
3453 	board_desc_t	*bdesc;
3454 
3455 	if (!DRMACH_IS_BOARD_ID(id))
3456 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3457 	bp = id;
3458 	board = bp->bnum;
3459 
3460 	ph = drmach_pda_open();
3461 	if (ph == NULL)
3462 		return (DRMACH_INTERNAL_ERROR());
3463 
3464 	if (pda_board_present(ph, board) == 0) {
3465 		cmn_err(CE_CONT, "board %d is MISSING\n", board);
3466 		pda_close(ph);
3467 		return (DRMACH_INTERNAL_ERROR());
3468 	}
3469 
3470 	cmn_err(CE_CONT, "board %d is PRESENT\n", board);
3471 
3472 	bdesc = (board_desc_t *)pda_get_board_info(ph, board);
3473 	if (bdesc == NULL) {
3474 		cmn_err(CE_CONT,
3475 		    "no board descriptor found for board %d\n",
3476 		    board);
3477 		pda_close(ph);
3478 		return (DRMACH_INTERNAL_ERROR());
3479 	}
3480 
3481 	/* make sure definition in platmod is in sync with pda */
3482 	ASSERT(MAX_PROCMODS == MAX_CPU_UNITS_PER_BOARD);
3483 
3484 	for (i = 0; i < MAX_PROCMODS; i++) {
3485 		if (BDA_NBL(bdesc->bda_proc, i) == BDAN_GOOD)
3486 			cmn_err(CE_CONT,
3487 			    "proc %d.%d PRESENT\n", board, i);
3488 		else
3489 			cmn_err(CE_CONT,
3490 			    "proc %d.%d MISSING\n", board, i);
3491 	}
3492 
3493 	for (i = 0; i < MAX_MGROUPS; i++) {
3494 		if (BDA_NBL(bdesc->bda_mgroup, i) == BDAN_GOOD)
3495 			cmn_err(CE_CONT,
3496 			    "mgroup %d.%d PRESENT\n", board, i);
3497 		else
3498 			cmn_err(CE_CONT,
3499 			    "mgroup %d.%d MISSING\n", board, i);
3500 	}
3501 
3502 	/* make sure definition in platmod is in sync with pda */
3503 	ASSERT(MAX_IOCS == MAX_IO_UNITS_PER_BOARD);
3504 
3505 	for (i = 0; i < MAX_IOCS; i++) {
3506 		int	s;
3507 
3508 		if (BDA_NBL(bdesc->bda_ioc, i) == BDAN_GOOD) {
3509 			cmn_err(CE_CONT,
3510 			    "ioc %d.%d PRESENT\n", board, i);
3511 			for (s = 0; s < MAX_SLOTS_PER_IOC; s++) {
3512 				if (BDA_NBL(bdesc->bda_ios[i], s) != BDAN_GOOD)
3513 					continue;
3514 				cmn_err(CE_CONT,
3515 				    "..scard %d.%d.%d PRESENT\n",
3516 				    board, i, s);
3517 			}
3518 		} else {
3519 			cmn_err(CE_CONT,
3520 			    "ioc %d.%d MISSING\n",
3521 			    board, i);
3522 		}
3523 	}
3524 
3525 	cmn_err(CE_CONT,
3526 	    "board %d memsize = %d pages\n",
3527 	    board, pda_get_mem_size(ph, board));
3528 
3529 	pda_close(ph);
3530 
3531 	return (NULL);
3532 }
3533 
3534 /*ARGSUSED*/
3535 sbd_error_t *
3536 drmach_pt_readmem(drmachid_t id, drmach_opts_t *opts)
3537 {
3538 	struct memlist	*ml;
3539 	uint64_t	src_pa;
3540 	uint64_t	dst_pa;
3541 	uint64_t	dst;
3542 
3543 	dst_pa = va_to_pa(&dst);
3544 
3545 	memlist_read_lock();
3546 	for (ml = phys_install; ml; ml = ml->ml_next) {
3547 		uint64_t	nbytes;
3548 
3549 		src_pa = ml->ml_address;
3550 		nbytes = ml->ml_size;
3551 
3552 		while (nbytes != 0ull) {
3553 
3554 			/* copy 32 bytes at arc_pa to dst_pa */
3555 			bcopy32_il(src_pa, dst_pa);
3556 
3557 			/* increment by 32 bytes */
3558 			src_pa += (4 * sizeof (uint64_t));
3559 
3560 			/* decrement by 32 bytes */
3561 			nbytes -= (4 * sizeof (uint64_t));
3562 		}
3563 	}
3564 	memlist_read_unlock();
3565 
3566 	return (NULL);
3567 }
3568 
3569 static struct {
3570 	const char	*name;
3571 	sbd_error_t	*(*handler)(drmachid_t id, drmach_opts_t *opts);
3572 } drmach_pt_arr[] = {
3573 	{ "juggle",		drmach_pt_juggle_bootproc	},
3574 	{ "pda",		drmach_pt_dump_pdainfo		},
3575 	{ "readmem",		drmach_pt_readmem		},
3576 
3577 	/* the following line must always be last */
3578 	{ NULL,			NULL				}
3579 };
3580 
3581 /*ARGSUSED*/
3582 sbd_error_t *
3583 drmach_passthru(drmachid_t id, drmach_opts_t *opts)
3584 {
3585 	int		i;
3586 	sbd_error_t	*err;
3587 
3588 	i = 0;
3589 	while (drmach_pt_arr[i].name != NULL) {
3590 		int len = strlen(drmach_pt_arr[i].name);
3591 
3592 		if (strncmp(drmach_pt_arr[i].name, opts->copts, len) == 0)
3593 			break;
3594 
3595 		i += 1;
3596 	}
3597 
3598 	if (drmach_pt_arr[i].name == NULL)
3599 		err = drerr_new(0, ESTF_UNKPTCMD, opts->copts);
3600 	else
3601 		err = (*drmach_pt_arr[i].handler)(id, opts);
3602 
3603 	return (err);
3604 }
3605 
3606 sbd_error_t *
3607 drmach_release(drmachid_t id)
3608 {
3609 	drmach_common_t *cp;
3610 	if (!DRMACH_IS_DEVICE_ID(id))
3611 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3612 	cp = id;
3613 
3614 	return (cp->release(id));
3615 }
3616 
3617 sbd_error_t *
3618 drmach_status(drmachid_t id, drmach_status_t *stat)
3619 {
3620 	drmach_common_t *cp;
3621 
3622 	if (!DRMACH_IS_ID(id))
3623 		return (drerr_new(0, ESTF_NOTID, NULL));
3624 	cp = id;
3625 
3626 	return (cp->status(id, stat));
3627 }
3628 
3629 sbd_error_t *
3630 drmach_unconfigure(drmachid_t id, int flags)
3631 {
3632 	drmach_device_t	*dp;
3633 	pnode_t		 nodeid;
3634 	dev_info_t	*dip, *fdip = NULL;
3635 
3636 	if (!DRMACH_IS_DEVICE_ID(id))
3637 		return (drerr_new(0, ESTF_INAPPROP, NULL));
3638 
3639 	dp = id;
3640 
3641 	nodeid = drmach_node_get_dnode(dp->node);
3642 	if (nodeid == OBP_NONODE)
3643 		return (DRMACH_INTERNAL_ERROR());
3644 
3645 	dip = e_ddi_nodeid_to_dip(nodeid);
3646 	if (dip == NULL)
3647 		return (NULL);
3648 
3649 	/*
3650 	 * Branch already held, so hold acquired in
3651 	 * e_ddi_nodeid_to_dip() can be released
3652 	 */
3653 	ddi_release_devi(dip);
3654 
3655 	if (flags & DEVI_BRANCH_DESTROY)
3656 		flags |= DEVI_BRANCH_EVENT;
3657 
3658 	/*
3659 	 * Force flag is no longer necessary. See starcat/io/drmach.c
3660 	 * for details.
3661 	 */
3662 	ASSERT(e_ddi_branch_held(dip));
3663 	if (e_ddi_branch_unconfigure(dip, &fdip, flags)) {
3664 		sbd_error_t	*err;
3665 		char		*path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3666 
3667 		/*
3668 		 * If non-NULL, fdip is returned held and must be released.
3669 		 */
3670 		if (fdip != NULL) {
3671 			(void) ddi_pathname(fdip, path);
3672 			ndi_rele_devi(fdip);
3673 		} else {
3674 			(void) ddi_pathname(dip, path);
3675 		}
3676 
3677 		err = drerr_new(1, ESTF_DRVFAIL, path);
3678 
3679 		kmem_free(path, MAXPATHLEN);
3680 
3681 		return (err);
3682 	}
3683 
3684 	return (NULL);
3685 }
3686 
3687 /*
3688  * drmach interfaces to legacy Starfire platmod logic
3689  * linkage via runtime symbol look up, called from plat_cpu_power*
3690  */
3691 
3692 /*
3693  * Start up a cpu.  It is possible that we're attempting to restart
3694  * the cpu after an UNCONFIGURE in which case the cpu will be
3695  * spinning in its cache.  So, all we have to do is wakeup him up.
3696  * Under normal circumstances the cpu will be coming from a previous
3697  * CONNECT and thus will be spinning in OBP.  In both cases, the
3698  * startup sequence is the same.
3699  */
3700 int
3701 drmach_cpu_poweron(struct cpu *cp)
3702 {
3703 	DRMACH_PR("drmach_cpu_poweron: starting cpuid %d\n", cp->cpu_id);
3704 
3705 	ASSERT(MUTEX_HELD(&cpu_lock));
3706 
3707 	if (drmach_cpu_start(cp) != 0)
3708 		return (EBUSY);
3709 	else
3710 		return (0);
3711 }
3712 
3713 int
3714 drmach_cpu_poweroff(struct cpu *cp)
3715 {
3716 	int		ntries, cnt;
3717 	processorid_t	cpuid = cp->cpu_id;
3718 	void		drmach_cpu_shutdown_self(void);
3719 
3720 	DRMACH_PR("drmach_cpu_poweroff: stopping cpuid %d\n", cp->cpu_id);
3721 
3722 	ASSERT(MUTEX_HELD(&cpu_lock));
3723 
3724 	/*
3725 	 * Capture all CPUs (except for detaching proc) to prevent
3726 	 * crosscalls to the detaching proc until it has cleared its
3727 	 * bit in cpu_ready_set.
3728 	 *
3729 	 * The CPU's remain paused and the prom_mutex is known to be free.
3730 	 * This prevents the x-trap victim from blocking when doing prom
3731 	 * IEEE-1275 calls at a high PIL level.
3732 	 */
3733 	promsafe_pause_cpus();
3734 
3735 	/*
3736 	 * Quiesce interrupts on the target CPU. We do this by setting
3737 	 * the CPU 'not ready'- (i.e. removing the CPU from cpu_ready_set) to
3738 	 * prevent it from receiving cross calls and cross traps.
3739 	 * This prevents the processor from receiving any new soft interrupts.
3740 	 */
3741 	mp_cpu_quiesce(cp);
3742 
3743 	/* setup xt_mb, will be cleared by drmach_shutdown_asm when ready */
3744 	drmach_xt_mb[cpuid] = 0x80;
3745 
3746 	xt_one_unchecked(cpuid, (xcfunc_t *)idle_stop_xcall,
3747 	    (uint64_t)drmach_cpu_shutdown_self, NULL);
3748 
3749 	ntries = drmach_cpu_ntries;
3750 	cnt = 0;
3751 	while (drmach_xt_mb[cpuid] && ntries) {
3752 		DELAY(drmach_cpu_delay);
3753 		ntries--;
3754 		cnt++;
3755 	}
3756 
3757 	drmach_xt_mb[cpuid] = 0;	/* steal the cache line back */
3758 
3759 	start_cpus();
3760 
3761 	DRMACH_PR("waited %d out of %d tries for "
3762 	    "drmach_cpu_shutdown_self on cpu%d",
3763 	    drmach_cpu_ntries - ntries, drmach_cpu_ntries, cp->cpu_id);
3764 
3765 	drmach_cpu_obp_detach(cpuid);
3766 
3767 	CPU_SIGNATURE(OS_SIG, SIGST_DETACHED, SIGSUBST_NULL, cpuid);
3768 
3769 	return (0);
3770 }
3771 
3772 /*ARGSUSED*/
3773 int
3774 drmach_verify_sr(dev_info_t *dip, int sflag)
3775 {
3776 	return (0);
3777 }
3778 
3779 void
3780 drmach_suspend_last(void)
3781 {
3782 }
3783 
3784 void
3785 drmach_resume_first(void)
3786 {
3787 }
3788 
3789 /*
3790  * Log a DR sysevent.
3791  * Return value: 0 success, non-zero failure.
3792  */
3793 int
3794 drmach_log_sysevent(int board, char *hint, int flag, int verbose)
3795 {
3796 	sysevent_t			*ev;
3797 	sysevent_id_t			eid;
3798 	int				rv, km_flag;
3799 	sysevent_value_t		evnt_val;
3800 	sysevent_attr_list_t		*evnt_attr_list = NULL;
3801 	char				attach_pnt[MAXNAMELEN];
3802 
3803 	km_flag = (flag == SE_SLEEP) ? KM_SLEEP : KM_NOSLEEP;
3804 	attach_pnt[0] = '\0';
3805 	if (drmach_board_name(board, attach_pnt, MAXNAMELEN)) {
3806 		rv = -1;
3807 		goto logexit;
3808 	}
3809 	if (verbose)
3810 		DRMACH_PR("drmach_log_sysevent: %s %s, flag: %d, verbose: %d\n",
3811 		    attach_pnt, hint, flag, verbose);
3812 
3813 	if ((ev = sysevent_alloc(EC_DR, ESC_DR_AP_STATE_CHANGE,
3814 	    SUNW_KERN_PUB"dr", km_flag)) == NULL) {
3815 		rv = -2;
3816 		goto logexit;
3817 	}
3818 	evnt_val.value_type = SE_DATA_TYPE_STRING;
3819 	evnt_val.value.sv_string = attach_pnt;
3820 	if ((rv = sysevent_add_attr(&evnt_attr_list, DR_AP_ID,
3821 	    &evnt_val, km_flag)) != 0)
3822 		goto logexit;
3823 
3824 	evnt_val.value_type = SE_DATA_TYPE_STRING;
3825 	evnt_val.value.sv_string = hint;
3826 	if ((rv = sysevent_add_attr(&evnt_attr_list, DR_HINT,
3827 	    &evnt_val, km_flag)) != 0) {
3828 		sysevent_free_attr(evnt_attr_list);
3829 		goto logexit;
3830 	}
3831 
3832 	(void) sysevent_attach_attributes(ev, evnt_attr_list);
3833 
3834 	/*
3835 	 * Log the event but do not sleep waiting for its
3836 	 * delivery. This provides insulation from syseventd.
3837 	 */
3838 	rv = log_sysevent(ev, SE_NOSLEEP, &eid);
3839 
3840 logexit:
3841 	if (ev)
3842 		sysevent_free(ev);
3843 	if ((rv != 0) && verbose)
3844 		cmn_err(CE_WARN,
3845 		    "drmach_log_sysevent failed (rv %d) for %s  %s\n",
3846 		    rv, attach_pnt, hint);
3847 
3848 	return (rv);
3849 }
3850 
3851 /*ARGSUSED*/
3852 int
3853 drmach_allow_memrange_modify(drmachid_t id)
3854 {
3855 	return (1);	/* TRUE */
3856 }
3857