xref: /titanic_50/usr/src/uts/common/os/devcfg.c (revision d2b5b2d357ee3172eacb6860be1891259902203d)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 #include <sys/note.h>
26 #include <sys/t_lock.h>
27 #include <sys/cmn_err.h>
28 #include <sys/instance.h>
29 #include <sys/conf.h>
30 #include <sys/stat.h>
31 #include <sys/ddi.h>
32 #include <sys/hwconf.h>
33 #include <sys/sunddi.h>
34 #include <sys/sunndi.h>
35 #include <sys/ddi_impldefs.h>
36 #include <sys/ndi_impldefs.h>
37 #include <sys/modctl.h>
38 #include <sys/contract/device_impl.h>
39 #include <sys/dacf.h>
40 #include <sys/promif.h>
41 #include <sys/pci.h>
42 #include <sys/cpuvar.h>
43 #include <sys/pathname.h>
44 #include <sys/taskq.h>
45 #include <sys/sysevent.h>
46 #include <sys/sunmdi.h>
47 #include <sys/stream.h>
48 #include <sys/strsubr.h>
49 #include <sys/fs/snode.h>
50 #include <sys/fs/dv_node.h>
51 #include <sys/reboot.h>
52 #include <sys/sysmacros.h>
53 #include <sys/systm.h>
54 #include <sys/fs/sdev_impl.h>
55 #include <sys/sunldi.h>
56 #include <sys/sunldi_impl.h>
57 #include <sys/bootprops.h>
58 #include <sys/varargs.h>
59 #include <sys/modhash.h>
60 #include <sys/instance.h>
61 
62 #if defined(__amd64) && !defined(__xpv)
63 #include <sys/iommulib.h>
64 #endif
65 
66 #ifdef DEBUG
67 int ddidebug = DDI_AUDIT;
68 #else
69 int ddidebug = 0;
70 #endif
71 
72 #define	MT_CONFIG_OP	0
73 #define	MT_UNCONFIG_OP	1
74 
75 /* Multi-threaded configuration */
76 struct mt_config_handle {
77 	kmutex_t mtc_lock;
78 	kcondvar_t mtc_cv;
79 	int mtc_thr_count;
80 	dev_info_t *mtc_pdip;	/* parent dip for mt_config_children */
81 	dev_info_t **mtc_fdip;	/* "a" dip where unconfigure failed */
82 	major_t mtc_parmajor;	/* parent major for mt_config_driver */
83 	major_t mtc_major;
84 	int mtc_flags;
85 	int mtc_op;		/* config or unconfig */
86 	int mtc_error;		/* operation error */
87 	struct brevq_node **mtc_brevqp;	/* outstanding branch events queue */
88 #ifdef DEBUG
89 	int total_time;
90 	timestruc_t start_time;
91 #endif /* DEBUG */
92 };
93 
94 struct devi_nodeid {
95 	pnode_t nodeid;
96 	dev_info_t *dip;
97 	struct devi_nodeid *next;
98 };
99 
100 struct devi_nodeid_list {
101 	kmutex_t dno_lock;		/* Protects other fields */
102 	struct devi_nodeid *dno_head;	/* list of devi nodeid elements */
103 	struct devi_nodeid *dno_free;	/* Free list */
104 	uint_t dno_list_length;		/* number of dips in list */
105 };
106 
107 /* used to keep track of branch remove events to be generated */
108 struct brevq_node {
109 	char *brn_deviname;
110 	struct brevq_node *brn_sibling;
111 	struct brevq_node *brn_child;
112 };
113 
114 static struct devi_nodeid_list devi_nodeid_list;
115 static struct devi_nodeid_list *devimap = &devi_nodeid_list;
116 
117 /*
118  * Well known nodes which are attached first at boot time.
119  */
120 dev_info_t *top_devinfo;		/* root of device tree */
121 dev_info_t *options_dip;
122 dev_info_t *pseudo_dip;
123 dev_info_t *clone_dip;
124 dev_info_t *scsi_vhci_dip;		/* MPXIO dip */
125 major_t clone_major;
126 
127 /*
128  * A non-global zone's /dev is derived from the device tree.
129  * This generation number serves to indicate when a zone's
130  * /dev may need to be updated.
131  */
132 volatile ulong_t devtree_gen;		/* generation number */
133 
134 /* block all future dev_info state changes */
135 hrtime_t volatile devinfo_freeze = 0;
136 
137 /* number of dev_info attaches/detaches currently in progress */
138 static ulong_t devinfo_attach_detach = 0;
139 
140 extern int	sys_shutdown;
141 extern kmutex_t global_vhci_lock;
142 
143 /* bitset of DS_SYSAVAIL & DS_RECONFIG - no races, no lock */
144 static int devname_state = 0;
145 
146 /*
147  * The devinfo snapshot cache and related variables.
148  * The only field in the di_cache structure that needs initialization
149  * is the mutex (cache_lock). However, since this is an adaptive mutex
150  * (MUTEX_DEFAULT) - it is automatically initialized by being allocated
151  * in zeroed memory (static storage class). Therefore no explicit
152  * initialization of the di_cache structure is needed.
153  */
154 struct di_cache	di_cache = {1};
155 int		di_cache_debug = 0;
156 
157 /* For ddvis, which needs pseudo children under PCI */
158 int pci_allow_pseudo_children = 0;
159 
160 /* Allow path-oriented alias driver binding on driver.conf enumerated nodes */
161 int driver_conf_allow_path_alias = 1;
162 
163 /*
164  * The following switch is for service people, in case a
165  * 3rd party driver depends on identify(9e) being called.
166  */
167 int identify_9e = 0;
168 
169 /*
170  * Add flag so behaviour of preventing attach for retired persistant nodes
171  * can be disabled.
172  */
173 int retire_prevents_attach = 1;
174 
175 int mtc_off;					/* turn off mt config */
176 
177 int quiesce_debug = 0;
178 
179 boolean_t ddi_aliases_present = B_FALSE;
180 ddi_alias_t ddi_aliases;
181 uint_t tsd_ddi_redirect;
182 
183 #define	DDI_ALIAS_HASH_SIZE	(2700)
184 
185 static kmem_cache_t *ddi_node_cache;		/* devinfo node cache */
186 static devinfo_log_header_t *devinfo_audit_log;	/* devinfo log */
187 static int devinfo_log_size;			/* size in pages */
188 
189 boolean_t ddi_err_panic = B_FALSE;
190 
191 static int lookup_compatible(dev_info_t *, uint_t);
192 static char *encode_composite_string(char **, uint_t, size_t *, uint_t);
193 static void link_to_driver_list(dev_info_t *);
194 static void unlink_from_driver_list(dev_info_t *);
195 static void add_to_dn_list(struct devnames *, dev_info_t *);
196 static void remove_from_dn_list(struct devnames *, dev_info_t *);
197 static dev_info_t *find_duplicate_child();
198 static void add_global_props(dev_info_t *);
199 static void remove_global_props(dev_info_t *);
200 static int uninit_node(dev_info_t *);
201 static void da_log_init(void);
202 static void da_log_enter(dev_info_t *);
203 static int walk_devs(dev_info_t *, int (*f)(dev_info_t *, void *), void *, int);
204 static int reset_nexus_flags(dev_info_t *, void *);
205 static void ddi_optimize_dtree(dev_info_t *);
206 static int is_leaf_node(dev_info_t *);
207 static struct mt_config_handle *mt_config_init(dev_info_t *, dev_info_t **,
208     int, major_t, int, struct brevq_node **);
209 static void mt_config_children(struct mt_config_handle *);
210 static void mt_config_driver(struct mt_config_handle *);
211 static int mt_config_fini(struct mt_config_handle *);
212 static int devi_unconfig_common(dev_info_t *, dev_info_t **, int, major_t,
213     struct brevq_node **);
214 static int
215 ndi_devi_config_obp_args(dev_info_t *parent, char *devnm,
216     dev_info_t **childp, int flags);
217 static void i_link_vhci_node(dev_info_t *);
218 static void ndi_devi_exit_and_wait(dev_info_t *dip,
219     int circular, clock_t end_time);
220 static int ndi_devi_unbind_driver(dev_info_t *dip);
221 
222 static int i_ddi_check_retire(dev_info_t *dip);
223 
224 static void quiesce_one_device(dev_info_t *, void *);
225 
226 dev_info_t *ddi_alias_redirect(char *alias);
227 char *ddi_curr_redirect(char *currpath);
228 
229 
230 /*
231  * dev_info cache and node management
232  */
233 
234 /* initialize dev_info node cache */
235 void
236 i_ddi_node_cache_init()
237 {
238 	ASSERT(ddi_node_cache == NULL);
239 	ddi_node_cache = kmem_cache_create("dev_info_node_cache",
240 	    sizeof (struct dev_info), 0, NULL, NULL, NULL, NULL, NULL, 0);
241 
242 	if (ddidebug & DDI_AUDIT)
243 		da_log_init();
244 }
245 
246 
247 /*
248  * Allocating a dev_info node, callable from interrupt context with KM_NOSLEEP
249  * The allocated node has a reference count of 0.
250  */
251 dev_info_t *
252 i_ddi_alloc_node(dev_info_t *pdip, char *node_name, pnode_t nodeid,
253     int instance, ddi_prop_t *sys_prop, int flag)
254 {
255 	struct dev_info *devi;
256 	struct devi_nodeid *elem;
257 	static char failed[] = "i_ddi_alloc_node: out of memory";
258 
259 	ASSERT(node_name != NULL);
260 
261 	if ((devi = kmem_cache_alloc(ddi_node_cache, flag)) == NULL) {
262 		cmn_err(CE_NOTE, failed);
263 		return (NULL);
264 	}
265 
266 	bzero(devi, sizeof (struct dev_info));
267 
268 	if (devinfo_audit_log) {
269 		devi->devi_audit = kmem_zalloc(sizeof (devinfo_audit_t), flag);
270 		if (devi->devi_audit == NULL)
271 			goto fail;
272 	}
273 
274 	if ((devi->devi_node_name = i_ddi_strdup(node_name, flag)) == NULL)
275 		goto fail;
276 
277 	/* default binding name is node name */
278 	devi->devi_binding_name = devi->devi_node_name;
279 	devi->devi_major = DDI_MAJOR_T_NONE;	/* unbound by default */
280 
281 	/*
282 	 * Make a copy of system property
283 	 */
284 	if (sys_prop &&
285 	    (devi->devi_sys_prop_ptr = i_ddi_prop_list_dup(sys_prop, flag))
286 	    == NULL)
287 		goto fail;
288 
289 	/*
290 	 * Assign devi_nodeid, devi_node_class, devi_node_attributes
291 	 * according to the following algorithm:
292 	 *
293 	 * nodeid arg			node class		node attributes
294 	 *
295 	 * DEVI_PSEUDO_NODEID		DDI_NC_PSEUDO		A
296 	 * DEVI_SID_NODEID		DDI_NC_PSEUDO		A,P
297 	 * DEVI_SID_HIDDEN_NODEID	DDI_NC_PSEUDO		A,P,H
298 	 * DEVI_SID_HP_NODEID		DDI_NC_PSEUDO		A,P,h
299 	 * DEVI_SID_HP_HIDDEN_NODEID	DDI_NC_PSEUDO		A,P,H,h
300 	 * other			DDI_NC_PROM		P
301 	 *
302 	 * Where A = DDI_AUTO_ASSIGNED_NODEID (auto-assign a nodeid)
303 	 * and	 P = DDI_PERSISTENT
304 	 * and	 H = DDI_HIDDEN_NODE
305 	 * and	 h = DDI_HOTPLUG_NODE
306 	 *
307 	 * auto-assigned nodeids are also auto-freed.
308 	 */
309 	devi->devi_node_attributes = 0;
310 	switch (nodeid) {
311 	case DEVI_SID_HIDDEN_NODEID:
312 		devi->devi_node_attributes |= DDI_HIDDEN_NODE;
313 		goto sid;
314 
315 	case DEVI_SID_HP_NODEID:
316 		devi->devi_node_attributes |= DDI_HOTPLUG_NODE;
317 		goto sid;
318 
319 	case DEVI_SID_HP_HIDDEN_NODEID:
320 		devi->devi_node_attributes |= DDI_HIDDEN_NODE;
321 		devi->devi_node_attributes |= DDI_HOTPLUG_NODE;
322 		goto sid;
323 
324 	case DEVI_SID_NODEID:
325 sid:		devi->devi_node_attributes |= DDI_PERSISTENT;
326 		if ((elem = kmem_zalloc(sizeof (*elem), flag)) == NULL)
327 			goto fail;
328 		/*FALLTHROUGH*/
329 
330 	case DEVI_PSEUDO_NODEID:
331 		devi->devi_node_attributes |= DDI_AUTO_ASSIGNED_NODEID;
332 		devi->devi_node_class = DDI_NC_PSEUDO;
333 		if (impl_ddi_alloc_nodeid(&devi->devi_nodeid)) {
334 			panic("i_ddi_alloc_node: out of nodeids");
335 			/*NOTREACHED*/
336 		}
337 		break;
338 
339 	default:
340 		if ((elem = kmem_zalloc(sizeof (*elem), flag)) == NULL)
341 			goto fail;
342 
343 		/*
344 		 * the nodetype is 'prom', try to 'take' the nodeid now.
345 		 * This requires memory allocation, so check for failure.
346 		 */
347 		if (impl_ddi_take_nodeid(nodeid, flag) != 0) {
348 			kmem_free(elem, sizeof (*elem));
349 			goto fail;
350 		}
351 
352 		devi->devi_nodeid = nodeid;
353 		devi->devi_node_class = DDI_NC_PROM;
354 		devi->devi_node_attributes = DDI_PERSISTENT;
355 		break;
356 	}
357 
358 	if (ndi_dev_is_persistent_node((dev_info_t *)devi)) {
359 		mutex_enter(&devimap->dno_lock);
360 		elem->next = devimap->dno_free;
361 		devimap->dno_free = elem;
362 		mutex_exit(&devimap->dno_lock);
363 	}
364 
365 	/*
366 	 * Instance is normally initialized to -1. In a few special
367 	 * cases, the caller may specify an instance (e.g. CPU nodes).
368 	 */
369 	devi->devi_instance = instance;
370 
371 	/*
372 	 * set parent and bus_ctl parent
373 	 */
374 	devi->devi_parent = DEVI(pdip);
375 	devi->devi_bus_ctl = DEVI(pdip);
376 
377 	NDI_CONFIG_DEBUG((CE_CONT,
378 	    "i_ddi_alloc_node: name=%s id=%d\n", node_name, devi->devi_nodeid));
379 
380 	cv_init(&(devi->devi_cv), NULL, CV_DEFAULT, NULL);
381 	mutex_init(&(devi->devi_lock), NULL, MUTEX_DEFAULT, NULL);
382 	mutex_init(&(devi->devi_pm_lock), NULL, MUTEX_DEFAULT, NULL);
383 	mutex_init(&(devi->devi_pm_busy_lock), NULL, MUTEX_DEFAULT, NULL);
384 
385 	RIO_TRACE((CE_NOTE, "i_ddi_alloc_node: Initing contract fields: "
386 	    "dip=%p, name=%s", (void *)devi, node_name));
387 
388 	mutex_init(&(devi->devi_ct_lock), NULL, MUTEX_DEFAULT, NULL);
389 	cv_init(&(devi->devi_ct_cv), NULL, CV_DEFAULT, NULL);
390 	devi->devi_ct_count = -1;	/* counter not in use if -1 */
391 	list_create(&(devi->devi_ct), sizeof (cont_device_t),
392 	    offsetof(cont_device_t, cond_next));
393 
394 	i_ddi_set_node_state((dev_info_t *)devi, DS_PROTO);
395 	da_log_enter((dev_info_t *)devi);
396 	return ((dev_info_t *)devi);
397 
398 fail:
399 	if (devi->devi_sys_prop_ptr)
400 		i_ddi_prop_list_delete(devi->devi_sys_prop_ptr);
401 	if (devi->devi_node_name)
402 		kmem_free(devi->devi_node_name, strlen(node_name) + 1);
403 	if (devi->devi_audit)
404 		kmem_free(devi->devi_audit, sizeof (devinfo_audit_t));
405 	kmem_cache_free(ddi_node_cache, devi);
406 	cmn_err(CE_NOTE, failed);
407 	return (NULL);
408 }
409 
410 /*
411  * free a dev_info structure.
412  * NB. Not callable from interrupt since impl_ddi_free_nodeid may block.
413  */
414 void
415 i_ddi_free_node(dev_info_t *dip)
416 {
417 	struct dev_info *devi = DEVI(dip);
418 	struct devi_nodeid *elem;
419 
420 	ASSERT(devi->devi_ref == 0);
421 	ASSERT(devi->devi_addr == NULL);
422 	ASSERT(devi->devi_node_state == DS_PROTO);
423 	ASSERT(devi->devi_child == NULL);
424 	ASSERT(devi->devi_hp_hdlp == NULL);
425 
426 	/* free devi_addr_buf allocated by ddi_set_name_addr() */
427 	if (devi->devi_addr_buf)
428 		kmem_free(devi->devi_addr_buf, 2 * MAXNAMELEN);
429 
430 	if (i_ndi_dev_is_auto_assigned_node(dip))
431 		impl_ddi_free_nodeid(DEVI(dip)->devi_nodeid);
432 
433 	if (ndi_dev_is_persistent_node(dip)) {
434 		mutex_enter(&devimap->dno_lock);
435 		ASSERT(devimap->dno_free);
436 		elem = devimap->dno_free;
437 		devimap->dno_free = elem->next;
438 		mutex_exit(&devimap->dno_lock);
439 		kmem_free(elem, sizeof (*elem));
440 	}
441 
442 	if (DEVI(dip)->devi_compat_names)
443 		kmem_free(DEVI(dip)->devi_compat_names,
444 		    DEVI(dip)->devi_compat_length);
445 	if (DEVI(dip)->devi_rebinding_name)
446 		kmem_free(DEVI(dip)->devi_rebinding_name,
447 		    strlen(DEVI(dip)->devi_rebinding_name) + 1);
448 
449 	ddi_prop_remove_all(dip);	/* remove driver properties */
450 	if (devi->devi_sys_prop_ptr)
451 		i_ddi_prop_list_delete(devi->devi_sys_prop_ptr);
452 	if (devi->devi_hw_prop_ptr)
453 		i_ddi_prop_list_delete(devi->devi_hw_prop_ptr);
454 
455 	if (DEVI(dip)->devi_devid_str)
456 		ddi_devid_str_free(DEVI(dip)->devi_devid_str);
457 
458 	i_ddi_set_node_state(dip, DS_INVAL);
459 	da_log_enter(dip);
460 	if (devi->devi_audit) {
461 		kmem_free(devi->devi_audit, sizeof (devinfo_audit_t));
462 	}
463 	if (devi->devi_device_class)
464 		kmem_free(devi->devi_device_class,
465 		    strlen(devi->devi_device_class) + 1);
466 	cv_destroy(&(devi->devi_cv));
467 	mutex_destroy(&(devi->devi_lock));
468 	mutex_destroy(&(devi->devi_pm_lock));
469 	mutex_destroy(&(devi->devi_pm_busy_lock));
470 
471 	RIO_TRACE((CE_NOTE, "i_ddi_free_node: destroying contract fields: "
472 	    "dip=%p", (void *)dip));
473 	contract_device_remove_dip(dip);
474 	ASSERT(devi->devi_ct_count == -1);
475 	ASSERT(list_is_empty(&(devi->devi_ct)));
476 	cv_destroy(&(devi->devi_ct_cv));
477 	list_destroy(&(devi->devi_ct));
478 	/* free this last since contract_device_remove_dip() uses it */
479 	mutex_destroy(&(devi->devi_ct_lock));
480 	RIO_TRACE((CE_NOTE, "i_ddi_free_node: destroyed all contract fields: "
481 	    "dip=%p, name=%s", (void *)dip, devi->devi_node_name));
482 
483 	kmem_free(devi->devi_node_name, strlen(devi->devi_node_name) + 1);
484 
485 	/* free event data */
486 	if (devi->devi_ev_path)
487 		kmem_free(devi->devi_ev_path, MAXPATHLEN);
488 
489 	kmem_cache_free(ddi_node_cache, devi);
490 }
491 
492 
493 /*
494  * Node state transitions
495  */
496 
497 /*
498  * Change the node name
499  */
500 int
501 ndi_devi_set_nodename(dev_info_t *dip, char *name, int flags)
502 {
503 	_NOTE(ARGUNUSED(flags))
504 	char *nname, *oname;
505 
506 	ASSERT(dip && name);
507 
508 	oname = DEVI(dip)->devi_node_name;
509 	if (strcmp(oname, name) == 0)
510 		return (DDI_SUCCESS);
511 
512 	/*
513 	 * pcicfg_fix_ethernet requires a name change after node
514 	 * is linked into the tree. When pcicfg is fixed, we
515 	 * should only allow name change in DS_PROTO state.
516 	 */
517 	if (i_ddi_node_state(dip) >= DS_BOUND) {
518 		/*
519 		 * Don't allow name change once node is bound
520 		 */
521 		cmn_err(CE_NOTE,
522 		    "ndi_devi_set_nodename: node already bound dip = %p,"
523 		    " %s -> %s", (void *)dip, ddi_node_name(dip), name);
524 		return (NDI_FAILURE);
525 	}
526 
527 	nname = i_ddi_strdup(name, KM_SLEEP);
528 	DEVI(dip)->devi_node_name = nname;
529 	i_ddi_set_binding_name(dip, nname);
530 	kmem_free(oname, strlen(oname) + 1);
531 
532 	da_log_enter(dip);
533 	return (NDI_SUCCESS);
534 }
535 
536 void
537 i_ddi_add_devimap(dev_info_t *dip)
538 {
539 	struct devi_nodeid *elem;
540 
541 	ASSERT(dip);
542 
543 	if (!ndi_dev_is_persistent_node(dip))
544 		return;
545 
546 	ASSERT(ddi_get_parent(dip) == NULL || (DEVI_VHCI_NODE(dip)) ||
547 	    DEVI_BUSY_OWNED(ddi_get_parent(dip)));
548 
549 	mutex_enter(&devimap->dno_lock);
550 
551 	ASSERT(devimap->dno_free);
552 
553 	elem = devimap->dno_free;
554 	devimap->dno_free = elem->next;
555 
556 	elem->nodeid = ddi_get_nodeid(dip);
557 	elem->dip = dip;
558 	elem->next = devimap->dno_head;
559 	devimap->dno_head = elem;
560 
561 	devimap->dno_list_length++;
562 
563 	mutex_exit(&devimap->dno_lock);
564 }
565 
566 static int
567 i_ddi_remove_devimap(dev_info_t *dip)
568 {
569 	struct devi_nodeid *prev, *elem;
570 	static const char *fcn = "i_ddi_remove_devimap";
571 
572 	ASSERT(dip);
573 
574 	if (!ndi_dev_is_persistent_node(dip))
575 		return (DDI_SUCCESS);
576 
577 	mutex_enter(&devimap->dno_lock);
578 
579 	/*
580 	 * The following check is done with dno_lock held
581 	 * to prevent race between dip removal and
582 	 * e_ddi_prom_node_to_dip()
583 	 */
584 	if (e_ddi_devi_holdcnt(dip)) {
585 		mutex_exit(&devimap->dno_lock);
586 		return (DDI_FAILURE);
587 	}
588 
589 	ASSERT(devimap->dno_head);
590 	ASSERT(devimap->dno_list_length > 0);
591 
592 	prev = NULL;
593 	for (elem = devimap->dno_head; elem; elem = elem->next) {
594 		if (elem->dip == dip) {
595 			ASSERT(elem->nodeid == ddi_get_nodeid(dip));
596 			break;
597 		}
598 		prev = elem;
599 	}
600 
601 	if (elem && prev)
602 		prev->next = elem->next;
603 	else if (elem)
604 		devimap->dno_head = elem->next;
605 	else
606 		panic("%s: devinfo node(%p) not found",
607 		    fcn, (void *)dip);
608 
609 	devimap->dno_list_length--;
610 
611 	elem->nodeid = 0;
612 	elem->dip = NULL;
613 
614 	elem->next = devimap->dno_free;
615 	devimap->dno_free = elem;
616 
617 	mutex_exit(&devimap->dno_lock);
618 
619 	return (DDI_SUCCESS);
620 }
621 
622 /*
623  * Link this node into the devinfo tree and add to orphan list
624  * Not callable from interrupt context
625  */
626 static void
627 link_node(dev_info_t *dip)
628 {
629 	struct dev_info *devi = DEVI(dip);
630 	struct dev_info *parent = devi->devi_parent;
631 	dev_info_t **dipp;
632 
633 	ASSERT(parent);	/* never called for root node */
634 
635 	NDI_CONFIG_DEBUG((CE_CONT, "link_node: parent = %s child = %s\n",
636 	    parent->devi_node_name, devi->devi_node_name));
637 
638 	/*
639 	 * Hold the global_vhci_lock before linking any direct
640 	 * children of rootnex driver. This special lock protects
641 	 * linking and unlinking for rootnext direct children.
642 	 */
643 	if ((dev_info_t *)parent == ddi_root_node())
644 		mutex_enter(&global_vhci_lock);
645 
646 	/*
647 	 * attach the node to end of the list unless the node is already there
648 	 */
649 	dipp = (dev_info_t **)(&DEVI(parent)->devi_child);
650 	while (*dipp && (*dipp != dip)) {
651 		dipp = (dev_info_t **)(&DEVI(*dipp)->devi_sibling);
652 	}
653 	ASSERT(*dipp == NULL);	/* node is not linked */
654 
655 	/*
656 	 * Now that we are in the tree, update the devi-nodeid map.
657 	 */
658 	i_ddi_add_devimap(dip);
659 
660 	/*
661 	 * This is a temporary workaround for Bug 4618861.
662 	 * We keep the scsi_vhci nexus node on the left side of the devinfo
663 	 * tree (under the root nexus driver), so that virtual nodes under
664 	 * scsi_vhci will be SUSPENDed first and RESUMEd last.	This ensures
665 	 * that the pHCI nodes are active during times when their clients
666 	 * may be depending on them.  This workaround embodies the knowledge
667 	 * that system PM and CPR both traverse the tree left-to-right during
668 	 * SUSPEND and right-to-left during RESUME.
669 	 * Extending the workaround to IB Nexus/VHCI
670 	 * driver also.
671 	 */
672 	if (strcmp(devi->devi_binding_name, "scsi_vhci") == 0) {
673 		/* Add scsi_vhci to beginning of list */
674 		ASSERT((dev_info_t *)parent == top_devinfo);
675 		/* scsi_vhci under rootnex */
676 		devi->devi_sibling = parent->devi_child;
677 		parent->devi_child = devi;
678 	} else if (strcmp(devi->devi_binding_name, "ib") == 0) {
679 		i_link_vhci_node(dip);
680 	} else {
681 		/* Add to end of list */
682 		*dipp = dip;
683 		DEVI(dip)->devi_sibling = NULL;
684 	}
685 
686 	/*
687 	 * Release the global_vhci_lock before linking any direct
688 	 * children of rootnex driver.
689 	 */
690 	if ((dev_info_t *)parent == ddi_root_node())
691 		mutex_exit(&global_vhci_lock);
692 
693 	/* persistent nodes go on orphan list */
694 	if (ndi_dev_is_persistent_node(dip))
695 		add_to_dn_list(&orphanlist, dip);
696 }
697 
698 /*
699  * Unlink this node from the devinfo tree
700  */
701 static int
702 unlink_node(dev_info_t *dip)
703 {
704 	struct dev_info *devi = DEVI(dip);
705 	struct dev_info *parent = devi->devi_parent;
706 	dev_info_t **dipp;
707 	ddi_hp_cn_handle_t *hdlp;
708 
709 	ASSERT(parent != NULL);
710 	ASSERT(devi->devi_node_state == DS_LINKED);
711 
712 	NDI_CONFIG_DEBUG((CE_CONT, "unlink_node: name = %s\n",
713 	    ddi_node_name(dip)));
714 
715 	/* check references */
716 	if (devi->devi_ref || i_ddi_remove_devimap(dip) != DDI_SUCCESS)
717 		return (DDI_FAILURE);
718 
719 	/*
720 	 * Hold the global_vhci_lock before linking any direct
721 	 * children of rootnex driver.
722 	 */
723 	if ((dev_info_t *)parent == ddi_root_node())
724 		mutex_enter(&global_vhci_lock);
725 
726 	dipp = (dev_info_t **)(&DEVI(parent)->devi_child);
727 	while (*dipp && (*dipp != dip)) {
728 		dipp = (dev_info_t **)(&DEVI(*dipp)->devi_sibling);
729 	}
730 	if (*dipp) {
731 		*dipp = (dev_info_t *)(devi->devi_sibling);
732 		devi->devi_sibling = NULL;
733 	} else {
734 		NDI_CONFIG_DEBUG((CE_NOTE, "unlink_node: %s not linked",
735 		    devi->devi_node_name));
736 	}
737 
738 	/*
739 	 * Release the global_vhci_lock before linking any direct
740 	 * children of rootnex driver.
741 	 */
742 	if ((dev_info_t *)parent == ddi_root_node())
743 		mutex_exit(&global_vhci_lock);
744 
745 	/* Remove node from orphan list */
746 	if (ndi_dev_is_persistent_node(dip)) {
747 		remove_from_dn_list(&orphanlist, dip);
748 	}
749 
750 	/* Update parent's hotplug handle list */
751 	for (hdlp = DEVI(parent)->devi_hp_hdlp; hdlp; hdlp = hdlp->next) {
752 		if (hdlp->cn_info.cn_child == dip)
753 			hdlp->cn_info.cn_child = NULL;
754 	}
755 	return (DDI_SUCCESS);
756 }
757 
758 /*
759  * Bind this devinfo node to a driver. If compat is NON-NULL, try that first.
760  * Else, use the node-name.
761  *
762  * NOTE: IEEE1275 specifies that nodename should be tried before compatible.
763  *	Solaris implementation binds nodename after compatible.
764  *
765  * If we find a binding,
766  * - set the binding name to the string,
767  * - set major number to driver major
768  *
769  * If we don't find a binding,
770  * - return failure
771  */
772 static int
773 bind_node(dev_info_t *dip)
774 {
775 	char *p = NULL;
776 	major_t major = DDI_MAJOR_T_NONE;
777 	struct dev_info *devi = DEVI(dip);
778 	dev_info_t *parent = ddi_get_parent(dip);
779 
780 	ASSERT(devi->devi_node_state == DS_LINKED);
781 
782 	NDI_CONFIG_DEBUG((CE_CONT, "bind_node: 0x%p(name = %s)\n",
783 	    (void *)dip, ddi_node_name(dip)));
784 
785 	mutex_enter(&DEVI(dip)->devi_lock);
786 	if (DEVI(dip)->devi_flags & DEVI_NO_BIND) {
787 		mutex_exit(&DEVI(dip)->devi_lock);
788 		return (DDI_FAILURE);
789 	}
790 	mutex_exit(&DEVI(dip)->devi_lock);
791 
792 	/* find the driver with most specific binding using compatible */
793 	major = ddi_compatible_driver_major(dip, &p);
794 	if (major == DDI_MAJOR_T_NONE)
795 		return (DDI_FAILURE);
796 
797 	devi->devi_major = major;
798 	if (p != NULL) {
799 		i_ddi_set_binding_name(dip, p);
800 		NDI_CONFIG_DEBUG((CE_CONT, "bind_node: %s bound to %s\n",
801 		    devi->devi_node_name, p));
802 	}
803 
804 	/* Link node to per-driver list */
805 	link_to_driver_list(dip);
806 
807 	/*
808 	 * reset parent flag so that nexus will merge .conf props
809 	 */
810 	if (ndi_dev_is_persistent_node(dip)) {
811 		mutex_enter(&DEVI(parent)->devi_lock);
812 		DEVI(parent)->devi_flags &=
813 		    ~(DEVI_ATTACHED_CHILDREN|DEVI_MADE_CHILDREN);
814 		mutex_exit(&DEVI(parent)->devi_lock);
815 	}
816 	return (DDI_SUCCESS);
817 }
818 
819 /*
820  * Unbind this devinfo node
821  * Called before the node is destroyed or driver is removed from system
822  */
823 static int
824 unbind_node(dev_info_t *dip)
825 {
826 	ASSERT(DEVI(dip)->devi_node_state == DS_BOUND);
827 	ASSERT(DEVI(dip)->devi_major != DDI_MAJOR_T_NONE);
828 
829 	/* check references */
830 	if (DEVI(dip)->devi_ref)
831 		return (DDI_FAILURE);
832 
833 	NDI_CONFIG_DEBUG((CE_CONT, "unbind_node: 0x%p(name = %s)\n",
834 	    (void *)dip, ddi_node_name(dip)));
835 
836 	unlink_from_driver_list(dip);
837 
838 	DEVI(dip)->devi_major = DDI_MAJOR_T_NONE;
839 	DEVI(dip)->devi_binding_name = DEVI(dip)->devi_node_name;
840 	return (DDI_SUCCESS);
841 }
842 
843 /*
844  * Initialize a node: calls the parent nexus' bus_ctl ops to do the operation.
845  * Must hold parent and per-driver list while calling this function.
846  * A successful init_node() returns with an active ndi_hold_devi() hold on
847  * the parent.
848  */
849 static int
850 init_node(dev_info_t *dip)
851 {
852 	int error;
853 	dev_info_t *pdip = ddi_get_parent(dip);
854 	int (*f)(dev_info_t *, dev_info_t *, ddi_ctl_enum_t, void *, void *);
855 	char *path;
856 	major_t	major;
857 	ddi_devid_t devid = NULL;
858 
859 	ASSERT(i_ddi_node_state(dip) == DS_BOUND);
860 
861 	/* should be DS_READY except for pcmcia ... */
862 	ASSERT(i_ddi_node_state(pdip) >= DS_PROBED);
863 
864 	path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
865 	(void) ddi_pathname(dip, path);
866 	NDI_CONFIG_DEBUG((CE_CONT, "init_node: entry: path %s 0x%p\n",
867 	    path, (void *)dip));
868 
869 	/*
870 	 * The parent must have a bus_ctl operation.
871 	 */
872 	if ((DEVI(pdip)->devi_ops->devo_bus_ops == NULL) ||
873 	    (f = DEVI(pdip)->devi_ops->devo_bus_ops->bus_ctl) == NULL) {
874 		error = DDI_FAILURE;
875 		goto out;
876 	}
877 
878 	add_global_props(dip);
879 
880 	/*
881 	 * Invoke the parent's bus_ctl operation with the DDI_CTLOPS_INITCHILD
882 	 * command to transform the child to canonical form 1. If there
883 	 * is an error, ddi_remove_child should be called, to clean up.
884 	 */
885 	error = (*f)(pdip, pdip, DDI_CTLOPS_INITCHILD, dip, NULL);
886 	if (error != DDI_SUCCESS) {
887 		NDI_CONFIG_DEBUG((CE_CONT, "init_node: %s 0x%p failed\n",
888 		    path, (void *)dip));
889 		remove_global_props(dip);
890 
891 		/*
892 		 * If a nexus INITCHILD implementation calls ddi_devid_regster()
893 		 * prior to setting devi_addr, the devid is not recorded in
894 		 * the devid cache (i.e. DEVI_CACHED_DEVID is not set).
895 		 * With mpxio, while the vhci client path may be missing
896 		 * from the cache, phci pathinfo paths may have already be
897 		 * added to the cache, against the client dip, by use of
898 		 * e_devid_cache_pathinfo().  Because of this, when INITCHILD
899 		 * of the client fails, we need to purge the client dip from
900 		 * the cache even if DEVI_CACHED_DEVID is not set - if only
901 		 * devi_devid_str is set.
902 		 */
903 		mutex_enter(&DEVI(dip)->devi_lock);
904 		if ((DEVI(dip)->devi_flags & DEVI_CACHED_DEVID) ||
905 		    DEVI(dip)->devi_devid_str) {
906 			DEVI(dip)->devi_flags &= ~DEVI_CACHED_DEVID;
907 			mutex_exit(&DEVI(dip)->devi_lock);
908 			ddi_devid_unregister(dip);
909 		} else
910 			mutex_exit(&DEVI(dip)->devi_lock);
911 
912 		/* in case nexus driver didn't clear this field */
913 		ddi_set_name_addr(dip, NULL);
914 		error = DDI_FAILURE;
915 		goto out;
916 	}
917 
918 	ndi_hold_devi(pdip);			/* initial hold of parent */
919 
920 	/* recompute path after initchild for @addr information */
921 	(void) ddi_pathname(dip, path);
922 
923 	/* Check for duplicate nodes */
924 	if (find_duplicate_child(pdip, dip) != NULL) {
925 		/*
926 		 * uninit_node() the duplicate - a successful uninit_node()
927 		 * will release inital hold of parent using ndi_rele_devi().
928 		 */
929 		if ((error = uninit_node(dip)) != DDI_SUCCESS) {
930 			ndi_rele_devi(pdip);	/* release initial hold */
931 			cmn_err(CE_WARN, "init_node: uninit of duplicate "
932 			    "node %s failed", path);
933 		}
934 		NDI_CONFIG_DEBUG((CE_CONT, "init_node: duplicate uninit "
935 		    "%s 0x%p%s\n", path, (void *)dip,
936 		    (error == DDI_SUCCESS) ? "" : " failed"));
937 		error = DDI_FAILURE;
938 		goto out;
939 	}
940 
941 	/*
942 	 * If a devid was registered for a DS_BOUND node then the devid_cache
943 	 * may not have captured the path. Detect this situation and ensure that
944 	 * the path enters the cache now that devi_addr is established.
945 	 */
946 	if (!(DEVI(dip)->devi_flags & DEVI_CACHED_DEVID) &&
947 	    (ddi_devid_get(dip, &devid) == DDI_SUCCESS)) {
948 		if (e_devid_cache_register(dip, devid) == DDI_SUCCESS) {
949 			mutex_enter(&DEVI(dip)->devi_lock);
950 			DEVI(dip)->devi_flags |= DEVI_CACHED_DEVID;
951 			mutex_exit(&DEVI(dip)->devi_lock);
952 		}
953 
954 		ddi_devid_free(devid);
955 	}
956 
957 	/*
958 	 * Check to see if we have a path-oriented driver alias that overrides
959 	 * the current driver binding. If so, we need to rebind. This check
960 	 * needs to be delayed until after a successful DDI_CTLOPS_INITCHILD,
961 	 * so the unit-address is established on the last component of the path.
962 	 *
963 	 * NOTE: Allowing a path-oriented alias to change the driver binding
964 	 * of a driver.conf node results in non-intuitive property behavior.
965 	 * We provide a tunable (driver_conf_allow_path_alias) to control
966 	 * this behavior. See uninit_node() for more details.
967 	 *
968 	 * NOTE: If you are adding a path-oriented alias for the boot device,
969 	 * and there is mismatch between OBP and the kernel in regard to
970 	 * generic name use, like "disk" .vs. "ssd", then you will need
971 	 * to add a path-oriented alias for both paths.
972 	 */
973 	major = ddi_name_to_major(path);
974 	if (driver_active(major) && (major != DEVI(dip)->devi_major) &&
975 	    (ndi_dev_is_persistent_node(dip) || driver_conf_allow_path_alias)) {
976 
977 		/* Mark node for rebind processing. */
978 		mutex_enter(&DEVI(dip)->devi_lock);
979 		DEVI(dip)->devi_flags |= DEVI_REBIND;
980 		mutex_exit(&DEVI(dip)->devi_lock);
981 
982 		/*
983 		 * Add an extra hold on the parent to prevent it from ever
984 		 * having a zero devi_ref during the child rebind process.
985 		 * This is necessary to ensure that the parent will never
986 		 * detach(9E) during the rebind.
987 		 */
988 		ndi_hold_devi(pdip);		/* extra hold of parent */
989 
990 		/*
991 		 * uninit_node() current binding - a successful uninit_node()
992 		 * will release extra hold of parent using ndi_rele_devi().
993 		 */
994 		if ((error = uninit_node(dip)) != DDI_SUCCESS) {
995 			ndi_rele_devi(pdip);	/* release extra hold */
996 			ndi_rele_devi(pdip);	/* release initial hold */
997 			cmn_err(CE_WARN, "init_node: uninit for rebind "
998 			    "of node %s failed", path);
999 			goto out;
1000 		}
1001 
1002 		/* Unbind: demote the node back to DS_LINKED.  */
1003 		if ((error = ndi_devi_unbind_driver(dip)) != DDI_SUCCESS) {
1004 			ndi_rele_devi(pdip);	/* release initial hold */
1005 			cmn_err(CE_WARN, "init_node: unbind for rebind "
1006 			    "of node %s failed", path);
1007 			goto out;
1008 		}
1009 
1010 		/* establish rebinding name */
1011 		if (DEVI(dip)->devi_rebinding_name == NULL)
1012 			DEVI(dip)->devi_rebinding_name =
1013 			    i_ddi_strdup(path, KM_SLEEP);
1014 
1015 		/*
1016 		 * Now that we are demoted and marked for rebind, repromote.
1017 		 * We need to do this in steps, instead of just calling
1018 		 * ddi_initchild, so that we can redo the merge operation
1019 		 * after we are rebound to the path-bound driver.
1020 		 *
1021 		 * Start by rebinding node to the path-bound driver.
1022 		 */
1023 		if ((error = ndi_devi_bind_driver(dip, 0)) != DDI_SUCCESS) {
1024 			ndi_rele_devi(pdip);	/* release initial hold */
1025 			cmn_err(CE_WARN, "init_node: rebind "
1026 			    "of node %s failed", path);
1027 			goto out;
1028 		}
1029 
1030 		/*
1031 		 * If the node is not a driver.conf node then merge
1032 		 * driver.conf properties from new path-bound driver.conf.
1033 		 */
1034 		if (ndi_dev_is_persistent_node(dip))
1035 			(void) i_ndi_make_spec_children(pdip, 0);
1036 
1037 		/*
1038 		 * Now that we have taken care of merge, repromote back
1039 		 * to DS_INITIALIZED.
1040 		 */
1041 		error = ddi_initchild(pdip, dip);
1042 		NDI_CONFIG_DEBUG((CE_CONT, "init_node: rebind "
1043 		    "%s 0x%p\n", path, (void *)dip));
1044 
1045 		/*
1046 		 * Release our initial hold. If ddi_initchild() was
1047 		 * successful then it will return with the active hold.
1048 		 */
1049 		ndi_rele_devi(pdip);
1050 		goto out;
1051 	}
1052 
1053 	/*
1054 	 * Apply multi-parent/deep-nexus optimization to the new node
1055 	 */
1056 	DEVI(dip)->devi_instance = e_ddi_assign_instance(dip);
1057 	ddi_optimize_dtree(dip);
1058 	error = DDI_SUCCESS;		/* return with active hold */
1059 
1060 out:	if (error != DDI_SUCCESS) {
1061 		/* On failure ensure that DEVI_REBIND is cleared */
1062 		mutex_enter(&DEVI(dip)->devi_lock);
1063 		DEVI(dip)->devi_flags &= ~DEVI_REBIND;
1064 		mutex_exit(&DEVI(dip)->devi_lock);
1065 	}
1066 	kmem_free(path, MAXPATHLEN);
1067 	return (error);
1068 }
1069 
1070 /*
1071  * Uninitialize node
1072  * The per-driver list must be held busy during the call.
1073  * A successful uninit_node() releases the init_node() hold on
1074  * the parent by calling ndi_rele_devi().
1075  */
1076 static int
1077 uninit_node(dev_info_t *dip)
1078 {
1079 	int node_state_entry;
1080 	dev_info_t *pdip;
1081 	struct dev_ops *ops;
1082 	int (*f)();
1083 	int error;
1084 	char *addr;
1085 
1086 	/*
1087 	 * Don't check for references here or else a ref-counted
1088 	 * dip cannot be downgraded by the framework.
1089 	 */
1090 	node_state_entry = i_ddi_node_state(dip);
1091 	ASSERT((node_state_entry == DS_BOUND) ||
1092 	    (node_state_entry == DS_INITIALIZED));
1093 	pdip = ddi_get_parent(dip);
1094 	ASSERT(pdip);
1095 
1096 	NDI_CONFIG_DEBUG((CE_CONT, "uninit_node: 0x%p(%s%d)\n",
1097 	    (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1098 
1099 	if (((ops = ddi_get_driver(pdip)) == NULL) ||
1100 	    (ops->devo_bus_ops == NULL) ||
1101 	    ((f = ops->devo_bus_ops->bus_ctl) == NULL)) {
1102 		return (DDI_FAILURE);
1103 	}
1104 
1105 	/*
1106 	 * save the @addr prior to DDI_CTLOPS_UNINITCHILD for use in
1107 	 * freeing the instance if it succeeds.
1108 	 */
1109 	if (node_state_entry == DS_INITIALIZED) {
1110 		addr = ddi_get_name_addr(dip);
1111 		if (addr)
1112 			addr = i_ddi_strdup(addr, KM_SLEEP);
1113 	} else {
1114 		addr = NULL;
1115 	}
1116 
1117 	error = (*f)(pdip, pdip, DDI_CTLOPS_UNINITCHILD, dip, (void *)NULL);
1118 	if (error == DDI_SUCCESS) {
1119 		/* ensure that devids are unregistered */
1120 		mutex_enter(&DEVI(dip)->devi_lock);
1121 		if ((DEVI(dip)->devi_flags & DEVI_CACHED_DEVID)) {
1122 			DEVI(dip)->devi_flags &= ~DEVI_CACHED_DEVID;
1123 			mutex_exit(&DEVI(dip)->devi_lock);
1124 			ddi_devid_unregister(dip);
1125 		} else
1126 			mutex_exit(&DEVI(dip)->devi_lock);
1127 
1128 		/* if uninitchild forgot to set devi_addr to NULL do it now */
1129 		ddi_set_name_addr(dip, NULL);
1130 
1131 		/*
1132 		 * Free instance number. This is a no-op if instance has
1133 		 * been kept by probe_node().  Avoid free when we are called
1134 		 * from init_node (DS_BOUND) because the instance has not yet
1135 		 * been assigned.
1136 		 */
1137 		if (node_state_entry == DS_INITIALIZED) {
1138 			e_ddi_free_instance(dip, addr);
1139 			DEVI(dip)->devi_instance = -1;
1140 		}
1141 
1142 		/* release the init_node hold */
1143 		ndi_rele_devi(pdip);
1144 
1145 		remove_global_props(dip);
1146 
1147 		/*
1148 		 * NOTE: The decision on whether to allow a path-oriented
1149 		 * rebind of a driver.conf enumerated node is made by
1150 		 * init_node() based on driver_conf_allow_path_alias. The
1151 		 * rebind code below prevents deletion of system properties
1152 		 * on driver.conf nodes.
1153 		 *
1154 		 * When driver_conf_allow_path_alias is set, property behavior
1155 		 * on rebound driver.conf file is non-intuitive. For a
1156 		 * driver.conf node, the unit-address properties come from
1157 		 * the driver.conf file as system properties. Removing system
1158 		 * properties from a driver.conf node makes the node
1159 		 * useless (we get node without unit-address properties) - so
1160 		 * we leave system properties in place. The result is a node
1161 		 * where system properties come from the node being rebound,
1162 		 * and global properties come from the driver.conf file
1163 		 * of the driver we are rebinding to.  If we could determine
1164 		 * that the path-oriented alias driver.conf file defined a
1165 		 * node at the same unit address, it would be best to use
1166 		 * that node and avoid the non-intuitive property behavior.
1167 		 * Unfortunately, the current "merge" code does not support
1168 		 * this, so we live with the non-intuitive property behavior.
1169 		 */
1170 		if (!((ndi_dev_is_persistent_node(dip) == 0) &&
1171 		    (DEVI(dip)->devi_flags & DEVI_REBIND)))
1172 			e_ddi_prop_remove_all(dip);
1173 	} else {
1174 		NDI_CONFIG_DEBUG((CE_CONT, "uninit_node failed: 0x%p(%s%d)\n",
1175 		    (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1176 	}
1177 
1178 	if (addr)
1179 		kmem_free(addr, strlen(addr) + 1);
1180 	return (error);
1181 }
1182 
1183 /*
1184  * Invoke driver's probe entry point to probe for existence of hardware.
1185  * Keep instance permanent for successful probe and leaf nodes.
1186  *
1187  * Per-driver list must be held busy while calling this function.
1188  */
1189 static int
1190 probe_node(dev_info_t *dip)
1191 {
1192 	int rv;
1193 
1194 	ASSERT(i_ddi_node_state(dip) == DS_INITIALIZED);
1195 
1196 	NDI_CONFIG_DEBUG((CE_CONT, "probe_node: 0x%p(%s%d)\n",
1197 	    (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1198 
1199 	/* temporarily hold the driver while we probe */
1200 	DEVI(dip)->devi_ops = ndi_hold_driver(dip);
1201 	if (DEVI(dip)->devi_ops == NULL) {
1202 		NDI_CONFIG_DEBUG((CE_CONT,
1203 		    "probe_node: 0x%p(%s%d) cannot load driver\n",
1204 		    (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1205 		return (DDI_FAILURE);
1206 	}
1207 
1208 	if (identify_9e != 0)
1209 		(void) devi_identify(dip);
1210 
1211 	rv = devi_probe(dip);
1212 
1213 	/* release the driver now that probe is complete */
1214 	ndi_rele_driver(dip);
1215 	DEVI(dip)->devi_ops = NULL;
1216 
1217 	switch (rv) {
1218 	case DDI_PROBE_SUCCESS:			/* found */
1219 	case DDI_PROBE_DONTCARE:		/* ddi_dev_is_sid */
1220 		e_ddi_keep_instance(dip);	/* persist instance */
1221 		rv = DDI_SUCCESS;
1222 		break;
1223 
1224 	case DDI_PROBE_PARTIAL:			/* maybe later */
1225 	case DDI_PROBE_FAILURE:			/* not found */
1226 		NDI_CONFIG_DEBUG((CE_CONT,
1227 		    "probe_node: 0x%p(%s%d) no hardware found%s\n",
1228 		    (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip),
1229 		    (rv == DDI_PROBE_PARTIAL) ? " yet" : ""));
1230 		rv = DDI_FAILURE;
1231 		break;
1232 
1233 	default:
1234 #ifdef	DEBUG
1235 		cmn_err(CE_WARN, "probe_node: %s%d: illegal probe(9E) value",
1236 		    ddi_driver_name(dip), ddi_get_instance(dip));
1237 #endif	/* DEBUG */
1238 		rv = DDI_FAILURE;
1239 		break;
1240 	}
1241 	return (rv);
1242 }
1243 
1244 /*
1245  * Unprobe a node. Simply reset the node state.
1246  * Per-driver list must be held busy while calling this function.
1247  */
1248 static int
1249 unprobe_node(dev_info_t *dip)
1250 {
1251 	ASSERT(i_ddi_node_state(dip) == DS_PROBED);
1252 
1253 	/*
1254 	 * Don't check for references here or else a ref-counted
1255 	 * dip cannot be downgraded by the framework.
1256 	 */
1257 
1258 	NDI_CONFIG_DEBUG((CE_CONT, "unprobe_node: 0x%p(name = %s)\n",
1259 	    (void *)dip, ddi_node_name(dip)));
1260 	return (DDI_SUCCESS);
1261 }
1262 
1263 /*
1264  * Attach devinfo node.
1265  * Per-driver list must be held busy.
1266  */
1267 static int
1268 attach_node(dev_info_t *dip)
1269 {
1270 	int rv;
1271 
1272 	ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(dip)));
1273 	ASSERT(i_ddi_node_state(dip) == DS_PROBED);
1274 
1275 	NDI_CONFIG_DEBUG((CE_CONT, "attach_node: 0x%p(%s%d)\n",
1276 	    (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1277 
1278 	/*
1279 	 * Tell mpxio framework that a node is about to online.
1280 	 */
1281 	if ((rv = mdi_devi_online(dip, 0)) != NDI_SUCCESS) {
1282 		return (DDI_FAILURE);
1283 	}
1284 
1285 	/* no recursive attachment */
1286 	ASSERT(DEVI(dip)->devi_ops == NULL);
1287 
1288 	/*
1289 	 * Hold driver the node is bound to.
1290 	 */
1291 	DEVI(dip)->devi_ops = ndi_hold_driver(dip);
1292 	if (DEVI(dip)->devi_ops == NULL) {
1293 		/*
1294 		 * We were able to load driver for probing, so we should
1295 		 * not get here unless something really bad happened.
1296 		 */
1297 		cmn_err(CE_WARN, "attach_node: no driver for major %d",
1298 		    DEVI(dip)->devi_major);
1299 		return (DDI_FAILURE);
1300 	}
1301 
1302 	if (NEXUS_DRV(DEVI(dip)->devi_ops))
1303 		DEVI(dip)->devi_taskq = ddi_taskq_create(dip,
1304 		    "nexus_enum_tq", 1,
1305 		    TASKQ_DEFAULTPRI, 0);
1306 
1307 	mutex_enter(&(DEVI(dip)->devi_lock));
1308 	DEVI_SET_ATTACHING(dip);
1309 	DEVI_SET_NEED_RESET(dip);
1310 	mutex_exit(&(DEVI(dip)->devi_lock));
1311 
1312 	rv = devi_attach(dip, DDI_ATTACH);
1313 
1314 	mutex_enter(&(DEVI(dip)->devi_lock));
1315 	DEVI_CLR_ATTACHING(dip);
1316 
1317 	if (rv != DDI_SUCCESS) {
1318 		DEVI_CLR_NEED_RESET(dip);
1319 		mutex_exit(&DEVI(dip)->devi_lock);
1320 
1321 		/*
1322 		 * Cleanup dacf reservations
1323 		 */
1324 		mutex_enter(&dacf_lock);
1325 		dacf_clr_rsrvs(dip, DACF_OPID_POSTATTACH);
1326 		dacf_clr_rsrvs(dip, DACF_OPID_PREDETACH);
1327 		mutex_exit(&dacf_lock);
1328 		if (DEVI(dip)->devi_taskq)
1329 			ddi_taskq_destroy(DEVI(dip)->devi_taskq);
1330 		ddi_remove_minor_node(dip, NULL);
1331 
1332 		/* release the driver if attach failed */
1333 		ndi_rele_driver(dip);
1334 		DEVI(dip)->devi_ops = NULL;
1335 		NDI_CONFIG_DEBUG((CE_CONT, "attach_node: 0x%p(%s%d) failed\n",
1336 		    (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1337 		return (DDI_FAILURE);
1338 	} else
1339 		mutex_exit(&DEVI(dip)->devi_lock);
1340 
1341 	/* successful attach, return with driver held */
1342 
1343 	return (DDI_SUCCESS);
1344 }
1345 
1346 /*
1347  * Detach devinfo node.
1348  * Per-driver list must be held busy.
1349  */
1350 static int
1351 detach_node(dev_info_t *dip, uint_t flag)
1352 {
1353 	struct devnames	*dnp;
1354 	int		rv;
1355 
1356 	ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(dip)));
1357 	ASSERT(i_ddi_node_state(dip) == DS_ATTACHED);
1358 
1359 	/* check references */
1360 	if (DEVI(dip)->devi_ref)
1361 		return (DDI_FAILURE);
1362 
1363 	NDI_CONFIG_DEBUG((CE_CONT, "detach_node: 0x%p(%s%d)\n",
1364 	    (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1365 
1366 	/*
1367 	 * NOTE: If we are processing a pHCI node then the calling code
1368 	 * must detect this and ndi_devi_enter() in (vHCI, parent(pHCI))
1369 	 * order unless pHCI and vHCI are siblings.  Code paths leading
1370 	 * here that must ensure this ordering include:
1371 	 * unconfig_immediate_children(), devi_unconfig_one(),
1372 	 * ndi_devi_unconfig_one(), ndi_devi_offline().
1373 	 */
1374 	ASSERT(!MDI_PHCI(dip) ||
1375 	    (ddi_get_parent(mdi_devi_get_vdip(dip)) == ddi_get_parent(dip)) ||
1376 	    DEVI_BUSY_OWNED(mdi_devi_get_vdip(dip)));
1377 
1378 	/* Offline the device node with the mpxio framework. */
1379 	if (mdi_devi_offline(dip, flag) != NDI_SUCCESS) {
1380 		return (DDI_FAILURE);
1381 	}
1382 
1383 	/* drain the taskq */
1384 	if (DEVI(dip)->devi_taskq)
1385 		ddi_taskq_wait(DEVI(dip)->devi_taskq);
1386 
1387 	rv = devi_detach(dip, DDI_DETACH);
1388 
1389 	if (rv != DDI_SUCCESS) {
1390 		NDI_CONFIG_DEBUG((CE_CONT,
1391 		    "detach_node: 0x%p(%s%d) failed\n",
1392 		    (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1393 		return (DDI_FAILURE);
1394 	}
1395 
1396 	mutex_enter(&(DEVI(dip)->devi_lock));
1397 	DEVI_CLR_NEED_RESET(dip);
1398 	mutex_exit(&(DEVI(dip)->devi_lock));
1399 
1400 #if defined(__amd64) && !defined(__xpv)
1401 	/*
1402 	 * Close any iommulib mediated linkage to an IOMMU
1403 	 */
1404 	if (IOMMU_USED(dip))
1405 		iommulib_nex_close(dip);
1406 #endif
1407 
1408 	/* destroy the taskq */
1409 	if (DEVI(dip)->devi_taskq) {
1410 		ddi_taskq_destroy(DEVI(dip)->devi_taskq);
1411 		DEVI(dip)->devi_taskq = NULL;
1412 	}
1413 
1414 	/* Cleanup dacf reservations */
1415 	mutex_enter(&dacf_lock);
1416 	dacf_clr_rsrvs(dip, DACF_OPID_POSTATTACH);
1417 	dacf_clr_rsrvs(dip, DACF_OPID_PREDETACH);
1418 	mutex_exit(&dacf_lock);
1419 
1420 	/* remove any additional flavors that were added */
1421 	if (DEVI(dip)->devi_flavorv_n > 1 && DEVI(dip)->devi_flavorv != NULL) {
1422 		kmem_free(DEVI(dip)->devi_flavorv,
1423 		    (DEVI(dip)->devi_flavorv_n - 1) * sizeof (void *));
1424 		DEVI(dip)->devi_flavorv = NULL;
1425 	}
1426 
1427 	/* Remove properties and minor nodes in case driver forgots */
1428 	ddi_remove_minor_node(dip, NULL);
1429 	ddi_prop_remove_all(dip);
1430 
1431 	/* a detached node can't have attached or .conf children */
1432 	mutex_enter(&DEVI(dip)->devi_lock);
1433 	DEVI(dip)->devi_flags &= ~(DEVI_MADE_CHILDREN|DEVI_ATTACHED_CHILDREN);
1434 	mutex_exit(&DEVI(dip)->devi_lock);
1435 
1436 	/*
1437 	 * If the instance has successfully detached in detach_driver() context,
1438 	 * clear DN_DRIVER_HELD for correct ddi_hold_installed_driver()
1439 	 * behavior. Consumers like qassociate() depend on this (via clnopen()).
1440 	 */
1441 	if (flag & NDI_DETACH_DRIVER) {
1442 		dnp = &(devnamesp[DEVI(dip)->devi_major]);
1443 		LOCK_DEV_OPS(&dnp->dn_lock);
1444 		dnp->dn_flags &= ~DN_DRIVER_HELD;
1445 		UNLOCK_DEV_OPS(&dnp->dn_lock);
1446 	}
1447 
1448 	/* successful detach, release the driver */
1449 	ndi_rele_driver(dip);
1450 	DEVI(dip)->devi_ops = NULL;
1451 	return (DDI_SUCCESS);
1452 }
1453 
1454 /*
1455  * Run dacf post_attach routines
1456  */
1457 static int
1458 postattach_node(dev_info_t *dip)
1459 {
1460 	int rval;
1461 
1462 	/*
1463 	 * For hotplug busses like USB, it's possible that devices
1464 	 * are removed but dip is still around. We don't want to
1465 	 * run dacf routines as part of detach failure recovery.
1466 	 *
1467 	 * Pretend success until we figure out how to prevent
1468 	 * access to such devinfo nodes.
1469 	 */
1470 	if (DEVI_IS_DEVICE_REMOVED(dip))
1471 		return (DDI_SUCCESS);
1472 
1473 	/*
1474 	 * if dacf_postattach failed, report it to the framework
1475 	 * so that it can be retried later at the open time.
1476 	 */
1477 	mutex_enter(&dacf_lock);
1478 	rval = dacfc_postattach(dip);
1479 	mutex_exit(&dacf_lock);
1480 
1481 	/*
1482 	 * Plumbing during postattach may fail because of the
1483 	 * underlying device is not ready. This will fail ndi_devi_config()
1484 	 * in dv_filldir() and a warning message is issued. The message
1485 	 * from here will explain what happened
1486 	 */
1487 	if (rval != DACF_SUCCESS) {
1488 		cmn_err(CE_WARN, "Postattach failed for %s%d\n",
1489 		    ddi_driver_name(dip), ddi_get_instance(dip));
1490 		return (DDI_FAILURE);
1491 	}
1492 
1493 	return (DDI_SUCCESS);
1494 }
1495 
1496 /*
1497  * Run dacf pre-detach routines
1498  */
1499 static int
1500 predetach_node(dev_info_t *dip, uint_t flag)
1501 {
1502 	int ret;
1503 
1504 	/*
1505 	 * Don't auto-detach if DDI_FORCEATTACH or DDI_NO_AUTODETACH
1506 	 * properties are set.
1507 	 */
1508 	if (flag & NDI_AUTODETACH) {
1509 		struct devnames *dnp;
1510 		int pflag = DDI_PROP_NOTPROM | DDI_PROP_DONTPASS;
1511 
1512 		if ((ddi_prop_get_int(DDI_DEV_T_ANY, dip,
1513 		    pflag, DDI_FORCEATTACH, 0) == 1) ||
1514 		    (ddi_prop_get_int(DDI_DEV_T_ANY, dip,
1515 		    pflag, DDI_NO_AUTODETACH, 0) == 1))
1516 			return (DDI_FAILURE);
1517 
1518 		/* check for driver global version of DDI_NO_AUTODETACH */
1519 		dnp = &devnamesp[DEVI(dip)->devi_major];
1520 		LOCK_DEV_OPS(&dnp->dn_lock);
1521 		if (dnp->dn_flags & DN_NO_AUTODETACH) {
1522 			UNLOCK_DEV_OPS(&dnp->dn_lock);
1523 			return (DDI_FAILURE);
1524 		}
1525 		UNLOCK_DEV_OPS(&dnp->dn_lock);
1526 	}
1527 
1528 	mutex_enter(&dacf_lock);
1529 	ret = dacfc_predetach(dip);
1530 	mutex_exit(&dacf_lock);
1531 
1532 	return (ret);
1533 }
1534 
1535 /*
1536  * Wrapper for making multiple state transitions
1537  */
1538 
1539 /*
1540  * i_ndi_config_node: upgrade dev_info node into a specified state.
1541  * It is a bit tricky because the locking protocol changes before and
1542  * after a node is bound to a driver. All locks are held external to
1543  * this function.
1544  */
1545 int
1546 i_ndi_config_node(dev_info_t *dip, ddi_node_state_t state, uint_t flag)
1547 {
1548 	_NOTE(ARGUNUSED(flag))
1549 	int rv = DDI_SUCCESS;
1550 
1551 	ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(dip)));
1552 
1553 	while ((i_ddi_node_state(dip) < state) && (rv == DDI_SUCCESS)) {
1554 
1555 		/* don't allow any more changes to the device tree */
1556 		if (devinfo_freeze) {
1557 			rv = DDI_FAILURE;
1558 			break;
1559 		}
1560 
1561 		switch (i_ddi_node_state(dip)) {
1562 		case DS_PROTO:
1563 			/*
1564 			 * only caller can reference this node, no external
1565 			 * locking needed.
1566 			 */
1567 			link_node(dip);
1568 			translate_devid((dev_info_t *)dip);
1569 			i_ddi_set_node_state(dip, DS_LINKED);
1570 			break;
1571 		case DS_LINKED:
1572 			/*
1573 			 * Three code path may attempt to bind a node:
1574 			 * - boot code
1575 			 * - add_drv
1576 			 * - hotplug thread
1577 			 * Boot code is single threaded, add_drv synchronize
1578 			 * on a userland lock, and hotplug synchronize on
1579 			 * hotplug_lk. There could be a race between add_drv
1580 			 * and hotplug thread. We'll live with this until the
1581 			 * conversion to top-down loading.
1582 			 */
1583 			if ((rv = bind_node(dip)) == DDI_SUCCESS)
1584 				i_ddi_set_node_state(dip, DS_BOUND);
1585 
1586 			break;
1587 		case DS_BOUND:
1588 			/*
1589 			 * The following transitions synchronizes on the
1590 			 * per-driver busy changing flag, since we already
1591 			 * have a driver.
1592 			 */
1593 			if ((rv = init_node(dip)) == DDI_SUCCESS)
1594 				i_ddi_set_node_state(dip, DS_INITIALIZED);
1595 			break;
1596 		case DS_INITIALIZED:
1597 			if ((rv = probe_node(dip)) == DDI_SUCCESS)
1598 				i_ddi_set_node_state(dip, DS_PROBED);
1599 			break;
1600 		case DS_PROBED:
1601 			/*
1602 			 * If node is retired and persistent, then prevent
1603 			 * attach. We can't do this for non-persistent nodes
1604 			 * as we would lose evidence that the node existed.
1605 			 */
1606 			if (i_ddi_check_retire(dip) == 1 &&
1607 			    ndi_dev_is_persistent_node(dip) &&
1608 			    retire_prevents_attach == 1) {
1609 				rv = DDI_FAILURE;
1610 				break;
1611 			}
1612 			atomic_add_long(&devinfo_attach_detach, 1);
1613 			if ((rv = attach_node(dip)) == DDI_SUCCESS)
1614 				i_ddi_set_node_state(dip, DS_ATTACHED);
1615 			atomic_add_long(&devinfo_attach_detach, -1);
1616 			break;
1617 		case DS_ATTACHED:
1618 			if ((rv = postattach_node(dip)) == DDI_SUCCESS)
1619 				i_ddi_set_node_state(dip, DS_READY);
1620 			break;
1621 		case DS_READY:
1622 			break;
1623 		default:
1624 			/* should never reach here */
1625 			ASSERT("unknown devinfo state");
1626 		}
1627 	}
1628 
1629 	if (ddidebug & DDI_AUDIT)
1630 		da_log_enter(dip);
1631 	return (rv);
1632 }
1633 
1634 /*
1635  * i_ndi_unconfig_node: downgrade dev_info node into a specified state.
1636  */
1637 int
1638 i_ndi_unconfig_node(dev_info_t *dip, ddi_node_state_t state, uint_t flag)
1639 {
1640 	int	rv = DDI_SUCCESS;
1641 
1642 	ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(dip)));
1643 
1644 	while ((i_ddi_node_state(dip) > state) && (rv == DDI_SUCCESS)) {
1645 
1646 		/* don't allow any more changes to the device tree */
1647 		if (devinfo_freeze) {
1648 			rv = DDI_FAILURE;
1649 			break;
1650 		}
1651 
1652 		switch (i_ddi_node_state(dip)) {
1653 		case DS_PROTO:
1654 			break;
1655 		case DS_LINKED:
1656 			/*
1657 			 * Persistent nodes are only removed by hotplug code
1658 			 * .conf nodes synchronizes on per-driver list.
1659 			 */
1660 			if ((rv = unlink_node(dip)) == DDI_SUCCESS)
1661 				i_ddi_set_node_state(dip, DS_PROTO);
1662 			break;
1663 		case DS_BOUND:
1664 			/*
1665 			 * The following transitions synchronizes on the
1666 			 * per-driver busy changing flag, since we already
1667 			 * have a driver.
1668 			 */
1669 			if ((rv = unbind_node(dip)) == DDI_SUCCESS)
1670 				i_ddi_set_node_state(dip, DS_LINKED);
1671 			break;
1672 		case DS_INITIALIZED:
1673 			if ((rv = uninit_node(dip)) == DDI_SUCCESS)
1674 				i_ddi_set_node_state(dip, DS_BOUND);
1675 			break;
1676 		case DS_PROBED:
1677 			if ((rv = unprobe_node(dip)) == DDI_SUCCESS)
1678 				i_ddi_set_node_state(dip, DS_INITIALIZED);
1679 			break;
1680 		case DS_ATTACHED:
1681 			atomic_add_long(&devinfo_attach_detach, 1);
1682 
1683 			mutex_enter(&(DEVI(dip)->devi_lock));
1684 			DEVI_SET_DETACHING(dip);
1685 			mutex_exit(&(DEVI(dip)->devi_lock));
1686 
1687 			membar_enter();	/* ensure visibility for hold_devi */
1688 
1689 			if ((rv = detach_node(dip, flag)) == DDI_SUCCESS)
1690 				i_ddi_set_node_state(dip, DS_PROBED);
1691 
1692 			mutex_enter(&(DEVI(dip)->devi_lock));
1693 			DEVI_CLR_DETACHING(dip);
1694 			mutex_exit(&(DEVI(dip)->devi_lock));
1695 
1696 			atomic_add_long(&devinfo_attach_detach, -1);
1697 			break;
1698 		case DS_READY:
1699 			if ((rv = predetach_node(dip, flag)) == DDI_SUCCESS)
1700 				i_ddi_set_node_state(dip, DS_ATTACHED);
1701 			break;
1702 		default:
1703 			ASSERT("unknown devinfo state");
1704 		}
1705 	}
1706 	da_log_enter(dip);
1707 	return (rv);
1708 }
1709 
1710 /*
1711  * ddi_initchild: transform node to DS_INITIALIZED state
1712  */
1713 int
1714 ddi_initchild(dev_info_t *parent, dev_info_t *proto)
1715 {
1716 	int ret, circ;
1717 
1718 	ndi_devi_enter(parent, &circ);
1719 	ret = i_ndi_config_node(proto, DS_INITIALIZED, 0);
1720 	ndi_devi_exit(parent, circ);
1721 
1722 	return (ret);
1723 }
1724 
1725 /*
1726  * ddi_uninitchild: transform node down to DS_BOUND state
1727  */
1728 int
1729 ddi_uninitchild(dev_info_t *dip)
1730 {
1731 	int ret, circ;
1732 	dev_info_t *parent = ddi_get_parent(dip);
1733 	ASSERT(parent);
1734 
1735 	ndi_devi_enter(parent, &circ);
1736 	ret = i_ndi_unconfig_node(dip, DS_BOUND, 0);
1737 	ndi_devi_exit(parent, circ);
1738 
1739 	return (ret);
1740 }
1741 
1742 /*
1743  * i_ddi_attachchild: transform node to DS_READY/i_ddi_devi_attached() state
1744  */
1745 static int
1746 i_ddi_attachchild(dev_info_t *dip)
1747 {
1748 	dev_info_t	*parent = ddi_get_parent(dip);
1749 	int		ret;
1750 
1751 	ASSERT(parent && DEVI_BUSY_OWNED(parent));
1752 
1753 	if ((i_ddi_node_state(dip) < DS_BOUND) || DEVI_IS_DEVICE_OFFLINE(dip))
1754 		return (DDI_FAILURE);
1755 
1756 	ret = i_ndi_config_node(dip, DS_READY, 0);
1757 	if (ret == NDI_SUCCESS) {
1758 		ret = DDI_SUCCESS;
1759 	} else {
1760 		/*
1761 		 * Take it down to DS_INITIALIZED so pm_pre_probe is run
1762 		 * on the next attach
1763 		 */
1764 		(void) i_ndi_unconfig_node(dip, DS_INITIALIZED, 0);
1765 		ret = DDI_FAILURE;
1766 	}
1767 
1768 	return (ret);
1769 }
1770 
1771 /*
1772  * i_ddi_detachchild: transform node down to DS_PROBED state
1773  *	If it fails, put it back to DS_READY state.
1774  * NOTE: A node that fails detach may be at DS_ATTACHED instead
1775  * of DS_READY for a small amount of time - this is the source of
1776  * transient DS_READY->DS_ATTACHED->DS_READY state changes.
1777  */
1778 static int
1779 i_ddi_detachchild(dev_info_t *dip, uint_t flags)
1780 {
1781 	dev_info_t	*parent = ddi_get_parent(dip);
1782 	int		ret;
1783 
1784 	ASSERT(parent && DEVI_BUSY_OWNED(parent));
1785 
1786 	ret = i_ndi_unconfig_node(dip, DS_PROBED, flags);
1787 	if (ret != DDI_SUCCESS)
1788 		(void) i_ndi_config_node(dip, DS_READY, 0);
1789 	else
1790 		/* allow pm_pre_probe to reestablish pm state */
1791 		(void) i_ndi_unconfig_node(dip, DS_INITIALIZED, 0);
1792 	return (ret);
1793 }
1794 
1795 /*
1796  * Add a child and bind to driver
1797  */
1798 dev_info_t *
1799 ddi_add_child(dev_info_t *pdip, char *name, uint_t nodeid, uint_t unit)
1800 {
1801 	int circ;
1802 	dev_info_t *dip;
1803 
1804 	/* allocate a new node */
1805 	dip = i_ddi_alloc_node(pdip, name, nodeid, (int)unit, NULL, KM_SLEEP);
1806 
1807 	ndi_devi_enter(pdip, &circ);
1808 	(void) i_ndi_config_node(dip, DS_BOUND, 0);
1809 	ndi_devi_exit(pdip, circ);
1810 	return (dip);
1811 }
1812 
1813 /*
1814  * ddi_remove_child: remove the dip. The parent must be attached and held
1815  */
1816 int
1817 ddi_remove_child(dev_info_t *dip, int dummy)
1818 {
1819 	_NOTE(ARGUNUSED(dummy))
1820 	int circ, ret;
1821 	dev_info_t *parent = ddi_get_parent(dip);
1822 	ASSERT(parent);
1823 
1824 	ndi_devi_enter(parent, &circ);
1825 
1826 	/*
1827 	 * If we still have children, for example SID nodes marked
1828 	 * as persistent but not attached, attempt to remove them.
1829 	 */
1830 	if (DEVI(dip)->devi_child) {
1831 		ret = ndi_devi_unconfig(dip, NDI_DEVI_REMOVE);
1832 		if (ret != NDI_SUCCESS) {
1833 			ndi_devi_exit(parent, circ);
1834 			return (DDI_FAILURE);
1835 		}
1836 		ASSERT(DEVI(dip)->devi_child == NULL);
1837 	}
1838 
1839 	ret = i_ndi_unconfig_node(dip, DS_PROTO, 0);
1840 	ndi_devi_exit(parent, circ);
1841 
1842 	if (ret != DDI_SUCCESS)
1843 		return (ret);
1844 
1845 	ASSERT(i_ddi_node_state(dip) == DS_PROTO);
1846 	i_ddi_free_node(dip);
1847 	return (DDI_SUCCESS);
1848 }
1849 
1850 /*
1851  * NDI wrappers for ref counting, node allocation, and transitions
1852  */
1853 
1854 /*
1855  * Hold/release the devinfo node itself.
1856  * Caller is assumed to prevent the devi from detaching during this call
1857  */
1858 void
1859 ndi_hold_devi(dev_info_t *dip)
1860 {
1861 	mutex_enter(&DEVI(dip)->devi_lock);
1862 	ASSERT(DEVI(dip)->devi_ref >= 0);
1863 	DEVI(dip)->devi_ref++;
1864 	membar_enter();			/* make sure stores are flushed */
1865 	mutex_exit(&DEVI(dip)->devi_lock);
1866 }
1867 
1868 void
1869 ndi_rele_devi(dev_info_t *dip)
1870 {
1871 	ASSERT(DEVI(dip)->devi_ref > 0);
1872 
1873 	mutex_enter(&DEVI(dip)->devi_lock);
1874 	DEVI(dip)->devi_ref--;
1875 	membar_enter();			/* make sure stores are flushed */
1876 	mutex_exit(&DEVI(dip)->devi_lock);
1877 }
1878 
1879 int
1880 e_ddi_devi_holdcnt(dev_info_t *dip)
1881 {
1882 	return (DEVI(dip)->devi_ref);
1883 }
1884 
1885 /*
1886  * Hold/release the driver the devinfo node is bound to.
1887  */
1888 struct dev_ops *
1889 ndi_hold_driver(dev_info_t *dip)
1890 {
1891 	if (i_ddi_node_state(dip) < DS_BOUND)
1892 		return (NULL);
1893 
1894 	ASSERT(DEVI(dip)->devi_major != -1);
1895 	return (mod_hold_dev_by_major(DEVI(dip)->devi_major));
1896 }
1897 
1898 void
1899 ndi_rele_driver(dev_info_t *dip)
1900 {
1901 	ASSERT(i_ddi_node_state(dip) >= DS_BOUND);
1902 	mod_rele_dev_by_major(DEVI(dip)->devi_major);
1903 }
1904 
1905 /*
1906  * Single thread entry into devinfo node for modifying its children (devinfo,
1907  * pathinfo, and minor). To verify in ASSERTS use DEVI_BUSY_OWNED macro.
1908  */
1909 void
1910 ndi_devi_enter(dev_info_t *dip, int *circular)
1911 {
1912 	struct dev_info *devi = DEVI(dip);
1913 	ASSERT(dip != NULL);
1914 
1915 	/* for vHCI, enforce (vHCI, pHCI) ndi_deve_enter() order */
1916 	ASSERT(!MDI_VHCI(dip) || (mdi_devi_pdip_entered(dip) == 0) ||
1917 	    DEVI_BUSY_OWNED(dip));
1918 
1919 	mutex_enter(&devi->devi_lock);
1920 	if (devi->devi_busy_thread == curthread) {
1921 		devi->devi_circular++;
1922 	} else {
1923 		while (DEVI_BUSY_CHANGING(devi) && !panicstr)
1924 			cv_wait(&(devi->devi_cv), &(devi->devi_lock));
1925 		if (panicstr) {
1926 			mutex_exit(&devi->devi_lock);
1927 			return;
1928 		}
1929 		devi->devi_flags |= DEVI_BUSY;
1930 		devi->devi_busy_thread = curthread;
1931 	}
1932 	*circular = devi->devi_circular;
1933 	mutex_exit(&devi->devi_lock);
1934 }
1935 
1936 /*
1937  * Release ndi_devi_enter or successful ndi_devi_tryenter.
1938  */
1939 void
1940 ndi_devi_exit(dev_info_t *dip, int circular)
1941 {
1942 	struct dev_info	*devi = DEVI(dip);
1943 	struct dev_info	*vdevi;
1944 	ASSERT(dip != NULL);
1945 
1946 	if (panicstr)
1947 		return;
1948 
1949 	mutex_enter(&(devi->devi_lock));
1950 	if (circular != 0) {
1951 		devi->devi_circular--;
1952 	} else {
1953 		devi->devi_flags &= ~DEVI_BUSY;
1954 		ASSERT(devi->devi_busy_thread == curthread);
1955 		devi->devi_busy_thread = NULL;
1956 		cv_broadcast(&(devi->devi_cv));
1957 	}
1958 	mutex_exit(&(devi->devi_lock));
1959 
1960 	/*
1961 	 * For pHCI exit we issue a broadcast to vHCI for ndi_devi_config_one()
1962 	 * doing cv_wait on vHCI.
1963 	 */
1964 	if (MDI_PHCI(dip)) {
1965 		vdevi = DEVI(mdi_devi_get_vdip(dip));
1966 		if (vdevi) {
1967 			mutex_enter(&(vdevi->devi_lock));
1968 			if (vdevi->devi_flags & DEVI_PHCI_SIGNALS_VHCI) {
1969 				vdevi->devi_flags &= ~DEVI_PHCI_SIGNALS_VHCI;
1970 				cv_broadcast(&(vdevi->devi_cv));
1971 			}
1972 			mutex_exit(&(vdevi->devi_lock));
1973 		}
1974 	}
1975 }
1976 
1977 /*
1978  * Release ndi_devi_enter and wait for possibility of new children, avoiding
1979  * possibility of missing broadcast before getting to cv_timedwait().
1980  */
1981 static void
1982 ndi_devi_exit_and_wait(dev_info_t *dip, int circular, clock_t end_time)
1983 {
1984 	struct dev_info	*devi = DEVI(dip);
1985 	ASSERT(dip != NULL);
1986 
1987 	if (panicstr)
1988 		return;
1989 
1990 	/*
1991 	 * We are called to wait for of a new child, and new child can
1992 	 * only be added if circular is zero.
1993 	 */
1994 	ASSERT(circular == 0);
1995 
1996 	/* like ndi_devi_exit with circular of zero */
1997 	mutex_enter(&(devi->devi_lock));
1998 	devi->devi_flags &= ~DEVI_BUSY;
1999 	ASSERT(devi->devi_busy_thread == curthread);
2000 	devi->devi_busy_thread = NULL;
2001 	cv_broadcast(&(devi->devi_cv));
2002 
2003 	/* now wait for new children while still holding devi_lock */
2004 	(void) cv_timedwait(&devi->devi_cv, &(devi->devi_lock), end_time);
2005 	mutex_exit(&(devi->devi_lock));
2006 }
2007 
2008 /*
2009  * Attempt to single thread entry into devinfo node for modifying its children.
2010  */
2011 int
2012 ndi_devi_tryenter(dev_info_t *dip, int *circular)
2013 {
2014 	int rval = 1;		   /* assume we enter */
2015 	struct dev_info *devi = DEVI(dip);
2016 	ASSERT(dip != NULL);
2017 
2018 	mutex_enter(&devi->devi_lock);
2019 	if (devi->devi_busy_thread == (void *)curthread) {
2020 		devi->devi_circular++;
2021 	} else {
2022 		if (!DEVI_BUSY_CHANGING(devi)) {
2023 			devi->devi_flags |= DEVI_BUSY;
2024 			devi->devi_busy_thread = (void *)curthread;
2025 		} else {
2026 			rval = 0;	/* devi is busy */
2027 		}
2028 	}
2029 	*circular = devi->devi_circular;
2030 	mutex_exit(&devi->devi_lock);
2031 	return (rval);
2032 }
2033 
2034 /*
2035  * Allocate and initialize a new dev_info structure.
2036  *
2037  * This routine may be called at interrupt time by a nexus in
2038  * response to a hotplug event, therefore memory allocations are
2039  * not allowed to sleep.
2040  */
2041 int
2042 ndi_devi_alloc(dev_info_t *parent, char *node_name, pnode_t nodeid,
2043     dev_info_t **ret_dip)
2044 {
2045 	ASSERT(node_name != NULL);
2046 	ASSERT(ret_dip != NULL);
2047 
2048 	*ret_dip = i_ddi_alloc_node(parent, node_name, nodeid, -1, NULL,
2049 	    KM_NOSLEEP);
2050 	if (*ret_dip == NULL) {
2051 		return (NDI_NOMEM);
2052 	}
2053 
2054 	return (NDI_SUCCESS);
2055 }
2056 
2057 /*
2058  * Allocate and initialize a new dev_info structure
2059  * This routine may sleep and should not be called at interrupt time
2060  */
2061 void
2062 ndi_devi_alloc_sleep(dev_info_t *parent, char *node_name, pnode_t nodeid,
2063     dev_info_t **ret_dip)
2064 {
2065 	ASSERT(node_name != NULL);
2066 	ASSERT(ret_dip != NULL);
2067 
2068 	*ret_dip = i_ddi_alloc_node(parent, node_name, nodeid, -1, NULL,
2069 	    KM_SLEEP);
2070 	ASSERT(*ret_dip);
2071 }
2072 
2073 /*
2074  * Remove an initialized (but not yet attached) dev_info
2075  * node from it's parent.
2076  */
2077 int
2078 ndi_devi_free(dev_info_t *dip)
2079 {
2080 	ASSERT(dip != NULL);
2081 
2082 	if (i_ddi_node_state(dip) >= DS_INITIALIZED)
2083 		return (DDI_FAILURE);
2084 
2085 	NDI_CONFIG_DEBUG((CE_CONT, "ndi_devi_free: %s%d (%p)\n",
2086 	    ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip));
2087 
2088 	(void) ddi_remove_child(dip, 0);
2089 
2090 	return (NDI_SUCCESS);
2091 }
2092 
2093 /*
2094  * ndi_devi_bind_driver() binds a driver to a given device. If it fails
2095  * to bind the driver, it returns an appropriate error back. Some drivers
2096  * may want to know if the actually failed to bind.
2097  */
2098 int
2099 ndi_devi_bind_driver(dev_info_t *dip, uint_t flags)
2100 {
2101 	int ret = NDI_FAILURE;
2102 	int circ;
2103 	dev_info_t *pdip = ddi_get_parent(dip);
2104 	ASSERT(pdip);
2105 
2106 	NDI_CONFIG_DEBUG((CE_CONT,
2107 	    "ndi_devi_bind_driver: %s%d (%p) flags: %x\n",
2108 	    ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip, flags));
2109 
2110 	ndi_devi_enter(pdip, &circ);
2111 	if (i_ndi_config_node(dip, DS_BOUND, flags) == DDI_SUCCESS)
2112 		ret = NDI_SUCCESS;
2113 	ndi_devi_exit(pdip, circ);
2114 
2115 	return (ret);
2116 }
2117 
2118 /*
2119  * ndi_devi_unbind_driver: unbind the dip
2120  */
2121 static int
2122 ndi_devi_unbind_driver(dev_info_t *dip)
2123 {
2124 	ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(dip)));
2125 
2126 	return (i_ndi_unconfig_node(dip, DS_LINKED, 0));
2127 }
2128 
2129 /*
2130  * Misc. help routines called by framework only
2131  */
2132 
2133 /*
2134  * Get the state of node
2135  */
2136 ddi_node_state_t
2137 i_ddi_node_state(dev_info_t *dip)
2138 {
2139 	return (DEVI(dip)->devi_node_state);
2140 }
2141 
2142 /*
2143  * Set the state of node
2144  */
2145 void
2146 i_ddi_set_node_state(dev_info_t *dip, ddi_node_state_t state)
2147 {
2148 	DEVI(dip)->devi_node_state = state;
2149 	membar_enter();			/* make sure stores are flushed */
2150 }
2151 
2152 /*
2153  * Determine if node is attached. The implementation accommodates transient
2154  * DS_READY->DS_ATTACHED->DS_READY state changes.  Outside this file, this
2155  * function should be instead of i_ddi_node_state() DS_ATTACHED/DS_READY
2156  * state checks.
2157  */
2158 int
2159 i_ddi_devi_attached(dev_info_t *dip)
2160 {
2161 	return (DEVI(dip)->devi_node_state >= DS_ATTACHED);
2162 }
2163 
2164 /*
2165  * Common function for finding a node in a sibling list given name and addr.
2166  *
2167  * By default, name is matched with devi_node_name. The following
2168  * alternative match strategies are supported:
2169  *
2170  *	FIND_NODE_BY_NODENAME: Match on node name - typical use.
2171  *
2172  *	FIND_NODE_BY_DRIVER: A match on driver name bound to node is conducted.
2173  *		This support is used for support of OBP generic names and
2174  *		for the conversion from driver names to generic names. When
2175  *		more consistency in the generic name environment is achieved
2176  *		(and not needed for upgrade) this support can be removed.
2177  *
2178  *	FIND_NODE_BY_ADDR: Match on just the addr.
2179  *		This support is only used/needed during boot to match
2180  *		a node bound via a path-based driver alias.
2181  *
2182  * If a child is not named (dev_addr == NULL), there are three
2183  * possible actions:
2184  *
2185  *	(1) skip it
2186  *	(2) FIND_ADDR_BY_INIT: bring child to DS_INITIALIZED state
2187  *	(3) FIND_ADDR_BY_CALLBACK: use a caller-supplied callback function
2188  */
2189 #define	FIND_NODE_BY_NODENAME	0x01
2190 #define	FIND_NODE_BY_DRIVER	0x02
2191 #define	FIND_NODE_BY_ADDR	0x04
2192 #define	FIND_ADDR_BY_INIT	0x10
2193 #define	FIND_ADDR_BY_CALLBACK	0x20
2194 
2195 static dev_info_t *
2196 find_sibling(dev_info_t *head, char *cname, char *caddr, uint_t flag,
2197     int (*callback)(dev_info_t *, char *, int))
2198 {
2199 	dev_info_t	*dip;
2200 	char		*addr, *buf;
2201 	major_t		major;
2202 	uint_t		by;
2203 
2204 	/* only one way to find a node */
2205 	by = flag &
2206 	    (FIND_NODE_BY_DRIVER | FIND_NODE_BY_NODENAME | FIND_NODE_BY_ADDR);
2207 	ASSERT(by && BIT_ONLYONESET(by));
2208 
2209 	/* only one way to name a node */
2210 	ASSERT(((flag & FIND_ADDR_BY_INIT) == 0) ||
2211 	    ((flag & FIND_ADDR_BY_CALLBACK) == 0));
2212 
2213 	if (by == FIND_NODE_BY_DRIVER) {
2214 		major = ddi_name_to_major(cname);
2215 		if (major == DDI_MAJOR_T_NONE)
2216 			return (NULL);
2217 	}
2218 
2219 	/* preallocate buffer of naming node by callback */
2220 	if (flag & FIND_ADDR_BY_CALLBACK)
2221 		buf = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2222 
2223 	/*
2224 	 * Walk the child list to find a match
2225 	 */
2226 	if (head == NULL)
2227 		return (NULL);
2228 	ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(head)));
2229 	for (dip = head; dip; dip = ddi_get_next_sibling(dip)) {
2230 		if (by == FIND_NODE_BY_NODENAME) {
2231 			/* match node name */
2232 			if (strcmp(cname, DEVI(dip)->devi_node_name) != 0)
2233 				continue;
2234 		} else if (by == FIND_NODE_BY_DRIVER) {
2235 			/* match driver major */
2236 			if (DEVI(dip)->devi_major != major)
2237 				continue;
2238 		}
2239 
2240 		if ((addr = DEVI(dip)->devi_addr) == NULL) {
2241 			/* name the child based on the flag */
2242 			if (flag & FIND_ADDR_BY_INIT) {
2243 				if (ddi_initchild(ddi_get_parent(dip), dip)
2244 				    != DDI_SUCCESS)
2245 					continue;
2246 				addr = DEVI(dip)->devi_addr;
2247 			} else if (flag & FIND_ADDR_BY_CALLBACK) {
2248 				if ((callback == NULL) || (callback(
2249 				    dip, buf, MAXNAMELEN) != DDI_SUCCESS))
2250 					continue;
2251 				addr = buf;
2252 			} else {
2253 				continue;	/* skip */
2254 			}
2255 		}
2256 
2257 		/* match addr */
2258 		ASSERT(addr != NULL);
2259 		if (strcmp(caddr, addr) == 0)
2260 			break;	/* node found */
2261 
2262 	}
2263 	if (flag & FIND_ADDR_BY_CALLBACK)
2264 		kmem_free(buf, MAXNAMELEN);
2265 	return (dip);
2266 }
2267 
2268 /*
2269  * Find child of pdip with name: cname@caddr
2270  * Called by init_node() to look for duplicate nodes
2271  */
2272 static dev_info_t *
2273 find_duplicate_child(dev_info_t *pdip, dev_info_t *dip)
2274 {
2275 	dev_info_t *dup;
2276 	char *cname = DEVI(dip)->devi_node_name;
2277 	char *caddr = DEVI(dip)->devi_addr;
2278 
2279 	/* search nodes before dip */
2280 	dup = find_sibling(ddi_get_child(pdip), cname, caddr,
2281 	    FIND_NODE_BY_NODENAME, NULL);
2282 	if (dup != dip)
2283 		return (dup);
2284 
2285 	/*
2286 	 * search nodes after dip; normally this is not needed,
2287 	 */
2288 	return (find_sibling(ddi_get_next_sibling(dip), cname, caddr,
2289 	    FIND_NODE_BY_NODENAME, NULL));
2290 }
2291 
2292 /*
2293  * Find a child of a given name and address, using a callback to name
2294  * unnamed children. cname is the binding name.
2295  */
2296 dev_info_t *
2297 ndi_devi_findchild_by_callback(dev_info_t *pdip, char *dname, char *ua,
2298     int (*make_ua)(dev_info_t *, char *, int))
2299 {
2300 	int	by = FIND_ADDR_BY_CALLBACK;
2301 
2302 	ASSERT(DEVI_BUSY_OWNED(pdip));
2303 	by |= dname ? FIND_NODE_BY_DRIVER : FIND_NODE_BY_ADDR;
2304 	return (find_sibling(ddi_get_child(pdip), dname, ua, by, make_ua));
2305 }
2306 
2307 /*
2308  * Find a child of a given name and address, invoking initchild to name
2309  * unnamed children. cname is the node name.
2310  */
2311 static dev_info_t *
2312 find_child_by_name(dev_info_t *pdip, char *cname, char *caddr)
2313 {
2314 	dev_info_t	*dip;
2315 
2316 	/* attempt search without changing state of preceding siblings */
2317 	dip = find_sibling(ddi_get_child(pdip), cname, caddr,
2318 	    FIND_NODE_BY_NODENAME, NULL);
2319 	if (dip)
2320 		return (dip);
2321 
2322 	return (find_sibling(ddi_get_child(pdip), cname, caddr,
2323 	    FIND_NODE_BY_NODENAME|FIND_ADDR_BY_INIT, NULL));
2324 }
2325 
2326 /*
2327  * Find a child of a given name and address, invoking initchild to name
2328  * unnamed children. cname is the node name.
2329  */
2330 static dev_info_t *
2331 find_child_by_driver(dev_info_t *pdip, char *cname, char *caddr)
2332 {
2333 	dev_info_t	*dip;
2334 
2335 	/* attempt search without changing state of preceding siblings */
2336 	dip = find_sibling(ddi_get_child(pdip), cname, caddr,
2337 	    FIND_NODE_BY_DRIVER, NULL);
2338 	if (dip)
2339 		return (dip);
2340 
2341 	return (find_sibling(ddi_get_child(pdip), cname, caddr,
2342 	    FIND_NODE_BY_DRIVER|FIND_ADDR_BY_INIT, NULL));
2343 }
2344 
2345 /*
2346  * Find a child of a given address, invoking initchild to name
2347  * unnamed children. cname is the node name.
2348  *
2349  * NOTE: This function is only used during boot. One would hope that
2350  * unique sibling unit-addresses on hardware branches of the tree would
2351  * be a requirement to avoid two drivers trying to control the same
2352  * piece of hardware. Unfortunately there are some cases where this
2353  * situation exists (/ssm@0,0/pci@1c,700000 /ssm@0,0/sghsc@1c,700000).
2354  * Until unit-address uniqueness of siblings is guaranteed, use of this
2355  * interface for purposes other than boot should be avoided.
2356  */
2357 static dev_info_t *
2358 find_child_by_addr(dev_info_t *pdip, char *caddr)
2359 {
2360 	dev_info_t	*dip;
2361 
2362 	/* return NULL if called without a unit-address */
2363 	if ((caddr == NULL) || (*caddr == '\0'))
2364 		return (NULL);
2365 
2366 	/* attempt search without changing state of preceding siblings */
2367 	dip = find_sibling(ddi_get_child(pdip), NULL, caddr,
2368 	    FIND_NODE_BY_ADDR, NULL);
2369 	if (dip)
2370 		return (dip);
2371 
2372 	return (find_sibling(ddi_get_child(pdip), NULL, caddr,
2373 	    FIND_NODE_BY_ADDR|FIND_ADDR_BY_INIT, NULL));
2374 }
2375 
2376 /*
2377  * Deleting a property list. Take care, since some property structures
2378  * may not be fully built.
2379  */
2380 void
2381 i_ddi_prop_list_delete(ddi_prop_t *prop)
2382 {
2383 	while (prop) {
2384 		ddi_prop_t *next = prop->prop_next;
2385 		if (prop->prop_name)
2386 			kmem_free(prop->prop_name, strlen(prop->prop_name) + 1);
2387 		if ((prop->prop_len != 0) && prop->prop_val)
2388 			kmem_free(prop->prop_val, prop->prop_len);
2389 		kmem_free(prop, sizeof (struct ddi_prop));
2390 		prop = next;
2391 	}
2392 }
2393 
2394 /*
2395  * Duplicate property list
2396  */
2397 ddi_prop_t *
2398 i_ddi_prop_list_dup(ddi_prop_t *prop, uint_t flag)
2399 {
2400 	ddi_prop_t *result, *prev, *copy;
2401 
2402 	if (prop == NULL)
2403 		return (NULL);
2404 
2405 	result = prev = NULL;
2406 	for (; prop != NULL; prop = prop->prop_next) {
2407 		ASSERT(prop->prop_name != NULL);
2408 		copy = kmem_zalloc(sizeof (struct ddi_prop), flag);
2409 		if (copy == NULL)
2410 			goto fail;
2411 
2412 		copy->prop_dev = prop->prop_dev;
2413 		copy->prop_flags = prop->prop_flags;
2414 		copy->prop_name = i_ddi_strdup(prop->prop_name, flag);
2415 		if (copy->prop_name == NULL)
2416 			goto fail;
2417 
2418 		if ((copy->prop_len = prop->prop_len) != 0) {
2419 			copy->prop_val = kmem_zalloc(prop->prop_len, flag);
2420 			if (copy->prop_val == NULL)
2421 				goto fail;
2422 
2423 			bcopy(prop->prop_val, copy->prop_val, prop->prop_len);
2424 		}
2425 
2426 		if (prev == NULL)
2427 			result = prev = copy;
2428 		else
2429 			prev->prop_next = copy;
2430 		prev = copy;
2431 	}
2432 	return (result);
2433 
2434 fail:
2435 	i_ddi_prop_list_delete(result);
2436 	return (NULL);
2437 }
2438 
2439 /*
2440  * Create a reference property list, currently used only for
2441  * driver global properties. Created with ref count of 1.
2442  */
2443 ddi_prop_list_t *
2444 i_ddi_prop_list_create(ddi_prop_t *props)
2445 {
2446 	ddi_prop_list_t *list = kmem_alloc(sizeof (*list), KM_SLEEP);
2447 	list->prop_list = props;
2448 	list->prop_ref = 1;
2449 	return (list);
2450 }
2451 
2452 /*
2453  * Increment/decrement reference count. The reference is
2454  * protected by dn_lock. The only interfaces modifying
2455  * dn_global_prop_ptr is in impl_make[free]_parlist().
2456  */
2457 void
2458 i_ddi_prop_list_hold(ddi_prop_list_t *prop_list, struct devnames *dnp)
2459 {
2460 	ASSERT(prop_list->prop_ref >= 0);
2461 	ASSERT(mutex_owned(&dnp->dn_lock));
2462 	prop_list->prop_ref++;
2463 }
2464 
2465 void
2466 i_ddi_prop_list_rele(ddi_prop_list_t *prop_list, struct devnames *dnp)
2467 {
2468 	ASSERT(prop_list->prop_ref > 0);
2469 	ASSERT(mutex_owned(&dnp->dn_lock));
2470 	prop_list->prop_ref--;
2471 
2472 	if (prop_list->prop_ref == 0) {
2473 		i_ddi_prop_list_delete(prop_list->prop_list);
2474 		kmem_free(prop_list, sizeof (*prop_list));
2475 	}
2476 }
2477 
2478 /*
2479  * Free table of classes by drivers
2480  */
2481 void
2482 i_ddi_free_exported_classes(char **classes, int n)
2483 {
2484 	if ((n == 0) || (classes == NULL))
2485 		return;
2486 
2487 	kmem_free(classes, n * sizeof (char *));
2488 }
2489 
2490 /*
2491  * Get all classes exported by dip
2492  */
2493 int
2494 i_ddi_get_exported_classes(dev_info_t *dip, char ***classes)
2495 {
2496 	extern void lock_hw_class_list();
2497 	extern void unlock_hw_class_list();
2498 	extern int get_class(const char *, char **);
2499 
2500 	static char *rootclass = "root";
2501 	int n = 0, nclass = 0;
2502 	char **buf;
2503 
2504 	ASSERT(i_ddi_node_state(dip) >= DS_BOUND);
2505 
2506 	if (dip == ddi_root_node())	/* rootnode exports class "root" */
2507 		nclass = 1;
2508 	lock_hw_class_list();
2509 	nclass += get_class(ddi_driver_name(dip), NULL);
2510 	if (nclass == 0) {
2511 		unlock_hw_class_list();
2512 		return (0);		/* no class exported */
2513 	}
2514 
2515 	*classes = buf = kmem_alloc(nclass * sizeof (char *), KM_SLEEP);
2516 	if (dip == ddi_root_node()) {
2517 		*buf++ = rootclass;
2518 		n = 1;
2519 	}
2520 	n += get_class(ddi_driver_name(dip), buf);
2521 	unlock_hw_class_list();
2522 
2523 	ASSERT(n == nclass);	/* make sure buf wasn't overrun */
2524 	return (nclass);
2525 }
2526 
2527 /*
2528  * Helper functions, returns NULL if no memory.
2529  */
2530 char *
2531 i_ddi_strdup(char *str, uint_t flag)
2532 {
2533 	char *copy;
2534 
2535 	if (str == NULL)
2536 		return (NULL);
2537 
2538 	copy = kmem_alloc(strlen(str) + 1, flag);
2539 	if (copy == NULL)
2540 		return (NULL);
2541 
2542 	(void) strcpy(copy, str);
2543 	return (copy);
2544 }
2545 
2546 /*
2547  * Load driver.conf file for major. Load all if major == -1.
2548  *
2549  * This is called
2550  * - early in boot after devnames array is initialized
2551  * - from vfs code when certain file systems are mounted
2552  * - from add_drv when a new driver is added
2553  */
2554 int
2555 i_ddi_load_drvconf(major_t major)
2556 {
2557 	extern int modrootloaded;
2558 
2559 	major_t low, high, m;
2560 
2561 	if (major == DDI_MAJOR_T_NONE) {
2562 		low = 0;
2563 		high = devcnt - 1;
2564 	} else {
2565 		if (major >= devcnt)
2566 			return (EINVAL);
2567 		low = high = major;
2568 	}
2569 
2570 	for (m = low; m <= high; m++) {
2571 		struct devnames *dnp = &devnamesp[m];
2572 		LOCK_DEV_OPS(&dnp->dn_lock);
2573 		dnp->dn_flags &= ~(DN_DRIVER_HELD|DN_DRIVER_INACTIVE);
2574 		(void) impl_make_parlist(m);
2575 		UNLOCK_DEV_OPS(&dnp->dn_lock);
2576 	}
2577 
2578 	if (modrootloaded) {
2579 		ddi_walk_devs(ddi_root_node(), reset_nexus_flags,
2580 		    (void *)(uintptr_t)major);
2581 	}
2582 
2583 	/* build dn_list from old entries in path_to_inst */
2584 	e_ddi_unorphan_instance_nos();
2585 	return (0);
2586 }
2587 
2588 /*
2589  * Unload a specific driver.conf.
2590  * Don't support unload all because it doesn't make any sense
2591  */
2592 int
2593 i_ddi_unload_drvconf(major_t major)
2594 {
2595 	int error;
2596 	struct devnames *dnp;
2597 
2598 	if (major >= devcnt)
2599 		return (EINVAL);
2600 
2601 	/*
2602 	 * Take the per-driver lock while unloading driver.conf
2603 	 */
2604 	dnp = &devnamesp[major];
2605 	LOCK_DEV_OPS(&dnp->dn_lock);
2606 	error = impl_free_parlist(major);
2607 	UNLOCK_DEV_OPS(&dnp->dn_lock);
2608 	return (error);
2609 }
2610 
2611 /*
2612  * Merge a .conf node. This is called by nexus drivers to augment
2613  * hw node with properties specified in driver.conf file. This function
2614  * takes a callback routine to name nexus children.
2615  * The parent node must be held busy.
2616  *
2617  * It returns DDI_SUCCESS if the node is merged and DDI_FAILURE otherwise.
2618  */
2619 int
2620 ndi_merge_node(dev_info_t *dip, int (*make_ua)(dev_info_t *, char *, int))
2621 {
2622 	dev_info_t *hwdip;
2623 
2624 	ASSERT(ndi_dev_is_persistent_node(dip) == 0);
2625 	ASSERT(ddi_get_name_addr(dip) != NULL);
2626 
2627 	hwdip = ndi_devi_findchild_by_callback(ddi_get_parent(dip),
2628 	    ddi_binding_name(dip), ddi_get_name_addr(dip), make_ua);
2629 
2630 	/*
2631 	 * Look for the hardware node that is the target of the merge;
2632 	 * return failure if not found.
2633 	 */
2634 	if ((hwdip == NULL) || (hwdip == dip)) {
2635 		char *buf = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2636 		NDI_CONFIG_DEBUG((CE_WARN, "No HW node to merge conf node %s",
2637 		    ddi_deviname(dip, buf)));
2638 		kmem_free(buf, MAXNAMELEN);
2639 		return (DDI_FAILURE);
2640 	}
2641 
2642 	/*
2643 	 * Make sure the hardware node is uninitialized and has no property.
2644 	 * This may not be the case if new .conf files are load after some
2645 	 * hardware nodes have already been initialized and attached.
2646 	 *
2647 	 * N.B. We return success here because the node was *intended*
2648 	 *	to be a merge node because there is a hw node with the name.
2649 	 */
2650 	mutex_enter(&DEVI(hwdip)->devi_lock);
2651 	if (ndi_dev_is_persistent_node(hwdip) == 0) {
2652 		char *buf;
2653 		mutex_exit(&DEVI(hwdip)->devi_lock);
2654 
2655 		buf = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2656 		NDI_CONFIG_DEBUG((CE_NOTE, "Duplicate .conf node %s",
2657 		    ddi_deviname(dip, buf)));
2658 		kmem_free(buf, MAXNAMELEN);
2659 		return (DDI_SUCCESS);
2660 	}
2661 
2662 	/*
2663 	 * If it is possible that the hardware has already been touched
2664 	 * then don't merge.
2665 	 */
2666 	if (i_ddi_node_state(hwdip) >= DS_INITIALIZED ||
2667 	    (DEVI(hwdip)->devi_sys_prop_ptr != NULL) ||
2668 	    (DEVI(hwdip)->devi_drv_prop_ptr != NULL)) {
2669 		char *buf;
2670 		mutex_exit(&DEVI(hwdip)->devi_lock);
2671 
2672 		buf = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2673 		NDI_CONFIG_DEBUG((CE_NOTE,
2674 		    "!Cannot merge .conf node %s with hw node %p "
2675 		    "-- not in proper state",
2676 		    ddi_deviname(dip, buf), (void *)hwdip));
2677 		kmem_free(buf, MAXNAMELEN);
2678 		return (DDI_SUCCESS);
2679 	}
2680 
2681 	mutex_enter(&DEVI(dip)->devi_lock);
2682 	DEVI(hwdip)->devi_sys_prop_ptr = DEVI(dip)->devi_sys_prop_ptr;
2683 	DEVI(hwdip)->devi_drv_prop_ptr = DEVI(dip)->devi_drv_prop_ptr;
2684 	DEVI(dip)->devi_sys_prop_ptr = NULL;
2685 	DEVI(dip)->devi_drv_prop_ptr = NULL;
2686 	mutex_exit(&DEVI(dip)->devi_lock);
2687 	mutex_exit(&DEVI(hwdip)->devi_lock);
2688 
2689 	return (DDI_SUCCESS);
2690 }
2691 
2692 /*
2693  * Merge a "wildcard" .conf node. This is called by nexus drivers to
2694  * augment a set of hw node with properties specified in driver.conf file.
2695  * The parent node must be held busy.
2696  *
2697  * There is no failure mode, since the nexus may or may not have child
2698  * node bound the driver specified by the wildcard node.
2699  */
2700 void
2701 ndi_merge_wildcard_node(dev_info_t *dip)
2702 {
2703 	dev_info_t *hwdip;
2704 	dev_info_t *pdip = ddi_get_parent(dip);
2705 	major_t major = ddi_driver_major(dip);
2706 
2707 	/* never attempt to merge a hw node */
2708 	ASSERT(ndi_dev_is_persistent_node(dip) == 0);
2709 	/* must be bound to a driver major number */
2710 	ASSERT(major != DDI_MAJOR_T_NONE);
2711 
2712 	/*
2713 	 * Walk the child list to find all nodes bound to major
2714 	 * and copy properties.
2715 	 */
2716 	mutex_enter(&DEVI(dip)->devi_lock);
2717 	ASSERT(DEVI_BUSY_OWNED(pdip));
2718 	for (hwdip = ddi_get_child(pdip); hwdip;
2719 	    hwdip = ddi_get_next_sibling(hwdip)) {
2720 		/*
2721 		 * Skip nodes not bound to same driver
2722 		 */
2723 		if (ddi_driver_major(hwdip) != major)
2724 			continue;
2725 
2726 		/*
2727 		 * Skip .conf nodes
2728 		 */
2729 		if (ndi_dev_is_persistent_node(hwdip) == 0)
2730 			continue;
2731 
2732 		/*
2733 		 * Make sure the node is uninitialized and has no property.
2734 		 */
2735 		mutex_enter(&DEVI(hwdip)->devi_lock);
2736 		if (i_ddi_node_state(hwdip) >= DS_INITIALIZED ||
2737 		    (DEVI(hwdip)->devi_sys_prop_ptr != NULL) ||
2738 		    (DEVI(hwdip)->devi_drv_prop_ptr != NULL)) {
2739 			mutex_exit(&DEVI(hwdip)->devi_lock);
2740 			NDI_CONFIG_DEBUG((CE_NOTE, "HW node %p state not "
2741 			    "suitable for merging wildcard conf node %s",
2742 			    (void *)hwdip, ddi_node_name(dip)));
2743 			continue;
2744 		}
2745 
2746 		DEVI(hwdip)->devi_sys_prop_ptr =
2747 		    i_ddi_prop_list_dup(DEVI(dip)->devi_sys_prop_ptr, KM_SLEEP);
2748 		DEVI(hwdip)->devi_drv_prop_ptr =
2749 		    i_ddi_prop_list_dup(DEVI(dip)->devi_drv_prop_ptr, KM_SLEEP);
2750 		mutex_exit(&DEVI(hwdip)->devi_lock);
2751 	}
2752 	mutex_exit(&DEVI(dip)->devi_lock);
2753 }
2754 
2755 /*
2756  * Return the major number based on the compatible property. This interface
2757  * may be used in situations where we are trying to detect if a better driver
2758  * now exists for a device, so it must use the 'compatible' property.  If
2759  * a non-NULL formp is specified and the binding was based on compatible then
2760  * return the pointer to the form used in *formp.
2761  */
2762 major_t
2763 ddi_compatible_driver_major(dev_info_t *dip, char **formp)
2764 {
2765 	struct dev_info *devi = DEVI(dip);
2766 	void		*compat;
2767 	size_t		len;
2768 	char		*p = NULL;
2769 	major_t		major = DDI_MAJOR_T_NONE;
2770 
2771 	if (formp)
2772 		*formp = NULL;
2773 
2774 	if (ddi_prop_exists(DDI_DEV_T_NONE, dip, DDI_PROP_DONTPASS,
2775 	    "ddi-assigned")) {
2776 		major = ddi_name_to_major("nulldriver");
2777 		return (major);
2778 	}
2779 
2780 	/*
2781 	 * Highest precedence binding is a path-oriented alias. Since this
2782 	 * requires a 'path', this type of binding occurs via more obtuse
2783 	 * 'rebind'. The need for a path-oriented alias 'rebind' is detected
2784 	 * after a successful DDI_CTLOPS_INITCHILD to another driver: this is
2785 	 * is the first point at which the unit-address (or instance) of the
2786 	 * last component of the path is available (even though the path is
2787 	 * bound to the wrong driver at this point).
2788 	 */
2789 	if (devi->devi_flags & DEVI_REBIND) {
2790 		p = devi->devi_rebinding_name;
2791 		major = ddi_name_to_major(p);
2792 		if (driver_active(major)) {
2793 			if (formp)
2794 				*formp = p;
2795 			return (major);
2796 		}
2797 
2798 		/*
2799 		 * If for some reason devi_rebinding_name no longer resolves
2800 		 * to a proper driver then clear DEVI_REBIND.
2801 		 */
2802 		mutex_enter(&devi->devi_lock);
2803 		devi->devi_flags &= ~DEVI_REBIND;
2804 		mutex_exit(&devi->devi_lock);
2805 	}
2806 
2807 	/* look up compatible property */
2808 	(void) lookup_compatible(dip, KM_SLEEP);
2809 	compat = (void *)(devi->devi_compat_names);
2810 	len = devi->devi_compat_length;
2811 
2812 	/* find the highest precedence compatible form with a driver binding */
2813 	while ((p = prom_decode_composite_string(compat, len, p)) != NULL) {
2814 		major = ddi_name_to_major(p);
2815 		if (driver_active(major)) {
2816 			if (formp)
2817 				*formp = p;
2818 			return (major);
2819 		}
2820 	}
2821 
2822 	/*
2823 	 * none of the compatible forms have a driver binding, see if
2824 	 * the node name has a driver binding.
2825 	 */
2826 	major = ddi_name_to_major(ddi_node_name(dip));
2827 	if (driver_active(major))
2828 		return (major);
2829 
2830 	/* no driver */
2831 	return (DDI_MAJOR_T_NONE);
2832 }
2833 
2834 /*
2835  * Static help functions
2836  */
2837 
2838 /*
2839  * lookup the "compatible" property and cache it's contents in the
2840  * device node.
2841  */
2842 static int
2843 lookup_compatible(dev_info_t *dip, uint_t flag)
2844 {
2845 	int rv;
2846 	int prop_flags;
2847 	uint_t ncompatstrs;
2848 	char **compatstrpp;
2849 	char *di_compat_strp;
2850 	size_t di_compat_strlen;
2851 
2852 	if (DEVI(dip)->devi_compat_names) {
2853 		return (DDI_SUCCESS);
2854 	}
2855 
2856 	prop_flags = DDI_PROP_TYPE_STRING | DDI_PROP_DONTPASS;
2857 
2858 	if (flag & KM_NOSLEEP) {
2859 		prop_flags |= DDI_PROP_DONTSLEEP;
2860 	}
2861 
2862 	if (ndi_dev_is_prom_node(dip) == 0) {
2863 		prop_flags |= DDI_PROP_NOTPROM;
2864 	}
2865 
2866 	rv = ddi_prop_lookup_common(DDI_DEV_T_ANY, dip, prop_flags,
2867 	    "compatible", &compatstrpp, &ncompatstrs,
2868 	    ddi_prop_fm_decode_strings);
2869 
2870 	if (rv == DDI_PROP_NOT_FOUND) {
2871 		return (DDI_SUCCESS);
2872 	}
2873 
2874 	if (rv != DDI_PROP_SUCCESS) {
2875 		return (DDI_FAILURE);
2876 	}
2877 
2878 	/*
2879 	 * encode the compatible property data in the dev_info node
2880 	 */
2881 	rv = DDI_SUCCESS;
2882 	if (ncompatstrs != 0) {
2883 		di_compat_strp = encode_composite_string(compatstrpp,
2884 		    ncompatstrs, &di_compat_strlen, flag);
2885 		if (di_compat_strp != NULL) {
2886 			DEVI(dip)->devi_compat_names = di_compat_strp;
2887 			DEVI(dip)->devi_compat_length = di_compat_strlen;
2888 		} else {
2889 			rv = DDI_FAILURE;
2890 		}
2891 	}
2892 	ddi_prop_free(compatstrpp);
2893 	return (rv);
2894 }
2895 
2896 /*
2897  * Create a composite string from a list of strings.
2898  *
2899  * A composite string consists of a single buffer containing one
2900  * or more NULL terminated strings.
2901  */
2902 static char *
2903 encode_composite_string(char **strings, uint_t nstrings, size_t *retsz,
2904     uint_t flag)
2905 {
2906 	uint_t index;
2907 	char  **strpp;
2908 	uint_t slen;
2909 	size_t cbuf_sz = 0;
2910 	char *cbuf_p;
2911 	char *cbuf_ip;
2912 
2913 	if (strings == NULL || nstrings == 0 || retsz == NULL) {
2914 		return (NULL);
2915 	}
2916 
2917 	for (index = 0, strpp = strings; index < nstrings; index++)
2918 		cbuf_sz += strlen(*(strpp++)) + 1;
2919 
2920 	if ((cbuf_p = kmem_alloc(cbuf_sz, flag)) == NULL) {
2921 		cmn_err(CE_NOTE,
2922 		    "?failed to allocate device node compatstr");
2923 		return (NULL);
2924 	}
2925 
2926 	cbuf_ip = cbuf_p;
2927 	for (index = 0, strpp = strings; index < nstrings; index++) {
2928 		slen = strlen(*strpp);
2929 		bcopy(*(strpp++), cbuf_ip, slen);
2930 		cbuf_ip += slen;
2931 		*(cbuf_ip++) = '\0';
2932 	}
2933 
2934 	*retsz = cbuf_sz;
2935 	return (cbuf_p);
2936 }
2937 
2938 static void
2939 link_to_driver_list(dev_info_t *dip)
2940 {
2941 	major_t major = DEVI(dip)->devi_major;
2942 	struct devnames *dnp;
2943 
2944 	ASSERT(major != DDI_MAJOR_T_NONE);
2945 
2946 	/*
2947 	 * Remove from orphan list
2948 	 */
2949 	if (ndi_dev_is_persistent_node(dip)) {
2950 		dnp = &orphanlist;
2951 		remove_from_dn_list(dnp, dip);
2952 	}
2953 
2954 	/*
2955 	 * Add to per driver list
2956 	 */
2957 	dnp = &devnamesp[major];
2958 	add_to_dn_list(dnp, dip);
2959 }
2960 
2961 static void
2962 unlink_from_driver_list(dev_info_t *dip)
2963 {
2964 	major_t major = DEVI(dip)->devi_major;
2965 	struct devnames *dnp;
2966 
2967 	ASSERT(major != DDI_MAJOR_T_NONE);
2968 
2969 	/*
2970 	 * Remove from per-driver list
2971 	 */
2972 	dnp = &devnamesp[major];
2973 	remove_from_dn_list(dnp, dip);
2974 
2975 	/*
2976 	 * Add to orphan list
2977 	 */
2978 	if (ndi_dev_is_persistent_node(dip)) {
2979 		dnp = &orphanlist;
2980 		add_to_dn_list(dnp, dip);
2981 	}
2982 }
2983 
2984 /*
2985  * scan the per-driver list looking for dev_info "dip"
2986  */
2987 static dev_info_t *
2988 in_dn_list(struct devnames *dnp, dev_info_t *dip)
2989 {
2990 	struct dev_info *idevi;
2991 
2992 	if ((idevi = DEVI(dnp->dn_head)) == NULL)
2993 		return (NULL);
2994 
2995 	while (idevi) {
2996 		if (idevi == DEVI(dip))
2997 			return (dip);
2998 		idevi = idevi->devi_next;
2999 	}
3000 	return (NULL);
3001 }
3002 
3003 /*
3004  * insert devinfo node 'dip' into the per-driver instance list
3005  * headed by 'dnp'
3006  *
3007  * Nodes on the per-driver list are ordered: HW - SID - PSEUDO.  The order is
3008  * required for merging of .conf file data to work properly.
3009  */
3010 static void
3011 add_to_ordered_dn_list(struct devnames *dnp, dev_info_t *dip)
3012 {
3013 	dev_info_t **dipp;
3014 
3015 	ASSERT(mutex_owned(&(dnp->dn_lock)));
3016 
3017 	dipp = &dnp->dn_head;
3018 	if (ndi_dev_is_prom_node(dip)) {
3019 		/*
3020 		 * Find the first non-prom node or end of list
3021 		 */
3022 		while (*dipp && (ndi_dev_is_prom_node(*dipp) != 0)) {
3023 			dipp = (dev_info_t **)&DEVI(*dipp)->devi_next;
3024 		}
3025 	} else if (ndi_dev_is_persistent_node(dip)) {
3026 		/*
3027 		 * Find the first non-persistent node
3028 		 */
3029 		while (*dipp && (ndi_dev_is_persistent_node(*dipp) != 0)) {
3030 			dipp = (dev_info_t **)&DEVI(*dipp)->devi_next;
3031 		}
3032 	} else {
3033 		/*
3034 		 * Find the end of the list
3035 		 */
3036 		while (*dipp) {
3037 			dipp = (dev_info_t **)&DEVI(*dipp)->devi_next;
3038 		}
3039 	}
3040 
3041 	DEVI(dip)->devi_next = DEVI(*dipp);
3042 	*dipp = dip;
3043 }
3044 
3045 /*
3046  * add a list of device nodes to the device node list in the
3047  * devnames structure
3048  */
3049 static void
3050 add_to_dn_list(struct devnames *dnp, dev_info_t *dip)
3051 {
3052 	/*
3053 	 * Look to see if node already exists
3054 	 */
3055 	LOCK_DEV_OPS(&(dnp->dn_lock));
3056 	if (in_dn_list(dnp, dip)) {
3057 		cmn_err(CE_NOTE, "add_to_dn_list: node %s already in list",
3058 		    DEVI(dip)->devi_node_name);
3059 	} else {
3060 		add_to_ordered_dn_list(dnp, dip);
3061 	}
3062 	UNLOCK_DEV_OPS(&(dnp->dn_lock));
3063 }
3064 
3065 static void
3066 remove_from_dn_list(struct devnames *dnp, dev_info_t *dip)
3067 {
3068 	dev_info_t **plist;
3069 
3070 	LOCK_DEV_OPS(&(dnp->dn_lock));
3071 
3072 	plist = (dev_info_t **)&dnp->dn_head;
3073 	while (*plist && (*plist != dip)) {
3074 		plist = (dev_info_t **)&DEVI(*plist)->devi_next;
3075 	}
3076 
3077 	if (*plist != NULL) {
3078 		ASSERT(*plist == dip);
3079 		*plist = (dev_info_t *)(DEVI(dip)->devi_next);
3080 		DEVI(dip)->devi_next = NULL;
3081 	} else {
3082 		NDI_CONFIG_DEBUG((CE_NOTE,
3083 		    "remove_from_dn_list: node %s not found in list",
3084 		    DEVI(dip)->devi_node_name));
3085 	}
3086 
3087 	UNLOCK_DEV_OPS(&(dnp->dn_lock));
3088 }
3089 
3090 /*
3091  * Add and remove reference driver global property list
3092  */
3093 static void
3094 add_global_props(dev_info_t *dip)
3095 {
3096 	struct devnames *dnp;
3097 	ddi_prop_list_t *plist;
3098 
3099 	ASSERT(DEVI(dip)->devi_global_prop_list == NULL);
3100 	ASSERT(DEVI(dip)->devi_major != DDI_MAJOR_T_NONE);
3101 
3102 	dnp = &devnamesp[DEVI(dip)->devi_major];
3103 	LOCK_DEV_OPS(&dnp->dn_lock);
3104 	plist = dnp->dn_global_prop_ptr;
3105 	if (plist == NULL) {
3106 		UNLOCK_DEV_OPS(&dnp->dn_lock);
3107 		return;
3108 	}
3109 	i_ddi_prop_list_hold(plist, dnp);
3110 	UNLOCK_DEV_OPS(&dnp->dn_lock);
3111 
3112 	mutex_enter(&DEVI(dip)->devi_lock);
3113 	DEVI(dip)->devi_global_prop_list = plist;
3114 	mutex_exit(&DEVI(dip)->devi_lock);
3115 }
3116 
3117 static void
3118 remove_global_props(dev_info_t *dip)
3119 {
3120 	ddi_prop_list_t *proplist;
3121 
3122 	mutex_enter(&DEVI(dip)->devi_lock);
3123 	proplist = DEVI(dip)->devi_global_prop_list;
3124 	DEVI(dip)->devi_global_prop_list = NULL;
3125 	mutex_exit(&DEVI(dip)->devi_lock);
3126 
3127 	if (proplist) {
3128 		major_t major;
3129 		struct devnames *dnp;
3130 
3131 		major = ddi_driver_major(dip);
3132 		ASSERT(major != DDI_MAJOR_T_NONE);
3133 		dnp = &devnamesp[major];
3134 		LOCK_DEV_OPS(&dnp->dn_lock);
3135 		i_ddi_prop_list_rele(proplist, dnp);
3136 		UNLOCK_DEV_OPS(&dnp->dn_lock);
3137 	}
3138 }
3139 
3140 #ifdef DEBUG
3141 /*
3142  * Set this variable to '0' to disable the optimization,
3143  * and to 2 to print debug message.
3144  */
3145 static int optimize_dtree = 1;
3146 
3147 static void
3148 debug_dtree(dev_info_t *devi, struct dev_info *adevi, char *service)
3149 {
3150 	char *adeviname, *buf;
3151 
3152 	/*
3153 	 * Don't print unless optimize dtree is set to 2+
3154 	 */
3155 	if (optimize_dtree <= 1)
3156 		return;
3157 
3158 	buf = kmem_alloc(MAXNAMELEN, KM_SLEEP);
3159 	adeviname = ddi_deviname((dev_info_t *)adevi, buf);
3160 	if (*adeviname == '\0')
3161 		adeviname = "root";
3162 
3163 	cmn_err(CE_CONT, "%s %s -> %s\n",
3164 	    ddi_deviname(devi, buf), service, adeviname);
3165 
3166 	kmem_free(buf, MAXNAMELEN);
3167 }
3168 #else /* DEBUG */
3169 #define	debug_dtree(a1, a2, a3)	 /* nothing */
3170 #endif	/* DEBUG */
3171 
3172 static void
3173 ddi_optimize_dtree(dev_info_t *devi)
3174 {
3175 	struct dev_info *pdevi;
3176 	struct bus_ops *b;
3177 
3178 	pdevi = DEVI(devi)->devi_parent;
3179 	ASSERT(pdevi);
3180 
3181 	/*
3182 	 * Set the unoptimized values
3183 	 */
3184 	DEVI(devi)->devi_bus_map_fault = pdevi;
3185 	DEVI(devi)->devi_bus_dma_map = pdevi;
3186 	DEVI(devi)->devi_bus_dma_allochdl = pdevi;
3187 	DEVI(devi)->devi_bus_dma_freehdl = pdevi;
3188 	DEVI(devi)->devi_bus_dma_bindhdl = pdevi;
3189 	DEVI(devi)->devi_bus_dma_bindfunc =
3190 	    pdevi->devi_ops->devo_bus_ops->bus_dma_bindhdl;
3191 	DEVI(devi)->devi_bus_dma_unbindhdl = pdevi;
3192 	DEVI(devi)->devi_bus_dma_unbindfunc =
3193 	    pdevi->devi_ops->devo_bus_ops->bus_dma_unbindhdl;
3194 	DEVI(devi)->devi_bus_dma_flush = pdevi;
3195 	DEVI(devi)->devi_bus_dma_win = pdevi;
3196 	DEVI(devi)->devi_bus_dma_ctl = pdevi;
3197 	DEVI(devi)->devi_bus_ctl = pdevi;
3198 
3199 #ifdef DEBUG
3200 	if (optimize_dtree == 0)
3201 		return;
3202 #endif /* DEBUG */
3203 
3204 	b = pdevi->devi_ops->devo_bus_ops;
3205 
3206 	if (i_ddi_map_fault == b->bus_map_fault) {
3207 		DEVI(devi)->devi_bus_map_fault = pdevi->devi_bus_map_fault;
3208 		debug_dtree(devi, DEVI(devi)->devi_bus_map_fault,
3209 		    "bus_map_fault");
3210 	}
3211 
3212 	if (ddi_dma_map == b->bus_dma_map) {
3213 		DEVI(devi)->devi_bus_dma_map = pdevi->devi_bus_dma_map;
3214 		debug_dtree(devi, DEVI(devi)->devi_bus_dma_map, "bus_dma_map");
3215 	}
3216 
3217 	if (ddi_dma_allochdl == b->bus_dma_allochdl) {
3218 		DEVI(devi)->devi_bus_dma_allochdl =
3219 		    pdevi->devi_bus_dma_allochdl;
3220 		debug_dtree(devi, DEVI(devi)->devi_bus_dma_allochdl,
3221 		    "bus_dma_allochdl");
3222 	}
3223 
3224 	if (ddi_dma_freehdl == b->bus_dma_freehdl) {
3225 		DEVI(devi)->devi_bus_dma_freehdl = pdevi->devi_bus_dma_freehdl;
3226 		debug_dtree(devi, DEVI(devi)->devi_bus_dma_freehdl,
3227 		    "bus_dma_freehdl");
3228 	}
3229 
3230 	if (ddi_dma_bindhdl == b->bus_dma_bindhdl) {
3231 		DEVI(devi)->devi_bus_dma_bindhdl = pdevi->devi_bus_dma_bindhdl;
3232 		DEVI(devi)->devi_bus_dma_bindfunc =
3233 		    pdevi->devi_bus_dma_bindhdl->devi_ops->
3234 		    devo_bus_ops->bus_dma_bindhdl;
3235 		debug_dtree(devi, DEVI(devi)->devi_bus_dma_bindhdl,
3236 		    "bus_dma_bindhdl");
3237 	}
3238 
3239 	if (ddi_dma_unbindhdl == b->bus_dma_unbindhdl) {
3240 		DEVI(devi)->devi_bus_dma_unbindhdl =
3241 		    pdevi->devi_bus_dma_unbindhdl;
3242 		DEVI(devi)->devi_bus_dma_unbindfunc =
3243 		    pdevi->devi_bus_dma_unbindhdl->devi_ops->
3244 		    devo_bus_ops->bus_dma_unbindhdl;
3245 		debug_dtree(devi, DEVI(devi)->devi_bus_dma_unbindhdl,
3246 		    "bus_dma_unbindhdl");
3247 	}
3248 
3249 	if (ddi_dma_flush == b->bus_dma_flush) {
3250 		DEVI(devi)->devi_bus_dma_flush = pdevi->devi_bus_dma_flush;
3251 		debug_dtree(devi, DEVI(devi)->devi_bus_dma_flush,
3252 		    "bus_dma_flush");
3253 	}
3254 
3255 	if (ddi_dma_win == b->bus_dma_win) {
3256 		DEVI(devi)->devi_bus_dma_win = pdevi->devi_bus_dma_win;
3257 		debug_dtree(devi, DEVI(devi)->devi_bus_dma_win,
3258 		    "bus_dma_win");
3259 	}
3260 
3261 	if (ddi_dma_mctl == b->bus_dma_ctl) {
3262 		DEVI(devi)->devi_bus_dma_ctl = pdevi->devi_bus_dma_ctl;
3263 		debug_dtree(devi, DEVI(devi)->devi_bus_dma_ctl, "bus_dma_ctl");
3264 	}
3265 
3266 	if (ddi_ctlops == b->bus_ctl) {
3267 		DEVI(devi)->devi_bus_ctl = pdevi->devi_bus_ctl;
3268 		debug_dtree(devi, DEVI(devi)->devi_bus_ctl, "bus_ctl");
3269 	}
3270 }
3271 
3272 #define	MIN_DEVINFO_LOG_SIZE	max_ncpus
3273 #define	MAX_DEVINFO_LOG_SIZE	max_ncpus * 10
3274 
3275 static void
3276 da_log_init()
3277 {
3278 	devinfo_log_header_t *dh;
3279 	int logsize = devinfo_log_size;
3280 
3281 	if (logsize == 0)
3282 		logsize = MIN_DEVINFO_LOG_SIZE;
3283 	else if (logsize > MAX_DEVINFO_LOG_SIZE)
3284 		logsize = MAX_DEVINFO_LOG_SIZE;
3285 
3286 	dh = kmem_alloc(logsize * PAGESIZE, KM_SLEEP);
3287 	mutex_init(&dh->dh_lock, NULL, MUTEX_DEFAULT, NULL);
3288 	dh->dh_max = ((logsize * PAGESIZE) - sizeof (*dh)) /
3289 	    sizeof (devinfo_audit_t) + 1;
3290 	dh->dh_curr = -1;
3291 	dh->dh_hits = 0;
3292 
3293 	devinfo_audit_log = dh;
3294 }
3295 
3296 /*
3297  * Log the stack trace in per-devinfo audit structure and also enter
3298  * it into a system wide log for recording the time history.
3299  */
3300 static void
3301 da_log_enter(dev_info_t *dip)
3302 {
3303 	devinfo_audit_t *da_log, *da = DEVI(dip)->devi_audit;
3304 	devinfo_log_header_t *dh = devinfo_audit_log;
3305 
3306 	if (devinfo_audit_log == NULL)
3307 		return;
3308 
3309 	ASSERT(da != NULL);
3310 
3311 	da->da_devinfo = dip;
3312 	da->da_timestamp = gethrtime();
3313 	da->da_thread = curthread;
3314 	da->da_node_state = DEVI(dip)->devi_node_state;
3315 	da->da_device_state = DEVI(dip)->devi_state;
3316 	da->da_depth = getpcstack(da->da_stack, DDI_STACK_DEPTH);
3317 
3318 	/*
3319 	 * Copy into common log and note the location for tracing history
3320 	 */
3321 	mutex_enter(&dh->dh_lock);
3322 	dh->dh_hits++;
3323 	dh->dh_curr++;
3324 	if (dh->dh_curr >= dh->dh_max)
3325 		dh->dh_curr -= dh->dh_max;
3326 	da_log = &dh->dh_entry[dh->dh_curr];
3327 	mutex_exit(&dh->dh_lock);
3328 
3329 	bcopy(da, da_log, sizeof (devinfo_audit_t));
3330 	da->da_lastlog = da_log;
3331 }
3332 
3333 static void
3334 attach_drivers()
3335 {
3336 	int i;
3337 	for (i = 0; i < devcnt; i++) {
3338 		struct devnames *dnp = &devnamesp[i];
3339 		if ((dnp->dn_flags & DN_FORCE_ATTACH) &&
3340 		    (ddi_hold_installed_driver((major_t)i) != NULL))
3341 			ddi_rele_driver((major_t)i);
3342 	}
3343 }
3344 
3345 /*
3346  * Launch a thread to force attach drivers. This avoids penalty on boot time.
3347  */
3348 void
3349 i_ddi_forceattach_drivers()
3350 {
3351 
3352 	/*
3353 	 * Attach IB VHCI driver before the force-attach thread attaches the
3354 	 * IB HCA driver. IB HCA driver will fail if IB Nexus has not yet
3355 	 * been attached.
3356 	 */
3357 	(void) ddi_hold_installed_driver(ddi_name_to_major("ib"));
3358 
3359 	(void) thread_create(NULL, 0, (void (*)())attach_drivers, NULL, 0, &p0,
3360 	    TS_RUN, minclsyspri);
3361 }
3362 
3363 /*
3364  * This is a private DDI interface for optimizing boot performance.
3365  * I/O subsystem initialization is considered complete when devfsadm
3366  * is executed.
3367  *
3368  * NOTE: The start of syseventd happens to be a convenient indicator
3369  *	of the completion of I/O initialization during boot.
3370  *	The implementation should be replaced by something more robust.
3371  */
3372 int
3373 i_ddi_io_initialized()
3374 {
3375 	extern int sysevent_daemon_init;
3376 	return (sysevent_daemon_init);
3377 }
3378 
3379 /*
3380  * May be used to determine system boot state
3381  * "Available" means the system is for the most part up
3382  * and initialized, with all system services either up or
3383  * capable of being started.  This state is set by devfsadm
3384  * during the boot process.  The /dev filesystem infers
3385  * from this when implicit reconfig can be performed,
3386  * ie, devfsadm can be invoked.  Please avoid making
3387  * further use of this unless it's really necessary.
3388  */
3389 int
3390 i_ddi_sysavail()
3391 {
3392 	return (devname_state & DS_SYSAVAIL);
3393 }
3394 
3395 /*
3396  * May be used to determine if boot is a reconfigure boot.
3397  */
3398 int
3399 i_ddi_reconfig()
3400 {
3401 	return (devname_state & DS_RECONFIG);
3402 }
3403 
3404 /*
3405  * Note system services are up, inform /dev.
3406  */
3407 void
3408 i_ddi_set_sysavail()
3409 {
3410 	if ((devname_state & DS_SYSAVAIL) == 0) {
3411 		devname_state |= DS_SYSAVAIL;
3412 		sdev_devstate_change();
3413 	}
3414 }
3415 
3416 /*
3417  * Note reconfiguration boot, inform /dev.
3418  */
3419 void
3420 i_ddi_set_reconfig()
3421 {
3422 	if ((devname_state & DS_RECONFIG) == 0) {
3423 		devname_state |= DS_RECONFIG;
3424 		sdev_devstate_change();
3425 	}
3426 }
3427 
3428 
3429 /*
3430  * device tree walking
3431  */
3432 
3433 struct walk_elem {
3434 	struct walk_elem *next;
3435 	dev_info_t *dip;
3436 };
3437 
3438 static void
3439 free_list(struct walk_elem *list)
3440 {
3441 	while (list) {
3442 		struct walk_elem *next = list->next;
3443 		kmem_free(list, sizeof (*list));
3444 		list = next;
3445 	}
3446 }
3447 
3448 static void
3449 append_node(struct walk_elem **list, dev_info_t *dip)
3450 {
3451 	struct walk_elem *tail;
3452 	struct walk_elem *elem = kmem_alloc(sizeof (*elem), KM_SLEEP);
3453 
3454 	elem->next = NULL;
3455 	elem->dip = dip;
3456 
3457 	if (*list == NULL) {
3458 		*list = elem;
3459 		return;
3460 	}
3461 
3462 	tail = *list;
3463 	while (tail->next)
3464 		tail = tail->next;
3465 
3466 	tail->next = elem;
3467 }
3468 
3469 /*
3470  * The implementation of ddi_walk_devs().
3471  */
3472 static int
3473 walk_devs(dev_info_t *dip, int (*f)(dev_info_t *, void *), void *arg,
3474     int do_locking)
3475 {
3476 	struct walk_elem *head = NULL;
3477 
3478 	/*
3479 	 * Do it in two passes. First pass invoke callback on each
3480 	 * dip on the sibling list. Second pass invoke callback on
3481 	 * children of each dip.
3482 	 */
3483 	while (dip) {
3484 		switch ((*f)(dip, arg)) {
3485 		case DDI_WALK_TERMINATE:
3486 			free_list(head);
3487 			return (DDI_WALK_TERMINATE);
3488 
3489 		case DDI_WALK_PRUNESIB:
3490 			/* ignore sibling by setting dip to NULL */
3491 			append_node(&head, dip);
3492 			dip = NULL;
3493 			break;
3494 
3495 		case DDI_WALK_PRUNECHILD:
3496 			/* don't worry about children */
3497 			dip = ddi_get_next_sibling(dip);
3498 			break;
3499 
3500 		case DDI_WALK_CONTINUE:
3501 		default:
3502 			append_node(&head, dip);
3503 			dip = ddi_get_next_sibling(dip);
3504 			break;
3505 		}
3506 
3507 	}
3508 
3509 	/* second pass */
3510 	while (head) {
3511 		int circ;
3512 		struct walk_elem *next = head->next;
3513 
3514 		if (do_locking)
3515 			ndi_devi_enter(head->dip, &circ);
3516 		if (walk_devs(ddi_get_child(head->dip), f, arg, do_locking) ==
3517 		    DDI_WALK_TERMINATE) {
3518 			if (do_locking)
3519 				ndi_devi_exit(head->dip, circ);
3520 			free_list(head);
3521 			return (DDI_WALK_TERMINATE);
3522 		}
3523 		if (do_locking)
3524 			ndi_devi_exit(head->dip, circ);
3525 		kmem_free(head, sizeof (*head));
3526 		head = next;
3527 	}
3528 
3529 	return (DDI_WALK_CONTINUE);
3530 }
3531 
3532 /*
3533  * This general-purpose routine traverses the tree of dev_info nodes,
3534  * starting from the given node, and calls the given function for each
3535  * node that it finds with the current node and the pointer arg (which
3536  * can point to a structure of information that the function
3537  * needs) as arguments.
3538  *
3539  * It does the walk a layer at a time, not depth-first. The given function
3540  * must return one of the following values:
3541  *	DDI_WALK_CONTINUE
3542  *	DDI_WALK_PRUNESIB
3543  *	DDI_WALK_PRUNECHILD
3544  *	DDI_WALK_TERMINATE
3545  *
3546  * N.B. Since we walk the sibling list, the caller must ensure that
3547  *	the parent of dip is held against changes, unless the parent
3548  *	is rootnode.  ndi_devi_enter() on the parent is sufficient.
3549  *
3550  *	To avoid deadlock situations, caller must not attempt to
3551  *	configure/unconfigure/remove device node in (*f)(), nor should
3552  *	it attempt to recurse on other nodes in the system. Any
3553  *	ndi_devi_enter() done by (*f)() must occur 'at-or-below' the
3554  *	node entered prior to ddi_walk_devs(). Furthermore, if (*f)()
3555  *	does any multi-threading (in framework *or* in driver) then the
3556  *	ndi_devi_enter() calls done by dependent threads must be
3557  *	'strictly-below'.
3558  *
3559  *	This is not callable from device autoconfiguration routines.
3560  *	They include, but not limited to, _init(9e), _fini(9e), probe(9e),
3561  *	attach(9e), and detach(9e).
3562  */
3563 
3564 void
3565 ddi_walk_devs(dev_info_t *dip, int (*f)(dev_info_t *, void *), void *arg)
3566 {
3567 
3568 	ASSERT(dip == NULL || ddi_get_parent(dip) == NULL ||
3569 	    DEVI_BUSY_OWNED(ddi_get_parent(dip)));
3570 
3571 	(void) walk_devs(dip, f, arg, 1);
3572 }
3573 
3574 /*
3575  * This is a general-purpose routine traverses the per-driver list
3576  * and calls the given function for each node. must return one of
3577  * the following values:
3578  *	DDI_WALK_CONTINUE
3579  *	DDI_WALK_TERMINATE
3580  *
3581  * N.B. The same restrictions from ddi_walk_devs() apply.
3582  */
3583 
3584 void
3585 e_ddi_walk_driver(char *drv, int (*f)(dev_info_t *, void *), void *arg)
3586 {
3587 	major_t major;
3588 	struct devnames *dnp;
3589 	dev_info_t *dip;
3590 
3591 	major = ddi_name_to_major(drv);
3592 	if (major == DDI_MAJOR_T_NONE)
3593 		return;
3594 
3595 	dnp = &devnamesp[major];
3596 	LOCK_DEV_OPS(&dnp->dn_lock);
3597 	dip = dnp->dn_head;
3598 	while (dip) {
3599 		ndi_hold_devi(dip);
3600 		UNLOCK_DEV_OPS(&dnp->dn_lock);
3601 		if ((*f)(dip, arg) == DDI_WALK_TERMINATE) {
3602 			ndi_rele_devi(dip);
3603 			return;
3604 		}
3605 		LOCK_DEV_OPS(&dnp->dn_lock);
3606 		ndi_rele_devi(dip);
3607 		dip = ddi_get_next(dip);
3608 	}
3609 	UNLOCK_DEV_OPS(&dnp->dn_lock);
3610 }
3611 
3612 /*
3613  * argument to i_find_devi, a devinfo node search callback function.
3614  */
3615 struct match_info {
3616 	dev_info_t	*dip;		/* result */
3617 	char		*nodename;	/* if non-null, nodename must match */
3618 	int		instance;	/* if != -1, instance must match */
3619 	int		attached;	/* if != 0, i_ddi_devi_attached() */
3620 };
3621 
3622 static int
3623 i_find_devi(dev_info_t *dip, void *arg)
3624 {
3625 	struct match_info *info = (struct match_info *)arg;
3626 
3627 	if (((info->nodename == NULL) ||
3628 	    (strcmp(ddi_node_name(dip), info->nodename) == 0)) &&
3629 	    ((info->instance == -1) ||
3630 	    (ddi_get_instance(dip) == info->instance)) &&
3631 	    ((info->attached == 0) || i_ddi_devi_attached(dip))) {
3632 		info->dip = dip;
3633 		ndi_hold_devi(dip);
3634 		return (DDI_WALK_TERMINATE);
3635 	}
3636 
3637 	return (DDI_WALK_CONTINUE);
3638 }
3639 
3640 /*
3641  * Find dip with a known node name and instance and return with it held
3642  */
3643 dev_info_t *
3644 ddi_find_devinfo(char *nodename, int instance, int attached)
3645 {
3646 	struct match_info	info;
3647 
3648 	info.nodename = nodename;
3649 	info.instance = instance;
3650 	info.attached = attached;
3651 	info.dip = NULL;
3652 
3653 	ddi_walk_devs(ddi_root_node(), i_find_devi, &info);
3654 	return (info.dip);
3655 }
3656 
3657 extern ib_boot_prop_t *iscsiboot_prop;
3658 static void
3659 i_ddi_parse_iscsi_name(char *name, char **nodename, char **addrname,
3660     char **minorname)
3661 {
3662 	char *cp, *colon;
3663 	static char nulladdrname[] = "";
3664 
3665 	/* default values */
3666 	if (nodename)
3667 		*nodename = name;
3668 	if (addrname)
3669 		*addrname = nulladdrname;
3670 	if (minorname)
3671 		*minorname = NULL;
3672 
3673 	cp = colon = name;
3674 	while (*cp != '\0') {
3675 		if (addrname && *cp == '@') {
3676 			*addrname = cp + 1;
3677 			*cp = '\0';
3678 		} else if (minorname && *cp == ':') {
3679 			*minorname = cp + 1;
3680 			colon = cp;
3681 		}
3682 		++cp;
3683 	}
3684 	if (colon != name) {
3685 		*colon = '\0';
3686 	}
3687 }
3688 
3689 /*
3690  * Parse for name, addr, and minor names. Some args may be NULL.
3691  */
3692 void
3693 i_ddi_parse_name(char *name, char **nodename, char **addrname, char **minorname)
3694 {
3695 	char *cp;
3696 	static char nulladdrname[] = "";
3697 
3698 	/* default values */
3699 	if (nodename)
3700 		*nodename = name;
3701 	if (addrname)
3702 		*addrname = nulladdrname;
3703 	if (minorname)
3704 		*minorname = NULL;
3705 
3706 	cp = name;
3707 	while (*cp != '\0') {
3708 		if (addrname && *cp == '@') {
3709 			*addrname = cp + 1;
3710 			*cp = '\0';
3711 		} else if (minorname && *cp == ':') {
3712 			*minorname = cp + 1;
3713 			*cp = '\0';
3714 		}
3715 		++cp;
3716 	}
3717 }
3718 
3719 static char *
3720 child_path_to_driver(dev_info_t *parent, char *child_name, char *unit_address)
3721 {
3722 	char *p, *drvname = NULL;
3723 	major_t maj;
3724 
3725 	/*
3726 	 * Construct the pathname and ask the implementation
3727 	 * if it can do a driver = f(pathname) for us, if not
3728 	 * we'll just default to using the node-name that
3729 	 * was given to us.  We want to do this first to
3730 	 * allow the platform to use 'generic' names for
3731 	 * legacy device drivers.
3732 	 */
3733 	p = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
3734 	(void) ddi_pathname(parent, p);
3735 	(void) strcat(p, "/");
3736 	(void) strcat(p, child_name);
3737 	if (unit_address && *unit_address) {
3738 		(void) strcat(p, "@");
3739 		(void) strcat(p, unit_address);
3740 	}
3741 
3742 	/*
3743 	 * Get the binding. If there is none, return the child_name
3744 	 * and let the caller deal with it.
3745 	 */
3746 	maj = path_to_major(p);
3747 
3748 	kmem_free(p, MAXPATHLEN);
3749 
3750 	if (maj != DDI_MAJOR_T_NONE)
3751 		drvname = ddi_major_to_name(maj);
3752 	if (drvname == NULL)
3753 		drvname = child_name;
3754 
3755 	return (drvname);
3756 }
3757 
3758 
3759 #define	PCI_EX_CLASS	"pciexclass"
3760 #define	PCI_EX		"pciex"
3761 #define	PCI_CLASS	"pciclass"
3762 #define	PCI		"pci"
3763 
3764 int
3765 ddi_is_pci_dip(dev_info_t *dip)
3766 {
3767 	char	*prop = NULL;
3768 
3769 	if (ddi_prop_lookup_string(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
3770 	    "compatible", &prop) == DDI_PROP_SUCCESS) {
3771 		ASSERT(prop);
3772 		if (strncmp(prop, PCI_EX_CLASS, sizeof (PCI_EX_CLASS) - 1)
3773 		    == 0 ||
3774 		    strncmp(prop, PCI_EX, sizeof (PCI_EX)- 1)
3775 		    == 0 ||
3776 		    strncmp(prop, PCI_CLASS, sizeof (PCI_CLASS) - 1)
3777 		    == 0 ||
3778 		    strncmp(prop, PCI, sizeof (PCI) - 1)
3779 		    == 0) {
3780 			ddi_prop_free(prop);
3781 			return (1);
3782 		}
3783 	}
3784 
3785 	if (prop != NULL) {
3786 		ddi_prop_free(prop);
3787 	}
3788 
3789 	return (0);
3790 }
3791 
3792 /*
3793  * Given the pathname of a device, fill in the dev_info_t value and/or the
3794  * dev_t value and/or the spectype, depending on which parameters are non-NULL.
3795  * If there is an error, this function returns -1.
3796  *
3797  * NOTE: If this function returns the dev_info_t structure, then it
3798  * does so with a hold on the devi. Caller should ensure that they get
3799  * decremented via ddi_release_devi() or ndi_rele_devi();
3800  *
3801  * This function can be invoked in the boot case for a pathname without
3802  * device argument (:xxxx), traditionally treated as a minor name.
3803  * In this case, we do the following
3804  * (1) search the minor node of type DDM_DEFAULT.
3805  * (2) if no DDM_DEFAULT minor exists, then the first non-alias minor is chosen.
3806  * (3) if neither exists, a dev_t is faked with minor number = instance.
3807  * As of S9 FCS, no instance of #1 exists. #2 is used by several platforms
3808  * to default the boot partition to :a possibly by other OBP definitions.
3809  * #3 is used for booting off network interfaces, most SPARC network
3810  * drivers support Style-2 only, so only DDM_ALIAS minor exists.
3811  *
3812  * It is possible for OBP to present device args at the end of the path as
3813  * well as in the middle. For example, with IB the following strings are
3814  * valid boot paths.
3815  *	a /pci@8,700000/ib@1,2:port=1,pkey=ff,dhcp,...
3816  *	b /pci@8,700000/ib@1,1:port=1/ioc@xxxxxx,yyyyyyy:dhcp
3817  * Case (a), we first look for minor node "port=1,pkey...".
3818  * Failing that, we will pass "port=1,pkey..." to the bus_config
3819  * entry point of ib (HCA) driver.
3820  * Case (b), configure ib@1,1 as usual. Then invoke ib's bus_config
3821  * with argument "ioc@xxxxxxx,yyyyyyy:port=1". After configuring
3822  * the ioc, look for minor node dhcp. If not found, pass ":dhcp"
3823  * to ioc's bus_config entry point.
3824  */
3825 int
3826 resolve_pathname(char *pathname,
3827 	dev_info_t **dipp, dev_t *devtp, int *spectypep)
3828 {
3829 	int			error;
3830 	dev_info_t		*parent, *child;
3831 	struct pathname		pn;
3832 	char			*component, *config_name;
3833 	char			*minorname = NULL;
3834 	char			*prev_minor = NULL;
3835 	dev_t			devt = NODEV;
3836 	int			spectype;
3837 	struct ddi_minor_data	*dmn;
3838 	int			circ;
3839 
3840 	if (*pathname != '/')
3841 		return (EINVAL);
3842 	parent = ddi_root_node();	/* Begin at the top of the tree */
3843 
3844 	if (error = pn_get(pathname, UIO_SYSSPACE, &pn))
3845 		return (error);
3846 	pn_skipslash(&pn);
3847 
3848 	ASSERT(i_ddi_devi_attached(parent));
3849 	ndi_hold_devi(parent);
3850 
3851 	component = kmem_alloc(MAXNAMELEN, KM_SLEEP);
3852 	config_name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
3853 
3854 	while (pn_pathleft(&pn)) {
3855 		/* remember prev minor (:xxx) in the middle of path */
3856 		if (minorname)
3857 			prev_minor = i_ddi_strdup(minorname, KM_SLEEP);
3858 
3859 		/* Get component and chop off minorname */
3860 		(void) pn_getcomponent(&pn, component);
3861 		if ((iscsiboot_prop != NULL) &&
3862 		    (strcmp((DEVI(parent)->devi_node_name), "iscsi") == 0)) {
3863 			i_ddi_parse_iscsi_name(component, NULL, NULL,
3864 			    &minorname);
3865 		} else {
3866 			i_ddi_parse_name(component, NULL, NULL, &minorname);
3867 		}
3868 		if (prev_minor == NULL) {
3869 			(void) snprintf(config_name, MAXNAMELEN, "%s",
3870 			    component);
3871 		} else {
3872 			(void) snprintf(config_name, MAXNAMELEN, "%s:%s",
3873 			    component, prev_minor);
3874 			kmem_free(prev_minor, strlen(prev_minor) + 1);
3875 			prev_minor = NULL;
3876 		}
3877 
3878 		/*
3879 		 * Find and configure the child
3880 		 */
3881 		if (ndi_devi_config_one(parent, config_name, &child,
3882 		    NDI_PROMNAME | NDI_NO_EVENT) != NDI_SUCCESS) {
3883 			ndi_rele_devi(parent);
3884 			pn_free(&pn);
3885 			kmem_free(component, MAXNAMELEN);
3886 			kmem_free(config_name, MAXNAMELEN);
3887 			return (-1);
3888 		}
3889 
3890 		ASSERT(i_ddi_devi_attached(child));
3891 		ndi_rele_devi(parent);
3892 		parent = child;
3893 		pn_skipslash(&pn);
3894 	}
3895 
3896 	/*
3897 	 * First look for a minor node matching minorname.
3898 	 * Failing that, try to pass minorname to bus_config().
3899 	 */
3900 	if (minorname && i_ddi_minorname_to_devtspectype(parent,
3901 	    minorname, &devt, &spectype) == DDI_FAILURE) {
3902 		(void) snprintf(config_name, MAXNAMELEN, "%s", minorname);
3903 		if (ndi_devi_config_obp_args(parent,
3904 		    config_name, &child, 0) != NDI_SUCCESS) {
3905 			ndi_rele_devi(parent);
3906 			pn_free(&pn);
3907 			kmem_free(component, MAXNAMELEN);
3908 			kmem_free(config_name, MAXNAMELEN);
3909 			NDI_CONFIG_DEBUG((CE_NOTE,
3910 			    "%s: minor node not found\n", pathname));
3911 			return (-1);
3912 		}
3913 		minorname = NULL;	/* look for default minor */
3914 		ASSERT(i_ddi_devi_attached(child));
3915 		ndi_rele_devi(parent);
3916 		parent = child;
3917 	}
3918 
3919 	if (devtp || spectypep) {
3920 		if (minorname == NULL) {
3921 			/*
3922 			 * Search for a default entry with an active
3923 			 * ndi_devi_enter to protect the devi_minor list.
3924 			 */
3925 			ndi_devi_enter(parent, &circ);
3926 			for (dmn = DEVI(parent)->devi_minor; dmn;
3927 			    dmn = dmn->next) {
3928 				if (dmn->type == DDM_DEFAULT) {
3929 					devt = dmn->ddm_dev;
3930 					spectype = dmn->ddm_spec_type;
3931 					break;
3932 				}
3933 			}
3934 
3935 			if (devt == NODEV) {
3936 				/*
3937 				 * No default minor node, try the first one;
3938 				 * else, assume 1-1 instance-minor mapping
3939 				 */
3940 				dmn = DEVI(parent)->devi_minor;
3941 				if (dmn && ((dmn->type == DDM_MINOR) ||
3942 				    (dmn->type == DDM_INTERNAL_PATH))) {
3943 					devt = dmn->ddm_dev;
3944 					spectype = dmn->ddm_spec_type;
3945 				} else {
3946 					devt = makedevice(
3947 					    DEVI(parent)->devi_major,
3948 					    ddi_get_instance(parent));
3949 					spectype = S_IFCHR;
3950 				}
3951 			}
3952 			ndi_devi_exit(parent, circ);
3953 		}
3954 		if (devtp)
3955 			*devtp = devt;
3956 		if (spectypep)
3957 			*spectypep = spectype;
3958 	}
3959 
3960 	pn_free(&pn);
3961 	kmem_free(component, MAXNAMELEN);
3962 	kmem_free(config_name, MAXNAMELEN);
3963 
3964 	/*
3965 	 * If there is no error, return the appropriate parameters
3966 	 */
3967 	if (dipp != NULL)
3968 		*dipp = parent;
3969 	else {
3970 		/*
3971 		 * We should really keep the ref count to keep the node from
3972 		 * detaching but ddi_pathname_to_dev_t() specifies a NULL dipp,
3973 		 * so we have no way of passing back the held dip.  Not holding
3974 		 * the dip allows detaches to occur - which can cause problems
3975 		 * for subsystems which call ddi_pathname_to_dev_t (console).
3976 		 *
3977 		 * Instead of holding the dip, we place a ddi-no-autodetach
3978 		 * property on the node to prevent auto detaching.
3979 		 *
3980 		 * The right fix is to remove ddi_pathname_to_dev_t and replace
3981 		 * it, and all references, with a call that specifies a dipp.
3982 		 * In addition, the callers of this new interfaces would then
3983 		 * need to call ndi_rele_devi when the reference is complete.
3984 		 *
3985 		 */
3986 		(void) ddi_prop_update_int(DDI_DEV_T_NONE, parent,
3987 		    DDI_NO_AUTODETACH, 1);
3988 		ndi_rele_devi(parent);
3989 	}
3990 
3991 	return (0);
3992 }
3993 
3994 /*
3995  * Given the pathname of a device, return the dev_t of the corresponding
3996  * device.  Returns NODEV on failure.
3997  *
3998  * Note that this call sets the DDI_NO_AUTODETACH property on the devinfo node.
3999  */
4000 dev_t
4001 ddi_pathname_to_dev_t(char *pathname)
4002 {
4003 	dev_t devt;
4004 	int error;
4005 
4006 	error = resolve_pathname(pathname, NULL, &devt, NULL);
4007 
4008 	return (error ? NODEV : devt);
4009 }
4010 
4011 /*
4012  * Translate a prom pathname to kernel devfs pathname.
4013  * Caller is assumed to allocate devfspath memory of
4014  * size at least MAXPATHLEN
4015  *
4016  * The prom pathname may not include minor name, but
4017  * devfs pathname has a minor name portion.
4018  */
4019 int
4020 i_ddi_prompath_to_devfspath(char *prompath, char *devfspath)
4021 {
4022 	dev_t		devt = (dev_t)NODEV;
4023 	dev_info_t	*dip = NULL;
4024 	char		*minor_name = NULL;
4025 	int		spectype;
4026 	int		error;
4027 	int		circ;
4028 
4029 	error = resolve_pathname(prompath, &dip, &devt, &spectype);
4030 	if (error)
4031 		return (DDI_FAILURE);
4032 	ASSERT(dip && devt != NODEV);
4033 
4034 	/*
4035 	 * Get in-kernel devfs pathname
4036 	 */
4037 	(void) ddi_pathname(dip, devfspath);
4038 
4039 	ndi_devi_enter(dip, &circ);
4040 	minor_name = i_ddi_devtspectype_to_minorname(dip, devt, spectype);
4041 	if (minor_name) {
4042 		(void) strcat(devfspath, ":");
4043 		(void) strcat(devfspath, minor_name);
4044 	} else {
4045 		/*
4046 		 * If minor_name is NULL, we have an alias minor node.
4047 		 * So manufacture a path to the corresponding clone minor.
4048 		 */
4049 		(void) snprintf(devfspath, MAXPATHLEN, "%s:%s",
4050 		    CLONE_PATH, ddi_driver_name(dip));
4051 	}
4052 	ndi_devi_exit(dip, circ);
4053 
4054 	/* release hold from resolve_pathname() */
4055 	ndi_rele_devi(dip);
4056 	return (0);
4057 }
4058 
4059 /*
4060  * This function is intended to identify drivers that must quiesce for fast
4061  * reboot to succeed.  It does not claim to have more knowledge about the device
4062  * than its driver.  If a driver has implemented quiesce(), it will be invoked;
4063  * if a so identified driver does not manage any device that needs to be
4064  * quiesced, it must explicitly set its devo_quiesce dev_op to
4065  * ddi_quiesce_not_needed.
4066  */
4067 static int skip_pseudo = 1;	/* Skip pseudo devices */
4068 static int skip_non_hw = 1;	/* Skip devices with no hardware property */
4069 static int
4070 should_implement_quiesce(dev_info_t *dip)
4071 {
4072 	struct dev_info *devi = DEVI(dip);
4073 	dev_info_t *pdip;
4074 
4075 	/*
4076 	 * If dip is pseudo and skip_pseudo is set, driver doesn't have to
4077 	 * implement quiesce().
4078 	 */
4079 	if (skip_pseudo &&
4080 	    strncmp(ddi_binding_name(dip), "pseudo", sizeof ("pseudo")) == 0)
4081 		return (0);
4082 
4083 	/*
4084 	 * If parent dip is pseudo and skip_pseudo is set, driver doesn't have
4085 	 * to implement quiesce().
4086 	 */
4087 	if (skip_pseudo && (pdip = ddi_get_parent(dip)) != NULL &&
4088 	    strncmp(ddi_binding_name(pdip), "pseudo", sizeof ("pseudo")) == 0)
4089 		return (0);
4090 
4091 	/*
4092 	 * If not attached, driver doesn't have to implement quiesce().
4093 	 */
4094 	if (!i_ddi_devi_attached(dip))
4095 		return (0);
4096 
4097 	/*
4098 	 * If dip has no hardware property and skip_non_hw is set,
4099 	 * driver doesn't have to implement quiesce().
4100 	 */
4101 	if (skip_non_hw && devi->devi_hw_prop_ptr == NULL)
4102 		return (0);
4103 
4104 	return (1);
4105 }
4106 
4107 static int
4108 driver_has_quiesce(struct dev_ops *ops)
4109 {
4110 	if ((ops->devo_rev >= 4) && (ops->devo_quiesce != nodev) &&
4111 	    (ops->devo_quiesce != NULL) && (ops->devo_quiesce != nulldev) &&
4112 	    (ops->devo_quiesce != ddi_quiesce_not_supported))
4113 		return (1);
4114 	else
4115 		return (0);
4116 }
4117 
4118 /*
4119  * Check to see if a driver has implemented the quiesce() DDI function.
4120  */
4121 int
4122 check_driver_quiesce(dev_info_t *dip, void *arg)
4123 {
4124 	struct dev_ops *ops;
4125 
4126 	if (!should_implement_quiesce(dip))
4127 		return (DDI_WALK_CONTINUE);
4128 
4129 	if ((ops = ddi_get_driver(dip)) == NULL)
4130 		return (DDI_WALK_CONTINUE);
4131 
4132 	if (driver_has_quiesce(ops)) {
4133 		if ((quiesce_debug & 0x2) == 0x2) {
4134 			if (ops->devo_quiesce == ddi_quiesce_not_needed)
4135 				cmn_err(CE_CONT, "%s does not need to be "
4136 				    "quiesced", ddi_driver_name(dip));
4137 			else
4138 				cmn_err(CE_CONT, "%s has quiesce routine",
4139 				    ddi_driver_name(dip));
4140 		}
4141 	} else {
4142 		if (arg != NULL)
4143 			*((int *)arg) = -1;
4144 		cmn_err(CE_WARN, "%s has no quiesce()", ddi_driver_name(dip));
4145 	}
4146 
4147 	return (DDI_WALK_CONTINUE);
4148 }
4149 
4150 /*
4151  * Quiesce device.
4152  */
4153 static void
4154 quiesce_one_device(dev_info_t *dip, void *arg)
4155 {
4156 	struct dev_ops *ops;
4157 	int should_quiesce = 0;
4158 
4159 	/*
4160 	 * If the device is not attached it doesn't need to be quiesced.
4161 	 */
4162 	if (!i_ddi_devi_attached(dip))
4163 		return;
4164 
4165 	if ((ops = ddi_get_driver(dip)) == NULL)
4166 		return;
4167 
4168 	should_quiesce = should_implement_quiesce(dip);
4169 
4170 	/*
4171 	 * If there's an implementation of quiesce(), always call it even if
4172 	 * some of the drivers don't have quiesce() or quiesce() have failed
4173 	 * so we can do force fast reboot.  The implementation of quiesce()
4174 	 * should not negatively affect a regular reboot.
4175 	 */
4176 	if (driver_has_quiesce(ops)) {
4177 		int rc = DDI_SUCCESS;
4178 
4179 		if (ops->devo_quiesce == ddi_quiesce_not_needed)
4180 			return;
4181 
4182 		rc = devi_quiesce(dip);
4183 
4184 		/* quiesce() should never fail */
4185 		ASSERT(rc == DDI_SUCCESS);
4186 
4187 		if (rc != DDI_SUCCESS && should_quiesce) {
4188 
4189 			if (arg != NULL)
4190 				*((int *)arg) = -1;
4191 		}
4192 	} else if (should_quiesce && arg != NULL) {
4193 		*((int *)arg) = -1;
4194 	}
4195 }
4196 
4197 /*
4198  * Traverse the dev info tree in a breadth-first manner so that we quiesce
4199  * children first.  All subtrees under the parent of dip will be quiesced.
4200  */
4201 void
4202 quiesce_devices(dev_info_t *dip, void *arg)
4203 {
4204 	/*
4205 	 * if we're reached here, the device tree better not be changing.
4206 	 * so either devinfo_freeze better be set or we better be panicing.
4207 	 */
4208 	ASSERT(devinfo_freeze || panicstr);
4209 
4210 	for (; dip != NULL; dip = ddi_get_next_sibling(dip)) {
4211 		quiesce_devices(ddi_get_child(dip), arg);
4212 
4213 		quiesce_one_device(dip, arg);
4214 	}
4215 }
4216 
4217 /*
4218  * Reset all the pure leaf drivers on the system at halt time
4219  */
4220 static int
4221 reset_leaf_device(dev_info_t *dip, void *arg)
4222 {
4223 	_NOTE(ARGUNUSED(arg))
4224 	struct dev_ops *ops;
4225 
4226 	/* if the device doesn't need to be reset then there's nothing to do */
4227 	if (!DEVI_NEED_RESET(dip))
4228 		return (DDI_WALK_CONTINUE);
4229 
4230 	/*
4231 	 * if the device isn't a char/block device or doesn't have a
4232 	 * reset entry point then there's nothing to do.
4233 	 */
4234 	ops = ddi_get_driver(dip);
4235 	if ((ops == NULL) || (ops->devo_cb_ops == NULL) ||
4236 	    (ops->devo_reset == nodev) || (ops->devo_reset == nulldev) ||
4237 	    (ops->devo_reset == NULL))
4238 		return (DDI_WALK_CONTINUE);
4239 
4240 	if (DEVI_IS_ATTACHING(dip) || DEVI_IS_DETACHING(dip)) {
4241 		static char path[MAXPATHLEN];
4242 
4243 		/*
4244 		 * bad news, this device has blocked in it's attach or
4245 		 * detach routine, which means it not safe to call it's
4246 		 * devo_reset() entry point.
4247 		 */
4248 		cmn_err(CE_WARN, "unable to reset device: %s",
4249 		    ddi_pathname(dip, path));
4250 		return (DDI_WALK_CONTINUE);
4251 	}
4252 
4253 	NDI_CONFIG_DEBUG((CE_NOTE, "resetting %s%d\n",
4254 	    ddi_driver_name(dip), ddi_get_instance(dip)));
4255 
4256 	(void) devi_reset(dip, DDI_RESET_FORCE);
4257 	return (DDI_WALK_CONTINUE);
4258 }
4259 
4260 void
4261 reset_leaves(void)
4262 {
4263 	/*
4264 	 * if we're reached here, the device tree better not be changing.
4265 	 * so either devinfo_freeze better be set or we better be panicing.
4266 	 */
4267 	ASSERT(devinfo_freeze || panicstr);
4268 
4269 	(void) walk_devs(top_devinfo, reset_leaf_device, NULL, 0);
4270 }
4271 
4272 
4273 /*
4274  * devtree_freeze() must be called before quiesce_devices() and reset_leaves()
4275  * during a normal system shutdown.  It attempts to ensure that there are no
4276  * outstanding attach or detach operations in progress when quiesce_devices() or
4277  * reset_leaves()is invoked.  It must be called before the system becomes
4278  * single-threaded because device attach and detach are multi-threaded
4279  * operations.	(note that during system shutdown the system doesn't actually
4280  * become single-thread since other threads still exist, but the shutdown thread
4281  * will disable preemption for itself, raise it's pil, and stop all the other
4282  * cpus in the system there by effectively making the system single-threaded.)
4283  */
4284 void
4285 devtree_freeze(void)
4286 {
4287 	int delayed = 0;
4288 
4289 	/* if we're panicing then the device tree isn't going to be changing */
4290 	if (panicstr)
4291 		return;
4292 
4293 	/* stop all dev_info state changes in the device tree */
4294 	devinfo_freeze = gethrtime();
4295 
4296 	/*
4297 	 * if we're not panicing and there are on-going attach or detach
4298 	 * operations, wait for up to 3 seconds for them to finish.  This
4299 	 * is a randomly chosen interval but this should be ok because:
4300 	 * - 3 seconds is very small relative to the deadman timer.
4301 	 * - normal attach and detach operations should be very quick.
4302 	 * - attach and detach operations are fairly rare.
4303 	 */
4304 	while (!panicstr && atomic_add_long_nv(&devinfo_attach_detach, 0) &&
4305 	    (delayed < 3)) {
4306 		delayed += 1;
4307 
4308 		/* do a sleeping wait for one second */
4309 		ASSERT(!servicing_interrupt());
4310 		delay(drv_usectohz(MICROSEC));
4311 	}
4312 }
4313 
4314 static int
4315 bind_dip(dev_info_t *dip, void *arg)
4316 {
4317 	_NOTE(ARGUNUSED(arg))
4318 	char	*path;
4319 	major_t	major, pmajor;
4320 
4321 	/*
4322 	 * If the node is currently bound to the wrong driver, try to unbind
4323 	 * so that we can rebind to the correct driver.
4324 	 */
4325 	if (i_ddi_node_state(dip) >= DS_BOUND) {
4326 		major = ddi_compatible_driver_major(dip, NULL);
4327 		if ((DEVI(dip)->devi_major == major) &&
4328 		    (i_ddi_node_state(dip) >= DS_INITIALIZED)) {
4329 			/*
4330 			 * Check for a path-oriented driver alias that
4331 			 * takes precedence over current driver binding.
4332 			 */
4333 			path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4334 			(void) ddi_pathname(dip, path);
4335 			pmajor = ddi_name_to_major(path);
4336 			if (driver_active(pmajor))
4337 				major = pmajor;
4338 			kmem_free(path, MAXPATHLEN);
4339 		}
4340 
4341 		/* attempt unbind if current driver is incorrect */
4342 		if (driver_active(major) &&
4343 		    (major != DEVI(dip)->devi_major))
4344 			(void) ndi_devi_unbind_driver(dip);
4345 	}
4346 
4347 	/* If unbound, try to bind to a driver */
4348 	if (i_ddi_node_state(dip) < DS_BOUND)
4349 		(void) ndi_devi_bind_driver(dip, 0);
4350 
4351 	return (DDI_WALK_CONTINUE);
4352 }
4353 
4354 void
4355 i_ddi_bind_devs(void)
4356 {
4357 	/* flush devfs so that ndi_devi_unbind_driver will work when possible */
4358 	(void) devfs_clean(top_devinfo, NULL, 0);
4359 
4360 	ddi_walk_devs(top_devinfo, bind_dip, (void *)NULL);
4361 }
4362 
4363 /* callback data for unbind_children_by_alias() */
4364 typedef struct unbind_data {
4365 	major_t	drv_major;
4366 	char	*drv_alias;
4367 	int	ndevs_bound;
4368 	int	unbind_errors;
4369 } unbind_data_t;
4370 
4371 /*
4372  * A utility function provided for testing and support convenience
4373  * Called for each device during an upgrade_drv -d bound to the alias
4374  * that cannot be unbound due to device in use.
4375  */
4376 static void
4377 unbind_alias_dev_in_use(dev_info_t *dip, char *alias)
4378 {
4379 	if (moddebug & MODDEBUG_BINDING) {
4380 		cmn_err(CE_CONT, "%s%d: state %d: bound to %s\n",
4381 		    ddi_driver_name(dip), ddi_get_instance(dip),
4382 		    i_ddi_node_state(dip), alias);
4383 	}
4384 }
4385 
4386 /*
4387  * walkdevs callback for unbind devices bound to specific driver
4388  * and alias.  Invoked within the context of update_drv -d <alias>.
4389  */
4390 static int
4391 unbind_children_by_alias(dev_info_t *dip, void *arg)
4392 {
4393 	int		circ;
4394 	dev_info_t	*cdip;
4395 	dev_info_t	*next;
4396 	unbind_data_t	*ub = (unbind_data_t *)(uintptr_t)arg;
4397 	int		rv;
4398 
4399 	/*
4400 	 * We are called from update_drv to try to unbind a specific
4401 	 * set of aliases for a driver.  Unbind what persistent nodes
4402 	 * we can, and return the number of nodes which cannot be unbound.
4403 	 * If not all nodes can be unbound, update_drv leaves the
4404 	 * state of the driver binding files unchanged, except in
4405 	 * the case of -f.
4406 	 */
4407 	ndi_devi_enter(dip, &circ);
4408 	for (cdip = ddi_get_child(dip); cdip; cdip = next) {
4409 		next = ddi_get_next_sibling(cdip);
4410 		if ((ddi_driver_major(cdip) != ub->drv_major) ||
4411 		    (strcmp(DEVI(cdip)->devi_node_name, ub->drv_alias) != 0))
4412 			continue;
4413 		if (i_ddi_node_state(cdip) >= DS_BOUND) {
4414 			rv = ndi_devi_unbind_driver(cdip);
4415 			if (rv != DDI_SUCCESS ||
4416 			    (i_ddi_node_state(cdip) >= DS_BOUND)) {
4417 				unbind_alias_dev_in_use(cdip, ub->drv_alias);
4418 				ub->ndevs_bound++;
4419 				continue;
4420 			}
4421 			if (ndi_dev_is_persistent_node(cdip) == 0)
4422 				(void) ddi_remove_child(cdip, 0);
4423 		}
4424 	}
4425 	ndi_devi_exit(dip, circ);
4426 
4427 	return (DDI_WALK_CONTINUE);
4428 }
4429 
4430 /*
4431  * Unbind devices by driver & alias
4432  * Context: update_drv [-f] -d -i <alias> <driver>
4433  */
4434 int
4435 i_ddi_unbind_devs_by_alias(major_t major, char *alias)
4436 {
4437 	unbind_data_t	*ub;
4438 	int		rv;
4439 
4440 	ub = kmem_zalloc(sizeof (*ub), KM_SLEEP);
4441 	ub->drv_major = major;
4442 	ub->drv_alias = alias;
4443 	ub->ndevs_bound = 0;
4444 	ub->unbind_errors = 0;
4445 
4446 	/* flush devfs so that ndi_devi_unbind_driver will work when possible */
4447 	(void) devfs_clean(top_devinfo, NULL, 0);
4448 	ddi_walk_devs(top_devinfo, unbind_children_by_alias,
4449 	    (void *)(uintptr_t)ub);
4450 
4451 	/* return the number of devices remaining bound to the alias */
4452 	rv = ub->ndevs_bound + ub->unbind_errors;
4453 	kmem_free(ub, sizeof (*ub));
4454 	return (rv);
4455 }
4456 
4457 /*
4458  * walkdevs callback for unbind devices by driver
4459  */
4460 static int
4461 unbind_children_by_driver(dev_info_t *dip, void *arg)
4462 {
4463 	int		circ;
4464 	dev_info_t	*cdip;
4465 	dev_info_t	*next;
4466 	major_t		major = (major_t)(uintptr_t)arg;
4467 	int		rv;
4468 
4469 	/*
4470 	 * We are called either from rem_drv or update_drv when reloading
4471 	 * a driver.conf file. In either case, we unbind persistent nodes
4472 	 * and destroy .conf nodes. In the case of rem_drv, this will be
4473 	 * the final state. In the case of update_drv,	i_ddi_bind_devs()
4474 	 * may be invoked later to re-enumerate (new) driver.conf rebind
4475 	 * persistent nodes.
4476 	 */
4477 	ndi_devi_enter(dip, &circ);
4478 	for (cdip = ddi_get_child(dip); cdip; cdip = next) {
4479 		next = ddi_get_next_sibling(cdip);
4480 		if (ddi_driver_major(cdip) != major)
4481 			continue;
4482 		if (i_ddi_node_state(cdip) >= DS_BOUND) {
4483 			rv = ndi_devi_unbind_driver(cdip);
4484 			if (rv == DDI_FAILURE ||
4485 			    (i_ddi_node_state(cdip) >= DS_BOUND))
4486 				continue;
4487 			if (ndi_dev_is_persistent_node(cdip) == 0)
4488 				(void) ddi_remove_child(cdip, 0);
4489 		}
4490 	}
4491 	ndi_devi_exit(dip, circ);
4492 
4493 	return (DDI_WALK_CONTINUE);
4494 }
4495 
4496 /*
4497  * Unbind devices by driver
4498  * Context: rem_drv or unload driver.conf
4499  */
4500 void
4501 i_ddi_unbind_devs(major_t major)
4502 {
4503 	/* flush devfs so that ndi_devi_unbind_driver will work when possible */
4504 	(void) devfs_clean(top_devinfo, NULL, 0);
4505 	ddi_walk_devs(top_devinfo, unbind_children_by_driver,
4506 	    (void *)(uintptr_t)major);
4507 }
4508 
4509 /*
4510  * I/O Hotplug control
4511  */
4512 
4513 /*
4514  * create and attach a dev_info node from a .conf file spec
4515  */
4516 static void
4517 init_spec_child(dev_info_t *pdip, struct hwc_spec *specp, uint_t flags)
4518 {
4519 	_NOTE(ARGUNUSED(flags))
4520 	dev_info_t *dip;
4521 	char *node_name;
4522 
4523 	if (((node_name = specp->hwc_devi_name) == NULL) ||
4524 	    (ddi_name_to_major(node_name) == DDI_MAJOR_T_NONE)) {
4525 		char *tmp = node_name;
4526 		if (tmp == NULL)
4527 			tmp = "<none>";
4528 		cmn_err(CE_CONT,
4529 		    "init_spec_child: parent=%s, bad spec (%s)\n",
4530 		    ddi_node_name(pdip), tmp);
4531 		return;
4532 	}
4533 
4534 	dip = i_ddi_alloc_node(pdip, node_name, (pnode_t)DEVI_PSEUDO_NODEID,
4535 	    -1, specp->hwc_devi_sys_prop_ptr, KM_SLEEP);
4536 
4537 	if (dip == NULL)
4538 		return;
4539 
4540 	if (ddi_initchild(pdip, dip) != DDI_SUCCESS)
4541 		(void) ddi_remove_child(dip, 0);
4542 }
4543 
4544 /*
4545  * Lookup hwc specs from hash tables and make children from the spec
4546  * Because some .conf children are "merge" nodes, we also initialize
4547  * .conf children to merge properties onto hardware nodes.
4548  *
4549  * The pdip must be held busy.
4550  */
4551 int
4552 i_ndi_make_spec_children(dev_info_t *pdip, uint_t flags)
4553 {
4554 	extern struct hwc_spec *hwc_get_child_spec(dev_info_t *, major_t);
4555 	int			circ;
4556 	struct hwc_spec		*list, *spec;
4557 
4558 	ndi_devi_enter(pdip, &circ);
4559 	if (DEVI(pdip)->devi_flags & DEVI_MADE_CHILDREN) {
4560 		ndi_devi_exit(pdip, circ);
4561 		return (DDI_SUCCESS);
4562 	}
4563 
4564 	list = hwc_get_child_spec(pdip, DDI_MAJOR_T_NONE);
4565 	for (spec = list; spec != NULL; spec = spec->hwc_next) {
4566 		init_spec_child(pdip, spec, flags);
4567 	}
4568 	hwc_free_spec_list(list);
4569 
4570 	mutex_enter(&DEVI(pdip)->devi_lock);
4571 	DEVI(pdip)->devi_flags |= DEVI_MADE_CHILDREN;
4572 	mutex_exit(&DEVI(pdip)->devi_lock);
4573 	ndi_devi_exit(pdip, circ);
4574 	return (DDI_SUCCESS);
4575 }
4576 
4577 /*
4578  * Run initchild on all child nodes such that instance assignment
4579  * for multiport network cards are contiguous.
4580  *
4581  * The pdip must be held busy.
4582  */
4583 static void
4584 i_ndi_init_hw_children(dev_info_t *pdip, uint_t flags)
4585 {
4586 	dev_info_t *dip;
4587 
4588 	ASSERT(DEVI(pdip)->devi_flags & DEVI_MADE_CHILDREN);
4589 
4590 	/* contiguous instance assignment */
4591 	e_ddi_enter_instance();
4592 	dip = ddi_get_child(pdip);
4593 	while (dip) {
4594 		if (ndi_dev_is_persistent_node(dip))
4595 			(void) i_ndi_config_node(dip, DS_INITIALIZED, flags);
4596 		dip = ddi_get_next_sibling(dip);
4597 	}
4598 	e_ddi_exit_instance();
4599 }
4600 
4601 /*
4602  * report device status
4603  */
4604 static void
4605 i_ndi_devi_report_status_change(dev_info_t *dip, char *path)
4606 {
4607 	char *status;
4608 
4609 	if (!DEVI_NEED_REPORT(dip) ||
4610 	    (i_ddi_node_state(dip) < DS_INITIALIZED) ||
4611 	    ndi_dev_is_hidden_node(dip)) {
4612 		return;
4613 	}
4614 
4615 	/* Invalidate the devinfo snapshot cache */
4616 	i_ddi_di_cache_invalidate();
4617 
4618 	if (DEVI_IS_DEVICE_REMOVED(dip)) {
4619 		status = "removed";
4620 	} else if (DEVI_IS_DEVICE_OFFLINE(dip)) {
4621 		status = "offline";
4622 	} else if (DEVI_IS_DEVICE_DOWN(dip)) {
4623 		status = "down";
4624 	} else if (DEVI_IS_BUS_QUIESCED(dip)) {
4625 		status = "quiesced";
4626 	} else if (DEVI_IS_BUS_DOWN(dip)) {
4627 		status = "down";
4628 	} else if (i_ddi_devi_attached(dip)) {
4629 		status = "online";
4630 	} else {
4631 		status = "unknown";
4632 	}
4633 
4634 	if (path == NULL) {
4635 		path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4636 		cmn_err(CE_CONT, "?%s (%s%d) %s\n",
4637 		    ddi_pathname(dip, path), ddi_driver_name(dip),
4638 		    ddi_get_instance(dip), status);
4639 		kmem_free(path, MAXPATHLEN);
4640 	} else {
4641 		cmn_err(CE_CONT, "?%s (%s%d) %s\n",
4642 		    path, ddi_driver_name(dip),
4643 		    ddi_get_instance(dip), status);
4644 	}
4645 
4646 	mutex_enter(&(DEVI(dip)->devi_lock));
4647 	DEVI_REPORT_DONE(dip);
4648 	mutex_exit(&(DEVI(dip)->devi_lock));
4649 }
4650 
4651 /*
4652  * log a notification that a dev_info node has been configured.
4653  */
4654 static int
4655 i_log_devfs_add_devinfo(dev_info_t *dip, uint_t flags)
4656 {
4657 	int			se_err;
4658 	char			*pathname;
4659 	sysevent_t		*ev;
4660 	sysevent_id_t		eid;
4661 	sysevent_value_t	se_val;
4662 	sysevent_attr_list_t	*ev_attr_list = NULL;
4663 	char			*class_name;
4664 	int			no_transport = 0;
4665 
4666 	ASSERT(dip && ddi_get_parent(dip) &&
4667 	    DEVI_BUSY_OWNED(ddi_get_parent(dip)));
4668 
4669 	/* do not generate ESC_DEVFS_DEVI_ADD event during boot */
4670 	if (!i_ddi_io_initialized())
4671 		return (DDI_SUCCESS);
4672 
4673 	/* Invalidate the devinfo snapshot cache */
4674 	i_ddi_di_cache_invalidate();
4675 
4676 	ev = sysevent_alloc(EC_DEVFS, ESC_DEVFS_DEVI_ADD, EP_DDI, SE_SLEEP);
4677 
4678 	pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4679 
4680 	(void) ddi_pathname(dip, pathname);
4681 	ASSERT(strlen(pathname));
4682 
4683 	se_val.value_type = SE_DATA_TYPE_STRING;
4684 	se_val.value.sv_string = pathname;
4685 	if (sysevent_add_attr(&ev_attr_list, DEVFS_PATHNAME,
4686 	    &se_val, SE_SLEEP) != 0) {
4687 		goto fail;
4688 	}
4689 
4690 	/* add the device class attribute */
4691 	if ((class_name = i_ddi_devi_class(dip)) != NULL) {
4692 		se_val.value_type = SE_DATA_TYPE_STRING;
4693 		se_val.value.sv_string = class_name;
4694 
4695 		if (sysevent_add_attr(&ev_attr_list,
4696 		    DEVFS_DEVI_CLASS, &se_val, SE_SLEEP) != 0) {
4697 			sysevent_free_attr(ev_attr_list);
4698 			goto fail;
4699 		}
4700 	}
4701 
4702 	/*
4703 	 * must log a branch event too unless NDI_BRANCH_EVENT_OP is set,
4704 	 * in which case the branch event will be logged by the caller
4705 	 * after the entire branch has been configured.
4706 	 */
4707 	if ((flags & NDI_BRANCH_EVENT_OP) == 0) {
4708 		/*
4709 		 * Instead of logging a separate branch event just add
4710 		 * DEVFS_BRANCH_EVENT attribute. It indicates devfsadmd to
4711 		 * generate a EC_DEV_BRANCH event.
4712 		 */
4713 		se_val.value_type = SE_DATA_TYPE_INT32;
4714 		se_val.value.sv_int32 = 1;
4715 		if (sysevent_add_attr(&ev_attr_list,
4716 		    DEVFS_BRANCH_EVENT, &se_val, SE_SLEEP) != 0) {
4717 			sysevent_free_attr(ev_attr_list);
4718 			goto fail;
4719 		}
4720 	}
4721 
4722 	if (sysevent_attach_attributes(ev, ev_attr_list) != 0) {
4723 		sysevent_free_attr(ev_attr_list);
4724 		goto fail;
4725 	}
4726 
4727 	if ((se_err = log_sysevent(ev, SE_SLEEP, &eid)) != 0) {
4728 		if (se_err == SE_NO_TRANSPORT)
4729 			no_transport = 1;
4730 		goto fail;
4731 	}
4732 
4733 	sysevent_free(ev);
4734 	kmem_free(pathname, MAXPATHLEN);
4735 
4736 	return (DDI_SUCCESS);
4737 
4738 fail:
4739 	cmn_err(CE_WARN, "failed to log ESC_DEVFS_DEVI_ADD event for %s%s",
4740 	    pathname, (no_transport) ? " (syseventd not responding)" : "");
4741 
4742 	cmn_err(CE_WARN, "/dev may not be current for driver %s. "
4743 	    "Run devfsadm -i %s",
4744 	    ddi_driver_name(dip), ddi_driver_name(dip));
4745 
4746 	sysevent_free(ev);
4747 	kmem_free(pathname, MAXPATHLEN);
4748 	return (DDI_SUCCESS);
4749 }
4750 
4751 /*
4752  * log a notification that a dev_info node has been unconfigured.
4753  */
4754 static int
4755 i_log_devfs_remove_devinfo(char *pathname, char *class_name, char *driver_name,
4756     int instance, uint_t flags)
4757 {
4758 	sysevent_t		*ev;
4759 	sysevent_id_t		eid;
4760 	sysevent_value_t	se_val;
4761 	sysevent_attr_list_t	*ev_attr_list = NULL;
4762 	int			se_err;
4763 	int			no_transport = 0;
4764 
4765 	if (!i_ddi_io_initialized())
4766 		return (DDI_SUCCESS);
4767 
4768 	/* Invalidate the devinfo snapshot cache */
4769 	i_ddi_di_cache_invalidate();
4770 
4771 	ev = sysevent_alloc(EC_DEVFS, ESC_DEVFS_DEVI_REMOVE, EP_DDI, SE_SLEEP);
4772 
4773 	se_val.value_type = SE_DATA_TYPE_STRING;
4774 	se_val.value.sv_string = pathname;
4775 	if (sysevent_add_attr(&ev_attr_list, DEVFS_PATHNAME,
4776 	    &se_val, SE_SLEEP) != 0) {
4777 		goto fail;
4778 	}
4779 
4780 	if (class_name) {
4781 		/* add the device class, driver name and instance attributes */
4782 
4783 		se_val.value_type = SE_DATA_TYPE_STRING;
4784 		se_val.value.sv_string = class_name;
4785 		if (sysevent_add_attr(&ev_attr_list,
4786 		    DEVFS_DEVI_CLASS, &se_val, SE_SLEEP) != 0) {
4787 			sysevent_free_attr(ev_attr_list);
4788 			goto fail;
4789 		}
4790 
4791 		se_val.value_type = SE_DATA_TYPE_STRING;
4792 		se_val.value.sv_string = driver_name;
4793 		if (sysevent_add_attr(&ev_attr_list,
4794 		    DEVFS_DRIVER_NAME, &se_val, SE_SLEEP) != 0) {
4795 			sysevent_free_attr(ev_attr_list);
4796 			goto fail;
4797 		}
4798 
4799 		se_val.value_type = SE_DATA_TYPE_INT32;
4800 		se_val.value.sv_int32 = instance;
4801 		if (sysevent_add_attr(&ev_attr_list,
4802 		    DEVFS_INSTANCE, &se_val, SE_SLEEP) != 0) {
4803 			sysevent_free_attr(ev_attr_list);
4804 			goto fail;
4805 		}
4806 	}
4807 
4808 	/*
4809 	 * must log a branch event too unless NDI_BRANCH_EVENT_OP is set,
4810 	 * in which case the branch event will be logged by the caller
4811 	 * after the entire branch has been unconfigured.
4812 	 */
4813 	if ((flags & NDI_BRANCH_EVENT_OP) == 0) {
4814 		/*
4815 		 * Instead of logging a separate branch event just add
4816 		 * DEVFS_BRANCH_EVENT attribute. It indicates devfsadmd to
4817 		 * generate a EC_DEV_BRANCH event.
4818 		 */
4819 		se_val.value_type = SE_DATA_TYPE_INT32;
4820 		se_val.value.sv_int32 = 1;
4821 		if (sysevent_add_attr(&ev_attr_list,
4822 		    DEVFS_BRANCH_EVENT, &se_val, SE_SLEEP) != 0) {
4823 			sysevent_free_attr(ev_attr_list);
4824 			goto fail;
4825 		}
4826 	}
4827 
4828 	if (sysevent_attach_attributes(ev, ev_attr_list) != 0) {
4829 		sysevent_free_attr(ev_attr_list);
4830 		goto fail;
4831 	}
4832 
4833 	if ((se_err = log_sysevent(ev, SE_SLEEP, &eid)) != 0) {
4834 		if (se_err == SE_NO_TRANSPORT)
4835 			no_transport = 1;
4836 		goto fail;
4837 	}
4838 
4839 	sysevent_free(ev);
4840 	return (DDI_SUCCESS);
4841 
4842 fail:
4843 	sysevent_free(ev);
4844 	cmn_err(CE_WARN, "failed to log ESC_DEVFS_DEVI_REMOVE event for %s%s",
4845 	    pathname, (no_transport) ? " (syseventd not responding)" : "");
4846 	return (DDI_SUCCESS);
4847 }
4848 
4849 static void
4850 i_ddi_log_devfs_device_remove(dev_info_t *dip)
4851 {
4852 	char	*path;
4853 
4854 	ASSERT(dip && ddi_get_parent(dip) &&
4855 	    DEVI_BUSY_OWNED(ddi_get_parent(dip)));
4856 	ASSERT(DEVI_IS_DEVICE_REMOVED(dip));
4857 
4858 	ASSERT(i_ddi_node_state(dip) >= DS_INITIALIZED);
4859 	if (i_ddi_node_state(dip) < DS_INITIALIZED)
4860 		return;
4861 
4862 	/* Inform LDI_EV_DEVICE_REMOVE callbacks. */
4863 	ldi_invoke_finalize(dip, DDI_DEV_T_ANY, 0, LDI_EV_DEVICE_REMOVE,
4864 	    LDI_EV_SUCCESS, NULL);
4865 
4866 	/* Generate EC_DEVFS_DEVI_REMOVE sysevent. */
4867 	path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4868 	(void) i_log_devfs_remove_devinfo(ddi_pathname(dip, path),
4869 	    i_ddi_devi_class(dip), (char *)ddi_driver_name(dip),
4870 	    ddi_get_instance(dip), 0);
4871 	kmem_free(path, MAXPATHLEN);
4872 }
4873 
4874 static void
4875 i_ddi_log_devfs_device_insert(dev_info_t *dip)
4876 {
4877 	ASSERT(dip && ddi_get_parent(dip) &&
4878 	    DEVI_BUSY_OWNED(ddi_get_parent(dip)));
4879 	ASSERT(!DEVI_IS_DEVICE_REMOVED(dip));
4880 
4881 	(void) i_log_devfs_add_devinfo(dip, 0);
4882 }
4883 
4884 
4885 /*
4886  * log an event that a dev_info branch has been configured or unconfigured.
4887  */
4888 static int
4889 i_log_devfs_branch(char *node_path, char *subclass)
4890 {
4891 	int se_err;
4892 	sysevent_t *ev;
4893 	sysevent_id_t eid;
4894 	sysevent_value_t se_val;
4895 	sysevent_attr_list_t *ev_attr_list = NULL;
4896 	int no_transport = 0;
4897 
4898 	/* do not generate the event during boot */
4899 	if (!i_ddi_io_initialized())
4900 		return (DDI_SUCCESS);
4901 
4902 	/* Invalidate the devinfo snapshot cache */
4903 	i_ddi_di_cache_invalidate();
4904 
4905 	ev = sysevent_alloc(EC_DEVFS, subclass, EP_DDI, SE_SLEEP);
4906 
4907 	se_val.value_type = SE_DATA_TYPE_STRING;
4908 	se_val.value.sv_string = node_path;
4909 
4910 	if (sysevent_add_attr(&ev_attr_list, DEVFS_PATHNAME,
4911 	    &se_val, SE_SLEEP) != 0) {
4912 		goto fail;
4913 	}
4914 
4915 	if (sysevent_attach_attributes(ev, ev_attr_list) != 0) {
4916 		sysevent_free_attr(ev_attr_list);
4917 		goto fail;
4918 	}
4919 
4920 	if ((se_err = log_sysevent(ev, SE_SLEEP, &eid)) != 0) {
4921 		if (se_err == SE_NO_TRANSPORT)
4922 			no_transport = 1;
4923 		goto fail;
4924 	}
4925 
4926 	sysevent_free(ev);
4927 	return (DDI_SUCCESS);
4928 
4929 fail:
4930 	cmn_err(CE_WARN, "failed to log %s branch event for %s%s",
4931 	    subclass, node_path,
4932 	    (no_transport) ? " (syseventd not responding)" : "");
4933 
4934 	sysevent_free(ev);
4935 	return (DDI_FAILURE);
4936 }
4937 
4938 /*
4939  * log an event that a dev_info tree branch has been configured.
4940  */
4941 static int
4942 i_log_devfs_branch_add(dev_info_t *dip)
4943 {
4944 	char *node_path;
4945 	int rv;
4946 
4947 	node_path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4948 	(void) ddi_pathname(dip, node_path);
4949 	rv = i_log_devfs_branch(node_path, ESC_DEVFS_BRANCH_ADD);
4950 	kmem_free(node_path, MAXPATHLEN);
4951 
4952 	return (rv);
4953 }
4954 
4955 /*
4956  * log an event that a dev_info tree branch has been unconfigured.
4957  */
4958 static int
4959 i_log_devfs_branch_remove(char *node_path)
4960 {
4961 	return (i_log_devfs_branch(node_path, ESC_DEVFS_BRANCH_REMOVE));
4962 }
4963 
4964 /*
4965  * enqueue the dip's deviname on the branch event queue.
4966  */
4967 static struct brevq_node *
4968 brevq_enqueue(struct brevq_node **brevqp, dev_info_t *dip,
4969     struct brevq_node *child)
4970 {
4971 	struct brevq_node *brn;
4972 	char *deviname;
4973 
4974 	deviname = kmem_alloc(MAXNAMELEN, KM_SLEEP);
4975 	(void) ddi_deviname(dip, deviname);
4976 
4977 	brn = kmem_zalloc(sizeof (*brn), KM_SLEEP);
4978 	brn->brn_deviname = i_ddi_strdup(deviname, KM_SLEEP);
4979 	kmem_free(deviname, MAXNAMELEN);
4980 	brn->brn_child = child;
4981 	brn->brn_sibling = *brevqp;
4982 	*brevqp = brn;
4983 
4984 	return (brn);
4985 }
4986 
4987 /*
4988  * free the memory allocated for the elements on the branch event queue.
4989  */
4990 static void
4991 free_brevq(struct brevq_node *brevq)
4992 {
4993 	struct brevq_node *brn, *next_brn;
4994 
4995 	for (brn = brevq; brn != NULL; brn = next_brn) {
4996 		next_brn = brn->brn_sibling;
4997 		ASSERT(brn->brn_child == NULL);
4998 		kmem_free(brn->brn_deviname, strlen(brn->brn_deviname) + 1);
4999 		kmem_free(brn, sizeof (*brn));
5000 	}
5001 }
5002 
5003 /*
5004  * log the events queued up on the branch event queue and free the
5005  * associated memory.
5006  *
5007  * node_path must have been allocated with at least MAXPATHLEN bytes.
5008  */
5009 static void
5010 log_and_free_brevq(char *node_path, struct brevq_node *brevq)
5011 {
5012 	struct brevq_node *brn;
5013 	char *p;
5014 
5015 	p = node_path + strlen(node_path);
5016 	for (brn = brevq; brn != NULL; brn = brn->brn_sibling) {
5017 		(void) strcpy(p, brn->brn_deviname);
5018 		(void) i_log_devfs_branch_remove(node_path);
5019 	}
5020 	*p = '\0';
5021 
5022 	free_brevq(brevq);
5023 }
5024 
5025 /*
5026  * log the events queued up on the branch event queue and free the
5027  * associated memory. Same as the previous function but operates on dip.
5028  */
5029 static void
5030 log_and_free_brevq_dip(dev_info_t *dip, struct brevq_node *brevq)
5031 {
5032 	char *path;
5033 
5034 	path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5035 	(void) ddi_pathname(dip, path);
5036 	log_and_free_brevq(path, brevq);
5037 	kmem_free(path, MAXPATHLEN);
5038 }
5039 
5040 /*
5041  * log the outstanding branch remove events for the grand children of the dip
5042  * and free the associated memory.
5043  */
5044 static void
5045 log_and_free_br_events_on_grand_children(dev_info_t *dip,
5046     struct brevq_node *brevq)
5047 {
5048 	struct brevq_node *brn;
5049 	char *path;
5050 	char *p;
5051 
5052 	path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5053 	(void) ddi_pathname(dip, path);
5054 	p = path + strlen(path);
5055 	for (brn = brevq; brn != NULL; brn = brn->brn_sibling) {
5056 		if (brn->brn_child) {
5057 			(void) strcpy(p, brn->brn_deviname);
5058 			/* now path contains the node path to the dip's child */
5059 			log_and_free_brevq(path, brn->brn_child);
5060 			brn->brn_child = NULL;
5061 		}
5062 	}
5063 	kmem_free(path, MAXPATHLEN);
5064 }
5065 
5066 /*
5067  * log and cleanup branch remove events for the grand children of the dip.
5068  */
5069 static void
5070 cleanup_br_events_on_grand_children(dev_info_t *dip, struct brevq_node **brevqp)
5071 {
5072 	dev_info_t *child;
5073 	struct brevq_node *brevq, *brn, *prev_brn, *next_brn;
5074 	char *path;
5075 	int circ;
5076 
5077 	path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5078 	prev_brn = NULL;
5079 	brevq = *brevqp;
5080 
5081 	ndi_devi_enter(dip, &circ);
5082 	for (brn = brevq; brn != NULL; brn = next_brn) {
5083 		next_brn = brn->brn_sibling;
5084 		for (child = ddi_get_child(dip); child != NULL;
5085 		    child = ddi_get_next_sibling(child)) {
5086 			if (i_ddi_node_state(child) >= DS_INITIALIZED) {
5087 				(void) ddi_deviname(child, path);
5088 				if (strcmp(path, brn->brn_deviname) == 0)
5089 					break;
5090 			}
5091 		}
5092 
5093 		if (child != NULL && !(DEVI_EVREMOVE(child))) {
5094 			/*
5095 			 * Event state is not REMOVE. So branch remove event
5096 			 * is not going be generated on brn->brn_child.
5097 			 * If any branch remove events were queued up on
5098 			 * brn->brn_child log them and remove the brn
5099 			 * from the queue.
5100 			 */
5101 			if (brn->brn_child) {
5102 				(void) ddi_pathname(dip, path);
5103 				(void) strcat(path, brn->brn_deviname);
5104 				log_and_free_brevq(path, brn->brn_child);
5105 			}
5106 
5107 			if (prev_brn)
5108 				prev_brn->brn_sibling = next_brn;
5109 			else
5110 				*brevqp = next_brn;
5111 
5112 			kmem_free(brn->brn_deviname,
5113 			    strlen(brn->brn_deviname) + 1);
5114 			kmem_free(brn, sizeof (*brn));
5115 		} else {
5116 			/*
5117 			 * Free up the outstanding branch remove events
5118 			 * queued on brn->brn_child since brn->brn_child
5119 			 * itself is eligible for branch remove event.
5120 			 */
5121 			if (brn->brn_child) {
5122 				free_brevq(brn->brn_child);
5123 				brn->brn_child = NULL;
5124 			}
5125 			prev_brn = brn;
5126 		}
5127 	}
5128 
5129 	ndi_devi_exit(dip, circ);
5130 	kmem_free(path, MAXPATHLEN);
5131 }
5132 
5133 static int
5134 need_remove_event(dev_info_t *dip, int flags)
5135 {
5136 	if ((flags & (NDI_NO_EVENT | NDI_AUTODETACH)) == 0 &&
5137 	    (flags & (NDI_DEVI_OFFLINE | NDI_UNCONFIG | NDI_DEVI_REMOVE)) &&
5138 	    !(DEVI_EVREMOVE(dip)))
5139 		return (1);
5140 	else
5141 		return (0);
5142 }
5143 
5144 /*
5145  * Unconfigure children/descendants of the dip.
5146  *
5147  * If the operation involves a branch event NDI_BRANCH_EVENT_OP is set
5148  * through out the unconfiguration. On successful return *brevqp is set to
5149  * a queue of dip's child devinames for which branch remove events need
5150  * to be generated.
5151  */
5152 static int
5153 devi_unconfig_branch(dev_info_t *dip, dev_info_t **dipp, int flags,
5154     struct brevq_node **brevqp)
5155 {
5156 	int rval;
5157 
5158 	*brevqp = NULL;
5159 
5160 	if ((!(flags & NDI_BRANCH_EVENT_OP)) && need_remove_event(dip, flags))
5161 		flags |= NDI_BRANCH_EVENT_OP;
5162 
5163 	if (flags & NDI_BRANCH_EVENT_OP) {
5164 		rval = devi_unconfig_common(dip, dipp, flags, DDI_MAJOR_T_NONE,
5165 		    brevqp);
5166 
5167 		if (rval != NDI_SUCCESS && (*brevqp)) {
5168 			log_and_free_brevq_dip(dip, *brevqp);
5169 			*brevqp = NULL;
5170 		}
5171 	} else
5172 		rval = devi_unconfig_common(dip, dipp, flags, DDI_MAJOR_T_NONE,
5173 		    NULL);
5174 
5175 	return (rval);
5176 }
5177 
5178 /*
5179  * If the dip is already bound to a driver transition to DS_INITIALIZED
5180  * in order to generate an event in the case where the node was left in
5181  * DS_BOUND state since boot (never got attached) and the node is now
5182  * being offlined.
5183  */
5184 static void
5185 init_bound_node_ev(dev_info_t *pdip, dev_info_t *dip, int flags)
5186 {
5187 	if (need_remove_event(dip, flags) &&
5188 	    i_ddi_node_state(dip) == DS_BOUND &&
5189 	    i_ddi_devi_attached(pdip) && !DEVI_IS_DEVICE_OFFLINE(dip))
5190 		(void) ddi_initchild(pdip, dip);
5191 }
5192 
5193 /*
5194  * attach a node/branch with parent already held busy
5195  */
5196 static int
5197 devi_attach_node(dev_info_t *dip, uint_t flags)
5198 {
5199 	dev_info_t *pdip = ddi_get_parent(dip);
5200 
5201 	ASSERT(pdip && DEVI_BUSY_OWNED(pdip));
5202 
5203 	mutex_enter(&(DEVI(dip)->devi_lock));
5204 	if (flags & NDI_DEVI_ONLINE) {
5205 		if (!i_ddi_devi_attached(dip))
5206 			DEVI_SET_REPORT(dip);
5207 		DEVI_SET_DEVICE_ONLINE(dip);
5208 	}
5209 	if (DEVI_IS_DEVICE_OFFLINE(dip)) {
5210 		mutex_exit(&(DEVI(dip)->devi_lock));
5211 		return (NDI_FAILURE);
5212 	}
5213 	mutex_exit(&(DEVI(dip)->devi_lock));
5214 
5215 	if (i_ddi_attachchild(dip) != DDI_SUCCESS) {
5216 		mutex_enter(&(DEVI(dip)->devi_lock));
5217 		DEVI_SET_EVUNINIT(dip);
5218 		mutex_exit(&(DEVI(dip)->devi_lock));
5219 
5220 		if (ndi_dev_is_persistent_node(dip))
5221 			(void) ddi_uninitchild(dip);
5222 		else {
5223 			/*
5224 			 * Delete .conf nodes and nodes that are not
5225 			 * well formed.
5226 			 */
5227 			(void) ddi_remove_child(dip, 0);
5228 		}
5229 		return (NDI_FAILURE);
5230 	}
5231 
5232 	i_ndi_devi_report_status_change(dip, NULL);
5233 
5234 	/*
5235 	 * log an event, but not during devfs lookups in which case
5236 	 * NDI_NO_EVENT is set.
5237 	 */
5238 	if ((flags & NDI_NO_EVENT) == 0 && !(DEVI_EVADD(dip))) {
5239 		(void) i_log_devfs_add_devinfo(dip, flags);
5240 
5241 		mutex_enter(&(DEVI(dip)->devi_lock));
5242 		DEVI_SET_EVADD(dip);
5243 		mutex_exit(&(DEVI(dip)->devi_lock));
5244 	} else if (!(flags & NDI_NO_EVENT_STATE_CHNG)) {
5245 		mutex_enter(&(DEVI(dip)->devi_lock));
5246 		DEVI_SET_EVADD(dip);
5247 		mutex_exit(&(DEVI(dip)->devi_lock));
5248 	}
5249 
5250 	return (NDI_SUCCESS);
5251 }
5252 
5253 /* internal function to config immediate children */
5254 static int
5255 config_immediate_children(dev_info_t *pdip, uint_t flags, major_t major)
5256 {
5257 	dev_info_t	*child, *next;
5258 	int		circ;
5259 
5260 	ASSERT(i_ddi_devi_attached(pdip));
5261 
5262 	if (!NEXUS_DRV(ddi_get_driver(pdip)))
5263 		return (NDI_SUCCESS);
5264 
5265 	NDI_CONFIG_DEBUG((CE_CONT,
5266 	    "config_immediate_children: %s%d (%p), flags=%x\n",
5267 	    ddi_driver_name(pdip), ddi_get_instance(pdip),
5268 	    (void *)pdip, flags));
5269 
5270 	ndi_devi_enter(pdip, &circ);
5271 
5272 	if (flags & NDI_CONFIG_REPROBE) {
5273 		mutex_enter(&DEVI(pdip)->devi_lock);
5274 		DEVI(pdip)->devi_flags &= ~DEVI_MADE_CHILDREN;
5275 		mutex_exit(&DEVI(pdip)->devi_lock);
5276 	}
5277 	(void) i_ndi_make_spec_children(pdip, flags);
5278 	i_ndi_init_hw_children(pdip, flags);
5279 
5280 	child = ddi_get_child(pdip);
5281 	while (child) {
5282 		/* NOTE: devi_attach_node() may remove the dip */
5283 		next = ddi_get_next_sibling(child);
5284 
5285 		/*
5286 		 * Configure all nexus nodes or leaf nodes with
5287 		 * matching driver major
5288 		 */
5289 		if ((major == DDI_MAJOR_T_NONE) ||
5290 		    (major == ddi_driver_major(child)) ||
5291 		    ((flags & NDI_CONFIG) && (is_leaf_node(child) == 0)))
5292 			(void) devi_attach_node(child, flags);
5293 		child = next;
5294 	}
5295 
5296 	ndi_devi_exit(pdip, circ);
5297 
5298 	return (NDI_SUCCESS);
5299 }
5300 
5301 /* internal function to config grand children */
5302 static int
5303 config_grand_children(dev_info_t *pdip, uint_t flags, major_t major)
5304 {
5305 	struct mt_config_handle *hdl;
5306 
5307 	/* multi-threaded configuration of child nexus */
5308 	hdl = mt_config_init(pdip, NULL, flags, major, MT_CONFIG_OP, NULL);
5309 	mt_config_children(hdl);
5310 
5311 	return (mt_config_fini(hdl));	/* wait for threads to exit */
5312 }
5313 
5314 /*
5315  * Common function for device tree configuration,
5316  * either BUS_CONFIG_ALL or BUS_CONFIG_DRIVER.
5317  * The NDI_CONFIG flag causes recursive configuration of
5318  * grandchildren, devfs usage should not recurse.
5319  */
5320 static int
5321 devi_config_common(dev_info_t *dip, int flags, major_t major)
5322 {
5323 	int error;
5324 	int (*f)();
5325 
5326 	if (!i_ddi_devi_attached(dip))
5327 		return (NDI_FAILURE);
5328 
5329 	if (pm_pre_config(dip, NULL) != DDI_SUCCESS)
5330 		return (NDI_FAILURE);
5331 
5332 	if ((DEVI(dip)->devi_ops->devo_bus_ops == NULL) ||
5333 	    (DEVI(dip)->devi_ops->devo_bus_ops->busops_rev < BUSO_REV_5) ||
5334 	    (f = DEVI(dip)->devi_ops->devo_bus_ops->bus_config) == NULL) {
5335 		error = config_immediate_children(dip, flags, major);
5336 	} else {
5337 		/* call bus_config entry point */
5338 		ddi_bus_config_op_t bus_op = (major == DDI_MAJOR_T_NONE) ?
5339 		    BUS_CONFIG_ALL : BUS_CONFIG_DRIVER;
5340 		error = (*f)(dip,
5341 		    flags, bus_op, (void *)(uintptr_t)major, NULL, 0);
5342 	}
5343 
5344 	if (error) {
5345 		pm_post_config(dip, NULL);
5346 		return (error);
5347 	}
5348 
5349 	/*
5350 	 * Some callers, notably SCSI, need to mark the devfs cache
5351 	 * to be rebuilt together with the config operation.
5352 	 */
5353 	if (flags & NDI_DEVFS_CLEAN)
5354 		(void) devfs_clean(dip, NULL, 0);
5355 
5356 	if (flags & NDI_CONFIG)
5357 		(void) config_grand_children(dip, flags, major);
5358 
5359 	pm_post_config(dip, NULL);
5360 
5361 	return (NDI_SUCCESS);
5362 }
5363 
5364 /*
5365  * Framework entry point for BUS_CONFIG_ALL
5366  */
5367 int
5368 ndi_devi_config(dev_info_t *dip, int flags)
5369 {
5370 	NDI_CONFIG_DEBUG((CE_CONT,
5371 	    "ndi_devi_config: par = %s%d (%p), flags = 0x%x\n",
5372 	    ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip, flags));
5373 
5374 	return (devi_config_common(dip, flags, DDI_MAJOR_T_NONE));
5375 }
5376 
5377 /*
5378  * Framework entry point for BUS_CONFIG_DRIVER, bound to major
5379  */
5380 int
5381 ndi_devi_config_driver(dev_info_t *dip, int flags, major_t major)
5382 {
5383 	/* don't abuse this function */
5384 	ASSERT(major != DDI_MAJOR_T_NONE);
5385 
5386 	NDI_CONFIG_DEBUG((CE_CONT,
5387 	    "ndi_devi_config_driver: par = %s%d (%p), flags = 0x%x\n",
5388 	    ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip, flags));
5389 
5390 	return (devi_config_common(dip, flags, major));
5391 }
5392 
5393 /*
5394  * Called by nexus drivers to configure its children.
5395  */
5396 static int
5397 devi_config_one(dev_info_t *pdip, char *devnm, dev_info_t **cdipp,
5398     uint_t flags, clock_t timeout)
5399 {
5400 	dev_info_t	*vdip = NULL;
5401 	char		*drivername = NULL;
5402 	int		find_by_addr = 0;
5403 	char		*name, *addr;
5404 	int		v_circ, p_circ;
5405 	clock_t		end_time;	/* 60 sec */
5406 	int		probed;
5407 	dev_info_t	*cdip;
5408 	mdi_pathinfo_t	*cpip;
5409 
5410 	*cdipp = NULL;
5411 
5412 	if (!NEXUS_DRV(ddi_get_driver(pdip)))
5413 		return (NDI_FAILURE);
5414 
5415 	/* split name into "name@addr" parts */
5416 	i_ddi_parse_name(devnm, &name, &addr, NULL);
5417 
5418 	/*
5419 	 * If the nexus is a pHCI and we are not processing a pHCI from
5420 	 * mdi bus_config code then we need to know the vHCI.
5421 	 */
5422 	if (MDI_PHCI(pdip))
5423 		vdip = mdi_devi_get_vdip(pdip);
5424 
5425 	/*
5426 	 * We may have a genericname on a system that creates drivername
5427 	 * nodes (from .conf files).  Find the drivername by nodeid. If we
5428 	 * can't find a node with devnm as the node name then we search by
5429 	 * drivername.	This allows an implementation to supply a genericly
5430 	 * named boot path (disk) and locate drivename nodes (sd).  The
5431 	 * NDI_PROMNAME flag does not apply to /devices/pseudo paths.
5432 	 */
5433 	if ((flags & NDI_PROMNAME) && (pdip != pseudo_dip)) {
5434 		drivername = child_path_to_driver(pdip, name, addr);
5435 		find_by_addr = 1;
5436 	}
5437 
5438 	/*
5439 	 * Determine end_time: This routine should *not* be called with a
5440 	 * constant non-zero timeout argument, the caller should be adjusting
5441 	 * the timeout argument relative to when it *started* its asynchronous
5442 	 * enumeration.
5443 	 */
5444 	if (timeout > 0)
5445 		end_time = ddi_get_lbolt() + timeout;
5446 
5447 	for (;;) {
5448 		/*
5449 		 * For pHCI, enter (vHCI, pHCI) and search for pathinfo/client
5450 		 * child - break out of for(;;) loop if child found.
5451 		 * NOTE: Lock order for ndi_devi_enter is (vHCI, pHCI).
5452 		 */
5453 		if (vdip) {
5454 			/* use mdi_devi_enter ordering */
5455 			ndi_devi_enter(vdip, &v_circ);
5456 			ndi_devi_enter(pdip, &p_circ);
5457 			cpip = mdi_pi_find(pdip, NULL, addr);
5458 			cdip = mdi_pi_get_client(cpip);
5459 			if (cdip)
5460 				break;
5461 		} else
5462 			ndi_devi_enter(pdip, &p_circ);
5463 
5464 		/*
5465 		 * When not a  vHCI or not all pHCI devices are required to
5466 		 * enumerated under the vHCI (NDI_MDI_FALLBACK) search for
5467 		 * devinfo child.
5468 		 */
5469 		if ((vdip == NULL) || (flags & NDI_MDI_FALLBACK)) {
5470 			/* determine if .conf nodes already built */
5471 			probed = (DEVI(pdip)->devi_flags & DEVI_MADE_CHILDREN);
5472 
5473 			/*
5474 			 * Search for child by name, if not found then search
5475 			 * for a node bound to the drivername driver with the
5476 			 * specified "@addr". Break out of for(;;) loop if
5477 			 * child found.  To support path-oriented aliases
5478 			 * binding on boot-device, we do a search_by_addr too.
5479 			 */
5480 again:			(void) i_ndi_make_spec_children(pdip, flags);
5481 			cdip = find_child_by_name(pdip, name, addr);
5482 			if ((cdip == NULL) && drivername)
5483 				cdip = find_child_by_driver(pdip,
5484 				    drivername, addr);
5485 			if ((cdip == NULL) && find_by_addr)
5486 				cdip = find_child_by_addr(pdip, addr);
5487 			if (cdip)
5488 				break;
5489 
5490 			/*
5491 			 * determine if we should reenumerate .conf nodes
5492 			 * and look for child again.
5493 			 */
5494 			if (probed &&
5495 			    i_ddi_io_initialized() &&
5496 			    (flags & NDI_CONFIG_REPROBE) &&
5497 			    ((timeout <= 0) || (ddi_get_lbolt() >= end_time))) {
5498 				probed = 0;
5499 				mutex_enter(&DEVI(pdip)->devi_lock);
5500 				DEVI(pdip)->devi_flags &= ~DEVI_MADE_CHILDREN;
5501 				mutex_exit(&DEVI(pdip)->devi_lock);
5502 				goto again;
5503 			}
5504 		}
5505 
5506 		/* break out of for(;;) if time expired */
5507 		if ((timeout <= 0) || (ddi_get_lbolt() >= end_time))
5508 			break;
5509 
5510 		/*
5511 		 * Child not found, exit and wait for asynchronous enumeration
5512 		 * to add child (or timeout). The addition of a new child (vhci
5513 		 * or phci) requires the asynchronous enumeration thread to
5514 		 * ndi_devi_enter/ndi_devi_exit. This exit will signal devi_cv
5515 		 * and cause us to return from ndi_devi_exit_and_wait, after
5516 		 * which we loop and search for the requested child again.
5517 		 */
5518 		NDI_DEBUG(flags, (CE_CONT,
5519 		    "%s%d: waiting for child %s@%s, timeout %ld",
5520 		    ddi_driver_name(pdip), ddi_get_instance(pdip),
5521 		    name, addr, timeout));
5522 		if (vdip) {
5523 			/*
5524 			 * Mark vHCI for pHCI ndi_devi_exit broadcast.
5525 			 */
5526 			mutex_enter(&DEVI(vdip)->devi_lock);
5527 			DEVI(vdip)->devi_flags |=
5528 			    DEVI_PHCI_SIGNALS_VHCI;
5529 			mutex_exit(&DEVI(vdip)->devi_lock);
5530 			ndi_devi_exit(pdip, p_circ);
5531 
5532 			/*
5533 			 * NB: There is a small race window from above
5534 			 * ndi_devi_exit() of pdip to cv_wait() in
5535 			 * ndi_devi_exit_and_wait() which can result in
5536 			 * not immediately finding a new pHCI child
5537 			 * of a pHCI that uses NDI_MDI_FAILBACK.
5538 			 */
5539 			ndi_devi_exit_and_wait(vdip, v_circ, end_time);
5540 		} else {
5541 			ndi_devi_exit_and_wait(pdip, p_circ, end_time);
5542 		}
5543 	}
5544 
5545 	/* done with paddr, fixup i_ddi_parse_name '@'->'\0' change */
5546 	if (addr && *addr != '\0')
5547 		*(addr - 1) = '@';
5548 
5549 	/* attach and hold the child, returning pointer to child */
5550 	if (cdip && (devi_attach_node(cdip, flags) == NDI_SUCCESS)) {
5551 		ndi_hold_devi(cdip);
5552 		*cdipp = cdip;
5553 	}
5554 
5555 	ndi_devi_exit(pdip, p_circ);
5556 	if (vdip)
5557 		ndi_devi_exit(vdip, v_circ);
5558 	return (*cdipp ? NDI_SUCCESS : NDI_FAILURE);
5559 }
5560 
5561 /*
5562  * Enumerate and attach a child specified by name 'devnm'.
5563  * Called by devfs lookup and DR to perform a BUS_CONFIG_ONE.
5564  * Note: devfs does not make use of NDI_CONFIG to configure
5565  * an entire branch.
5566  */
5567 int
5568 ndi_devi_config_one(dev_info_t *pdip, char *devnm, dev_info_t **dipp, int flags)
5569 {
5570 	int error;
5571 	int (*f)();
5572 	char *nmdup;
5573 	int duplen;
5574 	int branch_event = 0;
5575 
5576 	ASSERT(pdip);
5577 	ASSERT(devnm);
5578 	ASSERT(dipp);
5579 	ASSERT(i_ddi_devi_attached(pdip));
5580 
5581 	NDI_CONFIG_DEBUG((CE_CONT,
5582 	    "ndi_devi_config_one: par = %s%d (%p), child = %s\n",
5583 	    ddi_driver_name(pdip), ddi_get_instance(pdip),
5584 	    (void *)pdip, devnm));
5585 
5586 	*dipp = NULL;
5587 
5588 	if (pm_pre_config(pdip, devnm) != DDI_SUCCESS) {
5589 		cmn_err(CE_WARN, "preconfig failed: %s", devnm);
5590 		return (NDI_FAILURE);
5591 	}
5592 
5593 	if ((flags & (NDI_NO_EVENT | NDI_BRANCH_EVENT_OP)) == 0 &&
5594 	    (flags & NDI_CONFIG)) {
5595 		flags |= NDI_BRANCH_EVENT_OP;
5596 		branch_event = 1;
5597 	}
5598 
5599 	nmdup = strdup(devnm);
5600 	duplen = strlen(devnm) + 1;
5601 
5602 	if ((DEVI(pdip)->devi_ops->devo_bus_ops == NULL) ||
5603 	    (DEVI(pdip)->devi_ops->devo_bus_ops->busops_rev < BUSO_REV_5) ||
5604 	    (f = DEVI(pdip)->devi_ops->devo_bus_ops->bus_config) == NULL) {
5605 		error = devi_config_one(pdip, devnm, dipp, flags, 0);
5606 	} else {
5607 		/* call bus_config entry point */
5608 		error = (*f)(pdip, flags, BUS_CONFIG_ONE, (void *)devnm, dipp);
5609 	}
5610 
5611 	if (error) {
5612 		*dipp = NULL;
5613 	}
5614 
5615 	/*
5616 	 * if we fail to lookup and this could be an alias, lookup currdip
5617 	 * To prevent recursive lookups into the same hash table, only
5618 	 * do the currdip lookups once the hash table init is complete.
5619 	 * Use tsd so that redirection doesn't recurse
5620 	 */
5621 	if (error) {
5622 		char *alias = kmem_alloc(MAXPATHLEN, KM_NOSLEEP);
5623 		if (alias == NULL) {
5624 			ddi_err(DER_PANIC, pdip, "alias alloc failed: %s",
5625 			    nmdup);
5626 		}
5627 		(void) ddi_pathname(pdip, alias);
5628 		(void) strlcat(alias, "/", MAXPATHLEN);
5629 		(void) strlcat(alias, nmdup, MAXPATHLEN);
5630 
5631 		*dipp = ddi_alias_redirect(alias);
5632 		error = (*dipp ? NDI_SUCCESS : NDI_FAILURE);
5633 
5634 		kmem_free(alias, MAXPATHLEN);
5635 	}
5636 	kmem_free(nmdup, duplen);
5637 
5638 	if (error || !(flags & NDI_CONFIG)) {
5639 		pm_post_config(pdip, devnm);
5640 		return (error);
5641 	}
5642 
5643 	/*
5644 	 * DR usage (i.e. call with NDI_CONFIG) recursively configures
5645 	 * grandchildren, performing a BUS_CONFIG_ALL from the node attached
5646 	 * by the BUS_CONFIG_ONE.
5647 	 */
5648 	ASSERT(*dipp);
5649 	error = devi_config_common(*dipp, flags, DDI_MAJOR_T_NONE);
5650 
5651 	pm_post_config(pdip, devnm);
5652 
5653 	if (branch_event)
5654 		(void) i_log_devfs_branch_add(*dipp);
5655 
5656 	return (error);
5657 }
5658 
5659 /*
5660  * Enumerate and attach a child specified by name 'devnm'.
5661  * Called during configure the OBP options. This configures
5662  * only one node.
5663  */
5664 static int
5665 ndi_devi_config_obp_args(dev_info_t *parent, char *devnm,
5666     dev_info_t **childp, int flags)
5667 {
5668 	int error;
5669 	int (*f)();
5670 
5671 	ASSERT(childp);
5672 	ASSERT(i_ddi_devi_attached(parent));
5673 
5674 	NDI_CONFIG_DEBUG((CE_CONT, "ndi_devi_config_obp_args: "
5675 	    "par = %s%d (%p), child = %s\n", ddi_driver_name(parent),
5676 	    ddi_get_instance(parent), (void *)parent, devnm));
5677 
5678 	if ((DEVI(parent)->devi_ops->devo_bus_ops == NULL) ||
5679 	    (DEVI(parent)->devi_ops->devo_bus_ops->busops_rev < BUSO_REV_5) ||
5680 	    (f = DEVI(parent)->devi_ops->devo_bus_ops->bus_config) == NULL) {
5681 		error = NDI_FAILURE;
5682 	} else {
5683 		/* call bus_config entry point */
5684 		error = (*f)(parent, flags,
5685 		    BUS_CONFIG_OBP_ARGS, (void *)devnm, childp);
5686 	}
5687 	return (error);
5688 }
5689 
5690 /*
5691  * Pay attention, the following is a bit tricky:
5692  * There are three possible cases when constraints are applied
5693  *
5694  *	- A constraint is applied and the offline is disallowed.
5695  *	  Simply return failure and block the offline
5696  *
5697  *	- A constraint is applied and the offline is allowed.
5698  *	  Mark the dip as having passed the constraint and allow
5699  *	  offline to proceed.
5700  *
5701  *	- A constraint is not applied. Allow the offline to proceed for now.
5702  *
5703  * In the latter two cases we allow the offline to proceed. If the
5704  * offline succeeds (no users) everything is fine. It is ok for an unused
5705  * device to be offlined even if no constraints were imposed on the offline.
5706  * If the offline fails because there are users, we look at the constraint
5707  * flag on the dip. If the constraint flag is set (implying that it passed
5708  * a constraint) we allow the dip to be retired. If not, we don't allow
5709  * the retire. This ensures that we don't allow unconstrained retire.
5710  */
5711 int
5712 e_ddi_offline_notify(dev_info_t *dip)
5713 {
5714 	int retval;
5715 	int constraint;
5716 	int failure;
5717 
5718 	RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): entered: dip=%p",
5719 	    (void *) dip));
5720 
5721 	constraint = 0;
5722 	failure = 0;
5723 
5724 	/*
5725 	 * Start with userland constraints first - applied via device contracts
5726 	 */
5727 	retval = contract_device_offline(dip, DDI_DEV_T_ANY, 0);
5728 	switch (retval) {
5729 	case CT_NACK:
5730 		RIO_DEBUG((CE_NOTE, "Received NACK for dip=%p", (void *)dip));
5731 		failure = 1;
5732 		goto out;
5733 	case CT_ACK:
5734 		constraint = 1;
5735 		RIO_DEBUG((CE_NOTE, "Received ACK for dip=%p", (void *)dip));
5736 		break;
5737 	case CT_NONE:
5738 		/* no contracts */
5739 		RIO_DEBUG((CE_NOTE, "No contracts on dip=%p", (void *)dip));
5740 		break;
5741 	default:
5742 		ASSERT(retval == CT_NONE);
5743 	}
5744 
5745 	/*
5746 	 * Next, use LDI to impose kernel constraints
5747 	 */
5748 	retval = ldi_invoke_notify(dip, DDI_DEV_T_ANY, 0, LDI_EV_OFFLINE, NULL);
5749 	switch (retval) {
5750 	case LDI_EV_FAILURE:
5751 		contract_device_negend(dip, DDI_DEV_T_ANY, 0, CT_EV_FAILURE);
5752 		RIO_DEBUG((CE_NOTE, "LDI callback failed on dip=%p",
5753 		    (void *)dip));
5754 		failure = 1;
5755 		goto out;
5756 	case LDI_EV_SUCCESS:
5757 		constraint = 1;
5758 		RIO_DEBUG((CE_NOTE, "LDI callback success on dip=%p",
5759 		    (void *)dip));
5760 		break;
5761 	case LDI_EV_NONE:
5762 		/* no matching LDI callbacks */
5763 		RIO_DEBUG((CE_NOTE, "No LDI callbacks for dip=%p",
5764 		    (void *)dip));
5765 		break;
5766 	default:
5767 		ASSERT(retval == LDI_EV_NONE);
5768 	}
5769 
5770 out:
5771 	mutex_enter(&(DEVI(dip)->devi_lock));
5772 	if ((DEVI(dip)->devi_flags & DEVI_RETIRING) && failure) {
5773 		RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): setting "
5774 		    "BLOCKED flag. dip=%p", (void *)dip));
5775 		DEVI(dip)->devi_flags |= DEVI_R_BLOCKED;
5776 		if (DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT) {
5777 			RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): "
5778 			    "blocked. clearing RCM CONSTRAINT flag. dip=%p",
5779 			    (void *)dip));
5780 			DEVI(dip)->devi_flags &= ~DEVI_R_CONSTRAINT;
5781 		}
5782 	} else if ((DEVI(dip)->devi_flags & DEVI_RETIRING) && constraint) {
5783 		RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): setting "
5784 		    "CONSTRAINT flag. dip=%p", (void *)dip));
5785 		DEVI(dip)->devi_flags |= DEVI_R_CONSTRAINT;
5786 	} else if ((DEVI(dip)->devi_flags & DEVI_RETIRING) &&
5787 	    ((DEVI(dip)->devi_ops != NULL &&
5788 	    DEVI(dip)->devi_ops->devo_bus_ops != NULL) ||
5789 	    DEVI(dip)->devi_ref == 0)) {
5790 		/* also allow retire if nexus or if device is not in use */
5791 		RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): device not in "
5792 		    "use. Setting CONSTRAINT flag. dip=%p", (void *)dip));
5793 		DEVI(dip)->devi_flags |= DEVI_R_CONSTRAINT;
5794 	} else {
5795 		/*
5796 		 * Note: We cannot ASSERT here that DEVI_R_CONSTRAINT is
5797 		 * not set, since other sources (such as RCM) may have
5798 		 * set the flag.
5799 		 */
5800 		RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): not setting "
5801 		    "constraint flag. dip=%p", (void *)dip));
5802 	}
5803 	mutex_exit(&(DEVI(dip)->devi_lock));
5804 
5805 
5806 	RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): exit: dip=%p",
5807 	    (void *) dip));
5808 
5809 	return (failure ? DDI_FAILURE : DDI_SUCCESS);
5810 }
5811 
5812 void
5813 e_ddi_offline_finalize(dev_info_t *dip, int result)
5814 {
5815 	RIO_DEBUG((CE_NOTE, "e_ddi_offline_finalize(): entry: result=%s, "
5816 	    "dip=%p", result == DDI_SUCCESS ? "SUCCESS" : "FAILURE",
5817 	    (void *)dip));
5818 
5819 	contract_device_negend(dip, DDI_DEV_T_ANY, 0,  result == DDI_SUCCESS ?
5820 	    CT_EV_SUCCESS : CT_EV_FAILURE);
5821 
5822 	ldi_invoke_finalize(dip, DDI_DEV_T_ANY, 0,
5823 	    LDI_EV_OFFLINE, result == DDI_SUCCESS ?
5824 	    LDI_EV_SUCCESS : LDI_EV_FAILURE, NULL);
5825 
5826 	RIO_VERBOSE((CE_NOTE, "e_ddi_offline_finalize(): exit: dip=%p",
5827 	    (void *)dip));
5828 }
5829 
5830 void
5831 e_ddi_degrade_finalize(dev_info_t *dip)
5832 {
5833 	RIO_DEBUG((CE_NOTE, "e_ddi_degrade_finalize(): entry: "
5834 	    "result always = DDI_SUCCESS, dip=%p", (void *)dip));
5835 
5836 	contract_device_degrade(dip, DDI_DEV_T_ANY, 0);
5837 	contract_device_negend(dip, DDI_DEV_T_ANY, 0, CT_EV_SUCCESS);
5838 
5839 	ldi_invoke_finalize(dip, DDI_DEV_T_ANY, 0, LDI_EV_DEGRADE,
5840 	    LDI_EV_SUCCESS, NULL);
5841 
5842 	RIO_VERBOSE((CE_NOTE, "e_ddi_degrade_finalize(): exit: dip=%p",
5843 	    (void *)dip));
5844 }
5845 
5846 void
5847 e_ddi_undegrade_finalize(dev_info_t *dip)
5848 {
5849 	RIO_DEBUG((CE_NOTE, "e_ddi_undegrade_finalize(): entry: "
5850 	    "result always = DDI_SUCCESS, dip=%p", (void *)dip));
5851 
5852 	contract_device_undegrade(dip, DDI_DEV_T_ANY, 0);
5853 	contract_device_negend(dip, DDI_DEV_T_ANY, 0, CT_EV_SUCCESS);
5854 
5855 	RIO_VERBOSE((CE_NOTE, "e_ddi_undegrade_finalize(): exit: dip=%p",
5856 	    (void *)dip));
5857 }
5858 
5859 /*
5860  * detach a node with parent already held busy
5861  */
5862 static int
5863 devi_detach_node(dev_info_t *dip, uint_t flags)
5864 {
5865 	dev_info_t *pdip = ddi_get_parent(dip);
5866 	int ret = NDI_SUCCESS;
5867 	ddi_eventcookie_t cookie;
5868 	char *path = NULL;
5869 	char *class = NULL;
5870 	char *driver = NULL;
5871 	int instance = -1;
5872 	int post_event = 0;
5873 
5874 	ASSERT(pdip && DEVI_BUSY_OWNED(pdip));
5875 
5876 	/*
5877 	 * Invoke notify if offlining
5878 	 */
5879 	if (flags & NDI_DEVI_OFFLINE) {
5880 		RIO_DEBUG((CE_NOTE, "devi_detach_node: offlining dip=%p",
5881 		    (void *)dip));
5882 		if (e_ddi_offline_notify(dip) != DDI_SUCCESS) {
5883 			RIO_DEBUG((CE_NOTE, "devi_detach_node: offline NACKed"
5884 			    "dip=%p", (void *)dip));
5885 			return (NDI_FAILURE);
5886 		}
5887 	}
5888 
5889 	if (flags & NDI_POST_EVENT) {
5890 		if (i_ddi_devi_attached(pdip)) {
5891 			if (ddi_get_eventcookie(dip, DDI_DEVI_REMOVE_EVENT,
5892 			    &cookie) == NDI_SUCCESS)
5893 				(void) ndi_post_event(dip, dip, cookie, NULL);
5894 		}
5895 	}
5896 
5897 	if (i_ddi_detachchild(dip, flags) != DDI_SUCCESS) {
5898 		if (flags & NDI_DEVI_OFFLINE) {
5899 			RIO_DEBUG((CE_NOTE, "devi_detach_node: offline failed."
5900 			    " Calling e_ddi_offline_finalize with result=%d. "
5901 			    "dip=%p", DDI_FAILURE, (void *)dip));
5902 			e_ddi_offline_finalize(dip, DDI_FAILURE);
5903 		}
5904 		return (NDI_FAILURE);
5905 	}
5906 
5907 	if (flags & NDI_DEVI_OFFLINE) {
5908 		RIO_DEBUG((CE_NOTE, "devi_detach_node: offline succeeded."
5909 		    " Calling e_ddi_offline_finalize with result=%d, "
5910 		    "dip=%p", DDI_SUCCESS, (void *)dip));
5911 		e_ddi_offline_finalize(dip, DDI_SUCCESS);
5912 	}
5913 
5914 	if (flags & NDI_AUTODETACH)
5915 		return (NDI_SUCCESS);
5916 
5917 	/*
5918 	 * For DR, even bound nodes may need to have offline
5919 	 * flag set.
5920 	 */
5921 	if (flags & NDI_DEVI_OFFLINE) {
5922 		mutex_enter(&(DEVI(dip)->devi_lock));
5923 		DEVI_SET_DEVICE_OFFLINE(dip);
5924 		mutex_exit(&(DEVI(dip)->devi_lock));
5925 	}
5926 
5927 	if (i_ddi_node_state(dip) == DS_INITIALIZED) {
5928 		struct dev_info *devi = DEVI(dip);
5929 
5930 		if (devi->devi_ev_path == NULL) {
5931 			devi->devi_ev_path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5932 			(void) ddi_pathname(dip, devi->devi_ev_path);
5933 		}
5934 		if (flags & NDI_DEVI_OFFLINE)
5935 			i_ndi_devi_report_status_change(dip,
5936 			    devi->devi_ev_path);
5937 
5938 		if (need_remove_event(dip, flags)) {
5939 			/*
5940 			 * instance and path data are lost in call to
5941 			 * ddi_uninitchild
5942 			 */
5943 			devi->devi_ev_instance = ddi_get_instance(dip);
5944 
5945 			mutex_enter(&(DEVI(dip)->devi_lock));
5946 			DEVI_SET_EVREMOVE(dip);
5947 			mutex_exit(&(DEVI(dip)->devi_lock));
5948 		}
5949 	}
5950 
5951 	if (flags & (NDI_UNCONFIG | NDI_DEVI_REMOVE)) {
5952 		ret = ddi_uninitchild(dip);
5953 		if (ret == NDI_SUCCESS) {
5954 			/*
5955 			 * Remove uninitialized pseudo nodes because
5956 			 * system props are lost and the node cannot be
5957 			 * reattached.
5958 			 */
5959 			if (!ndi_dev_is_persistent_node(dip))
5960 				flags |= NDI_DEVI_REMOVE;
5961 
5962 			if (flags & NDI_DEVI_REMOVE) {
5963 				/*
5964 				 * NOTE: If there is a consumer of LDI events,
5965 				 * ddi_uninitchild above would have failed
5966 				 * because of active devi_ref from ldi_open().
5967 				 */
5968 
5969 				if (DEVI_EVREMOVE(dip)) {
5970 					path = i_ddi_strdup(
5971 					    DEVI(dip)->devi_ev_path,
5972 					    KM_SLEEP);
5973 					class =
5974 					    i_ddi_strdup(i_ddi_devi_class(dip),
5975 					    KM_SLEEP);
5976 					driver =
5977 					    i_ddi_strdup(
5978 					    (char *)ddi_driver_name(dip),
5979 					    KM_SLEEP);
5980 					instance = DEVI(dip)->devi_ev_instance;
5981 					post_event = 1;
5982 				}
5983 
5984 				ret = ddi_remove_child(dip, 0);
5985 				if (post_event && ret == NDI_SUCCESS) {
5986 					/* Generate EC_DEVFS_DEVI_REMOVE */
5987 					(void) i_log_devfs_remove_devinfo(path,
5988 					    class, driver, instance, flags);
5989 				}
5990 			}
5991 
5992 		}
5993 	}
5994 
5995 	if (path)
5996 		strfree(path);
5997 	if (class)
5998 		strfree(class);
5999 	if (driver)
6000 		strfree(driver);
6001 
6002 	return (ret);
6003 }
6004 
6005 /*
6006  * unconfigure immediate children of bus nexus device
6007  */
6008 static int
6009 unconfig_immediate_children(
6010 	dev_info_t *dip,
6011 	dev_info_t **dipp,
6012 	int flags,
6013 	major_t major)
6014 {
6015 	int rv = NDI_SUCCESS;
6016 	int circ, vcirc;
6017 	dev_info_t *child;
6018 	dev_info_t *vdip = NULL;
6019 	dev_info_t *next;
6020 
6021 	ASSERT(dipp == NULL || *dipp == NULL);
6022 
6023 	/*
6024 	 * Scan forward to see if we will be processing a pHCI child. If we
6025 	 * have a child that is a pHCI and vHCI and pHCI are not siblings then
6026 	 * enter vHCI before parent(pHCI) to prevent deadlock with mpxio
6027 	 * Client power management operations.
6028 	 */
6029 	ndi_devi_enter(dip, &circ);
6030 	for (child = ddi_get_child(dip); child;
6031 	    child = ddi_get_next_sibling(child)) {
6032 		/* skip same nodes we skip below */
6033 		if (((major != DDI_MAJOR_T_NONE) &&
6034 		    (major != ddi_driver_major(child))) ||
6035 		    ((flags & NDI_AUTODETACH) && !is_leaf_node(child)))
6036 			continue;
6037 
6038 		if (MDI_PHCI(child)) {
6039 			vdip = mdi_devi_get_vdip(child);
6040 			/*
6041 			 * If vHCI and vHCI is not a sibling of pHCI
6042 			 * then enter in (vHCI, parent(pHCI)) order.
6043 			 */
6044 			if (vdip && (ddi_get_parent(vdip) != dip)) {
6045 				ndi_devi_exit(dip, circ);
6046 
6047 				/* use mdi_devi_enter ordering */
6048 				ndi_devi_enter(vdip, &vcirc);
6049 				ndi_devi_enter(dip, &circ);
6050 				break;
6051 			} else
6052 				vdip = NULL;
6053 		}
6054 	}
6055 
6056 	child = ddi_get_child(dip);
6057 	while (child) {
6058 		next = ddi_get_next_sibling(child);
6059 
6060 		if ((major != DDI_MAJOR_T_NONE) &&
6061 		    (major != ddi_driver_major(child))) {
6062 			child = next;
6063 			continue;
6064 		}
6065 
6066 		/* skip nexus nodes during autodetach */
6067 		if ((flags & NDI_AUTODETACH) && !is_leaf_node(child)) {
6068 			child = next;
6069 			continue;
6070 		}
6071 
6072 		if (devi_detach_node(child, flags) != NDI_SUCCESS) {
6073 			if (dipp && *dipp == NULL) {
6074 				ndi_hold_devi(child);
6075 				*dipp = child;
6076 			}
6077 			rv = NDI_FAILURE;
6078 		}
6079 
6080 		/*
6081 		 * Continue upon failure--best effort algorithm
6082 		 */
6083 		child = next;
6084 	}
6085 
6086 	ndi_devi_exit(dip, circ);
6087 	if (vdip)
6088 		ndi_devi_exit(vdip, vcirc);
6089 
6090 	return (rv);
6091 }
6092 
6093 /*
6094  * unconfigure grand children of bus nexus device
6095  */
6096 static int
6097 unconfig_grand_children(
6098 	dev_info_t *dip,
6099 	dev_info_t **dipp,
6100 	int flags,
6101 	major_t major,
6102 	struct brevq_node **brevqp)
6103 {
6104 	struct mt_config_handle *hdl;
6105 
6106 	if (brevqp)
6107 		*brevqp = NULL;
6108 
6109 	/* multi-threaded configuration of child nexus */
6110 	hdl = mt_config_init(dip, dipp, flags, major, MT_UNCONFIG_OP, brevqp);
6111 	mt_config_children(hdl);
6112 
6113 	return (mt_config_fini(hdl));	/* wait for threads to exit */
6114 }
6115 
6116 /*
6117  * Unconfigure children/descendants of the dip.
6118  *
6119  * If brevqp is not NULL, on return *brevqp is set to a queue of dip's
6120  * child devinames for which branch remove events need to be generated.
6121  */
6122 static int
6123 devi_unconfig_common(
6124 	dev_info_t *dip,
6125 	dev_info_t **dipp,
6126 	int flags,
6127 	major_t major,
6128 	struct brevq_node **brevqp)
6129 {
6130 	int rv;
6131 	int pm_cookie;
6132 	int (*f)();
6133 	ddi_bus_config_op_t bus_op;
6134 
6135 	if (dipp)
6136 		*dipp = NULL;
6137 	if (brevqp)
6138 		*brevqp = NULL;
6139 
6140 	/*
6141 	 * Power up the dip if it is powered off.  If the flag bit
6142 	 * NDI_AUTODETACH is set and the dip is not at its full power,
6143 	 * skip the rest of the branch.
6144 	 */
6145 	if (pm_pre_unconfig(dip, flags, &pm_cookie, NULL) != DDI_SUCCESS)
6146 		return ((flags & NDI_AUTODETACH) ? NDI_SUCCESS :
6147 		    NDI_FAILURE);
6148 
6149 	/*
6150 	 * Some callers, notably SCSI, need to clear out the devfs
6151 	 * cache together with the unconfig to prevent stale entries.
6152 	 */
6153 	if (flags & NDI_DEVFS_CLEAN)
6154 		(void) devfs_clean(dip, NULL, 0);
6155 
6156 	rv = unconfig_grand_children(dip, dipp, flags, major, brevqp);
6157 
6158 	if ((rv != NDI_SUCCESS) && ((flags & NDI_AUTODETACH) == 0)) {
6159 		if (brevqp && *brevqp) {
6160 			log_and_free_br_events_on_grand_children(dip, *brevqp);
6161 			free_brevq(*brevqp);
6162 			*brevqp = NULL;
6163 		}
6164 		pm_post_unconfig(dip, pm_cookie, NULL);
6165 		return (rv);
6166 	}
6167 
6168 	if (dipp && *dipp) {
6169 		ndi_rele_devi(*dipp);
6170 		*dipp = NULL;
6171 	}
6172 
6173 	/*
6174 	 * It is possible to have a detached nexus with children
6175 	 * and grandchildren (for example: a branch consisting
6176 	 * entirely of bound nodes.) Since the nexus is detached
6177 	 * the bus_unconfig entry point cannot be used to remove
6178 	 * or unconfigure the descendants.
6179 	 */
6180 	if (!i_ddi_devi_attached(dip) ||
6181 	    (DEVI(dip)->devi_ops->devo_bus_ops == NULL) ||
6182 	    (DEVI(dip)->devi_ops->devo_bus_ops->busops_rev < BUSO_REV_5) ||
6183 	    (f = DEVI(dip)->devi_ops->devo_bus_ops->bus_unconfig) == NULL) {
6184 		rv = unconfig_immediate_children(dip, dipp, flags, major);
6185 	} else {
6186 		/*
6187 		 * call bus_unconfig entry point
6188 		 * It should reset nexus flags if unconfigure succeeds.
6189 		 */
6190 		bus_op = (major == DDI_MAJOR_T_NONE) ?
6191 		    BUS_UNCONFIG_ALL : BUS_UNCONFIG_DRIVER;
6192 		rv = (*f)(dip, flags, bus_op, (void *)(uintptr_t)major);
6193 	}
6194 
6195 	pm_post_unconfig(dip, pm_cookie, NULL);
6196 
6197 	if (brevqp && *brevqp)
6198 		cleanup_br_events_on_grand_children(dip, brevqp);
6199 
6200 	return (rv);
6201 }
6202 
6203 /*
6204  * called by devfs/framework to unconfigure children bound to major
6205  * If NDI_AUTODETACH is specified, this is invoked by either the
6206  * moduninstall daemon or the modunload -i 0 command.
6207  */
6208 int
6209 ndi_devi_unconfig_driver(dev_info_t *dip, int flags, major_t major)
6210 {
6211 	NDI_CONFIG_DEBUG((CE_CONT,
6212 	    "ndi_devi_unconfig_driver: par = %s%d (%p), flags = 0x%x\n",
6213 	    ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip, flags));
6214 
6215 	return (devi_unconfig_common(dip, NULL, flags, major, NULL));
6216 }
6217 
6218 int
6219 ndi_devi_unconfig(dev_info_t *dip, int flags)
6220 {
6221 	NDI_CONFIG_DEBUG((CE_CONT,
6222 	    "ndi_devi_unconfig: par = %s%d (%p), flags = 0x%x\n",
6223 	    ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip, flags));
6224 
6225 	return (devi_unconfig_common(dip, NULL, flags, DDI_MAJOR_T_NONE, NULL));
6226 }
6227 
6228 int
6229 e_ddi_devi_unconfig(dev_info_t *dip, dev_info_t **dipp, int flags)
6230 {
6231 	NDI_CONFIG_DEBUG((CE_CONT,
6232 	    "e_ddi_devi_unconfig: par = %s%d (%p), flags = 0x%x\n",
6233 	    ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip, flags));
6234 
6235 	return (devi_unconfig_common(dip, dipp, flags, DDI_MAJOR_T_NONE, NULL));
6236 }
6237 
6238 /*
6239  * Unconfigure child by name
6240  */
6241 static int
6242 devi_unconfig_one(dev_info_t *pdip, char *devnm, int flags)
6243 {
6244 	int		rv, circ;
6245 	dev_info_t	*child;
6246 	dev_info_t	*vdip = NULL;
6247 	int		v_circ;
6248 
6249 	ndi_devi_enter(pdip, &circ);
6250 	child = ndi_devi_findchild(pdip, devnm);
6251 
6252 	/*
6253 	 * If child is pHCI and vHCI and pHCI are not siblings then enter vHCI
6254 	 * before parent(pHCI) to avoid deadlock with mpxio Client power
6255 	 * management operations.
6256 	 */
6257 	if (child && MDI_PHCI(child)) {
6258 		vdip = mdi_devi_get_vdip(child);
6259 		if (vdip && (ddi_get_parent(vdip) != pdip)) {
6260 			ndi_devi_exit(pdip, circ);
6261 
6262 			/* use mdi_devi_enter ordering */
6263 			ndi_devi_enter(vdip, &v_circ);
6264 			ndi_devi_enter(pdip, &circ);
6265 			child = ndi_devi_findchild(pdip, devnm);
6266 		} else
6267 			vdip = NULL;
6268 	}
6269 
6270 	if (child) {
6271 		rv = devi_detach_node(child, flags);
6272 	} else {
6273 		NDI_CONFIG_DEBUG((CE_CONT,
6274 		    "devi_unconfig_one: %s not found\n", devnm));
6275 		rv = NDI_SUCCESS;
6276 	}
6277 
6278 	ndi_devi_exit(pdip, circ);
6279 	if (vdip)
6280 		ndi_devi_exit(vdip, v_circ);
6281 
6282 	return (rv);
6283 }
6284 
6285 int
6286 ndi_devi_unconfig_one(
6287 	dev_info_t *pdip,
6288 	char *devnm,
6289 	dev_info_t **dipp,
6290 	int flags)
6291 {
6292 	int		(*f)();
6293 	int		circ, rv;
6294 	int		pm_cookie;
6295 	dev_info_t	*child;
6296 	dev_info_t	*vdip = NULL;
6297 	int		v_circ;
6298 	struct brevq_node *brevq = NULL;
6299 
6300 	ASSERT(i_ddi_devi_attached(pdip));
6301 
6302 	NDI_CONFIG_DEBUG((CE_CONT,
6303 	    "ndi_devi_unconfig_one: par = %s%d (%p), child = %s\n",
6304 	    ddi_driver_name(pdip), ddi_get_instance(pdip),
6305 	    (void *)pdip, devnm));
6306 
6307 	if (pm_pre_unconfig(pdip, flags, &pm_cookie, devnm) != DDI_SUCCESS)
6308 		return (NDI_FAILURE);
6309 
6310 	if (dipp)
6311 		*dipp = NULL;
6312 
6313 	ndi_devi_enter(pdip, &circ);
6314 	child = ndi_devi_findchild(pdip, devnm);
6315 
6316 	/*
6317 	 * If child is pHCI and vHCI and pHCI are not siblings then enter vHCI
6318 	 * before parent(pHCI) to avoid deadlock with mpxio Client power
6319 	 * management operations.
6320 	 */
6321 	if (child && MDI_PHCI(child)) {
6322 		vdip = mdi_devi_get_vdip(child);
6323 		if (vdip && (ddi_get_parent(vdip) != pdip)) {
6324 			ndi_devi_exit(pdip, circ);
6325 
6326 			/* use mdi_devi_enter ordering */
6327 			ndi_devi_enter(vdip, &v_circ);
6328 			ndi_devi_enter(pdip, &circ);
6329 			child = ndi_devi_findchild(pdip, devnm);
6330 		} else
6331 			vdip = NULL;
6332 	}
6333 
6334 	if (child == NULL) {
6335 		NDI_CONFIG_DEBUG((CE_CONT, "ndi_devi_unconfig_one: %s"
6336 		    " not found\n", devnm));
6337 		rv = NDI_SUCCESS;
6338 		goto out;
6339 	}
6340 
6341 	/*
6342 	 * Unconfigure children/descendants of named child
6343 	 */
6344 	rv = devi_unconfig_branch(child, dipp, flags | NDI_UNCONFIG, &brevq);
6345 	if (rv != NDI_SUCCESS)
6346 		goto out;
6347 
6348 	init_bound_node_ev(pdip, child, flags);
6349 
6350 	if ((DEVI(pdip)->devi_ops->devo_bus_ops == NULL) ||
6351 	    (DEVI(pdip)->devi_ops->devo_bus_ops->busops_rev < BUSO_REV_5) ||
6352 	    (f = DEVI(pdip)->devi_ops->devo_bus_ops->bus_unconfig) == NULL) {
6353 		rv = devi_detach_node(child, flags);
6354 	} else {
6355 		/* call bus_config entry point */
6356 		rv = (*f)(pdip, flags, BUS_UNCONFIG_ONE, (void *)devnm);
6357 	}
6358 
6359 	if (brevq) {
6360 		if (rv != NDI_SUCCESS)
6361 			log_and_free_brevq_dip(child, brevq);
6362 		else
6363 			free_brevq(brevq);
6364 	}
6365 
6366 	if (dipp && rv != NDI_SUCCESS) {
6367 		ndi_hold_devi(child);
6368 		ASSERT(*dipp == NULL);
6369 		*dipp = child;
6370 	}
6371 
6372 out:
6373 	ndi_devi_exit(pdip, circ);
6374 	if (vdip)
6375 		ndi_devi_exit(vdip, v_circ);
6376 
6377 	pm_post_unconfig(pdip, pm_cookie, devnm);
6378 
6379 	return (rv);
6380 }
6381 
6382 struct async_arg {
6383 	dev_info_t *dip;
6384 	uint_t flags;
6385 };
6386 
6387 /*
6388  * Common async handler for:
6389  *	ndi_devi_bind_driver_async
6390  *	ndi_devi_online_async
6391  */
6392 static int
6393 i_ndi_devi_async_common(dev_info_t *dip, uint_t flags, void (*func)())
6394 {
6395 	int tqflag;
6396 	int kmflag;
6397 	struct async_arg *arg;
6398 	dev_info_t *pdip = ddi_get_parent(dip);
6399 
6400 	ASSERT(pdip);
6401 	ASSERT(DEVI(pdip)->devi_taskq);
6402 	ASSERT(ndi_dev_is_persistent_node(dip));
6403 
6404 	if (flags & NDI_NOSLEEP) {
6405 		kmflag = KM_NOSLEEP;
6406 		tqflag = TQ_NOSLEEP;
6407 	} else {
6408 		kmflag = KM_SLEEP;
6409 		tqflag = TQ_SLEEP;
6410 	}
6411 
6412 	arg = kmem_alloc(sizeof (*arg), kmflag);
6413 	if (arg == NULL)
6414 		goto fail;
6415 
6416 	arg->flags = flags;
6417 	arg->dip = dip;
6418 	if (ddi_taskq_dispatch(DEVI(pdip)->devi_taskq, func, arg, tqflag) ==
6419 	    DDI_SUCCESS) {
6420 		return (NDI_SUCCESS);
6421 	}
6422 
6423 fail:
6424 	NDI_CONFIG_DEBUG((CE_CONT, "%s%d: ddi_taskq_dispatch failed",
6425 	    ddi_driver_name(pdip), ddi_get_instance(pdip)));
6426 
6427 	if (arg)
6428 		kmem_free(arg, sizeof (*arg));
6429 	return (NDI_FAILURE);
6430 }
6431 
6432 static void
6433 i_ndi_devi_bind_driver_cb(struct async_arg *arg)
6434 {
6435 	(void) ndi_devi_bind_driver(arg->dip, arg->flags);
6436 	kmem_free(arg, sizeof (*arg));
6437 }
6438 
6439 int
6440 ndi_devi_bind_driver_async(dev_info_t *dip, uint_t flags)
6441 {
6442 	return (i_ndi_devi_async_common(dip, flags,
6443 	    (void (*)())i_ndi_devi_bind_driver_cb));
6444 }
6445 
6446 /*
6447  * place the devinfo in the ONLINE state.
6448  */
6449 int
6450 ndi_devi_online(dev_info_t *dip, uint_t flags)
6451 {
6452 	int circ, rv;
6453 	dev_info_t *pdip = ddi_get_parent(dip);
6454 	int branch_event = 0;
6455 
6456 	ASSERT(pdip);
6457 
6458 	NDI_CONFIG_DEBUG((CE_CONT, "ndi_devi_online: %s%d (%p)\n",
6459 	    ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip));
6460 
6461 	ndi_devi_enter(pdip, &circ);
6462 	/* bind child before merging .conf nodes */
6463 	rv = i_ndi_config_node(dip, DS_BOUND, flags);
6464 	if (rv != NDI_SUCCESS) {
6465 		ndi_devi_exit(pdip, circ);
6466 		return (rv);
6467 	}
6468 
6469 	/* merge .conf properties */
6470 	(void) i_ndi_make_spec_children(pdip, flags);
6471 
6472 	flags |= (NDI_DEVI_ONLINE | NDI_CONFIG);
6473 
6474 	if (flags & NDI_NO_EVENT) {
6475 		/*
6476 		 * Caller is specifically asking for not to generate an event.
6477 		 * Set the following flag so that devi_attach_node() don't
6478 		 * change the event state.
6479 		 */
6480 		flags |= NDI_NO_EVENT_STATE_CHNG;
6481 	}
6482 
6483 	if ((flags & (NDI_NO_EVENT | NDI_BRANCH_EVENT_OP)) == 0 &&
6484 	    ((flags & NDI_CONFIG) || DEVI_NEED_NDI_CONFIG(dip))) {
6485 		flags |= NDI_BRANCH_EVENT_OP;
6486 		branch_event = 1;
6487 	}
6488 
6489 	/*
6490 	 * devi_attach_node() may remove dip on failure
6491 	 */
6492 	if ((rv = devi_attach_node(dip, flags)) == NDI_SUCCESS) {
6493 		if ((flags & NDI_CONFIG) || DEVI_NEED_NDI_CONFIG(dip)) {
6494 			/*
6495 			 * Hold the attached dip, and exit the parent while
6496 			 * we drive configuration of children below the
6497 			 * attached dip.
6498 			 */
6499 			ndi_hold_devi(dip);
6500 			ndi_devi_exit(pdip, circ);
6501 
6502 			(void) ndi_devi_config(dip, flags);
6503 
6504 			ndi_devi_enter(pdip, &circ);
6505 			ndi_rele_devi(dip);
6506 		}
6507 
6508 		if (branch_event)
6509 			(void) i_log_devfs_branch_add(dip);
6510 	}
6511 
6512 	ndi_devi_exit(pdip, circ);
6513 
6514 	/*
6515 	 * Notify devfs that we have a new node. Devfs needs to invalidate
6516 	 * cached directory contents.
6517 	 *
6518 	 * For PCMCIA devices, it is possible the pdip is not fully
6519 	 * attached. In this case, calling back into devfs will
6520 	 * result in a loop or assertion error. Hence, the check
6521 	 * on node state.
6522 	 *
6523 	 * If we own parent lock, this is part of a branch operation.
6524 	 * We skip the devfs_clean() step because the cache invalidation
6525 	 * is done higher up in the device tree.
6526 	 */
6527 	if (rv == NDI_SUCCESS && i_ddi_devi_attached(pdip) &&
6528 	    !DEVI_BUSY_OWNED(pdip))
6529 		(void) devfs_clean(pdip, NULL, 0);
6530 	return (rv);
6531 }
6532 
6533 static void
6534 i_ndi_devi_online_cb(struct async_arg *arg)
6535 {
6536 	(void) ndi_devi_online(arg->dip, arg->flags);
6537 	kmem_free(arg, sizeof (*arg));
6538 }
6539 
6540 int
6541 ndi_devi_online_async(dev_info_t *dip, uint_t flags)
6542 {
6543 	/* mark child as need config if requested. */
6544 	if (flags & NDI_CONFIG) {
6545 		mutex_enter(&(DEVI(dip)->devi_lock));
6546 		DEVI_SET_NDI_CONFIG(dip);
6547 		mutex_exit(&(DEVI(dip)->devi_lock));
6548 	}
6549 
6550 	return (i_ndi_devi_async_common(dip, flags,
6551 	    (void (*)())i_ndi_devi_online_cb));
6552 }
6553 
6554 /*
6555  * Take a device node Offline
6556  * To take a device Offline means to detach the device instance from
6557  * the driver and prevent devfs requests from re-attaching the device
6558  * instance.
6559  *
6560  * The flag NDI_DEVI_REMOVE causes removes the device node from
6561  * the driver list and the device tree. In this case, the device
6562  * is assumed to be removed from the system.
6563  */
6564 int
6565 ndi_devi_offline(dev_info_t *dip, uint_t flags)
6566 {
6567 	int		circ, rval = 0;
6568 	dev_info_t	*pdip = ddi_get_parent(dip);
6569 	dev_info_t	*vdip = NULL;
6570 	int		v_circ;
6571 	struct brevq_node *brevq = NULL;
6572 
6573 	ASSERT(pdip);
6574 
6575 	flags |= NDI_DEVI_OFFLINE;
6576 
6577 	/*
6578 	 * If child is pHCI and vHCI and pHCI are not siblings then enter vHCI
6579 	 * before parent(pHCI) to avoid deadlock with mpxio Client power
6580 	 * management operations.
6581 	 */
6582 	if (MDI_PHCI(dip)) {
6583 		vdip = mdi_devi_get_vdip(dip);
6584 		if (vdip && (ddi_get_parent(vdip) != pdip))
6585 			ndi_devi_enter(vdip, &v_circ);
6586 		else
6587 			vdip = NULL;
6588 	}
6589 	ndi_devi_enter(pdip, &circ);
6590 
6591 	if (i_ddi_devi_attached(dip)) {
6592 		/*
6593 		 * If dip is in DS_READY state, there may be cached dv_nodes
6594 		 * referencing this dip, so we invoke devfs code path.
6595 		 * Note that we must release busy changing on pdip to
6596 		 * avoid deadlock against devfs.
6597 		 */
6598 		char *devname = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
6599 		(void) ddi_deviname(dip, devname);
6600 
6601 		ndi_devi_exit(pdip, circ);
6602 		if (vdip)
6603 			ndi_devi_exit(vdip, v_circ);
6604 
6605 		/*
6606 		 * If we are explictly told to clean, then clean. If we own the
6607 		 * parent lock then this is part of a branch operation, and we
6608 		 * skip the devfs_clean() step.
6609 		 *
6610 		 * NOTE: A thread performing a devfs file system lookup/
6611 		 * bus_config can't call devfs_clean to unconfig without
6612 		 * causing rwlock problems in devfs. For ndi_devi_offline, this
6613 		 * means that the NDI_DEVFS_CLEAN flag is safe from ioctl code
6614 		 * or from an async hotplug thread, but is not safe from a
6615 		 * nexus driver's bus_config implementation.
6616 		 */
6617 		if ((flags & NDI_DEVFS_CLEAN) ||
6618 		    (!DEVI_BUSY_OWNED(pdip)))
6619 			(void) devfs_clean(pdip, devname + 1, DV_CLEAN_FORCE);
6620 
6621 		kmem_free(devname, MAXNAMELEN + 1);
6622 
6623 		rval = devi_unconfig_branch(dip, NULL, flags|NDI_UNCONFIG,
6624 		    &brevq);
6625 
6626 		if (rval)
6627 			return (NDI_FAILURE);
6628 
6629 		if (vdip)
6630 			ndi_devi_enter(vdip, &v_circ);
6631 		ndi_devi_enter(pdip, &circ);
6632 	}
6633 
6634 	init_bound_node_ev(pdip, dip, flags);
6635 
6636 	rval = devi_detach_node(dip, flags);
6637 	if (brevq) {
6638 		if (rval != NDI_SUCCESS)
6639 			log_and_free_brevq_dip(dip, brevq);
6640 		else
6641 			free_brevq(brevq);
6642 	}
6643 
6644 	ndi_devi_exit(pdip, circ);
6645 	if (vdip)
6646 		ndi_devi_exit(vdip, v_circ);
6647 
6648 	return (rval);
6649 }
6650 
6651 /*
6652  * Find the child dev_info node of parent nexus 'p' whose unit address
6653  * matches "cname@caddr".  Recommend use of ndi_devi_findchild() instead.
6654  */
6655 dev_info_t *
6656 ndi_devi_find(dev_info_t *pdip, char *cname, char *caddr)
6657 {
6658 	dev_info_t *child;
6659 	int circ;
6660 
6661 	if (pdip == NULL || cname == NULL || caddr == NULL)
6662 		return ((dev_info_t *)NULL);
6663 
6664 	ndi_devi_enter(pdip, &circ);
6665 	child = find_sibling(ddi_get_child(pdip), cname, caddr,
6666 	    FIND_NODE_BY_NODENAME, NULL);
6667 	ndi_devi_exit(pdip, circ);
6668 	return (child);
6669 }
6670 
6671 /*
6672  * Find the child dev_info node of parent nexus 'p' whose unit address
6673  * matches devname "name@addr".  Permits caller to hold the parent.
6674  */
6675 dev_info_t *
6676 ndi_devi_findchild(dev_info_t *pdip, char *devname)
6677 {
6678 	dev_info_t *child;
6679 	char	*cname, *caddr;
6680 	char	*devstr;
6681 
6682 	ASSERT(DEVI_BUSY_OWNED(pdip));
6683 
6684 	devstr = i_ddi_strdup(devname, KM_SLEEP);
6685 	i_ddi_parse_name(devstr, &cname, &caddr, NULL);
6686 
6687 	if (cname == NULL || caddr == NULL) {
6688 		kmem_free(devstr, strlen(devname)+1);
6689 		return ((dev_info_t *)NULL);
6690 	}
6691 
6692 	child = find_sibling(ddi_get_child(pdip), cname, caddr,
6693 	    FIND_NODE_BY_NODENAME, NULL);
6694 	kmem_free(devstr, strlen(devname)+1);
6695 	return (child);
6696 }
6697 
6698 /*
6699  * Misc. routines called by framework only
6700  */
6701 
6702 /*
6703  * Clear the DEVI_MADE_CHILDREN/DEVI_ATTACHED_CHILDREN flags
6704  * if new child spec has been added.
6705  */
6706 static int
6707 reset_nexus_flags(dev_info_t *dip, void *arg)
6708 {
6709 	struct hwc_spec	*list;
6710 	int		circ;
6711 
6712 	if (((DEVI(dip)->devi_flags & DEVI_MADE_CHILDREN) == 0) ||
6713 	    ((list = hwc_get_child_spec(dip, (major_t)(uintptr_t)arg)) == NULL))
6714 		return (DDI_WALK_CONTINUE);
6715 
6716 	hwc_free_spec_list(list);
6717 
6718 	/* coordinate child state update */
6719 	ndi_devi_enter(dip, &circ);
6720 	mutex_enter(&DEVI(dip)->devi_lock);
6721 	DEVI(dip)->devi_flags &= ~(DEVI_MADE_CHILDREN | DEVI_ATTACHED_CHILDREN);
6722 	mutex_exit(&DEVI(dip)->devi_lock);
6723 	ndi_devi_exit(dip, circ);
6724 
6725 	return (DDI_WALK_CONTINUE);
6726 }
6727 
6728 /*
6729  * Helper functions, returns NULL if no memory.
6730  */
6731 
6732 /*
6733  * path_to_major:
6734  *
6735  * Return an alternate driver name binding for the leaf device
6736  * of the given pathname, if there is one. The purpose of this
6737  * function is to deal with generic pathnames. The default action
6738  * for platforms that can't do this (ie: x86 or any platform that
6739  * does not have prom_finddevice functionality, which matches
6740  * nodenames and unit-addresses without the drivers participation)
6741  * is to return DDI_MAJOR_T_NONE.
6742  *
6743  * Used in loadrootmodules() in the swapgeneric module to
6744  * associate a given pathname with a given leaf driver.
6745  *
6746  */
6747 major_t
6748 path_to_major(char *path)
6749 {
6750 	dev_info_t *dip;
6751 	char *p, *q;
6752 	pnode_t nodeid;
6753 	major_t major;
6754 
6755 	/* check for path-oriented alias */
6756 	major = ddi_name_to_major(path);
6757 	if (driver_active(major)) {
6758 		NDI_CONFIG_DEBUG((CE_NOTE, "path_to_major: %s path bound %s\n",
6759 		    path, ddi_major_to_name(major)));
6760 		return (major);
6761 	}
6762 
6763 	/*
6764 	 * Get the nodeid of the given pathname, if such a mapping exists.
6765 	 */
6766 	dip = NULL;
6767 	nodeid = prom_finddevice(path);
6768 	if (nodeid != OBP_BADNODE) {
6769 		/*
6770 		 * Find the nodeid in our copy of the device tree and return
6771 		 * whatever name we used to bind this node to a driver.
6772 		 */
6773 		dip = e_ddi_nodeid_to_dip(nodeid);
6774 	}
6775 
6776 	if (dip == NULL) {
6777 		NDI_CONFIG_DEBUG((CE_WARN,
6778 		    "path_to_major: can't bind <%s>\n", path));
6779 		return (DDI_MAJOR_T_NONE);
6780 	}
6781 
6782 	/*
6783 	 * If we're bound to something other than the nodename,
6784 	 * note that in the message buffer and system log.
6785 	 */
6786 	p = ddi_binding_name(dip);
6787 	q = ddi_node_name(dip);
6788 	if (p && q && (strcmp(p, q) != 0))
6789 		NDI_CONFIG_DEBUG((CE_NOTE, "path_to_major: %s bound to %s\n",
6790 		    path, p));
6791 
6792 	major = ddi_name_to_major(p);
6793 
6794 	ndi_rele_devi(dip);		/* release e_ddi_nodeid_to_dip hold */
6795 
6796 	return (major);
6797 }
6798 
6799 /*
6800  * Return the held dip for the specified major and instance, attempting to do
6801  * an attach if specified. Return NULL if the devi can't be found or put in
6802  * the proper state. The caller must release the hold via ddi_release_devi if
6803  * a non-NULL value is returned.
6804  *
6805  * Some callers expect to be able to perform a hold_devi() while in a context
6806  * where using ndi_devi_enter() to ensure the hold might cause deadlock (see
6807  * open-from-attach code in consconfig_dacf.c). Such special-case callers
6808  * must ensure that an ndi_devi_enter(parent)/ndi_hold_devi() from a safe
6809  * context is already active. The hold_devi() implementation must accommodate
6810  * these callers.
6811  */
6812 static dev_info_t *
6813 hold_devi(major_t major, int instance, int flags)
6814 {
6815 	struct devnames	*dnp;
6816 	dev_info_t	*dip;
6817 	char		*path;
6818 	char		*vpath;
6819 
6820 	if ((major >= devcnt) || (instance == -1))
6821 		return (NULL);
6822 
6823 	/* try to find the instance in the per driver list */
6824 	dnp = &(devnamesp[major]);
6825 	LOCK_DEV_OPS(&(dnp->dn_lock));
6826 	for (dip = dnp->dn_head; dip;
6827 	    dip = (dev_info_t *)DEVI(dip)->devi_next) {
6828 		/* skip node if instance field is not valid */
6829 		if (i_ddi_node_state(dip) < DS_INITIALIZED)
6830 			continue;
6831 
6832 		/* look for instance match */
6833 		if (DEVI(dip)->devi_instance == instance) {
6834 			/*
6835 			 * To accommodate callers that can't block in
6836 			 * ndi_devi_enter() we do an ndi_hold_devi(), and
6837 			 * afterwards check that the node is in a state where
6838 			 * the hold prevents detach(). If we did not manage to
6839 			 * prevent detach then we ndi_rele_devi() and perform
6840 			 * the slow path below (which can result in a blocking
6841 			 * ndi_devi_enter() while driving attach top-down).
6842 			 * This code depends on the ordering of
6843 			 * DEVI_SET_DETACHING and the devi_ref check in the
6844 			 * detach_node() code path.
6845 			 */
6846 			ndi_hold_devi(dip);
6847 			if (i_ddi_devi_attached(dip) &&
6848 			    !DEVI_IS_DETACHING(dip)) {
6849 				UNLOCK_DEV_OPS(&(dnp->dn_lock));
6850 				return (dip);	/* fast-path with devi held */
6851 			}
6852 			ndi_rele_devi(dip);
6853 
6854 			/* try slow-path */
6855 			dip = NULL;
6856 			break;
6857 		}
6858 	}
6859 	ASSERT(dip == NULL);
6860 	UNLOCK_DEV_OPS(&(dnp->dn_lock));
6861 
6862 	if (flags & E_DDI_HOLD_DEVI_NOATTACH)
6863 		return (NULL);		/* told not to drive attach */
6864 
6865 	/* slow-path may block, so it should not occur from interrupt */
6866 	ASSERT(!servicing_interrupt());
6867 	if (servicing_interrupt())
6868 		return (NULL);
6869 
6870 	/* reconstruct the path and drive attach by path through devfs. */
6871 	path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6872 	if (e_ddi_majorinstance_to_path(major, instance, path) == 0) {
6873 		dip = e_ddi_hold_devi_by_path(path, flags);
6874 
6875 		/*
6876 		 * Verify that we got the correct device - a path_to_inst file
6877 		 * with a bogus/corrupt path (or a nexus that changes its
6878 		 * unit-address format) could result in an incorrect answer
6879 		 *
6880 		 * Verify major, instance, and path.
6881 		 */
6882 		vpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6883 		if (dip &&
6884 		    ((DEVI(dip)->devi_major != major) ||
6885 		    ((DEVI(dip)->devi_instance != instance)) ||
6886 		    (strcmp(path, ddi_pathname(dip, vpath)) != 0))) {
6887 			ndi_rele_devi(dip);
6888 			dip = NULL;	/* no answer better than wrong answer */
6889 		}
6890 		kmem_free(vpath, MAXPATHLEN);
6891 	}
6892 	kmem_free(path, MAXPATHLEN);
6893 	return (dip);			/* with devi held */
6894 }
6895 
6896 /*
6897  * The {e_}ddi_hold_devi{_by_{instance|dev|path}} hold the devinfo node
6898  * associated with the specified arguments.  This hold should be released
6899  * by calling ddi_release_devi.
6900  *
6901  * The E_DDI_HOLD_DEVI_NOATTACH flag argument allows the caller to to specify
6902  * a failure return if the node is not already attached.
6903  *
6904  * NOTE: by the time we make e_ddi_hold_devi public, we should be able to reuse
6905  * ddi_hold_devi again.
6906  */
6907 dev_info_t *
6908 ddi_hold_devi_by_instance(major_t major, int instance, int flags)
6909 {
6910 	return (hold_devi(major, instance, flags));
6911 }
6912 
6913 dev_info_t *
6914 e_ddi_hold_devi_by_dev(dev_t dev, int flags)
6915 {
6916 	major_t	major = getmajor(dev);
6917 	dev_info_t	*dip;
6918 	struct dev_ops	*ops;
6919 	dev_info_t	*ddip = NULL;
6920 
6921 	dip = hold_devi(major, dev_to_instance(dev), flags);
6922 
6923 	/*
6924 	 * The rest of this routine is legacy support for drivers that
6925 	 * have broken DDI_INFO_DEVT2INSTANCE implementations but may have
6926 	 * functional DDI_INFO_DEVT2DEVINFO implementations.  This code will
6927 	 * diagnose inconsistency and, for maximum compatibility with legacy
6928 	 * drivers, give preference to the drivers DDI_INFO_DEVT2DEVINFO
6929 	 * implementation over the above derived dip based the driver's
6930 	 * DDI_INFO_DEVT2INSTANCE implementation. This legacy support should
6931 	 * be removed when DDI_INFO_DEVT2DEVINFO is deprecated.
6932 	 *
6933 	 * NOTE: The following code has a race condition. DEVT2DEVINFO
6934 	 *	returns a dip which is not held. By the time we ref ddip,
6935 	 *	it could have been freed. The saving grace is that for
6936 	 *	most drivers, the dip returned from hold_devi() is the
6937 	 *	same one as the one returned by DEVT2DEVINFO, so we are
6938 	 *	safe for drivers with the correct getinfo(9e) impl.
6939 	 */
6940 	if (((ops = ddi_hold_driver(major)) != NULL) &&
6941 	    CB_DRV_INSTALLED(ops) && ops->devo_getinfo)  {
6942 		if ((*ops->devo_getinfo)(NULL, DDI_INFO_DEVT2DEVINFO,
6943 		    (void *)dev, (void **)&ddip) != DDI_SUCCESS)
6944 			ddip = NULL;
6945 	}
6946 
6947 	/* give preference to the driver returned DEVT2DEVINFO dip */
6948 	if (ddip && (dip != ddip)) {
6949 #ifdef	DEBUG
6950 		cmn_err(CE_WARN, "%s: inconsistent getinfo(9E) implementation",
6951 		    ddi_driver_name(ddip));
6952 #endif	/* DEBUG */
6953 		ndi_hold_devi(ddip);
6954 		if (dip)
6955 			ndi_rele_devi(dip);
6956 		dip = ddip;
6957 	}
6958 
6959 	if (ops)
6960 		ddi_rele_driver(major);
6961 
6962 	return (dip);
6963 }
6964 
6965 /*
6966  * For compatibility only. Do not call this function!
6967  */
6968 dev_info_t *
6969 e_ddi_get_dev_info(dev_t dev, vtype_t type)
6970 {
6971 	dev_info_t *dip = NULL;
6972 	if (getmajor(dev) >= devcnt)
6973 		return (NULL);
6974 
6975 	switch (type) {
6976 	case VCHR:
6977 	case VBLK:
6978 		dip = e_ddi_hold_devi_by_dev(dev, 0);
6979 	default:
6980 		break;
6981 	}
6982 
6983 	/*
6984 	 * For compatibility reasons, we can only return the dip with
6985 	 * the driver ref count held. This is not a safe thing to do.
6986 	 * For certain broken third-party software, we are willing
6987 	 * to venture into unknown territory.
6988 	 */
6989 	if (dip) {
6990 		(void) ndi_hold_driver(dip);
6991 		ndi_rele_devi(dip);
6992 	}
6993 	return (dip);
6994 }
6995 
6996 dev_info_t *
6997 e_ddi_hold_devi_by_path(char *path, int flags)
6998 {
6999 	dev_info_t	*dip;
7000 
7001 	/* can't specify NOATTACH by path */
7002 	ASSERT(!(flags & E_DDI_HOLD_DEVI_NOATTACH));
7003 
7004 	return (resolve_pathname(path, &dip, NULL, NULL) ? NULL : dip);
7005 }
7006 
7007 void
7008 e_ddi_hold_devi(dev_info_t *dip)
7009 {
7010 	ndi_hold_devi(dip);
7011 }
7012 
7013 void
7014 ddi_release_devi(dev_info_t *dip)
7015 {
7016 	ndi_rele_devi(dip);
7017 }
7018 
7019 /*
7020  * Associate a streams queue with a devinfo node
7021  * NOTE: This function is called by STREAM driver's put procedure.
7022  *	It cannot block.
7023  */
7024 void
7025 ddi_assoc_queue_with_devi(queue_t *q, dev_info_t *dip)
7026 {
7027 	queue_t *rq = _RD(q);
7028 	struct stdata *stp;
7029 	vnode_t *vp;
7030 
7031 	/* set flag indicating that ddi_assoc_queue_with_devi was called */
7032 	mutex_enter(QLOCK(rq));
7033 	rq->q_flag |= _QASSOCIATED;
7034 	mutex_exit(QLOCK(rq));
7035 
7036 	/* get the vnode associated with the queue */
7037 	stp = STREAM(rq);
7038 	vp = stp->sd_vnode;
7039 	ASSERT(vp);
7040 
7041 	/* change the hardware association of the vnode */
7042 	spec_assoc_vp_with_devi(vp, dip);
7043 }
7044 
7045 /*
7046  * ddi_install_driver(name)
7047  *
7048  * Driver installation is currently a byproduct of driver loading.  This
7049  * may change.
7050  */
7051 int
7052 ddi_install_driver(char *name)
7053 {
7054 	major_t major = ddi_name_to_major(name);
7055 
7056 	if ((major == DDI_MAJOR_T_NONE) ||
7057 	    (ddi_hold_installed_driver(major) == NULL)) {
7058 		return (DDI_FAILURE);
7059 	}
7060 	ddi_rele_driver(major);
7061 	return (DDI_SUCCESS);
7062 }
7063 
7064 struct dev_ops *
7065 ddi_hold_driver(major_t major)
7066 {
7067 	return (mod_hold_dev_by_major(major));
7068 }
7069 
7070 
7071 void
7072 ddi_rele_driver(major_t major)
7073 {
7074 	mod_rele_dev_by_major(major);
7075 }
7076 
7077 
7078 /*
7079  * This is called during boot to force attachment order of special dips
7080  * dip must be referenced via ndi_hold_devi()
7081  */
7082 int
7083 i_ddi_attach_node_hierarchy(dev_info_t *dip)
7084 {
7085 	dev_info_t	*parent;
7086 	int		ret, circ;
7087 
7088 	/*
7089 	 * Recurse up until attached parent is found.
7090 	 */
7091 	if (i_ddi_devi_attached(dip))
7092 		return (DDI_SUCCESS);
7093 	parent = ddi_get_parent(dip);
7094 	if (i_ddi_attach_node_hierarchy(parent) != DDI_SUCCESS)
7095 		return (DDI_FAILURE);
7096 
7097 	/*
7098 	 * Come top-down, expanding .conf nodes under this parent
7099 	 * and driving attach.
7100 	 */
7101 	ndi_devi_enter(parent, &circ);
7102 	(void) i_ndi_make_spec_children(parent, 0);
7103 	ret = i_ddi_attachchild(dip);
7104 	ndi_devi_exit(parent, circ);
7105 
7106 	return (ret);
7107 }
7108 
7109 /* keep this function static */
7110 static int
7111 attach_driver_nodes(major_t major)
7112 {
7113 	struct devnames *dnp;
7114 	dev_info_t *dip;
7115 	int error = DDI_FAILURE;
7116 
7117 	dnp = &devnamesp[major];
7118 	LOCK_DEV_OPS(&dnp->dn_lock);
7119 	dip = dnp->dn_head;
7120 	while (dip) {
7121 		ndi_hold_devi(dip);
7122 		UNLOCK_DEV_OPS(&dnp->dn_lock);
7123 		if (i_ddi_attach_node_hierarchy(dip) == DDI_SUCCESS)
7124 			error = DDI_SUCCESS;
7125 		/*
7126 		 * Set the 'ddi-config-driver-node' property on a nexus
7127 		 * node to cause attach_driver_nodes() to configure all
7128 		 * immediate children of the nexus. This property should
7129 		 * be set on nodes with immediate children that bind to
7130 		 * the same driver as parent.
7131 		 */
7132 		if ((error == DDI_SUCCESS) && (ddi_prop_exists(DDI_DEV_T_ANY,
7133 		    dip, DDI_PROP_DONTPASS, "ddi-config-driver-node"))) {
7134 			(void) ndi_devi_config(dip, NDI_NO_EVENT);
7135 		}
7136 		LOCK_DEV_OPS(&dnp->dn_lock);
7137 		ndi_rele_devi(dip);
7138 		dip = ddi_get_next(dip);
7139 	}
7140 	if (error == DDI_SUCCESS)
7141 		dnp->dn_flags |= DN_NO_AUTODETACH;
7142 	UNLOCK_DEV_OPS(&dnp->dn_lock);
7143 
7144 
7145 	return (error);
7146 }
7147 
7148 /*
7149  * i_ddi_attach_hw_nodes configures and attaches all hw nodes
7150  * bound to a specific driver. This function replaces calls to
7151  * ddi_hold_installed_driver() for drivers with no .conf
7152  * enumerated nodes.
7153  *
7154  * This facility is typically called at boot time to attach
7155  * platform-specific hardware nodes, such as ppm nodes on xcal
7156  * and grover and keyswitch nodes on cherrystone. It does not
7157  * deal with .conf enumerated node. Calling it beyond the boot
7158  * process is strongly discouraged.
7159  */
7160 int
7161 i_ddi_attach_hw_nodes(char *driver)
7162 {
7163 	major_t major;
7164 
7165 	major = ddi_name_to_major(driver);
7166 	if (major == DDI_MAJOR_T_NONE)
7167 		return (DDI_FAILURE);
7168 
7169 	return (attach_driver_nodes(major));
7170 }
7171 
7172 /*
7173  * i_ddi_attach_pseudo_node configures pseudo drivers which
7174  * has a single node. The .conf nodes must be enumerated
7175  * before calling this interface. The dip is held attached
7176  * upon returning.
7177  *
7178  * This facility should only be called only at boot time
7179  * by the I/O framework.
7180  */
7181 dev_info_t *
7182 i_ddi_attach_pseudo_node(char *driver)
7183 {
7184 	major_t major;
7185 	dev_info_t *dip;
7186 
7187 	major = ddi_name_to_major(driver);
7188 	if (major == DDI_MAJOR_T_NONE)
7189 		return (NULL);
7190 
7191 	if (attach_driver_nodes(major) != DDI_SUCCESS)
7192 		return (NULL);
7193 
7194 	dip = devnamesp[major].dn_head;
7195 	ASSERT(dip && ddi_get_next(dip) == NULL);
7196 	ndi_hold_devi(dip);
7197 	return (dip);
7198 }
7199 
7200 static void
7201 diplist_to_parent_major(dev_info_t *head, char parents[])
7202 {
7203 	major_t major;
7204 	dev_info_t *dip, *pdip;
7205 
7206 	for (dip = head; dip != NULL; dip = ddi_get_next(dip)) {
7207 		pdip = ddi_get_parent(dip);
7208 		ASSERT(pdip);	/* disallow rootnex.conf nodes */
7209 		major = ddi_driver_major(pdip);
7210 		if ((major != DDI_MAJOR_T_NONE) && parents[major] == 0)
7211 			parents[major] = 1;
7212 	}
7213 }
7214 
7215 /*
7216  * Call ddi_hold_installed_driver() on each parent major
7217  * and invoke mt_config_driver() to attach child major.
7218  * This is part of the implementation of ddi_hold_installed_driver.
7219  */
7220 static int
7221 attach_driver_by_parent(major_t child_major, char parents[])
7222 {
7223 	major_t par_major;
7224 	struct mt_config_handle *hdl;
7225 	int flags = NDI_DEVI_PERSIST | NDI_NO_EVENT;
7226 
7227 	hdl = mt_config_init(NULL, NULL, flags, child_major, MT_CONFIG_OP,
7228 	    NULL);
7229 	for (par_major = 0; par_major < devcnt; par_major++) {
7230 		/* disallow recursion on the same driver */
7231 		if (parents[par_major] == 0 || par_major == child_major)
7232 			continue;
7233 		if (ddi_hold_installed_driver(par_major) == NULL)
7234 			continue;
7235 		hdl->mtc_parmajor = par_major;
7236 		mt_config_driver(hdl);
7237 		ddi_rele_driver(par_major);
7238 	}
7239 	(void) mt_config_fini(hdl);
7240 
7241 	return (i_ddi_devs_attached(child_major));
7242 }
7243 
7244 int
7245 i_ddi_devs_attached(major_t major)
7246 {
7247 	dev_info_t *dip;
7248 	struct devnames *dnp;
7249 	int error = DDI_FAILURE;
7250 
7251 	/* check for attached instances */
7252 	dnp = &devnamesp[major];
7253 	LOCK_DEV_OPS(&dnp->dn_lock);
7254 	for (dip = dnp->dn_head; dip != NULL; dip = ddi_get_next(dip)) {
7255 		if (i_ddi_devi_attached(dip)) {
7256 			error = DDI_SUCCESS;
7257 			break;
7258 		}
7259 	}
7260 	UNLOCK_DEV_OPS(&dnp->dn_lock);
7261 
7262 	return (error);
7263 }
7264 
7265 int
7266 i_ddi_minor_node_count(dev_info_t *ddip, const char *node_type)
7267 {
7268 	int			circ;
7269 	struct ddi_minor_data	*dp;
7270 	int			count = 0;
7271 
7272 	ndi_devi_enter(ddip, &circ);
7273 	for (dp = DEVI(ddip)->devi_minor; dp != NULL; dp = dp->next) {
7274 		if (strcmp(dp->ddm_node_type, node_type) == 0)
7275 			count++;
7276 	}
7277 	ndi_devi_exit(ddip, circ);
7278 	return (count);
7279 }
7280 
7281 /*
7282  * ddi_hold_installed_driver configures and attaches all
7283  * instances of the specified driver. To accomplish this
7284  * it configures and attaches all possible parents of
7285  * the driver, enumerated both in h/w nodes and in the
7286  * driver's .conf file.
7287  *
7288  * NOTE: This facility is for compatibility purposes only and will
7289  *	eventually go away. Its usage is strongly discouraged.
7290  */
7291 static void
7292 enter_driver(struct devnames *dnp)
7293 {
7294 	mutex_enter(&dnp->dn_lock);
7295 	ASSERT(dnp->dn_busy_thread != curthread);
7296 	while (dnp->dn_flags & DN_DRIVER_BUSY)
7297 		cv_wait(&dnp->dn_wait, &dnp->dn_lock);
7298 	dnp->dn_flags |= DN_DRIVER_BUSY;
7299 	dnp->dn_busy_thread = curthread;
7300 	mutex_exit(&dnp->dn_lock);
7301 }
7302 
7303 static void
7304 exit_driver(struct devnames *dnp)
7305 {
7306 	mutex_enter(&dnp->dn_lock);
7307 	ASSERT(dnp->dn_busy_thread == curthread);
7308 	dnp->dn_flags &= ~DN_DRIVER_BUSY;
7309 	dnp->dn_busy_thread = NULL;
7310 	cv_broadcast(&dnp->dn_wait);
7311 	mutex_exit(&dnp->dn_lock);
7312 }
7313 
7314 struct dev_ops *
7315 ddi_hold_installed_driver(major_t major)
7316 {
7317 	struct dev_ops *ops;
7318 	struct devnames *dnp;
7319 	char *parents;
7320 	int error;
7321 
7322 	ops = ddi_hold_driver(major);
7323 	if (ops == NULL)
7324 		return (NULL);
7325 
7326 	/*
7327 	 * Return immediately if all the attach operations associated
7328 	 * with a ddi_hold_installed_driver() call have already been done.
7329 	 */
7330 	dnp = &devnamesp[major];
7331 	enter_driver(dnp);
7332 	ASSERT(driver_active(major));
7333 
7334 	if (dnp->dn_flags & DN_DRIVER_HELD) {
7335 		exit_driver(dnp);
7336 		if (i_ddi_devs_attached(major) == DDI_SUCCESS)
7337 			return (ops);
7338 		ddi_rele_driver(major);
7339 		return (NULL);
7340 	}
7341 
7342 	LOCK_DEV_OPS(&dnp->dn_lock);
7343 	dnp->dn_flags |= (DN_DRIVER_HELD | DN_NO_AUTODETACH);
7344 	UNLOCK_DEV_OPS(&dnp->dn_lock);
7345 
7346 	DCOMPATPRINTF((CE_CONT,
7347 	    "ddi_hold_installed_driver: %s\n", dnp->dn_name));
7348 
7349 	/*
7350 	 * When the driver has no .conf children, it is sufficient
7351 	 * to attach existing nodes in the device tree. Nodes not
7352 	 * enumerated by the OBP are not attached.
7353 	 */
7354 	if (dnp->dn_pl == NULL) {
7355 		if (attach_driver_nodes(major) == DDI_SUCCESS) {
7356 			exit_driver(dnp);
7357 			return (ops);
7358 		}
7359 		exit_driver(dnp);
7360 		ddi_rele_driver(major);
7361 		return (NULL);
7362 	}
7363 
7364 	/*
7365 	 * Driver has .conf nodes. We find all possible parents
7366 	 * and recursively all ddi_hold_installed_driver on the
7367 	 * parent driver; then we invoke ndi_config_driver()
7368 	 * on all possible parent node in parallel to speed up
7369 	 * performance.
7370 	 */
7371 	parents = kmem_zalloc(devcnt * sizeof (char), KM_SLEEP);
7372 
7373 	LOCK_DEV_OPS(&dnp->dn_lock);
7374 	/* find .conf parents */
7375 	(void) impl_parlist_to_major(dnp->dn_pl, parents);
7376 	/* find hw node parents */
7377 	diplist_to_parent_major(dnp->dn_head, parents);
7378 	UNLOCK_DEV_OPS(&dnp->dn_lock);
7379 
7380 	error = attach_driver_by_parent(major, parents);
7381 	kmem_free(parents, devcnt * sizeof (char));
7382 	if (error == DDI_SUCCESS) {
7383 		exit_driver(dnp);
7384 		return (ops);
7385 	}
7386 
7387 	exit_driver(dnp);
7388 	ddi_rele_driver(major);
7389 	return (NULL);
7390 }
7391 
7392 /*
7393  * Default bus_config entry point for nexus drivers
7394  */
7395 int
7396 ndi_busop_bus_config(dev_info_t *pdip, uint_t flags, ddi_bus_config_op_t op,
7397     void *arg, dev_info_t **child, clock_t timeout)
7398 {
7399 	major_t major;
7400 
7401 	/*
7402 	 * A timeout of 30 minutes or more is probably a mistake
7403 	 * This is intended to catch uses where timeout is in
7404 	 * the wrong units.  timeout must be in units of ticks.
7405 	 */
7406 	ASSERT(timeout < SEC_TO_TICK(1800));
7407 
7408 	major = DDI_MAJOR_T_NONE;
7409 	switch (op) {
7410 	case BUS_CONFIG_ONE:
7411 		NDI_DEBUG(flags, (CE_CONT, "%s%d: bus config %s timeout=%ld\n",
7412 		    ddi_driver_name(pdip), ddi_get_instance(pdip),
7413 		    (char *)arg, timeout));
7414 		return (devi_config_one(pdip, (char *)arg, child, flags,
7415 		    timeout));
7416 
7417 	case BUS_CONFIG_DRIVER:
7418 		major = (major_t)(uintptr_t)arg;
7419 		/*FALLTHROUGH*/
7420 	case BUS_CONFIG_ALL:
7421 		NDI_DEBUG(flags, (CE_CONT, "%s%d: bus config timeout=%ld\n",
7422 		    ddi_driver_name(pdip), ddi_get_instance(pdip),
7423 		    timeout));
7424 		if (timeout > 0) {
7425 			NDI_DEBUG(flags, (CE_CONT,
7426 			    "%s%d: bus config all timeout=%ld\n",
7427 			    ddi_driver_name(pdip), ddi_get_instance(pdip),
7428 			    timeout));
7429 			delay(timeout);
7430 		}
7431 		return (config_immediate_children(pdip, flags, major));
7432 
7433 	default:
7434 		return (NDI_FAILURE);
7435 	}
7436 	/*NOTREACHED*/
7437 }
7438 
7439 /*
7440  * Default busop bus_unconfig handler for nexus drivers
7441  */
7442 int
7443 ndi_busop_bus_unconfig(dev_info_t *pdip, uint_t flags, ddi_bus_config_op_t op,
7444     void *arg)
7445 {
7446 	major_t major;
7447 
7448 	major = DDI_MAJOR_T_NONE;
7449 	switch (op) {
7450 	case BUS_UNCONFIG_ONE:
7451 		NDI_DEBUG(flags, (CE_CONT, "%s%d: bus unconfig %s\n",
7452 		    ddi_driver_name(pdip), ddi_get_instance(pdip),
7453 		    (char *)arg));
7454 		return (devi_unconfig_one(pdip, (char *)arg, flags));
7455 
7456 	case BUS_UNCONFIG_DRIVER:
7457 		major = (major_t)(uintptr_t)arg;
7458 		/*FALLTHROUGH*/
7459 	case BUS_UNCONFIG_ALL:
7460 		NDI_DEBUG(flags, (CE_CONT, "%s%d: bus unconfig all\n",
7461 		    ddi_driver_name(pdip), ddi_get_instance(pdip)));
7462 		return (unconfig_immediate_children(pdip, NULL, flags, major));
7463 
7464 	default:
7465 		return (NDI_FAILURE);
7466 	}
7467 	/*NOTREACHED*/
7468 }
7469 
7470 /*
7471  * dummy functions to be removed
7472  */
7473 void
7474 impl_rem_dev_props(dev_info_t *dip)
7475 {
7476 	_NOTE(ARGUNUSED(dip))
7477 	/* do nothing */
7478 }
7479 
7480 /*
7481  * Determine if a node is a leaf node. If not sure, return false (0).
7482  */
7483 static int
7484 is_leaf_node(dev_info_t *dip)
7485 {
7486 	major_t major = ddi_driver_major(dip);
7487 
7488 	if (major == DDI_MAJOR_T_NONE)
7489 		return (0);
7490 
7491 	return (devnamesp[major].dn_flags & DN_LEAF_DRIVER);
7492 }
7493 
7494 /*
7495  * Multithreaded [un]configuration
7496  */
7497 static struct mt_config_handle *
7498 mt_config_init(dev_info_t *pdip, dev_info_t **dipp, int flags,
7499     major_t major, int op, struct brevq_node **brevqp)
7500 {
7501 	struct mt_config_handle	*hdl = kmem_alloc(sizeof (*hdl), KM_SLEEP);
7502 
7503 	mutex_init(&hdl->mtc_lock, NULL, MUTEX_DEFAULT, NULL);
7504 	cv_init(&hdl->mtc_cv, NULL, CV_DEFAULT, NULL);
7505 	hdl->mtc_pdip = pdip;
7506 	hdl->mtc_fdip = dipp;
7507 	hdl->mtc_parmajor = DDI_MAJOR_T_NONE;
7508 	hdl->mtc_flags = flags;
7509 	hdl->mtc_major = major;
7510 	hdl->mtc_thr_count = 0;
7511 	hdl->mtc_op = op;
7512 	hdl->mtc_error = 0;
7513 	hdl->mtc_brevqp = brevqp;
7514 
7515 #ifdef DEBUG
7516 	gethrestime(&hdl->start_time);
7517 	hdl->total_time = 0;
7518 #endif /* DEBUG */
7519 
7520 	return (hdl);
7521 }
7522 
7523 #ifdef DEBUG
7524 static int
7525 time_diff_in_msec(timestruc_t start, timestruc_t end)
7526 {
7527 	int	nsec, sec;
7528 
7529 	sec = end.tv_sec - start.tv_sec;
7530 	nsec = end.tv_nsec - start.tv_nsec;
7531 	if (nsec < 0) {
7532 		nsec += NANOSEC;
7533 		sec -= 1;
7534 	}
7535 
7536 	return (sec * (NANOSEC >> 20) + (nsec >> 20));
7537 }
7538 
7539 #endif	/* DEBUG */
7540 
7541 static int
7542 mt_config_fini(struct mt_config_handle *hdl)
7543 {
7544 	int		rv;
7545 #ifdef DEBUG
7546 	int		real_time;
7547 	timestruc_t	end_time;
7548 #endif /* DEBUG */
7549 
7550 	mutex_enter(&hdl->mtc_lock);
7551 	while (hdl->mtc_thr_count > 0)
7552 		cv_wait(&hdl->mtc_cv, &hdl->mtc_lock);
7553 	rv = hdl->mtc_error;
7554 	mutex_exit(&hdl->mtc_lock);
7555 
7556 #ifdef DEBUG
7557 	gethrestime(&end_time);
7558 	real_time = time_diff_in_msec(hdl->start_time, end_time);
7559 	if ((ddidebug & DDI_MTCONFIG) && hdl->mtc_pdip)
7560 		cmn_err(CE_NOTE,
7561 		    "config %s%d: total time %d msec, real time %d msec",
7562 		    ddi_driver_name(hdl->mtc_pdip),
7563 		    ddi_get_instance(hdl->mtc_pdip),
7564 		    hdl->total_time, real_time);
7565 #endif /* DEBUG */
7566 
7567 	cv_destroy(&hdl->mtc_cv);
7568 	mutex_destroy(&hdl->mtc_lock);
7569 	kmem_free(hdl, sizeof (*hdl));
7570 
7571 	return (rv);
7572 }
7573 
7574 struct mt_config_data {
7575 	struct mt_config_handle	*mtc_hdl;
7576 	dev_info_t		*mtc_dip;
7577 	major_t			mtc_major;
7578 	int			mtc_flags;
7579 	struct brevq_node	*mtc_brn;
7580 	struct mt_config_data	*mtc_next;
7581 };
7582 
7583 static void
7584 mt_config_thread(void *arg)
7585 {
7586 	struct mt_config_data	*mcd = (struct mt_config_data *)arg;
7587 	struct mt_config_handle	*hdl = mcd->mtc_hdl;
7588 	dev_info_t		*dip = mcd->mtc_dip;
7589 	dev_info_t		*rdip, **dipp;
7590 	major_t			major = mcd->mtc_major;
7591 	int			flags = mcd->mtc_flags;
7592 	int			rv = 0;
7593 
7594 #ifdef DEBUG
7595 	timestruc_t start_time, end_time;
7596 	gethrestime(&start_time);
7597 #endif /* DEBUG */
7598 
7599 	rdip = NULL;
7600 	dipp = hdl->mtc_fdip ? &rdip : NULL;
7601 
7602 	switch (hdl->mtc_op) {
7603 	case MT_CONFIG_OP:
7604 		rv = devi_config_common(dip, flags, major);
7605 		break;
7606 	case MT_UNCONFIG_OP:
7607 		if (mcd->mtc_brn) {
7608 			struct brevq_node *brevq = NULL;
7609 			rv = devi_unconfig_common(dip, dipp, flags, major,
7610 			    &brevq);
7611 			mcd->mtc_brn->brn_child = brevq;
7612 		} else
7613 			rv = devi_unconfig_common(dip, dipp, flags, major,
7614 			    NULL);
7615 		break;
7616 	}
7617 
7618 	mutex_enter(&hdl->mtc_lock);
7619 #ifdef DEBUG
7620 	gethrestime(&end_time);
7621 	hdl->total_time += time_diff_in_msec(start_time, end_time);
7622 #endif /* DEBUG */
7623 
7624 	if ((rv != NDI_SUCCESS) && (hdl->mtc_error == 0)) {
7625 		hdl->mtc_error = rv;
7626 #ifdef	DEBUG
7627 		if ((ddidebug & DDI_DEBUG) && (major != DDI_MAJOR_T_NONE)) {
7628 			char	*path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
7629 
7630 			(void) ddi_pathname(dip, path);
7631 			cmn_err(CE_NOTE, "mt_config_thread: "
7632 			    "op %d.%d.%x at %s failed %d",
7633 			    hdl->mtc_op, major, flags, path, rv);
7634 			kmem_free(path, MAXPATHLEN);
7635 		}
7636 #endif	/* DEBUG */
7637 	}
7638 
7639 	if (hdl->mtc_fdip && *hdl->mtc_fdip == NULL) {
7640 		*hdl->mtc_fdip = rdip;
7641 		rdip = NULL;
7642 	}
7643 
7644 	if (rdip) {
7645 		ASSERT(rv != NDI_SUCCESS);
7646 		ndi_rele_devi(rdip);
7647 	}
7648 
7649 	ndi_rele_devi(dip);
7650 
7651 	if (--hdl->mtc_thr_count == 0)
7652 		cv_broadcast(&hdl->mtc_cv);
7653 	mutex_exit(&hdl->mtc_lock);
7654 	kmem_free(mcd, sizeof (*mcd));
7655 }
7656 
7657 /*
7658  * Multi-threaded config/unconfig of child nexus
7659  */
7660 static void
7661 mt_config_children(struct mt_config_handle *hdl)
7662 {
7663 	dev_info_t		*pdip = hdl->mtc_pdip;
7664 	major_t			major = hdl->mtc_major;
7665 	dev_info_t		*dip;
7666 	int			circ;
7667 	struct brevq_node	*brn;
7668 	struct mt_config_data	*mcd_head = NULL;
7669 	struct mt_config_data	*mcd_tail = NULL;
7670 	struct mt_config_data	*mcd;
7671 #ifdef DEBUG
7672 	timestruc_t		end_time;
7673 
7674 	/* Update total_time in handle */
7675 	gethrestime(&end_time);
7676 	hdl->total_time += time_diff_in_msec(hdl->start_time, end_time);
7677 #endif
7678 
7679 	ndi_devi_enter(pdip, &circ);
7680 	dip = ddi_get_child(pdip);
7681 	while (dip) {
7682 		if (hdl->mtc_op == MT_UNCONFIG_OP && hdl->mtc_brevqp &&
7683 		    !(DEVI_EVREMOVE(dip)) &&
7684 		    i_ddi_node_state(dip) >= DS_INITIALIZED) {
7685 			/*
7686 			 * Enqueue this dip's deviname.
7687 			 * No need to hold a lock while enqueuing since this
7688 			 * is the only thread doing the enqueue and no one
7689 			 * walks the queue while we are in multithreaded
7690 			 * unconfiguration.
7691 			 */
7692 			brn = brevq_enqueue(hdl->mtc_brevqp, dip, NULL);
7693 		} else
7694 			brn = NULL;
7695 
7696 		/*
7697 		 * Hold the child that we are processing so he does not get
7698 		 * removed. The corrisponding ndi_rele_devi() for children
7699 		 * that are not being skipped is done at the end of
7700 		 * mt_config_thread().
7701 		 */
7702 		ndi_hold_devi(dip);
7703 
7704 		/*
7705 		 * skip leaf nodes and (for configure) nodes not
7706 		 * fully attached.
7707 		 */
7708 		if (is_leaf_node(dip) ||
7709 		    (hdl->mtc_op == MT_CONFIG_OP &&
7710 		    i_ddi_node_state(dip) < DS_READY)) {
7711 			ndi_rele_devi(dip);
7712 			dip = ddi_get_next_sibling(dip);
7713 			continue;
7714 		}
7715 
7716 		mcd = kmem_alloc(sizeof (*mcd), KM_SLEEP);
7717 		mcd->mtc_dip = dip;
7718 		mcd->mtc_hdl = hdl;
7719 		mcd->mtc_brn = brn;
7720 
7721 		/*
7722 		 * Switch a 'driver' operation to an 'all' operation below a
7723 		 * node bound to the driver.
7724 		 */
7725 		if ((major == DDI_MAJOR_T_NONE) ||
7726 		    (major == ddi_driver_major(dip)))
7727 			mcd->mtc_major = DDI_MAJOR_T_NONE;
7728 		else
7729 			mcd->mtc_major = major;
7730 
7731 		/*
7732 		 * The unconfig-driver to unconfig-all conversion above
7733 		 * constitutes an autodetach for NDI_DETACH_DRIVER calls,
7734 		 * set NDI_AUTODETACH.
7735 		 */
7736 		mcd->mtc_flags = hdl->mtc_flags;
7737 		if ((mcd->mtc_flags & NDI_DETACH_DRIVER) &&
7738 		    (hdl->mtc_op == MT_UNCONFIG_OP) &&
7739 		    (major == ddi_driver_major(pdip)))
7740 			mcd->mtc_flags |= NDI_AUTODETACH;
7741 
7742 		mutex_enter(&hdl->mtc_lock);
7743 		hdl->mtc_thr_count++;
7744 		mutex_exit(&hdl->mtc_lock);
7745 
7746 		/*
7747 		 * Add to end of list to process after ndi_devi_exit to avoid
7748 		 * locking differences depending on value of mtc_off.
7749 		 */
7750 		mcd->mtc_next = NULL;
7751 		if (mcd_head == NULL)
7752 			mcd_head = mcd;
7753 		else
7754 			mcd_tail->mtc_next = mcd;
7755 		mcd_tail = mcd;
7756 
7757 		dip = ddi_get_next_sibling(dip);
7758 	}
7759 	ndi_devi_exit(pdip, circ);
7760 
7761 	/* go through the list of held children */
7762 	for (mcd = mcd_head; mcd; mcd = mcd_head) {
7763 		mcd_head = mcd->mtc_next;
7764 		if (mtc_off || (mcd->mtc_flags & NDI_MTC_OFF))
7765 			mt_config_thread(mcd);
7766 		else
7767 			(void) thread_create(NULL, 0, mt_config_thread, mcd,
7768 			    0, &p0, TS_RUN, minclsyspri);
7769 	}
7770 }
7771 
7772 static void
7773 mt_config_driver(struct mt_config_handle *hdl)
7774 {
7775 	major_t			par_major = hdl->mtc_parmajor;
7776 	major_t			major = hdl->mtc_major;
7777 	struct devnames		*dnp = &devnamesp[par_major];
7778 	dev_info_t		*dip;
7779 	struct mt_config_data	*mcd_head = NULL;
7780 	struct mt_config_data	*mcd_tail = NULL;
7781 	struct mt_config_data	*mcd;
7782 #ifdef DEBUG
7783 	timestruc_t		end_time;
7784 
7785 	/* Update total_time in handle */
7786 	gethrestime(&end_time);
7787 	hdl->total_time += time_diff_in_msec(hdl->start_time, end_time);
7788 #endif
7789 	ASSERT(par_major != DDI_MAJOR_T_NONE);
7790 	ASSERT(major != DDI_MAJOR_T_NONE);
7791 
7792 	LOCK_DEV_OPS(&dnp->dn_lock);
7793 	dip = devnamesp[par_major].dn_head;
7794 	while (dip) {
7795 		/*
7796 		 * Hold the child that we are processing so he does not get
7797 		 * removed. The corrisponding ndi_rele_devi() for children
7798 		 * that are not being skipped is done at the end of
7799 		 * mt_config_thread().
7800 		 */
7801 		ndi_hold_devi(dip);
7802 
7803 		/* skip leaf nodes and nodes not fully attached */
7804 		if (!i_ddi_devi_attached(dip) || is_leaf_node(dip)) {
7805 			ndi_rele_devi(dip);
7806 			dip = ddi_get_next(dip);
7807 			continue;
7808 		}
7809 
7810 		mcd = kmem_alloc(sizeof (*mcd), KM_SLEEP);
7811 		mcd->mtc_dip = dip;
7812 		mcd->mtc_hdl = hdl;
7813 		mcd->mtc_major = major;
7814 		mcd->mtc_flags = hdl->mtc_flags;
7815 
7816 		mutex_enter(&hdl->mtc_lock);
7817 		hdl->mtc_thr_count++;
7818 		mutex_exit(&hdl->mtc_lock);
7819 
7820 		/*
7821 		 * Add to end of list to process after UNLOCK_DEV_OPS to avoid
7822 		 * locking differences depending on value of mtc_off.
7823 		 */
7824 		mcd->mtc_next = NULL;
7825 		if (mcd_head == NULL)
7826 			mcd_head = mcd;
7827 		else
7828 			mcd_tail->mtc_next = mcd;
7829 		mcd_tail = mcd;
7830 
7831 		dip = ddi_get_next(dip);
7832 	}
7833 	UNLOCK_DEV_OPS(&dnp->dn_lock);
7834 
7835 	/* go through the list of held children */
7836 	for (mcd = mcd_head; mcd; mcd = mcd_head) {
7837 		mcd_head = mcd->mtc_next;
7838 		if (mtc_off || (mcd->mtc_flags & NDI_MTC_OFF))
7839 			mt_config_thread(mcd);
7840 		else
7841 			(void) thread_create(NULL, 0, mt_config_thread, mcd,
7842 			    0, &p0, TS_RUN, minclsyspri);
7843 	}
7844 }
7845 
7846 /*
7847  * Given the nodeid for a persistent (PROM or SID) node, return
7848  * the corresponding devinfo node
7849  * NOTE: This function will return NULL for .conf nodeids.
7850  */
7851 dev_info_t *
7852 e_ddi_nodeid_to_dip(pnode_t nodeid)
7853 {
7854 	dev_info_t		*dip = NULL;
7855 	struct devi_nodeid	*prev, *elem;
7856 
7857 	mutex_enter(&devimap->dno_lock);
7858 
7859 	prev = NULL;
7860 	for (elem = devimap->dno_head; elem; elem = elem->next) {
7861 		if (elem->nodeid == nodeid) {
7862 			ndi_hold_devi(elem->dip);
7863 			dip = elem->dip;
7864 			break;
7865 		}
7866 		prev = elem;
7867 	}
7868 
7869 	/*
7870 	 * Move to head for faster lookup next time
7871 	 */
7872 	if (elem && prev) {
7873 		prev->next = elem->next;
7874 		elem->next = devimap->dno_head;
7875 		devimap->dno_head = elem;
7876 	}
7877 
7878 	mutex_exit(&devimap->dno_lock);
7879 	return (dip);
7880 }
7881 
7882 static void
7883 free_cache_task(void *arg)
7884 {
7885 	ASSERT(arg == NULL);
7886 
7887 	mutex_enter(&di_cache.cache_lock);
7888 
7889 	/*
7890 	 * The cache can be invalidated without holding the lock
7891 	 * but it can be made valid again only while the lock is held.
7892 	 * So if the cache is invalid when the lock is held, it will
7893 	 * stay invalid until lock is released.
7894 	 */
7895 	if (!di_cache.cache_valid)
7896 		i_ddi_di_cache_free(&di_cache);
7897 
7898 	mutex_exit(&di_cache.cache_lock);
7899 
7900 	if (di_cache_debug)
7901 		cmn_err(CE_NOTE, "system_taskq: di_cache freed");
7902 }
7903 
7904 extern int modrootloaded;
7905 
7906 void
7907 i_ddi_di_cache_free(struct di_cache *cache)
7908 {
7909 	int	error;
7910 	extern int sys_shutdown;
7911 
7912 	ASSERT(mutex_owned(&cache->cache_lock));
7913 
7914 	if (cache->cache_size) {
7915 		ASSERT(cache->cache_size > 0);
7916 		ASSERT(cache->cache_data);
7917 
7918 		kmem_free(cache->cache_data, cache->cache_size);
7919 		cache->cache_data = NULL;
7920 		cache->cache_size = 0;
7921 
7922 		if (di_cache_debug)
7923 			cmn_err(CE_NOTE, "i_ddi_di_cache_free: freed cachemem");
7924 	} else {
7925 		ASSERT(cache->cache_data == NULL);
7926 		if (di_cache_debug)
7927 			cmn_err(CE_NOTE, "i_ddi_di_cache_free: NULL cache");
7928 	}
7929 
7930 	if (!modrootloaded || rootvp == NULL ||
7931 	    vn_is_readonly(rootvp) || sys_shutdown) {
7932 		if (di_cache_debug) {
7933 			cmn_err(CE_WARN, "/ not mounted/RDONLY. Skip unlink");
7934 		}
7935 		return;
7936 	}
7937 
7938 	error = vn_remove(DI_CACHE_FILE, UIO_SYSSPACE, RMFILE);
7939 	if (di_cache_debug && error && error != ENOENT) {
7940 		cmn_err(CE_WARN, "%s: unlink failed: %d", DI_CACHE_FILE, error);
7941 	} else if (di_cache_debug && !error) {
7942 		cmn_err(CE_NOTE, "i_ddi_di_cache_free: unlinked cache file");
7943 	}
7944 }
7945 
7946 void
7947 i_ddi_di_cache_invalidate()
7948 {
7949 	int	cache_valid;
7950 
7951 	if (!modrootloaded || !i_ddi_io_initialized()) {
7952 		if (di_cache_debug)
7953 			cmn_err(CE_NOTE, "I/O not inited. Skipping invalidate");
7954 		return;
7955 	}
7956 
7957 	/* Increment devtree generation number. */
7958 	atomic_inc_ulong(&devtree_gen);
7959 
7960 	/* Invalidate the in-core cache and dispatch free on valid->invalid */
7961 	cache_valid = atomic_swap_uint(&di_cache.cache_valid, 0);
7962 	if (cache_valid) {
7963 		/*
7964 		 * This is an optimization to start cleaning up a cached
7965 		 * snapshot early.  For this reason, it is OK for
7966 		 * taskq_dispatach to fail (and it is OK to not track calling
7967 		 * context relative to sleep, and assume NOSLEEP).
7968 		 */
7969 		(void) taskq_dispatch(system_taskq, free_cache_task, NULL,
7970 		    TQ_NOSLEEP);
7971 	}
7972 
7973 	if (di_cache_debug) {
7974 		cmn_err(CE_NOTE, "invalidation");
7975 	}
7976 }
7977 
7978 
7979 static void
7980 i_bind_vhci_node(dev_info_t *dip)
7981 {
7982 	DEVI(dip)->devi_major = ddi_name_to_major(ddi_node_name(dip));
7983 	i_ddi_set_node_state(dip, DS_BOUND);
7984 }
7985 
7986 static char vhci_node_addr[2];
7987 
7988 static int
7989 i_init_vhci_node(dev_info_t *dip)
7990 {
7991 	add_global_props(dip);
7992 	DEVI(dip)->devi_ops = ndi_hold_driver(dip);
7993 	if (DEVI(dip)->devi_ops == NULL)
7994 		return (-1);
7995 
7996 	DEVI(dip)->devi_instance = e_ddi_assign_instance(dip);
7997 	e_ddi_keep_instance(dip);
7998 	vhci_node_addr[0]	= '\0';
7999 	ddi_set_name_addr(dip, vhci_node_addr);
8000 	i_ddi_set_node_state(dip, DS_INITIALIZED);
8001 	return (0);
8002 }
8003 
8004 static void
8005 i_link_vhci_node(dev_info_t *dip)
8006 {
8007 	ASSERT(MUTEX_HELD(&global_vhci_lock));
8008 
8009 	/*
8010 	 * scsi_vhci should be kept left most of the device tree.
8011 	 */
8012 	if (scsi_vhci_dip) {
8013 		DEVI(dip)->devi_sibling = DEVI(scsi_vhci_dip)->devi_sibling;
8014 		DEVI(scsi_vhci_dip)->devi_sibling = DEVI(dip);
8015 	} else {
8016 		DEVI(dip)->devi_sibling = DEVI(top_devinfo)->devi_child;
8017 		DEVI(top_devinfo)->devi_child = DEVI(dip);
8018 	}
8019 }
8020 
8021 
8022 /*
8023  * This a special routine to enumerate vhci node (child of rootnex
8024  * node) without holding the ndi_devi_enter() lock. The device node
8025  * is allocated, initialized and brought into DS_READY state before
8026  * inserting into the device tree. The VHCI node is handcrafted
8027  * here to bring the node to DS_READY, similar to rootnex node.
8028  *
8029  * The global_vhci_lock protects linking the node into the device
8030  * as same lock is held before linking/unlinking any direct child
8031  * of rootnex children.
8032  *
8033  * This routine is a workaround to handle a possible deadlock
8034  * that occurs while trying to enumerate node in a different sub-tree
8035  * during _init/_attach entry points.
8036  */
8037 /*ARGSUSED*/
8038 dev_info_t *
8039 ndi_devi_config_vhci(char *drvname, int flags)
8040 {
8041 	struct devnames		*dnp;
8042 	dev_info_t		*dip;
8043 	major_t			major = ddi_name_to_major(drvname);
8044 
8045 	if (major == -1)
8046 		return (NULL);
8047 
8048 	/* Make sure we create the VHCI node only once */
8049 	dnp = &devnamesp[major];
8050 	LOCK_DEV_OPS(&dnp->dn_lock);
8051 	if (dnp->dn_head) {
8052 		dip = dnp->dn_head;
8053 		UNLOCK_DEV_OPS(&dnp->dn_lock);
8054 		return (dip);
8055 	}
8056 	UNLOCK_DEV_OPS(&dnp->dn_lock);
8057 
8058 	/* Allocate the VHCI node */
8059 	ndi_devi_alloc_sleep(top_devinfo, drvname, DEVI_SID_NODEID, &dip);
8060 	ndi_hold_devi(dip);
8061 
8062 	/* Mark the node as VHCI */
8063 	DEVI(dip)->devi_node_attributes |= DDI_VHCI_NODE;
8064 
8065 	i_ddi_add_devimap(dip);
8066 	i_bind_vhci_node(dip);
8067 	if (i_init_vhci_node(dip) == -1) {
8068 		ndi_rele_devi(dip);
8069 		(void) ndi_devi_free(dip);
8070 		return (NULL);
8071 	}
8072 
8073 	mutex_enter(&(DEVI(dip)->devi_lock));
8074 	DEVI_SET_ATTACHING(dip);
8075 	mutex_exit(&(DEVI(dip)->devi_lock));
8076 
8077 	if (devi_attach(dip, DDI_ATTACH) != DDI_SUCCESS) {
8078 		cmn_err(CE_CONT, "Could not attach %s driver", drvname);
8079 		e_ddi_free_instance(dip, vhci_node_addr);
8080 		ndi_rele_devi(dip);
8081 		(void) ndi_devi_free(dip);
8082 		return (NULL);
8083 	}
8084 	mutex_enter(&(DEVI(dip)->devi_lock));
8085 	DEVI_CLR_ATTACHING(dip);
8086 	mutex_exit(&(DEVI(dip)->devi_lock));
8087 
8088 	mutex_enter(&global_vhci_lock);
8089 	i_link_vhci_node(dip);
8090 	mutex_exit(&global_vhci_lock);
8091 	i_ddi_set_node_state(dip, DS_READY);
8092 
8093 	LOCK_DEV_OPS(&dnp->dn_lock);
8094 	dnp->dn_flags |= DN_DRIVER_HELD;
8095 	dnp->dn_head = dip;
8096 	UNLOCK_DEV_OPS(&dnp->dn_lock);
8097 
8098 	i_ndi_devi_report_status_change(dip, NULL);
8099 
8100 	return (dip);
8101 }
8102 
8103 /*
8104  * Maintain DEVI_DEVICE_REMOVED hotplug devi_state for remove/reinsert hotplug
8105  * of open devices. Currently, because of tight coupling between the devfs file
8106  * system and the Solaris device tree, a driver can't always make the device
8107  * tree state (esp devi_node_state) match device hardware hotplug state. Until
8108  * resolved, to overcome this deficiency we use the following interfaces that
8109  * maintain the DEVI_DEVICE_REMOVED devi_state status bit.  These interface
8110  * report current state, and drive operation (like events and cache
8111  * invalidation) when a driver changes remove/insert state of an open device.
8112  *
8113  * The ndi_devi_device_isremoved() returns 1 if the device is currently removed.
8114  *
8115  * The ndi_devi_device_remove() interface declares the device as removed, and
8116  * returns 1 if there was a state change associated with this declaration.
8117  *
8118  * The ndi_devi_device_insert() declares the device as inserted, and returns 1
8119  * if there was a state change associated with this declaration.
8120  */
8121 int
8122 ndi_devi_device_isremoved(dev_info_t *dip)
8123 {
8124 	return (DEVI_IS_DEVICE_REMOVED(dip));
8125 }
8126 
8127 int
8128 ndi_devi_device_remove(dev_info_t *dip)
8129 {
8130 	ASSERT(dip && ddi_get_parent(dip) &&
8131 	    DEVI_BUSY_OWNED(ddi_get_parent(dip)));
8132 
8133 	/* Return if already marked removed. */
8134 	if (ndi_devi_device_isremoved(dip))
8135 		return (0);
8136 
8137 	/* Mark the device as having been physically removed. */
8138 	mutex_enter(&(DEVI(dip)->devi_lock));
8139 	ndi_devi_set_hidden(dip);	/* invisible: lookup/snapshot */
8140 	DEVI_SET_DEVICE_REMOVED(dip);
8141 	DEVI_SET_EVREMOVE(dip);		/* this clears EVADD too */
8142 	mutex_exit(&(DEVI(dip)->devi_lock));
8143 
8144 	/* report remove (as 'removed') */
8145 	i_ndi_devi_report_status_change(dip, NULL);
8146 
8147 	/*
8148 	 * Invalidate the cache to ensure accurate
8149 	 * (di_state() & DI_DEVICE_REMOVED).
8150 	 */
8151 	i_ddi_di_cache_invalidate();
8152 
8153 	/*
8154 	 * Generate sysevent for those interested in removal (either
8155 	 * directly via private EC_DEVFS or indirectly via devfsadmd
8156 	 * generated EC_DEV). This will generate LDI DEVICE_REMOVE
8157 	 * event too.
8158 	 */
8159 	i_ddi_log_devfs_device_remove(dip);
8160 
8161 	return (1);		/* DEVICE_REMOVED state changed */
8162 }
8163 
8164 int
8165 ndi_devi_device_insert(dev_info_t *dip)
8166 {
8167 	ASSERT(dip && ddi_get_parent(dip) &&
8168 	    DEVI_BUSY_OWNED(ddi_get_parent(dip)));
8169 
8170 	/* Return if not marked removed. */
8171 	if (!ndi_devi_device_isremoved(dip))
8172 		return (0);
8173 
8174 	/* Mark the device as having been physically reinserted. */
8175 	mutex_enter(&(DEVI(dip)->devi_lock));
8176 	ndi_devi_clr_hidden(dip);	/* visible: lookup/snapshot */
8177 	DEVI_SET_DEVICE_REINSERTED(dip);
8178 	DEVI_SET_EVADD(dip);		/* this clears EVREMOVE too */
8179 	mutex_exit(&(DEVI(dip)->devi_lock));
8180 
8181 	/* report insert (as 'online') */
8182 	i_ndi_devi_report_status_change(dip, NULL);
8183 
8184 	/*
8185 	 * Invalidate the cache to ensure accurate
8186 	 * (di_state() & DI_DEVICE_REMOVED).
8187 	 */
8188 	i_ddi_di_cache_invalidate();
8189 
8190 	/*
8191 	 * Generate sysevent for those interested in removal (either directly
8192 	 * via EC_DEVFS or indirectly via devfsadmd generated EC_DEV).
8193 	 */
8194 	i_ddi_log_devfs_device_insert(dip);
8195 
8196 	return (1);		/* DEVICE_REMOVED state changed */
8197 }
8198 
8199 /*
8200  * ibt_hw_is_present() returns 0 when there is no IB hardware actively
8201  * running.  This is primarily useful for modules like rpcmod which
8202  * needs a quick check to decide whether or not it should try to use
8203  * InfiniBand
8204  */
8205 int ib_hw_status = 0;
8206 int
8207 ibt_hw_is_present()
8208 {
8209 	return (ib_hw_status);
8210 }
8211 
8212 /*
8213  * ASSERT that constraint flag is not set and then set the "retire attempt"
8214  * flag.
8215  */
8216 int
8217 e_ddi_mark_retiring(dev_info_t *dip, void *arg)
8218 {
8219 	char	**cons_array = (char **)arg;
8220 	char	*path;
8221 	int	constraint;
8222 	int	i;
8223 
8224 	constraint = 0;
8225 	if (cons_array) {
8226 		path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
8227 		(void) ddi_pathname(dip, path);
8228 		for (i = 0; cons_array[i] != NULL; i++) {
8229 			if (strcmp(path, cons_array[i]) == 0) {
8230 				constraint = 1;
8231 				break;
8232 			}
8233 		}
8234 		kmem_free(path, MAXPATHLEN);
8235 	}
8236 
8237 	mutex_enter(&DEVI(dip)->devi_lock);
8238 	ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT));
8239 	DEVI(dip)->devi_flags |= DEVI_RETIRING;
8240 	if (constraint)
8241 		DEVI(dip)->devi_flags |= DEVI_R_CONSTRAINT;
8242 	mutex_exit(&DEVI(dip)->devi_lock);
8243 
8244 	RIO_VERBOSE((CE_NOTE, "marked dip as undergoing retire process dip=%p",
8245 	    (void *)dip));
8246 
8247 	if (constraint)
8248 		RIO_DEBUG((CE_NOTE, "marked dip as constrained, dip=%p",
8249 		    (void *)dip));
8250 
8251 	if (MDI_PHCI(dip))
8252 		mdi_phci_mark_retiring(dip, cons_array);
8253 
8254 	return (DDI_WALK_CONTINUE);
8255 }
8256 
8257 static void
8258 free_array(char **cons_array)
8259 {
8260 	int	i;
8261 
8262 	if (cons_array == NULL)
8263 		return;
8264 
8265 	for (i = 0; cons_array[i] != NULL; i++) {
8266 		kmem_free(cons_array[i], strlen(cons_array[i]) + 1);
8267 	}
8268 	kmem_free(cons_array, (i+1) * sizeof (char *));
8269 }
8270 
8271 /*
8272  * Walk *every* node in subtree and check if it blocks, allows or has no
8273  * comment on a proposed retire.
8274  */
8275 int
8276 e_ddi_retire_notify(dev_info_t *dip, void *arg)
8277 {
8278 	int	*constraint = (int *)arg;
8279 
8280 	RIO_DEBUG((CE_NOTE, "retire notify: dip = %p", (void *)dip));
8281 
8282 	(void) e_ddi_offline_notify(dip);
8283 
8284 	mutex_enter(&(DEVI(dip)->devi_lock));
8285 	if (!(DEVI(dip)->devi_flags & DEVI_RETIRING)) {
8286 		RIO_DEBUG((CE_WARN, "retire notify: dip in retire "
8287 		    "subtree is not marked: dip = %p", (void *)dip));
8288 		*constraint = 0;
8289 	} else if (DEVI(dip)->devi_flags & DEVI_R_BLOCKED) {
8290 		ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT));
8291 		RIO_DEBUG((CE_NOTE, "retire notify: BLOCKED: dip = %p",
8292 		    (void *)dip));
8293 		*constraint = 0;
8294 	} else if (!(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT)) {
8295 		RIO_DEBUG((CE_NOTE, "retire notify: NO CONSTRAINT: "
8296 		    "dip = %p", (void *)dip));
8297 		*constraint = 0;
8298 	} else {
8299 		RIO_DEBUG((CE_NOTE, "retire notify: CONSTRAINT set: "
8300 		    "dip = %p", (void *)dip));
8301 	}
8302 	mutex_exit(&DEVI(dip)->devi_lock);
8303 
8304 	if (MDI_PHCI(dip))
8305 		mdi_phci_retire_notify(dip, constraint);
8306 
8307 	return (DDI_WALK_CONTINUE);
8308 }
8309 
8310 int
8311 e_ddi_retire_finalize(dev_info_t *dip, void *arg)
8312 {
8313 	int constraint = *(int *)arg;
8314 	int finalize;
8315 	int phci_only;
8316 
8317 	mutex_enter(&DEVI(dip)->devi_lock);
8318 	if (!(DEVI(dip)->devi_flags & DEVI_RETIRING)) {
8319 		RIO_DEBUG((CE_WARN,
8320 		    "retire: unmarked dip(%p) in retire subtree",
8321 		    (void *)dip));
8322 		ASSERT(!(DEVI(dip)->devi_flags & DEVI_RETIRED));
8323 		ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT));
8324 		ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_BLOCKED));
8325 		mutex_exit(&DEVI(dip)->devi_lock);
8326 		return (DDI_WALK_CONTINUE);
8327 	}
8328 
8329 	/*
8330 	 * retire the device if constraints have been applied
8331 	 * or if the device is not in use
8332 	 */
8333 	finalize = 0;
8334 	if (constraint) {
8335 		ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(dip)));
8336 
8337 		ASSERT(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT);
8338 		ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_BLOCKED));
8339 		DEVI(dip)->devi_flags &= ~DEVI_R_CONSTRAINT;
8340 		DEVI(dip)->devi_flags &= ~DEVI_RETIRING;
8341 		DEVI(dip)->devi_flags |= DEVI_RETIRED;
8342 		mutex_exit(&DEVI(dip)->devi_lock);
8343 		(void) spec_fence_snode(dip, NULL);
8344 		RIO_DEBUG((CE_NOTE, "Fenced off: dip = %p", (void *)dip));
8345 		e_ddi_offline_finalize(dip, DDI_SUCCESS);
8346 	} else {
8347 		if (DEVI(dip)->devi_flags & DEVI_R_BLOCKED) {
8348 			ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT));
8349 			DEVI(dip)->devi_flags &= ~DEVI_R_BLOCKED;
8350 			DEVI(dip)->devi_flags &= ~DEVI_RETIRING;
8351 			/* we have already finalized during notify */
8352 		} else if (DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT) {
8353 			DEVI(dip)->devi_flags &= ~DEVI_R_CONSTRAINT;
8354 			DEVI(dip)->devi_flags &= ~DEVI_RETIRING;
8355 			finalize = 1;
8356 		} else {
8357 			DEVI(dip)->devi_flags &= ~DEVI_RETIRING;
8358 			/*
8359 			 * even if no contracts, need to call finalize
8360 			 * to clear the contract barrier on the dip
8361 			 */
8362 			finalize = 1;
8363 		}
8364 		mutex_exit(&DEVI(dip)->devi_lock);
8365 		RIO_DEBUG((CE_NOTE, "finalize: NOT retired: dip = %p",
8366 		    (void *)dip));
8367 		if (finalize)
8368 			e_ddi_offline_finalize(dip, DDI_FAILURE);
8369 	}
8370 
8371 	/*
8372 	 * phci_only variable indicates no client checking, just
8373 	 * offline the PHCI. We set that to 0 to enable client
8374 	 * checking
8375 	 */
8376 	phci_only = 0;
8377 	if (MDI_PHCI(dip))
8378 		mdi_phci_retire_finalize(dip, phci_only, arg);
8379 
8380 	return (DDI_WALK_CONTINUE);
8381 }
8382 
8383 /*
8384  * Returns
8385  *	DDI_SUCCESS if constraints allow retire
8386  *	DDI_FAILURE if constraints don't allow retire.
8387  * cons_array is a NULL terminated array of node paths for
8388  * which constraints have already been applied.
8389  */
8390 int
8391 e_ddi_retire_device(char *path, char **cons_array)
8392 {
8393 	dev_info_t	*dip;
8394 	dev_info_t	*pdip;
8395 	int		circ;
8396 	int		circ2;
8397 	int		constraint;
8398 	char		*devnm;
8399 
8400 	/*
8401 	 * First, lookup the device
8402 	 */
8403 	dip = e_ddi_hold_devi_by_path(path, 0);
8404 	if (dip == NULL) {
8405 		/*
8406 		 * device does not exist. This device cannot be
8407 		 * a critical device since it is not in use. Thus
8408 		 * this device is always retireable. Return DDI_SUCCESS
8409 		 * to indicate this. If this device is ever
8410 		 * instantiated, I/O framework will consult the
8411 		 * the persistent retire store, mark it as
8412 		 * retired and fence it off.
8413 		 */
8414 		RIO_DEBUG((CE_NOTE, "Retire device: device doesn't exist."
8415 		    " NOP. Just returning SUCCESS. path=%s", path));
8416 		free_array(cons_array);
8417 		return (DDI_SUCCESS);
8418 	}
8419 
8420 	RIO_DEBUG((CE_NOTE, "Retire device: found dip = %p.", (void *)dip));
8421 
8422 	pdip = ddi_get_parent(dip);
8423 	ndi_hold_devi(pdip);
8424 
8425 	/*
8426 	 * Run devfs_clean() in case dip has no constraints and is
8427 	 * not in use, so is retireable but there are dv_nodes holding
8428 	 * ref-count on the dip. Note that devfs_clean() always returns
8429 	 * success.
8430 	 */
8431 	devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
8432 	(void) ddi_deviname(dip, devnm);
8433 	(void) devfs_clean(pdip, devnm + 1, DV_CLEAN_FORCE);
8434 	kmem_free(devnm, MAXNAMELEN + 1);
8435 
8436 	ndi_devi_enter(pdip, &circ);
8437 
8438 	/* release hold from e_ddi_hold_devi_by_path */
8439 	ndi_rele_devi(dip);
8440 
8441 	/*
8442 	 * If it cannot make a determination, is_leaf_node() assumes
8443 	 * dip is a nexus.
8444 	 */
8445 	(void) e_ddi_mark_retiring(dip, cons_array);
8446 	if (!is_leaf_node(dip)) {
8447 		ndi_devi_enter(dip, &circ2);
8448 		ddi_walk_devs(ddi_get_child(dip), e_ddi_mark_retiring,
8449 		    cons_array);
8450 		ndi_devi_exit(dip, circ2);
8451 	}
8452 	free_array(cons_array);
8453 
8454 	/*
8455 	 * apply constraints
8456 	 */
8457 	RIO_DEBUG((CE_NOTE, "retire: subtree retire notify: path = %s", path));
8458 
8459 	constraint = 1;	/* assume constraints allow retire */
8460 	(void) e_ddi_retire_notify(dip, &constraint);
8461 	if (!is_leaf_node(dip)) {
8462 		ndi_devi_enter(dip, &circ2);
8463 		ddi_walk_devs(ddi_get_child(dip), e_ddi_retire_notify,
8464 		    &constraint);
8465 		ndi_devi_exit(dip, circ2);
8466 	}
8467 
8468 	/*
8469 	 * Now finalize the retire
8470 	 */
8471 	(void) e_ddi_retire_finalize(dip, &constraint);
8472 	if (!is_leaf_node(dip)) {
8473 		ndi_devi_enter(dip, &circ2);
8474 		ddi_walk_devs(ddi_get_child(dip), e_ddi_retire_finalize,
8475 		    &constraint);
8476 		ndi_devi_exit(dip, circ2);
8477 	}
8478 
8479 	if (!constraint) {
8480 		RIO_DEBUG((CE_WARN, "retire failed: path = %s", path));
8481 	} else {
8482 		RIO_DEBUG((CE_NOTE, "retire succeeded: path = %s", path));
8483 	}
8484 
8485 	ndi_devi_exit(pdip, circ);
8486 	ndi_rele_devi(pdip);
8487 	return (constraint ? DDI_SUCCESS : DDI_FAILURE);
8488 }
8489 
8490 static int
8491 unmark_and_unfence(dev_info_t *dip, void *arg)
8492 {
8493 	char	*path = (char *)arg;
8494 
8495 	ASSERT(path);
8496 
8497 	(void) ddi_pathname(dip, path);
8498 
8499 	mutex_enter(&DEVI(dip)->devi_lock);
8500 	DEVI(dip)->devi_flags &= ~DEVI_RETIRED;
8501 	DEVI_SET_DEVICE_ONLINE(dip);
8502 	mutex_exit(&DEVI(dip)->devi_lock);
8503 
8504 	RIO_VERBOSE((CE_NOTE, "Cleared RETIRED flag: dip=%p, path=%s",
8505 	    (void *)dip, path));
8506 
8507 	(void) spec_unfence_snode(dip);
8508 	RIO_DEBUG((CE_NOTE, "Unfenced device: %s", path));
8509 
8510 	if (MDI_PHCI(dip))
8511 		mdi_phci_unretire(dip);
8512 
8513 	return (DDI_WALK_CONTINUE);
8514 }
8515 
8516 struct find_dip {
8517 	char	*fd_buf;
8518 	char	*fd_path;
8519 	dev_info_t *fd_dip;
8520 };
8521 
8522 static int
8523 find_dip_fcn(dev_info_t *dip, void *arg)
8524 {
8525 	struct find_dip *findp = (struct find_dip *)arg;
8526 
8527 	(void) ddi_pathname(dip, findp->fd_buf);
8528 
8529 	if (strcmp(findp->fd_path, findp->fd_buf) != 0)
8530 		return (DDI_WALK_CONTINUE);
8531 
8532 	ndi_hold_devi(dip);
8533 	findp->fd_dip = dip;
8534 
8535 	return (DDI_WALK_TERMINATE);
8536 }
8537 
8538 int
8539 e_ddi_unretire_device(char *path)
8540 {
8541 	int		circ;
8542 	int		circ2;
8543 	char		*path2;
8544 	dev_info_t	*pdip;
8545 	dev_info_t	*dip;
8546 	struct find_dip	 find_dip;
8547 
8548 	ASSERT(path);
8549 	ASSERT(*path == '/');
8550 
8551 	if (strcmp(path, "/") == 0) {
8552 		cmn_err(CE_WARN, "Root node cannot be retired. Skipping "
8553 		    "device unretire: %s", path);
8554 		return (0);
8555 	}
8556 
8557 	/*
8558 	 * We can't lookup the dip (corresponding to path) via
8559 	 * e_ddi_hold_devi_by_path() because the dip may be offline
8560 	 * and may not attach. Use ddi_walk_devs() instead;
8561 	 */
8562 	find_dip.fd_buf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
8563 	find_dip.fd_path = path;
8564 	find_dip.fd_dip = NULL;
8565 
8566 	pdip = ddi_root_node();
8567 
8568 	ndi_devi_enter(pdip, &circ);
8569 	ddi_walk_devs(ddi_get_child(pdip), find_dip_fcn, &find_dip);
8570 	ndi_devi_exit(pdip, circ);
8571 
8572 	kmem_free(find_dip.fd_buf, MAXPATHLEN);
8573 
8574 	if (find_dip.fd_dip == NULL) {
8575 		cmn_err(CE_WARN, "Device not found in device tree. Skipping "
8576 		    "device unretire: %s", path);
8577 		return (0);
8578 	}
8579 
8580 	dip = find_dip.fd_dip;
8581 
8582 	pdip = ddi_get_parent(dip);
8583 
8584 	ndi_hold_devi(pdip);
8585 
8586 	ndi_devi_enter(pdip, &circ);
8587 
8588 	path2 = kmem_alloc(MAXPATHLEN, KM_SLEEP);
8589 
8590 	(void) unmark_and_unfence(dip, path2);
8591 	if (!is_leaf_node(dip)) {
8592 		ndi_devi_enter(dip, &circ2);
8593 		ddi_walk_devs(ddi_get_child(dip), unmark_and_unfence, path2);
8594 		ndi_devi_exit(dip, circ2);
8595 	}
8596 
8597 	kmem_free(path2, MAXPATHLEN);
8598 
8599 	/* release hold from find_dip_fcn() */
8600 	ndi_rele_devi(dip);
8601 
8602 	ndi_devi_exit(pdip, circ);
8603 
8604 	ndi_rele_devi(pdip);
8605 
8606 	return (0);
8607 }
8608 
8609 /*
8610  * Called before attach on a dip that has been retired.
8611  */
8612 static int
8613 mark_and_fence(dev_info_t *dip, void *arg)
8614 {
8615 	char	*fencepath = (char *)arg;
8616 
8617 	/*
8618 	 * We have already decided to retire this device. The various
8619 	 * constraint checking should not be set.
8620 	 * NOTE that the retire flag may already be set due to
8621 	 * fenced -> detach -> fenced transitions.
8622 	 */
8623 	mutex_enter(&DEVI(dip)->devi_lock);
8624 	ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT));
8625 	ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_BLOCKED));
8626 	ASSERT(!(DEVI(dip)->devi_flags & DEVI_RETIRING));
8627 	DEVI(dip)->devi_flags |= DEVI_RETIRED;
8628 	mutex_exit(&DEVI(dip)->devi_lock);
8629 	RIO_VERBOSE((CE_NOTE, "marked as RETIRED dip=%p", (void *)dip));
8630 
8631 	if (fencepath) {
8632 		(void) spec_fence_snode(dip, NULL);
8633 		RIO_DEBUG((CE_NOTE, "Fenced: %s",
8634 		    ddi_pathname(dip, fencepath)));
8635 	}
8636 
8637 	return (DDI_WALK_CONTINUE);
8638 }
8639 
8640 /*
8641  * Checks the retire database and:
8642  *
8643  * - if device is present in the retire database, marks the device retired
8644  *   and fences it off.
8645  * - if device is not in retire database, allows the device to attach normally
8646  *
8647  * To be called only by framework attach code on first attach attempt.
8648  *
8649  */
8650 static int
8651 i_ddi_check_retire(dev_info_t *dip)
8652 {
8653 	char		*path;
8654 	dev_info_t	*pdip;
8655 	int		circ;
8656 	int		phci_only;
8657 	int		constraint;
8658 
8659 	pdip = ddi_get_parent(dip);
8660 
8661 	/*
8662 	 * Root dip is treated special and doesn't take this code path.
8663 	 * Also root can never be retired.
8664 	 */
8665 	ASSERT(pdip);
8666 	ASSERT(DEVI_BUSY_OWNED(pdip));
8667 	ASSERT(i_ddi_node_state(dip) < DS_ATTACHED);
8668 
8669 	path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
8670 
8671 	(void) ddi_pathname(dip, path);
8672 
8673 	RIO_VERBOSE((CE_NOTE, "Checking if dip should attach: dip=%p, path=%s",
8674 	    (void *)dip, path));
8675 
8676 	/*
8677 	 * Check if this device is in the "retired" store i.e.	should
8678 	 * be retired. If not, we have nothing to do.
8679 	 */
8680 	if (e_ddi_device_retired(path) == 0) {
8681 		RIO_VERBOSE((CE_NOTE, "device is NOT retired: path=%s", path));
8682 		if (DEVI(dip)->devi_flags & DEVI_RETIRED)
8683 			(void) e_ddi_unretire_device(path);
8684 		kmem_free(path, MAXPATHLEN);
8685 		return (0);
8686 	}
8687 
8688 	RIO_DEBUG((CE_NOTE, "attach: device is retired: path=%s", path));
8689 
8690 	/*
8691 	 * Mark dips and fence off snodes (if any)
8692 	 */
8693 	RIO_DEBUG((CE_NOTE, "attach: Mark and fence subtree: path=%s", path));
8694 	(void) mark_and_fence(dip, path);
8695 	if (!is_leaf_node(dip)) {
8696 		ndi_devi_enter(dip, &circ);
8697 		ddi_walk_devs(ddi_get_child(dip), mark_and_fence, path);
8698 		ndi_devi_exit(dip, circ);
8699 	}
8700 
8701 	kmem_free(path, MAXPATHLEN);
8702 
8703 	/*
8704 	 * We don't want to check the client. We just want to
8705 	 * offline the PHCI
8706 	 */
8707 	phci_only = 1;
8708 	constraint = 1;
8709 	if (MDI_PHCI(dip))
8710 		mdi_phci_retire_finalize(dip, phci_only, &constraint);
8711 	return (1);
8712 }
8713 
8714 
8715 #define	VAL_ALIAS(array, x)	(strlen(array[x].pair_alias))
8716 #define	VAL_CURR(array, x)	(strlen(array[x].pair_curr))
8717 #define	SWAP(array, x, y)			\
8718 {						\
8719 	alias_pair_t tmpair = array[x];		\
8720 	array[x] = array[y];			\
8721 	array[y] = tmpair;			\
8722 }
8723 
8724 static int
8725 partition_curr(alias_pair_t *array, int start, int end)
8726 {
8727 	int	i = start - 1;
8728 	int	j = end + 1;
8729 	int	pivot = start;
8730 
8731 	for (;;) {
8732 		do {
8733 			j--;
8734 		} while (VAL_CURR(array, j) > VAL_CURR(array, pivot));
8735 
8736 		do {
8737 			i++;
8738 		} while (VAL_CURR(array, i) < VAL_CURR(array, pivot));
8739 
8740 		if (i < j)
8741 			SWAP(array, i, j)
8742 		else
8743 			return (j);
8744 	}
8745 }
8746 
8747 static int
8748 partition_aliases(alias_pair_t *array, int start, int end)
8749 {
8750 	int	i = start - 1;
8751 	int	j = end + 1;
8752 	int	pivot = start;
8753 
8754 	for (;;) {
8755 		do {
8756 			j--;
8757 		} while (VAL_ALIAS(array, j) > VAL_ALIAS(array, pivot));
8758 
8759 		do {
8760 			i++;
8761 		} while (VAL_ALIAS(array, i) < VAL_ALIAS(array, pivot));
8762 
8763 		if (i < j)
8764 			SWAP(array, i, j)
8765 		else
8766 			return (j);
8767 	}
8768 }
8769 static void
8770 sort_alias_pairs(alias_pair_t *array, int start, int end)
8771 {
8772 	int mid;
8773 
8774 	if (start < end) {
8775 		mid = partition_aliases(array, start, end);
8776 		sort_alias_pairs(array, start, mid);
8777 		sort_alias_pairs(array, mid + 1, end);
8778 	}
8779 }
8780 
8781 static void
8782 sort_curr_pairs(alias_pair_t *array, int start, int end)
8783 {
8784 	int mid;
8785 
8786 	if (start < end) {
8787 		mid = partition_curr(array, start, end);
8788 		sort_curr_pairs(array, start, mid);
8789 		sort_curr_pairs(array, mid + 1, end);
8790 	}
8791 }
8792 
8793 static void
8794 create_sorted_pairs(plat_alias_t *pali, int npali)
8795 {
8796 	int		i;
8797 	int		j;
8798 	int		k;
8799 	int		count;
8800 
8801 	count = 0;
8802 	for (i = 0; i < npali; i++) {
8803 		count += pali[i].pali_naliases;
8804 	}
8805 
8806 	ddi_aliases.dali_alias_pairs = kmem_zalloc(
8807 	    (sizeof (alias_pair_t)) * count, KM_NOSLEEP);
8808 	if (ddi_aliases.dali_alias_pairs == NULL) {
8809 		cmn_err(CE_PANIC, "alias path-pair alloc failed");
8810 		/*NOTREACHED*/
8811 	}
8812 
8813 	ddi_aliases.dali_curr_pairs = kmem_zalloc(
8814 	    (sizeof (alias_pair_t)) * count, KM_NOSLEEP);
8815 	if (ddi_aliases.dali_curr_pairs == NULL) {
8816 		cmn_err(CE_PANIC, "curr path-pair alloc failed");
8817 		/*NOTREACHED*/
8818 	}
8819 
8820 	for (i = 0, k = 0; i < npali; i++) {
8821 		for (j = 0; j < pali[i].pali_naliases; j++, k++) {
8822 			ddi_aliases.dali_alias_pairs[k].pair_curr =
8823 			    ddi_aliases.dali_curr_pairs[k].pair_curr =
8824 			    pali[i].pali_current;
8825 			ddi_aliases.dali_alias_pairs[k].pair_alias =
8826 			    ddi_aliases.dali_curr_pairs[k].pair_alias =
8827 			    pali[i].pali_aliases[j];
8828 		}
8829 	}
8830 
8831 	ASSERT(k == count);
8832 
8833 	ddi_aliases.dali_num_pairs = count;
8834 
8835 	/* Now sort the array based on length of pair_alias */
8836 	sort_alias_pairs(ddi_aliases.dali_alias_pairs, 0, count - 1);
8837 	sort_curr_pairs(ddi_aliases.dali_curr_pairs, 0, count - 1);
8838 }
8839 
8840 void
8841 ddi_register_aliases(plat_alias_t *pali, uint64_t npali)
8842 {
8843 
8844 	ASSERT((pali == NULL) ^ (npali != 0));
8845 
8846 	if (npali == 0) {
8847 		ddi_err(DER_PANIC, NULL, "npali == 0");
8848 		/*NOTREACHED*/
8849 	}
8850 
8851 	if (ddi_aliases_present == B_TRUE) {
8852 		ddi_err(DER_PANIC, NULL, "multiple init");
8853 		/*NOTREACHED*/
8854 	}
8855 
8856 	ddi_aliases.dali_alias_TLB = mod_hash_create_strhash(
8857 	    "ddi-alias-tlb", DDI_ALIAS_HASH_SIZE, mod_hash_null_valdtor);
8858 	if (ddi_aliases.dali_alias_TLB == NULL) {
8859 		ddi_err(DER_PANIC, NULL, "alias TLB hash alloc failed");
8860 		/*NOTREACHED*/
8861 	}
8862 
8863 	ddi_aliases.dali_curr_TLB = mod_hash_create_strhash(
8864 	    "ddi-curr-tlb", DDI_ALIAS_HASH_SIZE, mod_hash_null_valdtor);
8865 	if (ddi_aliases.dali_curr_TLB == NULL) {
8866 		ddi_err(DER_PANIC, NULL, "curr TLB hash alloc failed");
8867 		/*NOTREACHED*/
8868 	}
8869 
8870 	create_sorted_pairs(pali, npali);
8871 
8872 	tsd_create(&tsd_ddi_redirect, NULL);
8873 
8874 	ddi_aliases_present = B_TRUE;
8875 }
8876 
8877 static dev_info_t *
8878 path_to_dip(char *path)
8879 {
8880 	dev_info_t	*currdip;
8881 	int		error;
8882 	char		*pdup;
8883 
8884 	pdup = ddi_strdup(path, KM_NOSLEEP);
8885 	if (pdup == NULL) {
8886 		cmn_err(CE_PANIC, "path strdup failed: %s", path);
8887 		/*NOTREACHED*/
8888 	}
8889 
8890 	error = resolve_pathname(pdup, &currdip, NULL, NULL);
8891 
8892 	kmem_free(pdup, strlen(path) + 1);
8893 
8894 	return (error ? NULL : currdip);
8895 }
8896 
8897 dev_info_t *
8898 ddi_alias_to_currdip(char *alias, int i)
8899 {
8900 	alias_pair_t *pair;
8901 	char *curr;
8902 	dev_info_t *currdip = NULL;
8903 	char *aliasdup;
8904 	int rv, len;
8905 
8906 	pair = &(ddi_aliases.dali_alias_pairs[i]);
8907 	len = strlen(pair->pair_alias);
8908 
8909 	curr = NULL;
8910 	aliasdup = ddi_strdup(alias, KM_NOSLEEP);
8911 	if (aliasdup == NULL) {
8912 		cmn_err(CE_PANIC, "aliasdup alloc failed");
8913 		/*NOTREACHED*/
8914 	}
8915 
8916 	if (strncmp(alias, pair->pair_alias, len)  != 0)
8917 		goto out;
8918 
8919 	if (alias[len] != '/' && alias[len] != '\0')
8920 		goto out;
8921 
8922 	curr = kmem_alloc(MAXPATHLEN, KM_NOSLEEP);
8923 	if (curr == NULL) {
8924 		cmn_err(CE_PANIC, "curr alloc failed");
8925 		/*NOTREACHED*/
8926 	}
8927 	(void) strlcpy(curr, pair->pair_curr, MAXPATHLEN);
8928 	if (alias[len] == '/') {
8929 		(void) strlcat(curr, "/", MAXPATHLEN);
8930 		(void) strlcat(curr, &alias[len + 1], MAXPATHLEN);
8931 	}
8932 
8933 	currdip = path_to_dip(curr);
8934 
8935 out:
8936 	if (currdip) {
8937 		rv = mod_hash_insert(ddi_aliases.dali_alias_TLB,
8938 		    (mod_hash_key_t)aliasdup, (mod_hash_val_t)curr);
8939 		if (rv != 0) {
8940 			kmem_free(curr, MAXPATHLEN);
8941 			strfree(aliasdup);
8942 		}
8943 	} else {
8944 		rv = mod_hash_insert(ddi_aliases.dali_alias_TLB,
8945 		    (mod_hash_key_t)aliasdup, (mod_hash_val_t)NULL);
8946 		if (rv != 0) {
8947 			strfree(aliasdup);
8948 		}
8949 		if (curr)
8950 			kmem_free(curr, MAXPATHLEN);
8951 	}
8952 
8953 	return (currdip);
8954 }
8955 
8956 char *
8957 ddi_curr_to_alias(char *curr, int i)
8958 {
8959 	alias_pair_t	*pair;
8960 	char		*alias;
8961 	char		*currdup;
8962 	int		len;
8963 	int		rv;
8964 
8965 	pair = &(ddi_aliases.dali_curr_pairs[i]);
8966 
8967 	len = strlen(pair->pair_curr);
8968 
8969 	alias = NULL;
8970 
8971 	currdup = ddi_strdup(curr, KM_NOSLEEP);
8972 	if (currdup == NULL) {
8973 		cmn_err(CE_PANIC, "currdup alloc failed");
8974 		/*NOTREACHED*/
8975 	}
8976 
8977 	if (strncmp(curr, pair->pair_curr, len) != 0)
8978 		goto out;
8979 
8980 	if (curr[len] != '/' && curr[len] != '\0')
8981 		goto out;
8982 
8983 	alias = kmem_alloc(MAXPATHLEN, KM_NOSLEEP);
8984 	if (alias == NULL) {
8985 		cmn_err(CE_PANIC, "alias alloc failed");
8986 		/*NOTREACHED*/
8987 	}
8988 
8989 	(void) strlcpy(alias, pair->pair_alias, MAXPATHLEN);
8990 	if (curr[len] == '/') {
8991 		(void) strlcat(alias, "/", MAXPATHLEN);
8992 		(void) strlcat(alias, &curr[len + 1], MAXPATHLEN);
8993 	}
8994 
8995 	if (e_ddi_path_to_instance(alias) == NULL) {
8996 		kmem_free(alias, MAXPATHLEN);
8997 		alias = NULL;
8998 	}
8999 
9000 out:
9001 	rv = mod_hash_insert(ddi_aliases.dali_curr_TLB,
9002 	    (mod_hash_key_t)currdup, (mod_hash_val_t)alias);
9003 	if (rv != 0) {
9004 		strfree(currdup);
9005 	}
9006 
9007 	return (alias);
9008 }
9009 
9010 dev_info_t *
9011 ddi_alias_redirect(char *alias)
9012 {
9013 	char		*curr;
9014 	dev_info_t	*currdip;
9015 	int		i;
9016 
9017 	if (ddi_aliases_present == B_FALSE)
9018 		return (NULL);
9019 
9020 	if (tsd_get(tsd_ddi_redirect))
9021 		return (NULL);
9022 
9023 	(void) tsd_set(tsd_ddi_redirect, (void *)1);
9024 
9025 	ASSERT(ddi_aliases.dali_alias_TLB);
9026 	ASSERT(ddi_aliases.dali_alias_pairs);
9027 
9028 	curr = NULL;
9029 	if (mod_hash_find(ddi_aliases.dali_alias_TLB,
9030 	    (mod_hash_key_t)alias, (mod_hash_val_t *)&curr) == 0) {
9031 		currdip = curr ? path_to_dip(curr) : NULL;
9032 		goto out;
9033 	}
9034 
9035 	/* The TLB has no translation, do it the hard way */
9036 	currdip = NULL;
9037 	for (i = ddi_aliases.dali_num_pairs - 1; i >= 0; i--) {
9038 		currdip = ddi_alias_to_currdip(alias, i);
9039 		if (currdip)
9040 			break;
9041 	}
9042 out:
9043 	(void) tsd_set(tsd_ddi_redirect, NULL);
9044 
9045 	return (currdip);
9046 }
9047 
9048 char *
9049 ddi_curr_redirect(char *curr)
9050 {
9051 	char 	*alias;
9052 	int i;
9053 
9054 	if (ddi_aliases_present == B_FALSE)
9055 		return (NULL);
9056 
9057 	if (tsd_get(tsd_ddi_redirect))
9058 		return (NULL);
9059 
9060 	(void) tsd_set(tsd_ddi_redirect, (void *)1);
9061 
9062 	ASSERT(ddi_aliases.dali_curr_TLB);
9063 	ASSERT(ddi_aliases.dali_curr_pairs);
9064 
9065 	alias = NULL;
9066 	if (mod_hash_find(ddi_aliases.dali_curr_TLB,
9067 	    (mod_hash_key_t)curr, (mod_hash_val_t *)&alias) == 0) {
9068 		goto out;
9069 	}
9070 
9071 
9072 	/* The TLB has no translation, do it the slow way */
9073 	alias = NULL;
9074 	for (i = ddi_aliases.dali_num_pairs - 1; i >= 0; i--) {
9075 		alias = ddi_curr_to_alias(curr, i);
9076 		if (alias)
9077 			break;
9078 	}
9079 
9080 out:
9081 	(void) tsd_set(tsd_ddi_redirect, NULL);
9082 
9083 	return (alias);
9084 }
9085 
9086 void
9087 ddi_err(ddi_err_t ade, dev_info_t *rdip, const char *fmt, ...)
9088 {
9089 	va_list ap;
9090 	char strbuf[256];
9091 	char *buf;
9092 	size_t buflen, tlen;
9093 	int ce;
9094 	int de;
9095 	const char *fmtbad = "Invalid arguments to ddi_err()";
9096 
9097 	de = DER_CONT;
9098 	strbuf[1] = '\0';
9099 
9100 	switch (ade) {
9101 	case DER_CONS:
9102 		strbuf[0] = '^';
9103 		break;
9104 	case DER_LOG:
9105 		strbuf[0] = '!';
9106 		break;
9107 	case DER_VERB:
9108 		strbuf[0] = '?';
9109 		break;
9110 	default:
9111 		strbuf[0] = '\0';
9112 		de = ade;
9113 		break;
9114 	}
9115 
9116 	tlen = strlen(strbuf);
9117 	buf = strbuf + tlen;
9118 	buflen = sizeof (strbuf) - tlen;
9119 
9120 	if (rdip && ddi_get_instance(rdip) == -1) {
9121 		(void) snprintf(buf, buflen, "%s: ",
9122 		    ddi_driver_name(rdip));
9123 	} else if (rdip) {
9124 		(void) snprintf(buf, buflen, "%s%d: ",
9125 		    ddi_driver_name(rdip), ddi_get_instance(rdip));
9126 	}
9127 
9128 	tlen = strlen(strbuf);
9129 	buf = strbuf + tlen;
9130 	buflen = sizeof (strbuf) - tlen;
9131 
9132 	va_start(ap, fmt);
9133 	switch (de) {
9134 	case DER_CONT:
9135 		(void) vsnprintf(buf, buflen, fmt, ap);
9136 		if (ade != DER_CONT) {
9137 			(void) strlcat(strbuf, "\n", sizeof (strbuf));
9138 		}
9139 		ce = CE_CONT;
9140 		break;
9141 	case DER_NOTE:
9142 		(void) vsnprintf(buf, buflen, fmt, ap);
9143 		ce = CE_NOTE;
9144 		break;
9145 	case DER_WARN:
9146 		(void) vsnprintf(buf, buflen, fmt, ap);
9147 		ce = CE_WARN;
9148 		break;
9149 	case DER_MODE:
9150 		(void) vsnprintf(buf, buflen, fmt, ap);
9151 		if (ddi_err_panic == B_TRUE) {
9152 			ce = CE_PANIC;
9153 		} else {
9154 			ce = CE_WARN;
9155 		}
9156 		break;
9157 	case DER_DEBUG:
9158 		(void) snprintf(buf, buflen, "DEBUG: ");
9159 		tlen = strlen("DEBUG: ");
9160 		(void) vsnprintf(buf + tlen, buflen - tlen, fmt, ap);
9161 		ce = CE_CONT;
9162 		break;
9163 	case DER_PANIC:
9164 		(void) vsnprintf(buf, buflen, fmt, ap);
9165 		ce = CE_PANIC;
9166 		break;
9167 	case DER_INVALID:
9168 	default:
9169 		(void) snprintf(buf, buflen, fmtbad);
9170 		tlen = strlen(fmtbad);
9171 		(void) vsnprintf(buf + tlen, buflen - tlen, fmt, ap);
9172 		ce = CE_PANIC;
9173 		break;
9174 	}
9175 	va_end(ap);
9176 
9177 	cmn_err(ce, strbuf);
9178 }
9179 
9180 /*ARGSUSED*/
9181 void
9182 ddi_mem_update(uint64_t addr, uint64_t size)
9183 {
9184 #if defined(__x86) && !defined(__xpv)
9185 	extern void immu_physmem_update(uint64_t addr, uint64_t size);
9186 	immu_physmem_update(addr, size);
9187 #else
9188 	/*LINTED*/
9189 	;
9190 #endif
9191 }
9192