xref: /titanic_41/usr/src/uts/common/os/sunmdi.c (revision 4496171313bed39e96f21bc2f9faf2868e267ae3)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 #pragma ident	"%Z%%M%	%I%	%E% SMI"
26 
27 /*
28  * Multipath driver interface (MDI) implementation; see mdi_impl.h for a more
29  * detailed discussion of the overall mpxio architecture.
30  *
31  * Default locking order:
32  *
33  * _NOTE(LOCK_ORDER(mdi_mutex, mdi_vhci:vh_phci_mutex);
34  * _NOTE(LOCK_ORDER(mdi_mutex, mdi_vhci:vh_client_mutex);
35  * _NOTE(LOCK_ORDER(mdi_vhci:vh_phci_mutex, mdi_phci::ph_mutex);
36  * _NOTE(LOCK_ORDER(mdi_vhci:vh_client_mutex, mdi_client::ct_mutex);
37  * _NOTE(LOCK_ORDER(mdi_phci::ph_mutex mdi_pathinfo::pi_mutex))
38  * _NOTE(LOCK_ORDER(mdi_phci::ph_mutex mdi_client::ct_mutex))
39  * _NOTE(LOCK_ORDER(mdi_client::ct_mutex mdi_pathinfo::pi_mutex))
40  */
41 
42 #include <sys/note.h>
43 #include <sys/types.h>
44 #include <sys/varargs.h>
45 #include <sys/param.h>
46 #include <sys/errno.h>
47 #include <sys/uio.h>
48 #include <sys/buf.h>
49 #include <sys/modctl.h>
50 #include <sys/open.h>
51 #include <sys/kmem.h>
52 #include <sys/poll.h>
53 #include <sys/conf.h>
54 #include <sys/bootconf.h>
55 #include <sys/cmn_err.h>
56 #include <sys/stat.h>
57 #include <sys/ddi.h>
58 #include <sys/sunddi.h>
59 #include <sys/ddipropdefs.h>
60 #include <sys/sunndi.h>
61 #include <sys/ndi_impldefs.h>
62 #include <sys/promif.h>
63 #include <sys/sunmdi.h>
64 #include <sys/mdi_impldefs.h>
65 #include <sys/taskq.h>
66 #include <sys/epm.h>
67 #include <sys/sunpm.h>
68 #include <sys/modhash.h>
69 #include <sys/disp.h>
70 #include <sys/autoconf.h>
71 #include <sys/sysmacros.h>
72 
73 #ifdef	DEBUG
74 #include <sys/debug.h>
75 int	mdi_debug = 1;
76 int	mdi_debug_logonly = 0;
77 #define	MDI_DEBUG(level, stmnt) \
78 	    if (mdi_debug >= (level)) i_mdi_log stmnt
79 static void i_mdi_log(int, dev_info_t *, const char *fmt, ...);
80 #else	/* !DEBUG */
81 #define	MDI_DEBUG(level, stmnt)
82 #endif	/* DEBUG */
83 
84 extern pri_t	minclsyspri;
85 extern int	modrootloaded;
86 
87 /*
88  * Global mutex:
89  * Protects vHCI list and structure members.
90  */
91 kmutex_t	mdi_mutex;
92 
93 /*
94  * Registered vHCI class driver lists
95  */
96 int		mdi_vhci_count;
97 mdi_vhci_t	*mdi_vhci_head;
98 mdi_vhci_t	*mdi_vhci_tail;
99 
100 /*
101  * Client Hash Table size
102  */
103 static int	mdi_client_table_size = CLIENT_HASH_TABLE_SIZE;
104 
105 /*
106  * taskq interface definitions
107  */
108 #define	MDI_TASKQ_N_THREADS	8
109 #define	MDI_TASKQ_PRI		minclsyspri
110 #define	MDI_TASKQ_MINALLOC	(4*mdi_taskq_n_threads)
111 #define	MDI_TASKQ_MAXALLOC	(500*mdi_taskq_n_threads)
112 
113 taskq_t				*mdi_taskq;
114 static uint_t			mdi_taskq_n_threads = MDI_TASKQ_N_THREADS;
115 
116 #define	TICKS_PER_SECOND	(drv_usectohz(1000000))
117 
118 /*
119  * The data should be "quiet" for this interval (in seconds) before the
120  * vhci cached data is flushed to the disk.
121  */
122 static int mdi_vhcache_flush_delay = 10;
123 
124 /* number of seconds the vhcache flush daemon will sleep idle before exiting */
125 static int mdi_vhcache_flush_daemon_idle_time = 60;
126 
127 /*
128  * MDI falls back to discovery of all paths when a bus_config_one fails.
129  * The following parameters can be used to tune this operation.
130  *
131  * mdi_path_discovery_boot
132  *	Number of times path discovery will be attempted during early boot.
133  *	Probably there is no reason to ever set this value to greater than one.
134  *
135  * mdi_path_discovery_postboot
136  *	Number of times path discovery will be attempted after early boot.
137  *	Set it to a minimum of two to allow for discovery of iscsi paths which
138  *	may happen very late during booting.
139  *
140  * mdi_path_discovery_interval
141  *	Minimum number of seconds MDI will wait between successive discovery
142  *	of all paths. Set it to -1 to disable discovery of all paths.
143  */
144 static int mdi_path_discovery_boot = 1;
145 static int mdi_path_discovery_postboot = 2;
146 static int mdi_path_discovery_interval = 10;
147 
148 /*
149  * number of seconds the asynchronous configuration thread will sleep idle
150  * before exiting.
151  */
152 static int mdi_async_config_idle_time = 600;
153 
154 static int mdi_bus_config_cache_hash_size = 256;
155 
156 /* turns off multithreaded configuration for certain operations */
157 static int mdi_mtc_off = 0;
158 
159 /*
160  * MDI component property name/value string definitions
161  */
162 const char 		*mdi_component_prop = "mpxio-component";
163 const char		*mdi_component_prop_vhci = "vhci";
164 const char		*mdi_component_prop_phci = "phci";
165 const char		*mdi_component_prop_client = "client";
166 
167 /*
168  * MDI client global unique identifier property name
169  */
170 const char		*mdi_client_guid_prop = "client-guid";
171 
172 /*
173  * MDI client load balancing property name/value string definitions
174  */
175 const char		*mdi_load_balance = "load-balance";
176 const char		*mdi_load_balance_none = "none";
177 const char		*mdi_load_balance_rr = "round-robin";
178 const char		*mdi_load_balance_lba = "logical-block";
179 
180 /*
181  * Obsolete vHCI class definition; to be removed after Leadville update
182  */
183 const char *mdi_vhci_class_scsi = MDI_HCI_CLASS_SCSI;
184 
185 static char vhci_greeting[] =
186 	"\tThere already exists one vHCI driver for class %s\n"
187 	"\tOnly one vHCI driver for each class is allowed\n";
188 
189 /*
190  * Static function prototypes
191  */
192 static int		i_mdi_phci_offline(dev_info_t *, uint_t);
193 static int		i_mdi_client_offline(dev_info_t *, uint_t);
194 static int		i_mdi_phci_pre_detach(dev_info_t *, ddi_detach_cmd_t);
195 static void		i_mdi_phci_post_detach(dev_info_t *,
196 			    ddi_detach_cmd_t, int);
197 static int		i_mdi_client_pre_detach(dev_info_t *,
198 			    ddi_detach_cmd_t);
199 static void		i_mdi_client_post_detach(dev_info_t *,
200 			    ddi_detach_cmd_t, int);
201 static void		i_mdi_pm_hold_pip(mdi_pathinfo_t *);
202 static void		i_mdi_pm_rele_pip(mdi_pathinfo_t *);
203 static int 		i_mdi_lba_lb(mdi_client_t *ct,
204 			    mdi_pathinfo_t **ret_pip, struct buf *buf);
205 static void		i_mdi_pm_hold_client(mdi_client_t *, int);
206 static void		i_mdi_pm_rele_client(mdi_client_t *, int);
207 static void		i_mdi_pm_reset_client(mdi_client_t *);
208 static int		i_mdi_power_all_phci(mdi_client_t *);
209 static void		i_mdi_log_sysevent(dev_info_t *, char *, char *);
210 
211 
212 /*
213  * Internal mdi_pathinfo node functions
214  */
215 static void		i_mdi_pi_kstat_destroy(mdi_pathinfo_t *);
216 
217 static mdi_vhci_t	*i_mdi_vhci_class2vhci(char *);
218 static mdi_vhci_t	*i_devi_get_vhci(dev_info_t *);
219 static mdi_phci_t	*i_devi_get_phci(dev_info_t *);
220 static void		i_mdi_phci_lock(mdi_phci_t *, mdi_pathinfo_t *);
221 static void		i_mdi_phci_unlock(mdi_phci_t *);
222 static mdi_pathinfo_t	*i_mdi_pi_alloc(mdi_phci_t *, char *, mdi_client_t *);
223 static void		i_mdi_phci_add_path(mdi_phci_t *, mdi_pathinfo_t *);
224 static void		i_mdi_client_add_path(mdi_client_t *, mdi_pathinfo_t *);
225 static void		i_mdi_pi_free(mdi_phci_t *ph, mdi_pathinfo_t *,
226 			    mdi_client_t *);
227 static void		i_mdi_phci_remove_path(mdi_phci_t *, mdi_pathinfo_t *);
228 static void		i_mdi_client_remove_path(mdi_client_t *,
229 			    mdi_pathinfo_t *);
230 
231 static int		i_mdi_pi_state_change(mdi_pathinfo_t *,
232 			    mdi_pathinfo_state_t, int);
233 static int		i_mdi_pi_offline(mdi_pathinfo_t *, int);
234 static dev_info_t	*i_mdi_devinfo_create(mdi_vhci_t *, char *, char *,
235 			    char **, int);
236 static dev_info_t	*i_mdi_devinfo_find(mdi_vhci_t *, char *, char *);
237 static int		i_mdi_devinfo_remove(dev_info_t *, dev_info_t *, int);
238 static int		i_mdi_is_child_present(dev_info_t *, dev_info_t *);
239 static mdi_client_t	*i_mdi_client_alloc(mdi_vhci_t *, char *, char *);
240 static void		i_mdi_client_enlist_table(mdi_vhci_t *, mdi_client_t *);
241 static void		i_mdi_client_delist_table(mdi_vhci_t *, mdi_client_t *);
242 static mdi_client_t	*i_mdi_client_find(mdi_vhci_t *, char *, char *);
243 static void		i_mdi_client_update_state(mdi_client_t *);
244 static int		i_mdi_client_compute_state(mdi_client_t *,
245 			    mdi_phci_t *);
246 static void		i_mdi_client_lock(mdi_client_t *, mdi_pathinfo_t *);
247 static void		i_mdi_client_unlock(mdi_client_t *);
248 static int		i_mdi_client_free(mdi_vhci_t *, mdi_client_t *);
249 static mdi_client_t	*i_devi_get_client(dev_info_t *);
250 /*
251  * NOTE: this will be removed once the NWS files are changed to use the new
252  * mdi_{enable,disable}_path interfaces
253  */
254 static int		i_mdi_pi_enable_disable(dev_info_t *, dev_info_t *,
255 				int, int);
256 static mdi_pathinfo_t 	*i_mdi_enable_disable_path(mdi_pathinfo_t *pip,
257 				mdi_vhci_t *vh, int flags, int op);
258 /*
259  * Failover related function prototypes
260  */
261 static int		i_mdi_failover(void *);
262 
263 /*
264  * misc internal functions
265  */
266 static int		i_mdi_get_hash_key(char *);
267 static int		i_map_nvlist_error_to_mdi(int);
268 static void		i_mdi_report_path_state(mdi_client_t *,
269 			    mdi_pathinfo_t *);
270 
271 static void		setup_vhci_cache(mdi_vhci_t *);
272 static int		destroy_vhci_cache(mdi_vhci_t *);
273 static int		stop_vhcache_async_threads(mdi_vhci_config_t *);
274 static boolean_t	stop_vhcache_flush_thread(void *, int);
275 static void		free_string_array(char **, int);
276 static void		free_vhcache_phci(mdi_vhcache_phci_t *);
277 static void		free_vhcache_pathinfo(mdi_vhcache_pathinfo_t *);
278 static void		free_vhcache_client(mdi_vhcache_client_t *);
279 static int		mainnvl_to_vhcache(mdi_vhci_cache_t *, nvlist_t *);
280 static nvlist_t		*vhcache_to_mainnvl(mdi_vhci_cache_t *);
281 static void		vhcache_phci_add(mdi_vhci_config_t *, mdi_phci_t *);
282 static void		vhcache_phci_remove(mdi_vhci_config_t *, mdi_phci_t *);
283 static void		vhcache_pi_add(mdi_vhci_config_t *,
284 			    struct mdi_pathinfo *);
285 static void		vhcache_pi_remove(mdi_vhci_config_t *,
286 			    struct mdi_pathinfo *);
287 static void		free_phclient_path_list(mdi_phys_path_t *);
288 static void		sort_vhcache_paths(mdi_vhcache_client_t *);
289 static int		flush_vhcache(mdi_vhci_config_t *, int);
290 static void		vhcache_dirty(mdi_vhci_config_t *);
291 static void		free_async_client_config(mdi_async_client_config_t *);
292 static void		single_threaded_vhconfig_enter(mdi_vhci_config_t *);
293 static void		single_threaded_vhconfig_exit(mdi_vhci_config_t *);
294 static nvlist_t		*read_on_disk_vhci_cache(char *);
295 extern int		fread_nvlist(char *, nvlist_t **);
296 extern int		fwrite_nvlist(char *, nvlist_t *);
297 
298 /* called once when first vhci registers with mdi */
299 static void
300 i_mdi_init()
301 {
302 	static int initialized = 0;
303 
304 	if (initialized)
305 		return;
306 	initialized = 1;
307 
308 	mutex_init(&mdi_mutex, NULL, MUTEX_DEFAULT, NULL);
309 	/*
310 	 * Create our taskq resources
311 	 */
312 	mdi_taskq = taskq_create("mdi_taskq", mdi_taskq_n_threads,
313 	    MDI_TASKQ_PRI, MDI_TASKQ_MINALLOC, MDI_TASKQ_MAXALLOC,
314 	    TASKQ_PREPOPULATE | TASKQ_CPR_SAFE);
315 	ASSERT(mdi_taskq != NULL);	/* taskq_create never fails */
316 }
317 
318 /*
319  * mdi_get_component_type():
320  *		Return mpxio component type
321  * Return Values:
322  *		MDI_COMPONENT_NONE
323  *		MDI_COMPONENT_VHCI
324  *		MDI_COMPONENT_PHCI
325  *		MDI_COMPONENT_CLIENT
326  * XXX This doesn't work under multi-level MPxIO and should be
327  *	removed when clients migrate mdi_component_is_*() interfaces.
328  */
329 int
330 mdi_get_component_type(dev_info_t *dip)
331 {
332 	return (DEVI(dip)->devi_mdi_component);
333 }
334 
335 /*
336  * mdi_vhci_register():
337  *		Register a vHCI module with the mpxio framework
338  *		mdi_vhci_register() is called by vHCI drivers to register the
339  *		'class_driver' vHCI driver and its MDI entrypoints with the
340  *		mpxio framework.  The vHCI driver must call this interface as
341  *		part of its attach(9e) handler.
342  *		Competing threads may try to attach mdi_vhci_register() as
343  *		the vHCI drivers are loaded and attached as a result of pHCI
344  *		driver instance registration (mdi_phci_register()) with the
345  *		framework.
346  * Return Values:
347  *		MDI_SUCCESS
348  *		MDI_FAILURE
349  */
350 /*ARGSUSED*/
351 int
352 mdi_vhci_register(char *class, dev_info_t *vdip, mdi_vhci_ops_t *vops,
353     int flags)
354 {
355 	mdi_vhci_t		*vh = NULL;
356 
357 	ASSERT(vops->vo_revision == MDI_VHCI_OPS_REV);
358 	ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(vdip)));
359 
360 	i_mdi_init();
361 
362 	mutex_enter(&mdi_mutex);
363 	/*
364 	 * Scan for already registered vhci
365 	 */
366 	for (vh = mdi_vhci_head; vh != NULL; vh = vh->vh_next) {
367 		if (strcmp(vh->vh_class, class) == 0) {
368 			/*
369 			 * vHCI has already been created.  Check for valid
370 			 * vHCI ops registration.  We only support one vHCI
371 			 * module per class
372 			 */
373 			if (vh->vh_ops != NULL) {
374 				mutex_exit(&mdi_mutex);
375 				cmn_err(CE_NOTE, vhci_greeting, class);
376 				return (MDI_FAILURE);
377 			}
378 			break;
379 		}
380 	}
381 
382 	/*
383 	 * if not yet created, create the vHCI component
384 	 */
385 	if (vh == NULL) {
386 		struct client_hash	*hash = NULL;
387 		char			*load_balance;
388 
389 		/*
390 		 * Allocate and initialize the mdi extensions
391 		 */
392 		vh = kmem_zalloc(sizeof (mdi_vhci_t), KM_SLEEP);
393 		hash = kmem_zalloc(mdi_client_table_size * sizeof (*hash),
394 		    KM_SLEEP);
395 		vh->vh_client_table = hash;
396 		vh->vh_class = kmem_zalloc(strlen(class) + 1, KM_SLEEP);
397 		(void) strcpy(vh->vh_class, class);
398 		vh->vh_lb = LOAD_BALANCE_RR;
399 		if (ddi_prop_lookup_string(DDI_DEV_T_ANY, vdip,
400 		    0, LOAD_BALANCE_PROP, &load_balance) == DDI_SUCCESS) {
401 			if (strcmp(load_balance, LOAD_BALANCE_PROP_NONE) == 0) {
402 				vh->vh_lb = LOAD_BALANCE_NONE;
403 			} else if (strcmp(load_balance, LOAD_BALANCE_PROP_LBA)
404 				    == 0) {
405 				vh->vh_lb = LOAD_BALANCE_LBA;
406 			}
407 			ddi_prop_free(load_balance);
408 		}
409 
410 		mutex_init(&vh->vh_phci_mutex, NULL, MUTEX_DEFAULT, NULL);
411 		mutex_init(&vh->vh_client_mutex, NULL, MUTEX_DEFAULT, NULL);
412 
413 		/*
414 		 * Store the vHCI ops vectors
415 		 */
416 		vh->vh_dip = vdip;
417 		vh->vh_ops = vops;
418 
419 		setup_vhci_cache(vh);
420 
421 		if (mdi_vhci_head == NULL) {
422 			mdi_vhci_head = vh;
423 		}
424 		if (mdi_vhci_tail) {
425 			mdi_vhci_tail->vh_next = vh;
426 		}
427 		mdi_vhci_tail = vh;
428 		mdi_vhci_count++;
429 	}
430 
431 	/*
432 	 * Claim the devfs node as a vhci component
433 	 */
434 	DEVI(vdip)->devi_mdi_component |= MDI_COMPONENT_VHCI;
435 
436 	/*
437 	 * Initialize our back reference from dev_info node
438 	 */
439 	DEVI(vdip)->devi_mdi_xhci = (caddr_t)vh;
440 	mutex_exit(&mdi_mutex);
441 	return (MDI_SUCCESS);
442 }
443 
444 /*
445  * mdi_vhci_unregister():
446  *		Unregister a vHCI module from mpxio framework
447  *		mdi_vhci_unregister() is called from the detach(9E) entrypoint
448  * 		of a vhci to unregister it from the framework.
449  * Return Values:
450  *		MDI_SUCCESS
451  *		MDI_FAILURE
452  */
453 /*ARGSUSED*/
454 int
455 mdi_vhci_unregister(dev_info_t *vdip, int flags)
456 {
457 	mdi_vhci_t	*found, *vh, *prev = NULL;
458 
459 	ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(vdip)));
460 
461 	/*
462 	 * Check for invalid VHCI
463 	 */
464 	if ((vh = i_devi_get_vhci(vdip)) == NULL)
465 		return (MDI_FAILURE);
466 
467 	/*
468 	 * Scan the list of registered vHCIs for a match
469 	 */
470 	mutex_enter(&mdi_mutex);
471 	for (found = mdi_vhci_head; found != NULL; found = found->vh_next) {
472 		if (found == vh)
473 			break;
474 		prev = found;
475 	}
476 
477 	if (found == NULL) {
478 		mutex_exit(&mdi_mutex);
479 		return (MDI_FAILURE);
480 	}
481 
482 	/*
483 	 * Check the vHCI, pHCI and client count. All the pHCIs and clients
484 	 * should have been unregistered, before a vHCI can be
485 	 * unregistered.
486 	 */
487 	MDI_VHCI_PHCI_LOCK(vh);
488 	if (vh->vh_refcnt || vh->vh_phci_count || vh->vh_client_count) {
489 		MDI_VHCI_PHCI_UNLOCK(vh);
490 		mutex_exit(&mdi_mutex);
491 		return (MDI_FAILURE);
492 	}
493 	MDI_VHCI_PHCI_UNLOCK(vh);
494 
495 	if (destroy_vhci_cache(vh) != MDI_SUCCESS) {
496 		mutex_exit(&mdi_mutex);
497 		return (MDI_FAILURE);
498 	}
499 
500 	/*
501 	 * Remove the vHCI from the global list
502 	 */
503 	if (vh == mdi_vhci_head) {
504 		mdi_vhci_head = vh->vh_next;
505 	} else {
506 		prev->vh_next = vh->vh_next;
507 	}
508 	if (vh == mdi_vhci_tail) {
509 		mdi_vhci_tail = prev;
510 	}
511 	mdi_vhci_count--;
512 	mutex_exit(&mdi_mutex);
513 
514 	vh->vh_ops = NULL;
515 	DEVI(vdip)->devi_mdi_component &= ~MDI_COMPONENT_VHCI;
516 	DEVI(vdip)->devi_mdi_xhci = NULL;
517 	kmem_free(vh->vh_class, strlen(vh->vh_class)+1);
518 	kmem_free(vh->vh_client_table,
519 	    mdi_client_table_size * sizeof (struct client_hash));
520 	mutex_destroy(&vh->vh_phci_mutex);
521 	mutex_destroy(&vh->vh_client_mutex);
522 
523 	kmem_free(vh, sizeof (mdi_vhci_t));
524 	return (MDI_SUCCESS);
525 }
526 
527 /*
528  * i_mdi_vhci_class2vhci():
529  *		Look for a matching vHCI module given a vHCI class name
530  * Return Values:
531  *		Handle to a vHCI component
532  *		NULL
533  */
534 static mdi_vhci_t *
535 i_mdi_vhci_class2vhci(char *class)
536 {
537 	mdi_vhci_t	*vh = NULL;
538 
539 	ASSERT(!MUTEX_HELD(&mdi_mutex));
540 
541 	mutex_enter(&mdi_mutex);
542 	for (vh = mdi_vhci_head; vh != NULL; vh = vh->vh_next) {
543 		if (strcmp(vh->vh_class, class) == 0) {
544 			break;
545 		}
546 	}
547 	mutex_exit(&mdi_mutex);
548 	return (vh);
549 }
550 
551 /*
552  * i_devi_get_vhci():
553  *		Utility function to get the handle to a vHCI component
554  * Return Values:
555  *		Handle to a vHCI component
556  *		NULL
557  */
558 mdi_vhci_t *
559 i_devi_get_vhci(dev_info_t *vdip)
560 {
561 	mdi_vhci_t	*vh = NULL;
562 	if (MDI_VHCI(vdip)) {
563 		vh = (mdi_vhci_t *)DEVI(vdip)->devi_mdi_xhci;
564 	}
565 	return (vh);
566 }
567 
568 /*
569  * mdi_phci_register():
570  *		Register a pHCI module with mpxio framework
571  *		mdi_phci_register() is called by pHCI drivers to register with
572  *		the mpxio framework and a specific 'class_driver' vHCI.  The
573  *		pHCI driver must call this interface as part of its attach(9e)
574  *		handler.
575  * Return Values:
576  *		MDI_SUCCESS
577  *		MDI_FAILURE
578  */
579 /*ARGSUSED*/
580 int
581 mdi_phci_register(char *class, dev_info_t *pdip, int flags)
582 {
583 	mdi_phci_t		*ph;
584 	mdi_vhci_t		*vh;
585 	char			*data;
586 	char			*pathname;
587 
588 	/*
589 	 * Some subsystems, like fcp, perform pHCI registration from a
590 	 * different thread than the one doing the pHCI attach(9E) - the
591 	 * driver attach code is waiting for this other thread to complete.
592 	 * This means we can only ASSERT DEVI_BUSY_CHANGING of parent
593 	 * (indicating that some thread has done an ndi_devi_enter of parent)
594 	 * not DEVI_BUSY_OWNED (which would indicate that we did the enter).
595 	 */
596 	ASSERT(DEVI_BUSY_CHANGING(ddi_get_parent(pdip)));
597 
598 	pathname = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
599 	(void) ddi_pathname(pdip, pathname);
600 
601 	/*
602 	 * Check for mpxio-disable property. Enable mpxio if the property is
603 	 * missing or not set to "yes".
604 	 * If the property is set to "yes" then emit a brief message.
605 	 */
606 	if ((ddi_prop_lookup_string(DDI_DEV_T_ANY, pdip, 0, "mpxio-disable",
607 	    &data) == DDI_SUCCESS)) {
608 		if (strcmp(data, "yes") == 0) {
609 			MDI_DEBUG(1, (CE_CONT, pdip,
610 			    "?%s (%s%d) multipath capabilities "
611 			    "disabled via %s.conf.\n", pathname,
612 			    ddi_driver_name(pdip), ddi_get_instance(pdip),
613 			    ddi_driver_name(pdip)));
614 			ddi_prop_free(data);
615 			kmem_free(pathname, MAXPATHLEN);
616 			return (MDI_FAILURE);
617 		}
618 		ddi_prop_free(data);
619 	}
620 
621 	kmem_free(pathname, MAXPATHLEN);
622 
623 	/*
624 	 * Search for a matching vHCI
625 	 */
626 	vh = (mdi_vhci_t *)i_mdi_vhci_class2vhci(class);
627 	if (vh == NULL) {
628 		return (MDI_FAILURE);
629 	}
630 
631 	ph = kmem_zalloc(sizeof (mdi_phci_t), KM_SLEEP);
632 	mutex_init(&ph->ph_mutex, NULL, MUTEX_DEFAULT, NULL);
633 	ph->ph_dip = pdip;
634 	ph->ph_vhci = vh;
635 	ph->ph_next = NULL;
636 	ph->ph_unstable = 0;
637 	ph->ph_vprivate = 0;
638 	cv_init(&ph->ph_unstable_cv, NULL, CV_DRIVER, NULL);
639 
640 	MDI_PHCI_LOCK(ph);
641 	MDI_PHCI_SET_POWER_UP(ph);
642 	MDI_PHCI_UNLOCK(ph);
643 	DEVI(pdip)->devi_mdi_component |= MDI_COMPONENT_PHCI;
644 	DEVI(pdip)->devi_mdi_xhci = (caddr_t)ph;
645 
646 	vhcache_phci_add(vh->vh_config, ph);
647 
648 	MDI_VHCI_PHCI_LOCK(vh);
649 	if (vh->vh_phci_head == NULL) {
650 		vh->vh_phci_head = ph;
651 	}
652 	if (vh->vh_phci_tail) {
653 		vh->vh_phci_tail->ph_next = ph;
654 	}
655 	vh->vh_phci_tail = ph;
656 	vh->vh_phci_count++;
657 	MDI_VHCI_PHCI_UNLOCK(vh);
658 
659 	i_mdi_log_sysevent(pdip, class, ESC_DDI_INITIATOR_REGISTER);
660 	return (MDI_SUCCESS);
661 }
662 
663 /*
664  * mdi_phci_unregister():
665  *		Unregister a pHCI module from mpxio framework
666  *		mdi_phci_unregister() is called by the pHCI drivers from their
667  *		detach(9E) handler to unregister their instances from the
668  *		framework.
669  * Return Values:
670  *		MDI_SUCCESS
671  *		MDI_FAILURE
672  */
673 /*ARGSUSED*/
674 int
675 mdi_phci_unregister(dev_info_t *pdip, int flags)
676 {
677 	mdi_vhci_t		*vh;
678 	mdi_phci_t		*ph;
679 	mdi_phci_t		*tmp;
680 	mdi_phci_t		*prev = NULL;
681 
682 	ASSERT(DEVI_BUSY_CHANGING(ddi_get_parent(pdip)));
683 
684 	ph = i_devi_get_phci(pdip);
685 	if (ph == NULL) {
686 		MDI_DEBUG(1, (CE_WARN, pdip,
687 		    "!pHCI unregister: Not a valid pHCI"));
688 		return (MDI_FAILURE);
689 	}
690 
691 	vh = ph->ph_vhci;
692 	ASSERT(vh != NULL);
693 	if (vh == NULL) {
694 		MDI_DEBUG(1, (CE_WARN, pdip,
695 		    "!pHCI unregister: Not a valid vHCI"));
696 		return (MDI_FAILURE);
697 	}
698 
699 	MDI_VHCI_PHCI_LOCK(vh);
700 	tmp = vh->vh_phci_head;
701 	while (tmp) {
702 		if (tmp == ph) {
703 			break;
704 		}
705 		prev = tmp;
706 		tmp = tmp->ph_next;
707 	}
708 
709 	if (ph == vh->vh_phci_head) {
710 		vh->vh_phci_head = ph->ph_next;
711 	} else {
712 		prev->ph_next = ph->ph_next;
713 	}
714 
715 	if (ph == vh->vh_phci_tail) {
716 		vh->vh_phci_tail = prev;
717 	}
718 
719 	vh->vh_phci_count--;
720 	MDI_VHCI_PHCI_UNLOCK(vh);
721 
722 	i_mdi_log_sysevent(pdip, ph->ph_vhci->vh_class,
723 	    ESC_DDI_INITIATOR_UNREGISTER);
724 	vhcache_phci_remove(vh->vh_config, ph);
725 	cv_destroy(&ph->ph_unstable_cv);
726 	mutex_destroy(&ph->ph_mutex);
727 	kmem_free(ph, sizeof (mdi_phci_t));
728 	DEVI(pdip)->devi_mdi_component &= ~MDI_COMPONENT_PHCI;
729 	DEVI(pdip)->devi_mdi_xhci = NULL;
730 	return (MDI_SUCCESS);
731 }
732 
733 /*
734  * i_devi_get_phci():
735  * 		Utility function to return the phci extensions.
736  */
737 static mdi_phci_t *
738 i_devi_get_phci(dev_info_t *pdip)
739 {
740 	mdi_phci_t	*ph = NULL;
741 	if (MDI_PHCI(pdip)) {
742 		ph = (mdi_phci_t *)DEVI(pdip)->devi_mdi_xhci;
743 	}
744 	return (ph);
745 }
746 
747 /*
748  * Single thread mdi entry into devinfo node for modifying its children.
749  * If necessary we perform an ndi_devi_enter of the vHCI before doing
750  * an ndi_devi_enter of 'dip'.  We maintain circular in two parts: one
751  * for the vHCI and one for the pHCI.
752  */
753 void
754 mdi_devi_enter(dev_info_t *phci_dip, int *circular)
755 {
756 	dev_info_t	*vdip;
757 	int		vcircular, pcircular;
758 
759 	/* Verify calling context */
760 	ASSERT(MDI_PHCI(phci_dip));
761 	vdip = mdi_devi_get_vdip(phci_dip);
762 	ASSERT(vdip);			/* A pHCI always has a vHCI */
763 
764 	/*
765 	 * If pHCI is detaching then the framework has already entered the
766 	 * vHCI on a threads that went down the code path leading to
767 	 * detach_node().  This framework enter of the vHCI during pHCI
768 	 * detach is done to avoid deadlock with vHCI power management
769 	 * operations which enter the vHCI and the enter down the path
770 	 * to the pHCI. If pHCI is detaching then we piggyback this calls
771 	 * enter of the vHCI on frameworks vHCI enter that has already
772 	 * occurred - this is OK because we know that the framework thread
773 	 * doing detach is waiting for our completion.
774 	 *
775 	 * We should DEVI_IS_DETACHING under an enter of the parent to avoid
776 	 * race with detach - but we can't do that because the framework has
777 	 * already entered the parent, so we have some complexity instead.
778 	 */
779 	for (;;) {
780 		if (ndi_devi_tryenter(vdip, &vcircular)) {
781 			ASSERT(vcircular != -1);
782 			if (DEVI_IS_DETACHING(phci_dip)) {
783 				ndi_devi_exit(vdip, vcircular);
784 				vcircular = -1;
785 			}
786 			break;
787 		} else if (DEVI_IS_DETACHING(phci_dip)) {
788 			vcircular = -1;
789 			break;
790 		} else {
791 			delay(1);
792 		}
793 	}
794 
795 	ndi_devi_enter(phci_dip, &pcircular);
796 	*circular = (vcircular << 16) | (pcircular & 0xFFFF);
797 }
798 
799 /*
800  * Release mdi_devi_enter or successful mdi_devi_tryenter.
801  */
802 void
803 mdi_devi_exit(dev_info_t *phci_dip, int circular)
804 {
805 	dev_info_t	*vdip;
806 	int		vcircular, pcircular;
807 
808 	/* Verify calling context */
809 	ASSERT(MDI_PHCI(phci_dip));
810 	vdip = mdi_devi_get_vdip(phci_dip);
811 	ASSERT(vdip);			/* A pHCI always has a vHCI */
812 
813 	/* extract two circular recursion values from single int */
814 	pcircular = (short)(circular & 0xFFFF);
815 	vcircular = (short)((circular >> 16) & 0xFFFF);
816 
817 	ndi_devi_exit(phci_dip, pcircular);
818 	if (vcircular != -1)
819 		ndi_devi_exit(vdip, vcircular);
820 }
821 
822 /*
823  * The functions mdi_devi_exit_phci() and mdi_devi_enter_phci() are used
824  * around a pHCI drivers calls to mdi_pi_online/offline, after holding
825  * the pathinfo node via mdi_hold_path/mdi_rele_path, to avoid deadlock
826  * with vHCI power management code during path online/offline.  Each
827  * mdi_devi_exit_phci must have a matching mdi_devi_enter_phci, and both must
828  * occur within the scope of an active mdi_devi_enter that establishes the
829  * circular value.
830  */
831 void
832 mdi_devi_exit_phci(dev_info_t *phci_dip, int circular)
833 {
834 	int		pcircular;
835 
836 	/* Verify calling context */
837 	ASSERT(MDI_PHCI(phci_dip));
838 
839 	pcircular = (short)(circular & 0xFFFF);
840 	ndi_devi_exit(phci_dip, pcircular);
841 }
842 
843 void
844 mdi_devi_enter_phci(dev_info_t *phci_dip, int *circular)
845 {
846 	int		pcircular;
847 
848 	/* Verify calling context */
849 	ASSERT(MDI_PHCI(phci_dip));
850 
851 	ndi_devi_enter(phci_dip, &pcircular);
852 
853 	/* verify matching mdi_devi_exit_phci/mdi_devi_enter_phci use */
854 	ASSERT(pcircular == ((short)(*circular & 0xFFFF)));
855 }
856 
857 /*
858  * mdi_devi_get_vdip():
859  *		given a pHCI dip return vHCI dip
860  */
861 dev_info_t *
862 mdi_devi_get_vdip(dev_info_t *pdip)
863 {
864 	mdi_phci_t	*ph;
865 
866 	ph = i_devi_get_phci(pdip);
867 	if (ph && ph->ph_vhci)
868 		return (ph->ph_vhci->vh_dip);
869 	return (NULL);
870 }
871 
872 /*
873  * mdi_devi_pdip_entered():
874  *		Return 1 if we are vHCI and have done an ndi_devi_enter
875  *		of a pHCI
876  */
877 int
878 mdi_devi_pdip_entered(dev_info_t *vdip)
879 {
880 	mdi_vhci_t	*vh;
881 	mdi_phci_t	*ph;
882 
883 	vh = i_devi_get_vhci(vdip);
884 	if (vh == NULL)
885 		return (0);
886 
887 	MDI_VHCI_PHCI_LOCK(vh);
888 	ph = vh->vh_phci_head;
889 	while (ph) {
890 		if (ph->ph_dip && DEVI_BUSY_OWNED(ph->ph_dip)) {
891 			MDI_VHCI_PHCI_UNLOCK(vh);
892 			return (1);
893 		}
894 		ph = ph->ph_next;
895 	}
896 	MDI_VHCI_PHCI_UNLOCK(vh);
897 	return (0);
898 }
899 
900 /*
901  * mdi_phci_path2devinfo():
902  * 		Utility function to search for a valid phci device given
903  *		the devfs pathname.
904  */
905 dev_info_t *
906 mdi_phci_path2devinfo(dev_info_t *vdip, caddr_t pathname)
907 {
908 	char		*temp_pathname;
909 	mdi_vhci_t	*vh;
910 	mdi_phci_t	*ph;
911 	dev_info_t 	*pdip = NULL;
912 
913 	vh = i_devi_get_vhci(vdip);
914 	ASSERT(vh != NULL);
915 
916 	if (vh == NULL) {
917 		/*
918 		 * Invalid vHCI component, return failure
919 		 */
920 		return (NULL);
921 	}
922 
923 	temp_pathname = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
924 	MDI_VHCI_PHCI_LOCK(vh);
925 	ph = vh->vh_phci_head;
926 	while (ph != NULL) {
927 		pdip = ph->ph_dip;
928 		ASSERT(pdip != NULL);
929 		*temp_pathname = '\0';
930 		(void) ddi_pathname(pdip, temp_pathname);
931 		if (strcmp(temp_pathname, pathname) == 0) {
932 			break;
933 		}
934 		ph = ph->ph_next;
935 	}
936 	if (ph == NULL) {
937 		pdip = NULL;
938 	}
939 	MDI_VHCI_PHCI_UNLOCK(vh);
940 	kmem_free(temp_pathname, MAXPATHLEN);
941 	return (pdip);
942 }
943 
944 /*
945  * mdi_phci_get_path_count():
946  * 		get number of path information nodes associated with a given
947  *		pHCI device.
948  */
949 int
950 mdi_phci_get_path_count(dev_info_t *pdip)
951 {
952 	mdi_phci_t	*ph;
953 	int		count = 0;
954 
955 	ph = i_devi_get_phci(pdip);
956 	if (ph != NULL) {
957 		count = ph->ph_path_count;
958 	}
959 	return (count);
960 }
961 
962 /*
963  * i_mdi_phci_lock():
964  *		Lock a pHCI device
965  * Return Values:
966  *		None
967  * Note:
968  *		The default locking order is:
969  *		_NOTE(LOCK_ORDER(mdi_phci::ph_mutex mdi_pathinfo::pi_mutex))
970  *		But there are number of situations where locks need to be
971  *		grabbed in reverse order.  This routine implements try and lock
972  *		mechanism depending on the requested parameter option.
973  */
974 static void
975 i_mdi_phci_lock(mdi_phci_t *ph, mdi_pathinfo_t *pip)
976 {
977 	if (pip) {
978 		/* Reverse locking is requested. */
979 		while (MDI_PHCI_TRYLOCK(ph) == 0) {
980 			/*
981 			 * tryenter failed. Try to grab again
982 			 * after a small delay
983 			 */
984 			MDI_PI_HOLD(pip);
985 			MDI_PI_UNLOCK(pip);
986 			delay(1);
987 			MDI_PI_LOCK(pip);
988 			MDI_PI_RELE(pip);
989 		}
990 	} else {
991 		MDI_PHCI_LOCK(ph);
992 	}
993 }
994 
995 /*
996  * i_mdi_phci_unlock():
997  *		Unlock the pHCI component
998  */
999 static void
1000 i_mdi_phci_unlock(mdi_phci_t *ph)
1001 {
1002 	MDI_PHCI_UNLOCK(ph);
1003 }
1004 
1005 /*
1006  * i_mdi_devinfo_create():
1007  *		create client device's devinfo node
1008  * Return Values:
1009  *		dev_info
1010  *		NULL
1011  * Notes:
1012  */
1013 static dev_info_t *
1014 i_mdi_devinfo_create(mdi_vhci_t *vh, char *name, char *guid,
1015 	char **compatible, int ncompatible)
1016 {
1017 	dev_info_t *cdip = NULL;
1018 
1019 	ASSERT(MDI_VHCI_CLIENT_LOCKED(vh));
1020 
1021 	/* Verify for duplicate entry */
1022 	cdip = i_mdi_devinfo_find(vh, name, guid);
1023 	ASSERT(cdip == NULL);
1024 	if (cdip) {
1025 		cmn_err(CE_WARN,
1026 		    "i_mdi_devinfo_create: client dip %p already exists",
1027 			(void *)cdip);
1028 	}
1029 
1030 	ndi_devi_alloc_sleep(vh->vh_dip, name, DEVI_SID_NODEID, &cdip);
1031 	if (cdip == NULL)
1032 		goto fail;
1033 
1034 	/*
1035 	 * Create component type and Global unique identifier
1036 	 * properties
1037 	 */
1038 	if (ndi_prop_update_string(DDI_DEV_T_NONE, cdip,
1039 	    MDI_CLIENT_GUID_PROP, guid) != DDI_PROP_SUCCESS) {
1040 		goto fail;
1041 	}
1042 
1043 	/* Decorate the node with compatible property */
1044 	if (compatible &&
1045 	    (ndi_prop_update_string_array(DDI_DEV_T_NONE, cdip,
1046 	    "compatible", compatible, ncompatible) != DDI_PROP_SUCCESS)) {
1047 		goto fail;
1048 	}
1049 
1050 	return (cdip);
1051 
1052 fail:
1053 	if (cdip) {
1054 		(void) ndi_prop_remove_all(cdip);
1055 		(void) ndi_devi_free(cdip);
1056 	}
1057 	return (NULL);
1058 }
1059 
1060 /*
1061  * i_mdi_devinfo_find():
1062  *		Find a matching devinfo node for given client node name
1063  *		and its guid.
1064  * Return Values:
1065  *		Handle to a dev_info node or NULL
1066  */
1067 static dev_info_t *
1068 i_mdi_devinfo_find(mdi_vhci_t *vh, caddr_t name, char *guid)
1069 {
1070 	char			*data;
1071 	dev_info_t 		*cdip = NULL;
1072 	dev_info_t 		*ndip = NULL;
1073 	int			circular;
1074 
1075 	ndi_devi_enter(vh->vh_dip, &circular);
1076 	ndip = (dev_info_t *)DEVI(vh->vh_dip)->devi_child;
1077 	while ((cdip = ndip) != NULL) {
1078 		ndip = (dev_info_t *)DEVI(cdip)->devi_sibling;
1079 
1080 		if (strcmp(DEVI(cdip)->devi_node_name, name)) {
1081 			continue;
1082 		}
1083 
1084 		if (ddi_prop_lookup_string(DDI_DEV_T_ANY, cdip,
1085 		    DDI_PROP_DONTPASS, MDI_CLIENT_GUID_PROP,
1086 		    &data) != DDI_PROP_SUCCESS) {
1087 			continue;
1088 		}
1089 
1090 		if (strcmp(data, guid) != 0) {
1091 			ddi_prop_free(data);
1092 			continue;
1093 		}
1094 		ddi_prop_free(data);
1095 		break;
1096 	}
1097 	ndi_devi_exit(vh->vh_dip, circular);
1098 	return (cdip);
1099 }
1100 
1101 /*
1102  * i_mdi_devinfo_remove():
1103  *		Remove a client device node
1104  */
1105 static int
1106 i_mdi_devinfo_remove(dev_info_t *vdip, dev_info_t *cdip, int flags)
1107 {
1108 	int	rv = MDI_SUCCESS;
1109 
1110 	if (i_mdi_is_child_present(vdip, cdip) == MDI_SUCCESS ||
1111 	    (flags & MDI_CLIENT_FLAGS_DEV_NOT_SUPPORTED)) {
1112 		rv = ndi_devi_offline(cdip, NDI_DEVI_REMOVE);
1113 		if (rv != NDI_SUCCESS) {
1114 			MDI_DEBUG(1, (CE_NOTE, NULL, "!i_mdi_devinfo_remove:"
1115 			    " failed. cdip = %p\n", (void *)cdip));
1116 		}
1117 		/*
1118 		 * Convert to MDI error code
1119 		 */
1120 		switch (rv) {
1121 		case NDI_SUCCESS:
1122 			rv = MDI_SUCCESS;
1123 			break;
1124 		case NDI_BUSY:
1125 			rv = MDI_BUSY;
1126 			break;
1127 		default:
1128 			rv = MDI_FAILURE;
1129 			break;
1130 		}
1131 	}
1132 	return (rv);
1133 }
1134 
1135 /*
1136  * i_devi_get_client()
1137  *		Utility function to get mpxio component extensions
1138  */
1139 static mdi_client_t *
1140 i_devi_get_client(dev_info_t *cdip)
1141 {
1142 	mdi_client_t	*ct = NULL;
1143 
1144 	if (MDI_CLIENT(cdip)) {
1145 		ct = (mdi_client_t *)DEVI(cdip)->devi_mdi_client;
1146 	}
1147 	return (ct);
1148 }
1149 
1150 /*
1151  * i_mdi_is_child_present():
1152  *		Search for the presence of client device dev_info node
1153  */
1154 static int
1155 i_mdi_is_child_present(dev_info_t *vdip, dev_info_t *cdip)
1156 {
1157 	int		rv = MDI_FAILURE;
1158 	struct dev_info	*dip;
1159 	int		circular;
1160 
1161 	ndi_devi_enter(vdip, &circular);
1162 	dip = DEVI(vdip)->devi_child;
1163 	while (dip) {
1164 		if (dip == DEVI(cdip)) {
1165 			rv = MDI_SUCCESS;
1166 			break;
1167 		}
1168 		dip = dip->devi_sibling;
1169 	}
1170 	ndi_devi_exit(vdip, circular);
1171 	return (rv);
1172 }
1173 
1174 
1175 /*
1176  * i_mdi_client_lock():
1177  *		Grab client component lock
1178  * Return Values:
1179  *		None
1180  * Note:
1181  *		The default locking order is:
1182  *		_NOTE(LOCK_ORDER(mdi_client::ct_mutex mdi_pathinfo::pi_mutex))
1183  *		But there are number of situations where locks need to be
1184  *		grabbed in reverse order.  This routine implements try and lock
1185  *		mechanism depending on the requested parameter option.
1186  */
1187 static void
1188 i_mdi_client_lock(mdi_client_t *ct, mdi_pathinfo_t *pip)
1189 {
1190 	if (pip) {
1191 		/*
1192 		 * Reverse locking is requested.
1193 		 */
1194 		while (MDI_CLIENT_TRYLOCK(ct) == 0) {
1195 			/*
1196 			 * tryenter failed. Try to grab again
1197 			 * after a small delay
1198 			 */
1199 			MDI_PI_HOLD(pip);
1200 			MDI_PI_UNLOCK(pip);
1201 			delay(1);
1202 			MDI_PI_LOCK(pip);
1203 			MDI_PI_RELE(pip);
1204 		}
1205 	} else {
1206 		MDI_CLIENT_LOCK(ct);
1207 	}
1208 }
1209 
1210 /*
1211  * i_mdi_client_unlock():
1212  *		Unlock a client component
1213  */
1214 static void
1215 i_mdi_client_unlock(mdi_client_t *ct)
1216 {
1217 	MDI_CLIENT_UNLOCK(ct);
1218 }
1219 
1220 /*
1221  * i_mdi_client_alloc():
1222  * 		Allocate and initialize a client structure.  Caller should
1223  *		hold the vhci client lock.
1224  * Return Values:
1225  *		Handle to a client component
1226  */
1227 /*ARGSUSED*/
1228 static mdi_client_t *
1229 i_mdi_client_alloc(mdi_vhci_t *vh, char *name, char *lguid)
1230 {
1231 	mdi_client_t	*ct;
1232 
1233 	ASSERT(MDI_VHCI_CLIENT_LOCKED(vh));
1234 
1235 	/*
1236 	 * Allocate and initialize a component structure.
1237 	 */
1238 	ct = kmem_zalloc(sizeof (*ct), KM_SLEEP);
1239 	mutex_init(&ct->ct_mutex, NULL, MUTEX_DEFAULT, NULL);
1240 	ct->ct_hnext = NULL;
1241 	ct->ct_hprev = NULL;
1242 	ct->ct_dip = NULL;
1243 	ct->ct_vhci = vh;
1244 	ct->ct_drvname = kmem_alloc(strlen(name) + 1, KM_SLEEP);
1245 	(void) strcpy(ct->ct_drvname, name);
1246 	ct->ct_guid = kmem_alloc(strlen(lguid) + 1, KM_SLEEP);
1247 	(void) strcpy(ct->ct_guid, lguid);
1248 	ct->ct_cprivate = NULL;
1249 	ct->ct_vprivate = NULL;
1250 	ct->ct_flags = 0;
1251 	ct->ct_state = MDI_CLIENT_STATE_FAILED;
1252 	MDI_CLIENT_LOCK(ct);
1253 	MDI_CLIENT_SET_OFFLINE(ct);
1254 	MDI_CLIENT_SET_DETACH(ct);
1255 	MDI_CLIENT_SET_POWER_UP(ct);
1256 	MDI_CLIENT_UNLOCK(ct);
1257 	ct->ct_failover_flags = 0;
1258 	ct->ct_failover_status = 0;
1259 	cv_init(&ct->ct_failover_cv, NULL, CV_DRIVER, NULL);
1260 	ct->ct_unstable = 0;
1261 	cv_init(&ct->ct_unstable_cv, NULL, CV_DRIVER, NULL);
1262 	cv_init(&ct->ct_powerchange_cv, NULL, CV_DRIVER, NULL);
1263 	ct->ct_lb = vh->vh_lb;
1264 	ct->ct_lb_args =  kmem_zalloc(sizeof (client_lb_args_t), KM_SLEEP);
1265 	ct->ct_lb_args->region_size = LOAD_BALANCE_DEFAULT_REGION_SIZE;
1266 	ct->ct_path_count = 0;
1267 	ct->ct_path_head = NULL;
1268 	ct->ct_path_tail = NULL;
1269 	ct->ct_path_last = NULL;
1270 
1271 	/*
1272 	 * Add this client component to our client hash queue
1273 	 */
1274 	i_mdi_client_enlist_table(vh, ct);
1275 	return (ct);
1276 }
1277 
1278 /*
1279  * i_mdi_client_enlist_table():
1280  *		Attach the client device to the client hash table. Caller
1281  *		should hold the vhci client lock.
1282  */
1283 static void
1284 i_mdi_client_enlist_table(mdi_vhci_t *vh, mdi_client_t *ct)
1285 {
1286 	int 			index;
1287 	struct client_hash	*head;
1288 
1289 	ASSERT(MDI_VHCI_CLIENT_LOCKED(vh));
1290 
1291 	index = i_mdi_get_hash_key(ct->ct_guid);
1292 	head = &vh->vh_client_table[index];
1293 	ct->ct_hnext = (mdi_client_t *)head->ct_hash_head;
1294 	head->ct_hash_head = ct;
1295 	head->ct_hash_count++;
1296 	vh->vh_client_count++;
1297 }
1298 
1299 /*
1300  * i_mdi_client_delist_table():
1301  *		Attach the client device to the client hash table.
1302  *		Caller should hold the vhci client lock.
1303  */
1304 static void
1305 i_mdi_client_delist_table(mdi_vhci_t *vh, mdi_client_t *ct)
1306 {
1307 	int			index;
1308 	char			*guid;
1309 	struct client_hash 	*head;
1310 	mdi_client_t		*next;
1311 	mdi_client_t		*last;
1312 
1313 	ASSERT(MDI_VHCI_CLIENT_LOCKED(vh));
1314 
1315 	guid = ct->ct_guid;
1316 	index = i_mdi_get_hash_key(guid);
1317 	head = &vh->vh_client_table[index];
1318 
1319 	last = NULL;
1320 	next = (mdi_client_t *)head->ct_hash_head;
1321 	while (next != NULL) {
1322 		if (next == ct) {
1323 			break;
1324 		}
1325 		last = next;
1326 		next = next->ct_hnext;
1327 	}
1328 
1329 	if (next) {
1330 		head->ct_hash_count--;
1331 		if (last == NULL) {
1332 			head->ct_hash_head = ct->ct_hnext;
1333 		} else {
1334 			last->ct_hnext = ct->ct_hnext;
1335 		}
1336 		ct->ct_hnext = NULL;
1337 		vh->vh_client_count--;
1338 	}
1339 }
1340 
1341 
1342 /*
1343  * i_mdi_client_free():
1344  *		Free a client component
1345  */
1346 static int
1347 i_mdi_client_free(mdi_vhci_t *vh, mdi_client_t *ct)
1348 {
1349 	int		rv = MDI_SUCCESS;
1350 	int		flags = ct->ct_flags;
1351 	dev_info_t	*cdip;
1352 	dev_info_t	*vdip;
1353 
1354 	ASSERT(MDI_VHCI_CLIENT_LOCKED(vh));
1355 
1356 	vdip = vh->vh_dip;
1357 	cdip = ct->ct_dip;
1358 
1359 	(void) ndi_prop_remove(DDI_DEV_T_NONE, cdip, MDI_CLIENT_GUID_PROP);
1360 	DEVI(cdip)->devi_mdi_component &= ~MDI_COMPONENT_CLIENT;
1361 	DEVI(cdip)->devi_mdi_client = NULL;
1362 
1363 	/*
1364 	 * Clear out back ref. to dev_info_t node
1365 	 */
1366 	ct->ct_dip = NULL;
1367 
1368 	/*
1369 	 * Remove this client from our hash queue
1370 	 */
1371 	i_mdi_client_delist_table(vh, ct);
1372 
1373 	/*
1374 	 * Uninitialize and free the component
1375 	 */
1376 	kmem_free(ct->ct_drvname, strlen(ct->ct_drvname) + 1);
1377 	kmem_free(ct->ct_guid, strlen(ct->ct_guid) + 1);
1378 	kmem_free(ct->ct_lb_args, sizeof (client_lb_args_t));
1379 	cv_destroy(&ct->ct_failover_cv);
1380 	cv_destroy(&ct->ct_unstable_cv);
1381 	cv_destroy(&ct->ct_powerchange_cv);
1382 	mutex_destroy(&ct->ct_mutex);
1383 	kmem_free(ct, sizeof (*ct));
1384 
1385 	if (cdip != NULL) {
1386 		MDI_VHCI_CLIENT_UNLOCK(vh);
1387 		(void) i_mdi_devinfo_remove(vdip, cdip, flags);
1388 		MDI_VHCI_CLIENT_LOCK(vh);
1389 	}
1390 	return (rv);
1391 }
1392 
1393 /*
1394  * i_mdi_client_find():
1395  * 		Find the client structure corresponding to a given guid
1396  *		Caller should hold the vhci client lock.
1397  */
1398 static mdi_client_t *
1399 i_mdi_client_find(mdi_vhci_t *vh, char *cname, char *guid)
1400 {
1401 	int			index;
1402 	struct client_hash	*head;
1403 	mdi_client_t		*ct;
1404 
1405 	ASSERT(MDI_VHCI_CLIENT_LOCKED(vh));
1406 
1407 	index = i_mdi_get_hash_key(guid);
1408 	head = &vh->vh_client_table[index];
1409 
1410 	ct = head->ct_hash_head;
1411 	while (ct != NULL) {
1412 		if (strcmp(ct->ct_guid, guid) == 0 &&
1413 		    (cname == NULL || strcmp(ct->ct_drvname, cname) == 0)) {
1414 			break;
1415 		}
1416 		ct = ct->ct_hnext;
1417 	}
1418 	return (ct);
1419 }
1420 
1421 /*
1422  * i_mdi_client_update_state():
1423  *		Compute and update client device state
1424  * Notes:
1425  *		A client device can be in any of three possible states:
1426  *
1427  *		MDI_CLIENT_STATE_OPTIMAL - Client in optimal state with more
1428  *		one online/standby paths. Can tolerate failures.
1429  *		MDI_CLIENT_STATE_DEGRADED - Client device in degraded state with
1430  *		no alternate paths available as standby. A failure on the online
1431  *		would result in loss of access to device data.
1432  *		MDI_CLIENT_STATE_FAILED - Client device in failed state with
1433  *		no paths available to access the device.
1434  */
1435 static void
1436 i_mdi_client_update_state(mdi_client_t *ct)
1437 {
1438 	int state;
1439 
1440 	ASSERT(MDI_CLIENT_LOCKED(ct));
1441 	state = i_mdi_client_compute_state(ct, NULL);
1442 	MDI_CLIENT_SET_STATE(ct, state);
1443 }
1444 
1445 /*
1446  * i_mdi_client_compute_state():
1447  *		Compute client device state
1448  *
1449  *		mdi_phci_t *	Pointer to pHCI structure which should
1450  *				while computing the new value.  Used by
1451  *				i_mdi_phci_offline() to find the new
1452  *				client state after DR of a pHCI.
1453  */
1454 static int
1455 i_mdi_client_compute_state(mdi_client_t *ct, mdi_phci_t *ph)
1456 {
1457 	int		state;
1458 	int		online_count = 0;
1459 	int		standby_count = 0;
1460 	mdi_pathinfo_t	*pip, *next;
1461 
1462 	ASSERT(MDI_CLIENT_LOCKED(ct));
1463 	pip = ct->ct_path_head;
1464 	while (pip != NULL) {
1465 		MDI_PI_LOCK(pip);
1466 		next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
1467 		if (MDI_PI(pip)->pi_phci == ph) {
1468 			MDI_PI_UNLOCK(pip);
1469 			pip = next;
1470 			continue;
1471 		}
1472 
1473 		if ((MDI_PI(pip)->pi_state & MDI_PATHINFO_STATE_MASK)
1474 				== MDI_PATHINFO_STATE_ONLINE)
1475 			online_count++;
1476 		else if ((MDI_PI(pip)->pi_state & MDI_PATHINFO_STATE_MASK)
1477 				== MDI_PATHINFO_STATE_STANDBY)
1478 			standby_count++;
1479 		MDI_PI_UNLOCK(pip);
1480 		pip = next;
1481 	}
1482 
1483 	if (online_count == 0) {
1484 		if (standby_count == 0) {
1485 			state = MDI_CLIENT_STATE_FAILED;
1486 			MDI_DEBUG(2, (CE_NOTE, NULL, "!client state: failed"
1487 			    " ct = %p\n", (void *)ct));
1488 		} else if (standby_count == 1) {
1489 			state = MDI_CLIENT_STATE_DEGRADED;
1490 		} else {
1491 			state = MDI_CLIENT_STATE_OPTIMAL;
1492 		}
1493 	} else if (online_count == 1) {
1494 		if (standby_count == 0) {
1495 			state = MDI_CLIENT_STATE_DEGRADED;
1496 		} else {
1497 			state = MDI_CLIENT_STATE_OPTIMAL;
1498 		}
1499 	} else {
1500 		state = MDI_CLIENT_STATE_OPTIMAL;
1501 	}
1502 	return (state);
1503 }
1504 
1505 /*
1506  * i_mdi_client2devinfo():
1507  *		Utility function
1508  */
1509 dev_info_t *
1510 i_mdi_client2devinfo(mdi_client_t *ct)
1511 {
1512 	return (ct->ct_dip);
1513 }
1514 
1515 /*
1516  * mdi_client_path2_devinfo():
1517  * 		Given the parent devinfo and child devfs pathname, search for
1518  *		a valid devfs node handle.
1519  */
1520 dev_info_t *
1521 mdi_client_path2devinfo(dev_info_t *vdip, char *pathname)
1522 {
1523 	dev_info_t 	*cdip = NULL;
1524 	dev_info_t 	*ndip = NULL;
1525 	char		*temp_pathname;
1526 	int		circular;
1527 
1528 	/*
1529 	 * Allocate temp buffer
1530 	 */
1531 	temp_pathname = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1532 
1533 	/*
1534 	 * Lock parent against changes
1535 	 */
1536 	ndi_devi_enter(vdip, &circular);
1537 	ndip = (dev_info_t *)DEVI(vdip)->devi_child;
1538 	while ((cdip = ndip) != NULL) {
1539 		ndip = (dev_info_t *)DEVI(cdip)->devi_sibling;
1540 
1541 		*temp_pathname = '\0';
1542 		(void) ddi_pathname(cdip, temp_pathname);
1543 		if (strcmp(temp_pathname, pathname) == 0) {
1544 			break;
1545 		}
1546 	}
1547 	/*
1548 	 * Release devinfo lock
1549 	 */
1550 	ndi_devi_exit(vdip, circular);
1551 
1552 	/*
1553 	 * Free the temp buffer
1554 	 */
1555 	kmem_free(temp_pathname, MAXPATHLEN);
1556 	return (cdip);
1557 }
1558 
1559 /*
1560  * mdi_client_get_path_count():
1561  * 		Utility function to get number of path information nodes
1562  *		associated with a given client device.
1563  */
1564 int
1565 mdi_client_get_path_count(dev_info_t *cdip)
1566 {
1567 	mdi_client_t	*ct;
1568 	int		count = 0;
1569 
1570 	ct = i_devi_get_client(cdip);
1571 	if (ct != NULL) {
1572 		count = ct->ct_path_count;
1573 	}
1574 	return (count);
1575 }
1576 
1577 
1578 /*
1579  * i_mdi_get_hash_key():
1580  * 		Create a hash using strings as keys
1581  *
1582  */
1583 static int
1584 i_mdi_get_hash_key(char *str)
1585 {
1586 	uint32_t	g, hash = 0;
1587 	char		*p;
1588 
1589 	for (p = str; *p != '\0'; p++) {
1590 		g = *p;
1591 		hash += g;
1592 	}
1593 	return (hash % (CLIENT_HASH_TABLE_SIZE - 1));
1594 }
1595 
1596 /*
1597  * mdi_get_lb_policy():
1598  * 		Get current load balancing policy for a given client device
1599  */
1600 client_lb_t
1601 mdi_get_lb_policy(dev_info_t *cdip)
1602 {
1603 	client_lb_t	lb = LOAD_BALANCE_NONE;
1604 	mdi_client_t	*ct;
1605 
1606 	ct = i_devi_get_client(cdip);
1607 	if (ct != NULL) {
1608 		lb = ct->ct_lb;
1609 	}
1610 	return (lb);
1611 }
1612 
1613 /*
1614  * mdi_set_lb_region_size():
1615  * 		Set current region size for the load-balance
1616  */
1617 int
1618 mdi_set_lb_region_size(dev_info_t *cdip, int region_size)
1619 {
1620 	mdi_client_t	*ct;
1621 	int		rv = MDI_FAILURE;
1622 
1623 	ct = i_devi_get_client(cdip);
1624 	if (ct != NULL && ct->ct_lb_args != NULL) {
1625 		ct->ct_lb_args->region_size = region_size;
1626 		rv = MDI_SUCCESS;
1627 	}
1628 	return (rv);
1629 }
1630 
1631 /*
1632  * mdi_Set_lb_policy():
1633  * 		Set current load balancing policy for a given client device
1634  */
1635 int
1636 mdi_set_lb_policy(dev_info_t *cdip, client_lb_t lb)
1637 {
1638 	mdi_client_t	*ct;
1639 	int		rv = MDI_FAILURE;
1640 
1641 	ct = i_devi_get_client(cdip);
1642 	if (ct != NULL) {
1643 		ct->ct_lb = lb;
1644 		rv = MDI_SUCCESS;
1645 	}
1646 	return (rv);
1647 }
1648 
1649 /*
1650  * mdi_failover():
1651  *		failover function called by the vHCI drivers to initiate
1652  *		a failover operation.  This is typically due to non-availability
1653  *		of online paths to route I/O requests.  Failover can be
1654  *		triggered through user application also.
1655  *
1656  *		The vHCI driver calls mdi_failover() to initiate a failover
1657  *		operation. mdi_failover() calls back into the vHCI driver's
1658  *		vo_failover() entry point to perform the actual failover
1659  *		operation.  The reason for requiring the vHCI driver to
1660  *		initiate failover by calling mdi_failover(), instead of directly
1661  *		executing vo_failover() itself, is to ensure that the mdi
1662  *		framework can keep track of the client state properly.
1663  *		Additionally, mdi_failover() provides as a convenience the
1664  *		option of performing the failover operation synchronously or
1665  *		asynchronously
1666  *
1667  *		Upon successful completion of the failover operation, the
1668  *		paths that were previously ONLINE will be in the STANDBY state,
1669  *		and the newly activated paths will be in the ONLINE state.
1670  *
1671  *		The flags modifier determines whether the activation is done
1672  *		synchronously: MDI_FAILOVER_SYNC
1673  * Return Values:
1674  *		MDI_SUCCESS
1675  *		MDI_FAILURE
1676  *		MDI_BUSY
1677  */
1678 /*ARGSUSED*/
1679 int
1680 mdi_failover(dev_info_t *vdip, dev_info_t *cdip, int flags)
1681 {
1682 	int			rv;
1683 	mdi_client_t		*ct;
1684 
1685 	ct = i_devi_get_client(cdip);
1686 	ASSERT(ct != NULL);
1687 	if (ct == NULL) {
1688 		/* cdip is not a valid client device. Nothing more to do. */
1689 		return (MDI_FAILURE);
1690 	}
1691 
1692 	MDI_CLIENT_LOCK(ct);
1693 
1694 	if (MDI_CLIENT_IS_PATH_FREE_IN_PROGRESS(ct)) {
1695 		/* A path to the client is being freed */
1696 		MDI_CLIENT_UNLOCK(ct);
1697 		return (MDI_BUSY);
1698 	}
1699 
1700 
1701 	if (MDI_CLIENT_IS_FAILED(ct)) {
1702 		/*
1703 		 * Client is in failed state. Nothing more to do.
1704 		 */
1705 		MDI_CLIENT_UNLOCK(ct);
1706 		return (MDI_FAILURE);
1707 	}
1708 
1709 	if (MDI_CLIENT_IS_FAILOVER_IN_PROGRESS(ct)) {
1710 		/*
1711 		 * Failover is already in progress; return BUSY
1712 		 */
1713 		MDI_CLIENT_UNLOCK(ct);
1714 		return (MDI_BUSY);
1715 	}
1716 	/*
1717 	 * Make sure that mdi_pathinfo node state changes are processed.
1718 	 * We do not allow failovers to progress while client path state
1719 	 * changes are in progress
1720 	 */
1721 	if (ct->ct_unstable) {
1722 		if (flags == MDI_FAILOVER_ASYNC) {
1723 			MDI_CLIENT_UNLOCK(ct);
1724 			return (MDI_BUSY);
1725 		} else {
1726 			while (ct->ct_unstable)
1727 				cv_wait(&ct->ct_unstable_cv, &ct->ct_mutex);
1728 		}
1729 	}
1730 
1731 	/*
1732 	 * Client device is in stable state. Before proceeding, perform sanity
1733 	 * checks again.
1734 	 */
1735 	if ((MDI_CLIENT_IS_DETACHED(ct)) || (MDI_CLIENT_IS_FAILED(ct)) ||
1736 	    (!i_ddi_devi_attached(ct->ct_dip))) {
1737 		/*
1738 		 * Client is in failed state. Nothing more to do.
1739 		 */
1740 		MDI_CLIENT_UNLOCK(ct);
1741 		return (MDI_FAILURE);
1742 	}
1743 
1744 	/*
1745 	 * Set the client state as failover in progress.
1746 	 */
1747 	MDI_CLIENT_SET_FAILOVER_IN_PROGRESS(ct);
1748 	ct->ct_failover_flags = flags;
1749 	MDI_CLIENT_UNLOCK(ct);
1750 
1751 	if (flags == MDI_FAILOVER_ASYNC) {
1752 		/*
1753 		 * Submit the initiate failover request via CPR safe
1754 		 * taskq threads.
1755 		 */
1756 		(void) taskq_dispatch(mdi_taskq, (task_func_t *)i_mdi_failover,
1757 		    ct, KM_SLEEP);
1758 		return (MDI_ACCEPT);
1759 	} else {
1760 		/*
1761 		 * Synchronous failover mode.  Typically invoked from the user
1762 		 * land.
1763 		 */
1764 		rv = i_mdi_failover(ct);
1765 	}
1766 	return (rv);
1767 }
1768 
1769 /*
1770  * i_mdi_failover():
1771  *		internal failover function. Invokes vHCI drivers failover
1772  *		callback function and process the failover status
1773  * Return Values:
1774  *		None
1775  *
1776  * Note: A client device in failover state can not be detached or freed.
1777  */
1778 static int
1779 i_mdi_failover(void *arg)
1780 {
1781 	int		rv = MDI_SUCCESS;
1782 	mdi_client_t	*ct = (mdi_client_t *)arg;
1783 	mdi_vhci_t	*vh = ct->ct_vhci;
1784 
1785 	ASSERT(!MDI_CLIENT_LOCKED(ct));
1786 
1787 	if (vh->vh_ops->vo_failover != NULL) {
1788 		/*
1789 		 * Call vHCI drivers callback routine
1790 		 */
1791 		rv = (*vh->vh_ops->vo_failover)(vh->vh_dip, ct->ct_dip,
1792 		    ct->ct_failover_flags);
1793 	}
1794 
1795 	MDI_CLIENT_LOCK(ct);
1796 	MDI_CLIENT_CLEAR_FAILOVER_IN_PROGRESS(ct);
1797 
1798 	/*
1799 	 * Save the failover return status
1800 	 */
1801 	ct->ct_failover_status = rv;
1802 
1803 	/*
1804 	 * As a result of failover, client status would have been changed.
1805 	 * Update the client state and wake up anyone waiting on this client
1806 	 * device.
1807 	 */
1808 	i_mdi_client_update_state(ct);
1809 
1810 	cv_broadcast(&ct->ct_failover_cv);
1811 	MDI_CLIENT_UNLOCK(ct);
1812 	return (rv);
1813 }
1814 
1815 /*
1816  * Load balancing is logical block.
1817  * IOs within the range described by region_size
1818  * would go on the same path. This would improve the
1819  * performance by cache-hit on some of the RAID devices.
1820  * Search only for online paths(At some point we
1821  * may want to balance across target ports).
1822  * If no paths are found then default to round-robin.
1823  */
1824 static int
1825 i_mdi_lba_lb(mdi_client_t *ct, mdi_pathinfo_t **ret_pip, struct buf *bp)
1826 {
1827 	int		path_index = -1;
1828 	int		online_path_count = 0;
1829 	int		online_nonpref_path_count = 0;
1830 	int 		region_size = ct->ct_lb_args->region_size;
1831 	mdi_pathinfo_t	*pip;
1832 	mdi_pathinfo_t	*next;
1833 	int		preferred, path_cnt;
1834 
1835 	pip = ct->ct_path_head;
1836 	while (pip) {
1837 		MDI_PI_LOCK(pip);
1838 		if (MDI_PI(pip)->pi_state ==
1839 		    MDI_PATHINFO_STATE_ONLINE && MDI_PI(pip)->pi_preferred) {
1840 			online_path_count++;
1841 		} else if (MDI_PI(pip)->pi_state ==
1842 		    MDI_PATHINFO_STATE_ONLINE && !MDI_PI(pip)->pi_preferred) {
1843 			online_nonpref_path_count++;
1844 		}
1845 		next = (mdi_pathinfo_t *)
1846 		    MDI_PI(pip)->pi_client_link;
1847 		MDI_PI_UNLOCK(pip);
1848 		pip = next;
1849 	}
1850 	/* if found any online/preferred then use this type */
1851 	if (online_path_count > 0) {
1852 		path_cnt = online_path_count;
1853 		preferred = 1;
1854 	} else if (online_nonpref_path_count > 0) {
1855 		path_cnt = online_nonpref_path_count;
1856 		preferred = 0;
1857 	} else {
1858 		path_cnt = 0;
1859 	}
1860 	if (path_cnt) {
1861 		path_index = (bp->b_blkno >> region_size) % path_cnt;
1862 		pip = ct->ct_path_head;
1863 		while (pip && path_index != -1) {
1864 			MDI_PI_LOCK(pip);
1865 			if (path_index == 0 &&
1866 			    (MDI_PI(pip)->pi_state ==
1867 			    MDI_PATHINFO_STATE_ONLINE) &&
1868 				MDI_PI(pip)->pi_preferred == preferred) {
1869 				MDI_PI_HOLD(pip);
1870 				MDI_PI_UNLOCK(pip);
1871 				*ret_pip = pip;
1872 				return (MDI_SUCCESS);
1873 			}
1874 			path_index --;
1875 			next = (mdi_pathinfo_t *)
1876 			    MDI_PI(pip)->pi_client_link;
1877 			MDI_PI_UNLOCK(pip);
1878 			pip = next;
1879 		}
1880 		if (pip == NULL) {
1881 			MDI_DEBUG(4, (CE_NOTE, NULL,
1882 			    "!lba %llx, no pip !!\n",
1883 				bp->b_lblkno));
1884 		} else {
1885 			MDI_DEBUG(4, (CE_NOTE, NULL,
1886 			    "!lba %llx, no pip for path_index, "
1887 			    "pip %p\n", bp->b_lblkno, (void *)pip));
1888 		}
1889 	}
1890 	return (MDI_FAILURE);
1891 }
1892 
1893 /*
1894  * mdi_select_path():
1895  *		select a path to access a client device.
1896  *
1897  *		mdi_select_path() function is called by the vHCI drivers to
1898  *		select a path to route the I/O request to.  The caller passes
1899  *		the block I/O data transfer structure ("buf") as one of the
1900  *		parameters.  The mpxio framework uses the buf structure
1901  *		contents to maintain per path statistics (total I/O size /
1902  *		count pending).  If more than one online paths are available to
1903  *		select, the framework automatically selects a suitable path
1904  *		for routing I/O request. If a failover operation is active for
1905  *		this client device the call shall be failed with MDI_BUSY error
1906  *		code.
1907  *
1908  *		By default this function returns a suitable path in online
1909  *		state based on the current load balancing policy.  Currently
1910  *		we support LOAD_BALANCE_NONE (Previously selected online path
1911  *		will continue to be used till the path is usable) and
1912  *		LOAD_BALANCE_RR (Online paths will be selected in a round
1913  *		robin fashion), LOAD_BALANCE_LB(Online paths will be selected
1914  *		based on the logical block).  The load balancing
1915  *		through vHCI drivers configuration file (driver.conf).
1916  *
1917  *		vHCI drivers may override this default behavior by specifying
1918  *		appropriate flags.  If start_pip is specified (non NULL) is
1919  *		used as start point to walk and find the next appropriate path.
1920  *		The following values are currently defined:
1921  *		MDI_SELECT_ONLINE_PATH (to select an ONLINE path) and/or
1922  *		MDI_SELECT_STANDBY_PATH (to select an STANDBY path).
1923  *
1924  *		The non-standard behavior is used by the scsi_vhci driver,
1925  *		whenever it has to use a STANDBY/FAULTED path.  Eg. during
1926  *		attach of client devices (to avoid an unnecessary failover
1927  *		when the STANDBY path comes up first), during failover
1928  *		(to activate a STANDBY path as ONLINE).
1929  *
1930  *		The selected path is returned in a a mdi_hold_path() state
1931  *		(pi_ref_cnt). Caller should release the hold by calling
1932  *		mdi_rele_path().
1933  *
1934  * Return Values:
1935  *		MDI_SUCCESS	- Completed successfully
1936  *		MDI_BUSY 	- Client device is busy failing over
1937  *		MDI_NOPATH	- Client device is online, but no valid path are
1938  *				  available to access this client device
1939  *		MDI_FAILURE	- Invalid client device or state
1940  *		MDI_DEVI_ONLINING
1941  *				- Client device (struct dev_info state) is in
1942  *				  onlining state.
1943  */
1944 
1945 /*ARGSUSED*/
1946 int
1947 mdi_select_path(dev_info_t *cdip, struct buf *bp, int flags,
1948     mdi_pathinfo_t *start_pip, mdi_pathinfo_t **ret_pip)
1949 {
1950 	mdi_client_t	*ct;
1951 	mdi_pathinfo_t	*pip;
1952 	mdi_pathinfo_t	*next;
1953 	mdi_pathinfo_t	*head;
1954 	mdi_pathinfo_t	*start;
1955 	client_lb_t	lbp;	/* load balancing policy */
1956 	int		sb = 1;	/* standard behavior */
1957 	int		preferred = 1;	/* preferred path */
1958 	int		cond, cont = 1;
1959 	int		retry = 0;
1960 
1961 	if (flags != 0) {
1962 		/*
1963 		 * disable default behavior
1964 		 */
1965 		sb = 0;
1966 	}
1967 
1968 	*ret_pip = NULL;
1969 	ct = i_devi_get_client(cdip);
1970 	if (ct == NULL) {
1971 		/* mdi extensions are NULL, Nothing more to do */
1972 		return (MDI_FAILURE);
1973 	}
1974 
1975 	MDI_CLIENT_LOCK(ct);
1976 
1977 	if (sb) {
1978 		if (MDI_CLIENT_IS_FAILED(ct)) {
1979 			/*
1980 			 * Client is not ready to accept any I/O requests.
1981 			 * Fail this request.
1982 			 */
1983 			MDI_DEBUG(2, (CE_NOTE, cdip, "!mdi_select_path: "
1984 			    "client state offline ct = %p\n", (void *)ct));
1985 			MDI_CLIENT_UNLOCK(ct);
1986 			return (MDI_FAILURE);
1987 		}
1988 
1989 		if (MDI_CLIENT_IS_FAILOVER_IN_PROGRESS(ct)) {
1990 			/*
1991 			 * Check for Failover is in progress. If so tell the
1992 			 * caller that this device is busy.
1993 			 */
1994 			MDI_DEBUG(2, (CE_NOTE, cdip, "!mdi_select_path: "
1995 			    "client failover in progress ct = %p\n",
1996 			    (void *)ct));
1997 			MDI_CLIENT_UNLOCK(ct);
1998 			return (MDI_BUSY);
1999 		}
2000 
2001 		/*
2002 		 * Check to see whether the client device is attached.
2003 		 * If not so, let the vHCI driver manually select a path
2004 		 * (standby) and let the probe/attach process to continue.
2005 		 */
2006 		if (MDI_CLIENT_IS_DETACHED(ct) || !i_ddi_devi_attached(cdip)) {
2007 			MDI_DEBUG(4, (CE_NOTE, cdip, "!Devi is onlining "
2008 			    "ct = %p\n", (void *)ct));
2009 			MDI_CLIENT_UNLOCK(ct);
2010 			return (MDI_DEVI_ONLINING);
2011 		}
2012 	}
2013 
2014 	/*
2015 	 * Cache in the client list head.  If head of the list is NULL
2016 	 * return MDI_NOPATH
2017 	 */
2018 	head = ct->ct_path_head;
2019 	if (head == NULL) {
2020 		MDI_CLIENT_UNLOCK(ct);
2021 		return (MDI_NOPATH);
2022 	}
2023 
2024 	/*
2025 	 * for non default behavior, bypass current
2026 	 * load balancing policy and always use LOAD_BALANCE_RR
2027 	 * except that the start point will be adjusted based
2028 	 * on the provided start_pip
2029 	 */
2030 	lbp = sb ? ct->ct_lb : LOAD_BALANCE_RR;
2031 
2032 	switch (lbp) {
2033 	case LOAD_BALANCE_NONE:
2034 		/*
2035 		 * Load balancing is None  or Alternate path mode
2036 		 * Start looking for a online mdi_pathinfo node starting from
2037 		 * last known selected path
2038 		 */
2039 		preferred = 1;
2040 		pip = (mdi_pathinfo_t *)ct->ct_path_last;
2041 		if (pip == NULL) {
2042 			pip = head;
2043 		}
2044 		start = pip;
2045 		do {
2046 			MDI_PI_LOCK(pip);
2047 			/*
2048 			 * No need to explicitly check if the path is disabled.
2049 			 * Since we are checking for state == ONLINE and the
2050 			 * same veriable is used for DISABLE/ENABLE information.
2051 			 */
2052 			if ((MDI_PI(pip)->pi_state  ==
2053 				MDI_PATHINFO_STATE_ONLINE) &&
2054 				preferred == MDI_PI(pip)->pi_preferred) {
2055 				/*
2056 				 * Return the path in hold state. Caller should
2057 				 * release the lock by calling mdi_rele_path()
2058 				 */
2059 				MDI_PI_HOLD(pip);
2060 				MDI_PI_UNLOCK(pip);
2061 				ct->ct_path_last = pip;
2062 				*ret_pip = pip;
2063 				MDI_CLIENT_UNLOCK(ct);
2064 				return (MDI_SUCCESS);
2065 			}
2066 
2067 			/*
2068 			 * Path is busy.
2069 			 */
2070 			if (MDI_PI_IS_DRV_DISABLE_TRANSIENT(pip) ||
2071 			    MDI_PI_IS_TRANSIENT(pip))
2072 				retry = 1;
2073 			/*
2074 			 * Keep looking for a next available online path
2075 			 */
2076 			next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
2077 			if (next == NULL) {
2078 				next = head;
2079 			}
2080 			MDI_PI_UNLOCK(pip);
2081 			pip = next;
2082 			if (start == pip && preferred) {
2083 				preferred = 0;
2084 			} else if (start == pip && !preferred) {
2085 				cont = 0;
2086 			}
2087 		} while (cont);
2088 		break;
2089 
2090 	case LOAD_BALANCE_LBA:
2091 		/*
2092 		 * Make sure we are looking
2093 		 * for an online path. Otherwise, if it is for a STANDBY
2094 		 * path request, it will go through and fetch an ONLINE
2095 		 * path which is not desirable.
2096 		 */
2097 		if ((ct->ct_lb_args != NULL) &&
2098 			    (ct->ct_lb_args->region_size) && bp &&
2099 				(sb || (flags == MDI_SELECT_ONLINE_PATH))) {
2100 			if (i_mdi_lba_lb(ct, ret_pip, bp)
2101 				    == MDI_SUCCESS) {
2102 				MDI_CLIENT_UNLOCK(ct);
2103 				return (MDI_SUCCESS);
2104 			}
2105 		}
2106 		/*  FALLTHROUGH */
2107 	case LOAD_BALANCE_RR:
2108 		/*
2109 		 * Load balancing is Round Robin. Start looking for a online
2110 		 * mdi_pathinfo node starting from last known selected path
2111 		 * as the start point.  If override flags are specified,
2112 		 * process accordingly.
2113 		 * If the search is already in effect(start_pip not null),
2114 		 * then lets just use the same path preference to continue the
2115 		 * traversal.
2116 		 */
2117 
2118 		if (start_pip != NULL) {
2119 			preferred = MDI_PI(start_pip)->pi_preferred;
2120 		} else {
2121 			preferred = 1;
2122 		}
2123 
2124 		start = sb ? (mdi_pathinfo_t *)ct->ct_path_last : start_pip;
2125 		if (start == NULL) {
2126 			pip = head;
2127 		} else {
2128 			pip = (mdi_pathinfo_t *)MDI_PI(start)->pi_client_link;
2129 			if (pip == NULL) {
2130 				if (!sb) {
2131 					if (preferred == 0) {
2132 						/*
2133 						 * Looks like we have completed
2134 						 * the traversal as preferred
2135 						 * value is 0. Time to bail out.
2136 						 */
2137 						*ret_pip = NULL;
2138 						MDI_CLIENT_UNLOCK(ct);
2139 						return (MDI_NOPATH);
2140 					} else {
2141 						/*
2142 						 * Looks like we reached the
2143 						 * end of the list. Lets enable
2144 						 * traversal of non preferred
2145 						 * paths.
2146 						 */
2147 						preferred = 0;
2148 					}
2149 				}
2150 				pip = head;
2151 			}
2152 		}
2153 		start = pip;
2154 		do {
2155 			MDI_PI_LOCK(pip);
2156 			if (sb) {
2157 				cond = ((MDI_PI(pip)->pi_state ==
2158 				    MDI_PATHINFO_STATE_ONLINE &&
2159 					MDI_PI(pip)->pi_preferred ==
2160 						preferred) ? 1 : 0);
2161 			} else {
2162 				if (flags == MDI_SELECT_ONLINE_PATH) {
2163 					cond = ((MDI_PI(pip)->pi_state ==
2164 					    MDI_PATHINFO_STATE_ONLINE &&
2165 						MDI_PI(pip)->pi_preferred ==
2166 						preferred) ? 1 : 0);
2167 				} else if (flags == MDI_SELECT_STANDBY_PATH) {
2168 					cond = ((MDI_PI(pip)->pi_state ==
2169 					    MDI_PATHINFO_STATE_STANDBY &&
2170 						MDI_PI(pip)->pi_preferred ==
2171 						preferred) ? 1 : 0);
2172 				} else if (flags == (MDI_SELECT_ONLINE_PATH |
2173 				    MDI_SELECT_STANDBY_PATH)) {
2174 					cond = (((MDI_PI(pip)->pi_state ==
2175 					    MDI_PATHINFO_STATE_ONLINE ||
2176 					    (MDI_PI(pip)->pi_state ==
2177 					    MDI_PATHINFO_STATE_STANDBY)) &&
2178 						MDI_PI(pip)->pi_preferred ==
2179 						preferred) ? 1 : 0);
2180 				} else if (flags ==
2181 					(MDI_SELECT_STANDBY_PATH |
2182 					MDI_SELECT_ONLINE_PATH |
2183 					MDI_SELECT_USER_DISABLE_PATH)) {
2184 					cond = (((MDI_PI(pip)->pi_state ==
2185 					    MDI_PATHINFO_STATE_ONLINE ||
2186 					    (MDI_PI(pip)->pi_state ==
2187 					    MDI_PATHINFO_STATE_STANDBY) ||
2188 						(MDI_PI(pip)->pi_state ==
2189 					    (MDI_PATHINFO_STATE_ONLINE|
2190 					    MDI_PATHINFO_STATE_USER_DISABLE)) ||
2191 						(MDI_PI(pip)->pi_state ==
2192 					    (MDI_PATHINFO_STATE_STANDBY |
2193 					    MDI_PATHINFO_STATE_USER_DISABLE)))&&
2194 						MDI_PI(pip)->pi_preferred ==
2195 						preferred) ? 1 : 0);
2196 				} else {
2197 					cond = 0;
2198 				}
2199 			}
2200 			/*
2201 			 * No need to explicitly check if the path is disabled.
2202 			 * Since we are checking for state == ONLINE and the
2203 			 * same veriable is used for DISABLE/ENABLE information.
2204 			 */
2205 			if (cond) {
2206 				/*
2207 				 * Return the path in hold state. Caller should
2208 				 * release the lock by calling mdi_rele_path()
2209 				 */
2210 				MDI_PI_HOLD(pip);
2211 				MDI_PI_UNLOCK(pip);
2212 				if (sb)
2213 					ct->ct_path_last = pip;
2214 				*ret_pip = pip;
2215 				MDI_CLIENT_UNLOCK(ct);
2216 				return (MDI_SUCCESS);
2217 			}
2218 			/*
2219 			 * Path is busy.
2220 			 */
2221 			if (MDI_PI_IS_DRV_DISABLE_TRANSIENT(pip) ||
2222 			    MDI_PI_IS_TRANSIENT(pip))
2223 				retry = 1;
2224 
2225 			/*
2226 			 * Keep looking for a next available online path
2227 			 */
2228 do_again:
2229 			next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
2230 			if (next == NULL) {
2231 				if (!sb) {
2232 					if (preferred == 1) {
2233 						/*
2234 						 * Looks like we reached the
2235 						 * end of the list. Lets enable
2236 						 * traversal of non preferred
2237 						 * paths.
2238 						 */
2239 						preferred = 0;
2240 						next = head;
2241 					} else {
2242 						/*
2243 						 * We have done both the passes
2244 						 * Preferred as well as for
2245 						 * Non-preferred. Bail out now.
2246 						 */
2247 						cont = 0;
2248 					}
2249 				} else {
2250 					/*
2251 					 * Standard behavior case.
2252 					 */
2253 					next = head;
2254 				}
2255 			}
2256 			MDI_PI_UNLOCK(pip);
2257 			if (cont == 0) {
2258 				break;
2259 			}
2260 			pip = next;
2261 
2262 			if (!sb) {
2263 				/*
2264 				 * We need to handle the selection of
2265 				 * non-preferred path in the following
2266 				 * case:
2267 				 *
2268 				 * +------+   +------+   +------+   +-----+
2269 				 * | A : 1| - | B : 1| - | C : 0| - |NULL |
2270 				 * +------+   +------+   +------+   +-----+
2271 				 *
2272 				 * If we start the search with B, we need to
2273 				 * skip beyond B to pick C which is non -
2274 				 * preferred in the second pass. The following
2275 				 * test, if true, will allow us to skip over
2276 				 * the 'start'(B in the example) to select
2277 				 * other non preferred elements.
2278 				 */
2279 				if ((start_pip != NULL) && (start_pip == pip) &&
2280 				    (MDI_PI(start_pip)->pi_preferred
2281 				    != preferred)) {
2282 					/*
2283 					 * try again after going past the start
2284 					 * pip
2285 					 */
2286 					MDI_PI_LOCK(pip);
2287 					goto do_again;
2288 				}
2289 			} else {
2290 				/*
2291 				 * Standard behavior case
2292 				 */
2293 				if (start == pip && preferred) {
2294 					/* look for nonpreferred paths */
2295 					preferred = 0;
2296 				} else if (start == pip && !preferred) {
2297 					/*
2298 					 * Exit condition
2299 					 */
2300 					cont = 0;
2301 				}
2302 			}
2303 		} while (cont);
2304 		break;
2305 	}
2306 
2307 	MDI_CLIENT_UNLOCK(ct);
2308 	if (retry == 1) {
2309 		return (MDI_BUSY);
2310 	} else {
2311 		return (MDI_NOPATH);
2312 	}
2313 }
2314 
2315 /*
2316  * For a client, return the next available path to any phci
2317  *
2318  * Note:
2319  *		Caller should hold the branch's devinfo node to get a consistent
2320  *		snap shot of the mdi_pathinfo nodes.
2321  *
2322  *		Please note that even the list is stable the mdi_pathinfo
2323  *		node state and properties are volatile.  The caller should lock
2324  *		and unlock the nodes by calling mdi_pi_lock() and
2325  *		mdi_pi_unlock() functions to get a stable properties.
2326  *
2327  *		If there is a need to use the nodes beyond the hold of the
2328  *		devinfo node period (For ex. I/O), then mdi_pathinfo node
2329  *		need to be held against unexpected removal by calling
2330  *		mdi_hold_path() and should be released by calling
2331  *		mdi_rele_path() on completion.
2332  */
2333 mdi_pathinfo_t *
2334 mdi_get_next_phci_path(dev_info_t *ct_dip, mdi_pathinfo_t *pip)
2335 {
2336 	mdi_client_t *ct;
2337 
2338 	if (!MDI_CLIENT(ct_dip))
2339 		return (NULL);
2340 
2341 	/*
2342 	 * Walk through client link
2343 	 */
2344 	ct = (mdi_client_t *)DEVI(ct_dip)->devi_mdi_client;
2345 	ASSERT(ct != NULL);
2346 
2347 	if (pip == NULL)
2348 		return ((mdi_pathinfo_t *)ct->ct_path_head);
2349 
2350 	return ((mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link);
2351 }
2352 
2353 /*
2354  * For a phci, return the next available path to any client
2355  * Note: ditto mdi_get_next_phci_path()
2356  */
2357 mdi_pathinfo_t *
2358 mdi_get_next_client_path(dev_info_t *ph_dip, mdi_pathinfo_t *pip)
2359 {
2360 	mdi_phci_t *ph;
2361 
2362 	if (!MDI_PHCI(ph_dip))
2363 		return (NULL);
2364 
2365 	/*
2366 	 * Walk through pHCI link
2367 	 */
2368 	ph = (mdi_phci_t *)DEVI(ph_dip)->devi_mdi_xhci;
2369 	ASSERT(ph != NULL);
2370 
2371 	if (pip == NULL)
2372 		return ((mdi_pathinfo_t *)ph->ph_path_head);
2373 
2374 	return ((mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link);
2375 }
2376 
2377 /*
2378  * mdi_hold_path():
2379  *		Hold the mdi_pathinfo node against unwanted unexpected free.
2380  * Return Values:
2381  *		None
2382  */
2383 void
2384 mdi_hold_path(mdi_pathinfo_t *pip)
2385 {
2386 	if (pip) {
2387 		MDI_PI_LOCK(pip);
2388 		MDI_PI_HOLD(pip);
2389 		MDI_PI_UNLOCK(pip);
2390 	}
2391 }
2392 
2393 
2394 /*
2395  * mdi_rele_path():
2396  *		Release the mdi_pathinfo node which was selected
2397  *		through mdi_select_path() mechanism or manually held by
2398  *		calling mdi_hold_path().
2399  * Return Values:
2400  *		None
2401  */
2402 void
2403 mdi_rele_path(mdi_pathinfo_t *pip)
2404 {
2405 	if (pip) {
2406 		MDI_PI_LOCK(pip);
2407 		MDI_PI_RELE(pip);
2408 		if (MDI_PI(pip)->pi_ref_cnt == 0) {
2409 			cv_broadcast(&MDI_PI(pip)->pi_ref_cv);
2410 		}
2411 		MDI_PI_UNLOCK(pip);
2412 	}
2413 }
2414 
2415 /*
2416  * mdi_pi_lock():
2417  * 		Lock the mdi_pathinfo node.
2418  * Note:
2419  *		The caller should release the lock by calling mdi_pi_unlock()
2420  */
2421 void
2422 mdi_pi_lock(mdi_pathinfo_t *pip)
2423 {
2424 	ASSERT(pip != NULL);
2425 	if (pip) {
2426 		MDI_PI_LOCK(pip);
2427 	}
2428 }
2429 
2430 
2431 /*
2432  * mdi_pi_unlock():
2433  * 		Unlock the mdi_pathinfo node.
2434  * Note:
2435  *		The mdi_pathinfo node should have been locked with mdi_pi_lock()
2436  */
2437 void
2438 mdi_pi_unlock(mdi_pathinfo_t *pip)
2439 {
2440 	ASSERT(pip != NULL);
2441 	if (pip) {
2442 		MDI_PI_UNLOCK(pip);
2443 	}
2444 }
2445 
2446 /*
2447  * mdi_pi_find():
2448  *		Search the list of mdi_pathinfo nodes attached to the
2449  *		pHCI/Client device node whose path address matches "paddr".
2450  *		Returns a pointer to the mdi_pathinfo node if a matching node is
2451  *		found.
2452  * Return Values:
2453  *		mdi_pathinfo node handle
2454  *		NULL
2455  * Notes:
2456  *		Caller need not hold any locks to call this function.
2457  */
2458 mdi_pathinfo_t *
2459 mdi_pi_find(dev_info_t *pdip, char *caddr, char *paddr)
2460 {
2461 	mdi_phci_t		*ph;
2462 	mdi_vhci_t		*vh;
2463 	mdi_client_t		*ct;
2464 	mdi_pathinfo_t		*pip = NULL;
2465 
2466 	MDI_DEBUG(2, (CE_NOTE, pdip, "!mdi_pi_find: %s %s",
2467 	    caddr ? caddr : "NULL", paddr ? paddr : "NULL"));
2468 	if ((pdip == NULL) || (paddr == NULL)) {
2469 		return (NULL);
2470 	}
2471 	ph = i_devi_get_phci(pdip);
2472 	if (ph == NULL) {
2473 		/*
2474 		 * Invalid pHCI device, Nothing more to do.
2475 		 */
2476 		MDI_DEBUG(2, (CE_WARN, pdip,
2477 		    "!mdi_pi_find: invalid phci"));
2478 		return (NULL);
2479 	}
2480 
2481 	vh = ph->ph_vhci;
2482 	if (vh == NULL) {
2483 		/*
2484 		 * Invalid vHCI device, Nothing more to do.
2485 		 */
2486 		MDI_DEBUG(2, (CE_WARN, pdip,
2487 		    "!mdi_pi_find: invalid vhci"));
2488 		return (NULL);
2489 	}
2490 
2491 	/*
2492 	 * Look for pathinfo node identified by paddr.
2493 	 */
2494 	if (caddr == NULL) {
2495 		/*
2496 		 * Find a mdi_pathinfo node under pHCI list for a matching
2497 		 * unit address.
2498 		 */
2499 		MDI_PHCI_LOCK(ph);
2500 		if (MDI_PHCI_IS_OFFLINE(ph)) {
2501 			MDI_DEBUG(2, (CE_WARN, pdip,
2502 			    "!mdi_pi_find: offline phci %p", (void *)ph));
2503 			MDI_PHCI_UNLOCK(ph);
2504 			return (NULL);
2505 		}
2506 		pip = (mdi_pathinfo_t *)ph->ph_path_head;
2507 
2508 		while (pip != NULL) {
2509 			if (strcmp(MDI_PI(pip)->pi_addr, paddr) == 0) {
2510 				break;
2511 			}
2512 			pip = (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
2513 		}
2514 		MDI_PHCI_UNLOCK(ph);
2515 		MDI_DEBUG(2, (CE_NOTE, pdip, "!mdi_pi_find: found %p",
2516 		    (void *)pip));
2517 		return (pip);
2518 	}
2519 
2520 	/*
2521 	 * XXX - Is the rest of the code in this function really necessary?
2522 	 * The consumers of mdi_pi_find() can search for the desired pathinfo
2523 	 * node by calling mdi_pi_find(pdip, NULL, paddr). Irrespective of
2524 	 * whether the search is based on the pathinfo nodes attached to
2525 	 * the pHCI or the client node, the result will be the same.
2526 	 */
2527 
2528 	/*
2529 	 * Find the client device corresponding to 'caddr'
2530 	 */
2531 	MDI_VHCI_CLIENT_LOCK(vh);
2532 
2533 	/*
2534 	 * XXX - Passing NULL to the following function works as long as the
2535 	 * the client addresses (caddr) are unique per vhci basis.
2536 	 */
2537 	ct = i_mdi_client_find(vh, NULL, caddr);
2538 	if (ct == NULL) {
2539 		/*
2540 		 * Client not found, Obviously mdi_pathinfo node has not been
2541 		 * created yet.
2542 		 */
2543 		MDI_VHCI_CLIENT_UNLOCK(vh);
2544 		MDI_DEBUG(2, (CE_NOTE, pdip, "!mdi_pi_find: client not "
2545 		    "found for caddr %s", caddr ? caddr : "NULL"));
2546 		return (NULL);
2547 	}
2548 
2549 	/*
2550 	 * Hold the client lock and look for a mdi_pathinfo node with matching
2551 	 * pHCI and paddr
2552 	 */
2553 	MDI_CLIENT_LOCK(ct);
2554 
2555 	/*
2556 	 * Release the global mutex as it is no more needed. Note: We always
2557 	 * respect the locking order while acquiring.
2558 	 */
2559 	MDI_VHCI_CLIENT_UNLOCK(vh);
2560 
2561 	pip = (mdi_pathinfo_t *)ct->ct_path_head;
2562 	while (pip != NULL) {
2563 		/*
2564 		 * Compare the unit address
2565 		 */
2566 		if ((MDI_PI(pip)->pi_phci == ph) &&
2567 		    strcmp(MDI_PI(pip)->pi_addr, paddr) == 0) {
2568 			break;
2569 		}
2570 		pip = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
2571 	}
2572 	MDI_CLIENT_UNLOCK(ct);
2573 	MDI_DEBUG(2, (CE_NOTE, pdip, "!mdi_pi_find: found:: %p", (void *)pip));
2574 	return (pip);
2575 }
2576 
2577 /*
2578  * mdi_pi_alloc():
2579  *		Allocate and initialize a new instance of a mdi_pathinfo node.
2580  *		The mdi_pathinfo node returned by this function identifies a
2581  *		unique device path is capable of having properties attached
2582  *		and passed to mdi_pi_online() to fully attach and online the
2583  *		path and client device node.
2584  *		The mdi_pathinfo node returned by this function must be
2585  *		destroyed using mdi_pi_free() if the path is no longer
2586  *		operational or if the caller fails to attach a client device
2587  *		node when calling mdi_pi_online(). The framework will not free
2588  *		the resources allocated.
2589  *		This function can be called from both interrupt and kernel
2590  *		contexts.  DDI_NOSLEEP flag should be used while calling
2591  *		from interrupt contexts.
2592  * Return Values:
2593  *		MDI_SUCCESS
2594  *		MDI_FAILURE
2595  *		MDI_NOMEM
2596  */
2597 /*ARGSUSED*/
2598 int
2599 mdi_pi_alloc_compatible(dev_info_t *pdip, char *cname, char *caddr, char *paddr,
2600     char **compatible, int ncompatible, int flags, mdi_pathinfo_t **ret_pip)
2601 {
2602 	mdi_vhci_t	*vh;
2603 	mdi_phci_t	*ph;
2604 	mdi_client_t	*ct;
2605 	mdi_pathinfo_t	*pip = NULL;
2606 	dev_info_t	*cdip;
2607 	int		rv = MDI_NOMEM;
2608 	int		path_allocated = 0;
2609 
2610 	MDI_DEBUG(2, (CE_NOTE, pdip, "!mdi_pi_alloc_compatible: %s %s %s",
2611 	    cname ? cname : "NULL", caddr ? caddr : "NULL",
2612 	    paddr ? paddr : "NULL"));
2613 
2614 	if (pdip == NULL || cname == NULL || caddr == NULL || paddr == NULL ||
2615 	    ret_pip == NULL) {
2616 		/* Nothing more to do */
2617 		return (MDI_FAILURE);
2618 	}
2619 
2620 	*ret_pip = NULL;
2621 
2622 	/* No allocations on detaching pHCI */
2623 	if (DEVI_IS_DETACHING(pdip)) {
2624 		/* Invalid pHCI device, return failure */
2625 		MDI_DEBUG(1, (CE_WARN, pdip,
2626 		    "!mdi_pi_alloc: detaching pHCI=%p", (void *)pdip));
2627 		return (MDI_FAILURE);
2628 	}
2629 
2630 	ph = i_devi_get_phci(pdip);
2631 	ASSERT(ph != NULL);
2632 	if (ph == NULL) {
2633 		/* Invalid pHCI device, return failure */
2634 		MDI_DEBUG(1, (CE_WARN, pdip,
2635 		    "!mdi_pi_alloc: invalid pHCI=%p", (void *)pdip));
2636 		return (MDI_FAILURE);
2637 	}
2638 
2639 	MDI_PHCI_LOCK(ph);
2640 	vh = ph->ph_vhci;
2641 	if (vh == NULL) {
2642 		/* Invalid vHCI device, return failure */
2643 		MDI_DEBUG(1, (CE_WARN, pdip,
2644 		    "!mdi_pi_alloc: invalid vHCI=%p", (void *)pdip));
2645 		MDI_PHCI_UNLOCK(ph);
2646 		return (MDI_FAILURE);
2647 	}
2648 
2649 	if (MDI_PHCI_IS_READY(ph) == 0) {
2650 		/*
2651 		 * Do not allow new node creation when pHCI is in
2652 		 * offline/suspended states
2653 		 */
2654 		MDI_DEBUG(1, (CE_WARN, pdip,
2655 		    "mdi_pi_alloc: pHCI=%p is not ready", (void *)ph));
2656 		MDI_PHCI_UNLOCK(ph);
2657 		return (MDI_BUSY);
2658 	}
2659 	MDI_PHCI_UNSTABLE(ph);
2660 	MDI_PHCI_UNLOCK(ph);
2661 
2662 	/* look for a matching client, create one if not found */
2663 	MDI_VHCI_CLIENT_LOCK(vh);
2664 	ct = i_mdi_client_find(vh, cname, caddr);
2665 	if (ct == NULL) {
2666 		ct = i_mdi_client_alloc(vh, cname, caddr);
2667 		ASSERT(ct != NULL);
2668 	}
2669 
2670 	if (ct->ct_dip == NULL) {
2671 		/*
2672 		 * Allocate a devinfo node
2673 		 */
2674 		ct->ct_dip = i_mdi_devinfo_create(vh, cname, caddr,
2675 		    compatible, ncompatible);
2676 		if (ct->ct_dip == NULL) {
2677 			(void) i_mdi_client_free(vh, ct);
2678 			goto fail;
2679 		}
2680 	}
2681 	cdip = ct->ct_dip;
2682 
2683 	DEVI(cdip)->devi_mdi_component |= MDI_COMPONENT_CLIENT;
2684 	DEVI(cdip)->devi_mdi_client = (caddr_t)ct;
2685 
2686 	MDI_CLIENT_LOCK(ct);
2687 	pip = (mdi_pathinfo_t *)ct->ct_path_head;
2688 	while (pip != NULL) {
2689 		/*
2690 		 * Compare the unit address
2691 		 */
2692 		if ((MDI_PI(pip)->pi_phci == ph) &&
2693 		    strcmp(MDI_PI(pip)->pi_addr, paddr) == 0) {
2694 			break;
2695 		}
2696 		pip = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
2697 	}
2698 	MDI_CLIENT_UNLOCK(ct);
2699 
2700 	if (pip == NULL) {
2701 		/*
2702 		 * This is a new path for this client device.  Allocate and
2703 		 * initialize a new pathinfo node
2704 		 */
2705 		pip = i_mdi_pi_alloc(ph, paddr, ct);
2706 		ASSERT(pip != NULL);
2707 		path_allocated = 1;
2708 	}
2709 	rv = MDI_SUCCESS;
2710 
2711 fail:
2712 	/*
2713 	 * Release the global mutex.
2714 	 */
2715 	MDI_VHCI_CLIENT_UNLOCK(vh);
2716 
2717 	/*
2718 	 * Mark the pHCI as stable
2719 	 */
2720 	MDI_PHCI_LOCK(ph);
2721 	MDI_PHCI_STABLE(ph);
2722 	MDI_PHCI_UNLOCK(ph);
2723 	*ret_pip = pip;
2724 
2725 	MDI_DEBUG(2, (CE_NOTE, pdip,
2726 	    "!mdi_pi_alloc_compatible: alloc %p", (void *)pip));
2727 
2728 	if (path_allocated)
2729 		vhcache_pi_add(vh->vh_config, MDI_PI(pip));
2730 
2731 	return (rv);
2732 }
2733 
2734 /*ARGSUSED*/
2735 int
2736 mdi_pi_alloc(dev_info_t *pdip, char *cname, char *caddr, char *paddr,
2737     int flags, mdi_pathinfo_t **ret_pip)
2738 {
2739 	return (mdi_pi_alloc_compatible(pdip, cname, caddr, paddr, NULL, 0,
2740 	    flags, ret_pip));
2741 }
2742 
2743 /*
2744  * i_mdi_pi_alloc():
2745  *		Allocate a mdi_pathinfo node and add to the pHCI path list
2746  * Return Values:
2747  *		mdi_pathinfo
2748  */
2749 /*ARGSUSED*/
2750 static mdi_pathinfo_t *
2751 i_mdi_pi_alloc(mdi_phci_t *ph, char *paddr, mdi_client_t *ct)
2752 {
2753 	mdi_pathinfo_t	*pip;
2754 	int		ct_circular;
2755 	int		ph_circular;
2756 	int		se_flag;
2757 	int		kmem_flag;
2758 
2759 	ASSERT(MDI_VHCI_CLIENT_LOCKED(ph->ph_vhci));
2760 
2761 	pip = kmem_zalloc(sizeof (struct mdi_pathinfo), KM_SLEEP);
2762 	mutex_init(&MDI_PI(pip)->pi_mutex, NULL, MUTEX_DEFAULT, NULL);
2763 	MDI_PI(pip)->pi_state = MDI_PATHINFO_STATE_INIT |
2764 	    MDI_PATHINFO_STATE_TRANSIENT;
2765 
2766 	if (MDI_PHCI_IS_USER_DISABLED(ph))
2767 		MDI_PI_SET_USER_DISABLE(pip);
2768 
2769 	if (MDI_PHCI_IS_DRV_DISABLED_TRANSIENT(ph))
2770 		MDI_PI_SET_DRV_DISABLE_TRANS(pip);
2771 
2772 	if (MDI_PHCI_IS_DRV_DISABLED(ph))
2773 		MDI_PI_SET_DRV_DISABLE(pip);
2774 
2775 	MDI_PI(pip)->pi_old_state = MDI_PATHINFO_STATE_INIT;
2776 	cv_init(&MDI_PI(pip)->pi_state_cv, NULL, CV_DEFAULT, NULL);
2777 	MDI_PI(pip)->pi_client = ct;
2778 	MDI_PI(pip)->pi_phci = ph;
2779 	MDI_PI(pip)->pi_addr = kmem_alloc(strlen(paddr) + 1, KM_SLEEP);
2780 	(void) strcpy(MDI_PI(pip)->pi_addr, paddr);
2781 	(void) nvlist_alloc(&MDI_PI(pip)->pi_prop, NV_UNIQUE_NAME, KM_SLEEP);
2782 	ASSERT(MDI_PI(pip)->pi_prop != NULL);
2783 	MDI_PI(pip)->pi_pprivate = NULL;
2784 	MDI_PI(pip)->pi_cprivate = NULL;
2785 	MDI_PI(pip)->pi_vprivate = NULL;
2786 	MDI_PI(pip)->pi_client_link = NULL;
2787 	MDI_PI(pip)->pi_phci_link = NULL;
2788 	MDI_PI(pip)->pi_ref_cnt = 0;
2789 	MDI_PI(pip)->pi_kstats = NULL;
2790 	MDI_PI(pip)->pi_preferred = 1;
2791 	cv_init(&MDI_PI(pip)->pi_ref_cv, NULL, CV_DEFAULT, NULL);
2792 
2793 	/*
2794 	 * Lock both dev_info nodes against changes in parallel.
2795 	 *
2796 	 * The ndi_devi_enter(Client), is atypical since the client is a leaf.
2797 	 * This atypical operation is done to synchronize pathinfo nodes
2798 	 * during devinfo snapshot (see di_register_pip) by 'pretending' that
2799 	 * the pathinfo nodes are children of the Client.
2800 	 */
2801 	ndi_devi_enter(ct->ct_dip, &ct_circular);
2802 	ndi_devi_enter(ph->ph_dip, &ph_circular);
2803 
2804 	i_mdi_phci_add_path(ph, pip);
2805 	i_mdi_client_add_path(ct, pip);
2806 
2807 	ndi_devi_exit(ph->ph_dip, ph_circular);
2808 	ndi_devi_exit(ct->ct_dip, ct_circular);
2809 
2810 	/* determine interrupt context */
2811 	se_flag = (servicing_interrupt()) ? SE_NOSLEEP : SE_SLEEP;
2812 	kmem_flag = (se_flag == SE_SLEEP) ? KM_SLEEP : KM_NOSLEEP;
2813 
2814 	i_ddi_di_cache_invalidate(kmem_flag);
2815 
2816 	return (pip);
2817 }
2818 
2819 /*
2820  * i_mdi_phci_add_path():
2821  * 		Add a mdi_pathinfo node to pHCI list.
2822  * Notes:
2823  *		Caller should per-pHCI mutex
2824  */
2825 static void
2826 i_mdi_phci_add_path(mdi_phci_t *ph, mdi_pathinfo_t *pip)
2827 {
2828 	ASSERT(DEVI_BUSY_OWNED(ph->ph_dip));
2829 
2830 	MDI_PHCI_LOCK(ph);
2831 	if (ph->ph_path_head == NULL) {
2832 		ph->ph_path_head = pip;
2833 	} else {
2834 		MDI_PI(ph->ph_path_tail)->pi_phci_link = MDI_PI(pip);
2835 	}
2836 	ph->ph_path_tail = pip;
2837 	ph->ph_path_count++;
2838 	MDI_PHCI_UNLOCK(ph);
2839 }
2840 
2841 /*
2842  * i_mdi_client_add_path():
2843  *		Add mdi_pathinfo node to client list
2844  */
2845 static void
2846 i_mdi_client_add_path(mdi_client_t *ct, mdi_pathinfo_t *pip)
2847 {
2848 	ASSERT(DEVI_BUSY_OWNED(ct->ct_dip));
2849 
2850 	MDI_CLIENT_LOCK(ct);
2851 	if (ct->ct_path_head == NULL) {
2852 		ct->ct_path_head = pip;
2853 	} else {
2854 		MDI_PI(ct->ct_path_tail)->pi_client_link = MDI_PI(pip);
2855 	}
2856 	ct->ct_path_tail = pip;
2857 	ct->ct_path_count++;
2858 	MDI_CLIENT_UNLOCK(ct);
2859 }
2860 
2861 /*
2862  * mdi_pi_free():
2863  *		Free the mdi_pathinfo node and also client device node if this
2864  *		is the last path to the device
2865  * Return Values:
2866  *		MDI_SUCCESS
2867  *		MDI_FAILURE
2868  *		MDI_BUSY
2869  */
2870 /*ARGSUSED*/
2871 int
2872 mdi_pi_free(mdi_pathinfo_t *pip, int flags)
2873 {
2874 	int		rv = MDI_SUCCESS;
2875 	mdi_vhci_t	*vh;
2876 	mdi_phci_t	*ph;
2877 	mdi_client_t	*ct;
2878 	int		(*f)();
2879 	int		client_held = 0;
2880 
2881 	MDI_PI_LOCK(pip);
2882 	ph = MDI_PI(pip)->pi_phci;
2883 	ASSERT(ph != NULL);
2884 	if (ph == NULL) {
2885 		/*
2886 		 * Invalid pHCI device, return failure
2887 		 */
2888 		MDI_DEBUG(1, (CE_WARN, NULL,
2889 		    "!mdi_pi_free: invalid pHCI pip=%p", (void *)pip));
2890 		MDI_PI_UNLOCK(pip);
2891 		return (MDI_FAILURE);
2892 	}
2893 
2894 	vh = ph->ph_vhci;
2895 	ASSERT(vh != NULL);
2896 	if (vh == NULL) {
2897 		/* Invalid pHCI device, return failure */
2898 		MDI_DEBUG(1, (CE_WARN, NULL,
2899 		    "!mdi_pi_free: invalid vHCI pip=%p", (void *)pip));
2900 		MDI_PI_UNLOCK(pip);
2901 		return (MDI_FAILURE);
2902 	}
2903 
2904 	ct = MDI_PI(pip)->pi_client;
2905 	ASSERT(ct != NULL);
2906 	if (ct == NULL) {
2907 		/*
2908 		 * Invalid Client device, return failure
2909 		 */
2910 		MDI_DEBUG(1, (CE_WARN, NULL,
2911 		    "!mdi_pi_free: invalid client pip=%p", (void *)pip));
2912 		MDI_PI_UNLOCK(pip);
2913 		return (MDI_FAILURE);
2914 	}
2915 
2916 	/*
2917 	 * Check to see for busy condition.  A mdi_pathinfo can only be freed
2918 	 * if the node state is either offline or init and the reference count
2919 	 * is zero.
2920 	 */
2921 	if (!(MDI_PI_IS_OFFLINE(pip) || MDI_PI_IS_INIT(pip) ||
2922 	    MDI_PI_IS_INITING(pip))) {
2923 		/*
2924 		 * Node is busy
2925 		 */
2926 		MDI_DEBUG(1, (CE_WARN, ct->ct_dip,
2927 		    "!mdi_pi_free: pathinfo node is busy pip=%p", (void *)pip));
2928 		MDI_PI_UNLOCK(pip);
2929 		return (MDI_BUSY);
2930 	}
2931 
2932 	while (MDI_PI(pip)->pi_ref_cnt != 0) {
2933 		/*
2934 		 * Give a chance for pending I/Os to complete.
2935 		 */
2936 		MDI_DEBUG(1, (CE_NOTE, ct->ct_dip, "!mdi_pi_free: "
2937 		    "%d cmds still pending on path: %p\n",
2938 		    MDI_PI(pip)->pi_ref_cnt, (void *)pip));
2939 		if (cv_timedwait(&MDI_PI(pip)->pi_ref_cv,
2940 		    &MDI_PI(pip)->pi_mutex,
2941 		    ddi_get_lbolt() + drv_usectohz(60 * 1000000)) == -1) {
2942 			/*
2943 			 * The timeout time reached without ref_cnt being zero
2944 			 * being signaled.
2945 			 */
2946 			MDI_DEBUG(1, (CE_NOTE, ct->ct_dip,
2947 			    "!mdi_pi_free: "
2948 			    "Timeout reached on path %p without the cond\n",
2949 			    (void *)pip));
2950 			MDI_DEBUG(1, (CE_NOTE, ct->ct_dip,
2951 			    "!mdi_pi_free: "
2952 			    "%d cmds still pending on path: %p\n",
2953 			    MDI_PI(pip)->pi_ref_cnt, (void *)pip));
2954 			MDI_PI_UNLOCK(pip);
2955 			return (MDI_BUSY);
2956 		}
2957 	}
2958 	if (MDI_PI(pip)->pi_pm_held) {
2959 		client_held = 1;
2960 	}
2961 	MDI_PI_UNLOCK(pip);
2962 
2963 	vhcache_pi_remove(vh->vh_config, MDI_PI(pip));
2964 
2965 	MDI_CLIENT_LOCK(ct);
2966 
2967 	/* Prevent further failovers till MDI_VHCI_CLIENT_LOCK is held */
2968 	MDI_CLIENT_SET_PATH_FREE_IN_PROGRESS(ct);
2969 
2970 	/*
2971 	 * Wait till failover is complete before removing this node.
2972 	 */
2973 	while (MDI_CLIENT_IS_FAILOVER_IN_PROGRESS(ct))
2974 		cv_wait(&ct->ct_failover_cv, &ct->ct_mutex);
2975 
2976 	MDI_CLIENT_UNLOCK(ct);
2977 	MDI_VHCI_CLIENT_LOCK(vh);
2978 	MDI_CLIENT_LOCK(ct);
2979 	MDI_CLIENT_CLEAR_PATH_FREE_IN_PROGRESS(ct);
2980 
2981 	if (!MDI_PI_IS_INITING(pip)) {
2982 		f = vh->vh_ops->vo_pi_uninit;
2983 		if (f != NULL) {
2984 			rv = (*f)(vh->vh_dip, pip, 0);
2985 		}
2986 	}
2987 	/*
2988 	 * If vo_pi_uninit() completed successfully.
2989 	 */
2990 	if (rv == MDI_SUCCESS) {
2991 		if (client_held) {
2992 			MDI_DEBUG(4, (CE_NOTE, ct->ct_dip, "mdi_pi_free "
2993 			    "i_mdi_pm_rele_client\n"));
2994 			i_mdi_pm_rele_client(ct, 1);
2995 		}
2996 		i_mdi_pi_free(ph, pip, ct);
2997 		if (ct->ct_path_count == 0) {
2998 			/*
2999 			 * Client lost its last path.
3000 			 * Clean up the client device
3001 			 */
3002 			MDI_CLIENT_UNLOCK(ct);
3003 			(void) i_mdi_client_free(ct->ct_vhci, ct);
3004 			MDI_VHCI_CLIENT_UNLOCK(vh);
3005 			return (rv);
3006 		}
3007 	}
3008 	MDI_CLIENT_UNLOCK(ct);
3009 	MDI_VHCI_CLIENT_UNLOCK(vh);
3010 
3011 	if (rv == MDI_FAILURE)
3012 		vhcache_pi_add(vh->vh_config, MDI_PI(pip));
3013 
3014 	return (rv);
3015 }
3016 
3017 /*
3018  * i_mdi_pi_free():
3019  *		Free the mdi_pathinfo node
3020  */
3021 static void
3022 i_mdi_pi_free(mdi_phci_t *ph, mdi_pathinfo_t *pip, mdi_client_t *ct)
3023 {
3024 	int	ct_circular;
3025 	int	ph_circular;
3026 	int	se_flag;
3027 	int	kmem_flag;
3028 
3029 	ASSERT(MDI_CLIENT_LOCKED(ct));
3030 
3031 	/*
3032 	 * remove any per-path kstats
3033 	 */
3034 	i_mdi_pi_kstat_destroy(pip);
3035 
3036 	/* See comments in i_mdi_pi_alloc() */
3037 	ndi_devi_enter(ct->ct_dip, &ct_circular);
3038 	ndi_devi_enter(ph->ph_dip, &ph_circular);
3039 
3040 	i_mdi_client_remove_path(ct, pip);
3041 	i_mdi_phci_remove_path(ph, pip);
3042 
3043 	ndi_devi_exit(ph->ph_dip, ph_circular);
3044 	ndi_devi_exit(ct->ct_dip, ct_circular);
3045 
3046 	/* determine interrupt context */
3047 	se_flag = (servicing_interrupt()) ? SE_NOSLEEP : SE_SLEEP;
3048 	kmem_flag = (se_flag == SE_SLEEP) ? KM_SLEEP : KM_NOSLEEP;
3049 
3050 	i_ddi_di_cache_invalidate(kmem_flag);
3051 
3052 	mutex_destroy(&MDI_PI(pip)->pi_mutex);
3053 	cv_destroy(&MDI_PI(pip)->pi_state_cv);
3054 	cv_destroy(&MDI_PI(pip)->pi_ref_cv);
3055 	if (MDI_PI(pip)->pi_addr) {
3056 		kmem_free(MDI_PI(pip)->pi_addr,
3057 		    strlen(MDI_PI(pip)->pi_addr) + 1);
3058 		MDI_PI(pip)->pi_addr = NULL;
3059 	}
3060 
3061 	if (MDI_PI(pip)->pi_prop) {
3062 		(void) nvlist_free(MDI_PI(pip)->pi_prop);
3063 		MDI_PI(pip)->pi_prop = NULL;
3064 	}
3065 	kmem_free(pip, sizeof (struct mdi_pathinfo));
3066 }
3067 
3068 
3069 /*
3070  * i_mdi_phci_remove_path():
3071  * 		Remove a mdi_pathinfo node from pHCI list.
3072  * Notes:
3073  *		Caller should hold per-pHCI mutex
3074  */
3075 static void
3076 i_mdi_phci_remove_path(mdi_phci_t *ph, mdi_pathinfo_t *pip)
3077 {
3078 	mdi_pathinfo_t	*prev = NULL;
3079 	mdi_pathinfo_t	*path = NULL;
3080 
3081 	ASSERT(DEVI_BUSY_OWNED(ph->ph_dip));
3082 
3083 	MDI_PHCI_LOCK(ph);
3084 	path = ph->ph_path_head;
3085 	while (path != NULL) {
3086 		if (path == pip) {
3087 			break;
3088 		}
3089 		prev = path;
3090 		path = (mdi_pathinfo_t *)MDI_PI(path)->pi_phci_link;
3091 	}
3092 
3093 	if (path) {
3094 		ph->ph_path_count--;
3095 		if (prev) {
3096 			MDI_PI(prev)->pi_phci_link = MDI_PI(path)->pi_phci_link;
3097 		} else {
3098 			ph->ph_path_head =
3099 			    (mdi_pathinfo_t *)MDI_PI(path)->pi_phci_link;
3100 		}
3101 		if (ph->ph_path_tail == path) {
3102 			ph->ph_path_tail = prev;
3103 		}
3104 	}
3105 
3106 	/*
3107 	 * Clear the pHCI link
3108 	 */
3109 	MDI_PI(pip)->pi_phci_link = NULL;
3110 	MDI_PI(pip)->pi_phci = NULL;
3111 	MDI_PHCI_UNLOCK(ph);
3112 }
3113 
3114 /*
3115  * i_mdi_client_remove_path():
3116  * 		Remove a mdi_pathinfo node from client path list.
3117  */
3118 static void
3119 i_mdi_client_remove_path(mdi_client_t *ct, mdi_pathinfo_t *pip)
3120 {
3121 	mdi_pathinfo_t	*prev = NULL;
3122 	mdi_pathinfo_t	*path;
3123 
3124 	ASSERT(DEVI_BUSY_OWNED(ct->ct_dip));
3125 
3126 	ASSERT(MDI_CLIENT_LOCKED(ct));
3127 	path = ct->ct_path_head;
3128 	while (path != NULL) {
3129 		if (path == pip) {
3130 			break;
3131 		}
3132 		prev = path;
3133 		path = (mdi_pathinfo_t *)MDI_PI(path)->pi_client_link;
3134 	}
3135 
3136 	if (path) {
3137 		ct->ct_path_count--;
3138 		if (prev) {
3139 			MDI_PI(prev)->pi_client_link =
3140 			    MDI_PI(path)->pi_client_link;
3141 		} else {
3142 			ct->ct_path_head =
3143 			    (mdi_pathinfo_t *)MDI_PI(path)->pi_client_link;
3144 		}
3145 		if (ct->ct_path_tail == path) {
3146 			ct->ct_path_tail = prev;
3147 		}
3148 		if (ct->ct_path_last == path) {
3149 			ct->ct_path_last = ct->ct_path_head;
3150 		}
3151 	}
3152 	MDI_PI(pip)->pi_client_link = NULL;
3153 	MDI_PI(pip)->pi_client = NULL;
3154 }
3155 
3156 /*
3157  * i_mdi_pi_state_change():
3158  *		online a mdi_pathinfo node
3159  *
3160  * Return Values:
3161  *		MDI_SUCCESS
3162  *		MDI_FAILURE
3163  */
3164 /*ARGSUSED*/
3165 static int
3166 i_mdi_pi_state_change(mdi_pathinfo_t *pip, mdi_pathinfo_state_t state, int flag)
3167 {
3168 	int		rv = MDI_SUCCESS;
3169 	mdi_vhci_t	*vh;
3170 	mdi_phci_t	*ph;
3171 	mdi_client_t	*ct;
3172 	int		(*f)();
3173 	dev_info_t	*cdip;
3174 
3175 	MDI_PI_LOCK(pip);
3176 
3177 	ph = MDI_PI(pip)->pi_phci;
3178 	ASSERT(ph);
3179 	if (ph == NULL) {
3180 		/*
3181 		 * Invalid pHCI device, fail the request
3182 		 */
3183 		MDI_PI_UNLOCK(pip);
3184 		MDI_DEBUG(1, (CE_WARN, NULL,
3185 		    "!mdi_pi_state_change: invalid phci pip=%p", (void *)pip));
3186 		return (MDI_FAILURE);
3187 	}
3188 
3189 	vh = ph->ph_vhci;
3190 	ASSERT(vh);
3191 	if (vh == NULL) {
3192 		/*
3193 		 * Invalid vHCI device, fail the request
3194 		 */
3195 		MDI_PI_UNLOCK(pip);
3196 		MDI_DEBUG(1, (CE_WARN, NULL,
3197 		    "!mdi_pi_state_change: invalid vhci pip=%p", (void *)pip));
3198 		return (MDI_FAILURE);
3199 	}
3200 
3201 	ct = MDI_PI(pip)->pi_client;
3202 	ASSERT(ct != NULL);
3203 	if (ct == NULL) {
3204 		/*
3205 		 * Invalid client device, fail the request
3206 		 */
3207 		MDI_PI_UNLOCK(pip);
3208 		MDI_DEBUG(1, (CE_WARN, NULL,
3209 		    "!mdi_pi_state_change: invalid client pip=%p",
3210 		    (void *)pip));
3211 		return (MDI_FAILURE);
3212 	}
3213 
3214 	/*
3215 	 * If this path has not been initialized yet, Callback vHCI driver's
3216 	 * pathinfo node initialize entry point
3217 	 */
3218 
3219 	if (MDI_PI_IS_INITING(pip)) {
3220 		MDI_PI_UNLOCK(pip);
3221 		f = vh->vh_ops->vo_pi_init;
3222 		if (f != NULL) {
3223 			rv = (*f)(vh->vh_dip, pip, 0);
3224 			if (rv != MDI_SUCCESS) {
3225 				MDI_DEBUG(1, (CE_WARN, ct->ct_dip,
3226 				    "!vo_pi_init: failed vHCI=0x%p, pip=0x%p",
3227 				    (void *)vh, (void *)pip));
3228 				return (MDI_FAILURE);
3229 			}
3230 		}
3231 		MDI_PI_LOCK(pip);
3232 		MDI_PI_CLEAR_TRANSIENT(pip);
3233 	}
3234 
3235 	/*
3236 	 * Do not allow state transition when pHCI is in offline/suspended
3237 	 * states
3238 	 */
3239 	i_mdi_phci_lock(ph, pip);
3240 	if (MDI_PHCI_IS_READY(ph) == 0) {
3241 		MDI_DEBUG(1, (CE_WARN, ct->ct_dip,
3242 		    "!mdi_pi_state_change: pHCI not ready, pHCI=%p",
3243 		    (void *)ph));
3244 		MDI_PI_UNLOCK(pip);
3245 		i_mdi_phci_unlock(ph);
3246 		return (MDI_BUSY);
3247 	}
3248 	MDI_PHCI_UNSTABLE(ph);
3249 	i_mdi_phci_unlock(ph);
3250 
3251 	/*
3252 	 * Check if mdi_pathinfo state is in transient state.
3253 	 * If yes, offlining is in progress and wait till transient state is
3254 	 * cleared.
3255 	 */
3256 	if (MDI_PI_IS_TRANSIENT(pip)) {
3257 		while (MDI_PI_IS_TRANSIENT(pip)) {
3258 			cv_wait(&MDI_PI(pip)->pi_state_cv,
3259 			    &MDI_PI(pip)->pi_mutex);
3260 		}
3261 	}
3262 
3263 	/*
3264 	 * Grab the client lock in reverse order sequence and release the
3265 	 * mdi_pathinfo mutex.
3266 	 */
3267 	i_mdi_client_lock(ct, pip);
3268 	MDI_PI_UNLOCK(pip);
3269 
3270 	/*
3271 	 * Wait till failover state is cleared
3272 	 */
3273 	while (MDI_CLIENT_IS_FAILOVER_IN_PROGRESS(ct))
3274 		cv_wait(&ct->ct_failover_cv, &ct->ct_mutex);
3275 
3276 	/*
3277 	 * Mark the mdi_pathinfo node state as transient
3278 	 */
3279 	MDI_PI_LOCK(pip);
3280 	switch (state) {
3281 	case MDI_PATHINFO_STATE_ONLINE:
3282 		MDI_PI_SET_ONLINING(pip);
3283 		break;
3284 
3285 	case MDI_PATHINFO_STATE_STANDBY:
3286 		MDI_PI_SET_STANDBYING(pip);
3287 		break;
3288 
3289 	case MDI_PATHINFO_STATE_FAULT:
3290 		/*
3291 		 * Mark the pathinfo state as FAULTED
3292 		 */
3293 		MDI_PI_SET_FAULTING(pip);
3294 		MDI_PI_ERRSTAT(pip, MDI_PI_HARDERR);
3295 		break;
3296 
3297 	case MDI_PATHINFO_STATE_OFFLINE:
3298 		/*
3299 		 * ndi_devi_offline() cannot hold pip or ct locks.
3300 		 */
3301 		MDI_PI_UNLOCK(pip);
3302 		/*
3303 		 * Do not offline if path will become last path and path
3304 		 * is busy for user initiated events.
3305 		 */
3306 		cdip = ct->ct_dip;
3307 		if ((flag & NDI_DEVI_REMOVE) &&
3308 		    (MDI_CLIENT_STATE(ct) == MDI_CLIENT_STATE_DEGRADED)) {
3309 			i_mdi_client_unlock(ct);
3310 			rv = ndi_devi_offline(cdip, 0);
3311 			if (rv != NDI_SUCCESS) {
3312 				/*
3313 				 * Convert to MDI error code
3314 				 */
3315 				switch (rv) {
3316 				case NDI_BUSY:
3317 					rv = MDI_BUSY;
3318 					break;
3319 				default:
3320 					rv = MDI_FAILURE;
3321 					break;
3322 				}
3323 				goto state_change_exit;
3324 			} else {
3325 				i_mdi_client_lock(ct, NULL);
3326 			}
3327 		}
3328 		/*
3329 		 * Mark the mdi_pathinfo node state as transient
3330 		 */
3331 		MDI_PI_LOCK(pip);
3332 		MDI_PI_SET_OFFLINING(pip);
3333 		break;
3334 	}
3335 	MDI_PI_UNLOCK(pip);
3336 	MDI_CLIENT_UNSTABLE(ct);
3337 	i_mdi_client_unlock(ct);
3338 
3339 	f = vh->vh_ops->vo_pi_state_change;
3340 	if (f != NULL)
3341 		rv = (*f)(vh->vh_dip, pip, state, 0, flag);
3342 
3343 	MDI_CLIENT_LOCK(ct);
3344 	MDI_PI_LOCK(pip);
3345 	if (rv == MDI_NOT_SUPPORTED) {
3346 		MDI_CLIENT_SET_DEV_NOT_SUPPORTED(ct);
3347 	}
3348 	if (rv != MDI_SUCCESS) {
3349 		MDI_DEBUG(2, (CE_WARN, ct->ct_dip,
3350 		    "!vo_pi_state_change: failed rv = %x", rv));
3351 	}
3352 	if (MDI_PI_IS_TRANSIENT(pip)) {
3353 		if (rv == MDI_SUCCESS) {
3354 			MDI_PI_CLEAR_TRANSIENT(pip);
3355 		} else {
3356 			MDI_PI(pip)->pi_state = MDI_PI_OLD_STATE(pip);
3357 		}
3358 	}
3359 
3360 	/*
3361 	 * Wake anyone waiting for this mdi_pathinfo node
3362 	 */
3363 	cv_broadcast(&MDI_PI(pip)->pi_state_cv);
3364 	MDI_PI_UNLOCK(pip);
3365 
3366 	/*
3367 	 * Mark the client device as stable
3368 	 */
3369 	MDI_CLIENT_STABLE(ct);
3370 	if (rv == MDI_SUCCESS) {
3371 		if (ct->ct_unstable == 0) {
3372 			cdip = ct->ct_dip;
3373 
3374 			/*
3375 			 * Onlining the mdi_pathinfo node will impact the
3376 			 * client state Update the client and dev_info node
3377 			 * state accordingly
3378 			 */
3379 			rv = NDI_SUCCESS;
3380 			i_mdi_client_update_state(ct);
3381 			switch (MDI_CLIENT_STATE(ct)) {
3382 			case MDI_CLIENT_STATE_OPTIMAL:
3383 			case MDI_CLIENT_STATE_DEGRADED:
3384 				if (cdip && !i_ddi_devi_attached(cdip) &&
3385 				    ((state == MDI_PATHINFO_STATE_ONLINE) ||
3386 				    (state == MDI_PATHINFO_STATE_STANDBY))) {
3387 
3388 					/*
3389 					 * Must do ndi_devi_online() through
3390 					 * hotplug thread for deferred
3391 					 * attach mechanism to work
3392 					 */
3393 					MDI_CLIENT_UNLOCK(ct);
3394 					rv = ndi_devi_online(cdip, 0);
3395 					MDI_CLIENT_LOCK(ct);
3396 					if ((rv != NDI_SUCCESS) &&
3397 					    (MDI_CLIENT_STATE(ct) ==
3398 					    MDI_CLIENT_STATE_DEGRADED)) {
3399 						/*
3400 						 * ndi_devi_online failed.
3401 						 * Reset client flags to
3402 						 * offline.
3403 						 */
3404 						MDI_DEBUG(1, (CE_WARN, cdip,
3405 						    "!ndi_devi_online: failed "
3406 						    " Error: %x", rv));
3407 						MDI_CLIENT_SET_OFFLINE(ct);
3408 					}
3409 					if (rv != NDI_SUCCESS) {
3410 						/* Reset the path state */
3411 						MDI_PI_LOCK(pip);
3412 						MDI_PI(pip)->pi_state =
3413 						    MDI_PI_OLD_STATE(pip);
3414 						MDI_PI_UNLOCK(pip);
3415 					}
3416 				}
3417 				break;
3418 
3419 			case MDI_CLIENT_STATE_FAILED:
3420 				/*
3421 				 * This is the last path case for
3422 				 * non-user initiated events.
3423 				 */
3424 				if (((flag & NDI_DEVI_REMOVE) == 0) &&
3425 				    cdip && (i_ddi_node_state(cdip) >=
3426 				    DS_INITIALIZED)) {
3427 					MDI_CLIENT_UNLOCK(ct);
3428 					rv = ndi_devi_offline(cdip, 0);
3429 					MDI_CLIENT_LOCK(ct);
3430 
3431 					if (rv != NDI_SUCCESS) {
3432 						/*
3433 						 * ndi_devi_offline failed.
3434 						 * Reset client flags to
3435 						 * online as the path could not
3436 						 * be offlined.
3437 						 */
3438 						MDI_DEBUG(1, (CE_WARN, cdip,
3439 						    "!ndi_devi_offline: failed "
3440 						    " Error: %x", rv));
3441 						MDI_CLIENT_SET_ONLINE(ct);
3442 					}
3443 				}
3444 				break;
3445 			}
3446 			/*
3447 			 * Convert to MDI error code
3448 			 */
3449 			switch (rv) {
3450 			case NDI_SUCCESS:
3451 				MDI_CLIENT_SET_REPORT_DEV_NEEDED(ct);
3452 				i_mdi_report_path_state(ct, pip);
3453 				rv = MDI_SUCCESS;
3454 				break;
3455 			case NDI_BUSY:
3456 				rv = MDI_BUSY;
3457 				break;
3458 			default:
3459 				rv = MDI_FAILURE;
3460 				break;
3461 			}
3462 		}
3463 	}
3464 	MDI_CLIENT_UNLOCK(ct);
3465 
3466 state_change_exit:
3467 	/*
3468 	 * Mark the pHCI as stable again.
3469 	 */
3470 	MDI_PHCI_LOCK(ph);
3471 	MDI_PHCI_STABLE(ph);
3472 	MDI_PHCI_UNLOCK(ph);
3473 	return (rv);
3474 }
3475 
3476 /*
3477  * mdi_pi_online():
3478  *		Place the path_info node in the online state.  The path is
3479  *		now available to be selected by mdi_select_path() for
3480  *		transporting I/O requests to client devices.
3481  * Return Values:
3482  *		MDI_SUCCESS
3483  *		MDI_FAILURE
3484  */
3485 int
3486 mdi_pi_online(mdi_pathinfo_t *pip, int flags)
3487 {
3488 	mdi_client_t	*ct = MDI_PI(pip)->pi_client;
3489 	int		client_held = 0;
3490 	int		rv;
3491 
3492 	ASSERT(ct != NULL);
3493 	rv = i_mdi_pi_state_change(pip, MDI_PATHINFO_STATE_ONLINE, flags);
3494 	if (rv != MDI_SUCCESS)
3495 		return (rv);
3496 
3497 	MDI_PI_LOCK(pip);
3498 	if (MDI_PI(pip)->pi_pm_held == 0) {
3499 		MDI_DEBUG(4, (CE_NOTE, ct->ct_dip, "mdi_pi_online "
3500 		    "i_mdi_pm_hold_pip %p\n", (void *)pip));
3501 		i_mdi_pm_hold_pip(pip);
3502 		client_held = 1;
3503 	}
3504 	MDI_PI_UNLOCK(pip);
3505 
3506 	if (client_held) {
3507 		MDI_CLIENT_LOCK(ct);
3508 		if (ct->ct_power_cnt == 0) {
3509 			rv = i_mdi_power_all_phci(ct);
3510 		}
3511 
3512 		MDI_DEBUG(4, (CE_NOTE, ct->ct_dip, "mdi_pi_online "
3513 		    "i_mdi_pm_hold_client %p\n", (void *)ct));
3514 		i_mdi_pm_hold_client(ct, 1);
3515 		MDI_CLIENT_UNLOCK(ct);
3516 	}
3517 
3518 	return (rv);
3519 }
3520 
3521 /*
3522  * mdi_pi_standby():
3523  *		Place the mdi_pathinfo node in standby state
3524  *
3525  * Return Values:
3526  *		MDI_SUCCESS
3527  *		MDI_FAILURE
3528  */
3529 int
3530 mdi_pi_standby(mdi_pathinfo_t *pip, int flags)
3531 {
3532 	return (i_mdi_pi_state_change(pip, MDI_PATHINFO_STATE_STANDBY, flags));
3533 }
3534 
3535 /*
3536  * mdi_pi_fault():
3537  *		Place the mdi_pathinfo node in fault'ed state
3538  * Return Values:
3539  *		MDI_SUCCESS
3540  *		MDI_FAILURE
3541  */
3542 int
3543 mdi_pi_fault(mdi_pathinfo_t *pip, int flags)
3544 {
3545 	return (i_mdi_pi_state_change(pip, MDI_PATHINFO_STATE_FAULT, flags));
3546 }
3547 
3548 /*
3549  * mdi_pi_offline():
3550  *		Offline a mdi_pathinfo node.
3551  * Return Values:
3552  *		MDI_SUCCESS
3553  *		MDI_FAILURE
3554  */
3555 int
3556 mdi_pi_offline(mdi_pathinfo_t *pip, int flags)
3557 {
3558 	int	ret, client_held = 0;
3559 	mdi_client_t	*ct;
3560 
3561 	ret = i_mdi_pi_state_change(pip, MDI_PATHINFO_STATE_OFFLINE, flags);
3562 
3563 	if (ret == MDI_SUCCESS) {
3564 		MDI_PI_LOCK(pip);
3565 		if (MDI_PI(pip)->pi_pm_held) {
3566 			client_held = 1;
3567 		}
3568 		MDI_PI_UNLOCK(pip);
3569 
3570 		if (client_held) {
3571 			ct = MDI_PI(pip)->pi_client;
3572 			MDI_CLIENT_LOCK(ct);
3573 			MDI_DEBUG(4, (CE_NOTE, ct->ct_dip,
3574 			    "mdi_pi_offline i_mdi_pm_rele_client\n"));
3575 			i_mdi_pm_rele_client(ct, 1);
3576 			MDI_CLIENT_UNLOCK(ct);
3577 		}
3578 	}
3579 
3580 	return (ret);
3581 }
3582 
3583 /*
3584  * i_mdi_pi_offline():
3585  *		Offline a mdi_pathinfo node and call the vHCI driver's callback
3586  */
3587 static int
3588 i_mdi_pi_offline(mdi_pathinfo_t *pip, int flags)
3589 {
3590 	dev_info_t	*vdip = NULL;
3591 	mdi_vhci_t	*vh = NULL;
3592 	mdi_client_t	*ct = NULL;
3593 	int		(*f)();
3594 	int		rv;
3595 
3596 	MDI_PI_LOCK(pip);
3597 	ct = MDI_PI(pip)->pi_client;
3598 	ASSERT(ct != NULL);
3599 
3600 	while (MDI_PI(pip)->pi_ref_cnt != 0) {
3601 		/*
3602 		 * Give a chance for pending I/Os to complete.
3603 		 */
3604 		MDI_DEBUG(1, (CE_NOTE, ct->ct_dip, "!i_mdi_pi_offline: "
3605 		    "%d cmds still pending on path: %p\n",
3606 		    MDI_PI(pip)->pi_ref_cnt, (void *)pip));
3607 		if (cv_timedwait(&MDI_PI(pip)->pi_ref_cv,
3608 		    &MDI_PI(pip)->pi_mutex,
3609 		    ddi_get_lbolt() + drv_usectohz(60 * 1000000)) == -1) {
3610 			/*
3611 			 * The timeout time reached without ref_cnt being zero
3612 			 * being signaled.
3613 			 */
3614 			MDI_DEBUG(1, (CE_NOTE, ct->ct_dip, "!i_mdi_pi_offline: "
3615 			    "Timeout reached on path %p without the cond\n",
3616 			    (void *)pip));
3617 			MDI_DEBUG(1, (CE_NOTE, ct->ct_dip, "!i_mdi_pi_offline: "
3618 			    "%d cmds still pending on path: %p\n",
3619 			    MDI_PI(pip)->pi_ref_cnt, (void *)pip));
3620 		}
3621 	}
3622 	vh = ct->ct_vhci;
3623 	vdip = vh->vh_dip;
3624 
3625 	/*
3626 	 * Notify vHCI that has registered this event
3627 	 */
3628 	ASSERT(vh->vh_ops);
3629 	f = vh->vh_ops->vo_pi_state_change;
3630 
3631 	if (f != NULL) {
3632 		MDI_PI_UNLOCK(pip);
3633 		if ((rv = (*f)(vdip, pip, MDI_PATHINFO_STATE_OFFLINE, 0,
3634 		    flags)) != MDI_SUCCESS) {
3635 			MDI_DEBUG(1, (CE_WARN, ct->ct_dip,
3636 			    "!vo_path_offline failed "
3637 			    "vdip %p, pip %p", (void *)vdip, (void *)pip));
3638 		}
3639 		MDI_PI_LOCK(pip);
3640 	}
3641 
3642 	/*
3643 	 * Set the mdi_pathinfo node state and clear the transient condition
3644 	 */
3645 	MDI_PI_SET_OFFLINE(pip);
3646 	cv_broadcast(&MDI_PI(pip)->pi_state_cv);
3647 	MDI_PI_UNLOCK(pip);
3648 
3649 	MDI_CLIENT_LOCK(ct);
3650 	if (rv == MDI_SUCCESS) {
3651 		if (ct->ct_unstable == 0) {
3652 			dev_info_t	*cdip = ct->ct_dip;
3653 
3654 			/*
3655 			 * Onlining the mdi_pathinfo node will impact the
3656 			 * client state Update the client and dev_info node
3657 			 * state accordingly
3658 			 */
3659 			i_mdi_client_update_state(ct);
3660 			rv = NDI_SUCCESS;
3661 			if (MDI_CLIENT_STATE(ct) == MDI_CLIENT_STATE_FAILED) {
3662 				if (cdip &&
3663 				    (i_ddi_node_state(cdip) >=
3664 				    DS_INITIALIZED)) {
3665 					MDI_CLIENT_UNLOCK(ct);
3666 					rv = ndi_devi_offline(cdip, 0);
3667 					MDI_CLIENT_LOCK(ct);
3668 					if (rv != NDI_SUCCESS) {
3669 						/*
3670 						 * ndi_devi_offline failed.
3671 						 * Reset client flags to
3672 						 * online.
3673 						 */
3674 						MDI_DEBUG(4, (CE_WARN, cdip,
3675 						    "!ndi_devi_offline: failed "
3676 						    " Error: %x", rv));
3677 						MDI_CLIENT_SET_ONLINE(ct);
3678 					}
3679 				}
3680 			}
3681 			/*
3682 			 * Convert to MDI error code
3683 			 */
3684 			switch (rv) {
3685 			case NDI_SUCCESS:
3686 				rv = MDI_SUCCESS;
3687 				break;
3688 			case NDI_BUSY:
3689 				rv = MDI_BUSY;
3690 				break;
3691 			default:
3692 				rv = MDI_FAILURE;
3693 				break;
3694 			}
3695 		}
3696 		MDI_CLIENT_SET_REPORT_DEV_NEEDED(ct);
3697 		i_mdi_report_path_state(ct, pip);
3698 	}
3699 
3700 	MDI_CLIENT_UNLOCK(ct);
3701 
3702 	/*
3703 	 * Change in the mdi_pathinfo node state will impact the client state
3704 	 */
3705 	MDI_DEBUG(2, (CE_NOTE, NULL, "!i_mdi_pi_offline ct = %p pip = %p",
3706 	    (void *)ct, (void *)pip));
3707 	return (rv);
3708 }
3709 
3710 
3711 /*
3712  * mdi_pi_get_addr():
3713  *		Get the unit address associated with a mdi_pathinfo node
3714  *
3715  * Return Values:
3716  *		char *
3717  */
3718 char *
3719 mdi_pi_get_addr(mdi_pathinfo_t *pip)
3720 {
3721 	if (pip == NULL)
3722 		return (NULL);
3723 
3724 	return (MDI_PI(pip)->pi_addr);
3725 }
3726 
3727 /*
3728  * mdi_pi_get_client():
3729  *		Get the client devinfo associated with a mdi_pathinfo node
3730  *
3731  * Return Values:
3732  *		Handle to client device dev_info node
3733  */
3734 dev_info_t *
3735 mdi_pi_get_client(mdi_pathinfo_t *pip)
3736 {
3737 	dev_info_t	*dip = NULL;
3738 	if (pip) {
3739 		dip = MDI_PI(pip)->pi_client->ct_dip;
3740 	}
3741 	return (dip);
3742 }
3743 
3744 /*
3745  * mdi_pi_get_phci():
3746  *		Get the pHCI devinfo associated with the mdi_pathinfo node
3747  * Return Values:
3748  *		Handle to dev_info node
3749  */
3750 dev_info_t *
3751 mdi_pi_get_phci(mdi_pathinfo_t *pip)
3752 {
3753 	dev_info_t	*dip = NULL;
3754 	if (pip) {
3755 		dip = MDI_PI(pip)->pi_phci->ph_dip;
3756 	}
3757 	return (dip);
3758 }
3759 
3760 /*
3761  * mdi_pi_get_client_private():
3762  *		Get the client private information associated with the
3763  *		mdi_pathinfo node
3764  */
3765 void *
3766 mdi_pi_get_client_private(mdi_pathinfo_t *pip)
3767 {
3768 	void *cprivate = NULL;
3769 	if (pip) {
3770 		cprivate = MDI_PI(pip)->pi_cprivate;
3771 	}
3772 	return (cprivate);
3773 }
3774 
3775 /*
3776  * mdi_pi_set_client_private():
3777  *		Set the client private information in the mdi_pathinfo node
3778  */
3779 void
3780 mdi_pi_set_client_private(mdi_pathinfo_t *pip, void *priv)
3781 {
3782 	if (pip) {
3783 		MDI_PI(pip)->pi_cprivate = priv;
3784 	}
3785 }
3786 
3787 /*
3788  * mdi_pi_get_phci_private():
3789  *		Get the pHCI private information associated with the
3790  *		mdi_pathinfo node
3791  */
3792 caddr_t
3793 mdi_pi_get_phci_private(mdi_pathinfo_t *pip)
3794 {
3795 	caddr_t	pprivate = NULL;
3796 	if (pip) {
3797 		pprivate = MDI_PI(pip)->pi_pprivate;
3798 	}
3799 	return (pprivate);
3800 }
3801 
3802 /*
3803  * mdi_pi_set_phci_private():
3804  *		Set the pHCI private information in the mdi_pathinfo node
3805  */
3806 void
3807 mdi_pi_set_phci_private(mdi_pathinfo_t *pip, caddr_t priv)
3808 {
3809 	if (pip) {
3810 		MDI_PI(pip)->pi_pprivate = priv;
3811 	}
3812 }
3813 
3814 /*
3815  * mdi_pi_get_state():
3816  *		Get the mdi_pathinfo node state. Transient states are internal
3817  *		and not provided to the users
3818  */
3819 mdi_pathinfo_state_t
3820 mdi_pi_get_state(mdi_pathinfo_t *pip)
3821 {
3822 	mdi_pathinfo_state_t    state = MDI_PATHINFO_STATE_INIT;
3823 
3824 	if (pip) {
3825 		if (MDI_PI_IS_TRANSIENT(pip)) {
3826 			/*
3827 			 * mdi_pathinfo is in state transition.  Return the
3828 			 * last good state.
3829 			 */
3830 			state = MDI_PI_OLD_STATE(pip);
3831 		} else {
3832 			state = MDI_PI_STATE(pip);
3833 		}
3834 	}
3835 	return (state);
3836 }
3837 
3838 /*
3839  * Note that the following function needs to be the new interface for
3840  * mdi_pi_get_state when mpxio gets integrated to ON.
3841  */
3842 int
3843 mdi_pi_get_state2(mdi_pathinfo_t *pip, mdi_pathinfo_state_t *state,
3844 		uint32_t *ext_state)
3845 {
3846 	*state = MDI_PATHINFO_STATE_INIT;
3847 
3848 	if (pip) {
3849 		if (MDI_PI_IS_TRANSIENT(pip)) {
3850 			/*
3851 			 * mdi_pathinfo is in state transition.  Return the
3852 			 * last good state.
3853 			 */
3854 			*state = MDI_PI_OLD_STATE(pip);
3855 			*ext_state = MDI_PI_OLD_EXT_STATE(pip);
3856 		} else {
3857 			*state = MDI_PI_STATE(pip);
3858 			*ext_state = MDI_PI_EXT_STATE(pip);
3859 		}
3860 	}
3861 	return (MDI_SUCCESS);
3862 }
3863 
3864 /*
3865  * mdi_pi_get_preferred:
3866  *	Get the preferred path flag
3867  */
3868 int
3869 mdi_pi_get_preferred(mdi_pathinfo_t *pip)
3870 {
3871 	if (pip) {
3872 		return (MDI_PI(pip)->pi_preferred);
3873 	}
3874 	return (0);
3875 }
3876 
3877 /*
3878  * mdi_pi_set_preferred:
3879  *	Set the preferred path flag
3880  */
3881 void
3882 mdi_pi_set_preferred(mdi_pathinfo_t *pip, int preferred)
3883 {
3884 	if (pip) {
3885 		MDI_PI(pip)->pi_preferred = preferred;
3886 	}
3887 }
3888 
3889 /*
3890  * mdi_pi_set_state():
3891  *		Set the mdi_pathinfo node state
3892  */
3893 void
3894 mdi_pi_set_state(mdi_pathinfo_t *pip, mdi_pathinfo_state_t state)
3895 {
3896 	uint32_t	ext_state;
3897 
3898 	if (pip) {
3899 		ext_state = MDI_PI(pip)->pi_state & MDI_PATHINFO_EXT_STATE_MASK;
3900 		MDI_PI(pip)->pi_state = state;
3901 		MDI_PI(pip)->pi_state |= ext_state;
3902 	}
3903 }
3904 
3905 /*
3906  * Property functions:
3907  */
3908 int
3909 i_map_nvlist_error_to_mdi(int val)
3910 {
3911 	int rv;
3912 
3913 	switch (val) {
3914 	case 0:
3915 		rv = DDI_PROP_SUCCESS;
3916 		break;
3917 	case EINVAL:
3918 	case ENOTSUP:
3919 		rv = DDI_PROP_INVAL_ARG;
3920 		break;
3921 	case ENOMEM:
3922 		rv = DDI_PROP_NO_MEMORY;
3923 		break;
3924 	default:
3925 		rv = DDI_PROP_NOT_FOUND;
3926 		break;
3927 	}
3928 	return (rv);
3929 }
3930 
3931 /*
3932  * mdi_pi_get_next_prop():
3933  * 		Property walk function.  The caller should hold mdi_pi_lock()
3934  *		and release by calling mdi_pi_unlock() at the end of walk to
3935  *		get a consistent value.
3936  */
3937 nvpair_t *
3938 mdi_pi_get_next_prop(mdi_pathinfo_t *pip, nvpair_t *prev)
3939 {
3940 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
3941 		return (NULL);
3942 	}
3943 	ASSERT(MDI_PI_LOCKED(pip));
3944 	return (nvlist_next_nvpair(MDI_PI(pip)->pi_prop, prev));
3945 }
3946 
3947 /*
3948  * mdi_prop_remove():
3949  * 		Remove the named property from the named list.
3950  */
3951 int
3952 mdi_prop_remove(mdi_pathinfo_t *pip, char *name)
3953 {
3954 	if (pip == NULL) {
3955 		return (DDI_PROP_NOT_FOUND);
3956 	}
3957 	ASSERT(!MDI_PI_LOCKED(pip));
3958 	MDI_PI_LOCK(pip);
3959 	if (MDI_PI(pip)->pi_prop == NULL) {
3960 		MDI_PI_UNLOCK(pip);
3961 		return (DDI_PROP_NOT_FOUND);
3962 	}
3963 	if (name) {
3964 		(void) nvlist_remove_all(MDI_PI(pip)->pi_prop, name);
3965 	} else {
3966 		char		nvp_name[MAXNAMELEN];
3967 		nvpair_t	*nvp;
3968 		nvp = nvlist_next_nvpair(MDI_PI(pip)->pi_prop, NULL);
3969 		while (nvp) {
3970 			nvpair_t	*next;
3971 			next = nvlist_next_nvpair(MDI_PI(pip)->pi_prop, nvp);
3972 			(void) snprintf(nvp_name, MAXNAMELEN, "%s",
3973 			    nvpair_name(nvp));
3974 			(void) nvlist_remove_all(MDI_PI(pip)->pi_prop,
3975 			    nvp_name);
3976 			nvp = next;
3977 		}
3978 	}
3979 	MDI_PI_UNLOCK(pip);
3980 	return (DDI_PROP_SUCCESS);
3981 }
3982 
3983 /*
3984  * mdi_prop_size():
3985  * 		Get buffer size needed to pack the property data.
3986  * 		Caller should hold the mdi_pathinfo_t lock to get a consistent
3987  *		buffer size.
3988  */
3989 int
3990 mdi_prop_size(mdi_pathinfo_t *pip, size_t *buflenp)
3991 {
3992 	int	rv;
3993 	size_t	bufsize;
3994 
3995 	*buflenp = 0;
3996 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
3997 		return (DDI_PROP_NOT_FOUND);
3998 	}
3999 	ASSERT(MDI_PI_LOCKED(pip));
4000 	rv = nvlist_size(MDI_PI(pip)->pi_prop,
4001 	    &bufsize, NV_ENCODE_NATIVE);
4002 	*buflenp = bufsize;
4003 	return (i_map_nvlist_error_to_mdi(rv));
4004 }
4005 
4006 /*
4007  * mdi_prop_pack():
4008  * 		pack the property list.  The caller should hold the
4009  *		mdi_pathinfo_t node to get a consistent data
4010  */
4011 int
4012 mdi_prop_pack(mdi_pathinfo_t *pip, char **bufp, uint_t buflen)
4013 {
4014 	int	rv;
4015 	size_t	bufsize;
4016 
4017 	if ((pip == NULL) || MDI_PI(pip)->pi_prop == NULL) {
4018 		return (DDI_PROP_NOT_FOUND);
4019 	}
4020 
4021 	ASSERT(MDI_PI_LOCKED(pip));
4022 
4023 	bufsize = buflen;
4024 	rv = nvlist_pack(MDI_PI(pip)->pi_prop, bufp, (size_t *)&bufsize,
4025 	    NV_ENCODE_NATIVE, KM_SLEEP);
4026 
4027 	return (i_map_nvlist_error_to_mdi(rv));
4028 }
4029 
4030 /*
4031  * mdi_prop_update_byte():
4032  *		Create/Update a byte property
4033  */
4034 int
4035 mdi_prop_update_byte(mdi_pathinfo_t *pip, char *name, uchar_t data)
4036 {
4037 	int rv;
4038 
4039 	if (pip == NULL) {
4040 		return (DDI_PROP_INVAL_ARG);
4041 	}
4042 	ASSERT(!MDI_PI_LOCKED(pip));
4043 	MDI_PI_LOCK(pip);
4044 	if (MDI_PI(pip)->pi_prop == NULL) {
4045 		MDI_PI_UNLOCK(pip);
4046 		return (DDI_PROP_NOT_FOUND);
4047 	}
4048 	rv = nvlist_add_byte(MDI_PI(pip)->pi_prop, name, data);
4049 	MDI_PI_UNLOCK(pip);
4050 	return (i_map_nvlist_error_to_mdi(rv));
4051 }
4052 
4053 /*
4054  * mdi_prop_update_byte_array():
4055  *		Create/Update a byte array property
4056  */
4057 int
4058 mdi_prop_update_byte_array(mdi_pathinfo_t *pip, char *name, uchar_t *data,
4059     uint_t nelements)
4060 {
4061 	int rv;
4062 
4063 	if (pip == NULL) {
4064 		return (DDI_PROP_INVAL_ARG);
4065 	}
4066 	ASSERT(!MDI_PI_LOCKED(pip));
4067 	MDI_PI_LOCK(pip);
4068 	if (MDI_PI(pip)->pi_prop == NULL) {
4069 		MDI_PI_UNLOCK(pip);
4070 		return (DDI_PROP_NOT_FOUND);
4071 	}
4072 	rv = nvlist_add_byte_array(MDI_PI(pip)->pi_prop, name, data, nelements);
4073 	MDI_PI_UNLOCK(pip);
4074 	return (i_map_nvlist_error_to_mdi(rv));
4075 }
4076 
4077 /*
4078  * mdi_prop_update_int():
4079  *		Create/Update a 32 bit integer property
4080  */
4081 int
4082 mdi_prop_update_int(mdi_pathinfo_t *pip, char *name, int data)
4083 {
4084 	int rv;
4085 
4086 	if (pip == NULL) {
4087 		return (DDI_PROP_INVAL_ARG);
4088 	}
4089 	ASSERT(!MDI_PI_LOCKED(pip));
4090 	MDI_PI_LOCK(pip);
4091 	if (MDI_PI(pip)->pi_prop == NULL) {
4092 		MDI_PI_UNLOCK(pip);
4093 		return (DDI_PROP_NOT_FOUND);
4094 	}
4095 	rv = nvlist_add_int32(MDI_PI(pip)->pi_prop, name, (int32_t)data);
4096 	MDI_PI_UNLOCK(pip);
4097 	return (i_map_nvlist_error_to_mdi(rv));
4098 }
4099 
4100 /*
4101  * mdi_prop_update_int64():
4102  *		Create/Update a 64 bit integer property
4103  */
4104 int
4105 mdi_prop_update_int64(mdi_pathinfo_t *pip, char *name, int64_t data)
4106 {
4107 	int rv;
4108 
4109 	if (pip == NULL) {
4110 		return (DDI_PROP_INVAL_ARG);
4111 	}
4112 	ASSERT(!MDI_PI_LOCKED(pip));
4113 	MDI_PI_LOCK(pip);
4114 	if (MDI_PI(pip)->pi_prop == NULL) {
4115 		MDI_PI_UNLOCK(pip);
4116 		return (DDI_PROP_NOT_FOUND);
4117 	}
4118 	rv = nvlist_add_int64(MDI_PI(pip)->pi_prop, name, data);
4119 	MDI_PI_UNLOCK(pip);
4120 	return (i_map_nvlist_error_to_mdi(rv));
4121 }
4122 
4123 /*
4124  * mdi_prop_update_int_array():
4125  *		Create/Update a int array property
4126  */
4127 int
4128 mdi_prop_update_int_array(mdi_pathinfo_t *pip, char *name, int *data,
4129 	    uint_t nelements)
4130 {
4131 	int rv;
4132 
4133 	if (pip == NULL) {
4134 		return (DDI_PROP_INVAL_ARG);
4135 	}
4136 	ASSERT(!MDI_PI_LOCKED(pip));
4137 	MDI_PI_LOCK(pip);
4138 	if (MDI_PI(pip)->pi_prop == NULL) {
4139 		MDI_PI_UNLOCK(pip);
4140 		return (DDI_PROP_NOT_FOUND);
4141 	}
4142 	rv = nvlist_add_int32_array(MDI_PI(pip)->pi_prop, name, (int32_t *)data,
4143 	    nelements);
4144 	MDI_PI_UNLOCK(pip);
4145 	return (i_map_nvlist_error_to_mdi(rv));
4146 }
4147 
4148 /*
4149  * mdi_prop_update_string():
4150  *		Create/Update a string property
4151  */
4152 int
4153 mdi_prop_update_string(mdi_pathinfo_t *pip, char *name, char *data)
4154 {
4155 	int rv;
4156 
4157 	if (pip == NULL) {
4158 		return (DDI_PROP_INVAL_ARG);
4159 	}
4160 	ASSERT(!MDI_PI_LOCKED(pip));
4161 	MDI_PI_LOCK(pip);
4162 	if (MDI_PI(pip)->pi_prop == NULL) {
4163 		MDI_PI_UNLOCK(pip);
4164 		return (DDI_PROP_NOT_FOUND);
4165 	}
4166 	rv = nvlist_add_string(MDI_PI(pip)->pi_prop, name, data);
4167 	MDI_PI_UNLOCK(pip);
4168 	return (i_map_nvlist_error_to_mdi(rv));
4169 }
4170 
4171 /*
4172  * mdi_prop_update_string_array():
4173  *		Create/Update a string array property
4174  */
4175 int
4176 mdi_prop_update_string_array(mdi_pathinfo_t *pip, char *name, char **data,
4177     uint_t nelements)
4178 {
4179 	int rv;
4180 
4181 	if (pip == NULL) {
4182 		return (DDI_PROP_INVAL_ARG);
4183 	}
4184 	ASSERT(!MDI_PI_LOCKED(pip));
4185 	MDI_PI_LOCK(pip);
4186 	if (MDI_PI(pip)->pi_prop == NULL) {
4187 		MDI_PI_UNLOCK(pip);
4188 		return (DDI_PROP_NOT_FOUND);
4189 	}
4190 	rv = nvlist_add_string_array(MDI_PI(pip)->pi_prop, name, data,
4191 	    nelements);
4192 	MDI_PI_UNLOCK(pip);
4193 	return (i_map_nvlist_error_to_mdi(rv));
4194 }
4195 
4196 /*
4197  * mdi_prop_lookup_byte():
4198  * 		Look for byte property identified by name.  The data returned
4199  *		is the actual property and valid as long as mdi_pathinfo_t node
4200  *		is alive.
4201  */
4202 int
4203 mdi_prop_lookup_byte(mdi_pathinfo_t *pip, char *name, uchar_t *data)
4204 {
4205 	int rv;
4206 
4207 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4208 		return (DDI_PROP_NOT_FOUND);
4209 	}
4210 	rv = nvlist_lookup_byte(MDI_PI(pip)->pi_prop, name, data);
4211 	return (i_map_nvlist_error_to_mdi(rv));
4212 }
4213 
4214 
4215 /*
4216  * mdi_prop_lookup_byte_array():
4217  * 		Look for byte array property identified by name.  The data
4218  *		returned is the actual property and valid as long as
4219  *		mdi_pathinfo_t node is alive.
4220  */
4221 int
4222 mdi_prop_lookup_byte_array(mdi_pathinfo_t *pip, char *name, uchar_t **data,
4223     uint_t *nelements)
4224 {
4225 	int rv;
4226 
4227 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4228 		return (DDI_PROP_NOT_FOUND);
4229 	}
4230 	rv = nvlist_lookup_byte_array(MDI_PI(pip)->pi_prop, name, data,
4231 	    nelements);
4232 	return (i_map_nvlist_error_to_mdi(rv));
4233 }
4234 
4235 /*
4236  * mdi_prop_lookup_int():
4237  * 		Look for int property identified by name.  The data returned
4238  *		is the actual property and valid as long as mdi_pathinfo_t
4239  *		node is alive.
4240  */
4241 int
4242 mdi_prop_lookup_int(mdi_pathinfo_t *pip, char *name, int *data)
4243 {
4244 	int rv;
4245 
4246 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4247 		return (DDI_PROP_NOT_FOUND);
4248 	}
4249 	rv = nvlist_lookup_int32(MDI_PI(pip)->pi_prop, name, (int32_t *)data);
4250 	return (i_map_nvlist_error_to_mdi(rv));
4251 }
4252 
4253 /*
4254  * mdi_prop_lookup_int64():
4255  * 		Look for int64 property identified by name.  The data returned
4256  *		is the actual property and valid as long as mdi_pathinfo_t node
4257  *		is alive.
4258  */
4259 int
4260 mdi_prop_lookup_int64(mdi_pathinfo_t *pip, char *name, int64_t *data)
4261 {
4262 	int rv;
4263 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4264 		return (DDI_PROP_NOT_FOUND);
4265 	}
4266 	rv = nvlist_lookup_int64(MDI_PI(pip)->pi_prop, name, data);
4267 	return (i_map_nvlist_error_to_mdi(rv));
4268 }
4269 
4270 /*
4271  * mdi_prop_lookup_int_array():
4272  * 		Look for int array property identified by name.  The data
4273  *		returned is the actual property and valid as long as
4274  *		mdi_pathinfo_t node is alive.
4275  */
4276 int
4277 mdi_prop_lookup_int_array(mdi_pathinfo_t *pip, char *name, int **data,
4278     uint_t *nelements)
4279 {
4280 	int rv;
4281 
4282 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4283 		return (DDI_PROP_NOT_FOUND);
4284 	}
4285 	rv = nvlist_lookup_int32_array(MDI_PI(pip)->pi_prop, name,
4286 	    (int32_t **)data, nelements);
4287 	return (i_map_nvlist_error_to_mdi(rv));
4288 }
4289 
4290 /*
4291  * mdi_prop_lookup_string():
4292  * 		Look for string property identified by name.  The data
4293  *		returned is the actual property and valid as long as
4294  *		mdi_pathinfo_t node is alive.
4295  */
4296 int
4297 mdi_prop_lookup_string(mdi_pathinfo_t *pip, char *name, char **data)
4298 {
4299 	int rv;
4300 
4301 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4302 		return (DDI_PROP_NOT_FOUND);
4303 	}
4304 	rv = nvlist_lookup_string(MDI_PI(pip)->pi_prop, name, data);
4305 	return (i_map_nvlist_error_to_mdi(rv));
4306 }
4307 
4308 /*
4309  * mdi_prop_lookup_string_array():
4310  * 		Look for string array property identified by name.  The data
4311  *		returned is the actual property and valid as long as
4312  *		mdi_pathinfo_t node is alive.
4313  */
4314 int
4315 mdi_prop_lookup_string_array(mdi_pathinfo_t *pip, char *name, char ***data,
4316     uint_t *nelements)
4317 {
4318 	int rv;
4319 
4320 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4321 		return (DDI_PROP_NOT_FOUND);
4322 	}
4323 	rv = nvlist_lookup_string_array(MDI_PI(pip)->pi_prop, name, data,
4324 	    nelements);
4325 	return (i_map_nvlist_error_to_mdi(rv));
4326 }
4327 
4328 /*
4329  * mdi_prop_free():
4330  * 		Symmetrical function to ddi_prop_free(). nvlist_lookup_xx()
4331  *		functions return the pointer to actual property data and not a
4332  *		copy of it.  So the data returned is valid as long as
4333  *		mdi_pathinfo_t node is valid.
4334  */
4335 /*ARGSUSED*/
4336 int
4337 mdi_prop_free(void *data)
4338 {
4339 	return (DDI_PROP_SUCCESS);
4340 }
4341 
4342 /*ARGSUSED*/
4343 static void
4344 i_mdi_report_path_state(mdi_client_t *ct, mdi_pathinfo_t *pip)
4345 {
4346 	char		*phci_path, *ct_path;
4347 	char		*ct_status;
4348 	char		*status;
4349 	dev_info_t	*dip = ct->ct_dip;
4350 	char		lb_buf[64];
4351 
4352 	ASSERT(MDI_CLIENT_LOCKED(ct));
4353 	if ((dip == NULL) || (ddi_get_instance(dip) == -1) ||
4354 	    (MDI_CLIENT_IS_REPORT_DEV_NEEDED(ct) == 0)) {
4355 		return;
4356 	}
4357 	if (MDI_CLIENT_STATE(ct) == MDI_CLIENT_STATE_OPTIMAL) {
4358 		ct_status = "optimal";
4359 	} else if (MDI_CLIENT_STATE(ct) == MDI_CLIENT_STATE_DEGRADED) {
4360 		ct_status = "degraded";
4361 	} else if (MDI_CLIENT_STATE(ct) == MDI_CLIENT_STATE_FAILED) {
4362 		ct_status = "failed";
4363 	} else {
4364 		ct_status = "unknown";
4365 	}
4366 
4367 	if (MDI_PI_IS_OFFLINE(pip)) {
4368 		status = "offline";
4369 	} else if (MDI_PI_IS_ONLINE(pip)) {
4370 		status = "online";
4371 	} else if (MDI_PI_IS_STANDBY(pip)) {
4372 		status = "standby";
4373 	} else if (MDI_PI_IS_FAULT(pip)) {
4374 		status = "faulted";
4375 	} else {
4376 		status = "unknown";
4377 	}
4378 
4379 	if (ct->ct_lb == LOAD_BALANCE_LBA) {
4380 		(void) snprintf(lb_buf, sizeof (lb_buf),
4381 		    "%s, region-size: %d", mdi_load_balance_lba,
4382 			ct->ct_lb_args->region_size);
4383 	} else if (ct->ct_lb == LOAD_BALANCE_NONE) {
4384 		(void) snprintf(lb_buf, sizeof (lb_buf),
4385 		    "%s", mdi_load_balance_none);
4386 	} else {
4387 		(void) snprintf(lb_buf, sizeof (lb_buf), "%s",
4388 		    mdi_load_balance_rr);
4389 	}
4390 
4391 	if (dip) {
4392 		ct_path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4393 		phci_path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4394 		cmn_err(CE_CONT, "?%s (%s%d) multipath status: %s, "
4395 		    "path %s (%s%d) to target address: %s is %s"
4396 		    " Load balancing: %s\n",
4397 		    ddi_pathname(dip, ct_path), ddi_driver_name(dip),
4398 		    ddi_get_instance(dip), ct_status,
4399 		    ddi_pathname(MDI_PI(pip)->pi_phci->ph_dip, phci_path),
4400 		    ddi_driver_name(MDI_PI(pip)->pi_phci->ph_dip),
4401 		    ddi_get_instance(MDI_PI(pip)->pi_phci->ph_dip),
4402 		    MDI_PI(pip)->pi_addr, status, lb_buf);
4403 		kmem_free(phci_path, MAXPATHLEN);
4404 		kmem_free(ct_path, MAXPATHLEN);
4405 		MDI_CLIENT_CLEAR_REPORT_DEV_NEEDED(ct);
4406 	}
4407 }
4408 
4409 #ifdef	DEBUG
4410 /*
4411  * i_mdi_log():
4412  *		Utility function for error message management
4413  *
4414  */
4415 /*PRINTFLIKE3*/
4416 static void
4417 i_mdi_log(int level, dev_info_t *dip, const char *fmt, ...)
4418 {
4419 	char		name[MAXNAMELEN];
4420 	char		buf[MAXNAMELEN];
4421 	char		*bp;
4422 	va_list		ap;
4423 	int		log_only = 0;
4424 	int		boot_only = 0;
4425 	int		console_only = 0;
4426 
4427 	if (dip) {
4428 		(void) snprintf(name, MAXNAMELEN, "%s%d: ",
4429 		    ddi_node_name(dip), ddi_get_instance(dip));
4430 	} else {
4431 		name[0] = 0;
4432 	}
4433 
4434 	va_start(ap, fmt);
4435 	(void) vsnprintf(buf, MAXNAMELEN, fmt, ap);
4436 	va_end(ap);
4437 
4438 	switch (buf[0]) {
4439 	case '!':
4440 		bp = &buf[1];
4441 		log_only = 1;
4442 		break;
4443 	case '?':
4444 		bp = &buf[1];
4445 		boot_only = 1;
4446 		break;
4447 	case '^':
4448 		bp = &buf[1];
4449 		console_only = 1;
4450 		break;
4451 	default:
4452 		bp = buf;
4453 		break;
4454 	}
4455 	if (mdi_debug_logonly) {
4456 		log_only = 1;
4457 		boot_only = 0;
4458 		console_only = 0;
4459 	}
4460 
4461 	switch (level) {
4462 	case CE_NOTE:
4463 		level = CE_CONT;
4464 		/* FALLTHROUGH */
4465 	case CE_CONT:
4466 	case CE_WARN:
4467 	case CE_PANIC:
4468 		if (boot_only) {
4469 			cmn_err(level, "?mdi: %s%s", name, bp);
4470 		} else if (console_only) {
4471 			cmn_err(level, "^mdi: %s%s", name, bp);
4472 		} else if (log_only) {
4473 			cmn_err(level, "!mdi: %s%s", name, bp);
4474 		} else {
4475 			cmn_err(level, "mdi: %s%s", name, bp);
4476 		}
4477 		break;
4478 	default:
4479 		cmn_err(level, "mdi: %s%s", name, bp);
4480 		break;
4481 	}
4482 }
4483 #endif	/* DEBUG */
4484 
4485 void
4486 i_mdi_client_online(dev_info_t *ct_dip)
4487 {
4488 	mdi_client_t	*ct;
4489 
4490 	/*
4491 	 * Client online notification. Mark client state as online
4492 	 * restore our binding with dev_info node
4493 	 */
4494 	ct = i_devi_get_client(ct_dip);
4495 	ASSERT(ct != NULL);
4496 	MDI_CLIENT_LOCK(ct);
4497 	MDI_CLIENT_SET_ONLINE(ct);
4498 	/* catch for any memory leaks */
4499 	ASSERT((ct->ct_dip == NULL) || (ct->ct_dip == ct_dip));
4500 	ct->ct_dip = ct_dip;
4501 
4502 	if (ct->ct_power_cnt == 0)
4503 		(void) i_mdi_power_all_phci(ct);
4504 
4505 	MDI_DEBUG(4, (CE_NOTE, ct_dip, "i_mdi_client_online "
4506 	    "i_mdi_pm_hold_client %p\n", (void *)ct));
4507 	i_mdi_pm_hold_client(ct, 1);
4508 
4509 	MDI_CLIENT_UNLOCK(ct);
4510 }
4511 
4512 void
4513 i_mdi_phci_online(dev_info_t *ph_dip)
4514 {
4515 	mdi_phci_t	*ph;
4516 
4517 	/* pHCI online notification. Mark state accordingly */
4518 	ph = i_devi_get_phci(ph_dip);
4519 	ASSERT(ph != NULL);
4520 	MDI_PHCI_LOCK(ph);
4521 	MDI_PHCI_SET_ONLINE(ph);
4522 	MDI_PHCI_UNLOCK(ph);
4523 }
4524 
4525 /*
4526  * mdi_devi_online():
4527  * 		Online notification from NDI framework on pHCI/client
4528  *		device online.
4529  * Return Values:
4530  *		NDI_SUCCESS
4531  *		MDI_FAILURE
4532  */
4533 /*ARGSUSED*/
4534 int
4535 mdi_devi_online(dev_info_t *dip, uint_t flags)
4536 {
4537 	if (MDI_PHCI(dip)) {
4538 		i_mdi_phci_online(dip);
4539 	}
4540 
4541 	if (MDI_CLIENT(dip)) {
4542 		i_mdi_client_online(dip);
4543 	}
4544 	return (NDI_SUCCESS);
4545 }
4546 
4547 /*
4548  * mdi_devi_offline():
4549  * 		Offline notification from NDI framework on pHCI/Client device
4550  *		offline.
4551  *
4552  * Return Values:
4553  *		NDI_SUCCESS
4554  *		NDI_FAILURE
4555  */
4556 /*ARGSUSED*/
4557 int
4558 mdi_devi_offline(dev_info_t *dip, uint_t flags)
4559 {
4560 	int		rv = NDI_SUCCESS;
4561 
4562 	if (MDI_CLIENT(dip)) {
4563 		rv = i_mdi_client_offline(dip, flags);
4564 		if (rv != NDI_SUCCESS)
4565 			return (rv);
4566 	}
4567 
4568 	if (MDI_PHCI(dip)) {
4569 		rv = i_mdi_phci_offline(dip, flags);
4570 
4571 		if ((rv != NDI_SUCCESS) && MDI_CLIENT(dip)) {
4572 			/* set client back online */
4573 			i_mdi_client_online(dip);
4574 		}
4575 	}
4576 
4577 	return (rv);
4578 }
4579 
4580 /*ARGSUSED*/
4581 static int
4582 i_mdi_phci_offline(dev_info_t *dip, uint_t flags)
4583 {
4584 	int		rv = NDI_SUCCESS;
4585 	mdi_phci_t	*ph;
4586 	mdi_client_t	*ct;
4587 	mdi_pathinfo_t	*pip;
4588 	mdi_pathinfo_t	*next;
4589 	mdi_pathinfo_t	*failed_pip = NULL;
4590 	dev_info_t	*cdip;
4591 
4592 	/*
4593 	 * pHCI component offline notification
4594 	 * Make sure that this pHCI instance is free to be offlined.
4595 	 * If it is OK to proceed, Offline and remove all the child
4596 	 * mdi_pathinfo nodes.  This process automatically offlines
4597 	 * corresponding client devices, for which this pHCI provides
4598 	 * critical services.
4599 	 */
4600 	ph = i_devi_get_phci(dip);
4601 	MDI_DEBUG(2, (CE_NOTE, dip, "!mdi_phci_offline called %p %p\n",
4602 	    (void *)dip, (void *)ph));
4603 	if (ph == NULL) {
4604 		return (rv);
4605 	}
4606 
4607 	MDI_PHCI_LOCK(ph);
4608 
4609 	if (MDI_PHCI_IS_OFFLINE(ph)) {
4610 		MDI_DEBUG(1, (CE_WARN, dip, "!pHCI %p already offlined",
4611 		    (void *)ph));
4612 		MDI_PHCI_UNLOCK(ph);
4613 		return (NDI_SUCCESS);
4614 	}
4615 
4616 	/*
4617 	 * Check to see if the pHCI can be offlined
4618 	 */
4619 	if (ph->ph_unstable) {
4620 		MDI_DEBUG(1, (CE_WARN, dip,
4621 		    "!One or more target devices are in transient "
4622 		    "state. This device can not be removed at "
4623 		    "this moment. Please try again later."));
4624 		MDI_PHCI_UNLOCK(ph);
4625 		return (NDI_BUSY);
4626 	}
4627 
4628 	pip = ph->ph_path_head;
4629 	while (pip != NULL) {
4630 		MDI_PI_LOCK(pip);
4631 		next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
4632 
4633 		/*
4634 		 * The mdi_pathinfo state is OK. Check the client state.
4635 		 * If failover in progress fail the pHCI from offlining
4636 		 */
4637 		ct = MDI_PI(pip)->pi_client;
4638 		i_mdi_client_lock(ct, pip);
4639 		if ((MDI_CLIENT_IS_FAILOVER_IN_PROGRESS(ct)) ||
4640 		    (ct->ct_unstable)) {
4641 			/*
4642 			 * Failover is in progress, Fail the DR
4643 			 */
4644 			MDI_DEBUG(1, (CE_WARN, dip,
4645 			    "!pHCI device (%s%d) is Busy. %s",
4646 			    ddi_driver_name(dip), ddi_get_instance(dip),
4647 			    "This device can not be removed at "
4648 			    "this moment. Please try again later."));
4649 			MDI_PI_UNLOCK(pip);
4650 			i_mdi_client_unlock(ct);
4651 			MDI_PHCI_UNLOCK(ph);
4652 			return (NDI_BUSY);
4653 		}
4654 		MDI_PI_UNLOCK(pip);
4655 
4656 		/*
4657 		 * Check to see of we are removing the last path of this
4658 		 * client device...
4659 		 */
4660 		cdip = ct->ct_dip;
4661 		if (cdip && (i_ddi_node_state(cdip) >= DS_INITIALIZED) &&
4662 		    (i_mdi_client_compute_state(ct, ph) ==
4663 		    MDI_CLIENT_STATE_FAILED)) {
4664 			i_mdi_client_unlock(ct);
4665 			MDI_PHCI_UNLOCK(ph);
4666 			if (ndi_devi_offline(cdip, 0) != NDI_SUCCESS) {
4667 				/*
4668 				 * ndi_devi_offline() failed.
4669 				 * This pHCI provides the critical path
4670 				 * to one or more client devices.
4671 				 * Return busy.
4672 				 */
4673 				MDI_PHCI_LOCK(ph);
4674 				MDI_DEBUG(1, (CE_WARN, dip,
4675 				    "!pHCI device (%s%d) is Busy. %s",
4676 				    ddi_driver_name(dip), ddi_get_instance(dip),
4677 				    "This device can not be removed at "
4678 				    "this moment. Please try again later."));
4679 				failed_pip = pip;
4680 				break;
4681 			} else {
4682 				MDI_PHCI_LOCK(ph);
4683 				pip = next;
4684 			}
4685 		} else {
4686 			i_mdi_client_unlock(ct);
4687 			pip = next;
4688 		}
4689 	}
4690 
4691 	if (failed_pip) {
4692 		pip = ph->ph_path_head;
4693 		while (pip != failed_pip) {
4694 			MDI_PI_LOCK(pip);
4695 			next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
4696 			ct = MDI_PI(pip)->pi_client;
4697 			i_mdi_client_lock(ct, pip);
4698 			cdip = ct->ct_dip;
4699 			switch (MDI_CLIENT_STATE(ct)) {
4700 			case MDI_CLIENT_STATE_OPTIMAL:
4701 			case MDI_CLIENT_STATE_DEGRADED:
4702 				if (cdip) {
4703 					MDI_PI_UNLOCK(pip);
4704 					i_mdi_client_unlock(ct);
4705 					MDI_PHCI_UNLOCK(ph);
4706 					(void) ndi_devi_online(cdip, 0);
4707 					MDI_PHCI_LOCK(ph);
4708 					pip = next;
4709 					continue;
4710 				}
4711 				break;
4712 
4713 			case MDI_CLIENT_STATE_FAILED:
4714 				if (cdip) {
4715 					MDI_PI_UNLOCK(pip);
4716 					i_mdi_client_unlock(ct);
4717 					MDI_PHCI_UNLOCK(ph);
4718 					(void) ndi_devi_offline(cdip, 0);
4719 					MDI_PHCI_LOCK(ph);
4720 					pip = next;
4721 					continue;
4722 				}
4723 				break;
4724 			}
4725 			MDI_PI_UNLOCK(pip);
4726 			i_mdi_client_unlock(ct);
4727 			pip = next;
4728 		}
4729 		MDI_PHCI_UNLOCK(ph);
4730 		return (NDI_BUSY);
4731 	}
4732 
4733 	/*
4734 	 * Mark the pHCI as offline
4735 	 */
4736 	MDI_PHCI_SET_OFFLINE(ph);
4737 
4738 	/*
4739 	 * Mark the child mdi_pathinfo nodes as transient
4740 	 */
4741 	pip = ph->ph_path_head;
4742 	while (pip != NULL) {
4743 		MDI_PI_LOCK(pip);
4744 		next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
4745 		MDI_PI_SET_OFFLINING(pip);
4746 		MDI_PI_UNLOCK(pip);
4747 		pip = next;
4748 	}
4749 	MDI_PHCI_UNLOCK(ph);
4750 	/*
4751 	 * Give a chance for any pending commands to execute
4752 	 */
4753 	delay(1);
4754 	MDI_PHCI_LOCK(ph);
4755 	pip = ph->ph_path_head;
4756 	while (pip != NULL) {
4757 		next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
4758 		(void) i_mdi_pi_offline(pip, flags);
4759 		MDI_PI_LOCK(pip);
4760 		ct = MDI_PI(pip)->pi_client;
4761 		if (!MDI_PI_IS_OFFLINE(pip)) {
4762 			MDI_DEBUG(1, (CE_WARN, dip,
4763 			    "!pHCI device (%s%d) is Busy. %s",
4764 			    ddi_driver_name(dip), ddi_get_instance(dip),
4765 			    "This device can not be removed at "
4766 			    "this moment. Please try again later."));
4767 			MDI_PI_UNLOCK(pip);
4768 			MDI_PHCI_SET_ONLINE(ph);
4769 			MDI_PHCI_UNLOCK(ph);
4770 			return (NDI_BUSY);
4771 		}
4772 		MDI_PI_UNLOCK(pip);
4773 		pip = next;
4774 	}
4775 	MDI_PHCI_UNLOCK(ph);
4776 
4777 	return (rv);
4778 }
4779 
4780 /*ARGSUSED*/
4781 static int
4782 i_mdi_client_offline(dev_info_t *dip, uint_t flags)
4783 {
4784 	int		rv = NDI_SUCCESS;
4785 	mdi_client_t	*ct;
4786 
4787 	/*
4788 	 * Client component to go offline.  Make sure that we are
4789 	 * not in failing over state and update client state
4790 	 * accordingly
4791 	 */
4792 	ct = i_devi_get_client(dip);
4793 	MDI_DEBUG(2, (CE_NOTE, dip, "!i_mdi_client_offline called %p %p\n",
4794 	    (void *)dip, (void *)ct));
4795 	if (ct != NULL) {
4796 		MDI_CLIENT_LOCK(ct);
4797 		if (ct->ct_unstable) {
4798 			/*
4799 			 * One or more paths are in transient state,
4800 			 * Dont allow offline of a client device
4801 			 */
4802 			MDI_DEBUG(1, (CE_WARN, dip,
4803 			    "!One or more paths to this device is "
4804 			    "in transient state. This device can not "
4805 			    "be removed at this moment. "
4806 			    "Please try again later."));
4807 			MDI_CLIENT_UNLOCK(ct);
4808 			return (NDI_BUSY);
4809 		}
4810 		if (MDI_CLIENT_IS_FAILOVER_IN_PROGRESS(ct)) {
4811 			/*
4812 			 * Failover is in progress, Dont allow DR of
4813 			 * a client device
4814 			 */
4815 			MDI_DEBUG(1, (CE_WARN, dip,
4816 			    "!Client device (%s%d) is Busy. %s",
4817 			    ddi_driver_name(dip), ddi_get_instance(dip),
4818 			    "This device can not be removed at "
4819 			    "this moment. Please try again later."));
4820 			MDI_CLIENT_UNLOCK(ct);
4821 			return (NDI_BUSY);
4822 		}
4823 		MDI_CLIENT_SET_OFFLINE(ct);
4824 
4825 		/*
4826 		 * Unbind our relationship with the dev_info node
4827 		 */
4828 		if (flags & NDI_DEVI_REMOVE) {
4829 			ct->ct_dip = NULL;
4830 		}
4831 		MDI_CLIENT_UNLOCK(ct);
4832 	}
4833 	return (rv);
4834 }
4835 
4836 /*
4837  * mdi_pre_attach():
4838  *		Pre attach() notification handler
4839  */
4840 /*ARGSUSED*/
4841 int
4842 mdi_pre_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
4843 {
4844 	/* don't support old DDI_PM_RESUME */
4845 	if ((DEVI(dip)->devi_mdi_component != MDI_COMPONENT_NONE) &&
4846 	    (cmd == DDI_PM_RESUME))
4847 		return (DDI_FAILURE);
4848 
4849 	return (DDI_SUCCESS);
4850 }
4851 
4852 /*
4853  * mdi_post_attach():
4854  *		Post attach() notification handler
4855  */
4856 /*ARGSUSED*/
4857 void
4858 mdi_post_attach(dev_info_t *dip, ddi_attach_cmd_t cmd, int error)
4859 {
4860 	mdi_phci_t	*ph;
4861 	mdi_client_t	*ct;
4862 	mdi_vhci_t	*vh;
4863 
4864 	if (MDI_PHCI(dip)) {
4865 		ph = i_devi_get_phci(dip);
4866 		ASSERT(ph != NULL);
4867 
4868 		MDI_PHCI_LOCK(ph);
4869 		switch (cmd) {
4870 		case DDI_ATTACH:
4871 			MDI_DEBUG(2, (CE_NOTE, dip,
4872 			    "!pHCI post_attach: called %p\n", (void *)ph));
4873 			if (error == DDI_SUCCESS) {
4874 				MDI_PHCI_SET_ATTACH(ph);
4875 			} else {
4876 				MDI_DEBUG(1, (CE_NOTE, dip,
4877 				    "!pHCI post_attach: failed error=%d\n",
4878 				    error));
4879 				MDI_PHCI_SET_DETACH(ph);
4880 			}
4881 			break;
4882 
4883 		case DDI_RESUME:
4884 			MDI_DEBUG(2, (CE_NOTE, dip,
4885 			    "!pHCI post_resume: called %p\n", (void *)ph));
4886 			if (error == DDI_SUCCESS) {
4887 				MDI_PHCI_SET_RESUME(ph);
4888 			} else {
4889 				MDI_DEBUG(1, (CE_NOTE, dip,
4890 				    "!pHCI post_resume: failed error=%d\n",
4891 				    error));
4892 				MDI_PHCI_SET_SUSPEND(ph);
4893 			}
4894 			break;
4895 		}
4896 		MDI_PHCI_UNLOCK(ph);
4897 	}
4898 
4899 	if (MDI_CLIENT(dip)) {
4900 		ct = i_devi_get_client(dip);
4901 		ASSERT(ct != NULL);
4902 
4903 		MDI_CLIENT_LOCK(ct);
4904 		switch (cmd) {
4905 		case DDI_ATTACH:
4906 			MDI_DEBUG(2, (CE_NOTE, dip,
4907 			    "!Client post_attach: called %p\n", (void *)ct));
4908 			if (error != DDI_SUCCESS) {
4909 				MDI_DEBUG(1, (CE_NOTE, dip,
4910 				    "!Client post_attach: failed error=%d\n",
4911 				    error));
4912 				MDI_CLIENT_SET_DETACH(ct);
4913 				MDI_DEBUG(4, (CE_WARN, dip,
4914 				    "mdi_post_attach i_mdi_pm_reset_client\n"));
4915 				i_mdi_pm_reset_client(ct);
4916 				break;
4917 			}
4918 
4919 			/*
4920 			 * Client device has successfully attached, inform
4921 			 * the vhci.
4922 			 */
4923 			vh = ct->ct_vhci;
4924 			if (vh->vh_ops->vo_client_attached)
4925 				(*vh->vh_ops->vo_client_attached)(dip);
4926 
4927 			MDI_CLIENT_SET_ATTACH(ct);
4928 			break;
4929 
4930 		case DDI_RESUME:
4931 			MDI_DEBUG(2, (CE_NOTE, dip,
4932 			    "!Client post_attach: called %p\n", (void *)ct));
4933 			if (error == DDI_SUCCESS) {
4934 				MDI_CLIENT_SET_RESUME(ct);
4935 			} else {
4936 				MDI_DEBUG(1, (CE_NOTE, dip,
4937 				    "!Client post_resume: failed error=%d\n",
4938 				    error));
4939 				MDI_CLIENT_SET_SUSPEND(ct);
4940 			}
4941 			break;
4942 		}
4943 		MDI_CLIENT_UNLOCK(ct);
4944 	}
4945 }
4946 
4947 /*
4948  * mdi_pre_detach():
4949  *		Pre detach notification handler
4950  */
4951 /*ARGSUSED*/
4952 int
4953 mdi_pre_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
4954 {
4955 	int rv = DDI_SUCCESS;
4956 
4957 	if (MDI_CLIENT(dip)) {
4958 		(void) i_mdi_client_pre_detach(dip, cmd);
4959 	}
4960 
4961 	if (MDI_PHCI(dip)) {
4962 		rv = i_mdi_phci_pre_detach(dip, cmd);
4963 	}
4964 
4965 	return (rv);
4966 }
4967 
4968 /*ARGSUSED*/
4969 static int
4970 i_mdi_phci_pre_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
4971 {
4972 	int		rv = DDI_SUCCESS;
4973 	mdi_phci_t	*ph;
4974 	mdi_client_t	*ct;
4975 	mdi_pathinfo_t	*pip;
4976 	mdi_pathinfo_t	*failed_pip = NULL;
4977 	mdi_pathinfo_t	*next;
4978 
4979 	ph = i_devi_get_phci(dip);
4980 	if (ph == NULL) {
4981 		return (rv);
4982 	}
4983 
4984 	MDI_PHCI_LOCK(ph);
4985 	switch (cmd) {
4986 	case DDI_DETACH:
4987 		MDI_DEBUG(2, (CE_NOTE, dip,
4988 		    "!pHCI pre_detach: called %p\n", (void *)ph));
4989 		if (!MDI_PHCI_IS_OFFLINE(ph)) {
4990 			/*
4991 			 * mdi_pathinfo nodes are still attached to
4992 			 * this pHCI. Fail the detach for this pHCI.
4993 			 */
4994 			MDI_DEBUG(2, (CE_WARN, dip,
4995 			    "!pHCI pre_detach: "
4996 			    "mdi_pathinfo nodes are still attached "
4997 			    "%p\n", (void *)ph));
4998 			rv = DDI_FAILURE;
4999 			break;
5000 		}
5001 		MDI_PHCI_SET_DETACH(ph);
5002 		break;
5003 
5004 	case DDI_SUSPEND:
5005 		/*
5006 		 * pHCI is getting suspended.  Since mpxio client
5007 		 * devices may not be suspended at this point, to avoid
5008 		 * a potential stack overflow, it is important to suspend
5009 		 * client devices before pHCI can be suspended.
5010 		 */
5011 
5012 		MDI_DEBUG(2, (CE_NOTE, dip,
5013 		    "!pHCI pre_suspend: called %p\n", (void *)ph));
5014 		/*
5015 		 * Suspend all the client devices accessible through this pHCI
5016 		 */
5017 		pip = ph->ph_path_head;
5018 		while (pip != NULL && rv == DDI_SUCCESS) {
5019 			dev_info_t *cdip;
5020 			MDI_PI_LOCK(pip);
5021 			next =
5022 			    (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
5023 			ct = MDI_PI(pip)->pi_client;
5024 			i_mdi_client_lock(ct, pip);
5025 			cdip = ct->ct_dip;
5026 			MDI_PI_UNLOCK(pip);
5027 			if ((MDI_CLIENT_IS_DETACHED(ct) == 0) &&
5028 			    MDI_CLIENT_IS_SUSPENDED(ct) == 0) {
5029 				i_mdi_client_unlock(ct);
5030 				if ((rv = devi_detach(cdip, DDI_SUSPEND)) !=
5031 				    DDI_SUCCESS) {
5032 					/*
5033 					 * Suspend of one of the client
5034 					 * device has failed.
5035 					 */
5036 					MDI_DEBUG(1, (CE_WARN, dip,
5037 					    "!Suspend of device (%s%d) failed.",
5038 					    ddi_driver_name(cdip),
5039 					    ddi_get_instance(cdip)));
5040 					failed_pip = pip;
5041 					break;
5042 				}
5043 			} else {
5044 				i_mdi_client_unlock(ct);
5045 			}
5046 			pip = next;
5047 		}
5048 
5049 		if (rv == DDI_SUCCESS) {
5050 			/*
5051 			 * Suspend of client devices is complete. Proceed
5052 			 * with pHCI suspend.
5053 			 */
5054 			MDI_PHCI_SET_SUSPEND(ph);
5055 		} else {
5056 			/*
5057 			 * Revert back all the suspended client device states
5058 			 * to converse.
5059 			 */
5060 			pip = ph->ph_path_head;
5061 			while (pip != failed_pip) {
5062 				dev_info_t *cdip;
5063 				MDI_PI_LOCK(pip);
5064 				next =
5065 				    (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
5066 				ct = MDI_PI(pip)->pi_client;
5067 				i_mdi_client_lock(ct, pip);
5068 				cdip = ct->ct_dip;
5069 				MDI_PI_UNLOCK(pip);
5070 				if (MDI_CLIENT_IS_SUSPENDED(ct)) {
5071 					i_mdi_client_unlock(ct);
5072 					(void) devi_attach(cdip, DDI_RESUME);
5073 				} else {
5074 					i_mdi_client_unlock(ct);
5075 				}
5076 				pip = next;
5077 			}
5078 		}
5079 		break;
5080 
5081 	default:
5082 		rv = DDI_FAILURE;
5083 		break;
5084 	}
5085 	MDI_PHCI_UNLOCK(ph);
5086 	return (rv);
5087 }
5088 
5089 /*ARGSUSED*/
5090 static int
5091 i_mdi_client_pre_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
5092 {
5093 	int		rv = DDI_SUCCESS;
5094 	mdi_client_t	*ct;
5095 
5096 	ct = i_devi_get_client(dip);
5097 	if (ct == NULL) {
5098 		return (rv);
5099 	}
5100 
5101 	MDI_CLIENT_LOCK(ct);
5102 	switch (cmd) {
5103 	case DDI_DETACH:
5104 		MDI_DEBUG(2, (CE_NOTE, dip,
5105 		    "!Client pre_detach: called %p\n", (void *)ct));
5106 		MDI_CLIENT_SET_DETACH(ct);
5107 		break;
5108 
5109 	case DDI_SUSPEND:
5110 		MDI_DEBUG(2, (CE_NOTE, dip,
5111 		    "!Client pre_suspend: called %p\n", (void *)ct));
5112 		MDI_CLIENT_SET_SUSPEND(ct);
5113 		break;
5114 
5115 	default:
5116 		rv = DDI_FAILURE;
5117 		break;
5118 	}
5119 	MDI_CLIENT_UNLOCK(ct);
5120 	return (rv);
5121 }
5122 
5123 /*
5124  * mdi_post_detach():
5125  *		Post detach notification handler
5126  */
5127 /*ARGSUSED*/
5128 void
5129 mdi_post_detach(dev_info_t *dip, ddi_detach_cmd_t cmd, int error)
5130 {
5131 	/*
5132 	 * Detach/Suspend of mpxio component failed. Update our state
5133 	 * too
5134 	 */
5135 	if (MDI_PHCI(dip))
5136 		i_mdi_phci_post_detach(dip, cmd, error);
5137 
5138 	if (MDI_CLIENT(dip))
5139 		i_mdi_client_post_detach(dip, cmd, error);
5140 }
5141 
5142 /*ARGSUSED*/
5143 static void
5144 i_mdi_phci_post_detach(dev_info_t *dip, ddi_detach_cmd_t cmd, int error)
5145 {
5146 	mdi_phci_t	*ph;
5147 
5148 	/*
5149 	 * Detach/Suspend of phci component failed. Update our state
5150 	 * too
5151 	 */
5152 	ph = i_devi_get_phci(dip);
5153 	if (ph == NULL) {
5154 		return;
5155 	}
5156 
5157 	MDI_PHCI_LOCK(ph);
5158 	/*
5159 	 * Detach of pHCI failed. Restore back converse
5160 	 * state
5161 	 */
5162 	switch (cmd) {
5163 	case DDI_DETACH:
5164 		MDI_DEBUG(2, (CE_NOTE, dip,
5165 		    "!pHCI post_detach: called %p\n", (void *)ph));
5166 		if (error != DDI_SUCCESS)
5167 			MDI_PHCI_SET_ATTACH(ph);
5168 		break;
5169 
5170 	case DDI_SUSPEND:
5171 		MDI_DEBUG(2, (CE_NOTE, dip,
5172 		    "!pHCI post_suspend: called %p\n", (void *)ph));
5173 		if (error != DDI_SUCCESS)
5174 			MDI_PHCI_SET_RESUME(ph);
5175 		break;
5176 	}
5177 	MDI_PHCI_UNLOCK(ph);
5178 }
5179 
5180 /*ARGSUSED*/
5181 static void
5182 i_mdi_client_post_detach(dev_info_t *dip, ddi_detach_cmd_t cmd, int error)
5183 {
5184 	mdi_client_t	*ct;
5185 
5186 	ct = i_devi_get_client(dip);
5187 	if (ct == NULL) {
5188 		return;
5189 	}
5190 	MDI_CLIENT_LOCK(ct);
5191 	/*
5192 	 * Detach of Client failed. Restore back converse
5193 	 * state
5194 	 */
5195 	switch (cmd) {
5196 	case DDI_DETACH:
5197 		MDI_DEBUG(2, (CE_NOTE, dip,
5198 		    "!Client post_detach: called %p\n", (void *)ct));
5199 		if (DEVI_IS_ATTACHING(ct->ct_dip)) {
5200 			MDI_DEBUG(4, (CE_NOTE, dip, "i_mdi_client_post_detach "
5201 			    "i_mdi_pm_rele_client\n"));
5202 			i_mdi_pm_rele_client(ct, ct->ct_path_count);
5203 		} else {
5204 			MDI_DEBUG(4, (CE_NOTE, dip, "i_mdi_client_post_detach "
5205 			    "i_mdi_pm_reset_client\n"));
5206 			i_mdi_pm_reset_client(ct);
5207 		}
5208 		if (error != DDI_SUCCESS)
5209 			MDI_CLIENT_SET_ATTACH(ct);
5210 		break;
5211 
5212 	case DDI_SUSPEND:
5213 		MDI_DEBUG(2, (CE_NOTE, dip,
5214 		    "!Client post_suspend: called %p\n", (void *)ct));
5215 		if (error != DDI_SUCCESS)
5216 			MDI_CLIENT_SET_RESUME(ct);
5217 		break;
5218 	}
5219 	MDI_CLIENT_UNLOCK(ct);
5220 }
5221 
5222 int
5223 mdi_pi_kstat_exists(mdi_pathinfo_t *pip)
5224 {
5225 	return (MDI_PI(pip)->pi_kstats ? 1 : 0);
5226 }
5227 
5228 /*
5229  * create and install per-path (client - pHCI) statistics
5230  * I/O stats supported: nread, nwritten, reads, and writes
5231  * Error stats - hard errors, soft errors, & transport errors
5232  */
5233 int
5234 mdi_pi_kstat_create(mdi_pathinfo_t *pip, char *ksname)
5235 {
5236 	kstat_t			*kiosp, *kerrsp;
5237 	struct pi_errs		*nsp;
5238 	struct mdi_pi_kstats	*mdi_statp;
5239 
5240 	if (MDI_PI(pip)->pi_kstats != NULL)
5241 		return (MDI_SUCCESS);
5242 
5243 	if ((kiosp = kstat_create("mdi", 0, ksname, "iopath",
5244 	    KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT)) == NULL) {
5245 		return (MDI_FAILURE);
5246 	}
5247 
5248 	(void) strcat(ksname, ",err");
5249 	kerrsp = kstat_create("mdi", 0, ksname, "iopath_errors",
5250 	    KSTAT_TYPE_NAMED,
5251 	    sizeof (struct pi_errs) / sizeof (kstat_named_t), 0);
5252 	if (kerrsp == NULL) {
5253 		kstat_delete(kiosp);
5254 		return (MDI_FAILURE);
5255 	}
5256 
5257 	nsp = (struct pi_errs *)kerrsp->ks_data;
5258 	kstat_named_init(&nsp->pi_softerrs, "Soft Errors", KSTAT_DATA_UINT32);
5259 	kstat_named_init(&nsp->pi_harderrs, "Hard Errors", KSTAT_DATA_UINT32);
5260 	kstat_named_init(&nsp->pi_transerrs, "Transport Errors",
5261 	    KSTAT_DATA_UINT32);
5262 	kstat_named_init(&nsp->pi_icnt_busy, "Interconnect Busy",
5263 	    KSTAT_DATA_UINT32);
5264 	kstat_named_init(&nsp->pi_icnt_errors, "Interconnect Errors",
5265 	    KSTAT_DATA_UINT32);
5266 	kstat_named_init(&nsp->pi_phci_rsrc, "pHCI No Resources",
5267 	    KSTAT_DATA_UINT32);
5268 	kstat_named_init(&nsp->pi_phci_localerr, "pHCI Local Errors",
5269 	    KSTAT_DATA_UINT32);
5270 	kstat_named_init(&nsp->pi_phci_invstate, "pHCI Invalid State",
5271 	    KSTAT_DATA_UINT32);
5272 	kstat_named_init(&nsp->pi_failedfrom, "Failed From",
5273 	    KSTAT_DATA_UINT32);
5274 	kstat_named_init(&nsp->pi_failedto, "Failed To", KSTAT_DATA_UINT32);
5275 
5276 	mdi_statp = kmem_alloc(sizeof (*mdi_statp), KM_SLEEP);
5277 	mdi_statp->pi_kstat_ref = 1;
5278 	mdi_statp->pi_kstat_iostats = kiosp;
5279 	mdi_statp->pi_kstat_errstats = kerrsp;
5280 	kstat_install(kiosp);
5281 	kstat_install(kerrsp);
5282 	MDI_PI(pip)->pi_kstats = mdi_statp;
5283 	return (MDI_SUCCESS);
5284 }
5285 
5286 /*
5287  * destroy per-path properties
5288  */
5289 static void
5290 i_mdi_pi_kstat_destroy(mdi_pathinfo_t *pip)
5291 {
5292 
5293 	struct mdi_pi_kstats *mdi_statp;
5294 
5295 	if (MDI_PI(pip)->pi_kstats == NULL)
5296 		return;
5297 	if ((mdi_statp = MDI_PI(pip)->pi_kstats) == NULL)
5298 		return;
5299 
5300 	MDI_PI(pip)->pi_kstats = NULL;
5301 
5302 	/*
5303 	 * the kstat may be shared between multiple pathinfo nodes
5304 	 * decrement this pathinfo's usage, removing the kstats
5305 	 * themselves when the last pathinfo reference is removed.
5306 	 */
5307 	ASSERT(mdi_statp->pi_kstat_ref > 0);
5308 	if (--mdi_statp->pi_kstat_ref != 0)
5309 		return;
5310 
5311 	kstat_delete(mdi_statp->pi_kstat_iostats);
5312 	kstat_delete(mdi_statp->pi_kstat_errstats);
5313 	kmem_free(mdi_statp, sizeof (*mdi_statp));
5314 }
5315 
5316 /*
5317  * update I/O paths KSTATS
5318  */
5319 void
5320 mdi_pi_kstat_iosupdate(mdi_pathinfo_t *pip, struct buf *bp)
5321 {
5322 	kstat_t *iostatp;
5323 	size_t xfer_cnt;
5324 
5325 	ASSERT(pip != NULL);
5326 
5327 	/*
5328 	 * I/O can be driven across a path prior to having path
5329 	 * statistics available, i.e. probe(9e).
5330 	 */
5331 	if (bp != NULL && MDI_PI(pip)->pi_kstats != NULL) {
5332 		iostatp = MDI_PI(pip)->pi_kstats->pi_kstat_iostats;
5333 		xfer_cnt = bp->b_bcount - bp->b_resid;
5334 		if (bp->b_flags & B_READ) {
5335 			KSTAT_IO_PTR(iostatp)->reads++;
5336 			KSTAT_IO_PTR(iostatp)->nread += xfer_cnt;
5337 		} else {
5338 			KSTAT_IO_PTR(iostatp)->writes++;
5339 			KSTAT_IO_PTR(iostatp)->nwritten += xfer_cnt;
5340 		}
5341 	}
5342 }
5343 
5344 /*
5345  * Enable the path(specific client/target/initiator)
5346  * Enabling a path means that MPxIO may select the enabled path for routing
5347  * future I/O requests, subject to other path state constraints.
5348  */
5349 int
5350 mdi_pi_enable_path(mdi_pathinfo_t *pip, int flags)
5351 {
5352 	mdi_phci_t	*ph;
5353 
5354 	ph = i_devi_get_phci(mdi_pi_get_phci(pip));
5355 	if (ph == NULL) {
5356 		MDI_DEBUG(1, (CE_NOTE, NULL, "!mdi_pi_enable_path:"
5357 			" failed. pip: %p ph = NULL\n", (void *)pip));
5358 		return (MDI_FAILURE);
5359 	}
5360 
5361 	(void) i_mdi_enable_disable_path(pip, ph->ph_vhci, flags,
5362 		MDI_ENABLE_OP);
5363 	MDI_DEBUG(5, (CE_NOTE, NULL, "!mdi_pi_enable_path:"
5364 		" Returning success pip = %p. ph = %p\n",
5365 		(void *)pip, (void *)ph));
5366 	return (MDI_SUCCESS);
5367 
5368 }
5369 
5370 /*
5371  * Disable the path (specific client/target/initiator)
5372  * Disabling a path means that MPxIO will not select the disabled path for
5373  * routing any new I/O requests.
5374  */
5375 int
5376 mdi_pi_disable_path(mdi_pathinfo_t *pip, int flags)
5377 {
5378 	mdi_phci_t	*ph;
5379 
5380 	ph = i_devi_get_phci(mdi_pi_get_phci(pip));
5381 	if (ph == NULL) {
5382 		MDI_DEBUG(1, (CE_NOTE, NULL, "!mdi_pi_disable_path:"
5383 			" failed. pip: %p ph = NULL\n", (void *)pip));
5384 		return (MDI_FAILURE);
5385 	}
5386 
5387 	(void) i_mdi_enable_disable_path(pip,
5388 			ph->ph_vhci, flags, MDI_DISABLE_OP);
5389 	MDI_DEBUG(5, (CE_NOTE, NULL, "!mdi_pi_disable_path:"
5390 		"Returning success pip = %p. ph = %p",
5391 		(void *)pip, (void *)ph));
5392 	return (MDI_SUCCESS);
5393 }
5394 
5395 /*
5396  * disable the path to a particular pHCI (pHCI specified in the phci_path
5397  * argument) for a particular client (specified in the client_path argument).
5398  * Disabling a path means that MPxIO will not select the disabled path for
5399  * routing any new I/O requests.
5400  * NOTE: this will be removed once the NWS files are changed to use the new
5401  * mdi_{enable,disable}_path interfaces
5402  */
5403 int
5404 mdi_pi_disable(dev_info_t *cdip, dev_info_t *pdip, int flags)
5405 {
5406 	return (i_mdi_pi_enable_disable(cdip, pdip, flags, MDI_DISABLE_OP));
5407 }
5408 
5409 /*
5410  * Enable the path to a particular pHCI (pHCI specified in the phci_path
5411  * argument) for a particular client (specified in the client_path argument).
5412  * Enabling a path means that MPxIO may select the enabled path for routing
5413  * future I/O requests, subject to other path state constraints.
5414  * NOTE: this will be removed once the NWS files are changed to use the new
5415  * mdi_{enable,disable}_path interfaces
5416  */
5417 
5418 int
5419 mdi_pi_enable(dev_info_t *cdip, dev_info_t *pdip, int flags)
5420 {
5421 	return (i_mdi_pi_enable_disable(cdip, pdip, flags, MDI_ENABLE_OP));
5422 }
5423 
5424 /*
5425  * Common routine for doing enable/disable.
5426  */
5427 static mdi_pathinfo_t *
5428 i_mdi_enable_disable_path(mdi_pathinfo_t *pip, mdi_vhci_t *vh, int flags,
5429 		int op)
5430 {
5431 	int		sync_flag = 0;
5432 	int		rv;
5433 	mdi_pathinfo_t 	*next;
5434 	int		(*f)() = NULL;
5435 
5436 	f = vh->vh_ops->vo_pi_state_change;
5437 
5438 	sync_flag = (flags << 8) & 0xf00;
5439 
5440 	/*
5441 	 * Do a callback into the mdi consumer to let it
5442 	 * know that path is about to get enabled/disabled.
5443 	 */
5444 	if (f != NULL) {
5445 		rv = (*f)(vh->vh_dip, pip, 0,
5446 			MDI_PI_EXT_STATE(pip),
5447 			MDI_EXT_STATE_CHANGE | sync_flag |
5448 			op | MDI_BEFORE_STATE_CHANGE);
5449 		if (rv != MDI_SUCCESS) {
5450 			MDI_DEBUG(2, (CE_WARN, vh->vh_dip,
5451 			"!vo_pi_state_change: failed rv = %x", rv));
5452 		}
5453 	}
5454 	MDI_PI_LOCK(pip);
5455 	next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
5456 
5457 	switch (flags) {
5458 		case USER_DISABLE:
5459 			if (op == MDI_DISABLE_OP) {
5460 				MDI_PI_SET_USER_DISABLE(pip);
5461 			} else {
5462 				MDI_PI_SET_USER_ENABLE(pip);
5463 			}
5464 			break;
5465 		case DRIVER_DISABLE:
5466 			if (op == MDI_DISABLE_OP) {
5467 				MDI_PI_SET_DRV_DISABLE(pip);
5468 			} else {
5469 				MDI_PI_SET_DRV_ENABLE(pip);
5470 			}
5471 			break;
5472 		case DRIVER_DISABLE_TRANSIENT:
5473 			if (op == MDI_DISABLE_OP && rv == MDI_SUCCESS) {
5474 				MDI_PI_SET_DRV_DISABLE_TRANS(pip);
5475 			} else {
5476 				MDI_PI_SET_DRV_ENABLE_TRANS(pip);
5477 			}
5478 			break;
5479 	}
5480 	MDI_PI_UNLOCK(pip);
5481 	/*
5482 	 * Do a callback into the mdi consumer to let it
5483 	 * know that path is now enabled/disabled.
5484 	 */
5485 	if (f != NULL) {
5486 		rv = (*f)(vh->vh_dip, pip, 0,
5487 			MDI_PI_EXT_STATE(pip),
5488 			MDI_EXT_STATE_CHANGE | sync_flag |
5489 			op | MDI_AFTER_STATE_CHANGE);
5490 		if (rv != MDI_SUCCESS) {
5491 			MDI_DEBUG(2, (CE_WARN, vh->vh_dip,
5492 			"!vo_pi_state_change: failed rv = %x", rv));
5493 		}
5494 	}
5495 	return (next);
5496 }
5497 
5498 /*
5499  * Common routine for doing enable/disable.
5500  * NOTE: this will be removed once the NWS files are changed to use the new
5501  * mdi_{enable,disable}_path has been putback
5502  */
5503 int
5504 i_mdi_pi_enable_disable(dev_info_t *cdip, dev_info_t *pdip, int flags, int op)
5505 {
5506 
5507 	mdi_phci_t	*ph;
5508 	mdi_vhci_t	*vh = NULL;
5509 	mdi_client_t	*ct;
5510 	mdi_pathinfo_t	*next, *pip;
5511 	int		found_it;
5512 
5513 	ph = i_devi_get_phci(pdip);
5514 	MDI_DEBUG(5, (CE_NOTE, NULL, "!i_mdi_pi_enable_disable: "
5515 		"Op = %d pdip = %p cdip = %p\n", op, (void *)pdip,
5516 		(void *)cdip));
5517 	if (ph == NULL) {
5518 		MDI_DEBUG(1, (CE_NOTE, NULL, "!i_mdi_pi_enable_disable:"
5519 			"Op %d failed. ph = NULL\n", op));
5520 		return (MDI_FAILURE);
5521 	}
5522 
5523 	if ((op != MDI_ENABLE_OP) && (op != MDI_DISABLE_OP)) {
5524 		MDI_DEBUG(1, (CE_NOTE, NULL, "!i_mdi_pi_enable_disable: "
5525 			"Op Invalid operation = %d\n", op));
5526 		return (MDI_FAILURE);
5527 	}
5528 
5529 	vh = ph->ph_vhci;
5530 
5531 	if (cdip == NULL) {
5532 		/*
5533 		 * Need to mark the Phci as enabled/disabled.
5534 		 */
5535 		MDI_DEBUG(3, (CE_NOTE, NULL, "!i_mdi_pi_enable_disable: "
5536 		"Op %d for the phci\n", op));
5537 		MDI_PHCI_LOCK(ph);
5538 		switch (flags) {
5539 			case USER_DISABLE:
5540 				if (op == MDI_DISABLE_OP) {
5541 					MDI_PHCI_SET_USER_DISABLE(ph);
5542 				} else {
5543 					MDI_PHCI_SET_USER_ENABLE(ph);
5544 				}
5545 				break;
5546 			case DRIVER_DISABLE:
5547 				if (op == MDI_DISABLE_OP) {
5548 					MDI_PHCI_SET_DRV_DISABLE(ph);
5549 				} else {
5550 					MDI_PHCI_SET_DRV_ENABLE(ph);
5551 				}
5552 				break;
5553 			case DRIVER_DISABLE_TRANSIENT:
5554 				if (op == MDI_DISABLE_OP) {
5555 					MDI_PHCI_SET_DRV_DISABLE_TRANSIENT(ph);
5556 				} else {
5557 					MDI_PHCI_SET_DRV_ENABLE_TRANSIENT(ph);
5558 				}
5559 				break;
5560 			default:
5561 				MDI_PHCI_UNLOCK(ph);
5562 				MDI_DEBUG(1, (CE_NOTE, NULL,
5563 				"!i_mdi_pi_enable_disable:"
5564 				" Invalid flag argument= %d\n", flags));
5565 		}
5566 
5567 		/*
5568 		 * Phci has been disabled. Now try to enable/disable
5569 		 * path info's to each client.
5570 		 */
5571 		pip = ph->ph_path_head;
5572 		while (pip != NULL) {
5573 			pip = i_mdi_enable_disable_path(pip, vh, flags, op);
5574 		}
5575 		MDI_PHCI_UNLOCK(ph);
5576 	} else {
5577 
5578 		/*
5579 		 * Disable a specific client.
5580 		 */
5581 		ct = i_devi_get_client(cdip);
5582 		if (ct == NULL) {
5583 			MDI_DEBUG(1, (CE_NOTE, NULL,
5584 			"!i_mdi_pi_enable_disable:"
5585 			" failed. ct = NULL operation = %d\n", op));
5586 			return (MDI_FAILURE);
5587 		}
5588 
5589 		MDI_CLIENT_LOCK(ct);
5590 		pip = ct->ct_path_head;
5591 		found_it = 0;
5592 		while (pip != NULL) {
5593 			MDI_PI_LOCK(pip);
5594 			next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
5595 			if (MDI_PI(pip)->pi_phci == ph) {
5596 				MDI_PI_UNLOCK(pip);
5597 				found_it = 1;
5598 				break;
5599 			}
5600 			MDI_PI_UNLOCK(pip);
5601 			pip = next;
5602 		}
5603 
5604 
5605 		MDI_CLIENT_UNLOCK(ct);
5606 		if (found_it == 0) {
5607 			MDI_DEBUG(1, (CE_NOTE, NULL,
5608 			"!i_mdi_pi_enable_disable:"
5609 			" failed. Could not find corresponding pip\n"));
5610 			return (MDI_FAILURE);
5611 		}
5612 
5613 		(void) i_mdi_enable_disable_path(pip, vh, flags, op);
5614 	}
5615 
5616 	MDI_DEBUG(5, (CE_NOTE, NULL, "!i_mdi_pi_enable_disable: "
5617 		"Op %d Returning success pdip = %p cdip = %p\n",
5618 		op, (void *)pdip, (void *)cdip));
5619 	return (MDI_SUCCESS);
5620 }
5621 
5622 /*
5623  * Ensure phci powered up
5624  */
5625 static void
5626 i_mdi_pm_hold_pip(mdi_pathinfo_t *pip)
5627 {
5628 	dev_info_t	*ph_dip;
5629 
5630 	ASSERT(pip != NULL);
5631 	ASSERT(MDI_PI_LOCKED(pip));
5632 
5633 	if (MDI_PI(pip)->pi_pm_held) {
5634 		return;
5635 	}
5636 
5637 	ph_dip = mdi_pi_get_phci(pip);
5638 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "i_mdi_pm_hold_pip for %s%d %p\n",
5639 	    ddi_get_name(ph_dip), ddi_get_instance(ph_dip), (void *)pip));
5640 	if (ph_dip == NULL) {
5641 		return;
5642 	}
5643 
5644 	MDI_PI_UNLOCK(pip);
5645 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "kidsupcnt was %d\n",
5646 	    DEVI(ph_dip)->devi_pm_kidsupcnt));
5647 
5648 	pm_hold_power(ph_dip);
5649 
5650 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "kidsupcnt is %d\n",
5651 	    DEVI(ph_dip)->devi_pm_kidsupcnt));
5652 	MDI_PI_LOCK(pip);
5653 
5654 	/* If PM_GET_PM_INFO is NULL the pm_hold_power above was a noop */
5655 	if (DEVI(ph_dip)->devi_pm_info)
5656 		MDI_PI(pip)->pi_pm_held = 1;
5657 }
5658 
5659 /*
5660  * Allow phci powered down
5661  */
5662 static void
5663 i_mdi_pm_rele_pip(mdi_pathinfo_t *pip)
5664 {
5665 	dev_info_t	*ph_dip = NULL;
5666 
5667 	ASSERT(pip != NULL);
5668 	ASSERT(MDI_PI_LOCKED(pip));
5669 
5670 	if (MDI_PI(pip)->pi_pm_held == 0) {
5671 		return;
5672 	}
5673 
5674 	ph_dip = mdi_pi_get_phci(pip);
5675 	ASSERT(ph_dip != NULL);
5676 
5677 	MDI_PI_UNLOCK(pip);
5678 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "i_mdi_pm_rele_pip for %s%d %p\n",
5679 	    ddi_get_name(ph_dip), ddi_get_instance(ph_dip), (void *)pip));
5680 
5681 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "kidsupcnt was %d\n",
5682 	    DEVI(ph_dip)->devi_pm_kidsupcnt));
5683 	pm_rele_power(ph_dip);
5684 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "kidsupcnt is %d\n",
5685 	    DEVI(ph_dip)->devi_pm_kidsupcnt));
5686 
5687 	MDI_PI_LOCK(pip);
5688 	MDI_PI(pip)->pi_pm_held = 0;
5689 }
5690 
5691 static void
5692 i_mdi_pm_hold_client(mdi_client_t *ct, int incr)
5693 {
5694 	ASSERT(MDI_CLIENT_LOCKED(ct));
5695 
5696 	ct->ct_power_cnt += incr;
5697 	MDI_DEBUG(4, (CE_NOTE, ct->ct_dip, "i_mdi_pm_hold_client %p "
5698 	    "ct_power_cnt = %d incr = %d\n", (void *)ct,
5699 	    ct->ct_power_cnt, incr));
5700 	ASSERT(ct->ct_power_cnt >= 0);
5701 }
5702 
5703 static void
5704 i_mdi_rele_all_phci(mdi_client_t *ct)
5705 {
5706 	mdi_pathinfo_t  *pip;
5707 
5708 	ASSERT(MDI_CLIENT_LOCKED(ct));
5709 	pip = (mdi_pathinfo_t *)ct->ct_path_head;
5710 	while (pip != NULL) {
5711 		mdi_hold_path(pip);
5712 		MDI_PI_LOCK(pip);
5713 		i_mdi_pm_rele_pip(pip);
5714 		MDI_PI_UNLOCK(pip);
5715 		mdi_rele_path(pip);
5716 		pip = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
5717 	}
5718 }
5719 
5720 static void
5721 i_mdi_pm_rele_client(mdi_client_t *ct, int decr)
5722 {
5723 	ASSERT(MDI_CLIENT_LOCKED(ct));
5724 
5725 	if (i_ddi_devi_attached(ct->ct_dip)) {
5726 		ct->ct_power_cnt -= decr;
5727 		MDI_DEBUG(4, (CE_NOTE, ct->ct_dip, "i_mdi_pm_rele_client %p "
5728 		    "ct_power_cnt = %d decr = %d\n",
5729 		    (void *)ct, ct->ct_power_cnt, decr));
5730 	}
5731 
5732 	ASSERT(ct->ct_power_cnt >= 0);
5733 	if (ct->ct_power_cnt == 0) {
5734 		i_mdi_rele_all_phci(ct);
5735 		return;
5736 	}
5737 }
5738 
5739 static void
5740 i_mdi_pm_reset_client(mdi_client_t *ct)
5741 {
5742 	MDI_DEBUG(4, (CE_NOTE, ct->ct_dip, "i_mdi_pm_reset_client %p "
5743 	    "ct_power_cnt = %d\n", (void *)ct, ct->ct_power_cnt));
5744 	ASSERT(MDI_CLIENT_LOCKED(ct));
5745 	ct->ct_power_cnt = 0;
5746 	i_mdi_rele_all_phci(ct);
5747 	ct->ct_powercnt_config = 0;
5748 	ct->ct_powercnt_unconfig = 0;
5749 	ct->ct_powercnt_reset = 1;
5750 }
5751 
5752 static int
5753 i_mdi_power_one_phci(mdi_pathinfo_t *pip)
5754 {
5755 	int		ret;
5756 	dev_info_t	*ph_dip;
5757 
5758 	MDI_PI_LOCK(pip);
5759 	i_mdi_pm_hold_pip(pip);
5760 
5761 	ph_dip = mdi_pi_get_phci(pip);
5762 	MDI_PI_UNLOCK(pip);
5763 
5764 	/* bring all components of phci to full power */
5765 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "i_mdi_power_one_phci "
5766 	    "pm_powerup for %s%d %p\n", ddi_get_name(ph_dip),
5767 	    ddi_get_instance(ph_dip), (void *)pip));
5768 
5769 	ret = pm_powerup(ph_dip);
5770 
5771 	if (ret == DDI_FAILURE) {
5772 		MDI_DEBUG(4, (CE_NOTE, ph_dip, "i_mdi_power_one_phci "
5773 		    "pm_powerup FAILED for %s%d %p\n",
5774 		    ddi_get_name(ph_dip), ddi_get_instance(ph_dip),
5775 		    (void *)pip));
5776 
5777 		MDI_PI_LOCK(pip);
5778 		i_mdi_pm_rele_pip(pip);
5779 		MDI_PI_UNLOCK(pip);
5780 		return (MDI_FAILURE);
5781 	}
5782 
5783 	return (MDI_SUCCESS);
5784 }
5785 
5786 static int
5787 i_mdi_power_all_phci(mdi_client_t *ct)
5788 {
5789 	mdi_pathinfo_t  *pip;
5790 	int		succeeded = 0;
5791 
5792 	ASSERT(MDI_CLIENT_LOCKED(ct));
5793 	pip = (mdi_pathinfo_t *)ct->ct_path_head;
5794 	while (pip != NULL) {
5795 		/*
5796 		 * Don't power if MDI_PATHINFO_STATE_FAULT
5797 		 * or MDI_PATHINFO_STATE_OFFLINE.
5798 		 */
5799 		if (MDI_PI_IS_INIT(pip) ||
5800 		    MDI_PI_IS_ONLINE(pip) || MDI_PI_IS_STANDBY(pip)) {
5801 			mdi_hold_path(pip);
5802 			MDI_CLIENT_UNLOCK(ct);
5803 			if (i_mdi_power_one_phci(pip) == MDI_SUCCESS)
5804 				succeeded = 1;
5805 
5806 			ASSERT(ct == MDI_PI(pip)->pi_client);
5807 			MDI_CLIENT_LOCK(ct);
5808 			mdi_rele_path(pip);
5809 		}
5810 		pip = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
5811 	}
5812 
5813 	return (succeeded ? MDI_SUCCESS : MDI_FAILURE);
5814 }
5815 
5816 /*
5817  * mdi_bus_power():
5818  *		1. Place the phci(s) into powered up state so that
5819  *		   client can do power management
5820  *		2. Ensure phci powered up as client power managing
5821  * Return Values:
5822  *		MDI_SUCCESS
5823  *		MDI_FAILURE
5824  */
5825 int
5826 mdi_bus_power(dev_info_t *parent, void *impl_arg, pm_bus_power_op_t op,
5827     void *arg, void *result)
5828 {
5829 	int			ret = MDI_SUCCESS;
5830 	pm_bp_child_pwrchg_t	*bpc;
5831 	mdi_client_t		*ct;
5832 	dev_info_t		*cdip;
5833 	pm_bp_has_changed_t	*bphc;
5834 
5835 	/*
5836 	 * BUS_POWER_NOINVOL not supported
5837 	 */
5838 	if (op == BUS_POWER_NOINVOL)
5839 		return (MDI_FAILURE);
5840 
5841 	/*
5842 	 * ignore other OPs.
5843 	 * return quickly to save cou cycles on the ct processing
5844 	 */
5845 	switch (op) {
5846 	case BUS_POWER_PRE_NOTIFICATION:
5847 	case BUS_POWER_POST_NOTIFICATION:
5848 		bpc = (pm_bp_child_pwrchg_t *)arg;
5849 		cdip = bpc->bpc_dip;
5850 		break;
5851 	case BUS_POWER_HAS_CHANGED:
5852 		bphc = (pm_bp_has_changed_t *)arg;
5853 		cdip = bphc->bphc_dip;
5854 		break;
5855 	default:
5856 		return (pm_busop_bus_power(parent, impl_arg, op, arg, result));
5857 	}
5858 
5859 	ASSERT(MDI_CLIENT(cdip));
5860 
5861 	ct = i_devi_get_client(cdip);
5862 	if (ct == NULL)
5863 		return (MDI_FAILURE);
5864 
5865 	/*
5866 	 * wait till the mdi_pathinfo node state change are processed
5867 	 */
5868 	MDI_CLIENT_LOCK(ct);
5869 	switch (op) {
5870 	case BUS_POWER_PRE_NOTIFICATION:
5871 		MDI_DEBUG(4, (CE_NOTE, bpc->bpc_dip, "mdi_bus_power "
5872 		    "BUS_POWER_PRE_NOTIFICATION:"
5873 		    "%s@%s, olevel=%d, nlevel=%d, comp=%d\n",
5874 		    PM_NAME(bpc->bpc_dip), PM_ADDR(bpc->bpc_dip),
5875 		    bpc->bpc_olevel, bpc->bpc_nlevel, bpc->bpc_comp));
5876 
5877 		/* serialize power level change per client */
5878 		while (MDI_CLIENT_IS_POWER_TRANSITION(ct))
5879 			cv_wait(&ct->ct_powerchange_cv, &ct->ct_mutex);
5880 
5881 		MDI_CLIENT_SET_POWER_TRANSITION(ct);
5882 
5883 		if (ct->ct_power_cnt == 0) {
5884 			ret = i_mdi_power_all_phci(ct);
5885 		}
5886 
5887 		/*
5888 		 * if new_level > 0:
5889 		 *	- hold phci(s)
5890 		 *	- power up phci(s) if not already
5891 		 * ignore power down
5892 		 */
5893 		if (bpc->bpc_nlevel > 0) {
5894 			if (!DEVI_IS_ATTACHING(ct->ct_dip)) {
5895 				MDI_DEBUG(4, (CE_NOTE, bpc->bpc_dip,
5896 				    "mdi_bus_power i_mdi_pm_hold_client\n"));
5897 				i_mdi_pm_hold_client(ct, ct->ct_path_count);
5898 			}
5899 		}
5900 		break;
5901 	case BUS_POWER_POST_NOTIFICATION:
5902 		MDI_DEBUG(4, (CE_NOTE, bpc->bpc_dip, "mdi_bus_power "
5903 		    "BUS_POWER_POST_NOTIFICATION:"
5904 		    "%s@%s, olevel=%d, nlevel=%d, comp=%d result=%d\n",
5905 		    PM_NAME(bpc->bpc_dip), PM_ADDR(bpc->bpc_dip),
5906 		    bpc->bpc_olevel, bpc->bpc_nlevel, bpc->bpc_comp,
5907 		    *(int *)result));
5908 
5909 		if (*(int *)result == DDI_SUCCESS) {
5910 			if (bpc->bpc_nlevel > 0) {
5911 				MDI_CLIENT_SET_POWER_UP(ct);
5912 			} else {
5913 				MDI_CLIENT_SET_POWER_DOWN(ct);
5914 			}
5915 		}
5916 
5917 		/* release the hold we did in pre-notification */
5918 		if (bpc->bpc_nlevel > 0 && (*(int *)result != DDI_SUCCESS) &&
5919 		    !DEVI_IS_ATTACHING(ct->ct_dip)) {
5920 			MDI_DEBUG(4, (CE_NOTE, bpc->bpc_dip,
5921 			    "mdi_bus_power i_mdi_pm_rele_client\n"));
5922 			i_mdi_pm_rele_client(ct, ct->ct_path_count);
5923 		}
5924 
5925 		if (bpc->bpc_nlevel == 0 && (*(int *)result == DDI_SUCCESS)) {
5926 			/* another thread might started attaching */
5927 			if (DEVI_IS_ATTACHING(ct->ct_dip)) {
5928 				MDI_DEBUG(4, (CE_NOTE, bpc->bpc_dip,
5929 				    "mdi_bus_power i_mdi_pm_rele_client\n"));
5930 				i_mdi_pm_rele_client(ct, ct->ct_path_count);
5931 			/* detaching has been taken care in pm_post_unconfig */
5932 			} else if (!DEVI_IS_DETACHING(ct->ct_dip)) {
5933 				MDI_DEBUG(4, (CE_NOTE, bpc->bpc_dip,
5934 				    "mdi_bus_power i_mdi_pm_reset_client\n"));
5935 				i_mdi_pm_reset_client(ct);
5936 			}
5937 		}
5938 
5939 		MDI_CLIENT_CLEAR_POWER_TRANSITION(ct);
5940 		cv_broadcast(&ct->ct_powerchange_cv);
5941 
5942 		break;
5943 
5944 	/* need to do more */
5945 	case BUS_POWER_HAS_CHANGED:
5946 		MDI_DEBUG(4, (CE_NOTE, bphc->bphc_dip, "mdi_bus_power "
5947 		    "BUS_POWER_HAS_CHANGED:"
5948 		    "%s@%s, olevel=%d, nlevel=%d, comp=%d\n",
5949 		    PM_NAME(bphc->bphc_dip), PM_ADDR(bphc->bphc_dip),
5950 		    bphc->bphc_olevel, bphc->bphc_nlevel, bphc->bphc_comp));
5951 
5952 		if (bphc->bphc_nlevel > 0 &&
5953 		    bphc->bphc_nlevel > bphc->bphc_olevel) {
5954 			if (ct->ct_power_cnt == 0) {
5955 				ret = i_mdi_power_all_phci(ct);
5956 			}
5957 			MDI_DEBUG(4, (CE_NOTE, bphc->bphc_dip,
5958 			    "mdi_bus_power i_mdi_pm_hold_client\n"));
5959 			i_mdi_pm_hold_client(ct, ct->ct_path_count);
5960 		}
5961 
5962 		if (bphc->bphc_nlevel == 0 && bphc->bphc_olevel != -1) {
5963 			MDI_DEBUG(4, (CE_NOTE, bphc->bphc_dip,
5964 			    "mdi_bus_power i_mdi_pm_rele_client\n"));
5965 			i_mdi_pm_rele_client(ct, ct->ct_path_count);
5966 		}
5967 		break;
5968 	}
5969 
5970 	MDI_CLIENT_UNLOCK(ct);
5971 	return (ret);
5972 }
5973 
5974 static int
5975 i_mdi_pm_pre_config_one(dev_info_t *child)
5976 {
5977 	int		ret = MDI_SUCCESS;
5978 	mdi_client_t	*ct;
5979 
5980 	ct = i_devi_get_client(child);
5981 	if (ct == NULL)
5982 		return (MDI_FAILURE);
5983 
5984 	MDI_CLIENT_LOCK(ct);
5985 	while (MDI_CLIENT_IS_POWER_TRANSITION(ct))
5986 		cv_wait(&ct->ct_powerchange_cv, &ct->ct_mutex);
5987 
5988 	if (!MDI_CLIENT_IS_FAILED(ct)) {
5989 		MDI_CLIENT_UNLOCK(ct);
5990 		MDI_DEBUG(4, (CE_NOTE, child,
5991 		    "i_mdi_pm_pre_config_one already configured\n"));
5992 		return (MDI_SUCCESS);
5993 	}
5994 
5995 	if (ct->ct_powercnt_config) {
5996 		MDI_CLIENT_UNLOCK(ct);
5997 		MDI_DEBUG(4, (CE_NOTE, child,
5998 		    "i_mdi_pm_pre_config_one ALREADY held\n"));
5999 		return (MDI_SUCCESS);
6000 	}
6001 
6002 	if (ct->ct_power_cnt == 0) {
6003 		ret = i_mdi_power_all_phci(ct);
6004 	}
6005 	MDI_DEBUG(4, (CE_NOTE, child,
6006 	    "i_mdi_pm_pre_config_one i_mdi_pm_hold_client\n"));
6007 	i_mdi_pm_hold_client(ct, ct->ct_path_count);
6008 	ct->ct_powercnt_config = 1;
6009 	ct->ct_powercnt_reset = 0;
6010 	MDI_CLIENT_UNLOCK(ct);
6011 	return (ret);
6012 }
6013 
6014 static int
6015 i_mdi_pm_pre_config(dev_info_t *vdip, dev_info_t *child)
6016 {
6017 	int			ret = MDI_SUCCESS;
6018 	dev_info_t		*cdip;
6019 	int			circ;
6020 
6021 	ASSERT(MDI_VHCI(vdip));
6022 
6023 	/* ndi_devi_config_one */
6024 	if (child) {
6025 		ASSERT(DEVI_BUSY_OWNED(vdip));
6026 		return (i_mdi_pm_pre_config_one(child));
6027 	}
6028 
6029 	/* devi_config_common */
6030 	ndi_devi_enter(vdip, &circ);
6031 	cdip = ddi_get_child(vdip);
6032 	while (cdip) {
6033 		dev_info_t *next = ddi_get_next_sibling(cdip);
6034 
6035 		ret = i_mdi_pm_pre_config_one(cdip);
6036 		if (ret != MDI_SUCCESS)
6037 			break;
6038 		cdip = next;
6039 	}
6040 	ndi_devi_exit(vdip, circ);
6041 	return (ret);
6042 }
6043 
6044 static int
6045 i_mdi_pm_pre_unconfig_one(dev_info_t *child, int *held, int flags)
6046 {
6047 	int		ret = MDI_SUCCESS;
6048 	mdi_client_t	*ct;
6049 
6050 	ct = i_devi_get_client(child);
6051 	if (ct == NULL)
6052 		return (MDI_FAILURE);
6053 
6054 	MDI_CLIENT_LOCK(ct);
6055 	while (MDI_CLIENT_IS_POWER_TRANSITION(ct))
6056 		cv_wait(&ct->ct_powerchange_cv, &ct->ct_mutex);
6057 
6058 	if (!i_ddi_devi_attached(ct->ct_dip)) {
6059 		MDI_DEBUG(4, (CE_NOTE, child,
6060 		    "i_mdi_pm_pre_unconfig node detached already\n"));
6061 		MDI_CLIENT_UNLOCK(ct);
6062 		return (MDI_SUCCESS);
6063 	}
6064 
6065 	if (MDI_CLIENT_IS_POWERED_DOWN(ct) &&
6066 	    (flags & NDI_AUTODETACH)) {
6067 		MDI_DEBUG(4, (CE_NOTE, child,
6068 		    "i_mdi_pm_pre_unconfig auto-modunload\n"));
6069 		MDI_CLIENT_UNLOCK(ct);
6070 		return (MDI_FAILURE);
6071 	}
6072 
6073 	if (ct->ct_powercnt_unconfig) {
6074 		MDI_DEBUG(4, (CE_NOTE, child,
6075 		    "i_mdi_pm_pre_unconfig ct_powercnt_held\n"));
6076 		MDI_CLIENT_UNLOCK(ct);
6077 		*held = 1;
6078 		return (MDI_SUCCESS);
6079 	}
6080 
6081 	if (ct->ct_power_cnt == 0) {
6082 		ret = i_mdi_power_all_phci(ct);
6083 	}
6084 	MDI_DEBUG(4, (CE_NOTE, child,
6085 	    "i_mdi_pm_pre_unconfig i_mdi_pm_hold_client\n"));
6086 	i_mdi_pm_hold_client(ct, ct->ct_path_count);
6087 	ct->ct_powercnt_unconfig = 1;
6088 	ct->ct_powercnt_reset = 0;
6089 	MDI_CLIENT_UNLOCK(ct);
6090 	if (ret == MDI_SUCCESS)
6091 		*held = 1;
6092 	return (ret);
6093 }
6094 
6095 static int
6096 i_mdi_pm_pre_unconfig(dev_info_t *vdip, dev_info_t *child, int *held,
6097     int flags)
6098 {
6099 	int			ret = MDI_SUCCESS;
6100 	dev_info_t		*cdip;
6101 	int			circ;
6102 
6103 	ASSERT(MDI_VHCI(vdip));
6104 	*held = 0;
6105 
6106 	/* ndi_devi_unconfig_one */
6107 	if (child) {
6108 		ASSERT(DEVI_BUSY_OWNED(vdip));
6109 		return (i_mdi_pm_pre_unconfig_one(child, held, flags));
6110 	}
6111 
6112 	/* devi_unconfig_common */
6113 	ndi_devi_enter(vdip, &circ);
6114 	cdip = ddi_get_child(vdip);
6115 	while (cdip) {
6116 		dev_info_t *next = ddi_get_next_sibling(cdip);
6117 
6118 		ret = i_mdi_pm_pre_unconfig_one(cdip, held, flags);
6119 		cdip = next;
6120 	}
6121 	ndi_devi_exit(vdip, circ);
6122 
6123 	if (*held)
6124 		ret = MDI_SUCCESS;
6125 
6126 	return (ret);
6127 }
6128 
6129 static void
6130 i_mdi_pm_post_config_one(dev_info_t *child)
6131 {
6132 	mdi_client_t	*ct;
6133 
6134 	ct = i_devi_get_client(child);
6135 	if (ct == NULL)
6136 		return;
6137 
6138 	MDI_CLIENT_LOCK(ct);
6139 	while (MDI_CLIENT_IS_POWER_TRANSITION(ct))
6140 		cv_wait(&ct->ct_powerchange_cv, &ct->ct_mutex);
6141 
6142 	if (ct->ct_powercnt_reset || !ct->ct_powercnt_config) {
6143 		MDI_DEBUG(4, (CE_NOTE, child,
6144 		    "i_mdi_pm_post_config_one NOT configured\n"));
6145 		MDI_CLIENT_UNLOCK(ct);
6146 		return;
6147 	}
6148 
6149 	/* client has not been updated */
6150 	if (MDI_CLIENT_IS_FAILED(ct)) {
6151 		MDI_DEBUG(4, (CE_NOTE, child,
6152 		    "i_mdi_pm_post_config_one NOT configured\n"));
6153 		MDI_CLIENT_UNLOCK(ct);
6154 		return;
6155 	}
6156 
6157 	/* another thread might have powered it down or detached it */
6158 	if ((MDI_CLIENT_IS_POWERED_DOWN(ct) &&
6159 	    !DEVI_IS_ATTACHING(ct->ct_dip)) ||
6160 	    (!i_ddi_devi_attached(ct->ct_dip) &&
6161 	    !DEVI_IS_ATTACHING(ct->ct_dip))) {
6162 		MDI_DEBUG(4, (CE_NOTE, child,
6163 		    "i_mdi_pm_post_config i_mdi_pm_reset_client\n"));
6164 		i_mdi_pm_reset_client(ct);
6165 	} else {
6166 		mdi_pathinfo_t  *pip, *next;
6167 		int	valid_path_count = 0;
6168 
6169 		MDI_DEBUG(4, (CE_NOTE, child,
6170 		    "i_mdi_pm_post_config i_mdi_pm_rele_client\n"));
6171 		pip = ct->ct_path_head;
6172 		while (pip != NULL) {
6173 			MDI_PI_LOCK(pip);
6174 			next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
6175 			if (MDI_PI_IS_ONLINE(pip) || MDI_PI_IS_STANDBY(pip))
6176 				valid_path_count ++;
6177 			MDI_PI_UNLOCK(pip);
6178 			pip = next;
6179 		}
6180 		i_mdi_pm_rele_client(ct, valid_path_count);
6181 	}
6182 	ct->ct_powercnt_config = 0;
6183 	MDI_CLIENT_UNLOCK(ct);
6184 }
6185 
6186 static void
6187 i_mdi_pm_post_config(dev_info_t *vdip, dev_info_t *child)
6188 {
6189 	int		circ;
6190 	dev_info_t	*cdip;
6191 
6192 	ASSERT(MDI_VHCI(vdip));
6193 
6194 	/* ndi_devi_config_one */
6195 	if (child) {
6196 		ASSERT(DEVI_BUSY_OWNED(vdip));
6197 		i_mdi_pm_post_config_one(child);
6198 		return;
6199 	}
6200 
6201 	/* devi_config_common */
6202 	ndi_devi_enter(vdip, &circ);
6203 	cdip = ddi_get_child(vdip);
6204 	while (cdip) {
6205 		dev_info_t *next = ddi_get_next_sibling(cdip);
6206 
6207 		i_mdi_pm_post_config_one(cdip);
6208 		cdip = next;
6209 	}
6210 	ndi_devi_exit(vdip, circ);
6211 }
6212 
6213 static void
6214 i_mdi_pm_post_unconfig_one(dev_info_t *child)
6215 {
6216 	mdi_client_t	*ct;
6217 
6218 	ct = i_devi_get_client(child);
6219 	if (ct == NULL)
6220 		return;
6221 
6222 	MDI_CLIENT_LOCK(ct);
6223 	while (MDI_CLIENT_IS_POWER_TRANSITION(ct))
6224 		cv_wait(&ct->ct_powerchange_cv, &ct->ct_mutex);
6225 
6226 	if (!ct->ct_powercnt_unconfig || ct->ct_powercnt_reset) {
6227 		MDI_DEBUG(4, (CE_NOTE, child,
6228 		    "i_mdi_pm_post_unconfig NOT held\n"));
6229 		MDI_CLIENT_UNLOCK(ct);
6230 		return;
6231 	}
6232 
6233 	/* failure detaching or another thread just attached it */
6234 	if ((MDI_CLIENT_IS_POWERED_DOWN(ct) &&
6235 	    i_ddi_devi_attached(ct->ct_dip)) ||
6236 	    (!i_ddi_devi_attached(ct->ct_dip) &&
6237 	    !DEVI_IS_ATTACHING(ct->ct_dip))) {
6238 		MDI_DEBUG(4, (CE_NOTE, child,
6239 		    "i_mdi_pm_post_unconfig i_mdi_pm_reset_client\n"));
6240 		i_mdi_pm_reset_client(ct);
6241 	} else {
6242 		mdi_pathinfo_t  *pip, *next;
6243 		int	valid_path_count = 0;
6244 
6245 		MDI_DEBUG(4, (CE_NOTE, child,
6246 		    "i_mdi_pm_post_unconfig i_mdi_pm_rele_client\n"));
6247 		pip = ct->ct_path_head;
6248 		while (pip != NULL) {
6249 			MDI_PI_LOCK(pip);
6250 			next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
6251 			if (MDI_PI_IS_ONLINE(pip) || MDI_PI_IS_STANDBY(pip))
6252 				valid_path_count ++;
6253 			MDI_PI_UNLOCK(pip);
6254 			pip = next;
6255 		}
6256 		i_mdi_pm_rele_client(ct, valid_path_count);
6257 		ct->ct_powercnt_unconfig = 0;
6258 	}
6259 
6260 	MDI_CLIENT_UNLOCK(ct);
6261 }
6262 
6263 static void
6264 i_mdi_pm_post_unconfig(dev_info_t *vdip, dev_info_t *child, int held)
6265 {
6266 	int			circ;
6267 	dev_info_t		*cdip;
6268 
6269 	ASSERT(MDI_VHCI(vdip));
6270 
6271 	if (!held) {
6272 		MDI_DEBUG(4, (CE_NOTE, vdip,
6273 		    "i_mdi_pm_post_unconfig held = %d\n", held));
6274 		return;
6275 	}
6276 
6277 	if (child) {
6278 		ASSERT(DEVI_BUSY_OWNED(vdip));
6279 		i_mdi_pm_post_unconfig_one(child);
6280 		return;
6281 	}
6282 
6283 	ndi_devi_enter(vdip, &circ);
6284 	cdip = ddi_get_child(vdip);
6285 	while (cdip) {
6286 		dev_info_t *next = ddi_get_next_sibling(cdip);
6287 
6288 		i_mdi_pm_post_unconfig_one(cdip);
6289 		cdip = next;
6290 	}
6291 	ndi_devi_exit(vdip, circ);
6292 }
6293 
6294 int
6295 mdi_power(dev_info_t *vdip, mdi_pm_op_t op, void *args, char *devnm, int flags)
6296 {
6297 	int			circ, ret = MDI_SUCCESS;
6298 	dev_info_t		*client_dip = NULL;
6299 	mdi_client_t		*ct;
6300 
6301 	/*
6302 	 * Handling ndi_devi_config_one and ndi_devi_unconfig_one.
6303 	 * Power up pHCI for the named client device.
6304 	 * Note: Before the client is enumerated under vhci by phci,
6305 	 * client_dip can be NULL. Then proceed to power up all the
6306 	 * pHCIs.
6307 	 */
6308 	if (devnm != NULL) {
6309 		ndi_devi_enter(vdip, &circ);
6310 		client_dip = ndi_devi_findchild(vdip, devnm);
6311 	}
6312 
6313 	MDI_DEBUG(4, (CE_NOTE, vdip, "mdi_power op = %d %s %p\n",
6314 	    op, devnm ? devnm : "NULL", (void *)client_dip));
6315 
6316 	switch (op) {
6317 	case MDI_PM_PRE_CONFIG:
6318 		ret = i_mdi_pm_pre_config(vdip, client_dip);
6319 		break;
6320 
6321 	case MDI_PM_PRE_UNCONFIG:
6322 		ret = i_mdi_pm_pre_unconfig(vdip, client_dip, (int *)args,
6323 		    flags);
6324 		break;
6325 
6326 	case MDI_PM_POST_CONFIG:
6327 		i_mdi_pm_post_config(vdip, client_dip);
6328 		break;
6329 
6330 	case MDI_PM_POST_UNCONFIG:
6331 		i_mdi_pm_post_unconfig(vdip, client_dip, *(int *)args);
6332 		break;
6333 
6334 	case MDI_PM_HOLD_POWER:
6335 	case MDI_PM_RELE_POWER:
6336 		ASSERT(args);
6337 
6338 		client_dip = (dev_info_t *)args;
6339 		ASSERT(MDI_CLIENT(client_dip));
6340 
6341 		ct = i_devi_get_client(client_dip);
6342 		MDI_CLIENT_LOCK(ct);
6343 
6344 		if (op == MDI_PM_HOLD_POWER) {
6345 			if (ct->ct_power_cnt == 0) {
6346 				(void) i_mdi_power_all_phci(ct);
6347 				MDI_DEBUG(4, (CE_NOTE, client_dip,
6348 				    "mdi_power i_mdi_pm_hold_client\n"));
6349 				i_mdi_pm_hold_client(ct, ct->ct_path_count);
6350 			}
6351 		} else {
6352 			if (DEVI_IS_ATTACHING(ct->ct_dip)) {
6353 				MDI_DEBUG(4, (CE_NOTE, client_dip,
6354 				    "mdi_power i_mdi_pm_rele_client\n"));
6355 				i_mdi_pm_rele_client(ct, ct->ct_path_count);
6356 			} else {
6357 				MDI_DEBUG(4, (CE_NOTE, client_dip,
6358 				    "mdi_power i_mdi_pm_reset_client\n"));
6359 				i_mdi_pm_reset_client(ct);
6360 			}
6361 		}
6362 
6363 		MDI_CLIENT_UNLOCK(ct);
6364 		break;
6365 
6366 	default:
6367 		break;
6368 	}
6369 
6370 	if (devnm)
6371 		ndi_devi_exit(vdip, circ);
6372 
6373 	return (ret);
6374 }
6375 
6376 int
6377 mdi_component_is_vhci(dev_info_t *dip, const char **mdi_class)
6378 {
6379 	mdi_vhci_t *vhci;
6380 
6381 	if (!MDI_VHCI(dip))
6382 		return (MDI_FAILURE);
6383 
6384 	if (mdi_class) {
6385 		vhci = DEVI(dip)->devi_mdi_xhci;
6386 		ASSERT(vhci);
6387 		*mdi_class = vhci->vh_class;
6388 	}
6389 
6390 	return (MDI_SUCCESS);
6391 }
6392 
6393 int
6394 mdi_component_is_phci(dev_info_t *dip, const char **mdi_class)
6395 {
6396 	mdi_phci_t *phci;
6397 
6398 	if (!MDI_PHCI(dip))
6399 		return (MDI_FAILURE);
6400 
6401 	if (mdi_class) {
6402 		phci = DEVI(dip)->devi_mdi_xhci;
6403 		ASSERT(phci);
6404 		*mdi_class = phci->ph_vhci->vh_class;
6405 	}
6406 
6407 	return (MDI_SUCCESS);
6408 }
6409 
6410 int
6411 mdi_component_is_client(dev_info_t *dip, const char **mdi_class)
6412 {
6413 	mdi_client_t *client;
6414 
6415 	if (!MDI_CLIENT(dip))
6416 		return (MDI_FAILURE);
6417 
6418 	if (mdi_class) {
6419 		client = DEVI(dip)->devi_mdi_client;
6420 		ASSERT(client);
6421 		*mdi_class = client->ct_vhci->vh_class;
6422 	}
6423 
6424 	return (MDI_SUCCESS);
6425 }
6426 
6427 void *
6428 mdi_client_get_vhci_private(dev_info_t *dip)
6429 {
6430 	ASSERT(mdi_component_is_client(dip, NULL) == MDI_SUCCESS);
6431 	if (mdi_component_is_client(dip, NULL) == MDI_SUCCESS) {
6432 		mdi_client_t	*ct;
6433 		ct = i_devi_get_client(dip);
6434 		return (ct->ct_vprivate);
6435 	}
6436 	return (NULL);
6437 }
6438 
6439 void
6440 mdi_client_set_vhci_private(dev_info_t *dip, void *data)
6441 {
6442 	ASSERT(mdi_component_is_client(dip, NULL) == MDI_SUCCESS);
6443 	if (mdi_component_is_client(dip, NULL) == MDI_SUCCESS) {
6444 		mdi_client_t	*ct;
6445 		ct = i_devi_get_client(dip);
6446 		ct->ct_vprivate = data;
6447 	}
6448 }
6449 /*
6450  * mdi_pi_get_vhci_private():
6451  *		Get the vhci private information associated with the
6452  *		mdi_pathinfo node
6453  */
6454 void *
6455 mdi_pi_get_vhci_private(mdi_pathinfo_t *pip)
6456 {
6457 	caddr_t	vprivate = NULL;
6458 	if (pip) {
6459 		vprivate = MDI_PI(pip)->pi_vprivate;
6460 	}
6461 	return (vprivate);
6462 }
6463 
6464 /*
6465  * mdi_pi_set_vhci_private():
6466  *		Set the vhci private information in the mdi_pathinfo node
6467  */
6468 void
6469 mdi_pi_set_vhci_private(mdi_pathinfo_t *pip, void *priv)
6470 {
6471 	if (pip) {
6472 		MDI_PI(pip)->pi_vprivate = priv;
6473 	}
6474 }
6475 
6476 /*
6477  * mdi_phci_get_vhci_private():
6478  *		Get the vhci private information associated with the
6479  *		mdi_phci node
6480  */
6481 void *
6482 mdi_phci_get_vhci_private(dev_info_t *dip)
6483 {
6484 	ASSERT(mdi_component_is_phci(dip, NULL) == MDI_SUCCESS);
6485 	if (mdi_component_is_phci(dip, NULL) == MDI_SUCCESS) {
6486 		mdi_phci_t	*ph;
6487 		ph = i_devi_get_phci(dip);
6488 		return (ph->ph_vprivate);
6489 	}
6490 	return (NULL);
6491 }
6492 
6493 /*
6494  * mdi_phci_set_vhci_private():
6495  *		Set the vhci private information in the mdi_phci node
6496  */
6497 void
6498 mdi_phci_set_vhci_private(dev_info_t *dip, void *priv)
6499 {
6500 	ASSERT(mdi_component_is_phci(dip, NULL) == MDI_SUCCESS);
6501 	if (mdi_component_is_phci(dip, NULL) == MDI_SUCCESS) {
6502 		mdi_phci_t	*ph;
6503 		ph = i_devi_get_phci(dip);
6504 		ph->ph_vprivate = priv;
6505 	}
6506 }
6507 
6508 /*
6509  * List of vhci class names:
6510  * A vhci class name must be in this list only if the corresponding vhci
6511  * driver intends to use the mdi provided bus config implementation
6512  * (i.e., mdi_vhci_bus_config()).
6513  */
6514 static char *vhci_class_list[] = { MDI_HCI_CLASS_SCSI, MDI_HCI_CLASS_IB };
6515 #define	N_VHCI_CLASSES	(sizeof (vhci_class_list) / sizeof (char *))
6516 
6517 /*
6518  * During boot time, the on-disk vhci cache for every vhci class is read
6519  * in the form of an nvlist and stored here.
6520  */
6521 static nvlist_t *vhcache_nvl[N_VHCI_CLASSES];
6522 
6523 /* nvpair names in vhci cache nvlist */
6524 #define	MDI_VHCI_CACHE_VERSION	1
6525 #define	MDI_NVPNAME_VERSION	"version"
6526 #define	MDI_NVPNAME_PHCIS	"phcis"
6527 #define	MDI_NVPNAME_CTADDRMAP	"clientaddrmap"
6528 
6529 /*
6530  * Given vhci class name, return its on-disk vhci cache filename.
6531  * Memory for the returned filename which includes the full path is allocated
6532  * by this function.
6533  */
6534 static char *
6535 vhclass2vhcache_filename(char *vhclass)
6536 {
6537 	char *filename;
6538 	int len;
6539 	static char *fmt = "/etc/devices/mdi_%s_cache";
6540 
6541 	/*
6542 	 * fmt contains the on-disk vhci cache file name format;
6543 	 * for scsi_vhci the filename is "/etc/devices/mdi_scsi_vhci_cache".
6544 	 */
6545 
6546 	/* the -1 below is to account for "%s" in the format string */
6547 	len = strlen(fmt) + strlen(vhclass) - 1;
6548 	filename = kmem_alloc(len, KM_SLEEP);
6549 	(void) snprintf(filename, len, fmt, vhclass);
6550 	ASSERT(len == (strlen(filename) + 1));
6551 	return (filename);
6552 }
6553 
6554 /*
6555  * initialize the vhci cache related data structures and read the on-disk
6556  * vhci cached data into memory.
6557  */
6558 static void
6559 setup_vhci_cache(mdi_vhci_t *vh)
6560 {
6561 	mdi_vhci_config_t *vhc;
6562 	mdi_vhci_cache_t *vhcache;
6563 	int i;
6564 	nvlist_t *nvl = NULL;
6565 
6566 	vhc = kmem_zalloc(sizeof (mdi_vhci_config_t), KM_SLEEP);
6567 	vh->vh_config = vhc;
6568 	vhcache = &vhc->vhc_vhcache;
6569 
6570 	vhc->vhc_vhcache_filename = vhclass2vhcache_filename(vh->vh_class);
6571 
6572 	mutex_init(&vhc->vhc_lock, NULL, MUTEX_DEFAULT, NULL);
6573 	cv_init(&vhc->vhc_cv, NULL, CV_DRIVER, NULL);
6574 
6575 	rw_init(&vhcache->vhcache_lock, NULL, RW_DRIVER, NULL);
6576 
6577 	/*
6578 	 * Create string hash; same as mod_hash_create_strhash() except that
6579 	 * we use NULL key destructor.
6580 	 */
6581 	vhcache->vhcache_client_hash = mod_hash_create_extended(vh->vh_class,
6582 	    mdi_bus_config_cache_hash_size,
6583 	    mod_hash_null_keydtor, mod_hash_null_valdtor,
6584 	    mod_hash_bystr, NULL, mod_hash_strkey_cmp, KM_SLEEP);
6585 
6586 	/*
6587 	 * The on-disk vhci cache is read during booting prior to the
6588 	 * lights-out period by mdi_read_devices_files().
6589 	 */
6590 	for (i = 0; i < N_VHCI_CLASSES; i++) {
6591 		if (strcmp(vhci_class_list[i], vh->vh_class) == 0) {
6592 			nvl = vhcache_nvl[i];
6593 			vhcache_nvl[i] = NULL;
6594 			break;
6595 		}
6596 	}
6597 
6598 	/*
6599 	 * this is to cover the case of some one manually causing unloading
6600 	 * (or detaching) and reloading (or attaching) of a vhci driver.
6601 	 */
6602 	if (nvl == NULL && modrootloaded)
6603 		nvl = read_on_disk_vhci_cache(vh->vh_class);
6604 
6605 	if (nvl != NULL) {
6606 		rw_enter(&vhcache->vhcache_lock, RW_WRITER);
6607 		if (mainnvl_to_vhcache(vhcache, nvl) == MDI_SUCCESS)
6608 			vhcache->vhcache_flags |= MDI_VHCI_CACHE_SETUP_DONE;
6609 		else  {
6610 			cmn_err(CE_WARN,
6611 			    "%s: data file corrupted, will recreate\n",
6612 			    vhc->vhc_vhcache_filename);
6613 		}
6614 		rw_exit(&vhcache->vhcache_lock);
6615 		nvlist_free(nvl);
6616 	}
6617 
6618 	vhc->vhc_cbid = callb_add(stop_vhcache_flush_thread, vhc,
6619 	    CB_CL_UADMIN_PRE_VFS, "mdi_vhcache_flush");
6620 
6621 	vhc->vhc_path_discovery_boot = mdi_path_discovery_boot;
6622 	vhc->vhc_path_discovery_postboot = mdi_path_discovery_postboot;
6623 }
6624 
6625 /*
6626  * free all vhci cache related resources
6627  */
6628 static int
6629 destroy_vhci_cache(mdi_vhci_t *vh)
6630 {
6631 	mdi_vhci_config_t *vhc = vh->vh_config;
6632 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
6633 	mdi_vhcache_phci_t *cphci, *cphci_next;
6634 	mdi_vhcache_client_t *cct, *cct_next;
6635 	mdi_vhcache_pathinfo_t *cpi, *cpi_next;
6636 
6637 	if (stop_vhcache_async_threads(vhc) != MDI_SUCCESS)
6638 		return (MDI_FAILURE);
6639 
6640 	kmem_free(vhc->vhc_vhcache_filename,
6641 	    strlen(vhc->vhc_vhcache_filename) + 1);
6642 
6643 	mod_hash_destroy_strhash(vhcache->vhcache_client_hash);
6644 
6645 	for (cphci = vhcache->vhcache_phci_head; cphci != NULL;
6646 	    cphci = cphci_next) {
6647 		cphci_next = cphci->cphci_next;
6648 		free_vhcache_phci(cphci);
6649 	}
6650 
6651 	for (cct = vhcache->vhcache_client_head; cct != NULL; cct = cct_next) {
6652 		cct_next = cct->cct_next;
6653 		for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi_next) {
6654 			cpi_next = cpi->cpi_next;
6655 			free_vhcache_pathinfo(cpi);
6656 		}
6657 		free_vhcache_client(cct);
6658 	}
6659 
6660 	rw_destroy(&vhcache->vhcache_lock);
6661 
6662 	mutex_destroy(&vhc->vhc_lock);
6663 	cv_destroy(&vhc->vhc_cv);
6664 	kmem_free(vhc, sizeof (mdi_vhci_config_t));
6665 	return (MDI_SUCCESS);
6666 }
6667 
6668 /*
6669  * Stop all vhci cache related async threads and free their resources.
6670  */
6671 static int
6672 stop_vhcache_async_threads(mdi_vhci_config_t *vhc)
6673 {
6674 	mdi_async_client_config_t *acc, *acc_next;
6675 
6676 	mutex_enter(&vhc->vhc_lock);
6677 	vhc->vhc_flags |= MDI_VHC_EXIT;
6678 	ASSERT(vhc->vhc_acc_thrcount >= 0);
6679 	cv_broadcast(&vhc->vhc_cv);
6680 
6681 	while ((vhc->vhc_flags & MDI_VHC_VHCACHE_FLUSH_THREAD) ||
6682 	    vhc->vhc_acc_thrcount != 0) {
6683 		mutex_exit(&vhc->vhc_lock);
6684 		delay(1);
6685 		mutex_enter(&vhc->vhc_lock);
6686 	}
6687 
6688 	vhc->vhc_flags &= ~MDI_VHC_EXIT;
6689 
6690 	for (acc = vhc->vhc_acc_list_head; acc != NULL; acc = acc_next) {
6691 		acc_next = acc->acc_next;
6692 		free_async_client_config(acc);
6693 	}
6694 	vhc->vhc_acc_list_head = NULL;
6695 	vhc->vhc_acc_list_tail = NULL;
6696 	vhc->vhc_acc_count = 0;
6697 
6698 	if (vhc->vhc_flags & MDI_VHC_VHCACHE_DIRTY) {
6699 		vhc->vhc_flags &= ~MDI_VHC_VHCACHE_DIRTY;
6700 		mutex_exit(&vhc->vhc_lock);
6701 		if (flush_vhcache(vhc, 0) != MDI_SUCCESS) {
6702 			vhcache_dirty(vhc);
6703 			return (MDI_FAILURE);
6704 		}
6705 	} else
6706 		mutex_exit(&vhc->vhc_lock);
6707 
6708 	if (callb_delete(vhc->vhc_cbid) != 0)
6709 		return (MDI_FAILURE);
6710 
6711 	return (MDI_SUCCESS);
6712 }
6713 
6714 /*
6715  * Stop vhci cache flush thread
6716  */
6717 /* ARGSUSED */
6718 static boolean_t
6719 stop_vhcache_flush_thread(void *arg, int code)
6720 {
6721 	mdi_vhci_config_t *vhc = (mdi_vhci_config_t *)arg;
6722 
6723 	mutex_enter(&vhc->vhc_lock);
6724 	vhc->vhc_flags |= MDI_VHC_EXIT;
6725 	cv_broadcast(&vhc->vhc_cv);
6726 
6727 	while (vhc->vhc_flags & MDI_VHC_VHCACHE_FLUSH_THREAD) {
6728 		mutex_exit(&vhc->vhc_lock);
6729 		delay(1);
6730 		mutex_enter(&vhc->vhc_lock);
6731 	}
6732 
6733 	if (vhc->vhc_flags & MDI_VHC_VHCACHE_DIRTY) {
6734 		vhc->vhc_flags &= ~MDI_VHC_VHCACHE_DIRTY;
6735 		mutex_exit(&vhc->vhc_lock);
6736 		(void) flush_vhcache(vhc, 1);
6737 	} else
6738 		mutex_exit(&vhc->vhc_lock);
6739 
6740 	return (B_TRUE);
6741 }
6742 
6743 /*
6744  * Enqueue the vhcache phci (cphci) at the tail of the list
6745  */
6746 static void
6747 enqueue_vhcache_phci(mdi_vhci_cache_t *vhcache, mdi_vhcache_phci_t *cphci)
6748 {
6749 	cphci->cphci_next = NULL;
6750 	if (vhcache->vhcache_phci_head == NULL)
6751 		vhcache->vhcache_phci_head = cphci;
6752 	else
6753 		vhcache->vhcache_phci_tail->cphci_next = cphci;
6754 	vhcache->vhcache_phci_tail = cphci;
6755 }
6756 
6757 /*
6758  * Enqueue the vhcache pathinfo (cpi) at the tail of the list
6759  */
6760 static void
6761 enqueue_tail_vhcache_pathinfo(mdi_vhcache_client_t *cct,
6762     mdi_vhcache_pathinfo_t *cpi)
6763 {
6764 	cpi->cpi_next = NULL;
6765 	if (cct->cct_cpi_head == NULL)
6766 		cct->cct_cpi_head = cpi;
6767 	else
6768 		cct->cct_cpi_tail->cpi_next = cpi;
6769 	cct->cct_cpi_tail = cpi;
6770 }
6771 
6772 /*
6773  * Enqueue the vhcache pathinfo (cpi) at the correct location in the
6774  * ordered list. All cpis which do not have MDI_CPI_HINT_PATH_DOES_NOT_EXIST
6775  * flag set come at the beginning of the list. All cpis which have this
6776  * flag set come at the end of the list.
6777  */
6778 static void
6779 enqueue_vhcache_pathinfo(mdi_vhcache_client_t *cct,
6780     mdi_vhcache_pathinfo_t *newcpi)
6781 {
6782 	mdi_vhcache_pathinfo_t *cpi, *prev_cpi;
6783 
6784 	if (cct->cct_cpi_head == NULL ||
6785 	    (newcpi->cpi_flags & MDI_CPI_HINT_PATH_DOES_NOT_EXIST))
6786 		enqueue_tail_vhcache_pathinfo(cct, newcpi);
6787 	else {
6788 		for (cpi = cct->cct_cpi_head, prev_cpi = NULL; cpi != NULL &&
6789 		    !(cpi->cpi_flags & MDI_CPI_HINT_PATH_DOES_NOT_EXIST);
6790 		    prev_cpi = cpi, cpi = cpi->cpi_next)
6791 			;
6792 
6793 		if (prev_cpi == NULL)
6794 			cct->cct_cpi_head = newcpi;
6795 		else
6796 			prev_cpi->cpi_next = newcpi;
6797 
6798 		newcpi->cpi_next = cpi;
6799 
6800 		if (cpi == NULL)
6801 			cct->cct_cpi_tail = newcpi;
6802 	}
6803 }
6804 
6805 /*
6806  * Enqueue the vhcache client (cct) at the tail of the list
6807  */
6808 static void
6809 enqueue_vhcache_client(mdi_vhci_cache_t *vhcache,
6810     mdi_vhcache_client_t *cct)
6811 {
6812 	cct->cct_next = NULL;
6813 	if (vhcache->vhcache_client_head == NULL)
6814 		vhcache->vhcache_client_head = cct;
6815 	else
6816 		vhcache->vhcache_client_tail->cct_next = cct;
6817 	vhcache->vhcache_client_tail = cct;
6818 }
6819 
6820 static void
6821 free_string_array(char **str, int nelem)
6822 {
6823 	int i;
6824 
6825 	if (str) {
6826 		for (i = 0; i < nelem; i++) {
6827 			if (str[i])
6828 				kmem_free(str[i], strlen(str[i]) + 1);
6829 		}
6830 		kmem_free(str, sizeof (char *) * nelem);
6831 	}
6832 }
6833 
6834 static void
6835 free_vhcache_phci(mdi_vhcache_phci_t *cphci)
6836 {
6837 	kmem_free(cphci->cphci_path, strlen(cphci->cphci_path) + 1);
6838 	kmem_free(cphci, sizeof (*cphci));
6839 }
6840 
6841 static void
6842 free_vhcache_pathinfo(mdi_vhcache_pathinfo_t *cpi)
6843 {
6844 	kmem_free(cpi->cpi_addr, strlen(cpi->cpi_addr) + 1);
6845 	kmem_free(cpi, sizeof (*cpi));
6846 }
6847 
6848 static void
6849 free_vhcache_client(mdi_vhcache_client_t *cct)
6850 {
6851 	kmem_free(cct->cct_name_addr, strlen(cct->cct_name_addr) + 1);
6852 	kmem_free(cct, sizeof (*cct));
6853 }
6854 
6855 static char *
6856 vhcache_mknameaddr(char *ct_name, char *ct_addr, int *ret_len)
6857 {
6858 	char *name_addr;
6859 	int len;
6860 
6861 	len = strlen(ct_name) + strlen(ct_addr) + 2;
6862 	name_addr = kmem_alloc(len, KM_SLEEP);
6863 	(void) snprintf(name_addr, len, "%s@%s", ct_name, ct_addr);
6864 
6865 	if (ret_len)
6866 		*ret_len = len;
6867 	return (name_addr);
6868 }
6869 
6870 /*
6871  * Copy the contents of paddrnvl to vhci cache.
6872  * paddrnvl nvlist contains path information for a vhci client.
6873  * See the comment in mainnvl_to_vhcache() for the format of this nvlist.
6874  */
6875 static void
6876 paddrnvl_to_vhcache(nvlist_t *nvl, mdi_vhcache_phci_t *cphci_list[],
6877     mdi_vhcache_client_t *cct)
6878 {
6879 	nvpair_t *nvp = NULL;
6880 	mdi_vhcache_pathinfo_t *cpi;
6881 	uint_t nelem;
6882 	uint32_t *val;
6883 
6884 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
6885 		ASSERT(nvpair_type(nvp) == DATA_TYPE_UINT32_ARRAY);
6886 		cpi = kmem_zalloc(sizeof (*cpi), KM_SLEEP);
6887 		cpi->cpi_addr = i_ddi_strdup(nvpair_name(nvp), KM_SLEEP);
6888 		(void) nvpair_value_uint32_array(nvp, &val, &nelem);
6889 		ASSERT(nelem == 2);
6890 		cpi->cpi_cphci = cphci_list[val[0]];
6891 		cpi->cpi_flags = val[1];
6892 		enqueue_tail_vhcache_pathinfo(cct, cpi);
6893 	}
6894 }
6895 
6896 /*
6897  * Copy the contents of caddrmapnvl to vhci cache.
6898  * caddrmapnvl nvlist contains vhci client address to phci client address
6899  * mappings. See the comment in mainnvl_to_vhcache() for the format of
6900  * this nvlist.
6901  */
6902 static void
6903 caddrmapnvl_to_vhcache(mdi_vhci_cache_t *vhcache, nvlist_t *nvl,
6904     mdi_vhcache_phci_t *cphci_list[])
6905 {
6906 	nvpair_t *nvp = NULL;
6907 	nvlist_t *paddrnvl;
6908 	mdi_vhcache_client_t *cct;
6909 
6910 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
6911 		ASSERT(nvpair_type(nvp) == DATA_TYPE_NVLIST);
6912 		cct = kmem_zalloc(sizeof (*cct), KM_SLEEP);
6913 		cct->cct_name_addr = i_ddi_strdup(nvpair_name(nvp), KM_SLEEP);
6914 		(void) nvpair_value_nvlist(nvp, &paddrnvl);
6915 		paddrnvl_to_vhcache(paddrnvl, cphci_list, cct);
6916 		/* the client must contain at least one path */
6917 		ASSERT(cct->cct_cpi_head != NULL);
6918 
6919 		enqueue_vhcache_client(vhcache, cct);
6920 		(void) mod_hash_insert(vhcache->vhcache_client_hash,
6921 		    (mod_hash_key_t)cct->cct_name_addr, (mod_hash_val_t)cct);
6922 	}
6923 }
6924 
6925 /*
6926  * Copy the contents of the main nvlist to vhci cache.
6927  *
6928  * VHCI busconfig cached data is stored in the form of a nvlist on the disk.
6929  * The nvlist contains the mappings between the vhci client addresses and
6930  * their corresponding phci client addresses.
6931  *
6932  * The structure of the nvlist is as follows:
6933  *
6934  * Main nvlist:
6935  *	NAME		TYPE		DATA
6936  *	version		int32		version number
6937  *	phcis		string array	array of phci paths
6938  *	clientaddrmap	nvlist_t	c2paddrs_nvl (see below)
6939  *
6940  * structure of c2paddrs_nvl:
6941  *	NAME		TYPE		DATA
6942  *	caddr1		nvlist_t	paddrs_nvl1
6943  *	caddr2		nvlist_t	paddrs_nvl2
6944  *	...
6945  * where caddr1, caddr2, ... are vhci client name and addresses in the
6946  * form of "<clientname>@<clientaddress>".
6947  * (for example: "ssd@2000002037cd9f72");
6948  * paddrs_nvl1, paddrs_nvl2, .. are nvlists that contain path information.
6949  *
6950  * structure of paddrs_nvl:
6951  *	NAME		TYPE		DATA
6952  *	pi_addr1	uint32_array	(phci-id, cpi_flags)
6953  *	pi_addr2	uint32_array	(phci-id, cpi_flags)
6954  *	...
6955  * where pi_addr1, pi_addr2, ... are bus specific addresses of pathinfo nodes
6956  * (so called pi_addrs, for example: "w2100002037cd9f72,0");
6957  * phci-ids are integers that identify PHCIs to which the
6958  * the bus specific address belongs to. These integers are used as an index
6959  * into to the phcis string array in the main nvlist to get the PHCI path.
6960  */
6961 static int
6962 mainnvl_to_vhcache(mdi_vhci_cache_t *vhcache, nvlist_t *nvl)
6963 {
6964 	char **phcis, **phci_namep;
6965 	uint_t nphcis;
6966 	mdi_vhcache_phci_t *cphci, **cphci_list;
6967 	nvlist_t *caddrmapnvl;
6968 	int32_t ver;
6969 	int i;
6970 	size_t cphci_list_size;
6971 
6972 	ASSERT(RW_WRITE_HELD(&vhcache->vhcache_lock));
6973 
6974 	if (nvlist_lookup_int32(nvl, MDI_NVPNAME_VERSION, &ver) != 0 ||
6975 	    ver != MDI_VHCI_CACHE_VERSION)
6976 		return (MDI_FAILURE);
6977 
6978 	if (nvlist_lookup_string_array(nvl, MDI_NVPNAME_PHCIS, &phcis,
6979 	    &nphcis) != 0)
6980 		return (MDI_SUCCESS);
6981 
6982 	ASSERT(nphcis > 0);
6983 
6984 	cphci_list_size = sizeof (mdi_vhcache_phci_t *) * nphcis;
6985 	cphci_list = kmem_alloc(cphci_list_size, KM_SLEEP);
6986 	for (i = 0, phci_namep = phcis; i < nphcis; i++, phci_namep++) {
6987 		cphci = kmem_zalloc(sizeof (mdi_vhcache_phci_t), KM_SLEEP);
6988 		cphci->cphci_path = i_ddi_strdup(*phci_namep, KM_SLEEP);
6989 		enqueue_vhcache_phci(vhcache, cphci);
6990 		cphci_list[i] = cphci;
6991 	}
6992 
6993 	ASSERT(vhcache->vhcache_phci_head != NULL);
6994 
6995 	if (nvlist_lookup_nvlist(nvl, MDI_NVPNAME_CTADDRMAP, &caddrmapnvl) == 0)
6996 		caddrmapnvl_to_vhcache(vhcache, caddrmapnvl, cphci_list);
6997 
6998 	kmem_free(cphci_list, cphci_list_size);
6999 	return (MDI_SUCCESS);
7000 }
7001 
7002 /*
7003  * Build paddrnvl for the specified client using the information in the
7004  * vhci cache and add it to the caddrmapnnvl.
7005  * Returns 0 on success, errno on failure.
7006  */
7007 static int
7008 vhcache_to_paddrnvl(mdi_vhci_cache_t *vhcache, mdi_vhcache_client_t *cct,
7009     nvlist_t *caddrmapnvl)
7010 {
7011 	mdi_vhcache_pathinfo_t *cpi;
7012 	nvlist_t *nvl;
7013 	int err;
7014 	uint32_t val[2];
7015 
7016 	ASSERT(RW_LOCK_HELD(&vhcache->vhcache_lock));
7017 
7018 	if ((err = nvlist_alloc(&nvl, 0, KM_SLEEP)) != 0)
7019 		return (err);
7020 
7021 	for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi->cpi_next) {
7022 		val[0] = cpi->cpi_cphci->cphci_id;
7023 		val[1] = cpi->cpi_flags;
7024 		if ((err = nvlist_add_uint32_array(nvl, cpi->cpi_addr, val, 2))
7025 		    != 0)
7026 			goto out;
7027 	}
7028 
7029 	err = nvlist_add_nvlist(caddrmapnvl, cct->cct_name_addr, nvl);
7030 out:
7031 	nvlist_free(nvl);
7032 	return (err);
7033 }
7034 
7035 /*
7036  * Build caddrmapnvl using the information in the vhci cache
7037  * and add it to the mainnvl.
7038  * Returns 0 on success, errno on failure.
7039  */
7040 static int
7041 vhcache_to_caddrmapnvl(mdi_vhci_cache_t *vhcache, nvlist_t *mainnvl)
7042 {
7043 	mdi_vhcache_client_t *cct;
7044 	nvlist_t *nvl;
7045 	int err;
7046 
7047 	ASSERT(RW_LOCK_HELD(&vhcache->vhcache_lock));
7048 
7049 	if ((err = nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP)) != 0)
7050 		return (err);
7051 
7052 	for (cct = vhcache->vhcache_client_head; cct != NULL;
7053 	    cct = cct->cct_next) {
7054 		if ((err = vhcache_to_paddrnvl(vhcache, cct, nvl)) != 0)
7055 			goto out;
7056 	}
7057 
7058 	err = nvlist_add_nvlist(mainnvl, MDI_NVPNAME_CTADDRMAP, nvl);
7059 out:
7060 	nvlist_free(nvl);
7061 	return (err);
7062 }
7063 
7064 /*
7065  * Build nvlist using the information in the vhci cache.
7066  * See the comment in mainnvl_to_vhcache() for the format of the nvlist.
7067  * Returns nvl on success, NULL on failure.
7068  */
7069 static nvlist_t *
7070 vhcache_to_mainnvl(mdi_vhci_cache_t *vhcache)
7071 {
7072 	mdi_vhcache_phci_t *cphci;
7073 	uint_t phci_count;
7074 	char **phcis;
7075 	nvlist_t *nvl;
7076 	int err, i;
7077 
7078 	if ((err = nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP)) != 0) {
7079 		nvl = NULL;
7080 		goto out;
7081 	}
7082 
7083 	if ((err = nvlist_add_int32(nvl, MDI_NVPNAME_VERSION,
7084 	    MDI_VHCI_CACHE_VERSION)) != 0)
7085 		goto out;
7086 
7087 	rw_enter(&vhcache->vhcache_lock, RW_READER);
7088 	if (vhcache->vhcache_phci_head == NULL) {
7089 		rw_exit(&vhcache->vhcache_lock);
7090 		return (nvl);
7091 	}
7092 
7093 	phci_count = 0;
7094 	for (cphci = vhcache->vhcache_phci_head; cphci != NULL;
7095 	    cphci = cphci->cphci_next)
7096 		cphci->cphci_id = phci_count++;
7097 
7098 	/* build phci pathname list */
7099 	phcis = kmem_alloc(sizeof (char *) * phci_count, KM_SLEEP);
7100 	for (cphci = vhcache->vhcache_phci_head, i = 0; cphci != NULL;
7101 	    cphci = cphci->cphci_next, i++)
7102 		phcis[i] = i_ddi_strdup(cphci->cphci_path, KM_SLEEP);
7103 
7104 	err = nvlist_add_string_array(nvl, MDI_NVPNAME_PHCIS, phcis,
7105 	    phci_count);
7106 	free_string_array(phcis, phci_count);
7107 
7108 	if (err == 0 &&
7109 	    (err = vhcache_to_caddrmapnvl(vhcache, nvl)) == 0) {
7110 		rw_exit(&vhcache->vhcache_lock);
7111 		return (nvl);
7112 	}
7113 
7114 	rw_exit(&vhcache->vhcache_lock);
7115 out:
7116 	if (nvl)
7117 		nvlist_free(nvl);
7118 	return (NULL);
7119 }
7120 
7121 /*
7122  * Lookup vhcache phci structure for the specified phci path.
7123  */
7124 static mdi_vhcache_phci_t *
7125 lookup_vhcache_phci_by_name(mdi_vhci_cache_t *vhcache, char *phci_path)
7126 {
7127 	mdi_vhcache_phci_t *cphci;
7128 
7129 	ASSERT(RW_LOCK_HELD(&vhcache->vhcache_lock));
7130 
7131 	for (cphci = vhcache->vhcache_phci_head; cphci != NULL;
7132 	    cphci = cphci->cphci_next) {
7133 		if (strcmp(cphci->cphci_path, phci_path) == 0)
7134 			return (cphci);
7135 	}
7136 
7137 	return (NULL);
7138 }
7139 
7140 /*
7141  * Lookup vhcache phci structure for the specified phci.
7142  */
7143 static mdi_vhcache_phci_t *
7144 lookup_vhcache_phci_by_addr(mdi_vhci_cache_t *vhcache, mdi_phci_t *ph)
7145 {
7146 	mdi_vhcache_phci_t *cphci;
7147 
7148 	ASSERT(RW_LOCK_HELD(&vhcache->vhcache_lock));
7149 
7150 	for (cphci = vhcache->vhcache_phci_head; cphci != NULL;
7151 	    cphci = cphci->cphci_next) {
7152 		if (cphci->cphci_phci == ph)
7153 			return (cphci);
7154 	}
7155 
7156 	return (NULL);
7157 }
7158 
7159 /*
7160  * Add the specified phci to the vhci cache if not already present.
7161  */
7162 static void
7163 vhcache_phci_add(mdi_vhci_config_t *vhc, mdi_phci_t *ph)
7164 {
7165 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7166 	mdi_vhcache_phci_t *cphci;
7167 	char *pathname;
7168 	int cache_updated;
7169 
7170 	rw_enter(&vhcache->vhcache_lock, RW_WRITER);
7171 
7172 	pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
7173 	(void) ddi_pathname(ph->ph_dip, pathname);
7174 	if ((cphci = lookup_vhcache_phci_by_name(vhcache, pathname))
7175 	    != NULL) {
7176 		cphci->cphci_phci = ph;
7177 		cache_updated = 0;
7178 	} else {
7179 		cphci = kmem_zalloc(sizeof (*cphci), KM_SLEEP);
7180 		cphci->cphci_path = i_ddi_strdup(pathname, KM_SLEEP);
7181 		cphci->cphci_phci = ph;
7182 		enqueue_vhcache_phci(vhcache, cphci);
7183 		cache_updated = 1;
7184 	}
7185 
7186 	rw_exit(&vhcache->vhcache_lock);
7187 
7188 	/*
7189 	 * Since a new phci has been added, reset
7190 	 * vhc_path_discovery_cutoff_time to allow for discovery of paths
7191 	 * during next vhcache_discover_paths().
7192 	 */
7193 	mutex_enter(&vhc->vhc_lock);
7194 	vhc->vhc_path_discovery_cutoff_time = 0;
7195 	mutex_exit(&vhc->vhc_lock);
7196 
7197 	kmem_free(pathname, MAXPATHLEN);
7198 	if (cache_updated)
7199 		vhcache_dirty(vhc);
7200 }
7201 
7202 /*
7203  * Remove the reference to the specified phci from the vhci cache.
7204  */
7205 static void
7206 vhcache_phci_remove(mdi_vhci_config_t *vhc, mdi_phci_t *ph)
7207 {
7208 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7209 	mdi_vhcache_phci_t *cphci;
7210 
7211 	rw_enter(&vhcache->vhcache_lock, RW_WRITER);
7212 	if ((cphci = lookup_vhcache_phci_by_addr(vhcache, ph)) != NULL) {
7213 		/* do not remove the actual mdi_vhcache_phci structure */
7214 		cphci->cphci_phci = NULL;
7215 	}
7216 	rw_exit(&vhcache->vhcache_lock);
7217 }
7218 
7219 static void
7220 init_vhcache_lookup_token(mdi_vhcache_lookup_token_t *dst,
7221     mdi_vhcache_lookup_token_t *src)
7222 {
7223 	if (src == NULL) {
7224 		dst->lt_cct = NULL;
7225 		dst->lt_cct_lookup_time = 0;
7226 	} else {
7227 		dst->lt_cct = src->lt_cct;
7228 		dst->lt_cct_lookup_time = src->lt_cct_lookup_time;
7229 	}
7230 }
7231 
7232 /*
7233  * Look up vhcache client for the specified client.
7234  */
7235 static mdi_vhcache_client_t *
7236 lookup_vhcache_client(mdi_vhci_cache_t *vhcache, char *ct_name, char *ct_addr,
7237     mdi_vhcache_lookup_token_t *token)
7238 {
7239 	mod_hash_val_t hv;
7240 	char *name_addr;
7241 	int len;
7242 
7243 	ASSERT(RW_LOCK_HELD(&vhcache->vhcache_lock));
7244 
7245 	/*
7246 	 * If no vhcache clean occurred since the last lookup, we can
7247 	 * simply return the cct from the last lookup operation.
7248 	 * It works because ccts are never freed except during the vhcache
7249 	 * cleanup operation.
7250 	 */
7251 	if (token != NULL &&
7252 	    vhcache->vhcache_clean_time < token->lt_cct_lookup_time)
7253 		return (token->lt_cct);
7254 
7255 	name_addr = vhcache_mknameaddr(ct_name, ct_addr, &len);
7256 	if (mod_hash_find(vhcache->vhcache_client_hash,
7257 	    (mod_hash_key_t)name_addr, &hv) == 0) {
7258 		if (token) {
7259 			token->lt_cct = (mdi_vhcache_client_t *)hv;
7260 			token->lt_cct_lookup_time = lbolt64;
7261 		}
7262 	} else {
7263 		if (token) {
7264 			token->lt_cct = NULL;
7265 			token->lt_cct_lookup_time = 0;
7266 		}
7267 		hv = NULL;
7268 	}
7269 	kmem_free(name_addr, len);
7270 	return ((mdi_vhcache_client_t *)hv);
7271 }
7272 
7273 /*
7274  * Add the specified path to the vhci cache if not already present.
7275  * Also add the vhcache client for the client corresponding to this path
7276  * if it doesn't already exist.
7277  */
7278 static void
7279 vhcache_pi_add(mdi_vhci_config_t *vhc, struct mdi_pathinfo *pip)
7280 {
7281 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7282 	mdi_vhcache_client_t *cct;
7283 	mdi_vhcache_pathinfo_t *cpi;
7284 	mdi_phci_t *ph = pip->pi_phci;
7285 	mdi_client_t *ct = pip->pi_client;
7286 	int cache_updated = 0;
7287 
7288 	rw_enter(&vhcache->vhcache_lock, RW_WRITER);
7289 
7290 	/* if vhcache client for this pip doesn't already exist, add it */
7291 	if ((cct = lookup_vhcache_client(vhcache, ct->ct_drvname, ct->ct_guid,
7292 	    NULL)) == NULL) {
7293 		cct = kmem_zalloc(sizeof (*cct), KM_SLEEP);
7294 		cct->cct_name_addr = vhcache_mknameaddr(ct->ct_drvname,
7295 		    ct->ct_guid, NULL);
7296 		enqueue_vhcache_client(vhcache, cct);
7297 		(void) mod_hash_insert(vhcache->vhcache_client_hash,
7298 		    (mod_hash_key_t)cct->cct_name_addr, (mod_hash_val_t)cct);
7299 		cache_updated = 1;
7300 	}
7301 
7302 	for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi->cpi_next) {
7303 		if (cpi->cpi_cphci->cphci_phci == ph &&
7304 		    strcmp(cpi->cpi_addr, pip->pi_addr) == 0) {
7305 			cpi->cpi_pip = pip;
7306 			if (cpi->cpi_flags & MDI_CPI_HINT_PATH_DOES_NOT_EXIST) {
7307 				cpi->cpi_flags &=
7308 				    ~MDI_CPI_HINT_PATH_DOES_NOT_EXIST;
7309 				sort_vhcache_paths(cct);
7310 				cache_updated = 1;
7311 			}
7312 			break;
7313 		}
7314 	}
7315 
7316 	if (cpi == NULL) {
7317 		cpi = kmem_zalloc(sizeof (*cpi), KM_SLEEP);
7318 		cpi->cpi_addr = i_ddi_strdup(pip->pi_addr, KM_SLEEP);
7319 		cpi->cpi_cphci = lookup_vhcache_phci_by_addr(vhcache, ph);
7320 		ASSERT(cpi->cpi_cphci != NULL);
7321 		cpi->cpi_pip = pip;
7322 		enqueue_vhcache_pathinfo(cct, cpi);
7323 		cache_updated = 1;
7324 	}
7325 
7326 	rw_exit(&vhcache->vhcache_lock);
7327 
7328 	if (cache_updated)
7329 		vhcache_dirty(vhc);
7330 }
7331 
7332 /*
7333  * Remove the reference to the specified path from the vhci cache.
7334  */
7335 static void
7336 vhcache_pi_remove(mdi_vhci_config_t *vhc, struct mdi_pathinfo *pip)
7337 {
7338 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7339 	mdi_client_t *ct = pip->pi_client;
7340 	mdi_vhcache_client_t *cct;
7341 	mdi_vhcache_pathinfo_t *cpi;
7342 
7343 	rw_enter(&vhcache->vhcache_lock, RW_WRITER);
7344 	if ((cct = lookup_vhcache_client(vhcache, ct->ct_drvname, ct->ct_guid,
7345 	    NULL)) != NULL) {
7346 		for (cpi = cct->cct_cpi_head; cpi != NULL;
7347 		    cpi = cpi->cpi_next) {
7348 			if (cpi->cpi_pip == pip) {
7349 				cpi->cpi_pip = NULL;
7350 				break;
7351 			}
7352 		}
7353 	}
7354 	rw_exit(&vhcache->vhcache_lock);
7355 }
7356 
7357 /*
7358  * Flush the vhci cache to disk.
7359  * Returns MDI_SUCCESS on success, MDI_FAILURE on failure.
7360  */
7361 static int
7362 flush_vhcache(mdi_vhci_config_t *vhc, int force_flag)
7363 {
7364 	nvlist_t *nvl;
7365 	int err;
7366 	int rv;
7367 
7368 	/*
7369 	 * It is possible that the system may shutdown before
7370 	 * i_ddi_io_initialized (during stmsboot for example). To allow for
7371 	 * flushing the cache in this case do not check for
7372 	 * i_ddi_io_initialized when force flag is set.
7373 	 */
7374 	if (force_flag == 0 && !i_ddi_io_initialized())
7375 		return (MDI_FAILURE);
7376 
7377 	if ((nvl = vhcache_to_mainnvl(&vhc->vhc_vhcache)) != NULL) {
7378 		err = fwrite_nvlist(vhc->vhc_vhcache_filename, nvl);
7379 		nvlist_free(nvl);
7380 	} else
7381 		err = EFAULT;
7382 
7383 	rv = MDI_SUCCESS;
7384 	mutex_enter(&vhc->vhc_lock);
7385 	if (err != 0) {
7386 		if (err == EROFS) {
7387 			vhc->vhc_flags |= MDI_VHC_READONLY_FS;
7388 			vhc->vhc_flags &= ~(MDI_VHC_VHCACHE_FLUSH_ERROR |
7389 			    MDI_VHC_VHCACHE_DIRTY);
7390 		} else {
7391 			if (!(vhc->vhc_flags & MDI_VHC_VHCACHE_FLUSH_ERROR)) {
7392 				cmn_err(CE_CONT, "%s: update failed\n",
7393 				    vhc->vhc_vhcache_filename);
7394 				vhc->vhc_flags |= MDI_VHC_VHCACHE_FLUSH_ERROR;
7395 			}
7396 			rv = MDI_FAILURE;
7397 		}
7398 	} else if (vhc->vhc_flags & MDI_VHC_VHCACHE_FLUSH_ERROR) {
7399 		cmn_err(CE_CONT,
7400 		    "%s: update now ok\n", vhc->vhc_vhcache_filename);
7401 		vhc->vhc_flags &= ~MDI_VHC_VHCACHE_FLUSH_ERROR;
7402 	}
7403 	mutex_exit(&vhc->vhc_lock);
7404 
7405 	return (rv);
7406 }
7407 
7408 /*
7409  * Call flush_vhcache() to flush the vhci cache at the scheduled time.
7410  * Exits itself if left idle for the idle timeout period.
7411  */
7412 static void
7413 vhcache_flush_thread(void *arg)
7414 {
7415 	mdi_vhci_config_t *vhc = (mdi_vhci_config_t *)arg;
7416 	clock_t idle_time, quit_at_ticks;
7417 	callb_cpr_t cprinfo;
7418 
7419 	/* number of seconds to sleep idle before exiting */
7420 	idle_time = mdi_vhcache_flush_daemon_idle_time * TICKS_PER_SECOND;
7421 
7422 	CALLB_CPR_INIT(&cprinfo, &vhc->vhc_lock, callb_generic_cpr,
7423 	    "mdi_vhcache_flush");
7424 	mutex_enter(&vhc->vhc_lock);
7425 	for (; ; ) {
7426 		while (!(vhc->vhc_flags & MDI_VHC_EXIT) &&
7427 		    (vhc->vhc_flags & MDI_VHC_VHCACHE_DIRTY)) {
7428 			if (ddi_get_lbolt() < vhc->vhc_flush_at_ticks) {
7429 				CALLB_CPR_SAFE_BEGIN(&cprinfo);
7430 				(void) cv_timedwait(&vhc->vhc_cv,
7431 				    &vhc->vhc_lock, vhc->vhc_flush_at_ticks);
7432 				CALLB_CPR_SAFE_END(&cprinfo, &vhc->vhc_lock);
7433 			} else {
7434 				vhc->vhc_flags &= ~MDI_VHC_VHCACHE_DIRTY;
7435 				mutex_exit(&vhc->vhc_lock);
7436 
7437 				if (flush_vhcache(vhc, 0) != MDI_SUCCESS)
7438 					vhcache_dirty(vhc);
7439 
7440 				mutex_enter(&vhc->vhc_lock);
7441 			}
7442 		}
7443 
7444 		quit_at_ticks = ddi_get_lbolt() + idle_time;
7445 
7446 		while (!(vhc->vhc_flags & MDI_VHC_EXIT) &&
7447 		    !(vhc->vhc_flags & MDI_VHC_VHCACHE_DIRTY) &&
7448 		    ddi_get_lbolt() < quit_at_ticks) {
7449 			CALLB_CPR_SAFE_BEGIN(&cprinfo);
7450 			(void) cv_timedwait(&vhc->vhc_cv, &vhc->vhc_lock,
7451 			    quit_at_ticks);
7452 			CALLB_CPR_SAFE_END(&cprinfo, &vhc->vhc_lock);
7453 		}
7454 
7455 		if ((vhc->vhc_flags & MDI_VHC_EXIT) ||
7456 		    !(vhc->vhc_flags & MDI_VHC_VHCACHE_DIRTY))
7457 			goto out;
7458 	}
7459 
7460 out:
7461 	vhc->vhc_flags &= ~MDI_VHC_VHCACHE_FLUSH_THREAD;
7462 	/* CALLB_CPR_EXIT releases the vhc->vhc_lock */
7463 	CALLB_CPR_EXIT(&cprinfo);
7464 }
7465 
7466 /*
7467  * Make vhci cache dirty and schedule flushing by vhcache flush thread.
7468  */
7469 static void
7470 vhcache_dirty(mdi_vhci_config_t *vhc)
7471 {
7472 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7473 	int create_thread;
7474 
7475 	rw_enter(&vhcache->vhcache_lock, RW_READER);
7476 	/* do not flush cache until the cache is fully built */
7477 	if (!(vhcache->vhcache_flags & MDI_VHCI_CACHE_SETUP_DONE)) {
7478 		rw_exit(&vhcache->vhcache_lock);
7479 		return;
7480 	}
7481 	rw_exit(&vhcache->vhcache_lock);
7482 
7483 	mutex_enter(&vhc->vhc_lock);
7484 	if (vhc->vhc_flags & MDI_VHC_READONLY_FS) {
7485 		mutex_exit(&vhc->vhc_lock);
7486 		return;
7487 	}
7488 
7489 	vhc->vhc_flags |= MDI_VHC_VHCACHE_DIRTY;
7490 	vhc->vhc_flush_at_ticks = ddi_get_lbolt() +
7491 	    mdi_vhcache_flush_delay * TICKS_PER_SECOND;
7492 	if (vhc->vhc_flags & MDI_VHC_VHCACHE_FLUSH_THREAD) {
7493 		cv_broadcast(&vhc->vhc_cv);
7494 		create_thread = 0;
7495 	} else {
7496 		vhc->vhc_flags |= MDI_VHC_VHCACHE_FLUSH_THREAD;
7497 		create_thread = 1;
7498 	}
7499 	mutex_exit(&vhc->vhc_lock);
7500 
7501 	if (create_thread)
7502 		(void) thread_create(NULL, 0, vhcache_flush_thread, vhc,
7503 		    0, &p0, TS_RUN, minclsyspri);
7504 }
7505 
7506 /*
7507  * phci bus config structure - one for for each phci bus config operation that
7508  * we initiate on behalf of a vhci.
7509  */
7510 typedef struct mdi_phci_bus_config_s {
7511 	char *phbc_phci_path;
7512 	struct mdi_vhci_bus_config_s *phbc_vhbusconfig;	/* vhci bus config */
7513 	struct mdi_phci_bus_config_s *phbc_next;
7514 } mdi_phci_bus_config_t;
7515 
7516 /* vhci bus config structure - one for each vhci bus config operation */
7517 typedef struct mdi_vhci_bus_config_s {
7518 	ddi_bus_config_op_t vhbc_op;	/* bus config op */
7519 	major_t vhbc_op_major;		/* bus config op major */
7520 	uint_t vhbc_op_flags;		/* bus config op flags */
7521 	kmutex_t vhbc_lock;
7522 	kcondvar_t vhbc_cv;
7523 	int vhbc_thr_count;
7524 } mdi_vhci_bus_config_t;
7525 
7526 /*
7527  * bus config the specified phci
7528  */
7529 static void
7530 bus_config_phci(void *arg)
7531 {
7532 	mdi_phci_bus_config_t *phbc = (mdi_phci_bus_config_t *)arg;
7533 	mdi_vhci_bus_config_t *vhbc = phbc->phbc_vhbusconfig;
7534 	dev_info_t *ph_dip;
7535 
7536 	/*
7537 	 * first configure all path components upto phci and then configure
7538 	 * the phci children.
7539 	 */
7540 	if ((ph_dip = e_ddi_hold_devi_by_path(phbc->phbc_phci_path, 0))
7541 	    != NULL) {
7542 		if (vhbc->vhbc_op == BUS_CONFIG_DRIVER ||
7543 		    vhbc->vhbc_op == BUS_UNCONFIG_DRIVER) {
7544 			(void) ndi_devi_config_driver(ph_dip,
7545 			    vhbc->vhbc_op_flags,
7546 			    vhbc->vhbc_op_major);
7547 		} else
7548 			(void) ndi_devi_config(ph_dip,
7549 			    vhbc->vhbc_op_flags);
7550 
7551 		/* release the hold that e_ddi_hold_devi_by_path() placed */
7552 		ndi_rele_devi(ph_dip);
7553 	}
7554 
7555 	kmem_free(phbc->phbc_phci_path, strlen(phbc->phbc_phci_path) + 1);
7556 	kmem_free(phbc, sizeof (*phbc));
7557 
7558 	mutex_enter(&vhbc->vhbc_lock);
7559 	vhbc->vhbc_thr_count--;
7560 	if (vhbc->vhbc_thr_count == 0)
7561 		cv_broadcast(&vhbc->vhbc_cv);
7562 	mutex_exit(&vhbc->vhbc_lock);
7563 }
7564 
7565 /*
7566  * Bus config all phcis associated with the vhci in parallel.
7567  * op must be BUS_CONFIG_DRIVER or BUS_CONFIG_ALL.
7568  */
7569 static void
7570 bus_config_all_phcis(mdi_vhci_cache_t *vhcache, uint_t flags,
7571     ddi_bus_config_op_t op, major_t maj)
7572 {
7573 	mdi_phci_bus_config_t *phbc_head = NULL, *phbc, *phbc_next;
7574 	mdi_vhci_bus_config_t *vhbc;
7575 	mdi_vhcache_phci_t *cphci;
7576 
7577 	rw_enter(&vhcache->vhcache_lock, RW_READER);
7578 	if (vhcache->vhcache_phci_head == NULL) {
7579 		rw_exit(&vhcache->vhcache_lock);
7580 		return;
7581 	}
7582 
7583 	vhbc = kmem_zalloc(sizeof (*vhbc), KM_SLEEP);
7584 
7585 	for (cphci = vhcache->vhcache_phci_head; cphci != NULL;
7586 	    cphci = cphci->cphci_next) {
7587 		phbc = kmem_zalloc(sizeof (*phbc), KM_SLEEP);
7588 		phbc->phbc_phci_path = i_ddi_strdup(cphci->cphci_path,
7589 		    KM_SLEEP);
7590 		phbc->phbc_vhbusconfig = vhbc;
7591 		phbc->phbc_next = phbc_head;
7592 		phbc_head = phbc;
7593 		vhbc->vhbc_thr_count++;
7594 	}
7595 	rw_exit(&vhcache->vhcache_lock);
7596 
7597 	vhbc->vhbc_op = op;
7598 	vhbc->vhbc_op_major = maj;
7599 	vhbc->vhbc_op_flags = NDI_NO_EVENT |
7600 	    (flags & (NDI_CONFIG_REPROBE | NDI_DRV_CONF_REPROBE));
7601 	mutex_init(&vhbc->vhbc_lock, NULL, MUTEX_DEFAULT, NULL);
7602 	cv_init(&vhbc->vhbc_cv, NULL, CV_DRIVER, NULL);
7603 
7604 	/* now create threads to initiate bus config on all phcis in parallel */
7605 	for (phbc = phbc_head; phbc != NULL; phbc = phbc_next) {
7606 		phbc_next = phbc->phbc_next;
7607 		if (mdi_mtc_off)
7608 			bus_config_phci((void *)phbc);
7609 		else
7610 			(void) thread_create(NULL, 0, bus_config_phci, phbc,
7611 			    0, &p0, TS_RUN, minclsyspri);
7612 	}
7613 
7614 	mutex_enter(&vhbc->vhbc_lock);
7615 	/* wait until all threads exit */
7616 	while (vhbc->vhbc_thr_count > 0)
7617 		cv_wait(&vhbc->vhbc_cv, &vhbc->vhbc_lock);
7618 	mutex_exit(&vhbc->vhbc_lock);
7619 
7620 	mutex_destroy(&vhbc->vhbc_lock);
7621 	cv_destroy(&vhbc->vhbc_cv);
7622 	kmem_free(vhbc, sizeof (*vhbc));
7623 }
7624 
7625 /*
7626  * Single threaded version of bus_config_all_phcis()
7627  */
7628 static void
7629 st_bus_config_all_phcis(mdi_vhci_config_t *vhc, uint_t flags,
7630     ddi_bus_config_op_t op, major_t maj)
7631 {
7632 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7633 
7634 	single_threaded_vhconfig_enter(vhc);
7635 	bus_config_all_phcis(vhcache, flags, op, maj);
7636 	single_threaded_vhconfig_exit(vhc);
7637 }
7638 
7639 /*
7640  * Perform BUS_CONFIG_ONE on the specified child of the phci.
7641  * The path includes the child component in addition to the phci path.
7642  */
7643 static int
7644 bus_config_one_phci_child(char *path)
7645 {
7646 	dev_info_t *ph_dip, *child;
7647 	char *devnm;
7648 	int rv = MDI_FAILURE;
7649 
7650 	/* extract the child component of the phci */
7651 	devnm = strrchr(path, '/');
7652 	*devnm++ = '\0';
7653 
7654 	/*
7655 	 * first configure all path components upto phci and then
7656 	 * configure the phci child.
7657 	 */
7658 	if ((ph_dip = e_ddi_hold_devi_by_path(path, 0)) != NULL) {
7659 		if (ndi_devi_config_one(ph_dip, devnm, &child, NDI_NO_EVENT) ==
7660 		    NDI_SUCCESS) {
7661 			/*
7662 			 * release the hold that ndi_devi_config_one() placed
7663 			 */
7664 			ndi_rele_devi(child);
7665 			rv = MDI_SUCCESS;
7666 		}
7667 
7668 		/* release the hold that e_ddi_hold_devi_by_path() placed */
7669 		ndi_rele_devi(ph_dip);
7670 	}
7671 
7672 	devnm--;
7673 	*devnm = '/';
7674 	return (rv);
7675 }
7676 
7677 /*
7678  * Build a list of phci client paths for the specified vhci client.
7679  * The list includes only those phci client paths which aren't configured yet.
7680  */
7681 static mdi_phys_path_t *
7682 build_phclient_path_list(mdi_vhcache_client_t *cct, char *ct_name)
7683 {
7684 	mdi_vhcache_pathinfo_t *cpi;
7685 	mdi_phys_path_t *pp_head = NULL, *pp_tail = NULL, *pp;
7686 	int config_path, len;
7687 
7688 	for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi->cpi_next) {
7689 		/*
7690 		 * include only those paths that aren't configured.
7691 		 */
7692 		config_path = 0;
7693 		if (cpi->cpi_pip == NULL)
7694 			config_path = 1;
7695 		else {
7696 			MDI_PI_LOCK(cpi->cpi_pip);
7697 			if (MDI_PI_IS_INIT(cpi->cpi_pip))
7698 				config_path = 1;
7699 			MDI_PI_UNLOCK(cpi->cpi_pip);
7700 		}
7701 
7702 		if (config_path) {
7703 			pp = kmem_alloc(sizeof (*pp), KM_SLEEP);
7704 			len = strlen(cpi->cpi_cphci->cphci_path) +
7705 			    strlen(ct_name) + strlen(cpi->cpi_addr) + 3;
7706 			pp->phys_path = kmem_alloc(len, KM_SLEEP);
7707 			(void) snprintf(pp->phys_path, len, "%s/%s@%s",
7708 			    cpi->cpi_cphci->cphci_path, ct_name,
7709 			    cpi->cpi_addr);
7710 			pp->phys_path_next = NULL;
7711 
7712 			if (pp_head == NULL)
7713 				pp_head = pp;
7714 			else
7715 				pp_tail->phys_path_next = pp;
7716 			pp_tail = pp;
7717 		}
7718 	}
7719 
7720 	return (pp_head);
7721 }
7722 
7723 /*
7724  * Free the memory allocated for phci client path list.
7725  */
7726 static void
7727 free_phclient_path_list(mdi_phys_path_t *pp_head)
7728 {
7729 	mdi_phys_path_t *pp, *pp_next;
7730 
7731 	for (pp = pp_head; pp != NULL; pp = pp_next) {
7732 		pp_next = pp->phys_path_next;
7733 		kmem_free(pp->phys_path, strlen(pp->phys_path) + 1);
7734 		kmem_free(pp, sizeof (*pp));
7735 	}
7736 }
7737 
7738 /*
7739  * Allocated async client structure and initialize with the specified values.
7740  */
7741 static mdi_async_client_config_t *
7742 alloc_async_client_config(char *ct_name, char *ct_addr,
7743     mdi_phys_path_t *pp_head, mdi_vhcache_lookup_token_t *tok)
7744 {
7745 	mdi_async_client_config_t *acc;
7746 
7747 	acc = kmem_alloc(sizeof (*acc), KM_SLEEP);
7748 	acc->acc_ct_name = i_ddi_strdup(ct_name, KM_SLEEP);
7749 	acc->acc_ct_addr = i_ddi_strdup(ct_addr, KM_SLEEP);
7750 	acc->acc_phclient_path_list_head = pp_head;
7751 	init_vhcache_lookup_token(&acc->acc_token, tok);
7752 	acc->acc_next = NULL;
7753 	return (acc);
7754 }
7755 
7756 /*
7757  * Free the memory allocated for the async client structure and their members.
7758  */
7759 static void
7760 free_async_client_config(mdi_async_client_config_t *acc)
7761 {
7762 	if (acc->acc_phclient_path_list_head)
7763 		free_phclient_path_list(acc->acc_phclient_path_list_head);
7764 	kmem_free(acc->acc_ct_name, strlen(acc->acc_ct_name) + 1);
7765 	kmem_free(acc->acc_ct_addr, strlen(acc->acc_ct_addr) + 1);
7766 	kmem_free(acc, sizeof (*acc));
7767 }
7768 
7769 /*
7770  * Sort vhcache pathinfos (cpis) of the specified client.
7771  * All cpis which do not have MDI_CPI_HINT_PATH_DOES_NOT_EXIST
7772  * flag set come at the beginning of the list. All cpis which have this
7773  * flag set come at the end of the list.
7774  */
7775 static void
7776 sort_vhcache_paths(mdi_vhcache_client_t *cct)
7777 {
7778 	mdi_vhcache_pathinfo_t *cpi, *cpi_next, *cpi_head;
7779 
7780 	cpi_head = cct->cct_cpi_head;
7781 	cct->cct_cpi_head = cct->cct_cpi_tail = NULL;
7782 	for (cpi = cpi_head; cpi != NULL; cpi = cpi_next) {
7783 		cpi_next = cpi->cpi_next;
7784 		enqueue_vhcache_pathinfo(cct, cpi);
7785 	}
7786 }
7787 
7788 /*
7789  * Verify whether MDI_CPI_HINT_PATH_DOES_NOT_EXIST flag setting is correct for
7790  * every vhcache pathinfo of the specified client. If not adjust the flag
7791  * setting appropriately.
7792  *
7793  * Note that MDI_CPI_HINT_PATH_DOES_NOT_EXIST flag is persisted in the
7794  * on-disk vhci cache. So every time this flag is updated the cache must be
7795  * flushed.
7796  */
7797 static void
7798 adjust_sort_vhcache_paths(mdi_vhci_config_t *vhc, char *ct_name, char *ct_addr,
7799     mdi_vhcache_lookup_token_t *tok)
7800 {
7801 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7802 	mdi_vhcache_client_t *cct;
7803 	mdi_vhcache_pathinfo_t *cpi;
7804 
7805 	rw_enter(&vhcache->vhcache_lock, RW_READER);
7806 	if ((cct = lookup_vhcache_client(vhcache, ct_name, ct_addr, tok))
7807 	    == NULL) {
7808 		rw_exit(&vhcache->vhcache_lock);
7809 		return;
7810 	}
7811 
7812 	/*
7813 	 * to avoid unnecessary on-disk cache updates, first check if an
7814 	 * update is really needed. If no update is needed simply return.
7815 	 */
7816 	for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi->cpi_next) {
7817 		if ((cpi->cpi_pip != NULL &&
7818 		    (cpi->cpi_flags & MDI_CPI_HINT_PATH_DOES_NOT_EXIST)) ||
7819 		    (cpi->cpi_pip == NULL &&
7820 		    !(cpi->cpi_flags & MDI_CPI_HINT_PATH_DOES_NOT_EXIST))) {
7821 			break;
7822 		}
7823 	}
7824 	if (cpi == NULL) {
7825 		rw_exit(&vhcache->vhcache_lock);
7826 		return;
7827 	}
7828 
7829 	if (rw_tryupgrade(&vhcache->vhcache_lock) == 0) {
7830 		rw_exit(&vhcache->vhcache_lock);
7831 		rw_enter(&vhcache->vhcache_lock, RW_WRITER);
7832 		if ((cct = lookup_vhcache_client(vhcache, ct_name, ct_addr,
7833 		    tok)) == NULL) {
7834 			rw_exit(&vhcache->vhcache_lock);
7835 			return;
7836 		}
7837 	}
7838 
7839 	for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi->cpi_next) {
7840 		if (cpi->cpi_pip != NULL)
7841 			cpi->cpi_flags &= ~MDI_CPI_HINT_PATH_DOES_NOT_EXIST;
7842 		else
7843 			cpi->cpi_flags |= MDI_CPI_HINT_PATH_DOES_NOT_EXIST;
7844 	}
7845 	sort_vhcache_paths(cct);
7846 
7847 	rw_exit(&vhcache->vhcache_lock);
7848 	vhcache_dirty(vhc);
7849 }
7850 
7851 /*
7852  * Configure all specified paths of the client.
7853  */
7854 static void
7855 config_client_paths_sync(mdi_vhci_config_t *vhc, char *ct_name, char *ct_addr,
7856     mdi_phys_path_t *pp_head, mdi_vhcache_lookup_token_t *tok)
7857 {
7858 	mdi_phys_path_t *pp;
7859 
7860 	for (pp = pp_head; pp != NULL; pp = pp->phys_path_next)
7861 		(void) bus_config_one_phci_child(pp->phys_path);
7862 	adjust_sort_vhcache_paths(vhc, ct_name, ct_addr, tok);
7863 }
7864 
7865 /*
7866  * Dequeue elements from vhci async client config list and bus configure
7867  * their corresponding phci clients.
7868  */
7869 static void
7870 config_client_paths_thread(void *arg)
7871 {
7872 	mdi_vhci_config_t *vhc = (mdi_vhci_config_t *)arg;
7873 	mdi_async_client_config_t *acc;
7874 	clock_t quit_at_ticks;
7875 	clock_t idle_time = mdi_async_config_idle_time * TICKS_PER_SECOND;
7876 	callb_cpr_t cprinfo;
7877 
7878 	CALLB_CPR_INIT(&cprinfo, &vhc->vhc_lock, callb_generic_cpr,
7879 	    "mdi_config_client_paths");
7880 
7881 	for (; ; ) {
7882 		quit_at_ticks = ddi_get_lbolt() + idle_time;
7883 
7884 		mutex_enter(&vhc->vhc_lock);
7885 		while (!(vhc->vhc_flags & MDI_VHC_EXIT) &&
7886 		    vhc->vhc_acc_list_head == NULL &&
7887 		    ddi_get_lbolt() < quit_at_ticks) {
7888 			CALLB_CPR_SAFE_BEGIN(&cprinfo);
7889 			(void) cv_timedwait(&vhc->vhc_cv, &vhc->vhc_lock,
7890 			    quit_at_ticks);
7891 			CALLB_CPR_SAFE_END(&cprinfo, &vhc->vhc_lock);
7892 		}
7893 
7894 		if ((vhc->vhc_flags & MDI_VHC_EXIT) ||
7895 		    vhc->vhc_acc_list_head == NULL)
7896 			goto out;
7897 
7898 		acc = vhc->vhc_acc_list_head;
7899 		vhc->vhc_acc_list_head = acc->acc_next;
7900 		if (vhc->vhc_acc_list_head == NULL)
7901 			vhc->vhc_acc_list_tail = NULL;
7902 		vhc->vhc_acc_count--;
7903 		mutex_exit(&vhc->vhc_lock);
7904 
7905 		config_client_paths_sync(vhc, acc->acc_ct_name,
7906 		    acc->acc_ct_addr, acc->acc_phclient_path_list_head,
7907 		    &acc->acc_token);
7908 
7909 		free_async_client_config(acc);
7910 	}
7911 
7912 out:
7913 	vhc->vhc_acc_thrcount--;
7914 	/* CALLB_CPR_EXIT releases the vhc->vhc_lock */
7915 	CALLB_CPR_EXIT(&cprinfo);
7916 }
7917 
7918 /*
7919  * Arrange for all the phci client paths (pp_head) for the specified client
7920  * to be bus configured asynchronously by a thread.
7921  */
7922 static void
7923 config_client_paths_async(mdi_vhci_config_t *vhc, char *ct_name, char *ct_addr,
7924     mdi_phys_path_t *pp_head, mdi_vhcache_lookup_token_t *tok)
7925 {
7926 	mdi_async_client_config_t *acc, *newacc;
7927 	int create_thread;
7928 
7929 	if (pp_head == NULL)
7930 		return;
7931 
7932 	if (mdi_mtc_off) {
7933 		config_client_paths_sync(vhc, ct_name, ct_addr, pp_head, tok);
7934 		free_phclient_path_list(pp_head);
7935 		return;
7936 	}
7937 
7938 	newacc = alloc_async_client_config(ct_name, ct_addr, pp_head, tok);
7939 	ASSERT(newacc);
7940 
7941 	mutex_enter(&vhc->vhc_lock);
7942 	for (acc = vhc->vhc_acc_list_head; acc != NULL; acc = acc->acc_next) {
7943 		if (strcmp(ct_name, acc->acc_ct_name) == 0 &&
7944 		    strcmp(ct_addr, acc->acc_ct_addr) == 0) {
7945 			free_async_client_config(newacc);
7946 			mutex_exit(&vhc->vhc_lock);
7947 			return;
7948 		}
7949 	}
7950 
7951 	if (vhc->vhc_acc_list_head == NULL)
7952 		vhc->vhc_acc_list_head = newacc;
7953 	else
7954 		vhc->vhc_acc_list_tail->acc_next = newacc;
7955 	vhc->vhc_acc_list_tail = newacc;
7956 	vhc->vhc_acc_count++;
7957 	if (vhc->vhc_acc_count <= vhc->vhc_acc_thrcount) {
7958 		cv_broadcast(&vhc->vhc_cv);
7959 		create_thread = 0;
7960 	} else {
7961 		vhc->vhc_acc_thrcount++;
7962 		create_thread = 1;
7963 	}
7964 	mutex_exit(&vhc->vhc_lock);
7965 
7966 	if (create_thread)
7967 		(void) thread_create(NULL, 0, config_client_paths_thread, vhc,
7968 		    0, &p0, TS_RUN, minclsyspri);
7969 }
7970 
7971 /*
7972  * Return number of online paths for the specified client.
7973  */
7974 static int
7975 nonline_paths(mdi_vhcache_client_t *cct)
7976 {
7977 	mdi_vhcache_pathinfo_t *cpi;
7978 	int online_count = 0;
7979 
7980 	for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi->cpi_next) {
7981 		if (cpi->cpi_pip != NULL) {
7982 			MDI_PI_LOCK(cpi->cpi_pip);
7983 			if (cpi->cpi_pip->pi_state == MDI_PATHINFO_STATE_ONLINE)
7984 				online_count++;
7985 			MDI_PI_UNLOCK(cpi->cpi_pip);
7986 		}
7987 	}
7988 
7989 	return (online_count);
7990 }
7991 
7992 /*
7993  * Bus configure all paths for the specified vhci client.
7994  * If at least one path for the client is already online, the remaining paths
7995  * will be configured asynchronously. Otherwise, it synchronously configures
7996  * the paths until at least one path is online and then rest of the paths
7997  * will be configured asynchronously.
7998  */
7999 static void
8000 config_client_paths(mdi_vhci_config_t *vhc, char *ct_name, char *ct_addr)
8001 {
8002 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
8003 	mdi_phys_path_t *pp_head, *pp;
8004 	mdi_vhcache_client_t *cct;
8005 	mdi_vhcache_lookup_token_t tok;
8006 
8007 	ASSERT(RW_LOCK_HELD(&vhcache->vhcache_lock));
8008 
8009 	init_vhcache_lookup_token(&tok, NULL);
8010 
8011 	if (ct_name == NULL || ct_addr == NULL ||
8012 	    (cct = lookup_vhcache_client(vhcache, ct_name, ct_addr, &tok))
8013 	    == NULL ||
8014 	    (pp_head = build_phclient_path_list(cct, ct_name)) == NULL) {
8015 		rw_exit(&vhcache->vhcache_lock);
8016 		return;
8017 	}
8018 
8019 	/* if at least one path is online, configure the rest asynchronously */
8020 	if (nonline_paths(cct) > 0) {
8021 		rw_exit(&vhcache->vhcache_lock);
8022 		config_client_paths_async(vhc, ct_name, ct_addr, pp_head, &tok);
8023 		return;
8024 	}
8025 
8026 	rw_exit(&vhcache->vhcache_lock);
8027 
8028 	for (pp = pp_head; pp != NULL; pp = pp->phys_path_next) {
8029 		if (bus_config_one_phci_child(pp->phys_path) == MDI_SUCCESS) {
8030 			rw_enter(&vhcache->vhcache_lock, RW_READER);
8031 
8032 			if ((cct = lookup_vhcache_client(vhcache, ct_name,
8033 			    ct_addr, &tok)) == NULL) {
8034 				rw_exit(&vhcache->vhcache_lock);
8035 				goto out;
8036 			}
8037 
8038 			if (nonline_paths(cct) > 0 &&
8039 			    pp->phys_path_next != NULL) {
8040 				rw_exit(&vhcache->vhcache_lock);
8041 				config_client_paths_async(vhc, ct_name, ct_addr,
8042 				    pp->phys_path_next, &tok);
8043 				pp->phys_path_next = NULL;
8044 				goto out;
8045 			}
8046 
8047 			rw_exit(&vhcache->vhcache_lock);
8048 		}
8049 	}
8050 
8051 	adjust_sort_vhcache_paths(vhc, ct_name, ct_addr, &tok);
8052 out:
8053 	free_phclient_path_list(pp_head);
8054 }
8055 
8056 static void
8057 single_threaded_vhconfig_enter(mdi_vhci_config_t *vhc)
8058 {
8059 	mutex_enter(&vhc->vhc_lock);
8060 	while (vhc->vhc_flags & MDI_VHC_SINGLE_THREADED)
8061 		cv_wait(&vhc->vhc_cv, &vhc->vhc_lock);
8062 	vhc->vhc_flags |= MDI_VHC_SINGLE_THREADED;
8063 	mutex_exit(&vhc->vhc_lock);
8064 }
8065 
8066 static void
8067 single_threaded_vhconfig_exit(mdi_vhci_config_t *vhc)
8068 {
8069 	mutex_enter(&vhc->vhc_lock);
8070 	vhc->vhc_flags &= ~MDI_VHC_SINGLE_THREADED;
8071 	cv_broadcast(&vhc->vhc_cv);
8072 	mutex_exit(&vhc->vhc_lock);
8073 }
8074 
8075 typedef struct mdi_phci_driver_info {
8076 	char	*phdriver_name;	/* name of the phci driver */
8077 
8078 	/* set to non zero if the phci driver supports root device */
8079 	int	phdriver_root_support;
8080 } mdi_phci_driver_info_t;
8081 
8082 /*
8083  * vhci class and root support capability of a phci driver can be
8084  * specified using ddi-vhci-class and ddi-no-root-support properties in the
8085  * phci driver.conf file. The built-in tables below contain this information
8086  * for those phci drivers whose driver.conf files don't yet contain this info.
8087  *
8088  * All phci drivers expect iscsi have root device support.
8089  */
8090 static mdi_phci_driver_info_t scsi_phci_driver_list[] = {
8091 	{ "fp", 1 },
8092 	{ "iscsi", 0 },
8093 	{ "ibsrp", 1 }
8094 	};
8095 
8096 static mdi_phci_driver_info_t ib_phci_driver_list[] = { "tavor", 1 };
8097 
8098 static void *
8099 mdi_realloc(void *old_ptr, size_t old_size, size_t new_size)
8100 {
8101 	void *new_ptr;
8102 
8103 	new_ptr = kmem_zalloc(new_size, KM_SLEEP);
8104 	if (old_ptr) {
8105 		bcopy(old_ptr, new_ptr, MIN(old_size, new_size));
8106 		kmem_free(old_ptr, old_size);
8107 	}
8108 	return (new_ptr);
8109 }
8110 
8111 static void
8112 add_to_phci_list(char ***driver_list, int **root_support_list,
8113     int *cur_elements, int *max_elements, char *driver_name, int root_support)
8114 {
8115 	ASSERT(*cur_elements <= *max_elements);
8116 	if (*cur_elements == *max_elements) {
8117 		*max_elements += 10;
8118 		*driver_list = mdi_realloc(*driver_list,
8119 		    sizeof (char *) * (*cur_elements),
8120 		    sizeof (char *) * (*max_elements));
8121 		*root_support_list = mdi_realloc(*root_support_list,
8122 		    sizeof (int) * (*cur_elements),
8123 		    sizeof (int) * (*max_elements));
8124 	}
8125 	(*driver_list)[*cur_elements] = i_ddi_strdup(driver_name, KM_SLEEP);
8126 	(*root_support_list)[*cur_elements] = root_support;
8127 	(*cur_elements)++;
8128 }
8129 
8130 static void
8131 get_phci_driver_list(char *vhci_class, char ***driver_list,
8132     int **root_support_list, int *cur_elements, int *max_elements)
8133 {
8134 	mdi_phci_driver_info_t	*st_driver_list, *p;
8135 	int		st_ndrivers, root_support, i, j, driver_conf_count;
8136 	major_t		m;
8137 	struct devnames	*dnp;
8138 	ddi_prop_t	*propp;
8139 
8140 	*driver_list = NULL;
8141 	*root_support_list = NULL;
8142 	*cur_elements = 0;
8143 	*max_elements = 0;
8144 
8145 	/* add the phci drivers derived from the phci driver.conf files */
8146 	for (m = 0; m < devcnt; m++) {
8147 		dnp = &devnamesp[m];
8148 
8149 		if (dnp->dn_flags & DN_PHCI_DRIVER) {
8150 			LOCK_DEV_OPS(&dnp->dn_lock);
8151 			if (dnp->dn_global_prop_ptr != NULL &&
8152 			    (propp = i_ddi_prop_search(DDI_DEV_T_ANY,
8153 			    DDI_VHCI_CLASS, DDI_PROP_TYPE_STRING,
8154 			    &dnp->dn_global_prop_ptr->prop_list)) != NULL &&
8155 			    strcmp(propp->prop_val, vhci_class) == 0) {
8156 
8157 				root_support = (i_ddi_prop_search(DDI_DEV_T_ANY,
8158 				    DDI_NO_ROOT_SUPPORT, DDI_PROP_TYPE_INT,
8159 				    &dnp->dn_global_prop_ptr->prop_list)
8160 				    == NULL) ? 1 : 0;
8161 
8162 				add_to_phci_list(driver_list, root_support_list,
8163 				    cur_elements, max_elements, dnp->dn_name,
8164 				    root_support);
8165 
8166 				UNLOCK_DEV_OPS(&dnp->dn_lock);
8167 			} else
8168 				UNLOCK_DEV_OPS(&dnp->dn_lock);
8169 		}
8170 	}
8171 
8172 	driver_conf_count = *cur_elements;
8173 
8174 	/* add the phci drivers specified in the built-in tables */
8175 	if (strcmp(vhci_class, MDI_HCI_CLASS_SCSI) == 0) {
8176 		st_driver_list = scsi_phci_driver_list;
8177 		st_ndrivers = sizeof (scsi_phci_driver_list) /
8178 		    sizeof (mdi_phci_driver_info_t);
8179 	} else if (strcmp(vhci_class, MDI_HCI_CLASS_IB) == 0) {
8180 		st_driver_list = ib_phci_driver_list;
8181 		st_ndrivers = sizeof (ib_phci_driver_list) /
8182 		    sizeof (mdi_phci_driver_info_t);
8183 	} else {
8184 		st_driver_list = NULL;
8185 		st_ndrivers = 0;
8186 	}
8187 
8188 	for (i = 0, p = st_driver_list; i < st_ndrivers; i++, p++) {
8189 		/* add this phci driver if not already added before */
8190 		for (j = 0; j < driver_conf_count; j++) {
8191 			if (strcmp((*driver_list)[j], p->phdriver_name) == 0)
8192 				break;
8193 		}
8194 		if (j == driver_conf_count) {
8195 			add_to_phci_list(driver_list, root_support_list,
8196 			    cur_elements, max_elements, p->phdriver_name,
8197 			    p->phdriver_root_support);
8198 		}
8199 	}
8200 }
8201 
8202 /*
8203  * Attach the phci driver instances associated with the specified vhci class.
8204  * If root is mounted attach all phci driver instances.
8205  * If root is not mounted, attach the instances of only those phci
8206  * drivers that have the root support.
8207  */
8208 static void
8209 attach_phci_drivers(char *vhci_class)
8210 {
8211 	char	**driver_list, **p;
8212 	int	*root_support_list;
8213 	int	cur_elements, max_elements, i;
8214 	major_t	m;
8215 
8216 	get_phci_driver_list(vhci_class, &driver_list, &root_support_list,
8217 	    &cur_elements, &max_elements);
8218 
8219 	for (i = 0; i < cur_elements; i++) {
8220 		if (modrootloaded || root_support_list[i]) {
8221 			m = ddi_name_to_major(driver_list[i]);
8222 			if (m != (major_t)-1 && ddi_hold_installed_driver(m))
8223 				ddi_rele_driver(m);
8224 		}
8225 	}
8226 
8227 	if (driver_list) {
8228 		for (i = 0, p = driver_list; i < cur_elements; i++, p++)
8229 			kmem_free(*p, strlen(*p) + 1);
8230 		kmem_free(driver_list, sizeof (char *) * max_elements);
8231 		kmem_free(root_support_list, sizeof (int) * max_elements);
8232 	}
8233 }
8234 
8235 /*
8236  * Build vhci cache:
8237  *
8238  * Attach phci driver instances and then drive BUS_CONFIG_ALL on
8239  * the phci driver instances. During this process the cache gets built.
8240  *
8241  * Cache is built fully if the root is mounted.
8242  * If the root is not mounted, phci drivers that do not have root support
8243  * are not attached. As a result the cache is built partially. The entries
8244  * in the cache reflect only those phci drivers that have root support.
8245  */
8246 static int
8247 build_vhci_cache(mdi_vhci_t *vh)
8248 {
8249 	mdi_vhci_config_t *vhc = vh->vh_config;
8250 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
8251 
8252 	single_threaded_vhconfig_enter(vhc);
8253 
8254 	rw_enter(&vhcache->vhcache_lock, RW_READER);
8255 	if (vhcache->vhcache_flags & MDI_VHCI_CACHE_SETUP_DONE) {
8256 		rw_exit(&vhcache->vhcache_lock);
8257 		single_threaded_vhconfig_exit(vhc);
8258 		return (0);
8259 	}
8260 	rw_exit(&vhcache->vhcache_lock);
8261 
8262 	attach_phci_drivers(vh->vh_class);
8263 	bus_config_all_phcis(vhcache, NDI_DRV_CONF_REPROBE | NDI_NO_EVENT,
8264 	    BUS_CONFIG_ALL, (major_t)-1);
8265 
8266 	rw_enter(&vhcache->vhcache_lock, RW_WRITER);
8267 	vhcache->vhcache_flags |= MDI_VHCI_CACHE_SETUP_DONE;
8268 	rw_exit(&vhcache->vhcache_lock);
8269 
8270 	single_threaded_vhconfig_exit(vhc);
8271 	vhcache_dirty(vhc);
8272 	return (1);
8273 }
8274 
8275 /*
8276  * Determine if discovery of paths is needed.
8277  */
8278 static int
8279 vhcache_do_discovery(mdi_vhci_config_t *vhc)
8280 {
8281 	int rv = 1;
8282 
8283 	mutex_enter(&vhc->vhc_lock);
8284 	if (i_ddi_io_initialized() == 0) {
8285 		if (vhc->vhc_path_discovery_boot > 0) {
8286 			vhc->vhc_path_discovery_boot--;
8287 			goto out;
8288 		}
8289 	} else {
8290 		if (vhc->vhc_path_discovery_postboot > 0) {
8291 			vhc->vhc_path_discovery_postboot--;
8292 			goto out;
8293 		}
8294 	}
8295 
8296 	/*
8297 	 * Do full path discovery at most once per mdi_path_discovery_interval.
8298 	 * This is to avoid a series of full path discoveries when opening
8299 	 * stale /dev/[r]dsk links.
8300 	 */
8301 	if (mdi_path_discovery_interval != -1 &&
8302 	    lbolt64 >= vhc->vhc_path_discovery_cutoff_time)
8303 		goto out;
8304 
8305 	rv = 0;
8306 out:
8307 	mutex_exit(&vhc->vhc_lock);
8308 	return (rv);
8309 }
8310 
8311 /*
8312  * Discover all paths:
8313  *
8314  * Attach phci driver instances and then drive BUS_CONFIG_ALL on all the phci
8315  * driver instances. During this process all paths will be discovered.
8316  */
8317 static int
8318 vhcache_discover_paths(mdi_vhci_t *vh)
8319 {
8320 	mdi_vhci_config_t *vhc = vh->vh_config;
8321 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
8322 	int rv = 0;
8323 
8324 	single_threaded_vhconfig_enter(vhc);
8325 
8326 	if (vhcache_do_discovery(vhc)) {
8327 		attach_phci_drivers(vh->vh_class);
8328 		bus_config_all_phcis(vhcache, NDI_DRV_CONF_REPROBE |
8329 		    NDI_NO_EVENT, BUS_CONFIG_ALL, (major_t)-1);
8330 
8331 		mutex_enter(&vhc->vhc_lock);
8332 		vhc->vhc_path_discovery_cutoff_time = lbolt64 +
8333 		    mdi_path_discovery_interval * TICKS_PER_SECOND;
8334 		mutex_exit(&vhc->vhc_lock);
8335 		rv = 1;
8336 	}
8337 
8338 	single_threaded_vhconfig_exit(vhc);
8339 	return (rv);
8340 }
8341 
8342 /*
8343  * Generic vhci bus config implementation:
8344  *
8345  * Parameters
8346  *	vdip	vhci dip
8347  *	flags	bus config flags
8348  *	op	bus config operation
8349  *	The remaining parameters are bus config operation specific
8350  *
8351  * for BUS_CONFIG_ONE
8352  *	arg	pointer to name@addr
8353  *	child	upon successful return from this function, *child will be
8354  *		set to the configured and held devinfo child node of vdip.
8355  *	ct_addr	pointer to client address (i.e. GUID)
8356  *
8357  * for BUS_CONFIG_DRIVER
8358  *	arg	major number of the driver
8359  *	child and ct_addr parameters are ignored
8360  *
8361  * for BUS_CONFIG_ALL
8362  *	arg, child, and ct_addr parameters are ignored
8363  *
8364  * Note that for the rest of the bus config operations, this function simply
8365  * calls the framework provided default bus config routine.
8366  */
8367 int
8368 mdi_vhci_bus_config(dev_info_t *vdip, uint_t flags, ddi_bus_config_op_t op,
8369     void *arg, dev_info_t **child, char *ct_addr)
8370 {
8371 	mdi_vhci_t *vh = i_devi_get_vhci(vdip);
8372 	mdi_vhci_config_t *vhc = vh->vh_config;
8373 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
8374 	int rv = 0;
8375 	int params_valid = 0;
8376 	char *cp;
8377 
8378 	/*
8379 	 * To bus config vhcis we relay operation, possibly using another
8380 	 * thread, to phcis. The phci driver then interacts with MDI to cause
8381 	 * vhci child nodes to be enumerated under the vhci node.  Adding a
8382 	 * vhci child requires an ndi_devi_enter of the vhci. Since another
8383 	 * thread may be adding the child, to avoid deadlock we can't wait
8384 	 * for the relayed operations to complete if we have already entered
8385 	 * the vhci node.
8386 	 */
8387 	if (DEVI_BUSY_OWNED(vdip)) {
8388 		MDI_DEBUG(2, (CE_NOTE, vdip, "!MDI: vhci bus config: "
8389 		    "vhci dip is busy owned %p\n", (void *)vdip));
8390 		goto default_bus_config;
8391 	}
8392 
8393 	rw_enter(&vhcache->vhcache_lock, RW_READER);
8394 	if (!(vhcache->vhcache_flags & MDI_VHCI_CACHE_SETUP_DONE)) {
8395 		rw_exit(&vhcache->vhcache_lock);
8396 		rv = build_vhci_cache(vh);
8397 		rw_enter(&vhcache->vhcache_lock, RW_READER);
8398 	}
8399 
8400 	switch (op) {
8401 	case BUS_CONFIG_ONE:
8402 		if (arg != NULL && ct_addr != NULL) {
8403 			/* extract node name */
8404 			cp = (char *)arg;
8405 			while (*cp != '\0' && *cp != '@')
8406 				cp++;
8407 			if (*cp == '@') {
8408 				params_valid = 1;
8409 				*cp = '\0';
8410 				config_client_paths(vhc, (char *)arg, ct_addr);
8411 				/* config_client_paths() releases cache_lock */
8412 				*cp = '@';
8413 				break;
8414 			}
8415 		}
8416 
8417 		rw_exit(&vhcache->vhcache_lock);
8418 		break;
8419 
8420 	case BUS_CONFIG_DRIVER:
8421 		rw_exit(&vhcache->vhcache_lock);
8422 		if (rv == 0)
8423 			st_bus_config_all_phcis(vhc, flags, op,
8424 			    (major_t)(uintptr_t)arg);
8425 		break;
8426 
8427 	case BUS_CONFIG_ALL:
8428 		rw_exit(&vhcache->vhcache_lock);
8429 		if (rv == 0)
8430 			st_bus_config_all_phcis(vhc, flags, op, -1);
8431 		break;
8432 
8433 	default:
8434 		rw_exit(&vhcache->vhcache_lock);
8435 		break;
8436 	}
8437 
8438 
8439 default_bus_config:
8440 	/*
8441 	 * All requested child nodes are enumerated under the vhci.
8442 	 * Now configure them.
8443 	 */
8444 	if (ndi_busop_bus_config(vdip, flags, op, arg, child, 0) ==
8445 	    NDI_SUCCESS) {
8446 		return (MDI_SUCCESS);
8447 	} else if (op == BUS_CONFIG_ONE && rv == 0 && params_valid) {
8448 		/* discover all paths and try configuring again */
8449 		if (vhcache_discover_paths(vh) &&
8450 		    ndi_busop_bus_config(vdip, flags, op, arg, child, 0) ==
8451 		    NDI_SUCCESS)
8452 			return (MDI_SUCCESS);
8453 	}
8454 
8455 	return (MDI_FAILURE);
8456 }
8457 
8458 /*
8459  * Read the on-disk vhci cache into an nvlist for the specified vhci class.
8460  */
8461 static nvlist_t *
8462 read_on_disk_vhci_cache(char *vhci_class)
8463 {
8464 	nvlist_t *nvl;
8465 	int err;
8466 	char *filename;
8467 
8468 	filename = vhclass2vhcache_filename(vhci_class);
8469 
8470 	if ((err = fread_nvlist(filename, &nvl)) == 0) {
8471 		kmem_free(filename, strlen(filename) + 1);
8472 		return (nvl);
8473 	} else if (err == EIO)
8474 		cmn_err(CE_WARN, "%s: I/O error, will recreate\n", filename);
8475 	else if (err == EINVAL)
8476 		cmn_err(CE_WARN,
8477 		    "%s: data file corrupted, will recreate\n", filename);
8478 
8479 	kmem_free(filename, strlen(filename) + 1);
8480 	return (NULL);
8481 }
8482 
8483 /*
8484  * Read on-disk vhci cache into nvlists for all vhci classes.
8485  * Called during booting by i_ddi_read_devices_files().
8486  */
8487 void
8488 mdi_read_devices_files(void)
8489 {
8490 	int i;
8491 
8492 	for (i = 0; i < N_VHCI_CLASSES; i++)
8493 		vhcache_nvl[i] = read_on_disk_vhci_cache(vhci_class_list[i]);
8494 }
8495 
8496 /*
8497  * Remove all stale entries from vhci cache.
8498  */
8499 static void
8500 clean_vhcache(mdi_vhci_config_t *vhc)
8501 {
8502 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
8503 	mdi_vhcache_phci_t *cphci, *cphci_head, *cphci_next;
8504 	mdi_vhcache_client_t *cct, *cct_head, *cct_next;
8505 	mdi_vhcache_pathinfo_t *cpi, *cpi_head, *cpi_next;
8506 
8507 	rw_enter(&vhcache->vhcache_lock, RW_WRITER);
8508 
8509 	cct_head = vhcache->vhcache_client_head;
8510 	vhcache->vhcache_client_head = vhcache->vhcache_client_tail = NULL;
8511 	for (cct = cct_head; cct != NULL; cct = cct_next) {
8512 		cct_next = cct->cct_next;
8513 
8514 		cpi_head = cct->cct_cpi_head;
8515 		cct->cct_cpi_head = cct->cct_cpi_tail = NULL;
8516 		for (cpi = cpi_head; cpi != NULL; cpi = cpi_next) {
8517 			cpi_next = cpi->cpi_next;
8518 			if (cpi->cpi_pip != NULL) {
8519 				ASSERT(cpi->cpi_cphci->cphci_phci != NULL);
8520 				enqueue_tail_vhcache_pathinfo(cct, cpi);
8521 			} else
8522 				free_vhcache_pathinfo(cpi);
8523 		}
8524 
8525 		if (cct->cct_cpi_head != NULL)
8526 			enqueue_vhcache_client(vhcache, cct);
8527 		else {
8528 			(void) mod_hash_destroy(vhcache->vhcache_client_hash,
8529 			    (mod_hash_key_t)cct->cct_name_addr);
8530 			free_vhcache_client(cct);
8531 		}
8532 	}
8533 
8534 	cphci_head = vhcache->vhcache_phci_head;
8535 	vhcache->vhcache_phci_head = vhcache->vhcache_phci_tail = NULL;
8536 	for (cphci = cphci_head; cphci != NULL; cphci = cphci_next) {
8537 		cphci_next = cphci->cphci_next;
8538 		if (cphci->cphci_phci != NULL)
8539 			enqueue_vhcache_phci(vhcache, cphci);
8540 		else
8541 			free_vhcache_phci(cphci);
8542 	}
8543 
8544 	vhcache->vhcache_clean_time = lbolt64;
8545 	rw_exit(&vhcache->vhcache_lock);
8546 	vhcache_dirty(vhc);
8547 }
8548 
8549 /*
8550  * Remove all stale entries from vhci cache.
8551  * Called by i_ddi_clean_devices_files() during the execution of devfsadm -C
8552  */
8553 void
8554 mdi_clean_vhcache(void)
8555 {
8556 	mdi_vhci_t *vh;
8557 
8558 	mutex_enter(&mdi_mutex);
8559 	for (vh = mdi_vhci_head; vh != NULL; vh = vh->vh_next) {
8560 		vh->vh_refcnt++;
8561 		mutex_exit(&mdi_mutex);
8562 		clean_vhcache(vh->vh_config);
8563 		mutex_enter(&mdi_mutex);
8564 		vh->vh_refcnt--;
8565 	}
8566 	mutex_exit(&mdi_mutex);
8567 }
8568 
8569 /*
8570  * mdi_vhci_walk_clients():
8571  *		Walker routine to traverse client dev_info nodes
8572  * ddi_walk_devs(ddi_get_child(vdip), f, arg) returns the entire tree
8573  * below the client, including nexus devices, which we dont want.
8574  * So we just traverse the immediate siblings, starting from 1st client.
8575  */
8576 void
8577 mdi_vhci_walk_clients(dev_info_t *vdip,
8578     int (*f)(dev_info_t *, void *), void *arg)
8579 {
8580 	mdi_vhci_t	*vh = i_devi_get_vhci(vdip);
8581 	dev_info_t	*cdip;
8582 	mdi_client_t	*ct;
8583 
8584 	MDI_VHCI_CLIENT_LOCK(vh);
8585 	cdip = ddi_get_child(vdip);
8586 	while (cdip) {
8587 		ct = i_devi_get_client(cdip);
8588 		MDI_CLIENT_LOCK(ct);
8589 
8590 		if (((*f)(cdip, arg)) == DDI_WALK_CONTINUE)
8591 			cdip = ddi_get_next_sibling(cdip);
8592 		else
8593 			cdip = NULL;
8594 
8595 		MDI_CLIENT_UNLOCK(ct);
8596 	}
8597 	MDI_VHCI_CLIENT_UNLOCK(vh);
8598 }
8599 
8600 /*
8601  * mdi_vhci_walk_phcis():
8602  *		Walker routine to traverse phci dev_info nodes
8603  */
8604 void
8605 mdi_vhci_walk_phcis(dev_info_t *vdip,
8606     int (*f)(dev_info_t *, void *), void *arg)
8607 {
8608 	mdi_vhci_t	*vh = i_devi_get_vhci(vdip);
8609 	mdi_phci_t	*ph, *next;
8610 
8611 	MDI_VHCI_PHCI_LOCK(vh);
8612 	ph = vh->vh_phci_head;
8613 	while (ph) {
8614 		MDI_PHCI_LOCK(ph);
8615 
8616 		if (((*f)(ph->ph_dip, arg)) == DDI_WALK_CONTINUE)
8617 			next = ph->ph_next;
8618 		else
8619 			next = NULL;
8620 
8621 		MDI_PHCI_UNLOCK(ph);
8622 		ph = next;
8623 	}
8624 	MDI_VHCI_PHCI_UNLOCK(vh);
8625 }
8626 
8627 
8628 /*
8629  * mdi_walk_vhcis():
8630  *		Walker routine to traverse vhci dev_info nodes
8631  */
8632 void
8633 mdi_walk_vhcis(int (*f)(dev_info_t *, void *), void *arg)
8634 {
8635 	mdi_vhci_t	*vh = NULL;
8636 
8637 	mutex_enter(&mdi_mutex);
8638 	/*
8639 	 * Scan for already registered vhci
8640 	 */
8641 	for (vh = mdi_vhci_head; vh != NULL; vh = vh->vh_next) {
8642 		vh->vh_refcnt++;
8643 		mutex_exit(&mdi_mutex);
8644 		if (((*f)(vh->vh_dip, arg)) != DDI_WALK_CONTINUE) {
8645 			mutex_enter(&mdi_mutex);
8646 			vh->vh_refcnt--;
8647 			break;
8648 		} else {
8649 			mutex_enter(&mdi_mutex);
8650 			vh->vh_refcnt--;
8651 		}
8652 	}
8653 
8654 	mutex_exit(&mdi_mutex);
8655 }
8656 
8657 /*
8658  * i_mdi_log_sysevent():
8659  *		Logs events for pickup by syseventd
8660  */
8661 static void
8662 i_mdi_log_sysevent(dev_info_t *dip, char *ph_vh_class, char *subclass)
8663 {
8664 	char		*path_name;
8665 	nvlist_t	*attr_list;
8666 
8667 	if (nvlist_alloc(&attr_list, NV_UNIQUE_NAME_TYPE,
8668 	    KM_SLEEP) != DDI_SUCCESS) {
8669 		goto alloc_failed;
8670 	}
8671 
8672 	path_name = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
8673 	(void) ddi_pathname(dip, path_name);
8674 
8675 	if (nvlist_add_string(attr_list, DDI_DRIVER_NAME,
8676 	    ddi_driver_name(dip)) != DDI_SUCCESS) {
8677 		goto error;
8678 	}
8679 
8680 	if (nvlist_add_int32(attr_list, DDI_DRIVER_MAJOR,
8681 	    (int32_t)ddi_driver_major(dip)) != DDI_SUCCESS) {
8682 		goto error;
8683 	}
8684 
8685 	if (nvlist_add_int32(attr_list, DDI_INSTANCE,
8686 	    (int32_t)ddi_get_instance(dip)) != DDI_SUCCESS) {
8687 		goto error;
8688 	}
8689 
8690 	if (nvlist_add_string(attr_list, DDI_PATHNAME,
8691 	    path_name) != DDI_SUCCESS) {
8692 		goto error;
8693 	}
8694 
8695 	if (nvlist_add_string(attr_list, DDI_CLASS,
8696 	    ph_vh_class) != DDI_SUCCESS) {
8697 		goto error;
8698 	}
8699 
8700 	(void) ddi_log_sysevent(dip, DDI_VENDOR_SUNW, EC_DDI, subclass,
8701 	    attr_list, NULL, DDI_SLEEP);
8702 
8703 error:
8704 	kmem_free(path_name, MAXPATHLEN);
8705 	nvlist_free(attr_list);
8706 	return;
8707 
8708 alloc_failed:
8709 	MDI_DEBUG(1, (CE_WARN, dip,
8710 	    "!i_mdi_log_sysevent: Unable to send sysevent"));
8711 }
8712 
8713 char **
8714 mdi_get_phci_driver_list(char *vhci_class, int	*ndrivers)
8715 {
8716 	char	**driver_list, **ret_driver_list = NULL;
8717 	int	*root_support_list;
8718 	int	cur_elements, max_elements;
8719 
8720 	get_phci_driver_list(vhci_class, &driver_list, &root_support_list,
8721 	    &cur_elements, &max_elements);
8722 
8723 
8724 	if (driver_list) {
8725 		kmem_free(root_support_list, sizeof (int) * max_elements);
8726 		ret_driver_list = mdi_realloc(driver_list, sizeof (char *)
8727 		    * max_elements, sizeof (char *) * cur_elements);
8728 	}
8729 	*ndrivers = cur_elements;
8730 
8731 	return (ret_driver_list);
8732 
8733 }
8734 
8735 void
8736 mdi_free_phci_driver_list(char **driver_list, int ndrivers)
8737 {
8738 	char	**p;
8739 	int	i;
8740 
8741 	if (driver_list) {
8742 		for (i = 0, p = driver_list; i < ndrivers; i++, p++)
8743 			kmem_free(*p, strlen(*p) + 1);
8744 		kmem_free(driver_list, sizeof (char *) * ndrivers);
8745 	}
8746 }
8747