xref: /illumos-gate/usr/src/uts/common/os/sunmdi.c (revision 5e3986cb9bc07bde1dfa1a7028d5ad3e501abae4)
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 
72 #ifdef	DEBUG
73 #include <sys/debug.h>
74 int	mdi_debug = 1;
75 int	mdi_debug_logonly = 0;
76 #define	MDI_DEBUG(level, stmnt) \
77 	    if (mdi_debug >= (level)) i_mdi_log stmnt
78 static void i_mdi_log(int, dev_info_t *, const char *fmt, ...);
79 #else	/* !DEBUG */
80 #define	MDI_DEBUG(level, stmnt)
81 #endif	/* DEBUG */
82 
83 extern pri_t	minclsyspri;
84 extern int	modrootloaded;
85 
86 /*
87  * Global mutex:
88  * Protects vHCI list and structure members.
89  */
90 kmutex_t	mdi_mutex;
91 
92 /*
93  * Registered vHCI class driver lists
94  */
95 int		mdi_vhci_count;
96 mdi_vhci_t	*mdi_vhci_head;
97 mdi_vhci_t	*mdi_vhci_tail;
98 
99 /*
100  * Client Hash Table size
101  */
102 static int	mdi_client_table_size = CLIENT_HASH_TABLE_SIZE;
103 
104 /*
105  * taskq interface definitions
106  */
107 #define	MDI_TASKQ_N_THREADS	8
108 #define	MDI_TASKQ_PRI		minclsyspri
109 #define	MDI_TASKQ_MINALLOC	(4*mdi_taskq_n_threads)
110 #define	MDI_TASKQ_MAXALLOC	(500*mdi_taskq_n_threads)
111 
112 taskq_t				*mdi_taskq;
113 static uint_t			mdi_taskq_n_threads = MDI_TASKQ_N_THREADS;
114 
115 #define	TICKS_PER_SECOND	(drv_usectohz(1000000))
116 
117 /*
118  * The data should be "quiet" for this interval (in seconds) before the
119  * vhci cached data is flushed to the disk.
120  */
121 static int mdi_vhcache_flush_delay = 10;
122 
123 /* number of seconds the vhcache flush daemon will sleep idle before exiting */
124 static int mdi_vhcache_flush_daemon_idle_time = 60;
125 
126 /*
127  * MDI falls back to discovery of all paths when a bus_config_one fails.
128  * The following parameters can be used to tune this operation.
129  *
130  * mdi_path_discovery_boot
131  *	Number of times path discovery will be attempted during early boot.
132  *	Probably there is no reason to ever set this value to greater than one.
133  *
134  * mdi_path_discovery_postboot
135  *	Number of times path discovery will be attempted after early boot.
136  *	Set it to a minimum of two to allow for discovery of iscsi paths which
137  *	may happen very late during booting.
138  *
139  * mdi_path_discovery_interval
140  *	Minimum number of seconds MDI will wait between successive discovery
141  *	of all paths. Set it to -1 to disable discovery of all paths.
142  */
143 static int mdi_path_discovery_boot = 1;
144 static int mdi_path_discovery_postboot = 2;
145 static int mdi_path_discovery_interval = 10;
146 
147 /*
148  * number of seconds the asynchronous configuration thread will sleep idle
149  * before exiting.
150  */
151 static int mdi_async_config_idle_time = 600;
152 
153 static int mdi_bus_config_cache_hash_size = 256;
154 
155 /* turns off multithreaded configuration for certain operations */
156 static int mdi_mtc_off = 0;
157 
158 /*
159  * MDI component property name/value string definitions
160  */
161 const char 		*mdi_component_prop = "mpxio-component";
162 const char		*mdi_component_prop_vhci = "vhci";
163 const char		*mdi_component_prop_phci = "phci";
164 const char		*mdi_component_prop_client = "client";
165 
166 /*
167  * MDI client global unique identifier property name
168  */
169 const char		*mdi_client_guid_prop = "client-guid";
170 
171 /*
172  * MDI client load balancing property name/value string definitions
173  */
174 const char		*mdi_load_balance = "load-balance";
175 const char		*mdi_load_balance_none = "none";
176 const char		*mdi_load_balance_rr = "round-robin";
177 const char		*mdi_load_balance_lba = "logical-block";
178 
179 /*
180  * Obsolete vHCI class definition; to be removed after Leadville update
181  */
182 const char *mdi_vhci_class_scsi = MDI_HCI_CLASS_SCSI;
183 
184 static char vhci_greeting[] =
185 	"\tThere already exists one vHCI driver for class %s\n"
186 	"\tOnly one vHCI driver for each class is allowed\n";
187 
188 /*
189  * Static function prototypes
190  */
191 static int		i_mdi_phci_offline(dev_info_t *, uint_t);
192 static int		i_mdi_client_offline(dev_info_t *, uint_t);
193 static int		i_mdi_phci_pre_detach(dev_info_t *, ddi_detach_cmd_t);
194 static void		i_mdi_phci_post_detach(dev_info_t *,
195 			    ddi_detach_cmd_t, int);
196 static int		i_mdi_client_pre_detach(dev_info_t *,
197 			    ddi_detach_cmd_t);
198 static void		i_mdi_client_post_detach(dev_info_t *,
199 			    ddi_detach_cmd_t, int);
200 static void		i_mdi_pm_hold_pip(mdi_pathinfo_t *);
201 static void		i_mdi_pm_rele_pip(mdi_pathinfo_t *);
202 static int 		i_mdi_lba_lb(mdi_client_t *ct,
203 			    mdi_pathinfo_t **ret_pip, struct buf *buf);
204 static void		i_mdi_pm_hold_client(mdi_client_t *, int);
205 static void		i_mdi_pm_rele_client(mdi_client_t *, int);
206 static void		i_mdi_pm_reset_client(mdi_client_t *);
207 static int		i_mdi_power_all_phci(mdi_client_t *);
208 static void		i_mdi_log_sysevent(dev_info_t *, char *, char *);
209 
210 
211 /*
212  * Internal mdi_pathinfo node functions
213  */
214 static int		i_mdi_pi_kstat_create(mdi_pathinfo_t *);
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 	dev_info_t *cdip;
3490 	int		client_held = 0;
3491 	int rv;
3492 
3493 	ASSERT(ct != NULL);
3494 	rv = i_mdi_pi_state_change(pip, MDI_PATHINFO_STATE_ONLINE, flags);
3495 	if (rv != MDI_SUCCESS)
3496 		return (rv);
3497 
3498 	MDI_PI_LOCK(pip);
3499 	if (MDI_PI(pip)->pi_pm_held == 0) {
3500 		MDI_DEBUG(4, (CE_NOTE, ct->ct_dip, "mdi_pi_online "
3501 		    "i_mdi_pm_hold_pip %p\n", (void *)pip));
3502 		i_mdi_pm_hold_pip(pip);
3503 		client_held = 1;
3504 	}
3505 	MDI_PI_UNLOCK(pip);
3506 
3507 	if (client_held) {
3508 		MDI_CLIENT_LOCK(ct);
3509 		if (ct->ct_power_cnt == 0) {
3510 			rv = i_mdi_power_all_phci(ct);
3511 		}
3512 
3513 		MDI_DEBUG(4, (CE_NOTE, ct->ct_dip, "mdi_pi_online "
3514 		    "i_mdi_pm_hold_client %p\n", (void *)ct));
3515 		i_mdi_pm_hold_client(ct, 1);
3516 		MDI_CLIENT_UNLOCK(ct);
3517 	}
3518 
3519 	/*
3520 	 * Create the per-path (pathinfo) IO and error kstats which
3521 	 * are reported via iostat(1m).
3522 	 *
3523 	 * Defer creating the per-path kstats if device is not yet
3524 	 * attached;  the names of the kstats are constructed in part
3525 	 * using the devices instance number which is assigned during
3526 	 * process of attaching the client device.
3527 	 *
3528 	 * The framework post_attach handler, mdi_post_attach(), is
3529 	 * is responsible for initializing the client's pathinfo list
3530 	 * once successfully attached.
3531 	 */
3532 	cdip = ct->ct_dip;
3533 	ASSERT(cdip);
3534 	if (cdip == NULL || !i_ddi_devi_attached(cdip))
3535 		return (rv);
3536 
3537 	MDI_CLIENT_LOCK(ct);
3538 	rv = i_mdi_pi_kstat_create(pip);
3539 	MDI_CLIENT_UNLOCK(ct);
3540 	return (rv);
3541 }
3542 
3543 /*
3544  * mdi_pi_standby():
3545  *		Place the mdi_pathinfo node in standby state
3546  *
3547  * Return Values:
3548  *		MDI_SUCCESS
3549  *		MDI_FAILURE
3550  */
3551 int
3552 mdi_pi_standby(mdi_pathinfo_t *pip, int flags)
3553 {
3554 	return (i_mdi_pi_state_change(pip, MDI_PATHINFO_STATE_STANDBY, flags));
3555 }
3556 
3557 /*
3558  * mdi_pi_fault():
3559  *		Place the mdi_pathinfo node in fault'ed state
3560  * Return Values:
3561  *		MDI_SUCCESS
3562  *		MDI_FAILURE
3563  */
3564 int
3565 mdi_pi_fault(mdi_pathinfo_t *pip, int flags)
3566 {
3567 	return (i_mdi_pi_state_change(pip, MDI_PATHINFO_STATE_FAULT, flags));
3568 }
3569 
3570 /*
3571  * mdi_pi_offline():
3572  *		Offline a mdi_pathinfo node.
3573  * Return Values:
3574  *		MDI_SUCCESS
3575  *		MDI_FAILURE
3576  */
3577 int
3578 mdi_pi_offline(mdi_pathinfo_t *pip, int flags)
3579 {
3580 	int	ret, client_held = 0;
3581 	mdi_client_t	*ct;
3582 
3583 	ret = i_mdi_pi_state_change(pip, MDI_PATHINFO_STATE_OFFLINE, flags);
3584 
3585 	if (ret == MDI_SUCCESS) {
3586 		MDI_PI_LOCK(pip);
3587 		if (MDI_PI(pip)->pi_pm_held) {
3588 			client_held = 1;
3589 		}
3590 		MDI_PI_UNLOCK(pip);
3591 
3592 		if (client_held) {
3593 			ct = MDI_PI(pip)->pi_client;
3594 			MDI_CLIENT_LOCK(ct);
3595 			MDI_DEBUG(4, (CE_NOTE, ct->ct_dip,
3596 			    "mdi_pi_offline i_mdi_pm_rele_client\n"));
3597 			i_mdi_pm_rele_client(ct, 1);
3598 			MDI_CLIENT_UNLOCK(ct);
3599 		}
3600 	}
3601 
3602 	return (ret);
3603 }
3604 
3605 /*
3606  * i_mdi_pi_offline():
3607  *		Offline a mdi_pathinfo node and call the vHCI driver's callback
3608  */
3609 static int
3610 i_mdi_pi_offline(mdi_pathinfo_t *pip, int flags)
3611 {
3612 	dev_info_t	*vdip = NULL;
3613 	mdi_vhci_t	*vh = NULL;
3614 	mdi_client_t	*ct = NULL;
3615 	int		(*f)();
3616 	int		rv;
3617 
3618 	MDI_PI_LOCK(pip);
3619 	ct = MDI_PI(pip)->pi_client;
3620 	ASSERT(ct != NULL);
3621 
3622 	while (MDI_PI(pip)->pi_ref_cnt != 0) {
3623 		/*
3624 		 * Give a chance for pending I/Os to complete.
3625 		 */
3626 		MDI_DEBUG(1, (CE_NOTE, ct->ct_dip, "!i_mdi_pi_offline: "
3627 		    "%d cmds still pending on path: %p\n",
3628 		    MDI_PI(pip)->pi_ref_cnt, (void *)pip));
3629 		if (cv_timedwait(&MDI_PI(pip)->pi_ref_cv,
3630 		    &MDI_PI(pip)->pi_mutex,
3631 		    ddi_get_lbolt() + drv_usectohz(60 * 1000000)) == -1) {
3632 			/*
3633 			 * The timeout time reached without ref_cnt being zero
3634 			 * being signaled.
3635 			 */
3636 			MDI_DEBUG(1, (CE_NOTE, ct->ct_dip, "!i_mdi_pi_offline: "
3637 			    "Timeout reached on path %p without the cond\n",
3638 			    (void *)pip));
3639 			MDI_DEBUG(1, (CE_NOTE, ct->ct_dip, "!i_mdi_pi_offline: "
3640 			    "%d cmds still pending on path: %p\n",
3641 			    MDI_PI(pip)->pi_ref_cnt, (void *)pip));
3642 		}
3643 	}
3644 	vh = ct->ct_vhci;
3645 	vdip = vh->vh_dip;
3646 
3647 	/*
3648 	 * Notify vHCI that has registered this event
3649 	 */
3650 	ASSERT(vh->vh_ops);
3651 	f = vh->vh_ops->vo_pi_state_change;
3652 
3653 	if (f != NULL) {
3654 		MDI_PI_UNLOCK(pip);
3655 		if ((rv = (*f)(vdip, pip, MDI_PATHINFO_STATE_OFFLINE, 0,
3656 		    flags)) != MDI_SUCCESS) {
3657 			MDI_DEBUG(1, (CE_WARN, ct->ct_dip,
3658 			    "!vo_path_offline failed "
3659 			    "vdip %p, pip %p", (void *)vdip, (void *)pip));
3660 		}
3661 		MDI_PI_LOCK(pip);
3662 	}
3663 
3664 	/*
3665 	 * Set the mdi_pathinfo node state and clear the transient condition
3666 	 */
3667 	MDI_PI_SET_OFFLINE(pip);
3668 	cv_broadcast(&MDI_PI(pip)->pi_state_cv);
3669 	MDI_PI_UNLOCK(pip);
3670 
3671 	MDI_CLIENT_LOCK(ct);
3672 	if (rv == MDI_SUCCESS) {
3673 		if (ct->ct_unstable == 0) {
3674 			dev_info_t	*cdip = ct->ct_dip;
3675 
3676 			/*
3677 			 * Onlining the mdi_pathinfo node will impact the
3678 			 * client state Update the client and dev_info node
3679 			 * state accordingly
3680 			 */
3681 			i_mdi_client_update_state(ct);
3682 			rv = NDI_SUCCESS;
3683 			if (MDI_CLIENT_STATE(ct) == MDI_CLIENT_STATE_FAILED) {
3684 				if (cdip &&
3685 				    (i_ddi_node_state(cdip) >=
3686 				    DS_INITIALIZED)) {
3687 					MDI_CLIENT_UNLOCK(ct);
3688 					rv = ndi_devi_offline(cdip, 0);
3689 					MDI_CLIENT_LOCK(ct);
3690 					if (rv != NDI_SUCCESS) {
3691 						/*
3692 						 * ndi_devi_offline failed.
3693 						 * Reset client flags to
3694 						 * online.
3695 						 */
3696 						MDI_DEBUG(4, (CE_WARN, cdip,
3697 						    "!ndi_devi_offline: failed "
3698 						    " Error: %x", rv));
3699 						MDI_CLIENT_SET_ONLINE(ct);
3700 					}
3701 				}
3702 			}
3703 			/*
3704 			 * Convert to MDI error code
3705 			 */
3706 			switch (rv) {
3707 			case NDI_SUCCESS:
3708 				rv = MDI_SUCCESS;
3709 				break;
3710 			case NDI_BUSY:
3711 				rv = MDI_BUSY;
3712 				break;
3713 			default:
3714 				rv = MDI_FAILURE;
3715 				break;
3716 			}
3717 		}
3718 		MDI_CLIENT_SET_REPORT_DEV_NEEDED(ct);
3719 		i_mdi_report_path_state(ct, pip);
3720 	}
3721 
3722 	MDI_CLIENT_UNLOCK(ct);
3723 
3724 	/*
3725 	 * Change in the mdi_pathinfo node state will impact the client state
3726 	 */
3727 	MDI_DEBUG(2, (CE_NOTE, NULL, "!i_mdi_pi_offline ct = %p pip = %p",
3728 	    (void *)ct, (void *)pip));
3729 	return (rv);
3730 }
3731 
3732 
3733 /*
3734  * mdi_pi_get_addr():
3735  *		Get the unit address associated with a mdi_pathinfo node
3736  *
3737  * Return Values:
3738  *		char *
3739  */
3740 char *
3741 mdi_pi_get_addr(mdi_pathinfo_t *pip)
3742 {
3743 	if (pip == NULL)
3744 		return (NULL);
3745 
3746 	return (MDI_PI(pip)->pi_addr);
3747 }
3748 
3749 /*
3750  * mdi_pi_get_client():
3751  *		Get the client devinfo associated with a mdi_pathinfo node
3752  *
3753  * Return Values:
3754  *		Handle to client device dev_info node
3755  */
3756 dev_info_t *
3757 mdi_pi_get_client(mdi_pathinfo_t *pip)
3758 {
3759 	dev_info_t	*dip = NULL;
3760 	if (pip) {
3761 		dip = MDI_PI(pip)->pi_client->ct_dip;
3762 	}
3763 	return (dip);
3764 }
3765 
3766 /*
3767  * mdi_pi_get_phci():
3768  *		Get the pHCI devinfo associated with the mdi_pathinfo node
3769  * Return Values:
3770  *		Handle to dev_info node
3771  */
3772 dev_info_t *
3773 mdi_pi_get_phci(mdi_pathinfo_t *pip)
3774 {
3775 	dev_info_t	*dip = NULL;
3776 	if (pip) {
3777 		dip = MDI_PI(pip)->pi_phci->ph_dip;
3778 	}
3779 	return (dip);
3780 }
3781 
3782 /*
3783  * mdi_pi_get_client_private():
3784  *		Get the client private information associated with the
3785  *		mdi_pathinfo node
3786  */
3787 void *
3788 mdi_pi_get_client_private(mdi_pathinfo_t *pip)
3789 {
3790 	void *cprivate = NULL;
3791 	if (pip) {
3792 		cprivate = MDI_PI(pip)->pi_cprivate;
3793 	}
3794 	return (cprivate);
3795 }
3796 
3797 /*
3798  * mdi_pi_set_client_private():
3799  *		Set the client private information in the mdi_pathinfo node
3800  */
3801 void
3802 mdi_pi_set_client_private(mdi_pathinfo_t *pip, void *priv)
3803 {
3804 	if (pip) {
3805 		MDI_PI(pip)->pi_cprivate = priv;
3806 	}
3807 }
3808 
3809 /*
3810  * mdi_pi_get_phci_private():
3811  *		Get the pHCI private information associated with the
3812  *		mdi_pathinfo node
3813  */
3814 caddr_t
3815 mdi_pi_get_phci_private(mdi_pathinfo_t *pip)
3816 {
3817 	caddr_t	pprivate = NULL;
3818 	if (pip) {
3819 		pprivate = MDI_PI(pip)->pi_pprivate;
3820 	}
3821 	return (pprivate);
3822 }
3823 
3824 /*
3825  * mdi_pi_set_phci_private():
3826  *		Set the pHCI private information in the mdi_pathinfo node
3827  */
3828 void
3829 mdi_pi_set_phci_private(mdi_pathinfo_t *pip, caddr_t priv)
3830 {
3831 	if (pip) {
3832 		MDI_PI(pip)->pi_pprivate = priv;
3833 	}
3834 }
3835 
3836 /*
3837  * mdi_pi_get_state():
3838  *		Get the mdi_pathinfo node state. Transient states are internal
3839  *		and not provided to the users
3840  */
3841 mdi_pathinfo_state_t
3842 mdi_pi_get_state(mdi_pathinfo_t *pip)
3843 {
3844 	mdi_pathinfo_state_t    state = MDI_PATHINFO_STATE_INIT;
3845 
3846 	if (pip) {
3847 		if (MDI_PI_IS_TRANSIENT(pip)) {
3848 			/*
3849 			 * mdi_pathinfo is in state transition.  Return the
3850 			 * last good state.
3851 			 */
3852 			state = MDI_PI_OLD_STATE(pip);
3853 		} else {
3854 			state = MDI_PI_STATE(pip);
3855 		}
3856 	}
3857 	return (state);
3858 }
3859 
3860 /*
3861  * Note that the following function needs to be the new interface for
3862  * mdi_pi_get_state when mpxio gets integrated to ON.
3863  */
3864 int
3865 mdi_pi_get_state2(mdi_pathinfo_t *pip, mdi_pathinfo_state_t *state,
3866 		uint32_t *ext_state)
3867 {
3868 	*state = MDI_PATHINFO_STATE_INIT;
3869 
3870 	if (pip) {
3871 		if (MDI_PI_IS_TRANSIENT(pip)) {
3872 			/*
3873 			 * mdi_pathinfo is in state transition.  Return the
3874 			 * last good state.
3875 			 */
3876 			*state = MDI_PI_OLD_STATE(pip);
3877 			*ext_state = MDI_PI_OLD_EXT_STATE(pip);
3878 		} else {
3879 			*state = MDI_PI_STATE(pip);
3880 			*ext_state = MDI_PI_EXT_STATE(pip);
3881 		}
3882 	}
3883 	return (MDI_SUCCESS);
3884 }
3885 
3886 /*
3887  * mdi_pi_get_preferred:
3888  *	Get the preferred path flag
3889  */
3890 int
3891 mdi_pi_get_preferred(mdi_pathinfo_t *pip)
3892 {
3893 	if (pip) {
3894 		return (MDI_PI(pip)->pi_preferred);
3895 	}
3896 	return (0);
3897 }
3898 
3899 /*
3900  * mdi_pi_set_preferred:
3901  *	Set the preferred path flag
3902  */
3903 void
3904 mdi_pi_set_preferred(mdi_pathinfo_t *pip, int preferred)
3905 {
3906 	if (pip) {
3907 		MDI_PI(pip)->pi_preferred = preferred;
3908 	}
3909 }
3910 
3911 /*
3912  * mdi_pi_set_state():
3913  *		Set the mdi_pathinfo node state
3914  */
3915 void
3916 mdi_pi_set_state(mdi_pathinfo_t *pip, mdi_pathinfo_state_t state)
3917 {
3918 	uint32_t	ext_state;
3919 
3920 	if (pip) {
3921 		ext_state = MDI_PI(pip)->pi_state & MDI_PATHINFO_EXT_STATE_MASK;
3922 		MDI_PI(pip)->pi_state = state;
3923 		MDI_PI(pip)->pi_state |= ext_state;
3924 	}
3925 }
3926 
3927 /*
3928  * Property functions:
3929  */
3930 int
3931 i_map_nvlist_error_to_mdi(int val)
3932 {
3933 	int rv;
3934 
3935 	switch (val) {
3936 	case 0:
3937 		rv = DDI_PROP_SUCCESS;
3938 		break;
3939 	case EINVAL:
3940 	case ENOTSUP:
3941 		rv = DDI_PROP_INVAL_ARG;
3942 		break;
3943 	case ENOMEM:
3944 		rv = DDI_PROP_NO_MEMORY;
3945 		break;
3946 	default:
3947 		rv = DDI_PROP_NOT_FOUND;
3948 		break;
3949 	}
3950 	return (rv);
3951 }
3952 
3953 /*
3954  * mdi_pi_get_next_prop():
3955  * 		Property walk function.  The caller should hold mdi_pi_lock()
3956  *		and release by calling mdi_pi_unlock() at the end of walk to
3957  *		get a consistent value.
3958  */
3959 nvpair_t *
3960 mdi_pi_get_next_prop(mdi_pathinfo_t *pip, nvpair_t *prev)
3961 {
3962 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
3963 		return (NULL);
3964 	}
3965 	ASSERT(MDI_PI_LOCKED(pip));
3966 	return (nvlist_next_nvpair(MDI_PI(pip)->pi_prop, prev));
3967 }
3968 
3969 /*
3970  * mdi_prop_remove():
3971  * 		Remove the named property from the named list.
3972  */
3973 int
3974 mdi_prop_remove(mdi_pathinfo_t *pip, char *name)
3975 {
3976 	if (pip == NULL) {
3977 		return (DDI_PROP_NOT_FOUND);
3978 	}
3979 	ASSERT(!MDI_PI_LOCKED(pip));
3980 	MDI_PI_LOCK(pip);
3981 	if (MDI_PI(pip)->pi_prop == NULL) {
3982 		MDI_PI_UNLOCK(pip);
3983 		return (DDI_PROP_NOT_FOUND);
3984 	}
3985 	if (name) {
3986 		(void) nvlist_remove_all(MDI_PI(pip)->pi_prop, name);
3987 	} else {
3988 		char		nvp_name[MAXNAMELEN];
3989 		nvpair_t	*nvp;
3990 		nvp = nvlist_next_nvpair(MDI_PI(pip)->pi_prop, NULL);
3991 		while (nvp) {
3992 			nvpair_t	*next;
3993 			next = nvlist_next_nvpair(MDI_PI(pip)->pi_prop, nvp);
3994 			(void) snprintf(nvp_name, MAXNAMELEN, "%s",
3995 			    nvpair_name(nvp));
3996 			(void) nvlist_remove_all(MDI_PI(pip)->pi_prop,
3997 			    nvp_name);
3998 			nvp = next;
3999 		}
4000 	}
4001 	MDI_PI_UNLOCK(pip);
4002 	return (DDI_PROP_SUCCESS);
4003 }
4004 
4005 /*
4006  * mdi_prop_size():
4007  * 		Get buffer size needed to pack the property data.
4008  * 		Caller should hold the mdi_pathinfo_t lock to get a consistent
4009  *		buffer size.
4010  */
4011 int
4012 mdi_prop_size(mdi_pathinfo_t *pip, size_t *buflenp)
4013 {
4014 	int	rv;
4015 	size_t	bufsize;
4016 
4017 	*buflenp = 0;
4018 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4019 		return (DDI_PROP_NOT_FOUND);
4020 	}
4021 	ASSERT(MDI_PI_LOCKED(pip));
4022 	rv = nvlist_size(MDI_PI(pip)->pi_prop,
4023 	    &bufsize, NV_ENCODE_NATIVE);
4024 	*buflenp = bufsize;
4025 	return (i_map_nvlist_error_to_mdi(rv));
4026 }
4027 
4028 /*
4029  * mdi_prop_pack():
4030  * 		pack the property list.  The caller should hold the
4031  *		mdi_pathinfo_t node to get a consistent data
4032  */
4033 int
4034 mdi_prop_pack(mdi_pathinfo_t *pip, char **bufp, uint_t buflen)
4035 {
4036 	int	rv;
4037 	size_t	bufsize;
4038 
4039 	if ((pip == NULL) || MDI_PI(pip)->pi_prop == NULL) {
4040 		return (DDI_PROP_NOT_FOUND);
4041 	}
4042 
4043 	ASSERT(MDI_PI_LOCKED(pip));
4044 
4045 	bufsize = buflen;
4046 	rv = nvlist_pack(MDI_PI(pip)->pi_prop, bufp, (size_t *)&bufsize,
4047 	    NV_ENCODE_NATIVE, KM_SLEEP);
4048 
4049 	return (i_map_nvlist_error_to_mdi(rv));
4050 }
4051 
4052 /*
4053  * mdi_prop_update_byte():
4054  *		Create/Update a byte property
4055  */
4056 int
4057 mdi_prop_update_byte(mdi_pathinfo_t *pip, char *name, uchar_t data)
4058 {
4059 	int rv;
4060 
4061 	if (pip == NULL) {
4062 		return (DDI_PROP_INVAL_ARG);
4063 	}
4064 	ASSERT(!MDI_PI_LOCKED(pip));
4065 	MDI_PI_LOCK(pip);
4066 	if (MDI_PI(pip)->pi_prop == NULL) {
4067 		MDI_PI_UNLOCK(pip);
4068 		return (DDI_PROP_NOT_FOUND);
4069 	}
4070 	rv = nvlist_add_byte(MDI_PI(pip)->pi_prop, name, data);
4071 	MDI_PI_UNLOCK(pip);
4072 	return (i_map_nvlist_error_to_mdi(rv));
4073 }
4074 
4075 /*
4076  * mdi_prop_update_byte_array():
4077  *		Create/Update a byte array property
4078  */
4079 int
4080 mdi_prop_update_byte_array(mdi_pathinfo_t *pip, char *name, uchar_t *data,
4081     uint_t nelements)
4082 {
4083 	int rv;
4084 
4085 	if (pip == NULL) {
4086 		return (DDI_PROP_INVAL_ARG);
4087 	}
4088 	ASSERT(!MDI_PI_LOCKED(pip));
4089 	MDI_PI_LOCK(pip);
4090 	if (MDI_PI(pip)->pi_prop == NULL) {
4091 		MDI_PI_UNLOCK(pip);
4092 		return (DDI_PROP_NOT_FOUND);
4093 	}
4094 	rv = nvlist_add_byte_array(MDI_PI(pip)->pi_prop, name, data, nelements);
4095 	MDI_PI_UNLOCK(pip);
4096 	return (i_map_nvlist_error_to_mdi(rv));
4097 }
4098 
4099 /*
4100  * mdi_prop_update_int():
4101  *		Create/Update a 32 bit integer property
4102  */
4103 int
4104 mdi_prop_update_int(mdi_pathinfo_t *pip, char *name, int data)
4105 {
4106 	int rv;
4107 
4108 	if (pip == NULL) {
4109 		return (DDI_PROP_INVAL_ARG);
4110 	}
4111 	ASSERT(!MDI_PI_LOCKED(pip));
4112 	MDI_PI_LOCK(pip);
4113 	if (MDI_PI(pip)->pi_prop == NULL) {
4114 		MDI_PI_UNLOCK(pip);
4115 		return (DDI_PROP_NOT_FOUND);
4116 	}
4117 	rv = nvlist_add_int32(MDI_PI(pip)->pi_prop, name, (int32_t)data);
4118 	MDI_PI_UNLOCK(pip);
4119 	return (i_map_nvlist_error_to_mdi(rv));
4120 }
4121 
4122 /*
4123  * mdi_prop_update_int64():
4124  *		Create/Update a 64 bit integer property
4125  */
4126 int
4127 mdi_prop_update_int64(mdi_pathinfo_t *pip, char *name, int64_t data)
4128 {
4129 	int rv;
4130 
4131 	if (pip == NULL) {
4132 		return (DDI_PROP_INVAL_ARG);
4133 	}
4134 	ASSERT(!MDI_PI_LOCKED(pip));
4135 	MDI_PI_LOCK(pip);
4136 	if (MDI_PI(pip)->pi_prop == NULL) {
4137 		MDI_PI_UNLOCK(pip);
4138 		return (DDI_PROP_NOT_FOUND);
4139 	}
4140 	rv = nvlist_add_int64(MDI_PI(pip)->pi_prop, name, data);
4141 	MDI_PI_UNLOCK(pip);
4142 	return (i_map_nvlist_error_to_mdi(rv));
4143 }
4144 
4145 /*
4146  * mdi_prop_update_int_array():
4147  *		Create/Update a int array property
4148  */
4149 int
4150 mdi_prop_update_int_array(mdi_pathinfo_t *pip, char *name, int *data,
4151 	    uint_t nelements)
4152 {
4153 	int rv;
4154 
4155 	if (pip == NULL) {
4156 		return (DDI_PROP_INVAL_ARG);
4157 	}
4158 	ASSERT(!MDI_PI_LOCKED(pip));
4159 	MDI_PI_LOCK(pip);
4160 	if (MDI_PI(pip)->pi_prop == NULL) {
4161 		MDI_PI_UNLOCK(pip);
4162 		return (DDI_PROP_NOT_FOUND);
4163 	}
4164 	rv = nvlist_add_int32_array(MDI_PI(pip)->pi_prop, name, (int32_t *)data,
4165 	    nelements);
4166 	MDI_PI_UNLOCK(pip);
4167 	return (i_map_nvlist_error_to_mdi(rv));
4168 }
4169 
4170 /*
4171  * mdi_prop_update_string():
4172  *		Create/Update a string property
4173  */
4174 int
4175 mdi_prop_update_string(mdi_pathinfo_t *pip, char *name, char *data)
4176 {
4177 	int rv;
4178 
4179 	if (pip == NULL) {
4180 		return (DDI_PROP_INVAL_ARG);
4181 	}
4182 	ASSERT(!MDI_PI_LOCKED(pip));
4183 	MDI_PI_LOCK(pip);
4184 	if (MDI_PI(pip)->pi_prop == NULL) {
4185 		MDI_PI_UNLOCK(pip);
4186 		return (DDI_PROP_NOT_FOUND);
4187 	}
4188 	rv = nvlist_add_string(MDI_PI(pip)->pi_prop, name, data);
4189 	MDI_PI_UNLOCK(pip);
4190 	return (i_map_nvlist_error_to_mdi(rv));
4191 }
4192 
4193 /*
4194  * mdi_prop_update_string_array():
4195  *		Create/Update a string array property
4196  */
4197 int
4198 mdi_prop_update_string_array(mdi_pathinfo_t *pip, char *name, char **data,
4199     uint_t nelements)
4200 {
4201 	int rv;
4202 
4203 	if (pip == NULL) {
4204 		return (DDI_PROP_INVAL_ARG);
4205 	}
4206 	ASSERT(!MDI_PI_LOCKED(pip));
4207 	MDI_PI_LOCK(pip);
4208 	if (MDI_PI(pip)->pi_prop == NULL) {
4209 		MDI_PI_UNLOCK(pip);
4210 		return (DDI_PROP_NOT_FOUND);
4211 	}
4212 	rv = nvlist_add_string_array(MDI_PI(pip)->pi_prop, name, data,
4213 	    nelements);
4214 	MDI_PI_UNLOCK(pip);
4215 	return (i_map_nvlist_error_to_mdi(rv));
4216 }
4217 
4218 /*
4219  * mdi_prop_lookup_byte():
4220  * 		Look for byte property identified by name.  The data returned
4221  *		is the actual property and valid as long as mdi_pathinfo_t node
4222  *		is alive.
4223  */
4224 int
4225 mdi_prop_lookup_byte(mdi_pathinfo_t *pip, char *name, uchar_t *data)
4226 {
4227 	int rv;
4228 
4229 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4230 		return (DDI_PROP_NOT_FOUND);
4231 	}
4232 	rv = nvlist_lookup_byte(MDI_PI(pip)->pi_prop, name, data);
4233 	return (i_map_nvlist_error_to_mdi(rv));
4234 }
4235 
4236 
4237 /*
4238  * mdi_prop_lookup_byte_array():
4239  * 		Look for byte array property identified by name.  The data
4240  *		returned is the actual property and valid as long as
4241  *		mdi_pathinfo_t node is alive.
4242  */
4243 int
4244 mdi_prop_lookup_byte_array(mdi_pathinfo_t *pip, char *name, uchar_t **data,
4245     uint_t *nelements)
4246 {
4247 	int rv;
4248 
4249 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4250 		return (DDI_PROP_NOT_FOUND);
4251 	}
4252 	rv = nvlist_lookup_byte_array(MDI_PI(pip)->pi_prop, name, data,
4253 	    nelements);
4254 	return (i_map_nvlist_error_to_mdi(rv));
4255 }
4256 
4257 /*
4258  * mdi_prop_lookup_int():
4259  * 		Look for int property identified by name.  The data returned
4260  *		is the actual property and valid as long as mdi_pathinfo_t
4261  *		node is alive.
4262  */
4263 int
4264 mdi_prop_lookup_int(mdi_pathinfo_t *pip, char *name, int *data)
4265 {
4266 	int rv;
4267 
4268 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4269 		return (DDI_PROP_NOT_FOUND);
4270 	}
4271 	rv = nvlist_lookup_int32(MDI_PI(pip)->pi_prop, name, (int32_t *)data);
4272 	return (i_map_nvlist_error_to_mdi(rv));
4273 }
4274 
4275 /*
4276  * mdi_prop_lookup_int64():
4277  * 		Look for int64 property identified by name.  The data returned
4278  *		is the actual property and valid as long as mdi_pathinfo_t node
4279  *		is alive.
4280  */
4281 int
4282 mdi_prop_lookup_int64(mdi_pathinfo_t *pip, char *name, int64_t *data)
4283 {
4284 	int rv;
4285 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4286 		return (DDI_PROP_NOT_FOUND);
4287 	}
4288 	rv = nvlist_lookup_int64(MDI_PI(pip)->pi_prop, name, data);
4289 	return (i_map_nvlist_error_to_mdi(rv));
4290 }
4291 
4292 /*
4293  * mdi_prop_lookup_int_array():
4294  * 		Look for int array property identified by name.  The data
4295  *		returned is the actual property and valid as long as
4296  *		mdi_pathinfo_t node is alive.
4297  */
4298 int
4299 mdi_prop_lookup_int_array(mdi_pathinfo_t *pip, char *name, int **data,
4300     uint_t *nelements)
4301 {
4302 	int rv;
4303 
4304 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4305 		return (DDI_PROP_NOT_FOUND);
4306 	}
4307 	rv = nvlist_lookup_int32_array(MDI_PI(pip)->pi_prop, name,
4308 	    (int32_t **)data, nelements);
4309 	return (i_map_nvlist_error_to_mdi(rv));
4310 }
4311 
4312 /*
4313  * mdi_prop_lookup_string():
4314  * 		Look for string property identified by name.  The data
4315  *		returned is the actual property and valid as long as
4316  *		mdi_pathinfo_t node is alive.
4317  */
4318 int
4319 mdi_prop_lookup_string(mdi_pathinfo_t *pip, char *name, char **data)
4320 {
4321 	int rv;
4322 
4323 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4324 		return (DDI_PROP_NOT_FOUND);
4325 	}
4326 	rv = nvlist_lookup_string(MDI_PI(pip)->pi_prop, name, data);
4327 	return (i_map_nvlist_error_to_mdi(rv));
4328 }
4329 
4330 /*
4331  * mdi_prop_lookup_string_array():
4332  * 		Look for string array property identified by name.  The data
4333  *		returned is the actual property and valid as long as
4334  *		mdi_pathinfo_t node is alive.
4335  */
4336 int
4337 mdi_prop_lookup_string_array(mdi_pathinfo_t *pip, char *name, char ***data,
4338     uint_t *nelements)
4339 {
4340 	int rv;
4341 
4342 	if ((pip == NULL) || (MDI_PI(pip)->pi_prop == NULL)) {
4343 		return (DDI_PROP_NOT_FOUND);
4344 	}
4345 	rv = nvlist_lookup_string_array(MDI_PI(pip)->pi_prop, name, data,
4346 	    nelements);
4347 	return (i_map_nvlist_error_to_mdi(rv));
4348 }
4349 
4350 /*
4351  * mdi_prop_free():
4352  * 		Symmetrical function to ddi_prop_free(). nvlist_lookup_xx()
4353  *		functions return the pointer to actual property data and not a
4354  *		copy of it.  So the data returned is valid as long as
4355  *		mdi_pathinfo_t node is valid.
4356  */
4357 /*ARGSUSED*/
4358 int
4359 mdi_prop_free(void *data)
4360 {
4361 	return (DDI_PROP_SUCCESS);
4362 }
4363 
4364 /*ARGSUSED*/
4365 static void
4366 i_mdi_report_path_state(mdi_client_t *ct, mdi_pathinfo_t *pip)
4367 {
4368 	char		*phci_path, *ct_path;
4369 	char		*ct_status;
4370 	char		*status;
4371 	dev_info_t	*dip = ct->ct_dip;
4372 	char		lb_buf[64];
4373 
4374 	ASSERT(MDI_CLIENT_LOCKED(ct));
4375 	if ((dip == NULL) || (ddi_get_instance(dip) == -1) ||
4376 	    (MDI_CLIENT_IS_REPORT_DEV_NEEDED(ct) == 0)) {
4377 		return;
4378 	}
4379 	if (MDI_CLIENT_STATE(ct) == MDI_CLIENT_STATE_OPTIMAL) {
4380 		ct_status = "optimal";
4381 	} else if (MDI_CLIENT_STATE(ct) == MDI_CLIENT_STATE_DEGRADED) {
4382 		ct_status = "degraded";
4383 	} else if (MDI_CLIENT_STATE(ct) == MDI_CLIENT_STATE_FAILED) {
4384 		ct_status = "failed";
4385 	} else {
4386 		ct_status = "unknown";
4387 	}
4388 
4389 	if (MDI_PI_IS_OFFLINE(pip)) {
4390 		status = "offline";
4391 	} else if (MDI_PI_IS_ONLINE(pip)) {
4392 		status = "online";
4393 	} else if (MDI_PI_IS_STANDBY(pip)) {
4394 		status = "standby";
4395 	} else if (MDI_PI_IS_FAULT(pip)) {
4396 		status = "faulted";
4397 	} else {
4398 		status = "unknown";
4399 	}
4400 
4401 	if (ct->ct_lb == LOAD_BALANCE_LBA) {
4402 		(void) snprintf(lb_buf, sizeof (lb_buf),
4403 		    "%s, region-size: %d", mdi_load_balance_lba,
4404 			ct->ct_lb_args->region_size);
4405 	} else if (ct->ct_lb == LOAD_BALANCE_NONE) {
4406 		(void) snprintf(lb_buf, sizeof (lb_buf),
4407 		    "%s", mdi_load_balance_none);
4408 	} else {
4409 		(void) snprintf(lb_buf, sizeof (lb_buf), "%s",
4410 		    mdi_load_balance_rr);
4411 	}
4412 
4413 	if (dip) {
4414 		ct_path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4415 		phci_path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4416 		cmn_err(CE_CONT, "?%s (%s%d) multipath status: %s, "
4417 		    "path %s (%s%d) to target address: %s is %s"
4418 		    " Load balancing: %s\n",
4419 		    ddi_pathname(dip, ct_path), ddi_driver_name(dip),
4420 		    ddi_get_instance(dip), ct_status,
4421 		    ddi_pathname(MDI_PI(pip)->pi_phci->ph_dip, phci_path),
4422 		    ddi_driver_name(MDI_PI(pip)->pi_phci->ph_dip),
4423 		    ddi_get_instance(MDI_PI(pip)->pi_phci->ph_dip),
4424 		    MDI_PI(pip)->pi_addr, status, lb_buf);
4425 		kmem_free(phci_path, MAXPATHLEN);
4426 		kmem_free(ct_path, MAXPATHLEN);
4427 		MDI_CLIENT_CLEAR_REPORT_DEV_NEEDED(ct);
4428 	}
4429 }
4430 
4431 #ifdef	DEBUG
4432 /*
4433  * i_mdi_log():
4434  *		Utility function for error message management
4435  *
4436  */
4437 /*PRINTFLIKE3*/
4438 static void
4439 i_mdi_log(int level, dev_info_t *dip, const char *fmt, ...)
4440 {
4441 	char		name[MAXNAMELEN];
4442 	char		buf[MAXNAMELEN];
4443 	char		*bp;
4444 	va_list		ap;
4445 	int		log_only = 0;
4446 	int		boot_only = 0;
4447 	int		console_only = 0;
4448 
4449 	if (dip) {
4450 		(void) snprintf(name, MAXNAMELEN, "%s%d: ",
4451 		    ddi_node_name(dip), ddi_get_instance(dip));
4452 	} else {
4453 		name[0] = 0;
4454 	}
4455 
4456 	va_start(ap, fmt);
4457 	(void) vsnprintf(buf, MAXNAMELEN, fmt, ap);
4458 	va_end(ap);
4459 
4460 	switch (buf[0]) {
4461 	case '!':
4462 		bp = &buf[1];
4463 		log_only = 1;
4464 		break;
4465 	case '?':
4466 		bp = &buf[1];
4467 		boot_only = 1;
4468 		break;
4469 	case '^':
4470 		bp = &buf[1];
4471 		console_only = 1;
4472 		break;
4473 	default:
4474 		bp = buf;
4475 		break;
4476 	}
4477 	if (mdi_debug_logonly) {
4478 		log_only = 1;
4479 		boot_only = 0;
4480 		console_only = 0;
4481 	}
4482 
4483 	switch (level) {
4484 	case CE_NOTE:
4485 		level = CE_CONT;
4486 		/* FALLTHROUGH */
4487 	case CE_CONT:
4488 	case CE_WARN:
4489 	case CE_PANIC:
4490 		if (boot_only) {
4491 			cmn_err(level, "?mdi: %s%s", name, bp);
4492 		} else if (console_only) {
4493 			cmn_err(level, "^mdi: %s%s", name, bp);
4494 		} else if (log_only) {
4495 			cmn_err(level, "!mdi: %s%s", name, bp);
4496 		} else {
4497 			cmn_err(level, "mdi: %s%s", name, bp);
4498 		}
4499 		break;
4500 	default:
4501 		cmn_err(level, "mdi: %s%s", name, bp);
4502 		break;
4503 	}
4504 }
4505 #endif	/* DEBUG */
4506 
4507 void
4508 i_mdi_client_online(dev_info_t *ct_dip)
4509 {
4510 	mdi_client_t	*ct;
4511 
4512 	/*
4513 	 * Client online notification. Mark client state as online
4514 	 * restore our binding with dev_info node
4515 	 */
4516 	ct = i_devi_get_client(ct_dip);
4517 	ASSERT(ct != NULL);
4518 	MDI_CLIENT_LOCK(ct);
4519 	MDI_CLIENT_SET_ONLINE(ct);
4520 	/* catch for any memory leaks */
4521 	ASSERT((ct->ct_dip == NULL) || (ct->ct_dip == ct_dip));
4522 	ct->ct_dip = ct_dip;
4523 
4524 	if (ct->ct_power_cnt == 0)
4525 		(void) i_mdi_power_all_phci(ct);
4526 
4527 	MDI_DEBUG(4, (CE_NOTE, ct_dip, "i_mdi_client_online "
4528 	    "i_mdi_pm_hold_client %p\n", (void *)ct));
4529 	i_mdi_pm_hold_client(ct, 1);
4530 
4531 	MDI_CLIENT_UNLOCK(ct);
4532 }
4533 
4534 void
4535 i_mdi_phci_online(dev_info_t *ph_dip)
4536 {
4537 	mdi_phci_t	*ph;
4538 
4539 	/* pHCI online notification. Mark state accordingly */
4540 	ph = i_devi_get_phci(ph_dip);
4541 	ASSERT(ph != NULL);
4542 	MDI_PHCI_LOCK(ph);
4543 	MDI_PHCI_SET_ONLINE(ph);
4544 	MDI_PHCI_UNLOCK(ph);
4545 }
4546 
4547 /*
4548  * mdi_devi_online():
4549  * 		Online notification from NDI framework on pHCI/client
4550  *		device online.
4551  * Return Values:
4552  *		NDI_SUCCESS
4553  *		MDI_FAILURE
4554  */
4555 /*ARGSUSED*/
4556 int
4557 mdi_devi_online(dev_info_t *dip, uint_t flags)
4558 {
4559 	if (MDI_PHCI(dip)) {
4560 		i_mdi_phci_online(dip);
4561 	}
4562 
4563 	if (MDI_CLIENT(dip)) {
4564 		i_mdi_client_online(dip);
4565 	}
4566 	return (NDI_SUCCESS);
4567 }
4568 
4569 /*
4570  * mdi_devi_offline():
4571  * 		Offline notification from NDI framework on pHCI/Client device
4572  *		offline.
4573  *
4574  * Return Values:
4575  *		NDI_SUCCESS
4576  *		NDI_FAILURE
4577  */
4578 /*ARGSUSED*/
4579 int
4580 mdi_devi_offline(dev_info_t *dip, uint_t flags)
4581 {
4582 	int		rv = NDI_SUCCESS;
4583 
4584 	if (MDI_CLIENT(dip)) {
4585 		rv = i_mdi_client_offline(dip, flags);
4586 		if (rv != NDI_SUCCESS)
4587 			return (rv);
4588 	}
4589 
4590 	if (MDI_PHCI(dip)) {
4591 		rv = i_mdi_phci_offline(dip, flags);
4592 
4593 		if ((rv != NDI_SUCCESS) && MDI_CLIENT(dip)) {
4594 			/* set client back online */
4595 			i_mdi_client_online(dip);
4596 		}
4597 	}
4598 
4599 	return (rv);
4600 }
4601 
4602 /*ARGSUSED*/
4603 static int
4604 i_mdi_phci_offline(dev_info_t *dip, uint_t flags)
4605 {
4606 	int		rv = NDI_SUCCESS;
4607 	mdi_phci_t	*ph;
4608 	mdi_client_t	*ct;
4609 	mdi_pathinfo_t	*pip;
4610 	mdi_pathinfo_t	*next;
4611 	mdi_pathinfo_t	*failed_pip = NULL;
4612 	dev_info_t	*cdip;
4613 
4614 	/*
4615 	 * pHCI component offline notification
4616 	 * Make sure that this pHCI instance is free to be offlined.
4617 	 * If it is OK to proceed, Offline and remove all the child
4618 	 * mdi_pathinfo nodes.  This process automatically offlines
4619 	 * corresponding client devices, for which this pHCI provides
4620 	 * critical services.
4621 	 */
4622 	ph = i_devi_get_phci(dip);
4623 	MDI_DEBUG(2, (CE_NOTE, dip, "!mdi_phci_offline called %p %p\n",
4624 	    (void *)dip, (void *)ph));
4625 	if (ph == NULL) {
4626 		return (rv);
4627 	}
4628 
4629 	MDI_PHCI_LOCK(ph);
4630 
4631 	if (MDI_PHCI_IS_OFFLINE(ph)) {
4632 		MDI_DEBUG(1, (CE_WARN, dip, "!pHCI %p already offlined",
4633 		    (void *)ph));
4634 		MDI_PHCI_UNLOCK(ph);
4635 		return (NDI_SUCCESS);
4636 	}
4637 
4638 	/*
4639 	 * Check to see if the pHCI can be offlined
4640 	 */
4641 	if (ph->ph_unstable) {
4642 		MDI_DEBUG(1, (CE_WARN, dip,
4643 		    "!One or more target devices are in transient "
4644 		    "state. This device can not be removed at "
4645 		    "this moment. Please try again later."));
4646 		MDI_PHCI_UNLOCK(ph);
4647 		return (NDI_BUSY);
4648 	}
4649 
4650 	pip = ph->ph_path_head;
4651 	while (pip != NULL) {
4652 		MDI_PI_LOCK(pip);
4653 		next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
4654 
4655 		/*
4656 		 * The mdi_pathinfo state is OK. Check the client state.
4657 		 * If failover in progress fail the pHCI from offlining
4658 		 */
4659 		ct = MDI_PI(pip)->pi_client;
4660 		i_mdi_client_lock(ct, pip);
4661 		if ((MDI_CLIENT_IS_FAILOVER_IN_PROGRESS(ct)) ||
4662 		    (ct->ct_unstable)) {
4663 			/*
4664 			 * Failover is in progress, Fail the DR
4665 			 */
4666 			MDI_DEBUG(1, (CE_WARN, dip,
4667 			    "!pHCI device (%s%d) is Busy. %s",
4668 			    ddi_driver_name(dip), ddi_get_instance(dip),
4669 			    "This device can not be removed at "
4670 			    "this moment. Please try again later."));
4671 			MDI_PI_UNLOCK(pip);
4672 			i_mdi_client_unlock(ct);
4673 			MDI_PHCI_UNLOCK(ph);
4674 			return (NDI_BUSY);
4675 		}
4676 		MDI_PI_UNLOCK(pip);
4677 
4678 		/*
4679 		 * Check to see of we are removing the last path of this
4680 		 * client device...
4681 		 */
4682 		cdip = ct->ct_dip;
4683 		if (cdip && (i_ddi_node_state(cdip) >= DS_INITIALIZED) &&
4684 		    (i_mdi_client_compute_state(ct, ph) ==
4685 		    MDI_CLIENT_STATE_FAILED)) {
4686 			i_mdi_client_unlock(ct);
4687 			MDI_PHCI_UNLOCK(ph);
4688 			if (ndi_devi_offline(cdip, 0) != NDI_SUCCESS) {
4689 				/*
4690 				 * ndi_devi_offline() failed.
4691 				 * This pHCI provides the critical path
4692 				 * to one or more client devices.
4693 				 * Return busy.
4694 				 */
4695 				MDI_PHCI_LOCK(ph);
4696 				MDI_DEBUG(1, (CE_WARN, dip,
4697 				    "!pHCI device (%s%d) is Busy. %s",
4698 				    ddi_driver_name(dip), ddi_get_instance(dip),
4699 				    "This device can not be removed at "
4700 				    "this moment. Please try again later."));
4701 				failed_pip = pip;
4702 				break;
4703 			} else {
4704 				MDI_PHCI_LOCK(ph);
4705 				pip = next;
4706 			}
4707 		} else {
4708 			i_mdi_client_unlock(ct);
4709 			pip = next;
4710 		}
4711 	}
4712 
4713 	if (failed_pip) {
4714 		pip = ph->ph_path_head;
4715 		while (pip != failed_pip) {
4716 			MDI_PI_LOCK(pip);
4717 			next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
4718 			ct = MDI_PI(pip)->pi_client;
4719 			i_mdi_client_lock(ct, pip);
4720 			cdip = ct->ct_dip;
4721 			switch (MDI_CLIENT_STATE(ct)) {
4722 			case MDI_CLIENT_STATE_OPTIMAL:
4723 			case MDI_CLIENT_STATE_DEGRADED:
4724 				if (cdip) {
4725 					MDI_PI_UNLOCK(pip);
4726 					i_mdi_client_unlock(ct);
4727 					MDI_PHCI_UNLOCK(ph);
4728 					(void) ndi_devi_online(cdip, 0);
4729 					MDI_PHCI_LOCK(ph);
4730 					pip = next;
4731 					continue;
4732 				}
4733 				break;
4734 
4735 			case MDI_CLIENT_STATE_FAILED:
4736 				if (cdip) {
4737 					MDI_PI_UNLOCK(pip);
4738 					i_mdi_client_unlock(ct);
4739 					MDI_PHCI_UNLOCK(ph);
4740 					(void) ndi_devi_offline(cdip, 0);
4741 					MDI_PHCI_LOCK(ph);
4742 					pip = next;
4743 					continue;
4744 				}
4745 				break;
4746 			}
4747 			MDI_PI_UNLOCK(pip);
4748 			i_mdi_client_unlock(ct);
4749 			pip = next;
4750 		}
4751 		MDI_PHCI_UNLOCK(ph);
4752 		return (NDI_BUSY);
4753 	}
4754 
4755 	/*
4756 	 * Mark the pHCI as offline
4757 	 */
4758 	MDI_PHCI_SET_OFFLINE(ph);
4759 
4760 	/*
4761 	 * Mark the child mdi_pathinfo nodes as transient
4762 	 */
4763 	pip = ph->ph_path_head;
4764 	while (pip != NULL) {
4765 		MDI_PI_LOCK(pip);
4766 		next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
4767 		MDI_PI_SET_OFFLINING(pip);
4768 		MDI_PI_UNLOCK(pip);
4769 		pip = next;
4770 	}
4771 	MDI_PHCI_UNLOCK(ph);
4772 	/*
4773 	 * Give a chance for any pending commands to execute
4774 	 */
4775 	delay(1);
4776 	MDI_PHCI_LOCK(ph);
4777 	pip = ph->ph_path_head;
4778 	while (pip != NULL) {
4779 		next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
4780 		(void) i_mdi_pi_offline(pip, flags);
4781 		MDI_PI_LOCK(pip);
4782 		ct = MDI_PI(pip)->pi_client;
4783 		if (!MDI_PI_IS_OFFLINE(pip)) {
4784 			MDI_DEBUG(1, (CE_WARN, dip,
4785 			    "!pHCI device (%s%d) is Busy. %s",
4786 			    ddi_driver_name(dip), ddi_get_instance(dip),
4787 			    "This device can not be removed at "
4788 			    "this moment. Please try again later."));
4789 			MDI_PI_UNLOCK(pip);
4790 			MDI_PHCI_SET_ONLINE(ph);
4791 			MDI_PHCI_UNLOCK(ph);
4792 			return (NDI_BUSY);
4793 		}
4794 		MDI_PI_UNLOCK(pip);
4795 		pip = next;
4796 	}
4797 	MDI_PHCI_UNLOCK(ph);
4798 
4799 	return (rv);
4800 }
4801 
4802 /*ARGSUSED*/
4803 static int
4804 i_mdi_client_offline(dev_info_t *dip, uint_t flags)
4805 {
4806 	int		rv = NDI_SUCCESS;
4807 	mdi_client_t	*ct;
4808 
4809 	/*
4810 	 * Client component to go offline.  Make sure that we are
4811 	 * not in failing over state and update client state
4812 	 * accordingly
4813 	 */
4814 	ct = i_devi_get_client(dip);
4815 	MDI_DEBUG(2, (CE_NOTE, dip, "!i_mdi_client_offline called %p %p\n",
4816 	    (void *)dip, (void *)ct));
4817 	if (ct != NULL) {
4818 		MDI_CLIENT_LOCK(ct);
4819 		if (ct->ct_unstable) {
4820 			/*
4821 			 * One or more paths are in transient state,
4822 			 * Dont allow offline of a client device
4823 			 */
4824 			MDI_DEBUG(1, (CE_WARN, dip,
4825 			    "!One or more paths to this device is "
4826 			    "in transient state. This device can not "
4827 			    "be removed at this moment. "
4828 			    "Please try again later."));
4829 			MDI_CLIENT_UNLOCK(ct);
4830 			return (NDI_BUSY);
4831 		}
4832 		if (MDI_CLIENT_IS_FAILOVER_IN_PROGRESS(ct)) {
4833 			/*
4834 			 * Failover is in progress, Dont allow DR of
4835 			 * a client device
4836 			 */
4837 			MDI_DEBUG(1, (CE_WARN, dip,
4838 			    "!Client device (%s%d) is Busy. %s",
4839 			    ddi_driver_name(dip), ddi_get_instance(dip),
4840 			    "This device can not be removed at "
4841 			    "this moment. Please try again later."));
4842 			MDI_CLIENT_UNLOCK(ct);
4843 			return (NDI_BUSY);
4844 		}
4845 		MDI_CLIENT_SET_OFFLINE(ct);
4846 
4847 		/*
4848 		 * Unbind our relationship with the dev_info node
4849 		 */
4850 		if (flags & NDI_DEVI_REMOVE) {
4851 			ct->ct_dip = NULL;
4852 		}
4853 		MDI_CLIENT_UNLOCK(ct);
4854 	}
4855 	return (rv);
4856 }
4857 
4858 /*
4859  * mdi_pre_attach():
4860  *		Pre attach() notification handler
4861  */
4862 /*ARGSUSED*/
4863 int
4864 mdi_pre_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
4865 {
4866 	/* don't support old DDI_PM_RESUME */
4867 	if ((DEVI(dip)->devi_mdi_component != MDI_COMPONENT_NONE) &&
4868 	    (cmd == DDI_PM_RESUME))
4869 		return (DDI_FAILURE);
4870 
4871 	return (DDI_SUCCESS);
4872 }
4873 
4874 /*
4875  * mdi_post_attach():
4876  *		Post attach() notification handler
4877  */
4878 /*ARGSUSED*/
4879 void
4880 mdi_post_attach(dev_info_t *dip, ddi_attach_cmd_t cmd, int error)
4881 {
4882 	mdi_phci_t	*ph;
4883 	mdi_client_t	*ct;
4884 	mdi_pathinfo_t	*pip;
4885 
4886 	if (MDI_PHCI(dip)) {
4887 		ph = i_devi_get_phci(dip);
4888 		ASSERT(ph != NULL);
4889 
4890 		MDI_PHCI_LOCK(ph);
4891 		switch (cmd) {
4892 		case DDI_ATTACH:
4893 			MDI_DEBUG(2, (CE_NOTE, dip,
4894 			    "!pHCI post_attach: called %p\n", (void *)ph));
4895 			if (error == DDI_SUCCESS) {
4896 				MDI_PHCI_SET_ATTACH(ph);
4897 			} else {
4898 				MDI_DEBUG(1, (CE_NOTE, dip,
4899 				    "!pHCI post_attach: failed error=%d\n",
4900 				    error));
4901 				MDI_PHCI_SET_DETACH(ph);
4902 			}
4903 			break;
4904 
4905 		case DDI_RESUME:
4906 			MDI_DEBUG(2, (CE_NOTE, dip,
4907 			    "!pHCI post_resume: called %p\n", (void *)ph));
4908 			if (error == DDI_SUCCESS) {
4909 				MDI_PHCI_SET_RESUME(ph);
4910 			} else {
4911 				MDI_DEBUG(1, (CE_NOTE, dip,
4912 				    "!pHCI post_resume: failed error=%d\n",
4913 				    error));
4914 				MDI_PHCI_SET_SUSPEND(ph);
4915 			}
4916 			break;
4917 		}
4918 		MDI_PHCI_UNLOCK(ph);
4919 	}
4920 
4921 	if (MDI_CLIENT(dip)) {
4922 		ct = i_devi_get_client(dip);
4923 		ASSERT(ct != NULL);
4924 
4925 		MDI_CLIENT_LOCK(ct);
4926 		switch (cmd) {
4927 		case DDI_ATTACH:
4928 			MDI_DEBUG(2, (CE_NOTE, dip,
4929 			    "!Client post_attach: called %p\n", (void *)ct));
4930 			if (error != DDI_SUCCESS) {
4931 				MDI_DEBUG(1, (CE_NOTE, dip,
4932 				    "!Client post_attach: failed error=%d\n",
4933 				    error));
4934 				MDI_CLIENT_SET_DETACH(ct);
4935 				MDI_DEBUG(4, (CE_WARN, dip,
4936 				    "mdi_post_attach i_mdi_pm_reset_client\n"));
4937 				i_mdi_pm_reset_client(ct);
4938 				break;
4939 			}
4940 
4941 			/*
4942 			 * Client device has successfully attached.
4943 			 * Create kstats for any pathinfo structures
4944 			 * initially associated with this client.
4945 			 */
4946 			for (pip = ct->ct_path_head; pip != NULL;
4947 			    pip = (mdi_pathinfo_t *)
4948 			    MDI_PI(pip)->pi_client_link) {
4949 				if (!MDI_PI_IS_OFFLINE(pip)) {
4950 					(void) i_mdi_pi_kstat_create(pip);
4951 					i_mdi_report_path_state(ct, pip);
4952 				}
4953 			}
4954 			MDI_CLIENT_SET_ATTACH(ct);
4955 			break;
4956 
4957 		case DDI_RESUME:
4958 			MDI_DEBUG(2, (CE_NOTE, dip,
4959 			    "!Client post_attach: called %p\n", (void *)ct));
4960 			if (error == DDI_SUCCESS) {
4961 				MDI_CLIENT_SET_RESUME(ct);
4962 			} else {
4963 				MDI_DEBUG(1, (CE_NOTE, dip,
4964 				    "!Client post_resume: failed error=%d\n",
4965 				    error));
4966 				MDI_CLIENT_SET_SUSPEND(ct);
4967 			}
4968 			break;
4969 		}
4970 		MDI_CLIENT_UNLOCK(ct);
4971 	}
4972 }
4973 
4974 /*
4975  * mdi_pre_detach():
4976  *		Pre detach notification handler
4977  */
4978 /*ARGSUSED*/
4979 int
4980 mdi_pre_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
4981 {
4982 	int rv = DDI_SUCCESS;
4983 
4984 	if (MDI_CLIENT(dip)) {
4985 		(void) i_mdi_client_pre_detach(dip, cmd);
4986 	}
4987 
4988 	if (MDI_PHCI(dip)) {
4989 		rv = i_mdi_phci_pre_detach(dip, cmd);
4990 	}
4991 
4992 	return (rv);
4993 }
4994 
4995 /*ARGSUSED*/
4996 static int
4997 i_mdi_phci_pre_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
4998 {
4999 	int		rv = DDI_SUCCESS;
5000 	mdi_phci_t	*ph;
5001 	mdi_client_t	*ct;
5002 	mdi_pathinfo_t	*pip;
5003 	mdi_pathinfo_t	*failed_pip = NULL;
5004 	mdi_pathinfo_t	*next;
5005 
5006 	ph = i_devi_get_phci(dip);
5007 	if (ph == NULL) {
5008 		return (rv);
5009 	}
5010 
5011 	MDI_PHCI_LOCK(ph);
5012 	switch (cmd) {
5013 	case DDI_DETACH:
5014 		MDI_DEBUG(2, (CE_NOTE, dip,
5015 		    "!pHCI pre_detach: called %p\n", (void *)ph));
5016 		if (!MDI_PHCI_IS_OFFLINE(ph)) {
5017 			/*
5018 			 * mdi_pathinfo nodes are still attached to
5019 			 * this pHCI. Fail the detach for this pHCI.
5020 			 */
5021 			MDI_DEBUG(2, (CE_WARN, dip,
5022 			    "!pHCI pre_detach: "
5023 			    "mdi_pathinfo nodes are still attached "
5024 			    "%p\n", (void *)ph));
5025 			rv = DDI_FAILURE;
5026 			break;
5027 		}
5028 		MDI_PHCI_SET_DETACH(ph);
5029 		break;
5030 
5031 	case DDI_SUSPEND:
5032 		/*
5033 		 * pHCI is getting suspended.  Since mpxio client
5034 		 * devices may not be suspended at this point, to avoid
5035 		 * a potential stack overflow, it is important to suspend
5036 		 * client devices before pHCI can be suspended.
5037 		 */
5038 
5039 		MDI_DEBUG(2, (CE_NOTE, dip,
5040 		    "!pHCI pre_suspend: called %p\n", (void *)ph));
5041 		/*
5042 		 * Suspend all the client devices accessible through this pHCI
5043 		 */
5044 		pip = ph->ph_path_head;
5045 		while (pip != NULL && rv == DDI_SUCCESS) {
5046 			dev_info_t *cdip;
5047 			MDI_PI_LOCK(pip);
5048 			next =
5049 			    (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
5050 			ct = MDI_PI(pip)->pi_client;
5051 			i_mdi_client_lock(ct, pip);
5052 			cdip = ct->ct_dip;
5053 			MDI_PI_UNLOCK(pip);
5054 			if ((MDI_CLIENT_IS_DETACHED(ct) == 0) &&
5055 			    MDI_CLIENT_IS_SUSPENDED(ct) == 0) {
5056 				i_mdi_client_unlock(ct);
5057 				if ((rv = devi_detach(cdip, DDI_SUSPEND)) !=
5058 				    DDI_SUCCESS) {
5059 					/*
5060 					 * Suspend of one of the client
5061 					 * device has failed.
5062 					 */
5063 					MDI_DEBUG(1, (CE_WARN, dip,
5064 					    "!Suspend of device (%s%d) failed.",
5065 					    ddi_driver_name(cdip),
5066 					    ddi_get_instance(cdip)));
5067 					failed_pip = pip;
5068 					break;
5069 				}
5070 			} else {
5071 				i_mdi_client_unlock(ct);
5072 			}
5073 			pip = next;
5074 		}
5075 
5076 		if (rv == DDI_SUCCESS) {
5077 			/*
5078 			 * Suspend of client devices is complete. Proceed
5079 			 * with pHCI suspend.
5080 			 */
5081 			MDI_PHCI_SET_SUSPEND(ph);
5082 		} else {
5083 			/*
5084 			 * Revert back all the suspended client device states
5085 			 * to converse.
5086 			 */
5087 			pip = ph->ph_path_head;
5088 			while (pip != failed_pip) {
5089 				dev_info_t *cdip;
5090 				MDI_PI_LOCK(pip);
5091 				next =
5092 				    (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
5093 				ct = MDI_PI(pip)->pi_client;
5094 				i_mdi_client_lock(ct, pip);
5095 				cdip = ct->ct_dip;
5096 				MDI_PI_UNLOCK(pip);
5097 				if (MDI_CLIENT_IS_SUSPENDED(ct)) {
5098 					i_mdi_client_unlock(ct);
5099 					(void) devi_attach(cdip, DDI_RESUME);
5100 				} else {
5101 					i_mdi_client_unlock(ct);
5102 				}
5103 				pip = next;
5104 			}
5105 		}
5106 		break;
5107 
5108 	default:
5109 		rv = DDI_FAILURE;
5110 		break;
5111 	}
5112 	MDI_PHCI_UNLOCK(ph);
5113 	return (rv);
5114 }
5115 
5116 /*ARGSUSED*/
5117 static int
5118 i_mdi_client_pre_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
5119 {
5120 	int		rv = DDI_SUCCESS;
5121 	mdi_client_t	*ct;
5122 
5123 	ct = i_devi_get_client(dip);
5124 	if (ct == NULL) {
5125 		return (rv);
5126 	}
5127 
5128 	MDI_CLIENT_LOCK(ct);
5129 	switch (cmd) {
5130 	case DDI_DETACH:
5131 		MDI_DEBUG(2, (CE_NOTE, dip,
5132 		    "!Client pre_detach: called %p\n", (void *)ct));
5133 		MDI_CLIENT_SET_DETACH(ct);
5134 		break;
5135 
5136 	case DDI_SUSPEND:
5137 		MDI_DEBUG(2, (CE_NOTE, dip,
5138 		    "!Client pre_suspend: called %p\n", (void *)ct));
5139 		MDI_CLIENT_SET_SUSPEND(ct);
5140 		break;
5141 
5142 	default:
5143 		rv = DDI_FAILURE;
5144 		break;
5145 	}
5146 	MDI_CLIENT_UNLOCK(ct);
5147 	return (rv);
5148 }
5149 
5150 /*
5151  * mdi_post_detach():
5152  *		Post detach notification handler
5153  */
5154 /*ARGSUSED*/
5155 void
5156 mdi_post_detach(dev_info_t *dip, ddi_detach_cmd_t cmd, int error)
5157 {
5158 	/*
5159 	 * Detach/Suspend of mpxio component failed. Update our state
5160 	 * too
5161 	 */
5162 	if (MDI_PHCI(dip))
5163 		i_mdi_phci_post_detach(dip, cmd, error);
5164 
5165 	if (MDI_CLIENT(dip))
5166 		i_mdi_client_post_detach(dip, cmd, error);
5167 }
5168 
5169 /*ARGSUSED*/
5170 static void
5171 i_mdi_phci_post_detach(dev_info_t *dip, ddi_detach_cmd_t cmd, int error)
5172 {
5173 	mdi_phci_t	*ph;
5174 
5175 	/*
5176 	 * Detach/Suspend of phci component failed. Update our state
5177 	 * too
5178 	 */
5179 	ph = i_devi_get_phci(dip);
5180 	if (ph == NULL) {
5181 		return;
5182 	}
5183 
5184 	MDI_PHCI_LOCK(ph);
5185 	/*
5186 	 * Detach of pHCI failed. Restore back converse
5187 	 * state
5188 	 */
5189 	switch (cmd) {
5190 	case DDI_DETACH:
5191 		MDI_DEBUG(2, (CE_NOTE, dip,
5192 		    "!pHCI post_detach: called %p\n", (void *)ph));
5193 		if (error != DDI_SUCCESS)
5194 			MDI_PHCI_SET_ATTACH(ph);
5195 		break;
5196 
5197 	case DDI_SUSPEND:
5198 		MDI_DEBUG(2, (CE_NOTE, dip,
5199 		    "!pHCI post_suspend: called %p\n", (void *)ph));
5200 		if (error != DDI_SUCCESS)
5201 			MDI_PHCI_SET_RESUME(ph);
5202 		break;
5203 	}
5204 	MDI_PHCI_UNLOCK(ph);
5205 }
5206 
5207 /*ARGSUSED*/
5208 static void
5209 i_mdi_client_post_detach(dev_info_t *dip, ddi_detach_cmd_t cmd, int error)
5210 {
5211 	mdi_client_t	*ct;
5212 
5213 	ct = i_devi_get_client(dip);
5214 	if (ct == NULL) {
5215 		return;
5216 	}
5217 	MDI_CLIENT_LOCK(ct);
5218 	/*
5219 	 * Detach of Client failed. Restore back converse
5220 	 * state
5221 	 */
5222 	switch (cmd) {
5223 	case DDI_DETACH:
5224 		MDI_DEBUG(2, (CE_NOTE, dip,
5225 		    "!Client post_detach: called %p\n", (void *)ct));
5226 		if (DEVI_IS_ATTACHING(ct->ct_dip)) {
5227 			MDI_DEBUG(4, (CE_NOTE, dip, "i_mdi_client_post_detach "
5228 			    "i_mdi_pm_rele_client\n"));
5229 			i_mdi_pm_rele_client(ct, ct->ct_path_count);
5230 		} else {
5231 			MDI_DEBUG(4, (CE_NOTE, dip, "i_mdi_client_post_detach "
5232 			    "i_mdi_pm_reset_client\n"));
5233 			i_mdi_pm_reset_client(ct);
5234 		}
5235 		if (error != DDI_SUCCESS)
5236 			MDI_CLIENT_SET_ATTACH(ct);
5237 		break;
5238 
5239 	case DDI_SUSPEND:
5240 		MDI_DEBUG(2, (CE_NOTE, dip,
5241 		    "!Client post_suspend: called %p\n", (void *)ct));
5242 		if (error != DDI_SUCCESS)
5243 			MDI_CLIENT_SET_RESUME(ct);
5244 		break;
5245 	}
5246 	MDI_CLIENT_UNLOCK(ct);
5247 }
5248 
5249 /*
5250  * create and install per-path (client - pHCI) statistics
5251  * I/O stats supported: nread, nwritten, reads, and writes
5252  * Error stats - hard errors, soft errors, & transport errors
5253  */
5254 static int
5255 i_mdi_pi_kstat_create(mdi_pathinfo_t *pip)
5256 {
5257 
5258 	dev_info_t *client = MDI_PI(pip)->pi_client->ct_dip;
5259 	dev_info_t *ppath = MDI_PI(pip)->pi_phci->ph_dip;
5260 	char ksname[KSTAT_STRLEN];
5261 	mdi_pathinfo_t *cpip;
5262 	const char *err_postfix = ",err";
5263 	kstat_t	*kiosp, *kerrsp;
5264 	struct pi_errs	*nsp;
5265 	struct mdi_pi_kstats *mdi_statp;
5266 
5267 	ASSERT(client != NULL && ppath != NULL);
5268 
5269 	ASSERT(MDI_CLIENT_LOCKED(MDI_PI(pip)->pi_client));
5270 
5271 	if (MDI_PI(pip)->pi_kstats != NULL)
5272 		return (MDI_SUCCESS);
5273 
5274 	for (cpip = MDI_PI(pip)->pi_client->ct_path_head; cpip != NULL;
5275 	    cpip = (mdi_pathinfo_t *)(MDI_PI(cpip)->pi_client_link)) {
5276 		if ((cpip == pip) || MDI_PI_IS_OFFLINE(pip))
5277 			continue;
5278 		/*
5279 		 * We have found a different path with same parent
5280 		 * kstats for a given client-pHCI are common
5281 		 */
5282 		if ((MDI_PI(cpip)->pi_phci->ph_dip == ppath) &&
5283 		    (MDI_PI(cpip)->pi_kstats != NULL)) {
5284 			MDI_PI(cpip)->pi_kstats->pi_kstat_ref++;
5285 			MDI_PI(pip)->pi_kstats = MDI_PI(cpip)->pi_kstats;
5286 			return (MDI_SUCCESS);
5287 		}
5288 	}
5289 
5290 	/*
5291 	 * stats are named as follows: TGTx.HBAy, e.g. "ssd0.fp0"
5292 	 * clamp length of name against max length of error kstat name
5293 	 */
5294 	if (snprintf(ksname, KSTAT_STRLEN, "%s%d.%s%d",
5295 	    ddi_driver_name(client), ddi_get_instance(client),
5296 	    ddi_driver_name(ppath), ddi_get_instance(ppath)) >
5297 	    (KSTAT_STRLEN - strlen(err_postfix))) {
5298 		return (MDI_FAILURE);
5299 	}
5300 	if ((kiosp = kstat_create("mdi", 0, ksname, "iopath",
5301 	    KSTAT_TYPE_IO, 1, 0)) == NULL) {
5302 		return (MDI_FAILURE);
5303 	}
5304 
5305 	(void) strcat(ksname, err_postfix);
5306 	kerrsp = kstat_create("mdi", 0, ksname, "iopath_errors",
5307 	    KSTAT_TYPE_NAMED,
5308 	    sizeof (struct pi_errs) / sizeof (kstat_named_t), 0);
5309 
5310 	if (kerrsp == NULL) {
5311 		kstat_delete(kiosp);
5312 		return (MDI_FAILURE);
5313 	}
5314 
5315 	nsp = (struct pi_errs *)kerrsp->ks_data;
5316 	kstat_named_init(&nsp->pi_softerrs, "Soft Errors", KSTAT_DATA_UINT32);
5317 	kstat_named_init(&nsp->pi_harderrs, "Hard Errors", KSTAT_DATA_UINT32);
5318 	kstat_named_init(&nsp->pi_transerrs, "Transport Errors",
5319 	    KSTAT_DATA_UINT32);
5320 	kstat_named_init(&nsp->pi_icnt_busy, "Interconnect Busy",
5321 	    KSTAT_DATA_UINT32);
5322 	kstat_named_init(&nsp->pi_icnt_errors, "Interconnect Errors",
5323 	    KSTAT_DATA_UINT32);
5324 	kstat_named_init(&nsp->pi_phci_rsrc, "pHCI No Resources",
5325 	    KSTAT_DATA_UINT32);
5326 	kstat_named_init(&nsp->pi_phci_localerr, "pHCI Local Errors",
5327 	    KSTAT_DATA_UINT32);
5328 	kstat_named_init(&nsp->pi_phci_invstate, "pHCI Invalid State",
5329 	    KSTAT_DATA_UINT32);
5330 	kstat_named_init(&nsp->pi_failedfrom, "Failed From",
5331 	    KSTAT_DATA_UINT32);
5332 	kstat_named_init(&nsp->pi_failedto, "Failed To", KSTAT_DATA_UINT32);
5333 
5334 	mdi_statp = kmem_alloc(sizeof (*mdi_statp), KM_SLEEP);
5335 	mdi_statp->pi_kstat_ref = 1;
5336 	mdi_statp->pi_kstat_iostats = kiosp;
5337 	mdi_statp->pi_kstat_errstats = kerrsp;
5338 	kstat_install(kiosp);
5339 	kstat_install(kerrsp);
5340 	MDI_PI(pip)->pi_kstats = mdi_statp;
5341 	return (MDI_SUCCESS);
5342 }
5343 
5344 /*
5345  * destroy per-path properties
5346  */
5347 static void
5348 i_mdi_pi_kstat_destroy(mdi_pathinfo_t *pip)
5349 {
5350 
5351 	struct mdi_pi_kstats *mdi_statp;
5352 
5353 	if ((mdi_statp = MDI_PI(pip)->pi_kstats) == NULL)
5354 		return;
5355 
5356 	MDI_PI(pip)->pi_kstats = NULL;
5357 
5358 	/*
5359 	 * the kstat may be shared between multiple pathinfo nodes
5360 	 * decrement this pathinfo's usage, removing the kstats
5361 	 * themselves when the last pathinfo reference is removed.
5362 	 */
5363 	ASSERT(mdi_statp->pi_kstat_ref > 0);
5364 	if (--mdi_statp->pi_kstat_ref != 0)
5365 		return;
5366 
5367 	kstat_delete(mdi_statp->pi_kstat_iostats);
5368 	kstat_delete(mdi_statp->pi_kstat_errstats);
5369 	kmem_free(mdi_statp, sizeof (*mdi_statp));
5370 }
5371 
5372 /*
5373  * update I/O paths KSTATS
5374  */
5375 void
5376 mdi_pi_kstat_iosupdate(mdi_pathinfo_t *pip, struct buf *bp)
5377 {
5378 	kstat_t *iostatp;
5379 	size_t xfer_cnt;
5380 
5381 	ASSERT(pip != NULL);
5382 
5383 	/*
5384 	 * I/O can be driven across a path prior to having path
5385 	 * statistics available, i.e. probe(9e).
5386 	 */
5387 	if (bp != NULL && MDI_PI(pip)->pi_kstats != NULL) {
5388 		iostatp = MDI_PI(pip)->pi_kstats->pi_kstat_iostats;
5389 		xfer_cnt = bp->b_bcount - bp->b_resid;
5390 		if (bp->b_flags & B_READ) {
5391 			KSTAT_IO_PTR(iostatp)->reads++;
5392 			KSTAT_IO_PTR(iostatp)->nread += xfer_cnt;
5393 		} else {
5394 			KSTAT_IO_PTR(iostatp)->writes++;
5395 			KSTAT_IO_PTR(iostatp)->nwritten += xfer_cnt;
5396 		}
5397 	}
5398 }
5399 
5400 /*
5401  * Enable the path(specific client/target/initiator)
5402  * Enabling a path means that MPxIO may select the enabled path for routing
5403  * future I/O requests, subject to other path state constraints.
5404  */
5405 int
5406 mdi_pi_enable_path(mdi_pathinfo_t *pip, int flags)
5407 {
5408 	mdi_phci_t	*ph;
5409 
5410 	ph = i_devi_get_phci(mdi_pi_get_phci(pip));
5411 	if (ph == NULL) {
5412 		MDI_DEBUG(1, (CE_NOTE, NULL, "!mdi_pi_enable_path:"
5413 			" failed. pip: %p ph = NULL\n", (void *)pip));
5414 		return (MDI_FAILURE);
5415 	}
5416 
5417 	(void) i_mdi_enable_disable_path(pip, ph->ph_vhci, flags,
5418 		MDI_ENABLE_OP);
5419 	MDI_DEBUG(5, (CE_NOTE, NULL, "!mdi_pi_enable_path:"
5420 		" Returning success pip = %p. ph = %p\n",
5421 		(void *)pip, (void *)ph));
5422 	return (MDI_SUCCESS);
5423 
5424 }
5425 
5426 /*
5427  * Disable the path (specific client/target/initiator)
5428  * Disabling a path means that MPxIO will not select the disabled path for
5429  * routing any new I/O requests.
5430  */
5431 int
5432 mdi_pi_disable_path(mdi_pathinfo_t *pip, int flags)
5433 {
5434 	mdi_phci_t	*ph;
5435 
5436 	ph = i_devi_get_phci(mdi_pi_get_phci(pip));
5437 	if (ph == NULL) {
5438 		MDI_DEBUG(1, (CE_NOTE, NULL, "!mdi_pi_disable_path:"
5439 			" failed. pip: %p ph = NULL\n", (void *)pip));
5440 		return (MDI_FAILURE);
5441 	}
5442 
5443 	(void) i_mdi_enable_disable_path(pip,
5444 			ph->ph_vhci, flags, MDI_DISABLE_OP);
5445 	MDI_DEBUG(5, (CE_NOTE, NULL, "!mdi_pi_disable_path:"
5446 		"Returning success pip = %p. ph = %p",
5447 		(void *)pip, (void *)ph));
5448 	return (MDI_SUCCESS);
5449 }
5450 
5451 /*
5452  * disable the path to a particular pHCI (pHCI specified in the phci_path
5453  * argument) for a particular client (specified in the client_path argument).
5454  * Disabling a path means that MPxIO will not select the disabled path for
5455  * routing any new I/O requests.
5456  * NOTE: this will be removed once the NWS files are changed to use the new
5457  * mdi_{enable,disable}_path interfaces
5458  */
5459 int
5460 mdi_pi_disable(dev_info_t *cdip, dev_info_t *pdip, int flags)
5461 {
5462 	return (i_mdi_pi_enable_disable(cdip, pdip, flags, MDI_DISABLE_OP));
5463 }
5464 
5465 /*
5466  * Enable the path to a particular pHCI (pHCI specified in the phci_path
5467  * argument) for a particular client (specified in the client_path argument).
5468  * Enabling a path means that MPxIO may select the enabled path for routing
5469  * future I/O requests, subject to other path state constraints.
5470  * NOTE: this will be removed once the NWS files are changed to use the new
5471  * mdi_{enable,disable}_path interfaces
5472  */
5473 
5474 int
5475 mdi_pi_enable(dev_info_t *cdip, dev_info_t *pdip, int flags)
5476 {
5477 	return (i_mdi_pi_enable_disable(cdip, pdip, flags, MDI_ENABLE_OP));
5478 }
5479 
5480 /*
5481  * Common routine for doing enable/disable.
5482  */
5483 static mdi_pathinfo_t *
5484 i_mdi_enable_disable_path(mdi_pathinfo_t *pip, mdi_vhci_t *vh, int flags,
5485 		int op)
5486 {
5487 	int		sync_flag = 0;
5488 	int		rv;
5489 	mdi_pathinfo_t 	*next;
5490 	int		(*f)() = NULL;
5491 
5492 	f = vh->vh_ops->vo_pi_state_change;
5493 
5494 	sync_flag = (flags << 8) & 0xf00;
5495 
5496 	/*
5497 	 * Do a callback into the mdi consumer to let it
5498 	 * know that path is about to get enabled/disabled.
5499 	 */
5500 	if (f != NULL) {
5501 		rv = (*f)(vh->vh_dip, pip, 0,
5502 			MDI_PI_EXT_STATE(pip),
5503 			MDI_EXT_STATE_CHANGE | sync_flag |
5504 			op | MDI_BEFORE_STATE_CHANGE);
5505 		if (rv != MDI_SUCCESS) {
5506 			MDI_DEBUG(2, (CE_WARN, vh->vh_dip,
5507 			"!vo_pi_state_change: failed rv = %x", rv));
5508 		}
5509 	}
5510 	MDI_PI_LOCK(pip);
5511 	next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_phci_link;
5512 
5513 	switch (flags) {
5514 		case USER_DISABLE:
5515 			if (op == MDI_DISABLE_OP) {
5516 				MDI_PI_SET_USER_DISABLE(pip);
5517 			} else {
5518 				MDI_PI_SET_USER_ENABLE(pip);
5519 			}
5520 			break;
5521 		case DRIVER_DISABLE:
5522 			if (op == MDI_DISABLE_OP) {
5523 				MDI_PI_SET_DRV_DISABLE(pip);
5524 			} else {
5525 				MDI_PI_SET_DRV_ENABLE(pip);
5526 			}
5527 			break;
5528 		case DRIVER_DISABLE_TRANSIENT:
5529 			if (op == MDI_DISABLE_OP && rv == MDI_SUCCESS) {
5530 				MDI_PI_SET_DRV_DISABLE_TRANS(pip);
5531 			} else {
5532 				MDI_PI_SET_DRV_ENABLE_TRANS(pip);
5533 			}
5534 			break;
5535 	}
5536 	MDI_PI_UNLOCK(pip);
5537 	/*
5538 	 * Do a callback into the mdi consumer to let it
5539 	 * know that path is now enabled/disabled.
5540 	 */
5541 	if (f != NULL) {
5542 		rv = (*f)(vh->vh_dip, pip, 0,
5543 			MDI_PI_EXT_STATE(pip),
5544 			MDI_EXT_STATE_CHANGE | sync_flag |
5545 			op | MDI_AFTER_STATE_CHANGE);
5546 		if (rv != MDI_SUCCESS) {
5547 			MDI_DEBUG(2, (CE_WARN, vh->vh_dip,
5548 			"!vo_pi_state_change: failed rv = %x", rv));
5549 		}
5550 	}
5551 	return (next);
5552 }
5553 
5554 /*
5555  * Common routine for doing enable/disable.
5556  * NOTE: this will be removed once the NWS files are changed to use the new
5557  * mdi_{enable,disable}_path has been putback
5558  */
5559 int
5560 i_mdi_pi_enable_disable(dev_info_t *cdip, dev_info_t *pdip, int flags, int op)
5561 {
5562 
5563 	mdi_phci_t	*ph;
5564 	mdi_vhci_t	*vh = NULL;
5565 	mdi_client_t	*ct;
5566 	mdi_pathinfo_t	*next, *pip;
5567 	int		found_it;
5568 
5569 	ph = i_devi_get_phci(pdip);
5570 	MDI_DEBUG(5, (CE_NOTE, NULL, "!i_mdi_pi_enable_disable: "
5571 		"Op = %d pdip = %p cdip = %p\n", op, (void *)pdip,
5572 		(void *)cdip));
5573 	if (ph == NULL) {
5574 		MDI_DEBUG(1, (CE_NOTE, NULL, "!i_mdi_pi_enable_disable:"
5575 			"Op %d failed. ph = NULL\n", op));
5576 		return (MDI_FAILURE);
5577 	}
5578 
5579 	if ((op != MDI_ENABLE_OP) && (op != MDI_DISABLE_OP)) {
5580 		MDI_DEBUG(1, (CE_NOTE, NULL, "!i_mdi_pi_enable_disable: "
5581 			"Op Invalid operation = %d\n", op));
5582 		return (MDI_FAILURE);
5583 	}
5584 
5585 	vh = ph->ph_vhci;
5586 
5587 	if (cdip == NULL) {
5588 		/*
5589 		 * Need to mark the Phci as enabled/disabled.
5590 		 */
5591 		MDI_DEBUG(3, (CE_NOTE, NULL, "!i_mdi_pi_enable_disable: "
5592 		"Op %d for the phci\n", op));
5593 		MDI_PHCI_LOCK(ph);
5594 		switch (flags) {
5595 			case USER_DISABLE:
5596 				if (op == MDI_DISABLE_OP) {
5597 					MDI_PHCI_SET_USER_DISABLE(ph);
5598 				} else {
5599 					MDI_PHCI_SET_USER_ENABLE(ph);
5600 				}
5601 				break;
5602 			case DRIVER_DISABLE:
5603 				if (op == MDI_DISABLE_OP) {
5604 					MDI_PHCI_SET_DRV_DISABLE(ph);
5605 				} else {
5606 					MDI_PHCI_SET_DRV_ENABLE(ph);
5607 				}
5608 				break;
5609 			case DRIVER_DISABLE_TRANSIENT:
5610 				if (op == MDI_DISABLE_OP) {
5611 					MDI_PHCI_SET_DRV_DISABLE_TRANSIENT(ph);
5612 				} else {
5613 					MDI_PHCI_SET_DRV_ENABLE_TRANSIENT(ph);
5614 				}
5615 				break;
5616 			default:
5617 				MDI_PHCI_UNLOCK(ph);
5618 				MDI_DEBUG(1, (CE_NOTE, NULL,
5619 				"!i_mdi_pi_enable_disable:"
5620 				" Invalid flag argument= %d\n", flags));
5621 		}
5622 
5623 		/*
5624 		 * Phci has been disabled. Now try to enable/disable
5625 		 * path info's to each client.
5626 		 */
5627 		pip = ph->ph_path_head;
5628 		while (pip != NULL) {
5629 			pip = i_mdi_enable_disable_path(pip, vh, flags, op);
5630 		}
5631 		MDI_PHCI_UNLOCK(ph);
5632 	} else {
5633 
5634 		/*
5635 		 * Disable a specific client.
5636 		 */
5637 		ct = i_devi_get_client(cdip);
5638 		if (ct == NULL) {
5639 			MDI_DEBUG(1, (CE_NOTE, NULL,
5640 			"!i_mdi_pi_enable_disable:"
5641 			" failed. ct = NULL operation = %d\n", op));
5642 			return (MDI_FAILURE);
5643 		}
5644 
5645 		MDI_CLIENT_LOCK(ct);
5646 		pip = ct->ct_path_head;
5647 		found_it = 0;
5648 		while (pip != NULL) {
5649 			MDI_PI_LOCK(pip);
5650 			next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
5651 			if (MDI_PI(pip)->pi_phci == ph) {
5652 				MDI_PI_UNLOCK(pip);
5653 				found_it = 1;
5654 				break;
5655 			}
5656 			MDI_PI_UNLOCK(pip);
5657 			pip = next;
5658 		}
5659 
5660 
5661 		MDI_CLIENT_UNLOCK(ct);
5662 		if (found_it == 0) {
5663 			MDI_DEBUG(1, (CE_NOTE, NULL,
5664 			"!i_mdi_pi_enable_disable:"
5665 			" failed. Could not find corresponding pip\n"));
5666 			return (MDI_FAILURE);
5667 		}
5668 
5669 		(void) i_mdi_enable_disable_path(pip, vh, flags, op);
5670 	}
5671 
5672 	MDI_DEBUG(5, (CE_NOTE, NULL, "!i_mdi_pi_enable_disable: "
5673 		"Op %d Returning success pdip = %p cdip = %p\n",
5674 		op, (void *)pdip, (void *)cdip));
5675 	return (MDI_SUCCESS);
5676 }
5677 
5678 /*
5679  * Ensure phci powered up
5680  */
5681 static void
5682 i_mdi_pm_hold_pip(mdi_pathinfo_t *pip)
5683 {
5684 	dev_info_t	*ph_dip;
5685 
5686 	ASSERT(pip != NULL);
5687 	ASSERT(MDI_PI_LOCKED(pip));
5688 
5689 	if (MDI_PI(pip)->pi_pm_held) {
5690 		return;
5691 	}
5692 
5693 	ph_dip = mdi_pi_get_phci(pip);
5694 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "i_mdi_pm_hold_pip for %s%d %p\n",
5695 	    ddi_get_name(ph_dip), ddi_get_instance(ph_dip), (void *)pip));
5696 	if (ph_dip == NULL) {
5697 		return;
5698 	}
5699 
5700 	MDI_PI_UNLOCK(pip);
5701 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "kidsupcnt was %d\n",
5702 	    DEVI(ph_dip)->devi_pm_kidsupcnt));
5703 
5704 	pm_hold_power(ph_dip);
5705 
5706 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "kidsupcnt is %d\n",
5707 	    DEVI(ph_dip)->devi_pm_kidsupcnt));
5708 	MDI_PI_LOCK(pip);
5709 
5710 	/* If PM_GET_PM_INFO is NULL the pm_hold_power above was a noop */
5711 	if (DEVI(ph_dip)->devi_pm_info)
5712 		MDI_PI(pip)->pi_pm_held = 1;
5713 }
5714 
5715 /*
5716  * Allow phci powered down
5717  */
5718 static void
5719 i_mdi_pm_rele_pip(mdi_pathinfo_t *pip)
5720 {
5721 	dev_info_t	*ph_dip = NULL;
5722 
5723 	ASSERT(pip != NULL);
5724 	ASSERT(MDI_PI_LOCKED(pip));
5725 
5726 	if (MDI_PI(pip)->pi_pm_held == 0) {
5727 		return;
5728 	}
5729 
5730 	ph_dip = mdi_pi_get_phci(pip);
5731 	ASSERT(ph_dip != NULL);
5732 
5733 	MDI_PI_UNLOCK(pip);
5734 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "i_mdi_pm_rele_pip for %s%d %p\n",
5735 	    ddi_get_name(ph_dip), ddi_get_instance(ph_dip), (void *)pip));
5736 
5737 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "kidsupcnt was %d\n",
5738 	    DEVI(ph_dip)->devi_pm_kidsupcnt));
5739 	pm_rele_power(ph_dip);
5740 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "kidsupcnt is %d\n",
5741 	    DEVI(ph_dip)->devi_pm_kidsupcnt));
5742 
5743 	MDI_PI_LOCK(pip);
5744 	MDI_PI(pip)->pi_pm_held = 0;
5745 }
5746 
5747 static void
5748 i_mdi_pm_hold_client(mdi_client_t *ct, int incr)
5749 {
5750 	ASSERT(MDI_CLIENT_LOCKED(ct));
5751 
5752 	ct->ct_power_cnt += incr;
5753 	MDI_DEBUG(4, (CE_NOTE, ct->ct_dip, "i_mdi_pm_hold_client %p "
5754 	    "ct_power_cnt = %d incr = %d\n", (void *)ct,
5755 	    ct->ct_power_cnt, incr));
5756 	ASSERT(ct->ct_power_cnt >= 0);
5757 }
5758 
5759 static void
5760 i_mdi_rele_all_phci(mdi_client_t *ct)
5761 {
5762 	mdi_pathinfo_t  *pip;
5763 
5764 	ASSERT(MDI_CLIENT_LOCKED(ct));
5765 	pip = (mdi_pathinfo_t *)ct->ct_path_head;
5766 	while (pip != NULL) {
5767 		mdi_hold_path(pip);
5768 		MDI_PI_LOCK(pip);
5769 		i_mdi_pm_rele_pip(pip);
5770 		MDI_PI_UNLOCK(pip);
5771 		mdi_rele_path(pip);
5772 		pip = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
5773 	}
5774 }
5775 
5776 static void
5777 i_mdi_pm_rele_client(mdi_client_t *ct, int decr)
5778 {
5779 	ASSERT(MDI_CLIENT_LOCKED(ct));
5780 
5781 	if (i_ddi_devi_attached(ct->ct_dip)) {
5782 		ct->ct_power_cnt -= decr;
5783 		MDI_DEBUG(4, (CE_NOTE, ct->ct_dip, "i_mdi_pm_rele_client %p "
5784 		    "ct_power_cnt = %d decr = %d\n",
5785 		    (void *)ct, ct->ct_power_cnt, decr));
5786 	}
5787 
5788 	ASSERT(ct->ct_power_cnt >= 0);
5789 	if (ct->ct_power_cnt == 0) {
5790 		i_mdi_rele_all_phci(ct);
5791 		return;
5792 	}
5793 }
5794 
5795 static void
5796 i_mdi_pm_reset_client(mdi_client_t *ct)
5797 {
5798 	MDI_DEBUG(4, (CE_NOTE, ct->ct_dip, "i_mdi_pm_reset_client %p "
5799 	    "ct_power_cnt = %d\n", (void *)ct, ct->ct_power_cnt));
5800 	ASSERT(MDI_CLIENT_LOCKED(ct));
5801 	ct->ct_power_cnt = 0;
5802 	i_mdi_rele_all_phci(ct);
5803 	ct->ct_powercnt_config = 0;
5804 	ct->ct_powercnt_unconfig = 0;
5805 	ct->ct_powercnt_reset = 1;
5806 }
5807 
5808 static int
5809 i_mdi_power_one_phci(mdi_pathinfo_t *pip)
5810 {
5811 	int		ret;
5812 	dev_info_t	*ph_dip;
5813 
5814 	MDI_PI_LOCK(pip);
5815 	i_mdi_pm_hold_pip(pip);
5816 
5817 	ph_dip = mdi_pi_get_phci(pip);
5818 	MDI_PI_UNLOCK(pip);
5819 
5820 	/* bring all components of phci to full power */
5821 	MDI_DEBUG(4, (CE_NOTE, ph_dip, "i_mdi_power_one_phci "
5822 	    "pm_powerup for %s%d %p\n", ddi_get_name(ph_dip),
5823 	    ddi_get_instance(ph_dip), (void *)pip));
5824 
5825 	ret = pm_powerup(ph_dip);
5826 
5827 	if (ret == DDI_FAILURE) {
5828 		MDI_DEBUG(4, (CE_NOTE, ph_dip, "i_mdi_power_one_phci "
5829 		    "pm_powerup FAILED for %s%d %p\n",
5830 		    ddi_get_name(ph_dip), ddi_get_instance(ph_dip),
5831 		    (void *)pip));
5832 
5833 		MDI_PI_LOCK(pip);
5834 		i_mdi_pm_rele_pip(pip);
5835 		MDI_PI_UNLOCK(pip);
5836 		return (MDI_FAILURE);
5837 	}
5838 
5839 	return (MDI_SUCCESS);
5840 }
5841 
5842 static int
5843 i_mdi_power_all_phci(mdi_client_t *ct)
5844 {
5845 	mdi_pathinfo_t  *pip;
5846 	int		succeeded = 0;
5847 
5848 	ASSERT(MDI_CLIENT_LOCKED(ct));
5849 	pip = (mdi_pathinfo_t *)ct->ct_path_head;
5850 	while (pip != NULL) {
5851 		/*
5852 		 * Don't power if MDI_PATHINFO_STATE_FAULT
5853 		 * or MDI_PATHINFO_STATE_OFFLINE.
5854 		 */
5855 		if (MDI_PI_IS_INIT(pip) ||
5856 		    MDI_PI_IS_ONLINE(pip) || MDI_PI_IS_STANDBY(pip)) {
5857 			mdi_hold_path(pip);
5858 			MDI_CLIENT_UNLOCK(ct);
5859 			if (i_mdi_power_one_phci(pip) == MDI_SUCCESS)
5860 				succeeded = 1;
5861 
5862 			ASSERT(ct == MDI_PI(pip)->pi_client);
5863 			MDI_CLIENT_LOCK(ct);
5864 			mdi_rele_path(pip);
5865 		}
5866 		pip = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
5867 	}
5868 
5869 	return (succeeded ? MDI_SUCCESS : MDI_FAILURE);
5870 }
5871 
5872 /*
5873  * mdi_bus_power():
5874  *		1. Place the phci(s) into powered up state so that
5875  *		   client can do power management
5876  *		2. Ensure phci powered up as client power managing
5877  * Return Values:
5878  *		MDI_SUCCESS
5879  *		MDI_FAILURE
5880  */
5881 int
5882 mdi_bus_power(dev_info_t *parent, void *impl_arg, pm_bus_power_op_t op,
5883     void *arg, void *result)
5884 {
5885 	int			ret = MDI_SUCCESS;
5886 	pm_bp_child_pwrchg_t	*bpc;
5887 	mdi_client_t		*ct;
5888 	dev_info_t		*cdip;
5889 	pm_bp_has_changed_t	*bphc;
5890 
5891 	/*
5892 	 * BUS_POWER_NOINVOL not supported
5893 	 */
5894 	if (op == BUS_POWER_NOINVOL)
5895 		return (MDI_FAILURE);
5896 
5897 	/*
5898 	 * ignore other OPs.
5899 	 * return quickly to save cou cycles on the ct processing
5900 	 */
5901 	switch (op) {
5902 	case BUS_POWER_PRE_NOTIFICATION:
5903 	case BUS_POWER_POST_NOTIFICATION:
5904 		bpc = (pm_bp_child_pwrchg_t *)arg;
5905 		cdip = bpc->bpc_dip;
5906 		break;
5907 	case BUS_POWER_HAS_CHANGED:
5908 		bphc = (pm_bp_has_changed_t *)arg;
5909 		cdip = bphc->bphc_dip;
5910 		break;
5911 	default:
5912 		return (pm_busop_bus_power(parent, impl_arg, op, arg, result));
5913 	}
5914 
5915 	ASSERT(MDI_CLIENT(cdip));
5916 
5917 	ct = i_devi_get_client(cdip);
5918 	if (ct == NULL)
5919 		return (MDI_FAILURE);
5920 
5921 	/*
5922 	 * wait till the mdi_pathinfo node state change are processed
5923 	 */
5924 	MDI_CLIENT_LOCK(ct);
5925 	switch (op) {
5926 	case BUS_POWER_PRE_NOTIFICATION:
5927 		MDI_DEBUG(4, (CE_NOTE, bpc->bpc_dip, "mdi_bus_power "
5928 		    "BUS_POWER_PRE_NOTIFICATION:"
5929 		    "%s@%s, olevel=%d, nlevel=%d, comp=%d\n",
5930 		    PM_NAME(bpc->bpc_dip), PM_ADDR(bpc->bpc_dip),
5931 		    bpc->bpc_olevel, bpc->bpc_nlevel, bpc->bpc_comp));
5932 
5933 		/* serialize power level change per client */
5934 		while (MDI_CLIENT_IS_POWER_TRANSITION(ct))
5935 			cv_wait(&ct->ct_powerchange_cv, &ct->ct_mutex);
5936 
5937 		MDI_CLIENT_SET_POWER_TRANSITION(ct);
5938 
5939 		if (ct->ct_power_cnt == 0) {
5940 			ret = i_mdi_power_all_phci(ct);
5941 		}
5942 
5943 		/*
5944 		 * if new_level > 0:
5945 		 *	- hold phci(s)
5946 		 *	- power up phci(s) if not already
5947 		 * ignore power down
5948 		 */
5949 		if (bpc->bpc_nlevel > 0) {
5950 			if (!DEVI_IS_ATTACHING(ct->ct_dip)) {
5951 				MDI_DEBUG(4, (CE_NOTE, bpc->bpc_dip,
5952 				    "mdi_bus_power i_mdi_pm_hold_client\n"));
5953 				i_mdi_pm_hold_client(ct, ct->ct_path_count);
5954 			}
5955 		}
5956 		break;
5957 	case BUS_POWER_POST_NOTIFICATION:
5958 		MDI_DEBUG(4, (CE_NOTE, bpc->bpc_dip, "mdi_bus_power "
5959 		    "BUS_POWER_POST_NOTIFICATION:"
5960 		    "%s@%s, olevel=%d, nlevel=%d, comp=%d result=%d\n",
5961 		    PM_NAME(bpc->bpc_dip), PM_ADDR(bpc->bpc_dip),
5962 		    bpc->bpc_olevel, bpc->bpc_nlevel, bpc->bpc_comp,
5963 		    *(int *)result));
5964 
5965 		if (*(int *)result == DDI_SUCCESS) {
5966 			if (bpc->bpc_nlevel > 0) {
5967 				MDI_CLIENT_SET_POWER_UP(ct);
5968 			} else {
5969 				MDI_CLIENT_SET_POWER_DOWN(ct);
5970 			}
5971 		}
5972 
5973 		/* release the hold we did in pre-notification */
5974 		if (bpc->bpc_nlevel > 0 && (*(int *)result != DDI_SUCCESS) &&
5975 		    !DEVI_IS_ATTACHING(ct->ct_dip)) {
5976 			MDI_DEBUG(4, (CE_NOTE, bpc->bpc_dip,
5977 			    "mdi_bus_power i_mdi_pm_rele_client\n"));
5978 			i_mdi_pm_rele_client(ct, ct->ct_path_count);
5979 		}
5980 
5981 		if (bpc->bpc_nlevel == 0 && (*(int *)result == DDI_SUCCESS)) {
5982 			/* another thread might started attaching */
5983 			if (DEVI_IS_ATTACHING(ct->ct_dip)) {
5984 				MDI_DEBUG(4, (CE_NOTE, bpc->bpc_dip,
5985 				    "mdi_bus_power i_mdi_pm_rele_client\n"));
5986 				i_mdi_pm_rele_client(ct, ct->ct_path_count);
5987 			/* detaching has been taken care in pm_post_unconfig */
5988 			} else if (!DEVI_IS_DETACHING(ct->ct_dip)) {
5989 				MDI_DEBUG(4, (CE_NOTE, bpc->bpc_dip,
5990 				    "mdi_bus_power i_mdi_pm_reset_client\n"));
5991 				i_mdi_pm_reset_client(ct);
5992 			}
5993 		}
5994 
5995 		MDI_CLIENT_CLEAR_POWER_TRANSITION(ct);
5996 		cv_broadcast(&ct->ct_powerchange_cv);
5997 
5998 		break;
5999 
6000 	/* need to do more */
6001 	case BUS_POWER_HAS_CHANGED:
6002 		MDI_DEBUG(4, (CE_NOTE, bphc->bphc_dip, "mdi_bus_power "
6003 		    "BUS_POWER_HAS_CHANGED:"
6004 		    "%s@%s, olevel=%d, nlevel=%d, comp=%d\n",
6005 		    PM_NAME(bphc->bphc_dip), PM_ADDR(bphc->bphc_dip),
6006 		    bphc->bphc_olevel, bphc->bphc_nlevel, bphc->bphc_comp));
6007 
6008 		if (bphc->bphc_nlevel > 0 &&
6009 		    bphc->bphc_nlevel > bphc->bphc_olevel) {
6010 			if (ct->ct_power_cnt == 0) {
6011 				ret = i_mdi_power_all_phci(ct);
6012 			}
6013 			MDI_DEBUG(4, (CE_NOTE, bphc->bphc_dip,
6014 			    "mdi_bus_power i_mdi_pm_hold_client\n"));
6015 			i_mdi_pm_hold_client(ct, ct->ct_path_count);
6016 		}
6017 
6018 		if (bphc->bphc_nlevel == 0 && bphc->bphc_olevel != -1) {
6019 			MDI_DEBUG(4, (CE_NOTE, bphc->bphc_dip,
6020 			    "mdi_bus_power i_mdi_pm_rele_client\n"));
6021 			i_mdi_pm_rele_client(ct, ct->ct_path_count);
6022 		}
6023 		break;
6024 	}
6025 
6026 	MDI_CLIENT_UNLOCK(ct);
6027 	return (ret);
6028 }
6029 
6030 static int
6031 i_mdi_pm_pre_config_one(dev_info_t *child)
6032 {
6033 	int		ret = MDI_SUCCESS;
6034 	mdi_client_t	*ct;
6035 
6036 	ct = i_devi_get_client(child);
6037 	if (ct == NULL)
6038 		return (MDI_FAILURE);
6039 
6040 	MDI_CLIENT_LOCK(ct);
6041 	while (MDI_CLIENT_IS_POWER_TRANSITION(ct))
6042 		cv_wait(&ct->ct_powerchange_cv, &ct->ct_mutex);
6043 
6044 	if (!MDI_CLIENT_IS_FAILED(ct)) {
6045 		MDI_CLIENT_UNLOCK(ct);
6046 		MDI_DEBUG(4, (CE_NOTE, child,
6047 		    "i_mdi_pm_pre_config_one already configured\n"));
6048 		return (MDI_SUCCESS);
6049 	}
6050 
6051 	if (ct->ct_powercnt_config) {
6052 		MDI_CLIENT_UNLOCK(ct);
6053 		MDI_DEBUG(4, (CE_NOTE, child,
6054 		    "i_mdi_pm_pre_config_one ALREADY held\n"));
6055 		return (MDI_SUCCESS);
6056 	}
6057 
6058 	if (ct->ct_power_cnt == 0) {
6059 		ret = i_mdi_power_all_phci(ct);
6060 	}
6061 	MDI_DEBUG(4, (CE_NOTE, child,
6062 	    "i_mdi_pm_pre_config_one i_mdi_pm_hold_client\n"));
6063 	i_mdi_pm_hold_client(ct, ct->ct_path_count);
6064 	ct->ct_powercnt_config = 1;
6065 	ct->ct_powercnt_reset = 0;
6066 	MDI_CLIENT_UNLOCK(ct);
6067 	return (ret);
6068 }
6069 
6070 static int
6071 i_mdi_pm_pre_config(dev_info_t *vdip, dev_info_t *child)
6072 {
6073 	int			ret = MDI_SUCCESS;
6074 	dev_info_t		*cdip;
6075 	int			circ;
6076 
6077 	ASSERT(MDI_VHCI(vdip));
6078 
6079 	/* ndi_devi_config_one */
6080 	if (child) {
6081 		ASSERT(DEVI_BUSY_OWNED(vdip));
6082 		return (i_mdi_pm_pre_config_one(child));
6083 	}
6084 
6085 	/* devi_config_common */
6086 	ndi_devi_enter(vdip, &circ);
6087 	cdip = ddi_get_child(vdip);
6088 	while (cdip) {
6089 		dev_info_t *next = ddi_get_next_sibling(cdip);
6090 
6091 		ret = i_mdi_pm_pre_config_one(cdip);
6092 		if (ret != MDI_SUCCESS)
6093 			break;
6094 		cdip = next;
6095 	}
6096 	ndi_devi_exit(vdip, circ);
6097 	return (ret);
6098 }
6099 
6100 static int
6101 i_mdi_pm_pre_unconfig_one(dev_info_t *child, int *held, int flags)
6102 {
6103 	int		ret = MDI_SUCCESS;
6104 	mdi_client_t	*ct;
6105 
6106 	ct = i_devi_get_client(child);
6107 	if (ct == NULL)
6108 		return (MDI_FAILURE);
6109 
6110 	MDI_CLIENT_LOCK(ct);
6111 	while (MDI_CLIENT_IS_POWER_TRANSITION(ct))
6112 		cv_wait(&ct->ct_powerchange_cv, &ct->ct_mutex);
6113 
6114 	if (!i_ddi_devi_attached(ct->ct_dip)) {
6115 		MDI_DEBUG(4, (CE_NOTE, child,
6116 		    "i_mdi_pm_pre_unconfig node detached already\n"));
6117 		MDI_CLIENT_UNLOCK(ct);
6118 		return (MDI_SUCCESS);
6119 	}
6120 
6121 	if (MDI_CLIENT_IS_POWERED_DOWN(ct) &&
6122 	    (flags & NDI_AUTODETACH)) {
6123 		MDI_DEBUG(4, (CE_NOTE, child,
6124 		    "i_mdi_pm_pre_unconfig auto-modunload\n"));
6125 		MDI_CLIENT_UNLOCK(ct);
6126 		return (MDI_FAILURE);
6127 	}
6128 
6129 	if (ct->ct_powercnt_unconfig) {
6130 		MDI_DEBUG(4, (CE_NOTE, child,
6131 		    "i_mdi_pm_pre_unconfig ct_powercnt_held\n"));
6132 		MDI_CLIENT_UNLOCK(ct);
6133 		*held = 1;
6134 		return (MDI_SUCCESS);
6135 	}
6136 
6137 	if (ct->ct_power_cnt == 0) {
6138 		ret = i_mdi_power_all_phci(ct);
6139 	}
6140 	MDI_DEBUG(4, (CE_NOTE, child,
6141 	    "i_mdi_pm_pre_unconfig i_mdi_pm_hold_client\n"));
6142 	i_mdi_pm_hold_client(ct, ct->ct_path_count);
6143 	ct->ct_powercnt_unconfig = 1;
6144 	ct->ct_powercnt_reset = 0;
6145 	MDI_CLIENT_UNLOCK(ct);
6146 	if (ret == MDI_SUCCESS)
6147 		*held = 1;
6148 	return (ret);
6149 }
6150 
6151 static int
6152 i_mdi_pm_pre_unconfig(dev_info_t *vdip, dev_info_t *child, int *held,
6153     int flags)
6154 {
6155 	int			ret = MDI_SUCCESS;
6156 	dev_info_t		*cdip;
6157 	int			circ;
6158 
6159 	ASSERT(MDI_VHCI(vdip));
6160 	*held = 0;
6161 
6162 	/* ndi_devi_unconfig_one */
6163 	if (child) {
6164 		ASSERT(DEVI_BUSY_OWNED(vdip));
6165 		return (i_mdi_pm_pre_unconfig_one(child, held, flags));
6166 	}
6167 
6168 	/* devi_unconfig_common */
6169 	ndi_devi_enter(vdip, &circ);
6170 	cdip = ddi_get_child(vdip);
6171 	while (cdip) {
6172 		dev_info_t *next = ddi_get_next_sibling(cdip);
6173 
6174 		ret = i_mdi_pm_pre_unconfig_one(cdip, held, flags);
6175 		cdip = next;
6176 	}
6177 	ndi_devi_exit(vdip, circ);
6178 
6179 	if (*held)
6180 		ret = MDI_SUCCESS;
6181 
6182 	return (ret);
6183 }
6184 
6185 static void
6186 i_mdi_pm_post_config_one(dev_info_t *child)
6187 {
6188 	mdi_client_t	*ct;
6189 
6190 	ct = i_devi_get_client(child);
6191 	if (ct == NULL)
6192 		return;
6193 
6194 	MDI_CLIENT_LOCK(ct);
6195 	while (MDI_CLIENT_IS_POWER_TRANSITION(ct))
6196 		cv_wait(&ct->ct_powerchange_cv, &ct->ct_mutex);
6197 
6198 	if (ct->ct_powercnt_reset || !ct->ct_powercnt_config) {
6199 		MDI_DEBUG(4, (CE_NOTE, child,
6200 		    "i_mdi_pm_post_config_one NOT configured\n"));
6201 		MDI_CLIENT_UNLOCK(ct);
6202 		return;
6203 	}
6204 
6205 	/* client has not been updated */
6206 	if (MDI_CLIENT_IS_FAILED(ct)) {
6207 		MDI_DEBUG(4, (CE_NOTE, child,
6208 		    "i_mdi_pm_post_config_one NOT configured\n"));
6209 		MDI_CLIENT_UNLOCK(ct);
6210 		return;
6211 	}
6212 
6213 	/* another thread might have powered it down or detached it */
6214 	if ((MDI_CLIENT_IS_POWERED_DOWN(ct) &&
6215 	    !DEVI_IS_ATTACHING(ct->ct_dip)) ||
6216 	    (!i_ddi_devi_attached(ct->ct_dip) &&
6217 	    !DEVI_IS_ATTACHING(ct->ct_dip))) {
6218 		MDI_DEBUG(4, (CE_NOTE, child,
6219 		    "i_mdi_pm_post_config i_mdi_pm_reset_client\n"));
6220 		i_mdi_pm_reset_client(ct);
6221 	} else {
6222 		mdi_pathinfo_t  *pip, *next;
6223 		int	valid_path_count = 0;
6224 
6225 		MDI_DEBUG(4, (CE_NOTE, child,
6226 		    "i_mdi_pm_post_config i_mdi_pm_rele_client\n"));
6227 		pip = ct->ct_path_head;
6228 		while (pip != NULL) {
6229 			MDI_PI_LOCK(pip);
6230 			next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
6231 			if (MDI_PI_IS_ONLINE(pip) || MDI_PI_IS_STANDBY(pip))
6232 				valid_path_count ++;
6233 			MDI_PI_UNLOCK(pip);
6234 			pip = next;
6235 		}
6236 		i_mdi_pm_rele_client(ct, valid_path_count);
6237 	}
6238 	ct->ct_powercnt_config = 0;
6239 	MDI_CLIENT_UNLOCK(ct);
6240 }
6241 
6242 static void
6243 i_mdi_pm_post_config(dev_info_t *vdip, dev_info_t *child)
6244 {
6245 	int		circ;
6246 	dev_info_t	*cdip;
6247 
6248 	ASSERT(MDI_VHCI(vdip));
6249 
6250 	/* ndi_devi_config_one */
6251 	if (child) {
6252 		ASSERT(DEVI_BUSY_OWNED(vdip));
6253 		i_mdi_pm_post_config_one(child);
6254 		return;
6255 	}
6256 
6257 	/* devi_config_common */
6258 	ndi_devi_enter(vdip, &circ);
6259 	cdip = ddi_get_child(vdip);
6260 	while (cdip) {
6261 		dev_info_t *next = ddi_get_next_sibling(cdip);
6262 
6263 		i_mdi_pm_post_config_one(cdip);
6264 		cdip = next;
6265 	}
6266 	ndi_devi_exit(vdip, circ);
6267 }
6268 
6269 static void
6270 i_mdi_pm_post_unconfig_one(dev_info_t *child)
6271 {
6272 	mdi_client_t	*ct;
6273 
6274 	ct = i_devi_get_client(child);
6275 	if (ct == NULL)
6276 		return;
6277 
6278 	MDI_CLIENT_LOCK(ct);
6279 	while (MDI_CLIENT_IS_POWER_TRANSITION(ct))
6280 		cv_wait(&ct->ct_powerchange_cv, &ct->ct_mutex);
6281 
6282 	if (!ct->ct_powercnt_unconfig || ct->ct_powercnt_reset) {
6283 		MDI_DEBUG(4, (CE_NOTE, child,
6284 		    "i_mdi_pm_post_unconfig NOT held\n"));
6285 		MDI_CLIENT_UNLOCK(ct);
6286 		return;
6287 	}
6288 
6289 	/* failure detaching or another thread just attached it */
6290 	if ((MDI_CLIENT_IS_POWERED_DOWN(ct) &&
6291 	    i_ddi_devi_attached(ct->ct_dip)) ||
6292 	    (!i_ddi_devi_attached(ct->ct_dip) &&
6293 	    !DEVI_IS_ATTACHING(ct->ct_dip))) {
6294 		MDI_DEBUG(4, (CE_NOTE, child,
6295 		    "i_mdi_pm_post_unconfig i_mdi_pm_reset_client\n"));
6296 		i_mdi_pm_reset_client(ct);
6297 	} else {
6298 		mdi_pathinfo_t  *pip, *next;
6299 		int	valid_path_count = 0;
6300 
6301 		MDI_DEBUG(4, (CE_NOTE, child,
6302 		    "i_mdi_pm_post_unconfig i_mdi_pm_rele_client\n"));
6303 		pip = ct->ct_path_head;
6304 		while (pip != NULL) {
6305 			MDI_PI_LOCK(pip);
6306 			next = (mdi_pathinfo_t *)MDI_PI(pip)->pi_client_link;
6307 			if (MDI_PI_IS_ONLINE(pip) || MDI_PI_IS_STANDBY(pip))
6308 				valid_path_count ++;
6309 			MDI_PI_UNLOCK(pip);
6310 			pip = next;
6311 		}
6312 		i_mdi_pm_rele_client(ct, valid_path_count);
6313 		ct->ct_powercnt_unconfig = 0;
6314 	}
6315 
6316 	MDI_CLIENT_UNLOCK(ct);
6317 }
6318 
6319 static void
6320 i_mdi_pm_post_unconfig(dev_info_t *vdip, dev_info_t *child, int held)
6321 {
6322 	int			circ;
6323 	dev_info_t		*cdip;
6324 
6325 	ASSERT(MDI_VHCI(vdip));
6326 
6327 	if (!held) {
6328 		MDI_DEBUG(4, (CE_NOTE, vdip,
6329 		    "i_mdi_pm_post_unconfig held = %d\n", held));
6330 		return;
6331 	}
6332 
6333 	if (child) {
6334 		ASSERT(DEVI_BUSY_OWNED(vdip));
6335 		i_mdi_pm_post_unconfig_one(child);
6336 		return;
6337 	}
6338 
6339 	ndi_devi_enter(vdip, &circ);
6340 	cdip = ddi_get_child(vdip);
6341 	while (cdip) {
6342 		dev_info_t *next = ddi_get_next_sibling(cdip);
6343 
6344 		i_mdi_pm_post_unconfig_one(cdip);
6345 		cdip = next;
6346 	}
6347 	ndi_devi_exit(vdip, circ);
6348 }
6349 
6350 int
6351 mdi_power(dev_info_t *vdip, mdi_pm_op_t op, void *args, char *devnm, int flags)
6352 {
6353 	int			circ, ret = MDI_SUCCESS;
6354 	dev_info_t		*client_dip = NULL;
6355 	mdi_client_t		*ct;
6356 
6357 	/*
6358 	 * Handling ndi_devi_config_one and ndi_devi_unconfig_one.
6359 	 * Power up pHCI for the named client device.
6360 	 * Note: Before the client is enumerated under vhci by phci,
6361 	 * client_dip can be NULL. Then proceed to power up all the
6362 	 * pHCIs.
6363 	 */
6364 	if (devnm != NULL) {
6365 		ndi_devi_enter(vdip, &circ);
6366 		client_dip = ndi_devi_findchild(vdip, devnm);
6367 	}
6368 
6369 	MDI_DEBUG(4, (CE_NOTE, vdip, "mdi_power op = %d %s %p\n",
6370 	    op, devnm ? devnm : "NULL", (void *)client_dip));
6371 
6372 	switch (op) {
6373 	case MDI_PM_PRE_CONFIG:
6374 		ret = i_mdi_pm_pre_config(vdip, client_dip);
6375 		break;
6376 
6377 	case MDI_PM_PRE_UNCONFIG:
6378 		ret = i_mdi_pm_pre_unconfig(vdip, client_dip, (int *)args,
6379 		    flags);
6380 		break;
6381 
6382 	case MDI_PM_POST_CONFIG:
6383 		i_mdi_pm_post_config(vdip, client_dip);
6384 		break;
6385 
6386 	case MDI_PM_POST_UNCONFIG:
6387 		i_mdi_pm_post_unconfig(vdip, client_dip, *(int *)args);
6388 		break;
6389 
6390 	case MDI_PM_HOLD_POWER:
6391 	case MDI_PM_RELE_POWER:
6392 		ASSERT(args);
6393 
6394 		client_dip = (dev_info_t *)args;
6395 		ASSERT(MDI_CLIENT(client_dip));
6396 
6397 		ct = i_devi_get_client(client_dip);
6398 		MDI_CLIENT_LOCK(ct);
6399 
6400 		if (op == MDI_PM_HOLD_POWER) {
6401 			if (ct->ct_power_cnt == 0) {
6402 				(void) i_mdi_power_all_phci(ct);
6403 				MDI_DEBUG(4, (CE_NOTE, client_dip,
6404 				    "mdi_power i_mdi_pm_hold_client\n"));
6405 				i_mdi_pm_hold_client(ct, ct->ct_path_count);
6406 			}
6407 		} else {
6408 			if (DEVI_IS_ATTACHING(ct->ct_dip)) {
6409 				MDI_DEBUG(4, (CE_NOTE, client_dip,
6410 				    "mdi_power i_mdi_pm_rele_client\n"));
6411 				i_mdi_pm_rele_client(ct, ct->ct_path_count);
6412 			} else {
6413 				MDI_DEBUG(4, (CE_NOTE, client_dip,
6414 				    "mdi_power i_mdi_pm_reset_client\n"));
6415 				i_mdi_pm_reset_client(ct);
6416 			}
6417 		}
6418 
6419 		MDI_CLIENT_UNLOCK(ct);
6420 		break;
6421 
6422 	default:
6423 		break;
6424 	}
6425 
6426 	if (devnm)
6427 		ndi_devi_exit(vdip, circ);
6428 
6429 	return (ret);
6430 }
6431 
6432 int
6433 mdi_component_is_vhci(dev_info_t *dip, const char **mdi_class)
6434 {
6435 	mdi_vhci_t *vhci;
6436 
6437 	if (!MDI_VHCI(dip))
6438 		return (MDI_FAILURE);
6439 
6440 	if (mdi_class) {
6441 		vhci = DEVI(dip)->devi_mdi_xhci;
6442 		ASSERT(vhci);
6443 		*mdi_class = vhci->vh_class;
6444 	}
6445 
6446 	return (MDI_SUCCESS);
6447 }
6448 
6449 int
6450 mdi_component_is_phci(dev_info_t *dip, const char **mdi_class)
6451 {
6452 	mdi_phci_t *phci;
6453 
6454 	if (!MDI_PHCI(dip))
6455 		return (MDI_FAILURE);
6456 
6457 	if (mdi_class) {
6458 		phci = DEVI(dip)->devi_mdi_xhci;
6459 		ASSERT(phci);
6460 		*mdi_class = phci->ph_vhci->vh_class;
6461 	}
6462 
6463 	return (MDI_SUCCESS);
6464 }
6465 
6466 int
6467 mdi_component_is_client(dev_info_t *dip, const char **mdi_class)
6468 {
6469 	mdi_client_t *client;
6470 
6471 	if (!MDI_CLIENT(dip))
6472 		return (MDI_FAILURE);
6473 
6474 	if (mdi_class) {
6475 		client = DEVI(dip)->devi_mdi_client;
6476 		ASSERT(client);
6477 		*mdi_class = client->ct_vhci->vh_class;
6478 	}
6479 
6480 	return (MDI_SUCCESS);
6481 }
6482 
6483 void *
6484 mdi_client_get_vhci_private(dev_info_t *dip)
6485 {
6486 	ASSERT(mdi_component_is_client(dip, NULL) == MDI_SUCCESS);
6487 	if (mdi_component_is_client(dip, NULL) == MDI_SUCCESS) {
6488 		mdi_client_t	*ct;
6489 		ct = i_devi_get_client(dip);
6490 		return (ct->ct_vprivate);
6491 	}
6492 	return (NULL);
6493 }
6494 
6495 void
6496 mdi_client_set_vhci_private(dev_info_t *dip, void *data)
6497 {
6498 	ASSERT(mdi_component_is_client(dip, NULL) == MDI_SUCCESS);
6499 	if (mdi_component_is_client(dip, NULL) == MDI_SUCCESS) {
6500 		mdi_client_t	*ct;
6501 		ct = i_devi_get_client(dip);
6502 		ct->ct_vprivate = data;
6503 	}
6504 }
6505 /*
6506  * mdi_pi_get_vhci_private():
6507  *		Get the vhci private information associated with the
6508  *		mdi_pathinfo node
6509  */
6510 void *
6511 mdi_pi_get_vhci_private(mdi_pathinfo_t *pip)
6512 {
6513 	caddr_t	vprivate = NULL;
6514 	if (pip) {
6515 		vprivate = MDI_PI(pip)->pi_vprivate;
6516 	}
6517 	return (vprivate);
6518 }
6519 
6520 /*
6521  * mdi_pi_set_vhci_private():
6522  *		Set the vhci private information in the mdi_pathinfo node
6523  */
6524 void
6525 mdi_pi_set_vhci_private(mdi_pathinfo_t *pip, void *priv)
6526 {
6527 	if (pip) {
6528 		MDI_PI(pip)->pi_vprivate = priv;
6529 	}
6530 }
6531 
6532 /*
6533  * mdi_phci_get_vhci_private():
6534  *		Get the vhci private information associated with the
6535  *		mdi_phci node
6536  */
6537 void *
6538 mdi_phci_get_vhci_private(dev_info_t *dip)
6539 {
6540 	ASSERT(mdi_component_is_phci(dip, NULL) == MDI_SUCCESS);
6541 	if (mdi_component_is_phci(dip, NULL) == MDI_SUCCESS) {
6542 		mdi_phci_t	*ph;
6543 		ph = i_devi_get_phci(dip);
6544 		return (ph->ph_vprivate);
6545 	}
6546 	return (NULL);
6547 }
6548 
6549 /*
6550  * mdi_phci_set_vhci_private():
6551  *		Set the vhci private information in the mdi_phci node
6552  */
6553 void
6554 mdi_phci_set_vhci_private(dev_info_t *dip, void *priv)
6555 {
6556 	ASSERT(mdi_component_is_phci(dip, NULL) == MDI_SUCCESS);
6557 	if (mdi_component_is_phci(dip, NULL) == MDI_SUCCESS) {
6558 		mdi_phci_t	*ph;
6559 		ph = i_devi_get_phci(dip);
6560 		ph->ph_vprivate = priv;
6561 	}
6562 }
6563 
6564 /*
6565  * List of vhci class names:
6566  * A vhci class name must be in this list only if the corresponding vhci
6567  * driver intends to use the mdi provided bus config implementation
6568  * (i.e., mdi_vhci_bus_config()).
6569  */
6570 static char *vhci_class_list[] = { MDI_HCI_CLASS_SCSI, MDI_HCI_CLASS_IB };
6571 #define	N_VHCI_CLASSES	(sizeof (vhci_class_list) / sizeof (char *))
6572 
6573 /*
6574  * During boot time, the on-disk vhci cache for every vhci class is read
6575  * in the form of an nvlist and stored here.
6576  */
6577 static nvlist_t *vhcache_nvl[N_VHCI_CLASSES];
6578 
6579 /* nvpair names in vhci cache nvlist */
6580 #define	MDI_VHCI_CACHE_VERSION	1
6581 #define	MDI_NVPNAME_VERSION	"version"
6582 #define	MDI_NVPNAME_PHCIS	"phcis"
6583 #define	MDI_NVPNAME_CTADDRMAP	"clientaddrmap"
6584 
6585 /*
6586  * Given vhci class name, return its on-disk vhci cache filename.
6587  * Memory for the returned filename which includes the full path is allocated
6588  * by this function.
6589  */
6590 static char *
6591 vhclass2vhcache_filename(char *vhclass)
6592 {
6593 	char *filename;
6594 	int len;
6595 	static char *fmt = "/etc/devices/mdi_%s_cache";
6596 
6597 	/*
6598 	 * fmt contains the on-disk vhci cache file name format;
6599 	 * for scsi_vhci the filename is "/etc/devices/mdi_scsi_vhci_cache".
6600 	 */
6601 
6602 	/* the -1 below is to account for "%s" in the format string */
6603 	len = strlen(fmt) + strlen(vhclass) - 1;
6604 	filename = kmem_alloc(len, KM_SLEEP);
6605 	(void) snprintf(filename, len, fmt, vhclass);
6606 	ASSERT(len == (strlen(filename) + 1));
6607 	return (filename);
6608 }
6609 
6610 /*
6611  * initialize the vhci cache related data structures and read the on-disk
6612  * vhci cached data into memory.
6613  */
6614 static void
6615 setup_vhci_cache(mdi_vhci_t *vh)
6616 {
6617 	mdi_vhci_config_t *vhc;
6618 	mdi_vhci_cache_t *vhcache;
6619 	int i;
6620 	nvlist_t *nvl = NULL;
6621 
6622 	vhc = kmem_zalloc(sizeof (mdi_vhci_config_t), KM_SLEEP);
6623 	vh->vh_config = vhc;
6624 	vhcache = &vhc->vhc_vhcache;
6625 
6626 	vhc->vhc_vhcache_filename = vhclass2vhcache_filename(vh->vh_class);
6627 
6628 	mutex_init(&vhc->vhc_lock, NULL, MUTEX_DEFAULT, NULL);
6629 	cv_init(&vhc->vhc_cv, NULL, CV_DRIVER, NULL);
6630 
6631 	rw_init(&vhcache->vhcache_lock, NULL, RW_DRIVER, NULL);
6632 
6633 	/*
6634 	 * Create string hash; same as mod_hash_create_strhash() except that
6635 	 * we use NULL key destructor.
6636 	 */
6637 	vhcache->vhcache_client_hash = mod_hash_create_extended(vh->vh_class,
6638 	    mdi_bus_config_cache_hash_size,
6639 	    mod_hash_null_keydtor, mod_hash_null_valdtor,
6640 	    mod_hash_bystr, NULL, mod_hash_strkey_cmp, KM_SLEEP);
6641 
6642 	/*
6643 	 * The on-disk vhci cache is read during booting prior to the
6644 	 * lights-out period by mdi_read_devices_files().
6645 	 */
6646 	for (i = 0; i < N_VHCI_CLASSES; i++) {
6647 		if (strcmp(vhci_class_list[i], vh->vh_class) == 0) {
6648 			nvl = vhcache_nvl[i];
6649 			vhcache_nvl[i] = NULL;
6650 			break;
6651 		}
6652 	}
6653 
6654 	/*
6655 	 * this is to cover the case of some one manually causing unloading
6656 	 * (or detaching) and reloading (or attaching) of a vhci driver.
6657 	 */
6658 	if (nvl == NULL && modrootloaded)
6659 		nvl = read_on_disk_vhci_cache(vh->vh_class);
6660 
6661 	if (nvl != NULL) {
6662 		rw_enter(&vhcache->vhcache_lock, RW_WRITER);
6663 		if (mainnvl_to_vhcache(vhcache, nvl) == MDI_SUCCESS)
6664 			vhcache->vhcache_flags |= MDI_VHCI_CACHE_SETUP_DONE;
6665 		else  {
6666 			cmn_err(CE_WARN,
6667 			    "%s: data file corrupted, will recreate\n",
6668 			    vhc->vhc_vhcache_filename);
6669 		}
6670 		rw_exit(&vhcache->vhcache_lock);
6671 		nvlist_free(nvl);
6672 	}
6673 
6674 	vhc->vhc_cbid = callb_add(stop_vhcache_flush_thread, vhc,
6675 	    CB_CL_UADMIN_PRE_VFS, "mdi_vhcache_flush");
6676 
6677 	vhc->vhc_path_discovery_boot = mdi_path_discovery_boot;
6678 	vhc->vhc_path_discovery_postboot = mdi_path_discovery_postboot;
6679 }
6680 
6681 /*
6682  * free all vhci cache related resources
6683  */
6684 static int
6685 destroy_vhci_cache(mdi_vhci_t *vh)
6686 {
6687 	mdi_vhci_config_t *vhc = vh->vh_config;
6688 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
6689 	mdi_vhcache_phci_t *cphci, *cphci_next;
6690 	mdi_vhcache_client_t *cct, *cct_next;
6691 	mdi_vhcache_pathinfo_t *cpi, *cpi_next;
6692 
6693 	if (stop_vhcache_async_threads(vhc) != MDI_SUCCESS)
6694 		return (MDI_FAILURE);
6695 
6696 	kmem_free(vhc->vhc_vhcache_filename,
6697 	    strlen(vhc->vhc_vhcache_filename) + 1);
6698 
6699 	mod_hash_destroy_strhash(vhcache->vhcache_client_hash);
6700 
6701 	for (cphci = vhcache->vhcache_phci_head; cphci != NULL;
6702 	    cphci = cphci_next) {
6703 		cphci_next = cphci->cphci_next;
6704 		free_vhcache_phci(cphci);
6705 	}
6706 
6707 	for (cct = vhcache->vhcache_client_head; cct != NULL; cct = cct_next) {
6708 		cct_next = cct->cct_next;
6709 		for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi_next) {
6710 			cpi_next = cpi->cpi_next;
6711 			free_vhcache_pathinfo(cpi);
6712 		}
6713 		free_vhcache_client(cct);
6714 	}
6715 
6716 	rw_destroy(&vhcache->vhcache_lock);
6717 
6718 	mutex_destroy(&vhc->vhc_lock);
6719 	cv_destroy(&vhc->vhc_cv);
6720 	kmem_free(vhc, sizeof (mdi_vhci_config_t));
6721 	return (MDI_SUCCESS);
6722 }
6723 
6724 /*
6725  * Stop all vhci cache related async threads and free their resources.
6726  */
6727 static int
6728 stop_vhcache_async_threads(mdi_vhci_config_t *vhc)
6729 {
6730 	mdi_async_client_config_t *acc, *acc_next;
6731 
6732 	mutex_enter(&vhc->vhc_lock);
6733 	vhc->vhc_flags |= MDI_VHC_EXIT;
6734 	ASSERT(vhc->vhc_acc_thrcount >= 0);
6735 	cv_broadcast(&vhc->vhc_cv);
6736 
6737 	while ((vhc->vhc_flags & MDI_VHC_VHCACHE_FLUSH_THREAD) ||
6738 	    vhc->vhc_acc_thrcount != 0) {
6739 		mutex_exit(&vhc->vhc_lock);
6740 		delay(1);
6741 		mutex_enter(&vhc->vhc_lock);
6742 	}
6743 
6744 	vhc->vhc_flags &= ~MDI_VHC_EXIT;
6745 
6746 	for (acc = vhc->vhc_acc_list_head; acc != NULL; acc = acc_next) {
6747 		acc_next = acc->acc_next;
6748 		free_async_client_config(acc);
6749 	}
6750 	vhc->vhc_acc_list_head = NULL;
6751 	vhc->vhc_acc_list_tail = NULL;
6752 	vhc->vhc_acc_count = 0;
6753 
6754 	if (vhc->vhc_flags & MDI_VHC_VHCACHE_DIRTY) {
6755 		vhc->vhc_flags &= ~MDI_VHC_VHCACHE_DIRTY;
6756 		mutex_exit(&vhc->vhc_lock);
6757 		if (flush_vhcache(vhc, 0) != MDI_SUCCESS) {
6758 			vhcache_dirty(vhc);
6759 			return (MDI_FAILURE);
6760 		}
6761 	} else
6762 		mutex_exit(&vhc->vhc_lock);
6763 
6764 	if (callb_delete(vhc->vhc_cbid) != 0)
6765 		return (MDI_FAILURE);
6766 
6767 	return (MDI_SUCCESS);
6768 }
6769 
6770 /*
6771  * Stop vhci cache flush thread
6772  */
6773 /* ARGSUSED */
6774 static boolean_t
6775 stop_vhcache_flush_thread(void *arg, int code)
6776 {
6777 	mdi_vhci_config_t *vhc = (mdi_vhci_config_t *)arg;
6778 
6779 	mutex_enter(&vhc->vhc_lock);
6780 	vhc->vhc_flags |= MDI_VHC_EXIT;
6781 	cv_broadcast(&vhc->vhc_cv);
6782 
6783 	while (vhc->vhc_flags & MDI_VHC_VHCACHE_FLUSH_THREAD) {
6784 		mutex_exit(&vhc->vhc_lock);
6785 		delay(1);
6786 		mutex_enter(&vhc->vhc_lock);
6787 	}
6788 
6789 	if (vhc->vhc_flags & MDI_VHC_VHCACHE_DIRTY) {
6790 		vhc->vhc_flags &= ~MDI_VHC_VHCACHE_DIRTY;
6791 		mutex_exit(&vhc->vhc_lock);
6792 		(void) flush_vhcache(vhc, 1);
6793 	} else
6794 		mutex_exit(&vhc->vhc_lock);
6795 
6796 	return (B_TRUE);
6797 }
6798 
6799 /*
6800  * Enqueue the vhcache phci (cphci) at the tail of the list
6801  */
6802 static void
6803 enqueue_vhcache_phci(mdi_vhci_cache_t *vhcache, mdi_vhcache_phci_t *cphci)
6804 {
6805 	cphci->cphci_next = NULL;
6806 	if (vhcache->vhcache_phci_head == NULL)
6807 		vhcache->vhcache_phci_head = cphci;
6808 	else
6809 		vhcache->vhcache_phci_tail->cphci_next = cphci;
6810 	vhcache->vhcache_phci_tail = cphci;
6811 }
6812 
6813 /*
6814  * Enqueue the vhcache pathinfo (cpi) at the tail of the list
6815  */
6816 static void
6817 enqueue_tail_vhcache_pathinfo(mdi_vhcache_client_t *cct,
6818     mdi_vhcache_pathinfo_t *cpi)
6819 {
6820 	cpi->cpi_next = NULL;
6821 	if (cct->cct_cpi_head == NULL)
6822 		cct->cct_cpi_head = cpi;
6823 	else
6824 		cct->cct_cpi_tail->cpi_next = cpi;
6825 	cct->cct_cpi_tail = cpi;
6826 }
6827 
6828 /*
6829  * Enqueue the vhcache pathinfo (cpi) at the correct location in the
6830  * ordered list. All cpis which do not have MDI_CPI_HINT_PATH_DOES_NOT_EXIST
6831  * flag set come at the beginning of the list. All cpis which have this
6832  * flag set come at the end of the list.
6833  */
6834 static void
6835 enqueue_vhcache_pathinfo(mdi_vhcache_client_t *cct,
6836     mdi_vhcache_pathinfo_t *newcpi)
6837 {
6838 	mdi_vhcache_pathinfo_t *cpi, *prev_cpi;
6839 
6840 	if (cct->cct_cpi_head == NULL ||
6841 	    (newcpi->cpi_flags & MDI_CPI_HINT_PATH_DOES_NOT_EXIST))
6842 		enqueue_tail_vhcache_pathinfo(cct, newcpi);
6843 	else {
6844 		for (cpi = cct->cct_cpi_head, prev_cpi = NULL; cpi != NULL &&
6845 		    !(cpi->cpi_flags & MDI_CPI_HINT_PATH_DOES_NOT_EXIST);
6846 		    prev_cpi = cpi, cpi = cpi->cpi_next)
6847 			;
6848 
6849 		if (prev_cpi == NULL)
6850 			cct->cct_cpi_head = newcpi;
6851 		else
6852 			prev_cpi->cpi_next = newcpi;
6853 
6854 		newcpi->cpi_next = cpi;
6855 
6856 		if (cpi == NULL)
6857 			cct->cct_cpi_tail = newcpi;
6858 	}
6859 }
6860 
6861 /*
6862  * Enqueue the vhcache client (cct) at the tail of the list
6863  */
6864 static void
6865 enqueue_vhcache_client(mdi_vhci_cache_t *vhcache,
6866     mdi_vhcache_client_t *cct)
6867 {
6868 	cct->cct_next = NULL;
6869 	if (vhcache->vhcache_client_head == NULL)
6870 		vhcache->vhcache_client_head = cct;
6871 	else
6872 		vhcache->vhcache_client_tail->cct_next = cct;
6873 	vhcache->vhcache_client_tail = cct;
6874 }
6875 
6876 static void
6877 free_string_array(char **str, int nelem)
6878 {
6879 	int i;
6880 
6881 	if (str) {
6882 		for (i = 0; i < nelem; i++) {
6883 			if (str[i])
6884 				kmem_free(str[i], strlen(str[i]) + 1);
6885 		}
6886 		kmem_free(str, sizeof (char *) * nelem);
6887 	}
6888 }
6889 
6890 static void
6891 free_vhcache_phci(mdi_vhcache_phci_t *cphci)
6892 {
6893 	kmem_free(cphci->cphci_path, strlen(cphci->cphci_path) + 1);
6894 	kmem_free(cphci, sizeof (*cphci));
6895 }
6896 
6897 static void
6898 free_vhcache_pathinfo(mdi_vhcache_pathinfo_t *cpi)
6899 {
6900 	kmem_free(cpi->cpi_addr, strlen(cpi->cpi_addr) + 1);
6901 	kmem_free(cpi, sizeof (*cpi));
6902 }
6903 
6904 static void
6905 free_vhcache_client(mdi_vhcache_client_t *cct)
6906 {
6907 	kmem_free(cct->cct_name_addr, strlen(cct->cct_name_addr) + 1);
6908 	kmem_free(cct, sizeof (*cct));
6909 }
6910 
6911 static char *
6912 vhcache_mknameaddr(char *ct_name, char *ct_addr, int *ret_len)
6913 {
6914 	char *name_addr;
6915 	int len;
6916 
6917 	len = strlen(ct_name) + strlen(ct_addr) + 2;
6918 	name_addr = kmem_alloc(len, KM_SLEEP);
6919 	(void) snprintf(name_addr, len, "%s@%s", ct_name, ct_addr);
6920 
6921 	if (ret_len)
6922 		*ret_len = len;
6923 	return (name_addr);
6924 }
6925 
6926 /*
6927  * Copy the contents of paddrnvl to vhci cache.
6928  * paddrnvl nvlist contains path information for a vhci client.
6929  * See the comment in mainnvl_to_vhcache() for the format of this nvlist.
6930  */
6931 static void
6932 paddrnvl_to_vhcache(nvlist_t *nvl, mdi_vhcache_phci_t *cphci_list[],
6933     mdi_vhcache_client_t *cct)
6934 {
6935 	nvpair_t *nvp = NULL;
6936 	mdi_vhcache_pathinfo_t *cpi;
6937 	uint_t nelem;
6938 	uint32_t *val;
6939 
6940 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
6941 		ASSERT(nvpair_type(nvp) == DATA_TYPE_UINT32_ARRAY);
6942 		cpi = kmem_zalloc(sizeof (*cpi), KM_SLEEP);
6943 		cpi->cpi_addr = i_ddi_strdup(nvpair_name(nvp), KM_SLEEP);
6944 		(void) nvpair_value_uint32_array(nvp, &val, &nelem);
6945 		ASSERT(nelem == 2);
6946 		cpi->cpi_cphci = cphci_list[val[0]];
6947 		cpi->cpi_flags = val[1];
6948 		enqueue_tail_vhcache_pathinfo(cct, cpi);
6949 	}
6950 }
6951 
6952 /*
6953  * Copy the contents of caddrmapnvl to vhci cache.
6954  * caddrmapnvl nvlist contains vhci client address to phci client address
6955  * mappings. See the comment in mainnvl_to_vhcache() for the format of
6956  * this nvlist.
6957  */
6958 static void
6959 caddrmapnvl_to_vhcache(mdi_vhci_cache_t *vhcache, nvlist_t *nvl,
6960     mdi_vhcache_phci_t *cphci_list[])
6961 {
6962 	nvpair_t *nvp = NULL;
6963 	nvlist_t *paddrnvl;
6964 	mdi_vhcache_client_t *cct;
6965 
6966 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
6967 		ASSERT(nvpair_type(nvp) == DATA_TYPE_NVLIST);
6968 		cct = kmem_zalloc(sizeof (*cct), KM_SLEEP);
6969 		cct->cct_name_addr = i_ddi_strdup(nvpair_name(nvp), KM_SLEEP);
6970 		(void) nvpair_value_nvlist(nvp, &paddrnvl);
6971 		paddrnvl_to_vhcache(paddrnvl, cphci_list, cct);
6972 		/* the client must contain at least one path */
6973 		ASSERT(cct->cct_cpi_head != NULL);
6974 
6975 		enqueue_vhcache_client(vhcache, cct);
6976 		(void) mod_hash_insert(vhcache->vhcache_client_hash,
6977 		    (mod_hash_key_t)cct->cct_name_addr, (mod_hash_val_t)cct);
6978 	}
6979 }
6980 
6981 /*
6982  * Copy the contents of the main nvlist to vhci cache.
6983  *
6984  * VHCI busconfig cached data is stored in the form of a nvlist on the disk.
6985  * The nvlist contains the mappings between the vhci client addresses and
6986  * their corresponding phci client addresses.
6987  *
6988  * The structure of the nvlist is as follows:
6989  *
6990  * Main nvlist:
6991  *	NAME		TYPE		DATA
6992  *	version		int32		version number
6993  *	phcis		string array	array of phci paths
6994  *	clientaddrmap	nvlist_t	c2paddrs_nvl (see below)
6995  *
6996  * structure of c2paddrs_nvl:
6997  *	NAME		TYPE		DATA
6998  *	caddr1		nvlist_t	paddrs_nvl1
6999  *	caddr2		nvlist_t	paddrs_nvl2
7000  *	...
7001  * where caddr1, caddr2, ... are vhci client name and addresses in the
7002  * form of "<clientname>@<clientaddress>".
7003  * (for example: "ssd@2000002037cd9f72");
7004  * paddrs_nvl1, paddrs_nvl2, .. are nvlists that contain path information.
7005  *
7006  * structure of paddrs_nvl:
7007  *	NAME		TYPE		DATA
7008  *	pi_addr1	uint32_array	(phci-id, cpi_flags)
7009  *	pi_addr2	uint32_array	(phci-id, cpi_flags)
7010  *	...
7011  * where pi_addr1, pi_addr2, ... are bus specific addresses of pathinfo nodes
7012  * (so called pi_addrs, for example: "w2100002037cd9f72,0");
7013  * phci-ids are integers that identify PHCIs to which the
7014  * the bus specific address belongs to. These integers are used as an index
7015  * into to the phcis string array in the main nvlist to get the PHCI path.
7016  */
7017 static int
7018 mainnvl_to_vhcache(mdi_vhci_cache_t *vhcache, nvlist_t *nvl)
7019 {
7020 	char **phcis, **phci_namep;
7021 	uint_t nphcis;
7022 	mdi_vhcache_phci_t *cphci, **cphci_list;
7023 	nvlist_t *caddrmapnvl;
7024 	int32_t ver;
7025 	int i;
7026 	size_t cphci_list_size;
7027 
7028 	ASSERT(RW_WRITE_HELD(&vhcache->vhcache_lock));
7029 
7030 	if (nvlist_lookup_int32(nvl, MDI_NVPNAME_VERSION, &ver) != 0 ||
7031 	    ver != MDI_VHCI_CACHE_VERSION)
7032 		return (MDI_FAILURE);
7033 
7034 	if (nvlist_lookup_string_array(nvl, MDI_NVPNAME_PHCIS, &phcis,
7035 	    &nphcis) != 0)
7036 		return (MDI_SUCCESS);
7037 
7038 	ASSERT(nphcis > 0);
7039 
7040 	cphci_list_size = sizeof (mdi_vhcache_phci_t *) * nphcis;
7041 	cphci_list = kmem_alloc(cphci_list_size, KM_SLEEP);
7042 	for (i = 0, phci_namep = phcis; i < nphcis; i++, phci_namep++) {
7043 		cphci = kmem_zalloc(sizeof (mdi_vhcache_phci_t), KM_SLEEP);
7044 		cphci->cphci_path = i_ddi_strdup(*phci_namep, KM_SLEEP);
7045 		enqueue_vhcache_phci(vhcache, cphci);
7046 		cphci_list[i] = cphci;
7047 	}
7048 
7049 	ASSERT(vhcache->vhcache_phci_head != NULL);
7050 
7051 	if (nvlist_lookup_nvlist(nvl, MDI_NVPNAME_CTADDRMAP, &caddrmapnvl) == 0)
7052 		caddrmapnvl_to_vhcache(vhcache, caddrmapnvl, cphci_list);
7053 
7054 	kmem_free(cphci_list, cphci_list_size);
7055 	return (MDI_SUCCESS);
7056 }
7057 
7058 /*
7059  * Build paddrnvl for the specified client using the information in the
7060  * vhci cache and add it to the caddrmapnnvl.
7061  * Returns 0 on success, errno on failure.
7062  */
7063 static int
7064 vhcache_to_paddrnvl(mdi_vhci_cache_t *vhcache, mdi_vhcache_client_t *cct,
7065     nvlist_t *caddrmapnvl)
7066 {
7067 	mdi_vhcache_pathinfo_t *cpi;
7068 	nvlist_t *nvl;
7069 	int err;
7070 	uint32_t val[2];
7071 
7072 	ASSERT(RW_LOCK_HELD(&vhcache->vhcache_lock));
7073 
7074 	if ((err = nvlist_alloc(&nvl, 0, KM_SLEEP)) != 0)
7075 		return (err);
7076 
7077 	for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi->cpi_next) {
7078 		val[0] = cpi->cpi_cphci->cphci_id;
7079 		val[1] = cpi->cpi_flags;
7080 		if ((err = nvlist_add_uint32_array(nvl, cpi->cpi_addr, val, 2))
7081 		    != 0)
7082 			goto out;
7083 	}
7084 
7085 	err = nvlist_add_nvlist(caddrmapnvl, cct->cct_name_addr, nvl);
7086 out:
7087 	nvlist_free(nvl);
7088 	return (err);
7089 }
7090 
7091 /*
7092  * Build caddrmapnvl using the information in the vhci cache
7093  * and add it to the mainnvl.
7094  * Returns 0 on success, errno on failure.
7095  */
7096 static int
7097 vhcache_to_caddrmapnvl(mdi_vhci_cache_t *vhcache, nvlist_t *mainnvl)
7098 {
7099 	mdi_vhcache_client_t *cct;
7100 	nvlist_t *nvl;
7101 	int err;
7102 
7103 	ASSERT(RW_LOCK_HELD(&vhcache->vhcache_lock));
7104 
7105 	if ((err = nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP)) != 0)
7106 		return (err);
7107 
7108 	for (cct = vhcache->vhcache_client_head; cct != NULL;
7109 	    cct = cct->cct_next) {
7110 		if ((err = vhcache_to_paddrnvl(vhcache, cct, nvl)) != 0)
7111 			goto out;
7112 	}
7113 
7114 	err = nvlist_add_nvlist(mainnvl, MDI_NVPNAME_CTADDRMAP, nvl);
7115 out:
7116 	nvlist_free(nvl);
7117 	return (err);
7118 }
7119 
7120 /*
7121  * Build nvlist using the information in the vhci cache.
7122  * See the comment in mainnvl_to_vhcache() for the format of the nvlist.
7123  * Returns nvl on success, NULL on failure.
7124  */
7125 static nvlist_t *
7126 vhcache_to_mainnvl(mdi_vhci_cache_t *vhcache)
7127 {
7128 	mdi_vhcache_phci_t *cphci;
7129 	uint_t phci_count;
7130 	char **phcis;
7131 	nvlist_t *nvl;
7132 	int err, i;
7133 
7134 	if ((err = nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP)) != 0) {
7135 		nvl = NULL;
7136 		goto out;
7137 	}
7138 
7139 	if ((err = nvlist_add_int32(nvl, MDI_NVPNAME_VERSION,
7140 	    MDI_VHCI_CACHE_VERSION)) != 0)
7141 		goto out;
7142 
7143 	rw_enter(&vhcache->vhcache_lock, RW_READER);
7144 	if (vhcache->vhcache_phci_head == NULL) {
7145 		rw_exit(&vhcache->vhcache_lock);
7146 		return (nvl);
7147 	}
7148 
7149 	phci_count = 0;
7150 	for (cphci = vhcache->vhcache_phci_head; cphci != NULL;
7151 	    cphci = cphci->cphci_next)
7152 		cphci->cphci_id = phci_count++;
7153 
7154 	/* build phci pathname list */
7155 	phcis = kmem_alloc(sizeof (char *) * phci_count, KM_SLEEP);
7156 	for (cphci = vhcache->vhcache_phci_head, i = 0; cphci != NULL;
7157 	    cphci = cphci->cphci_next, i++)
7158 		phcis[i] = i_ddi_strdup(cphci->cphci_path, KM_SLEEP);
7159 
7160 	err = nvlist_add_string_array(nvl, MDI_NVPNAME_PHCIS, phcis,
7161 	    phci_count);
7162 	free_string_array(phcis, phci_count);
7163 
7164 	if (err == 0 &&
7165 	    (err = vhcache_to_caddrmapnvl(vhcache, nvl)) == 0) {
7166 		rw_exit(&vhcache->vhcache_lock);
7167 		return (nvl);
7168 	}
7169 
7170 	rw_exit(&vhcache->vhcache_lock);
7171 out:
7172 	if (nvl)
7173 		nvlist_free(nvl);
7174 	return (NULL);
7175 }
7176 
7177 /*
7178  * Lookup vhcache phci structure for the specified phci path.
7179  */
7180 static mdi_vhcache_phci_t *
7181 lookup_vhcache_phci_by_name(mdi_vhci_cache_t *vhcache, char *phci_path)
7182 {
7183 	mdi_vhcache_phci_t *cphci;
7184 
7185 	ASSERT(RW_LOCK_HELD(&vhcache->vhcache_lock));
7186 
7187 	for (cphci = vhcache->vhcache_phci_head; cphci != NULL;
7188 	    cphci = cphci->cphci_next) {
7189 		if (strcmp(cphci->cphci_path, phci_path) == 0)
7190 			return (cphci);
7191 	}
7192 
7193 	return (NULL);
7194 }
7195 
7196 /*
7197  * Lookup vhcache phci structure for the specified phci.
7198  */
7199 static mdi_vhcache_phci_t *
7200 lookup_vhcache_phci_by_addr(mdi_vhci_cache_t *vhcache, mdi_phci_t *ph)
7201 {
7202 	mdi_vhcache_phci_t *cphci;
7203 
7204 	ASSERT(RW_LOCK_HELD(&vhcache->vhcache_lock));
7205 
7206 	for (cphci = vhcache->vhcache_phci_head; cphci != NULL;
7207 	    cphci = cphci->cphci_next) {
7208 		if (cphci->cphci_phci == ph)
7209 			return (cphci);
7210 	}
7211 
7212 	return (NULL);
7213 }
7214 
7215 /*
7216  * Add the specified phci to the vhci cache if not already present.
7217  */
7218 static void
7219 vhcache_phci_add(mdi_vhci_config_t *vhc, mdi_phci_t *ph)
7220 {
7221 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7222 	mdi_vhcache_phci_t *cphci;
7223 	char *pathname;
7224 	int cache_updated;
7225 
7226 	rw_enter(&vhcache->vhcache_lock, RW_WRITER);
7227 
7228 	pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
7229 	(void) ddi_pathname(ph->ph_dip, pathname);
7230 	if ((cphci = lookup_vhcache_phci_by_name(vhcache, pathname))
7231 	    != NULL) {
7232 		cphci->cphci_phci = ph;
7233 		cache_updated = 0;
7234 	} else {
7235 		cphci = kmem_zalloc(sizeof (*cphci), KM_SLEEP);
7236 		cphci->cphci_path = i_ddi_strdup(pathname, KM_SLEEP);
7237 		cphci->cphci_phci = ph;
7238 		enqueue_vhcache_phci(vhcache, cphci);
7239 		cache_updated = 1;
7240 	}
7241 
7242 	rw_exit(&vhcache->vhcache_lock);
7243 
7244 	/*
7245 	 * Since a new phci has been added, reset
7246 	 * vhc_path_discovery_cutoff_time to allow for discovery of paths
7247 	 * during next vhcache_discover_paths().
7248 	 */
7249 	mutex_enter(&vhc->vhc_lock);
7250 	vhc->vhc_path_discovery_cutoff_time = 0;
7251 	mutex_exit(&vhc->vhc_lock);
7252 
7253 	kmem_free(pathname, MAXPATHLEN);
7254 	if (cache_updated)
7255 		vhcache_dirty(vhc);
7256 }
7257 
7258 /*
7259  * Remove the reference to the specified phci from the vhci cache.
7260  */
7261 static void
7262 vhcache_phci_remove(mdi_vhci_config_t *vhc, mdi_phci_t *ph)
7263 {
7264 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7265 	mdi_vhcache_phci_t *cphci;
7266 
7267 	rw_enter(&vhcache->vhcache_lock, RW_WRITER);
7268 	if ((cphci = lookup_vhcache_phci_by_addr(vhcache, ph)) != NULL) {
7269 		/* do not remove the actual mdi_vhcache_phci structure */
7270 		cphci->cphci_phci = NULL;
7271 	}
7272 	rw_exit(&vhcache->vhcache_lock);
7273 }
7274 
7275 static void
7276 init_vhcache_lookup_token(mdi_vhcache_lookup_token_t *dst,
7277     mdi_vhcache_lookup_token_t *src)
7278 {
7279 	if (src == NULL) {
7280 		dst->lt_cct = NULL;
7281 		dst->lt_cct_lookup_time = 0;
7282 	} else {
7283 		dst->lt_cct = src->lt_cct;
7284 		dst->lt_cct_lookup_time = src->lt_cct_lookup_time;
7285 	}
7286 }
7287 
7288 /*
7289  * Look up vhcache client for the specified client.
7290  */
7291 static mdi_vhcache_client_t *
7292 lookup_vhcache_client(mdi_vhci_cache_t *vhcache, char *ct_name, char *ct_addr,
7293     mdi_vhcache_lookup_token_t *token)
7294 {
7295 	mod_hash_val_t hv;
7296 	char *name_addr;
7297 	int len;
7298 
7299 	ASSERT(RW_LOCK_HELD(&vhcache->vhcache_lock));
7300 
7301 	/*
7302 	 * If no vhcache clean occurred since the last lookup, we can
7303 	 * simply return the cct from the last lookup operation.
7304 	 * It works because ccts are never freed except during the vhcache
7305 	 * cleanup operation.
7306 	 */
7307 	if (token != NULL &&
7308 	    vhcache->vhcache_clean_time < token->lt_cct_lookup_time)
7309 		return (token->lt_cct);
7310 
7311 	name_addr = vhcache_mknameaddr(ct_name, ct_addr, &len);
7312 	if (mod_hash_find(vhcache->vhcache_client_hash,
7313 	    (mod_hash_key_t)name_addr, &hv) == 0) {
7314 		if (token) {
7315 			token->lt_cct = (mdi_vhcache_client_t *)hv;
7316 			token->lt_cct_lookup_time = lbolt64;
7317 		}
7318 	} else {
7319 		if (token) {
7320 			token->lt_cct = NULL;
7321 			token->lt_cct_lookup_time = 0;
7322 		}
7323 		hv = NULL;
7324 	}
7325 	kmem_free(name_addr, len);
7326 	return ((mdi_vhcache_client_t *)hv);
7327 }
7328 
7329 /*
7330  * Add the specified path to the vhci cache if not already present.
7331  * Also add the vhcache client for the client corresponding to this path
7332  * if it doesn't already exist.
7333  */
7334 static void
7335 vhcache_pi_add(mdi_vhci_config_t *vhc, struct mdi_pathinfo *pip)
7336 {
7337 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7338 	mdi_vhcache_client_t *cct;
7339 	mdi_vhcache_pathinfo_t *cpi;
7340 	mdi_phci_t *ph = pip->pi_phci;
7341 	mdi_client_t *ct = pip->pi_client;
7342 	int cache_updated = 0;
7343 
7344 	rw_enter(&vhcache->vhcache_lock, RW_WRITER);
7345 
7346 	/* if vhcache client for this pip doesn't already exist, add it */
7347 	if ((cct = lookup_vhcache_client(vhcache, ct->ct_drvname, ct->ct_guid,
7348 	    NULL)) == NULL) {
7349 		cct = kmem_zalloc(sizeof (*cct), KM_SLEEP);
7350 		cct->cct_name_addr = vhcache_mknameaddr(ct->ct_drvname,
7351 		    ct->ct_guid, NULL);
7352 		enqueue_vhcache_client(vhcache, cct);
7353 		(void) mod_hash_insert(vhcache->vhcache_client_hash,
7354 		    (mod_hash_key_t)cct->cct_name_addr, (mod_hash_val_t)cct);
7355 		cache_updated = 1;
7356 	}
7357 
7358 	for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi->cpi_next) {
7359 		if (cpi->cpi_cphci->cphci_phci == ph &&
7360 		    strcmp(cpi->cpi_addr, pip->pi_addr) == 0) {
7361 			cpi->cpi_pip = pip;
7362 			if (cpi->cpi_flags & MDI_CPI_HINT_PATH_DOES_NOT_EXIST) {
7363 				cpi->cpi_flags &=
7364 				    ~MDI_CPI_HINT_PATH_DOES_NOT_EXIST;
7365 				sort_vhcache_paths(cct);
7366 				cache_updated = 1;
7367 			}
7368 			break;
7369 		}
7370 	}
7371 
7372 	if (cpi == NULL) {
7373 		cpi = kmem_zalloc(sizeof (*cpi), KM_SLEEP);
7374 		cpi->cpi_addr = i_ddi_strdup(pip->pi_addr, KM_SLEEP);
7375 		cpi->cpi_cphci = lookup_vhcache_phci_by_addr(vhcache, ph);
7376 		ASSERT(cpi->cpi_cphci != NULL);
7377 		cpi->cpi_pip = pip;
7378 		enqueue_vhcache_pathinfo(cct, cpi);
7379 		cache_updated = 1;
7380 	}
7381 
7382 	rw_exit(&vhcache->vhcache_lock);
7383 
7384 	if (cache_updated)
7385 		vhcache_dirty(vhc);
7386 }
7387 
7388 /*
7389  * Remove the reference to the specified path from the vhci cache.
7390  */
7391 static void
7392 vhcache_pi_remove(mdi_vhci_config_t *vhc, struct mdi_pathinfo *pip)
7393 {
7394 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7395 	mdi_client_t *ct = pip->pi_client;
7396 	mdi_vhcache_client_t *cct;
7397 	mdi_vhcache_pathinfo_t *cpi;
7398 
7399 	rw_enter(&vhcache->vhcache_lock, RW_WRITER);
7400 	if ((cct = lookup_vhcache_client(vhcache, ct->ct_drvname, ct->ct_guid,
7401 	    NULL)) != NULL) {
7402 		for (cpi = cct->cct_cpi_head; cpi != NULL;
7403 		    cpi = cpi->cpi_next) {
7404 			if (cpi->cpi_pip == pip) {
7405 				cpi->cpi_pip = NULL;
7406 				break;
7407 			}
7408 		}
7409 	}
7410 	rw_exit(&vhcache->vhcache_lock);
7411 }
7412 
7413 /*
7414  * Flush the vhci cache to disk.
7415  * Returns MDI_SUCCESS on success, MDI_FAILURE on failure.
7416  */
7417 static int
7418 flush_vhcache(mdi_vhci_config_t *vhc, int force_flag)
7419 {
7420 	nvlist_t *nvl;
7421 	int err;
7422 	int rv;
7423 
7424 	/*
7425 	 * It is possible that the system may shutdown before
7426 	 * i_ddi_io_initialized (during stmsboot for example). To allow for
7427 	 * flushing the cache in this case do not check for
7428 	 * i_ddi_io_initialized when force flag is set.
7429 	 */
7430 	if (force_flag == 0 && !i_ddi_io_initialized())
7431 		return (MDI_FAILURE);
7432 
7433 	if ((nvl = vhcache_to_mainnvl(&vhc->vhc_vhcache)) != NULL) {
7434 		err = fwrite_nvlist(vhc->vhc_vhcache_filename, nvl);
7435 		nvlist_free(nvl);
7436 	} else
7437 		err = EFAULT;
7438 
7439 	rv = MDI_SUCCESS;
7440 	mutex_enter(&vhc->vhc_lock);
7441 	if (err != 0) {
7442 		if (err == EROFS) {
7443 			vhc->vhc_flags |= MDI_VHC_READONLY_FS;
7444 			vhc->vhc_flags &= ~(MDI_VHC_VHCACHE_FLUSH_ERROR |
7445 			    MDI_VHC_VHCACHE_DIRTY);
7446 		} else {
7447 			if (!(vhc->vhc_flags & MDI_VHC_VHCACHE_FLUSH_ERROR)) {
7448 				cmn_err(CE_CONT, "%s: update failed\n",
7449 				    vhc->vhc_vhcache_filename);
7450 				vhc->vhc_flags |= MDI_VHC_VHCACHE_FLUSH_ERROR;
7451 			}
7452 			rv = MDI_FAILURE;
7453 		}
7454 	} else if (vhc->vhc_flags & MDI_VHC_VHCACHE_FLUSH_ERROR) {
7455 		cmn_err(CE_CONT,
7456 		    "%s: update now ok\n", vhc->vhc_vhcache_filename);
7457 		vhc->vhc_flags &= ~MDI_VHC_VHCACHE_FLUSH_ERROR;
7458 	}
7459 	mutex_exit(&vhc->vhc_lock);
7460 
7461 	return (rv);
7462 }
7463 
7464 /*
7465  * Call flush_vhcache() to flush the vhci cache at the scheduled time.
7466  * Exits itself if left idle for the idle timeout period.
7467  */
7468 static void
7469 vhcache_flush_thread(void *arg)
7470 {
7471 	mdi_vhci_config_t *vhc = (mdi_vhci_config_t *)arg;
7472 	clock_t idle_time, quit_at_ticks;
7473 	callb_cpr_t cprinfo;
7474 
7475 	/* number of seconds to sleep idle before exiting */
7476 	idle_time = mdi_vhcache_flush_daemon_idle_time * TICKS_PER_SECOND;
7477 
7478 	CALLB_CPR_INIT(&cprinfo, &vhc->vhc_lock, callb_generic_cpr,
7479 	    "mdi_vhcache_flush");
7480 	mutex_enter(&vhc->vhc_lock);
7481 	for (; ; ) {
7482 		while (!(vhc->vhc_flags & MDI_VHC_EXIT) &&
7483 		    (vhc->vhc_flags & MDI_VHC_VHCACHE_DIRTY)) {
7484 			if (ddi_get_lbolt() < vhc->vhc_flush_at_ticks) {
7485 				CALLB_CPR_SAFE_BEGIN(&cprinfo);
7486 				(void) cv_timedwait(&vhc->vhc_cv,
7487 				    &vhc->vhc_lock, vhc->vhc_flush_at_ticks);
7488 				CALLB_CPR_SAFE_END(&cprinfo, &vhc->vhc_lock);
7489 			} else {
7490 				vhc->vhc_flags &= ~MDI_VHC_VHCACHE_DIRTY;
7491 				mutex_exit(&vhc->vhc_lock);
7492 
7493 				if (flush_vhcache(vhc, 0) != MDI_SUCCESS)
7494 					vhcache_dirty(vhc);
7495 
7496 				mutex_enter(&vhc->vhc_lock);
7497 			}
7498 		}
7499 
7500 		quit_at_ticks = ddi_get_lbolt() + idle_time;
7501 
7502 		while (!(vhc->vhc_flags & MDI_VHC_EXIT) &&
7503 		    !(vhc->vhc_flags & MDI_VHC_VHCACHE_DIRTY) &&
7504 		    ddi_get_lbolt() < quit_at_ticks) {
7505 			CALLB_CPR_SAFE_BEGIN(&cprinfo);
7506 			(void) cv_timedwait(&vhc->vhc_cv, &vhc->vhc_lock,
7507 			    quit_at_ticks);
7508 			CALLB_CPR_SAFE_END(&cprinfo, &vhc->vhc_lock);
7509 		}
7510 
7511 		if ((vhc->vhc_flags & MDI_VHC_EXIT) ||
7512 		    !(vhc->vhc_flags & MDI_VHC_VHCACHE_DIRTY))
7513 			goto out;
7514 	}
7515 
7516 out:
7517 	vhc->vhc_flags &= ~MDI_VHC_VHCACHE_FLUSH_THREAD;
7518 	/* CALLB_CPR_EXIT releases the vhc->vhc_lock */
7519 	CALLB_CPR_EXIT(&cprinfo);
7520 }
7521 
7522 /*
7523  * Make vhci cache dirty and schedule flushing by vhcache flush thread.
7524  */
7525 static void
7526 vhcache_dirty(mdi_vhci_config_t *vhc)
7527 {
7528 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7529 	int create_thread;
7530 
7531 	rw_enter(&vhcache->vhcache_lock, RW_READER);
7532 	/* do not flush cache until the cache is fully built */
7533 	if (!(vhcache->vhcache_flags & MDI_VHCI_CACHE_SETUP_DONE)) {
7534 		rw_exit(&vhcache->vhcache_lock);
7535 		return;
7536 	}
7537 	rw_exit(&vhcache->vhcache_lock);
7538 
7539 	mutex_enter(&vhc->vhc_lock);
7540 	if (vhc->vhc_flags & MDI_VHC_READONLY_FS) {
7541 		mutex_exit(&vhc->vhc_lock);
7542 		return;
7543 	}
7544 
7545 	vhc->vhc_flags |= MDI_VHC_VHCACHE_DIRTY;
7546 	vhc->vhc_flush_at_ticks = ddi_get_lbolt() +
7547 	    mdi_vhcache_flush_delay * TICKS_PER_SECOND;
7548 	if (vhc->vhc_flags & MDI_VHC_VHCACHE_FLUSH_THREAD) {
7549 		cv_broadcast(&vhc->vhc_cv);
7550 		create_thread = 0;
7551 	} else {
7552 		vhc->vhc_flags |= MDI_VHC_VHCACHE_FLUSH_THREAD;
7553 		create_thread = 1;
7554 	}
7555 	mutex_exit(&vhc->vhc_lock);
7556 
7557 	if (create_thread)
7558 		(void) thread_create(NULL, 0, vhcache_flush_thread, vhc,
7559 		    0, &p0, TS_RUN, minclsyspri);
7560 }
7561 
7562 /*
7563  * phci bus config structure - one for for each phci bus config operation that
7564  * we initiate on behalf of a vhci.
7565  */
7566 typedef struct mdi_phci_bus_config_s {
7567 	char *phbc_phci_path;
7568 	struct mdi_vhci_bus_config_s *phbc_vhbusconfig;	/* vhci bus config */
7569 	struct mdi_phci_bus_config_s *phbc_next;
7570 } mdi_phci_bus_config_t;
7571 
7572 /* vhci bus config structure - one for each vhci bus config operation */
7573 typedef struct mdi_vhci_bus_config_s {
7574 	ddi_bus_config_op_t vhbc_op;	/* bus config op */
7575 	major_t vhbc_op_major;		/* bus config op major */
7576 	uint_t vhbc_op_flags;		/* bus config op flags */
7577 	kmutex_t vhbc_lock;
7578 	kcondvar_t vhbc_cv;
7579 	int vhbc_thr_count;
7580 } mdi_vhci_bus_config_t;
7581 
7582 /*
7583  * bus config the specified phci
7584  */
7585 static void
7586 bus_config_phci(void *arg)
7587 {
7588 	mdi_phci_bus_config_t *phbc = (mdi_phci_bus_config_t *)arg;
7589 	mdi_vhci_bus_config_t *vhbc = phbc->phbc_vhbusconfig;
7590 	dev_info_t *ph_dip;
7591 
7592 	/*
7593 	 * first configure all path components upto phci and then configure
7594 	 * the phci children.
7595 	 */
7596 	if ((ph_dip = e_ddi_hold_devi_by_path(phbc->phbc_phci_path, 0))
7597 	    != NULL) {
7598 		if (vhbc->vhbc_op == BUS_CONFIG_DRIVER ||
7599 		    vhbc->vhbc_op == BUS_UNCONFIG_DRIVER) {
7600 			(void) ndi_devi_config_driver(ph_dip,
7601 			    vhbc->vhbc_op_flags,
7602 			    vhbc->vhbc_op_major);
7603 		} else
7604 			(void) ndi_devi_config(ph_dip,
7605 			    vhbc->vhbc_op_flags);
7606 
7607 		/* release the hold that e_ddi_hold_devi_by_path() placed */
7608 		ndi_rele_devi(ph_dip);
7609 	}
7610 
7611 	kmem_free(phbc->phbc_phci_path, strlen(phbc->phbc_phci_path) + 1);
7612 	kmem_free(phbc, sizeof (*phbc));
7613 
7614 	mutex_enter(&vhbc->vhbc_lock);
7615 	vhbc->vhbc_thr_count--;
7616 	if (vhbc->vhbc_thr_count == 0)
7617 		cv_broadcast(&vhbc->vhbc_cv);
7618 	mutex_exit(&vhbc->vhbc_lock);
7619 }
7620 
7621 /*
7622  * Bus config all phcis associated with the vhci in parallel.
7623  * op must be BUS_CONFIG_DRIVER or BUS_CONFIG_ALL.
7624  */
7625 static void
7626 bus_config_all_phcis(mdi_vhci_cache_t *vhcache, uint_t flags,
7627     ddi_bus_config_op_t op, major_t maj)
7628 {
7629 	mdi_phci_bus_config_t *phbc_head = NULL, *phbc, *phbc_next;
7630 	mdi_vhci_bus_config_t *vhbc;
7631 	mdi_vhcache_phci_t *cphci;
7632 
7633 	rw_enter(&vhcache->vhcache_lock, RW_READER);
7634 	if (vhcache->vhcache_phci_head == NULL) {
7635 		rw_exit(&vhcache->vhcache_lock);
7636 		return;
7637 	}
7638 
7639 	vhbc = kmem_zalloc(sizeof (*vhbc), KM_SLEEP);
7640 
7641 	for (cphci = vhcache->vhcache_phci_head; cphci != NULL;
7642 	    cphci = cphci->cphci_next) {
7643 		phbc = kmem_zalloc(sizeof (*phbc), KM_SLEEP);
7644 		phbc->phbc_phci_path = i_ddi_strdup(cphci->cphci_path,
7645 		    KM_SLEEP);
7646 		phbc->phbc_vhbusconfig = vhbc;
7647 		phbc->phbc_next = phbc_head;
7648 		phbc_head = phbc;
7649 		vhbc->vhbc_thr_count++;
7650 	}
7651 	rw_exit(&vhcache->vhcache_lock);
7652 
7653 	vhbc->vhbc_op = op;
7654 	vhbc->vhbc_op_major = maj;
7655 	vhbc->vhbc_op_flags = NDI_NO_EVENT |
7656 	    (flags & (NDI_CONFIG_REPROBE | NDI_DRV_CONF_REPROBE));
7657 	mutex_init(&vhbc->vhbc_lock, NULL, MUTEX_DEFAULT, NULL);
7658 	cv_init(&vhbc->vhbc_cv, NULL, CV_DRIVER, NULL);
7659 
7660 	/* now create threads to initiate bus config on all phcis in parallel */
7661 	for (phbc = phbc_head; phbc != NULL; phbc = phbc_next) {
7662 		phbc_next = phbc->phbc_next;
7663 		if (mdi_mtc_off)
7664 			bus_config_phci((void *)phbc);
7665 		else
7666 			(void) thread_create(NULL, 0, bus_config_phci, phbc,
7667 			    0, &p0, TS_RUN, minclsyspri);
7668 	}
7669 
7670 	mutex_enter(&vhbc->vhbc_lock);
7671 	/* wait until all threads exit */
7672 	while (vhbc->vhbc_thr_count > 0)
7673 		cv_wait(&vhbc->vhbc_cv, &vhbc->vhbc_lock);
7674 	mutex_exit(&vhbc->vhbc_lock);
7675 
7676 	mutex_destroy(&vhbc->vhbc_lock);
7677 	cv_destroy(&vhbc->vhbc_cv);
7678 	kmem_free(vhbc, sizeof (*vhbc));
7679 }
7680 
7681 /*
7682  * Single threaded version of bus_config_all_phcis()
7683  */
7684 static void
7685 st_bus_config_all_phcis(mdi_vhci_config_t *vhc, uint_t flags,
7686     ddi_bus_config_op_t op, major_t maj)
7687 {
7688 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7689 
7690 	single_threaded_vhconfig_enter(vhc);
7691 	bus_config_all_phcis(vhcache, flags, op, maj);
7692 	single_threaded_vhconfig_exit(vhc);
7693 }
7694 
7695 /*
7696  * Perform BUS_CONFIG_ONE on the specified child of the phci.
7697  * The path includes the child component in addition to the phci path.
7698  */
7699 static int
7700 bus_config_one_phci_child(char *path)
7701 {
7702 	dev_info_t *ph_dip, *child;
7703 	char *devnm;
7704 	int rv = MDI_FAILURE;
7705 
7706 	/* extract the child component of the phci */
7707 	devnm = strrchr(path, '/');
7708 	*devnm++ = '\0';
7709 
7710 	/*
7711 	 * first configure all path components upto phci and then
7712 	 * configure the phci child.
7713 	 */
7714 	if ((ph_dip = e_ddi_hold_devi_by_path(path, 0)) != NULL) {
7715 		if (ndi_devi_config_one(ph_dip, devnm, &child, NDI_NO_EVENT) ==
7716 		    NDI_SUCCESS) {
7717 			/*
7718 			 * release the hold that ndi_devi_config_one() placed
7719 			 */
7720 			ndi_rele_devi(child);
7721 			rv = MDI_SUCCESS;
7722 		}
7723 
7724 		/* release the hold that e_ddi_hold_devi_by_path() placed */
7725 		ndi_rele_devi(ph_dip);
7726 	}
7727 
7728 	devnm--;
7729 	*devnm = '/';
7730 	return (rv);
7731 }
7732 
7733 /*
7734  * Build a list of phci client paths for the specified vhci client.
7735  * The list includes only those phci client paths which aren't configured yet.
7736  */
7737 static mdi_phys_path_t *
7738 build_phclient_path_list(mdi_vhcache_client_t *cct, char *ct_name)
7739 {
7740 	mdi_vhcache_pathinfo_t *cpi;
7741 	mdi_phys_path_t *pp_head = NULL, *pp_tail = NULL, *pp;
7742 	int config_path, len;
7743 
7744 	for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi->cpi_next) {
7745 		/*
7746 		 * include only those paths that aren't configured.
7747 		 */
7748 		config_path = 0;
7749 		if (cpi->cpi_pip == NULL)
7750 			config_path = 1;
7751 		else {
7752 			MDI_PI_LOCK(cpi->cpi_pip);
7753 			if (MDI_PI_IS_INIT(cpi->cpi_pip))
7754 				config_path = 1;
7755 			MDI_PI_UNLOCK(cpi->cpi_pip);
7756 		}
7757 
7758 		if (config_path) {
7759 			pp = kmem_alloc(sizeof (*pp), KM_SLEEP);
7760 			len = strlen(cpi->cpi_cphci->cphci_path) +
7761 			    strlen(ct_name) + strlen(cpi->cpi_addr) + 3;
7762 			pp->phys_path = kmem_alloc(len, KM_SLEEP);
7763 			(void) snprintf(pp->phys_path, len, "%s/%s@%s",
7764 			    cpi->cpi_cphci->cphci_path, ct_name,
7765 			    cpi->cpi_addr);
7766 			pp->phys_path_next = NULL;
7767 
7768 			if (pp_head == NULL)
7769 				pp_head = pp;
7770 			else
7771 				pp_tail->phys_path_next = pp;
7772 			pp_tail = pp;
7773 		}
7774 	}
7775 
7776 	return (pp_head);
7777 }
7778 
7779 /*
7780  * Free the memory allocated for phci client path list.
7781  */
7782 static void
7783 free_phclient_path_list(mdi_phys_path_t *pp_head)
7784 {
7785 	mdi_phys_path_t *pp, *pp_next;
7786 
7787 	for (pp = pp_head; pp != NULL; pp = pp_next) {
7788 		pp_next = pp->phys_path_next;
7789 		kmem_free(pp->phys_path, strlen(pp->phys_path) + 1);
7790 		kmem_free(pp, sizeof (*pp));
7791 	}
7792 }
7793 
7794 /*
7795  * Allocated async client structure and initialize with the specified values.
7796  */
7797 static mdi_async_client_config_t *
7798 alloc_async_client_config(char *ct_name, char *ct_addr,
7799     mdi_phys_path_t *pp_head, mdi_vhcache_lookup_token_t *tok)
7800 {
7801 	mdi_async_client_config_t *acc;
7802 
7803 	acc = kmem_alloc(sizeof (*acc), KM_SLEEP);
7804 	acc->acc_ct_name = i_ddi_strdup(ct_name, KM_SLEEP);
7805 	acc->acc_ct_addr = i_ddi_strdup(ct_addr, KM_SLEEP);
7806 	acc->acc_phclient_path_list_head = pp_head;
7807 	init_vhcache_lookup_token(&acc->acc_token, tok);
7808 	acc->acc_next = NULL;
7809 	return (acc);
7810 }
7811 
7812 /*
7813  * Free the memory allocated for the async client structure and their members.
7814  */
7815 static void
7816 free_async_client_config(mdi_async_client_config_t *acc)
7817 {
7818 	if (acc->acc_phclient_path_list_head)
7819 		free_phclient_path_list(acc->acc_phclient_path_list_head);
7820 	kmem_free(acc->acc_ct_name, strlen(acc->acc_ct_name) + 1);
7821 	kmem_free(acc->acc_ct_addr, strlen(acc->acc_ct_addr) + 1);
7822 	kmem_free(acc, sizeof (*acc));
7823 }
7824 
7825 /*
7826  * Sort vhcache pathinfos (cpis) of the specified client.
7827  * All cpis which do not have MDI_CPI_HINT_PATH_DOES_NOT_EXIST
7828  * flag set come at the beginning of the list. All cpis which have this
7829  * flag set come at the end of the list.
7830  */
7831 static void
7832 sort_vhcache_paths(mdi_vhcache_client_t *cct)
7833 {
7834 	mdi_vhcache_pathinfo_t *cpi, *cpi_next, *cpi_head;
7835 
7836 	cpi_head = cct->cct_cpi_head;
7837 	cct->cct_cpi_head = cct->cct_cpi_tail = NULL;
7838 	for (cpi = cpi_head; cpi != NULL; cpi = cpi_next) {
7839 		cpi_next = cpi->cpi_next;
7840 		enqueue_vhcache_pathinfo(cct, cpi);
7841 	}
7842 }
7843 
7844 /*
7845  * Verify whether MDI_CPI_HINT_PATH_DOES_NOT_EXIST flag setting is correct for
7846  * every vhcache pathinfo of the specified client. If not adjust the flag
7847  * setting appropriately.
7848  *
7849  * Note that MDI_CPI_HINT_PATH_DOES_NOT_EXIST flag is persisted in the
7850  * on-disk vhci cache. So every time this flag is updated the cache must be
7851  * flushed.
7852  */
7853 static void
7854 adjust_sort_vhcache_paths(mdi_vhci_config_t *vhc, char *ct_name, char *ct_addr,
7855     mdi_vhcache_lookup_token_t *tok)
7856 {
7857 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
7858 	mdi_vhcache_client_t *cct;
7859 	mdi_vhcache_pathinfo_t *cpi;
7860 
7861 	rw_enter(&vhcache->vhcache_lock, RW_READER);
7862 	if ((cct = lookup_vhcache_client(vhcache, ct_name, ct_addr, tok))
7863 	    == NULL) {
7864 		rw_exit(&vhcache->vhcache_lock);
7865 		return;
7866 	}
7867 
7868 	/*
7869 	 * to avoid unnecessary on-disk cache updates, first check if an
7870 	 * update is really needed. If no update is needed simply return.
7871 	 */
7872 	for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi->cpi_next) {
7873 		if ((cpi->cpi_pip != NULL &&
7874 		    (cpi->cpi_flags & MDI_CPI_HINT_PATH_DOES_NOT_EXIST)) ||
7875 		    (cpi->cpi_pip == NULL &&
7876 		    !(cpi->cpi_flags & MDI_CPI_HINT_PATH_DOES_NOT_EXIST))) {
7877 			break;
7878 		}
7879 	}
7880 	if (cpi == NULL) {
7881 		rw_exit(&vhcache->vhcache_lock);
7882 		return;
7883 	}
7884 
7885 	if (rw_tryupgrade(&vhcache->vhcache_lock) == 0) {
7886 		rw_exit(&vhcache->vhcache_lock);
7887 		rw_enter(&vhcache->vhcache_lock, RW_WRITER);
7888 		if ((cct = lookup_vhcache_client(vhcache, ct_name, ct_addr,
7889 		    tok)) == NULL) {
7890 			rw_exit(&vhcache->vhcache_lock);
7891 			return;
7892 		}
7893 	}
7894 
7895 	for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi->cpi_next) {
7896 		if (cpi->cpi_pip != NULL)
7897 			cpi->cpi_flags &= ~MDI_CPI_HINT_PATH_DOES_NOT_EXIST;
7898 		else
7899 			cpi->cpi_flags |= MDI_CPI_HINT_PATH_DOES_NOT_EXIST;
7900 	}
7901 	sort_vhcache_paths(cct);
7902 
7903 	rw_exit(&vhcache->vhcache_lock);
7904 	vhcache_dirty(vhc);
7905 }
7906 
7907 /*
7908  * Configure all specified paths of the client.
7909  */
7910 static void
7911 config_client_paths_sync(mdi_vhci_config_t *vhc, char *ct_name, char *ct_addr,
7912     mdi_phys_path_t *pp_head, mdi_vhcache_lookup_token_t *tok)
7913 {
7914 	mdi_phys_path_t *pp;
7915 
7916 	for (pp = pp_head; pp != NULL; pp = pp->phys_path_next)
7917 		(void) bus_config_one_phci_child(pp->phys_path);
7918 	adjust_sort_vhcache_paths(vhc, ct_name, ct_addr, tok);
7919 }
7920 
7921 /*
7922  * Dequeue elements from vhci async client config list and bus configure
7923  * their corresponding phci clients.
7924  */
7925 static void
7926 config_client_paths_thread(void *arg)
7927 {
7928 	mdi_vhci_config_t *vhc = (mdi_vhci_config_t *)arg;
7929 	mdi_async_client_config_t *acc;
7930 	clock_t quit_at_ticks;
7931 	clock_t idle_time = mdi_async_config_idle_time * TICKS_PER_SECOND;
7932 	callb_cpr_t cprinfo;
7933 
7934 	CALLB_CPR_INIT(&cprinfo, &vhc->vhc_lock, callb_generic_cpr,
7935 	    "mdi_config_client_paths");
7936 
7937 	for (; ; ) {
7938 		quit_at_ticks = ddi_get_lbolt() + idle_time;
7939 
7940 		mutex_enter(&vhc->vhc_lock);
7941 		while (!(vhc->vhc_flags & MDI_VHC_EXIT) &&
7942 		    vhc->vhc_acc_list_head == NULL &&
7943 		    ddi_get_lbolt() < quit_at_ticks) {
7944 			CALLB_CPR_SAFE_BEGIN(&cprinfo);
7945 			(void) cv_timedwait(&vhc->vhc_cv, &vhc->vhc_lock,
7946 			    quit_at_ticks);
7947 			CALLB_CPR_SAFE_END(&cprinfo, &vhc->vhc_lock);
7948 		}
7949 
7950 		if ((vhc->vhc_flags & MDI_VHC_EXIT) ||
7951 		    vhc->vhc_acc_list_head == NULL)
7952 			goto out;
7953 
7954 		acc = vhc->vhc_acc_list_head;
7955 		vhc->vhc_acc_list_head = acc->acc_next;
7956 		if (vhc->vhc_acc_list_head == NULL)
7957 			vhc->vhc_acc_list_tail = NULL;
7958 		vhc->vhc_acc_count--;
7959 		mutex_exit(&vhc->vhc_lock);
7960 
7961 		config_client_paths_sync(vhc, acc->acc_ct_name,
7962 		    acc->acc_ct_addr, acc->acc_phclient_path_list_head,
7963 		    &acc->acc_token);
7964 
7965 		free_async_client_config(acc);
7966 	}
7967 
7968 out:
7969 	vhc->vhc_acc_thrcount--;
7970 	/* CALLB_CPR_EXIT releases the vhc->vhc_lock */
7971 	CALLB_CPR_EXIT(&cprinfo);
7972 }
7973 
7974 /*
7975  * Arrange for all the phci client paths (pp_head) for the specified client
7976  * to be bus configured asynchronously by a thread.
7977  */
7978 static void
7979 config_client_paths_async(mdi_vhci_config_t *vhc, char *ct_name, char *ct_addr,
7980     mdi_phys_path_t *pp_head, mdi_vhcache_lookup_token_t *tok)
7981 {
7982 	mdi_async_client_config_t *acc, *newacc;
7983 	int create_thread;
7984 
7985 	if (pp_head == NULL)
7986 		return;
7987 
7988 	if (mdi_mtc_off) {
7989 		config_client_paths_sync(vhc, ct_name, ct_addr, pp_head, tok);
7990 		free_phclient_path_list(pp_head);
7991 		return;
7992 	}
7993 
7994 	newacc = alloc_async_client_config(ct_name, ct_addr, pp_head, tok);
7995 	ASSERT(newacc);
7996 
7997 	mutex_enter(&vhc->vhc_lock);
7998 	for (acc = vhc->vhc_acc_list_head; acc != NULL; acc = acc->acc_next) {
7999 		if (strcmp(ct_name, acc->acc_ct_name) == 0 &&
8000 		    strcmp(ct_addr, acc->acc_ct_addr) == 0) {
8001 			free_async_client_config(newacc);
8002 			mutex_exit(&vhc->vhc_lock);
8003 			return;
8004 		}
8005 	}
8006 
8007 	if (vhc->vhc_acc_list_head == NULL)
8008 		vhc->vhc_acc_list_head = newacc;
8009 	else
8010 		vhc->vhc_acc_list_tail->acc_next = newacc;
8011 	vhc->vhc_acc_list_tail = newacc;
8012 	vhc->vhc_acc_count++;
8013 	if (vhc->vhc_acc_count <= vhc->vhc_acc_thrcount) {
8014 		cv_broadcast(&vhc->vhc_cv);
8015 		create_thread = 0;
8016 	} else {
8017 		vhc->vhc_acc_thrcount++;
8018 		create_thread = 1;
8019 	}
8020 	mutex_exit(&vhc->vhc_lock);
8021 
8022 	if (create_thread)
8023 		(void) thread_create(NULL, 0, config_client_paths_thread, vhc,
8024 		    0, &p0, TS_RUN, minclsyspri);
8025 }
8026 
8027 /*
8028  * Return number of online paths for the specified client.
8029  */
8030 static int
8031 nonline_paths(mdi_vhcache_client_t *cct)
8032 {
8033 	mdi_vhcache_pathinfo_t *cpi;
8034 	int online_count = 0;
8035 
8036 	for (cpi = cct->cct_cpi_head; cpi != NULL; cpi = cpi->cpi_next) {
8037 		if (cpi->cpi_pip != NULL) {
8038 			MDI_PI_LOCK(cpi->cpi_pip);
8039 			if (cpi->cpi_pip->pi_state == MDI_PATHINFO_STATE_ONLINE)
8040 				online_count++;
8041 			MDI_PI_UNLOCK(cpi->cpi_pip);
8042 		}
8043 	}
8044 
8045 	return (online_count);
8046 }
8047 
8048 /*
8049  * Bus configure all paths for the specified vhci client.
8050  * If at least one path for the client is already online, the remaining paths
8051  * will be configured asynchronously. Otherwise, it synchronously configures
8052  * the paths until at least one path is online and then rest of the paths
8053  * will be configured asynchronously.
8054  */
8055 static void
8056 config_client_paths(mdi_vhci_config_t *vhc, char *ct_name, char *ct_addr)
8057 {
8058 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
8059 	mdi_phys_path_t *pp_head, *pp;
8060 	mdi_vhcache_client_t *cct;
8061 	mdi_vhcache_lookup_token_t tok;
8062 
8063 	ASSERT(RW_LOCK_HELD(&vhcache->vhcache_lock));
8064 
8065 	init_vhcache_lookup_token(&tok, NULL);
8066 
8067 	if (ct_name == NULL || ct_addr == NULL ||
8068 	    (cct = lookup_vhcache_client(vhcache, ct_name, ct_addr, &tok))
8069 	    == NULL ||
8070 	    (pp_head = build_phclient_path_list(cct, ct_name)) == NULL) {
8071 		rw_exit(&vhcache->vhcache_lock);
8072 		return;
8073 	}
8074 
8075 	/* if at least one path is online, configure the rest asynchronously */
8076 	if (nonline_paths(cct) > 0) {
8077 		rw_exit(&vhcache->vhcache_lock);
8078 		config_client_paths_async(vhc, ct_name, ct_addr, pp_head, &tok);
8079 		return;
8080 	}
8081 
8082 	rw_exit(&vhcache->vhcache_lock);
8083 
8084 	for (pp = pp_head; pp != NULL; pp = pp->phys_path_next) {
8085 		if (bus_config_one_phci_child(pp->phys_path) == MDI_SUCCESS) {
8086 			rw_enter(&vhcache->vhcache_lock, RW_READER);
8087 
8088 			if ((cct = lookup_vhcache_client(vhcache, ct_name,
8089 			    ct_addr, &tok)) == NULL) {
8090 				rw_exit(&vhcache->vhcache_lock);
8091 				goto out;
8092 			}
8093 
8094 			if (nonline_paths(cct) > 0 &&
8095 			    pp->phys_path_next != NULL) {
8096 				rw_exit(&vhcache->vhcache_lock);
8097 				config_client_paths_async(vhc, ct_name, ct_addr,
8098 				    pp->phys_path_next, &tok);
8099 				pp->phys_path_next = NULL;
8100 				goto out;
8101 			}
8102 
8103 			rw_exit(&vhcache->vhcache_lock);
8104 		}
8105 	}
8106 
8107 	adjust_sort_vhcache_paths(vhc, ct_name, ct_addr, &tok);
8108 out:
8109 	free_phclient_path_list(pp_head);
8110 }
8111 
8112 static void
8113 single_threaded_vhconfig_enter(mdi_vhci_config_t *vhc)
8114 {
8115 	mutex_enter(&vhc->vhc_lock);
8116 	while (vhc->vhc_flags & MDI_VHC_SINGLE_THREADED)
8117 		cv_wait(&vhc->vhc_cv, &vhc->vhc_lock);
8118 	vhc->vhc_flags |= MDI_VHC_SINGLE_THREADED;
8119 	mutex_exit(&vhc->vhc_lock);
8120 }
8121 
8122 static void
8123 single_threaded_vhconfig_exit(mdi_vhci_config_t *vhc)
8124 {
8125 	mutex_enter(&vhc->vhc_lock);
8126 	vhc->vhc_flags &= ~MDI_VHC_SINGLE_THREADED;
8127 	cv_broadcast(&vhc->vhc_cv);
8128 	mutex_exit(&vhc->vhc_lock);
8129 }
8130 
8131 typedef struct mdi_phci_driver_info {
8132 	char	*phdriver_name;	/* name of the phci driver */
8133 
8134 	/* set to non zero if the phci driver supports root device */
8135 	int	phdriver_root_support;
8136 } mdi_phci_driver_info_t;
8137 
8138 /*
8139  * vhci class and root support capability of a phci driver can be
8140  * specified using ddi-vhci-class and ddi-no-root-support properties in the
8141  * phci driver.conf file. The built-in tables below contain this information
8142  * for those phci drivers whose driver.conf files don't yet contain this info.
8143  *
8144  * All phci drivers expect iscsi have root device support.
8145  */
8146 static mdi_phci_driver_info_t scsi_phci_driver_list[] = {
8147 	{ "fp", 1 },
8148 	{ "iscsi", 0 },
8149 	{ "ibsrp", 1 }
8150 	};
8151 
8152 static mdi_phci_driver_info_t ib_phci_driver_list[] = { "tavor", 1 };
8153 
8154 static void *
8155 mdi_realloc(void *old_ptr, size_t old_size, size_t new_size)
8156 {
8157 	void *new_ptr;
8158 
8159 	new_ptr = kmem_zalloc(new_size, KM_SLEEP);
8160 	if (old_ptr) {
8161 		bcopy(old_ptr, new_ptr, old_size);
8162 		kmem_free(old_ptr, old_size);
8163 	}
8164 	return (new_ptr);
8165 }
8166 
8167 static void
8168 add_to_phci_list(char ***driver_list, int **root_support_list,
8169     int *cur_elements, int *max_elements, char *driver_name, int root_support)
8170 {
8171 	ASSERT(*cur_elements <= *max_elements);
8172 	if (*cur_elements == *max_elements) {
8173 		*max_elements += 10;
8174 		*driver_list = mdi_realloc(*driver_list,
8175 		    sizeof (char *) * (*cur_elements),
8176 		    sizeof (char *) * (*max_elements));
8177 		*root_support_list = mdi_realloc(*root_support_list,
8178 		    sizeof (int) * (*cur_elements),
8179 		    sizeof (int) * (*max_elements));
8180 	}
8181 	(*driver_list)[*cur_elements] = i_ddi_strdup(driver_name, KM_SLEEP);
8182 	(*root_support_list)[*cur_elements] = root_support;
8183 	(*cur_elements)++;
8184 }
8185 
8186 static void
8187 get_phci_driver_list(char *vhci_class, char ***driver_list,
8188     int **root_support_list, int *cur_elements, int *max_elements)
8189 {
8190 	mdi_phci_driver_info_t	*st_driver_list, *p;
8191 	int		st_ndrivers, root_support, i, j, driver_conf_count;
8192 	major_t		m;
8193 	struct devnames	*dnp;
8194 	ddi_prop_t	*propp;
8195 
8196 	*driver_list = NULL;
8197 	*root_support_list = NULL;
8198 	*cur_elements = 0;
8199 	*max_elements = 0;
8200 
8201 	/* add the phci drivers derived from the phci driver.conf files */
8202 	for (m = 0; m < devcnt; m++) {
8203 		dnp = &devnamesp[m];
8204 
8205 		if (dnp->dn_flags & DN_PHCI_DRIVER) {
8206 			LOCK_DEV_OPS(&dnp->dn_lock);
8207 			if (dnp->dn_global_prop_ptr != NULL &&
8208 			    (propp = i_ddi_prop_search(DDI_DEV_T_ANY,
8209 			    DDI_VHCI_CLASS, DDI_PROP_TYPE_STRING,
8210 			    &dnp->dn_global_prop_ptr->prop_list)) != NULL &&
8211 			    strcmp(propp->prop_val, vhci_class) == 0) {
8212 
8213 				root_support = (i_ddi_prop_search(DDI_DEV_T_ANY,
8214 				    DDI_NO_ROOT_SUPPORT, DDI_PROP_TYPE_INT,
8215 				    &dnp->dn_global_prop_ptr->prop_list)
8216 				    == NULL) ? 1 : 0;
8217 
8218 				add_to_phci_list(driver_list, root_support_list,
8219 				    cur_elements, max_elements, dnp->dn_name,
8220 				    root_support);
8221 
8222 				UNLOCK_DEV_OPS(&dnp->dn_lock);
8223 			} else
8224 				UNLOCK_DEV_OPS(&dnp->dn_lock);
8225 		}
8226 	}
8227 
8228 	driver_conf_count = *cur_elements;
8229 
8230 	/* add the phci drivers specified in the built-in tables */
8231 	if (strcmp(vhci_class, MDI_HCI_CLASS_SCSI) == 0) {
8232 		st_driver_list = scsi_phci_driver_list;
8233 		st_ndrivers = sizeof (scsi_phci_driver_list) /
8234 		    sizeof (mdi_phci_driver_info_t);
8235 	} else if (strcmp(vhci_class, MDI_HCI_CLASS_IB) == 0) {
8236 		st_driver_list = ib_phci_driver_list;
8237 		st_ndrivers = sizeof (ib_phci_driver_list) /
8238 		    sizeof (mdi_phci_driver_info_t);
8239 	} else {
8240 		st_driver_list = NULL;
8241 		st_ndrivers = 0;
8242 	}
8243 
8244 	for (i = 0, p = st_driver_list; i < st_ndrivers; i++, p++) {
8245 		/* add this phci driver if not already added before */
8246 		for (j = 0; j < driver_conf_count; j++) {
8247 			if (strcmp((*driver_list)[j], p->phdriver_name) == 0)
8248 				break;
8249 		}
8250 		if (j == driver_conf_count) {
8251 			add_to_phci_list(driver_list, root_support_list,
8252 			    cur_elements, max_elements, p->phdriver_name,
8253 			    p->phdriver_root_support);
8254 		}
8255 	}
8256 }
8257 
8258 /*
8259  * Attach the phci driver instances associated with the specified vhci class.
8260  * If root is mounted attach all phci driver instances.
8261  * If root is not mounted, attach the instances of only those phci
8262  * drivers that have the root support.
8263  */
8264 static void
8265 attach_phci_drivers(char *vhci_class)
8266 {
8267 	char	**driver_list, **p;
8268 	int	*root_support_list;
8269 	int	cur_elements, max_elements, i;
8270 	major_t	m;
8271 
8272 	get_phci_driver_list(vhci_class, &driver_list, &root_support_list,
8273 	    &cur_elements, &max_elements);
8274 
8275 	for (i = 0; i < cur_elements; i++) {
8276 		if (modrootloaded || root_support_list[i]) {
8277 			m = ddi_name_to_major(driver_list[i]);
8278 			if (m != (major_t)-1 && ddi_hold_installed_driver(m))
8279 				ddi_rele_driver(m);
8280 		}
8281 	}
8282 
8283 	if (driver_list) {
8284 		for (i = 0, p = driver_list; i < cur_elements; i++, p++)
8285 			kmem_free(*p, strlen(*p) + 1);
8286 		kmem_free(driver_list, sizeof (char *) * max_elements);
8287 		kmem_free(root_support_list, sizeof (int) * max_elements);
8288 	}
8289 }
8290 
8291 /*
8292  * Build vhci cache:
8293  *
8294  * Attach phci driver instances and then drive BUS_CONFIG_ALL on
8295  * the phci driver instances. During this process the cache gets built.
8296  *
8297  * Cache is built fully if the root is mounted.
8298  * If the root is not mounted, phci drivers that do not have root support
8299  * are not attached. As a result the cache is built partially. The entries
8300  * in the cache reflect only those phci drivers that have root support.
8301  */
8302 static int
8303 build_vhci_cache(mdi_vhci_t *vh)
8304 {
8305 	mdi_vhci_config_t *vhc = vh->vh_config;
8306 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
8307 
8308 	single_threaded_vhconfig_enter(vhc);
8309 
8310 	rw_enter(&vhcache->vhcache_lock, RW_READER);
8311 	if (vhcache->vhcache_flags & MDI_VHCI_CACHE_SETUP_DONE) {
8312 		rw_exit(&vhcache->vhcache_lock);
8313 		single_threaded_vhconfig_exit(vhc);
8314 		return (0);
8315 	}
8316 	rw_exit(&vhcache->vhcache_lock);
8317 
8318 	attach_phci_drivers(vh->vh_class);
8319 	bus_config_all_phcis(vhcache, NDI_DRV_CONF_REPROBE | NDI_NO_EVENT,
8320 	    BUS_CONFIG_ALL, (major_t)-1);
8321 
8322 	rw_enter(&vhcache->vhcache_lock, RW_WRITER);
8323 	vhcache->vhcache_flags |= MDI_VHCI_CACHE_SETUP_DONE;
8324 	rw_exit(&vhcache->vhcache_lock);
8325 
8326 	single_threaded_vhconfig_exit(vhc);
8327 	vhcache_dirty(vhc);
8328 	return (1);
8329 }
8330 
8331 /*
8332  * Determine if discovery of paths is needed.
8333  */
8334 static int
8335 vhcache_do_discovery(mdi_vhci_config_t *vhc)
8336 {
8337 	int rv = 1;
8338 
8339 	mutex_enter(&vhc->vhc_lock);
8340 	if (i_ddi_io_initialized() == 0) {
8341 		if (vhc->vhc_path_discovery_boot > 0) {
8342 			vhc->vhc_path_discovery_boot--;
8343 			goto out;
8344 		}
8345 	} else {
8346 		if (vhc->vhc_path_discovery_postboot > 0) {
8347 			vhc->vhc_path_discovery_postboot--;
8348 			goto out;
8349 		}
8350 	}
8351 
8352 	/*
8353 	 * Do full path discovery at most once per mdi_path_discovery_interval.
8354 	 * This is to avoid a series of full path discoveries when opening
8355 	 * stale /dev/[r]dsk links.
8356 	 */
8357 	if (mdi_path_discovery_interval != -1 &&
8358 	    lbolt64 >= vhc->vhc_path_discovery_cutoff_time)
8359 		goto out;
8360 
8361 	rv = 0;
8362 out:
8363 	mutex_exit(&vhc->vhc_lock);
8364 	return (rv);
8365 }
8366 
8367 /*
8368  * Discover all paths:
8369  *
8370  * Attach phci driver instances and then drive BUS_CONFIG_ALL on all the phci
8371  * driver instances. During this process all paths will be discovered.
8372  */
8373 static int
8374 vhcache_discover_paths(mdi_vhci_t *vh)
8375 {
8376 	mdi_vhci_config_t *vhc = vh->vh_config;
8377 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
8378 	int rv = 0;
8379 
8380 	single_threaded_vhconfig_enter(vhc);
8381 
8382 	if (vhcache_do_discovery(vhc)) {
8383 		attach_phci_drivers(vh->vh_class);
8384 		bus_config_all_phcis(vhcache, NDI_DRV_CONF_REPROBE |
8385 		    NDI_NO_EVENT, BUS_CONFIG_ALL, (major_t)-1);
8386 
8387 		mutex_enter(&vhc->vhc_lock);
8388 		vhc->vhc_path_discovery_cutoff_time = lbolt64 +
8389 		    mdi_path_discovery_interval * TICKS_PER_SECOND;
8390 		mutex_exit(&vhc->vhc_lock);
8391 		rv = 1;
8392 	}
8393 
8394 	single_threaded_vhconfig_exit(vhc);
8395 	return (rv);
8396 }
8397 
8398 /*
8399  * Generic vhci bus config implementation:
8400  *
8401  * Parameters
8402  *	vdip	vhci dip
8403  *	flags	bus config flags
8404  *	op	bus config operation
8405  *	The remaining parameters are bus config operation specific
8406  *
8407  * for BUS_CONFIG_ONE
8408  *	arg	pointer to name@addr
8409  *	child	upon successful return from this function, *child will be
8410  *		set to the configured and held devinfo child node of vdip.
8411  *	ct_addr	pointer to client address (i.e. GUID)
8412  *
8413  * for BUS_CONFIG_DRIVER
8414  *	arg	major number of the driver
8415  *	child and ct_addr parameters are ignored
8416  *
8417  * for BUS_CONFIG_ALL
8418  *	arg, child, and ct_addr parameters are ignored
8419  *
8420  * Note that for the rest of the bus config operations, this function simply
8421  * calls the framework provided default bus config routine.
8422  */
8423 int
8424 mdi_vhci_bus_config(dev_info_t *vdip, uint_t flags, ddi_bus_config_op_t op,
8425     void *arg, dev_info_t **child, char *ct_addr)
8426 {
8427 	mdi_vhci_t *vh = i_devi_get_vhci(vdip);
8428 	mdi_vhci_config_t *vhc = vh->vh_config;
8429 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
8430 	int rv = 0;
8431 	int params_valid = 0;
8432 	char *cp;
8433 
8434 	/*
8435 	 * To bus config vhcis we relay operation, possibly using another
8436 	 * thread, to phcis. The phci driver then interacts with MDI to cause
8437 	 * vhci child nodes to be enumerated under the vhci node.  Adding a
8438 	 * vhci child requires an ndi_devi_enter of the vhci. Since another
8439 	 * thread may be adding the child, to avoid deadlock we can't wait
8440 	 * for the relayed operations to complete if we have already entered
8441 	 * the vhci node.
8442 	 */
8443 	if (DEVI_BUSY_OWNED(vdip)) {
8444 		MDI_DEBUG(2, (CE_NOTE, vdip, "!MDI: vhci bus config: "
8445 		    "vhci dip is busy owned %p\n", (void *)vdip));
8446 		goto default_bus_config;
8447 	}
8448 
8449 	rw_enter(&vhcache->vhcache_lock, RW_READER);
8450 	if (!(vhcache->vhcache_flags & MDI_VHCI_CACHE_SETUP_DONE)) {
8451 		rw_exit(&vhcache->vhcache_lock);
8452 		rv = build_vhci_cache(vh);
8453 		rw_enter(&vhcache->vhcache_lock, RW_READER);
8454 	}
8455 
8456 	switch (op) {
8457 	case BUS_CONFIG_ONE:
8458 		if (arg != NULL && ct_addr != NULL) {
8459 			/* extract node name */
8460 			cp = (char *)arg;
8461 			while (*cp != '\0' && *cp != '@')
8462 				cp++;
8463 			if (*cp == '@') {
8464 				params_valid = 1;
8465 				*cp = '\0';
8466 				config_client_paths(vhc, (char *)arg, ct_addr);
8467 				/* config_client_paths() releases cache_lock */
8468 				*cp = '@';
8469 				break;
8470 			}
8471 		}
8472 
8473 		rw_exit(&vhcache->vhcache_lock);
8474 		break;
8475 
8476 	case BUS_CONFIG_DRIVER:
8477 		rw_exit(&vhcache->vhcache_lock);
8478 		if (rv == 0)
8479 			st_bus_config_all_phcis(vhc, flags, op,
8480 			    (major_t)(uintptr_t)arg);
8481 		break;
8482 
8483 	case BUS_CONFIG_ALL:
8484 		rw_exit(&vhcache->vhcache_lock);
8485 		if (rv == 0)
8486 			st_bus_config_all_phcis(vhc, flags, op, -1);
8487 		break;
8488 
8489 	default:
8490 		rw_exit(&vhcache->vhcache_lock);
8491 		break;
8492 	}
8493 
8494 
8495 default_bus_config:
8496 	/*
8497 	 * All requested child nodes are enumerated under the vhci.
8498 	 * Now configure them.
8499 	 */
8500 	if (ndi_busop_bus_config(vdip, flags, op, arg, child, 0) ==
8501 	    NDI_SUCCESS) {
8502 		return (MDI_SUCCESS);
8503 	} else if (op == BUS_CONFIG_ONE && rv == 0 && params_valid) {
8504 		/* discover all paths and try configuring again */
8505 		if (vhcache_discover_paths(vh) &&
8506 		    ndi_busop_bus_config(vdip, flags, op, arg, child, 0) ==
8507 		    NDI_SUCCESS)
8508 			return (MDI_SUCCESS);
8509 	}
8510 
8511 	return (MDI_FAILURE);
8512 }
8513 
8514 /*
8515  * Read the on-disk vhci cache into an nvlist for the specified vhci class.
8516  */
8517 static nvlist_t *
8518 read_on_disk_vhci_cache(char *vhci_class)
8519 {
8520 	nvlist_t *nvl;
8521 	int err;
8522 	char *filename;
8523 
8524 	filename = vhclass2vhcache_filename(vhci_class);
8525 
8526 	if ((err = fread_nvlist(filename, &nvl)) == 0) {
8527 		kmem_free(filename, strlen(filename) + 1);
8528 		return (nvl);
8529 	} else if (err == EIO)
8530 		cmn_err(CE_WARN, "%s: I/O error, will recreate\n", filename);
8531 	else if (err == EINVAL)
8532 		cmn_err(CE_WARN,
8533 		    "%s: data file corrupted, will recreate\n", filename);
8534 
8535 	kmem_free(filename, strlen(filename) + 1);
8536 	return (NULL);
8537 }
8538 
8539 /*
8540  * Read on-disk vhci cache into nvlists for all vhci classes.
8541  * Called during booting by i_ddi_read_devices_files().
8542  */
8543 void
8544 mdi_read_devices_files(void)
8545 {
8546 	int i;
8547 
8548 	for (i = 0; i < N_VHCI_CLASSES; i++)
8549 		vhcache_nvl[i] = read_on_disk_vhci_cache(vhci_class_list[i]);
8550 }
8551 
8552 /*
8553  * Remove all stale entries from vhci cache.
8554  */
8555 static void
8556 clean_vhcache(mdi_vhci_config_t *vhc)
8557 {
8558 	mdi_vhci_cache_t *vhcache = &vhc->vhc_vhcache;
8559 	mdi_vhcache_phci_t *cphci, *cphci_head, *cphci_next;
8560 	mdi_vhcache_client_t *cct, *cct_head, *cct_next;
8561 	mdi_vhcache_pathinfo_t *cpi, *cpi_head, *cpi_next;
8562 
8563 	rw_enter(&vhcache->vhcache_lock, RW_WRITER);
8564 
8565 	cct_head = vhcache->vhcache_client_head;
8566 	vhcache->vhcache_client_head = vhcache->vhcache_client_tail = NULL;
8567 	for (cct = cct_head; cct != NULL; cct = cct_next) {
8568 		cct_next = cct->cct_next;
8569 
8570 		cpi_head = cct->cct_cpi_head;
8571 		cct->cct_cpi_head = cct->cct_cpi_tail = NULL;
8572 		for (cpi = cpi_head; cpi != NULL; cpi = cpi_next) {
8573 			cpi_next = cpi->cpi_next;
8574 			if (cpi->cpi_pip != NULL) {
8575 				ASSERT(cpi->cpi_cphci->cphci_phci != NULL);
8576 				enqueue_tail_vhcache_pathinfo(cct, cpi);
8577 			} else
8578 				free_vhcache_pathinfo(cpi);
8579 		}
8580 
8581 		if (cct->cct_cpi_head != NULL)
8582 			enqueue_vhcache_client(vhcache, cct);
8583 		else {
8584 			(void) mod_hash_destroy(vhcache->vhcache_client_hash,
8585 			    (mod_hash_key_t)cct->cct_name_addr);
8586 			free_vhcache_client(cct);
8587 		}
8588 	}
8589 
8590 	cphci_head = vhcache->vhcache_phci_head;
8591 	vhcache->vhcache_phci_head = vhcache->vhcache_phci_tail = NULL;
8592 	for (cphci = cphci_head; cphci != NULL; cphci = cphci_next) {
8593 		cphci_next = cphci->cphci_next;
8594 		if (cphci->cphci_phci != NULL)
8595 			enqueue_vhcache_phci(vhcache, cphci);
8596 		else
8597 			free_vhcache_phci(cphci);
8598 	}
8599 
8600 	vhcache->vhcache_clean_time = lbolt64;
8601 	rw_exit(&vhcache->vhcache_lock);
8602 	vhcache_dirty(vhc);
8603 }
8604 
8605 /*
8606  * Remove all stale entries from vhci cache.
8607  * Called by i_ddi_clean_devices_files() during the execution of devfsadm -C
8608  */
8609 void
8610 mdi_clean_vhcache(void)
8611 {
8612 	mdi_vhci_t *vh;
8613 
8614 	mutex_enter(&mdi_mutex);
8615 	for (vh = mdi_vhci_head; vh != NULL; vh = vh->vh_next) {
8616 		vh->vh_refcnt++;
8617 		mutex_exit(&mdi_mutex);
8618 		clean_vhcache(vh->vh_config);
8619 		mutex_enter(&mdi_mutex);
8620 		vh->vh_refcnt--;
8621 	}
8622 	mutex_exit(&mdi_mutex);
8623 }
8624 
8625 /*
8626  * mdi_vhci_walk_clients():
8627  *		Walker routine to traverse client dev_info nodes
8628  * ddi_walk_devs(ddi_get_child(vdip), f, arg) returns the entire tree
8629  * below the client, including nexus devices, which we dont want.
8630  * So we just traverse the immediate siblings, starting from 1st client.
8631  */
8632 void
8633 mdi_vhci_walk_clients(dev_info_t *vdip,
8634     int (*f)(dev_info_t *, void *), void *arg)
8635 {
8636 	mdi_vhci_t	*vh = i_devi_get_vhci(vdip);
8637 	dev_info_t	*cdip;
8638 	mdi_client_t	*ct;
8639 
8640 	MDI_VHCI_CLIENT_LOCK(vh);
8641 	cdip = ddi_get_child(vdip);
8642 	while (cdip) {
8643 		ct = i_devi_get_client(cdip);
8644 		MDI_CLIENT_LOCK(ct);
8645 
8646 		if (((*f)(cdip, arg)) == DDI_WALK_CONTINUE)
8647 			cdip = ddi_get_next_sibling(cdip);
8648 		else
8649 			cdip = NULL;
8650 
8651 		MDI_CLIENT_UNLOCK(ct);
8652 	}
8653 	MDI_VHCI_CLIENT_UNLOCK(vh);
8654 }
8655 
8656 /*
8657  * mdi_vhci_walk_phcis():
8658  *		Walker routine to traverse phci dev_info nodes
8659  */
8660 void
8661 mdi_vhci_walk_phcis(dev_info_t *vdip,
8662     int (*f)(dev_info_t *, void *), void *arg)
8663 {
8664 	mdi_vhci_t	*vh = i_devi_get_vhci(vdip);
8665 	mdi_phci_t	*ph, *next;
8666 
8667 	MDI_VHCI_PHCI_LOCK(vh);
8668 	ph = vh->vh_phci_head;
8669 	while (ph) {
8670 		MDI_PHCI_LOCK(ph);
8671 
8672 		if (((*f)(ph->ph_dip, arg)) == DDI_WALK_CONTINUE)
8673 			next = ph->ph_next;
8674 		else
8675 			next = NULL;
8676 
8677 		MDI_PHCI_UNLOCK(ph);
8678 		ph = next;
8679 	}
8680 	MDI_VHCI_PHCI_UNLOCK(vh);
8681 }
8682 
8683 
8684 /*
8685  * mdi_walk_vhcis():
8686  *		Walker routine to traverse vhci dev_info nodes
8687  */
8688 void
8689 mdi_walk_vhcis(int (*f)(dev_info_t *, void *), void *arg)
8690 {
8691 	mdi_vhci_t	*vh = NULL;
8692 
8693 	mutex_enter(&mdi_mutex);
8694 	/*
8695 	 * Scan for already registered vhci
8696 	 */
8697 	for (vh = mdi_vhci_head; vh != NULL; vh = vh->vh_next) {
8698 		vh->vh_refcnt++;
8699 		mutex_exit(&mdi_mutex);
8700 		if (((*f)(vh->vh_dip, arg)) != DDI_WALK_CONTINUE) {
8701 			mutex_enter(&mdi_mutex);
8702 			vh->vh_refcnt--;
8703 			break;
8704 		} else {
8705 			mutex_enter(&mdi_mutex);
8706 			vh->vh_refcnt--;
8707 		}
8708 	}
8709 
8710 	mutex_exit(&mdi_mutex);
8711 }
8712 
8713 /*
8714  * i_mdi_log_sysevent():
8715  *		Logs events for pickup by syseventd
8716  */
8717 static void
8718 i_mdi_log_sysevent(dev_info_t *dip, char *ph_vh_class, char *subclass)
8719 {
8720 	char		*path_name;
8721 	nvlist_t	*attr_list;
8722 
8723 	if (nvlist_alloc(&attr_list, NV_UNIQUE_NAME_TYPE,
8724 	    KM_SLEEP) != DDI_SUCCESS) {
8725 		goto alloc_failed;
8726 	}
8727 
8728 	path_name = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
8729 	(void) ddi_pathname(dip, path_name);
8730 
8731 	if (nvlist_add_string(attr_list, DDI_DRIVER_NAME,
8732 	    ddi_driver_name(dip)) != DDI_SUCCESS) {
8733 		goto error;
8734 	}
8735 
8736 	if (nvlist_add_int32(attr_list, DDI_DRIVER_MAJOR,
8737 	    (int32_t)ddi_driver_major(dip)) != DDI_SUCCESS) {
8738 		goto error;
8739 	}
8740 
8741 	if (nvlist_add_int32(attr_list, DDI_INSTANCE,
8742 	    (int32_t)ddi_get_instance(dip)) != DDI_SUCCESS) {
8743 		goto error;
8744 	}
8745 
8746 	if (nvlist_add_string(attr_list, DDI_PATHNAME,
8747 	    path_name) != DDI_SUCCESS) {
8748 		goto error;
8749 	}
8750 
8751 	if (nvlist_add_string(attr_list, DDI_CLASS,
8752 	    ph_vh_class) != DDI_SUCCESS) {
8753 		goto error;
8754 	}
8755 
8756 	(void) ddi_log_sysevent(dip, DDI_VENDOR_SUNW, EC_DDI, subclass,
8757 	    attr_list, NULL, DDI_SLEEP);
8758 
8759 error:
8760 	kmem_free(path_name, MAXPATHLEN);
8761 	nvlist_free(attr_list);
8762 	return;
8763 
8764 alloc_failed:
8765 	MDI_DEBUG(1, (CE_WARN, dip,
8766 	    "!i_mdi_log_sysevent: Unable to send sysevent"));
8767 }
8768