xref: /titanic_52/usr/src/uts/sun4u/serengeti/io/sbdp_quiesce.c (revision 004388ebfdfe2ed7dfd2d153a876dfcc22d2c006)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * A CPR derivative specifically for sbd
31  */
32 
33 #include <sys/types.h>
34 #include <sys/systm.h>
35 #include <sys/machparam.h>
36 #include <sys/machsystm.h>
37 #include <sys/ddi.h>
38 #define	SUNDDI_IMPL
39 #include <sys/sunddi.h>
40 #include <sys/sunndi.h>
41 #include <sys/devctl.h>
42 #include <sys/time.h>
43 #include <sys/kmem.h>
44 #include <nfs/lm.h>
45 #include <sys/ddi_impldefs.h>
46 #include <sys/ndi_impldefs.h>
47 #include <sys/obpdefs.h>
48 #include <sys/cmn_err.h>
49 #include <sys/debug.h>
50 #include <sys/errno.h>
51 #include <sys/callb.h>
52 #include <sys/clock.h>
53 #include <sys/x_call.h>
54 #include <sys/cpuvar.h>
55 #include <sys/epm.h>
56 #include <sys/vfs.h>
57 
58 #ifdef DEBUG
59 #include <sys/note.h>
60 #endif
61 
62 #include <sys/promif.h>
63 #include <sys/conf.h>
64 #include <sys/cyclic.h>
65 
66 #include <sys/sbd_ioctl.h>
67 #include <sys/sbd.h>
68 #include <sys/sbdp_priv.h>
69 #include <sys/cpu_sgnblk_defs.h>
70 
71 static char *
72 sbdp_get_err_buf(sbd_error_t *ep)
73 {
74 	return (ep->e_rsc);
75 }
76 
77 extern void	e_ddi_enter_driver_list(struct devnames *dnp, int *listcnt);
78 extern void	e_ddi_exit_driver_list(struct devnames *dnp, int listcnt);
79 extern int	is_pseudo_device(dev_info_t *dip);
80 
81 extern kmutex_t	cpu_lock;
82 
83 static int	sbdp_is_real_device(dev_info_t *dip);
84 #ifdef DEBUG
85 static int	sbdp_bypass_device(char *dname);
86 #endif
87 static int	sbdp_check_dip(dev_info_t *dip, void *arg, uint_t ref);
88 
89 static int	sbdp_resolve_devname(dev_info_t *dip, char *buffer,
90 				char *alias);
91 
92 int sbdp_test_suspend(sbdp_handle_t *hp);
93 
94 #define	SR_STATE(srh)			((srh)->sr_suspend_state)
95 #define	SR_SET_STATE(srh, state)	(SR_STATE((srh)) = (state))
96 #define	SR_FAILED_DIP(srh)		((srh)->sr_failed_dip)
97 
98 #define	SR_FLAG_WATCHDOG	0x1
99 #define	SR_CHECK_FLAG(srh, flag)	((srh)->sr_flags & (flag))
100 #define	SR_SET_FLAG(srh, flag)		((srh)->sr_flags |= (flag))
101 #define	SR_CLEAR_FLAG(srh, flag)	((srh)->sr_flags &= ~(flag))
102 
103 #ifdef DEBUG
104 /*
105  * Just for testing. List of drivers to bypass when performing a suspend.
106  */
107 static char *sbdp_bypass_list[] = {
108 	/* "sgsbbc", this is an example when needed */
109 	""
110 };
111 #endif
112 
113 #define		SKIP_SYNC	/* bypass sync ops in sbdp_suspend */
114 
115 /*
116  * sbdp_skip_user_threads is used to control if user threads should
117  * be suspended.  If sbdp_skip_user_threads is true, the rest of the
118  * flags are not used; if it is false, sbdp_check_user_stop_result
119  * will be used to control whether or not we need to check suspend
120  * result, and sbdp_allow_blocked_threads will be used to control
121  * whether or not we allow suspend to continue if there are blocked
122  * threads.  We allow all combinations of sbdp_check_user_stop_result
123  * and sbdp_allow_block_threads, even though it might not make much
124  * sense to not allow block threads when we don't even check stop
125  * result.
126  */
127 static int	sbdp_skip_user_threads = 0;		/* default to FALSE */
128 static int	sbdp_check_user_stop_result = 1;	/* default to TRUE */
129 static int	sbdp_allow_blocked_threads = 1;		/* default to TRUE */
130 
131 
132 static void
133 sbdp_stop_intr(void)
134 {
135 	kpreempt_disable();
136 	cyclic_suspend();
137 }
138 
139 static void
140 sbdp_enable_intr(void)
141 {
142 	cyclic_resume();
143 	kpreempt_enable();
144 }
145 
146 sbdp_sr_handle_t *
147 sbdp_get_sr_handle(void)
148 {
149 	sbdp_sr_handle_t *srh;
150 	srh = kmem_zalloc(sizeof (sbdp_sr_handle_t), KM_SLEEP);
151 
152 	return (srh);
153 }
154 
155 void
156 sbdp_release_sr_handle(sbdp_sr_handle_t *srh)
157 {
158 	ASSERT(SR_FAILED_DIP(srh) == NULL);
159 	kmem_free((caddr_t)srh, sizeof (sbdp_sr_handle_t));
160 }
161 
162 static int
163 sbdp_is_real_device(dev_info_t *dip)
164 {
165 	struct regspec *regbuf = NULL;
166 	int length = 0;
167 	int rc;
168 
169 	if (ddi_get_driver(dip) == NULL)
170 		return (0);
171 
172 	if (DEVI(dip)->devi_pm_flags & (PMC_NEEDS_SR|PMC_PARENTAL_SR))
173 		return (1);
174 	if (DEVI(dip)->devi_pm_flags & PMC_NO_SR)
175 		return (0);
176 
177 	/*
178 	 * now the general case
179 	 */
180 	rc = ddi_getlongprop(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, "reg",
181 		(caddr_t)&regbuf, &length);
182 	ASSERT(rc != DDI_PROP_NO_MEMORY);
183 	if (rc != DDI_PROP_SUCCESS) {
184 		return (0);
185 	} else {
186 		if ((length > 0) && (regbuf != NULL))
187 			kmem_free(regbuf, length);
188 		return (1);
189 	}
190 }
191 
192 #ifdef DEBUG
193 static int
194 sbdp_bypass_device(char *dname)
195 {
196 	int i;
197 	char **lname;
198 	/* check the bypass list */
199 	for (i = 0, lname = &sbdp_bypass_list[i]; **lname != '\0'; lname++) {
200 		SBDP_DBG_QR("Checking %s\n", *lname);
201 		if (strcmp(dname, sbdp_bypass_list[i++]) == 0)
202 			return (1);
203 	}
204 	return (0);
205 }
206 #endif
207 
208 static int
209 sbdp_resolve_devname(dev_info_t *dip, char *buffer, char *alias)
210 {
211 	major_t	devmajor;
212 	char	*aka, *name;
213 
214 	*buffer = *alias = 0;
215 
216 	if (dip == NULL)
217 		return (-1);
218 
219 	if ((name = ddi_get_name(dip)) == NULL)
220 		name = "<null name>";
221 
222 	aka = name;
223 
224 	if ((devmajor = ddi_name_to_major(aka)) != -1)
225 		aka = ddi_major_to_name(devmajor);
226 
227 	(void) strcpy(buffer, name);
228 
229 	if (strcmp(name, aka))
230 		(void) strcpy(alias, aka);
231 	else
232 		*alias = 0;
233 
234 	return (0);
235 }
236 
237 typedef struct sbdp_ref {
238 	int *refcount;
239 	sbd_error_t *sep;
240 } sbdp_ref_t;
241 
242 static int
243 sbdp_check_dip(dev_info_t *dip, void *arg, uint_t ref)
244 {
245 	char		*dname;
246 	sbdp_ref_t	*sbrp = (sbdp_ref_t *)arg;
247 
248 	if (dip == NULL)
249 		return (DDI_WALK_CONTINUE);
250 
251 	ASSERT(sbrp->sep != NULL);
252 	ASSERT(sbrp->refcount != NULL);
253 
254 	if (!sbdp_is_real_device(dip))
255 		return (DDI_WALK_CONTINUE);
256 
257 	dname = ddi_binding_name(dip);
258 
259 	if ((strcmp(dname, "pciclass,060940") == 0) || (strcmp(dname,
260 	    "pciclass,060980") == 0)) {
261 		(void) ddi_pathname(dip, sbdp_get_err_buf(sbrp->sep));
262 		sbdp_set_err(sbrp->sep, ESBD_BUSY, NULL);
263 		(*sbrp->refcount)++;
264 		return (DDI_WALK_TERMINATE);
265 	}
266 
267 #ifdef DEBUG
268 	if (sbdp_bypass_device(dname))
269 		return (DDI_WALK_CONTINUE);
270 #endif
271 
272 	if (ref) {
273 		(*sbrp->refcount)++;
274 		SBDP_DBG_QR("\n%s (major# %d) is referenced\n",
275 			dname, ddi_name_to_major(dname));
276 		(void) ddi_pathname(dip, sbdp_get_err_buf(sbrp->sep));
277 		sbdp_set_err(sbrp->sep, ESBD_BUSY, NULL);
278 		return (DDI_WALK_TERMINATE);
279 	}
280 	return (DDI_WALK_CONTINUE);
281 }
282 
283 void
284 sbdp_check_devices(dev_info_t *dip, int *refcount, sbd_error_t *sep)
285 {
286 	sbdp_ref_t sbr;
287 
288 	sbr.refcount = refcount;
289 	sbr.sep = sep;
290 
291 	ASSERT(e_ddi_branch_held(dip));
292 
293 	(void) e_ddi_branch_referenced(dip, sbdp_check_dip, &sbr);
294 }
295 
296 static int
297 sbdp_suspend_devices(dev_info_t *dip, sbdp_sr_handle_t *srh)
298 {
299 	int circ;
300 	major_t	major;
301 	char	*dname;
302 
303 	for (; dip != NULL; dip = ddi_get_next_sibling(dip)) {
304 		char	d_name[40], d_alias[40], *d_info;
305 
306 		ndi_devi_enter(dip, &circ);
307 		if (sbdp_suspend_devices(ddi_get_child(dip), srh)) {
308 			ndi_devi_exit(dip, circ);
309 			return (ENXIO);
310 		}
311 		ndi_devi_exit(dip, circ);
312 
313 		if (!sbdp_is_real_device(dip))
314 				continue;
315 
316 		major = (major_t)-1;
317 		if ((dname = DEVI(dip)->devi_binding_name) != NULL)
318 			major = ddi_name_to_major(dname);
319 
320 #ifdef DEBUG
321 		if (sbdp_bypass_device(dname)) {
322 			SBDP_DBG_QR("bypassed suspend of %s (major# %d)\n",
323 			    dname, major);
324 			continue;
325 		}
326 #endif
327 
328 		if ((d_info = ddi_get_name_addr(dip)) == NULL)
329 			d_info = "<null>";
330 
331 		d_name[0] = 0;
332 		if (sbdp_resolve_devname(dip, d_name, d_alias) == 0) {
333 			if (d_alias[0] != 0) {
334 				SBDP_DBG_QR("\tsuspending %s@%s (aka %s)\n",
335 					d_name, d_info, d_alias);
336 			} else {
337 				SBDP_DBG_QR("\tsuspending %s@%s\n",
338 					d_name, d_info);
339 			}
340 		} else {
341 			SBDP_DBG_QR("\tsuspending %s@%s\n", dname, d_info);
342 		}
343 
344 		if (devi_detach(dip, DDI_SUSPEND) != DDI_SUCCESS) {
345 			(void) sprintf(sbdp_get_err_buf(&srh->sep),
346 			    "%d", major);
347 
348 			sbdp_set_err(&srh->sep, ESGT_SUSPEND, NULL);
349 			ndi_hold_devi(dip);
350 			SR_FAILED_DIP(srh) = dip;
351 			return (DDI_FAILURE);
352 		}
353 	}
354 
355 	return (DDI_SUCCESS);
356 }
357 
358 static void
359 sbdp_resume_devices(dev_info_t *start, sbdp_sr_handle_t *srh)
360 {
361 	int circ;
362 	dev_info_t	*dip, *next, *last = NULL;
363 	char		*bn;
364 	sbd_error_t	*sep;
365 
366 	sep = &srh->sep;
367 
368 	/* attach in reverse device tree order */
369 	while (last != start) {
370 		dip = start;
371 		next = ddi_get_next_sibling(dip);
372 		while (next != last && dip != SR_FAILED_DIP(srh)) {
373 			dip = next;
374 			next = ddi_get_next_sibling(dip);
375 		}
376 		if (dip == SR_FAILED_DIP(srh)) {
377 			/* Release hold acquired in sbdp_suspend_devices() */
378 			ndi_rele_devi(dip);
379 			SR_FAILED_DIP(srh) = NULL;
380 		} else if (sbdp_is_real_device(dip) &&
381 				SR_FAILED_DIP(srh) == NULL) {
382 
383 			if (DEVI(dip)->devi_binding_name != NULL) {
384 				bn = ddi_binding_name(dip);
385 			}
386 #ifdef DEBUG
387 			if (!sbdp_bypass_device(bn)) {
388 #else
389 			{
390 #endif
391 				char	d_name[40], d_alias[40], *d_info;
392 
393 				d_name[0] = 0;
394 				d_info = ddi_get_name_addr(dip);
395 				if (d_info == NULL)
396 					d_info = "<null>";
397 
398 				if (!sbdp_resolve_devname(dip, d_name,
399 								d_alias)) {
400 					if (d_alias[0] != 0) {
401 						SBDP_DBG_QR("\tresuming "
402 							"%s@%s (aka %s)\n",
403 							d_name, d_info,
404 							d_alias);
405 					} else {
406 						SBDP_DBG_QR("\tresuming "
407 							"%s@%s\n",
408 							d_name, d_info);
409 					}
410 				} else {
411 					SBDP_DBG_QR("\tresuming %s@%s\n",
412 						bn, d_info);
413 				}
414 
415 				if (devi_attach(dip, DDI_RESUME) !=
416 							DDI_SUCCESS) {
417 					/*
418 					 * Print a console warning,
419 					 * set an errno of ESGT_RESUME,
420 					 * and save the driver major
421 					 * number in the e_str.
422 					 */
423 
424 					(void) sprintf(sbdp_get_err_buf(sep),
425 					    "%s@%s",
426 					    d_name[0] ? d_name : bn, d_info);
427 					SBDP_DBG_QR("\tFAILED to resume "
428 						"%s\n", sbdp_get_err_buf(sep));
429 					sbdp_set_err(sep,
430 					    ESGT_RESUME, NULL);
431 				}
432 			}
433 		}
434 		ndi_devi_enter(dip, &circ);
435 		sbdp_resume_devices(ddi_get_child(dip), srh);
436 		ndi_devi_exit(dip, circ);
437 		last = dip;
438 	}
439 }
440 
441 /*
442  * True if thread is virtually stopped.  Similar to CPR_VSTOPPED
443  * but from DR point of view.  These user threads are waiting in
444  * the kernel.  Once they return from kernel, they will process
445  * the stop signal and stop.
446  */
447 #define	SBDP_VSTOPPED(t)			\
448 	((t)->t_state == TS_SLEEP &&		\
449 	(t)->t_wchan != NULL &&			\
450 	(t)->t_astflag &&		\
451 	((t)->t_proc_flag & TP_CHKPT))
452 
453 
454 static int
455 sbdp_stop_user_threads(sbdp_sr_handle_t *srh)
456 {
457 	int		count;
458 	char		cache_psargs[PSARGSZ];
459 	kthread_id_t	cache_tp;
460 	uint_t		cache_t_state;
461 	int		bailout;
462 	sbd_error_t	*sep;
463 	kthread_id_t 	tp;
464 
465 	extern void add_one_utstop();
466 	extern void utstop_timedwait(clock_t);
467 	extern void utstop_init(void);
468 
469 #define	SBDP_UTSTOP_RETRY	4
470 #define	SBDP_UTSTOP_WAIT	hz
471 
472 	if (sbdp_skip_user_threads)
473 		return (DDI_SUCCESS);
474 
475 	sep = &srh->sep;
476 	ASSERT(sep);
477 
478 	utstop_init();
479 
480 	/* we need to try a few times to get past fork, etc. */
481 	for (count = 0; count < SBDP_UTSTOP_RETRY; count++) {
482 		/* walk the entire threadlist */
483 		mutex_enter(&pidlock);
484 		for (tp = curthread->t_next; tp != curthread; tp = tp->t_next) {
485 			proc_t *p = ttoproc(tp);
486 
487 			/* handle kernel threads separately */
488 			if (p->p_as == &kas || p->p_stat == SZOMB)
489 				continue;
490 
491 			mutex_enter(&p->p_lock);
492 			thread_lock(tp);
493 
494 			if (tp->t_state == TS_STOPPED) {
495 				/* add another reason to stop this thread */
496 				tp->t_schedflag &= ~TS_RESUME;
497 			} else {
498 				tp->t_proc_flag |= TP_CHKPT;
499 
500 				thread_unlock(tp);
501 				mutex_exit(&p->p_lock);
502 				add_one_utstop();
503 				mutex_enter(&p->p_lock);
504 				thread_lock(tp);
505 
506 				aston(tp);
507 
508 				if (tp->t_state == TS_SLEEP &&
509 				    (tp->t_flag & T_WAKEABLE)) {
510 					setrun_locked(tp);
511 				}
512 
513 			}
514 
515 			/* grab thread if needed */
516 			if (tp->t_state == TS_ONPROC && tp->t_cpu != CPU)
517 				poke_cpu(tp->t_cpu->cpu_id);
518 
519 
520 			thread_unlock(tp);
521 			mutex_exit(&p->p_lock);
522 		}
523 		mutex_exit(&pidlock);
524 
525 
526 		/* let everything catch up */
527 		utstop_timedwait(count * count * SBDP_UTSTOP_WAIT);
528 
529 
530 		/* now, walk the threadlist again to see if we are done */
531 		mutex_enter(&pidlock);
532 		for (tp = curthread->t_next, bailout = 0;
533 			tp != curthread; tp = tp->t_next) {
534 			proc_t *p = ttoproc(tp);
535 
536 			/* handle kernel threads separately */
537 			if (p->p_as == &kas || p->p_stat == SZOMB)
538 				continue;
539 
540 			/*
541 			 * If this thread didn't stop, and we don't allow
542 			 * unstopped blocked threads, bail.
543 			 */
544 			thread_lock(tp);
545 			if (!CPR_ISTOPPED(tp) &&
546 			    !(sbdp_allow_blocked_threads &&
547 			    SBDP_VSTOPPED(tp))) {
548 
549 				/* nope, cache the details for later */
550 				bcopy(p->p_user.u_psargs, cache_psargs,
551 					sizeof (cache_psargs));
552 				cache_tp = tp;
553 				cache_t_state = tp->t_state;
554 				bailout = 1;
555 			}
556 			thread_unlock(tp);
557 		}
558 		mutex_exit(&pidlock);
559 
560 		/* were all the threads stopped? */
561 		if (!bailout)
562 			break;
563 	}
564 
565 	/* were we unable to stop all threads after a few tries? */
566 	if (bailout) {
567 		cmn_err(CE_NOTE, "process: %s id: %p state: %x\n",
568 			cache_psargs, cache_tp, cache_t_state);
569 
570 		(void) sprintf(sbdp_get_err_buf(sep), "%s", cache_psargs);
571 		sbdp_set_err(sep, ESGT_UTHREAD, NULL);
572 		return (ESRCH);
573 	}
574 
575 	return (DDI_SUCCESS);
576 }
577 
578 static void
579 sbdp_start_user_threads(void)
580 {
581 	kthread_id_t tp;
582 
583 	mutex_enter(&pidlock);
584 
585 	/* walk all threads and release them */
586 	for (tp = curthread->t_next; tp != curthread; tp = tp->t_next) {
587 		proc_t *p = ttoproc(tp);
588 
589 		/* skip kernel threads */
590 		if (ttoproc(tp)->p_as == &kas)
591 			continue;
592 
593 		mutex_enter(&p->p_lock);
594 		tp->t_proc_flag &= ~TP_CHKPT;
595 		mutex_exit(&p->p_lock);
596 
597 		thread_lock(tp);
598 		if (CPR_ISTOPPED(tp)) {
599 			/* back on the runq */
600 			tp->t_schedflag |= TS_RESUME;
601 			setrun_locked(tp);
602 		}
603 		thread_unlock(tp);
604 	}
605 
606 	mutex_exit(&pidlock);
607 }
608 
609 static void
610 sbdp_signal_user(int sig)
611 {
612 	struct proc *p;
613 
614 	mutex_enter(&pidlock);
615 
616 	for (p = practive; p != NULL; p = p->p_next) {
617 		/* only user threads */
618 		if (p->p_exec == NULL || p->p_stat == SZOMB ||
619 		    p == proc_init || p == ttoproc(curthread))
620 			continue;
621 
622 		mutex_enter(&p->p_lock);
623 		sigtoproc(p, NULL, sig);
624 		mutex_exit(&p->p_lock);
625 	}
626 
627 	mutex_exit(&pidlock);
628 
629 	/* add a bit of delay */
630 	delay(hz);
631 }
632 
633 static uint_t saved_watchdog_seconds;
634 
635 void
636 sbdp_resume(sbdp_sr_handle_t *srh)
637 {
638 	/*
639 	 * update the signature block
640 	 */
641 	CPU_SIGNATURE(OS_SIG, SIGST_RESUME_INPROGRESS, SIGSUBST_NULL,
642 	    CPU->cpu_id);
643 
644 	switch (SR_STATE(srh)) {
645 	case SBDP_SRSTATE_FULL:
646 
647 		ASSERT(MUTEX_HELD(&cpu_lock));
648 
649 		sbdp_enable_intr(); 	/* enable intr & clock */
650 
651 		/*
652 		 * release all the other cpus
653 		 * using start_cpus() vice sbdp_release_cpus()
654 		 */
655 		start_cpus();
656 		mutex_exit(&cpu_lock);
657 
658 		/*
659 		 * If we suspended hw watchdog at suspend,
660 		 * re-enable it now.
661 		 */
662 
663 		if (SR_CHECK_FLAG(srh, SR_FLAG_WATCHDOG)) {
664 			mutex_enter(&tod_lock);
665 			tod_ops.tod_set_watchdog_timer(
666 				saved_watchdog_seconds);
667 			mutex_exit(&tod_lock);
668 		}
669 
670 		/* FALLTHROUGH */
671 
672 	case SBDP_SRSTATE_DRIVER:
673 		/*
674 		 * resume devices: root node doesn't have to
675 		 * be held in any way.
676 		 */
677 		sbdp_resume_devices(ddi_root_node(), srh);
678 
679 		/*
680 		 * resume the lock manager
681 		 */
682 		lm_cprresume();
683 
684 		/* FALLTHROUGH */
685 
686 	case SBDP_SRSTATE_USER:
687 		/*
688 		 * finally, resume user threads
689 		 */
690 		if (!sbdp_skip_user_threads) {
691 			SBDP_DBG_QR("DR: resuming user threads...\n");
692 			sbdp_start_user_threads();
693 		}
694 		/* FALLTHROUGH */
695 
696 	case SBDP_SRSTATE_BEGIN:
697 	default:
698 		/*
699 		 * let those who care know that we've just resumed
700 		 */
701 		SBDP_DBG_QR("sending SIGTHAW...\n");
702 		sbdp_signal_user(SIGTHAW);
703 		break;
704 	}
705 
706 	i_ndi_allow_device_tree_changes(srh->sh_ndi);
707 
708 	/*
709 	 * update the signature block
710 	 */
711 	CPU_SIGNATURE(OS_SIG, SIGST_RUN, SIGSUBST_NULL, CPU->cpu_id);
712 
713 	SBDP_DBG_QR("DR: resume COMPLETED\n");
714 }
715 
716 int
717 sbdp_suspend(sbdp_sr_handle_t *srh)
718 {
719 	int force;
720 	int rc = DDI_SUCCESS;
721 
722 	force = (srh && (srh->sr_flags & SBDP_IOCTL_FLAG_FORCE));
723 
724 	/*
725 	 * if no force flag, check for unsafe drivers
726 	 */
727 	if (force) {
728 		SBDP_DBG_QR("\nsbdp_suspend invoked with force flag");
729 	}
730 
731 	/*
732 	 * update the signature block
733 	 */
734 	CPU_SIGNATURE(OS_SIG, SIGST_QUIESCE_INPROGRESS, SIGSUBST_NULL,
735 	    CPU->cpu_id);
736 
737 	i_ndi_block_device_tree_changes(&srh->sh_ndi);
738 
739 	/*
740 	 * first, stop all user threads
741 	 */
742 	SBDP_DBG_QR("SBDP: suspending user threads...\n");
743 	SR_SET_STATE(srh, SBDP_SRSTATE_USER);
744 	if (((rc = sbdp_stop_user_threads(srh)) != DDI_SUCCESS) &&
745 	    sbdp_check_user_stop_result) {
746 		sbdp_resume(srh);
747 		return (rc);
748 	}
749 
750 #ifndef	SKIP_SYNC
751 	/*
752 	 * This sync swap out all user pages
753 	 */
754 	vfs_sync(SYNC_ALL);
755 #endif
756 
757 	/*
758 	 * special treatment for lock manager
759 	 */
760 	lm_cprsuspend();
761 
762 #ifndef	SKIP_SYNC
763 	/*
764 	 * sync the file system in case we never make it back
765 	 */
766 	sync();
767 
768 #endif
769 	/*
770 	 * now suspend drivers
771 	 */
772 	SBDP_DBG_QR("SBDP: suspending drivers...\n");
773 	SR_SET_STATE(srh, SBDP_SRSTATE_DRIVER);
774 
775 	/*
776 	 * Root node doesn't have to be held in any way.
777 	 */
778 	if ((rc = sbdp_suspend_devices(ddi_root_node(), srh)) != DDI_SUCCESS) {
779 		sbdp_resume(srh);
780 		return (rc);
781 	}
782 
783 	/*
784 	 * finally, grab all cpus
785 	 */
786 	SR_SET_STATE(srh, SBDP_SRSTATE_FULL);
787 
788 	/*
789 	 * if watchdog was activated, disable it
790 	 */
791 	if (watchdog_activated) {
792 		mutex_enter(&tod_lock);
793 		saved_watchdog_seconds = tod_ops.tod_clear_watchdog_timer();
794 		mutex_exit(&tod_lock);
795 		SR_SET_FLAG(srh, SR_FLAG_WATCHDOG);
796 	} else {
797 		SR_CLEAR_FLAG(srh, SR_FLAG_WATCHDOG);
798 	}
799 
800 	mutex_enter(&cpu_lock);
801 	pause_cpus(NULL);
802 	sbdp_stop_intr();
803 
804 	/*
805 	 * update the signature block
806 	 */
807 	CPU_SIGNATURE(OS_SIG, SIGST_QUIESCED, SIGSUBST_NULL, CPU->cpu_id);
808 
809 	return (rc);
810 }
811 
812 /*ARGSUSED*/
813 int
814 sbdp_test_suspend(sbdp_handle_t *hp)
815 {
816 	sbdp_sr_handle_t	*srh;
817 	int			err;
818 
819 	SBDP_DBG_QR("%s...\n", "sbdp_test_suspend");
820 
821 	srh = sbdp_get_sr_handle();
822 
823 	srh->sr_flags = hp->h_flags;
824 
825 	if ((err = sbdp_suspend(srh)) == DDI_SUCCESS) {
826 		sbdp_resume(srh);
827 	} else {
828 		SBDP_DBG_MISC("sbdp_suspend() failed, err = 0x%x\n", err);
829 	}
830 	sbdp_release_sr_handle(srh);
831 
832 	return (0);
833 }
834 
835 #ifdef	DEBUG
836 int
837 sbdp_passthru_test_quiesce(sbdp_handle_t *hp, void *arg)
838 {
839 	_NOTE(ARGUNUSED(arg))
840 
841 	return (sbdp_test_suspend(hp));
842 }
843 #endif
844