xref: /titanic_50/usr/src/uts/common/disp/ts.c (revision 7c4f0c867215d12824a34c4005ebd6ba576f0a85)
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 (c) 1994, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright 2019 Joyent, Inc.
25  */
26 
27 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
28 /*	  All Rights Reserved	*/
29 
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/sysmacros.h>
33 #include <sys/cred.h>
34 #include <sys/proc.h>
35 #include <sys/session.h>
36 #include <sys/strsubr.h>
37 #include <sys/signal.h>
38 #include <sys/user.h>
39 #include <sys/priocntl.h>
40 #include <sys/class.h>
41 #include <sys/disp.h>
42 #include <sys/procset.h>
43 #include <sys/debug.h>
44 #include <sys/ts.h>
45 #include <sys/tspriocntl.h>
46 #include <sys/iapriocntl.h>
47 #include <sys/kmem.h>
48 #include <sys/errno.h>
49 #include <sys/cpuvar.h>
50 #include <sys/systm.h>		/* for lbolt */
51 #include <sys/vtrace.h>
52 #include <sys/vmsystm.h>
53 #include <sys/schedctl.h>
54 #include <sys/tnf_probe.h>
55 #include <sys/atomic.h>
56 #include <sys/policy.h>
57 #include <sys/sdt.h>
58 #include <sys/cpupart.h>
59 #include <vm/rm.h>
60 #include <vm/seg_kmem.h>
61 #include <sys/modctl.h>
62 #include <sys/cpucaps.h>
63 
64 static pri_t ts_init(id_t, int, classfuncs_t **);
65 
66 static struct sclass csw = {
67 	"TS",
68 	ts_init,
69 	0
70 };
71 
72 static struct modlsched modlsched = {
73 	&mod_schedops, "time sharing sched class", &csw
74 };
75 
76 static struct modlinkage modlinkage = {
77 	MODREV_1, (void *)&modlsched, NULL
78 };
79 
80 int
81 _init()
82 {
83 	return (mod_install(&modlinkage));
84 }
85 
86 int
87 _fini()
88 {
89 	return (EBUSY);		/* don't remove TS for now */
90 }
91 
92 int
93 _info(struct modinfo *modinfop)
94 {
95 	return (mod_info(&modlinkage, modinfop));
96 }
97 
98 /*
99  * Class specific code for the time-sharing class
100  */
101 
102 
103 /*
104  * Extern declarations for variables defined in the ts master file
105  */
106 #define	TSMAXUPRI 60
107 
108 pri_t	ts_maxupri = TSMAXUPRI;	/* max time-sharing user priority */
109 pri_t	ts_maxumdpri;		/* maximum user mode ts priority */
110 
111 pri_t	ia_maxupri = IAMAXUPRI;	/* max interactive user priority */
112 pri_t	ia_boost = IA_BOOST;	/* boost value for interactive */
113 
114 tsdpent_t  *ts_dptbl;	/* time-sharing disp parameter table */
115 pri_t	*ts_kmdpris;	/* array of global pris used by ts procs when */
116 			/*  sleeping or running in kernel after sleep */
117 
118 static id_t ia_cid;
119 
120 int ts_sleep_promote = 1;
121 
122 #define	tsmedumdpri	(ts_maxumdpri >> 1)
123 
124 #define	TS_NEWUMDPRI(tspp) \
125 { \
126 	pri_t pri; \
127 	pri = (tspp)->ts_cpupri + (tspp)->ts_upri + (tspp)->ts_boost; \
128 	if (pri > ts_maxumdpri) \
129 		(tspp)->ts_umdpri = ts_maxumdpri; \
130 	else if (pri < 0) \
131 		(tspp)->ts_umdpri = 0; \
132 	else \
133 		(tspp)->ts_umdpri = pri; \
134 	ASSERT((tspp)->ts_umdpri >= 0 && (tspp)->ts_umdpri <= ts_maxumdpri); \
135 }
136 
137 /*
138  * The tsproc_t structures are kept in an array of circular doubly linked
139  * lists.  A hash on the thread pointer is used to determine which list
140  * each thread should be placed.  Each list has a dummy "head" which is
141  * never removed, so the list is never empty.  ts_update traverses these
142  * lists to update the priorities of threads that have been waiting on
143  * the run queue.
144  */
145 
146 #define	TS_LISTS 16		/* number of lists, must be power of 2 */
147 
148 /* hash function, argument is a thread pointer */
149 #define	TS_LIST_HASH(tp)	(((uintptr_t)(tp) >> 9) & (TS_LISTS - 1))
150 
151 /* iterate to the next list */
152 #define	TS_LIST_NEXT(i)		(((i) + 1) & (TS_LISTS - 1))
153 
154 /*
155  * Insert thread into the appropriate tsproc list.
156  */
157 #define	TS_LIST_INSERT(tspp)				\
158 {							\
159 	int index = TS_LIST_HASH(tspp->ts_tp);		\
160 	kmutex_t *lockp = &ts_list_lock[index];		\
161 	tsproc_t *headp = &ts_plisthead[index];		\
162 	mutex_enter(lockp);				\
163 	tspp->ts_next = headp->ts_next;			\
164 	tspp->ts_prev = headp;				\
165 	headp->ts_next->ts_prev = tspp;			\
166 	headp->ts_next = tspp;				\
167 	mutex_exit(lockp);				\
168 }
169 
170 /*
171  * Remove thread from tsproc list.
172  */
173 #define	TS_LIST_DELETE(tspp)				\
174 {							\
175 	int index = TS_LIST_HASH(tspp->ts_tp);		\
176 	kmutex_t *lockp = &ts_list_lock[index];		\
177 	mutex_enter(lockp);				\
178 	tspp->ts_prev->ts_next = tspp->ts_next;		\
179 	tspp->ts_next->ts_prev = tspp->ts_prev;		\
180 	mutex_exit(lockp);				\
181 }
182 
183 
184 static int	ts_admin(caddr_t, cred_t *);
185 static int	ts_enterclass(kthread_t *, id_t, void *, cred_t *, void *);
186 static int	ts_fork(kthread_t *, kthread_t *, void *);
187 static int	ts_getclinfo(void *);
188 static int	ts_getclpri(pcpri_t *);
189 static int	ts_parmsin(void *);
190 static int	ts_parmsout(void *, pc_vaparms_t *);
191 static int	ts_vaparmsin(void *, pc_vaparms_t *);
192 static int	ts_vaparmsout(void *, pc_vaparms_t *);
193 static int	ts_parmsset(kthread_t *, void *, id_t, cred_t *);
194 static void	ts_exit(kthread_t *);
195 static int	ts_donice(kthread_t *, cred_t *, int, int *);
196 static int	ts_doprio(kthread_t *, cred_t *, int, int *);
197 static void	ts_exitclass(void *);
198 static int	ts_canexit(kthread_t *, cred_t *);
199 static void	ts_forkret(kthread_t *, kthread_t *);
200 static void	ts_nullsys();
201 static void	ts_parmsget(kthread_t *, void *);
202 static void	ts_preempt(kthread_t *);
203 static void	ts_setrun(kthread_t *);
204 static void	ts_sleep(kthread_t *);
205 static pri_t	ts_swapin(kthread_t *, int);
206 static pri_t	ts_swapout(kthread_t *, int);
207 static void	ts_tick(kthread_t *);
208 static void	ts_trapret(kthread_t *);
209 static void	ts_update(void *);
210 static int	ts_update_list(int);
211 static void	ts_wakeup(kthread_t *);
212 static pri_t	ts_globpri(kthread_t *);
213 static void	ts_yield(kthread_t *);
214 extern tsdpent_t *ts_getdptbl(void);
215 extern pri_t	*ts_getkmdpris(void);
216 extern pri_t	td_getmaxumdpri(void);
217 static int	ts_alloc(void **, int);
218 static void	ts_free(void *);
219 
220 pri_t		ia_init(id_t, int, classfuncs_t **);
221 static int	ia_getclinfo(void *);
222 static int	ia_getclpri(pcpri_t *);
223 static int	ia_parmsin(void *);
224 static int	ia_vaparmsin(void *, pc_vaparms_t *);
225 static int	ia_vaparmsout(void *, pc_vaparms_t *);
226 static int	ia_parmsset(kthread_t *, void *, id_t, cred_t *);
227 static void	ia_parmsget(kthread_t *, void *);
228 static void	ia_set_process_group(pid_t, pid_t, pid_t);
229 
230 static void	ts_change_priority(kthread_t *, tsproc_t *);
231 
232 static pri_t	ts_maxglobpri;	/* maximum global priority used by ts class */
233 static kmutex_t	ts_dptblock;	/* protects time sharing dispatch table */
234 static kmutex_t	ts_list_lock[TS_LISTS];	/* protects tsproc lists */
235 static tsproc_t	ts_plisthead[TS_LISTS];	/* dummy tsproc at head of lists */
236 
237 static gid_t	IA_gid = 0;
238 
239 static struct classfuncs ts_classfuncs = {
240 	/* class functions */
241 	ts_admin,
242 	ts_getclinfo,
243 	ts_parmsin,
244 	ts_parmsout,
245 	ts_vaparmsin,
246 	ts_vaparmsout,
247 	ts_getclpri,
248 	ts_alloc,
249 	ts_free,
250 
251 	/* thread functions */
252 	ts_enterclass,
253 	ts_exitclass,
254 	ts_canexit,
255 	ts_fork,
256 	ts_forkret,
257 	ts_parmsget,
258 	ts_parmsset,
259 	ts_nullsys,	/* stop */
260 	ts_exit,
261 	ts_nullsys,	/* active */
262 	ts_nullsys,	/* inactive */
263 	ts_swapin,
264 	ts_swapout,
265 	ts_trapret,
266 	ts_preempt,
267 	ts_setrun,
268 	ts_sleep,
269 	ts_tick,
270 	ts_wakeup,
271 	ts_donice,
272 	ts_globpri,
273 	ts_nullsys,	/* set_process_group */
274 	ts_yield,
275 	ts_doprio,
276 };
277 
278 /*
279  * ia_classfuncs is used for interactive class threads; IA threads are stored
280  * on the same class list as TS threads, and most of the class functions are
281  * identical, but a few have different enough functionality to require their
282  * own functions.
283  */
284 static struct classfuncs ia_classfuncs = {
285 	/* class functions */
286 	ts_admin,
287 	ia_getclinfo,
288 	ia_parmsin,
289 	ts_parmsout,
290 	ia_vaparmsin,
291 	ia_vaparmsout,
292 	ia_getclpri,
293 	ts_alloc,
294 	ts_free,
295 
296 	/* thread functions */
297 	ts_enterclass,
298 	ts_exitclass,
299 	ts_canexit,
300 	ts_fork,
301 	ts_forkret,
302 	ia_parmsget,
303 	ia_parmsset,
304 	ts_nullsys,	/* stop */
305 	ts_exit,
306 	ts_nullsys,	/* active */
307 	ts_nullsys,	/* inactive */
308 	ts_swapin,
309 	ts_swapout,
310 	ts_trapret,
311 	ts_preempt,
312 	ts_setrun,
313 	ts_sleep,
314 	ts_tick,
315 	ts_wakeup,
316 	ts_donice,
317 	ts_globpri,
318 	ia_set_process_group,
319 	ts_yield,
320 	ts_doprio,
321 };
322 
323 
324 /*
325  * Time sharing class initialization.  Called by dispinit() at boot time.
326  * We can ignore the clparmsz argument since we know that the smallest
327  * possible parameter buffer is big enough for us.
328  */
329 /* ARGSUSED */
330 static pri_t
331 ts_init(id_t cid, int clparmsz, classfuncs_t **clfuncspp)
332 {
333 	int i;
334 	extern pri_t ts_getmaxumdpri(void);
335 
336 	ts_dptbl = ts_getdptbl();
337 	ts_kmdpris = ts_getkmdpris();
338 	ts_maxumdpri = ts_getmaxumdpri();
339 	ts_maxglobpri = MAX(ts_kmdpris[0], ts_dptbl[ts_maxumdpri].ts_globpri);
340 
341 	/*
342 	 * Initialize the tsproc lists.
343 	 */
344 	for (i = 0; i < TS_LISTS; i++) {
345 		ts_plisthead[i].ts_next = ts_plisthead[i].ts_prev =
346 		    &ts_plisthead[i];
347 	}
348 
349 	/*
350 	 * We're required to return a pointer to our classfuncs
351 	 * structure and the highest global priority value we use.
352 	 */
353 	*clfuncspp = &ts_classfuncs;
354 	return (ts_maxglobpri);
355 }
356 
357 
358 /*
359  * Interactive class scheduler initialization
360  */
361 /* ARGSUSED */
362 pri_t
363 ia_init(id_t cid, int clparmsz, classfuncs_t **clfuncspp)
364 {
365 	/*
366 	 * We're required to return a pointer to our classfuncs
367 	 * structure and the highest global priority value we use.
368 	 */
369 	ia_cid = cid;
370 	*clfuncspp = &ia_classfuncs;
371 	return (ts_maxglobpri);
372 }
373 
374 
375 /*
376  * Get or reset the ts_dptbl values per the user's request.
377  */
378 static int
379 ts_admin(caddr_t uaddr, cred_t *reqpcredp)
380 {
381 	tsadmin_t	tsadmin;
382 	tsdpent_t	*tmpdpp;
383 	int		userdpsz;
384 	int		i;
385 	size_t		tsdpsz;
386 
387 	if (get_udatamodel() == DATAMODEL_NATIVE) {
388 		if (copyin(uaddr, &tsadmin, sizeof (tsadmin_t)))
389 			return (EFAULT);
390 	}
391 #ifdef _SYSCALL32_IMPL
392 	else {
393 		/* get tsadmin struct from ILP32 caller */
394 		tsadmin32_t tsadmin32;
395 		if (copyin(uaddr, &tsadmin32, sizeof (tsadmin32_t)))
396 			return (EFAULT);
397 		tsadmin.ts_dpents =
398 		    (struct tsdpent *)(uintptr_t)tsadmin32.ts_dpents;
399 		tsadmin.ts_ndpents = tsadmin32.ts_ndpents;
400 		tsadmin.ts_cmd = tsadmin32.ts_cmd;
401 	}
402 #endif /* _SYSCALL32_IMPL */
403 
404 	tsdpsz = (ts_maxumdpri + 1) * sizeof (tsdpent_t);
405 
406 	switch (tsadmin.ts_cmd) {
407 	case TS_GETDPSIZE:
408 		tsadmin.ts_ndpents = ts_maxumdpri + 1;
409 
410 		if (get_udatamodel() == DATAMODEL_NATIVE) {
411 			if (copyout(&tsadmin, uaddr, sizeof (tsadmin_t)))
412 				return (EFAULT);
413 		}
414 #ifdef _SYSCALL32_IMPL
415 		else {
416 			/* return tsadmin struct to ILP32 caller */
417 			tsadmin32_t tsadmin32;
418 			tsadmin32.ts_dpents =
419 			    (caddr32_t)(uintptr_t)tsadmin.ts_dpents;
420 			tsadmin32.ts_ndpents = tsadmin.ts_ndpents;
421 			tsadmin32.ts_cmd = tsadmin.ts_cmd;
422 			if (copyout(&tsadmin32, uaddr, sizeof (tsadmin32_t)))
423 				return (EFAULT);
424 		}
425 #endif /* _SYSCALL32_IMPL */
426 		break;
427 
428 	case TS_GETDPTBL:
429 		userdpsz = MIN(tsadmin.ts_ndpents * sizeof (tsdpent_t),
430 		    tsdpsz);
431 		if (copyout(ts_dptbl, tsadmin.ts_dpents, userdpsz))
432 			return (EFAULT);
433 
434 		tsadmin.ts_ndpents = userdpsz / sizeof (tsdpent_t);
435 
436 		if (get_udatamodel() == DATAMODEL_NATIVE) {
437 			if (copyout(&tsadmin, uaddr, sizeof (tsadmin_t)))
438 				return (EFAULT);
439 		}
440 #ifdef _SYSCALL32_IMPL
441 		else {
442 			/* return tsadmin struct to ILP32 callers */
443 			tsadmin32_t tsadmin32;
444 			tsadmin32.ts_dpents =
445 			    (caddr32_t)(uintptr_t)tsadmin.ts_dpents;
446 			tsadmin32.ts_ndpents = tsadmin.ts_ndpents;
447 			tsadmin32.ts_cmd = tsadmin.ts_cmd;
448 			if (copyout(&tsadmin32, uaddr, sizeof (tsadmin32_t)))
449 				return (EFAULT);
450 		}
451 #endif /* _SYSCALL32_IMPL */
452 		break;
453 
454 	case TS_SETDPTBL:
455 		/*
456 		 * We require that the requesting process has sufficient
457 		 * priveleges.  We also require that the table supplied by
458 		 * the user exactly match the current ts_dptbl in size.
459 		 */
460 		if (secpolicy_dispadm(reqpcredp) != 0)
461 			return (EPERM);
462 
463 		if (tsadmin.ts_ndpents * sizeof (tsdpent_t) != tsdpsz) {
464 			return (EINVAL);
465 		}
466 
467 		/*
468 		 * We read the user supplied table into a temporary buffer
469 		 * where it is validated before being copied over the
470 		 * ts_dptbl.
471 		 */
472 		tmpdpp = kmem_alloc(tsdpsz, KM_SLEEP);
473 		if (copyin((caddr_t)tsadmin.ts_dpents, (caddr_t)tmpdpp,
474 		    tsdpsz)) {
475 			kmem_free(tmpdpp, tsdpsz);
476 			return (EFAULT);
477 		}
478 		for (i = 0; i < tsadmin.ts_ndpents; i++) {
479 
480 			/*
481 			 * Validate the user supplied values.  All we are doing
482 			 * here is verifying that the values are within their
483 			 * allowable ranges and will not panic the system.  We
484 			 * make no attempt to ensure that the resulting
485 			 * configuration makes sense or results in reasonable
486 			 * performance.
487 			 */
488 			if (tmpdpp[i].ts_quantum <= 0) {
489 				kmem_free(tmpdpp, tsdpsz);
490 				return (EINVAL);
491 			}
492 			if (tmpdpp[i].ts_tqexp > ts_maxumdpri ||
493 			    tmpdpp[i].ts_tqexp < 0) {
494 				kmem_free(tmpdpp, tsdpsz);
495 				return (EINVAL);
496 			}
497 			if (tmpdpp[i].ts_slpret > ts_maxumdpri ||
498 			    tmpdpp[i].ts_slpret < 0) {
499 				kmem_free(tmpdpp, tsdpsz);
500 				return (EINVAL);
501 			}
502 			if (tmpdpp[i].ts_maxwait < 0) {
503 				kmem_free(tmpdpp, tsdpsz);
504 				return (EINVAL);
505 			}
506 			if (tmpdpp[i].ts_lwait > ts_maxumdpri ||
507 			    tmpdpp[i].ts_lwait < 0) {
508 				kmem_free(tmpdpp, tsdpsz);
509 				return (EINVAL);
510 			}
511 		}
512 
513 		/*
514 		 * Copy the user supplied values over the current ts_dptbl
515 		 * values.  The ts_globpri member is read-only so we don't
516 		 * overwrite it.
517 		 */
518 		mutex_enter(&ts_dptblock);
519 		for (i = 0; i < tsadmin.ts_ndpents; i++) {
520 			ts_dptbl[i].ts_quantum = tmpdpp[i].ts_quantum;
521 			ts_dptbl[i].ts_tqexp = tmpdpp[i].ts_tqexp;
522 			ts_dptbl[i].ts_slpret = tmpdpp[i].ts_slpret;
523 			ts_dptbl[i].ts_maxwait = tmpdpp[i].ts_maxwait;
524 			ts_dptbl[i].ts_lwait = tmpdpp[i].ts_lwait;
525 		}
526 		mutex_exit(&ts_dptblock);
527 		kmem_free(tmpdpp, tsdpsz);
528 		break;
529 
530 	default:
531 		return (EINVAL);
532 	}
533 	return (0);
534 }
535 
536 
537 /*
538  * Allocate a time-sharing class specific thread structure and
539  * initialize it with the parameters supplied. Also move the thread
540  * to specified time-sharing priority.
541  */
542 static int
543 ts_enterclass(kthread_t *t, id_t cid, void *parmsp, cred_t *reqpcredp,
544     void *bufp)
545 {
546 	tsparms_t	*tsparmsp = (tsparms_t *)parmsp;
547 	tsproc_t	*tspp;
548 	pri_t		reqtsuprilim;
549 	pri_t		reqtsupri;
550 	static uint32_t	tspexists = 0;	/* set on first occurrence of */
551 					/*   a time-sharing process */
552 
553 	tspp = (tsproc_t *)bufp;
554 	ASSERT(tspp != NULL);
555 
556 	/*
557 	 * Initialize the tsproc structure.
558 	 */
559 	tspp->ts_cpupri = tsmedumdpri;
560 	if (cid == ia_cid) {
561 		/*
562 		 * Check to make sure caller is either privileged or the
563 		 * window system.  When the window system is converted
564 		 * to using privileges, the second check can go away.
565 		 */
566 		if (reqpcredp != NULL && !groupmember(IA_gid, reqpcredp) &&
567 		    secpolicy_setpriority(reqpcredp) != 0)
568 			return (EPERM);
569 		/*
570 		 * Belongs to IA "class", so set appropriate flags.
571 		 * Mark as 'on' so it will not be a swap victim
572 		 * while forking.
573 		 */
574 		tspp->ts_flags = TSIA | TSIASET;
575 		tspp->ts_boost = ia_boost;
576 	} else {
577 		tspp->ts_flags = 0;
578 		tspp->ts_boost = 0;
579 	}
580 
581 	if (tsparmsp == NULL) {
582 		/*
583 		 * Use default values.
584 		 */
585 		tspp->ts_uprilim = tspp->ts_upri = 0;
586 		tspp->ts_nice = NZERO;
587 	} else {
588 		/*
589 		 * Use supplied values.
590 		 */
591 		if (tsparmsp->ts_uprilim == TS_NOCHANGE)
592 			reqtsuprilim = 0;
593 		else {
594 			if (tsparmsp->ts_uprilim > 0 &&
595 			    secpolicy_setpriority(reqpcredp) != 0)
596 				return (EPERM);
597 			reqtsuprilim = tsparmsp->ts_uprilim;
598 		}
599 
600 		if (tsparmsp->ts_upri == TS_NOCHANGE) {
601 			reqtsupri = reqtsuprilim;
602 		} else {
603 			if (tsparmsp->ts_upri > 0 &&
604 			    secpolicy_setpriority(reqpcredp) != 0)
605 				return (EPERM);
606 			/*
607 			 * Set the user priority to the requested value
608 			 * or the upri limit, whichever is lower.
609 			 */
610 			reqtsupri = tsparmsp->ts_upri;
611 			if (reqtsupri > reqtsuprilim)
612 				reqtsupri = reqtsuprilim;
613 		}
614 
615 
616 		tspp->ts_uprilim = reqtsuprilim;
617 		tspp->ts_upri = reqtsupri;
618 		tspp->ts_nice = NZERO - (NZERO * reqtsupri) / ts_maxupri;
619 	}
620 	TS_NEWUMDPRI(tspp);
621 
622 	tspp->ts_dispwait = 0;
623 	tspp->ts_timeleft = ts_dptbl[tspp->ts_cpupri].ts_quantum;
624 	tspp->ts_tp = t;
625 	cpucaps_sc_init(&tspp->ts_caps);
626 
627 	/*
628 	 * Reset priority. Process goes to a "user mode" priority
629 	 * here regardless of whether or not it has slept since
630 	 * entering the kernel.
631 	 */
632 	thread_lock(t);			/* get dispatcher lock on thread */
633 	t->t_clfuncs = &(sclass[cid].cl_funcs->thread);
634 	t->t_cid = cid;
635 	t->t_cldata = (void *)tspp;
636 	t->t_schedflag &= ~TS_RUNQMATCH;
637 	ts_change_priority(t, tspp);
638 	thread_unlock(t);
639 
640 	/*
641 	 * Link new structure into tsproc list.
642 	 */
643 	TS_LIST_INSERT(tspp);
644 
645 	/*
646 	 * If this is the first time-sharing thread to occur since
647 	 * boot we set up the initial call to ts_update() here.
648 	 * Use an atomic compare-and-swap since that's easier and
649 	 * faster than a mutex (but check with an ordinary load first
650 	 * since most of the time this will already be done).
651 	 */
652 	if (tspexists == 0 && atomic_cas_32(&tspexists, 0, 1) == 0)
653 		(void) timeout(ts_update, NULL, hz);
654 
655 	return (0);
656 }
657 
658 
659 /*
660  * Free tsproc structure of thread.
661  */
662 static void
663 ts_exitclass(void *procp)
664 {
665 	tsproc_t *tspp = (tsproc_t *)procp;
666 
667 	/* Remove tsproc_t structure from list */
668 	TS_LIST_DELETE(tspp);
669 	kmem_free(tspp, sizeof (tsproc_t));
670 }
671 
672 /* ARGSUSED */
673 static int
674 ts_canexit(kthread_t *t, cred_t *cred)
675 {
676 	/*
677 	 * A thread can always leave a TS/IA class
678 	 */
679 	return (0);
680 }
681 
682 static int
683 ts_fork(kthread_t *t, kthread_t *ct, void *bufp)
684 {
685 	tsproc_t	*ptspp;		/* ptr to parent's tsproc structure */
686 	tsproc_t	*ctspp;		/* ptr to child's tsproc structure */
687 
688 	ASSERT(MUTEX_HELD(&ttoproc(t)->p_lock));
689 
690 	ctspp = (tsproc_t *)bufp;
691 	ASSERT(ctspp != NULL);
692 	ptspp = (tsproc_t *)t->t_cldata;
693 	/*
694 	 * Initialize child's tsproc structure.
695 	 */
696 	thread_lock(t);
697 	ctspp->ts_timeleft = ts_dptbl[ptspp->ts_cpupri].ts_quantum;
698 	ctspp->ts_cpupri = ptspp->ts_cpupri;
699 	ctspp->ts_boost = ptspp->ts_boost;
700 	ctspp->ts_uprilim = ptspp->ts_uprilim;
701 	ctspp->ts_upri = ptspp->ts_upri;
702 	TS_NEWUMDPRI(ctspp);
703 	ctspp->ts_nice = ptspp->ts_nice;
704 	ctspp->ts_dispwait = 0;
705 	ctspp->ts_flags = ptspp->ts_flags & ~(TSBACKQ | TSRESTORE);
706 	ctspp->ts_tp = ct;
707 	cpucaps_sc_init(&ctspp->ts_caps);
708 	thread_unlock(t);
709 
710 	/*
711 	 * Link new structure into tsproc list.
712 	 */
713 	ct->t_cldata = (void *)ctspp;
714 	TS_LIST_INSERT(ctspp);
715 	return (0);
716 }
717 
718 
719 /*
720  * Child is placed at back of dispatcher queue and parent gives
721  * up processor so that the child runs first after the fork.
722  * This allows the child immediately execing to break the multiple
723  * use of copy on write pages with no disk home. The parent will
724  * get to steal them back rather than uselessly copying them.
725  */
726 static void
727 ts_forkret(kthread_t *t, kthread_t *ct)
728 {
729 	proc_t	*pp = ttoproc(t);
730 	proc_t	*cp = ttoproc(ct);
731 	tsproc_t *tspp;
732 
733 	ASSERT(t == curthread);
734 	ASSERT(MUTEX_HELD(&pidlock));
735 
736 	/*
737 	 * Grab the child's p_lock before dropping pidlock to ensure
738 	 * the process does not disappear before we set it running.
739 	 */
740 	mutex_enter(&cp->p_lock);
741 	continuelwps(cp);
742 	mutex_exit(&cp->p_lock);
743 
744 	mutex_enter(&pp->p_lock);
745 	mutex_exit(&pidlock);
746 	continuelwps(pp);
747 
748 	thread_lock(t);
749 	tspp = (tsproc_t *)(t->t_cldata);
750 	tspp->ts_cpupri = ts_dptbl[tspp->ts_cpupri].ts_tqexp;
751 	TS_NEWUMDPRI(tspp);
752 	tspp->ts_timeleft = ts_dptbl[tspp->ts_cpupri].ts_quantum;
753 	tspp->ts_dispwait = 0;
754 	t->t_pri = ts_dptbl[tspp->ts_umdpri].ts_globpri;
755 	ASSERT(t->t_pri >= 0 && t->t_pri <= ts_maxglobpri);
756 	THREAD_TRANSITION(t);
757 	ts_setrun(t);
758 	thread_unlock(t);
759 	/*
760 	 * Safe to drop p_lock now since since it is safe to change
761 	 * the scheduling class after this point.
762 	 */
763 	mutex_exit(&pp->p_lock);
764 
765 	swtch();
766 }
767 
768 
769 /*
770  * Get information about the time-sharing class into the buffer
771  * pointed to by tsinfop. The maximum configured user priority
772  * is the only information we supply.  ts_getclinfo() is called
773  * for TS threads, and ia_getclinfo() is called for IA threads.
774  */
775 static int
776 ts_getclinfo(void *infop)
777 {
778 	tsinfo_t *tsinfop = (tsinfo_t *)infop;
779 	tsinfop->ts_maxupri = ts_maxupri;
780 	return (0);
781 }
782 
783 static int
784 ia_getclinfo(void *infop)
785 {
786 	iainfo_t *iainfop = (iainfo_t *)infop;
787 	iainfop->ia_maxupri = ia_maxupri;
788 	return (0);
789 }
790 
791 
792 /*
793  * Return the user mode scheduling priority range.
794  */
795 static int
796 ts_getclpri(pcpri_t *pcprip)
797 {
798 	pcprip->pc_clpmax = ts_maxupri;
799 	pcprip->pc_clpmin = -ts_maxupri;
800 	return (0);
801 }
802 
803 
804 static int
805 ia_getclpri(pcpri_t *pcprip)
806 {
807 	pcprip->pc_clpmax = ia_maxupri;
808 	pcprip->pc_clpmin = -ia_maxupri;
809 	return (0);
810 }
811 
812 
813 static void
814 ts_nullsys()
815 {}
816 
817 
818 /*
819  * Get the time-sharing parameters of the thread pointed to by
820  * tsprocp into the buffer pointed to by tsparmsp.  ts_parmsget()
821  * is called for TS threads, and ia_parmsget() is called for IA
822  * threads.
823  */
824 static void
825 ts_parmsget(kthread_t *t, void *parmsp)
826 {
827 	tsproc_t *tspp = (tsproc_t *)t->t_cldata;
828 	tsparms_t *tsparmsp = (tsparms_t *)parmsp;
829 
830 	tsparmsp->ts_uprilim = tspp->ts_uprilim;
831 	tsparmsp->ts_upri = tspp->ts_upri;
832 }
833 
834 static void
835 ia_parmsget(kthread_t *t, void *parmsp)
836 {
837 	tsproc_t *tspp = (tsproc_t *)t->t_cldata;
838 	iaparms_t *iaparmsp = (iaparms_t *)parmsp;
839 
840 	iaparmsp->ia_uprilim = tspp->ts_uprilim;
841 	iaparmsp->ia_upri = tspp->ts_upri;
842 	if (tspp->ts_flags & TSIASET)
843 		iaparmsp->ia_mode = IA_SET_INTERACTIVE;
844 	else
845 		iaparmsp->ia_mode = IA_INTERACTIVE_OFF;
846 }
847 
848 
849 /*
850  * Check the validity of the time-sharing parameters in the buffer
851  * pointed to by tsparmsp.
852  * ts_parmsin() is called for TS threads, and ia_parmsin() is called
853  * for IA threads.
854  */
855 static int
856 ts_parmsin(void *parmsp)
857 {
858 	tsparms_t	*tsparmsp = (tsparms_t *)parmsp;
859 	/*
860 	 * Check validity of parameters.
861 	 */
862 	if ((tsparmsp->ts_uprilim > ts_maxupri ||
863 	    tsparmsp->ts_uprilim < -ts_maxupri) &&
864 	    tsparmsp->ts_uprilim != TS_NOCHANGE)
865 		return (EINVAL);
866 
867 	if ((tsparmsp->ts_upri > ts_maxupri ||
868 	    tsparmsp->ts_upri < -ts_maxupri) &&
869 	    tsparmsp->ts_upri != TS_NOCHANGE)
870 		return (EINVAL);
871 
872 	return (0);
873 }
874 
875 static int
876 ia_parmsin(void *parmsp)
877 {
878 	iaparms_t	*iaparmsp = (iaparms_t *)parmsp;
879 
880 	if ((iaparmsp->ia_uprilim > ia_maxupri ||
881 	    iaparmsp->ia_uprilim < -ia_maxupri) &&
882 	    iaparmsp->ia_uprilim != IA_NOCHANGE) {
883 		return (EINVAL);
884 	}
885 
886 	if ((iaparmsp->ia_upri > ia_maxupri ||
887 	    iaparmsp->ia_upri < -ia_maxupri) &&
888 	    iaparmsp->ia_upri != IA_NOCHANGE) {
889 		return (EINVAL);
890 	}
891 
892 	return (0);
893 }
894 
895 
896 /*
897  * Check the validity of the time-sharing parameters in the pc_vaparms_t
898  * structure vaparmsp and put them in the buffer pointed to by tsparmsp.
899  * pc_vaparms_t contains (key, value) pairs of parameter.
900  * ts_vaparmsin() is called for TS threads, and ia_vaparmsin() is called
901  * for IA threads. ts_vaparmsin() is the variable parameter version of
902  * ts_parmsin() and ia_vaparmsin() is the variable parameter version of
903  * ia_parmsin().
904  */
905 static int
906 ts_vaparmsin(void *parmsp, pc_vaparms_t *vaparmsp)
907 {
908 	tsparms_t	*tsparmsp = (tsparms_t *)parmsp;
909 	int		priflag = 0;
910 	int		limflag = 0;
911 	uint_t		cnt;
912 	pc_vaparm_t	*vpp = &vaparmsp->pc_parms[0];
913 
914 
915 	/*
916 	 * TS_NOCHANGE (-32768) is outside of the range of values for
917 	 * ts_uprilim and ts_upri. If the structure tsparms_t is changed,
918 	 * TS_NOCHANGE should be replaced by a flag word (in the same manner
919 	 * as in rt.c).
920 	 */
921 	tsparmsp->ts_uprilim = TS_NOCHANGE;
922 	tsparmsp->ts_upri = TS_NOCHANGE;
923 
924 	/*
925 	 * Get the varargs parameter and check validity of parameters.
926 	 */
927 	if (vaparmsp->pc_vaparmscnt > PC_VAPARMCNT)
928 		return (EINVAL);
929 
930 	for (cnt = 0; cnt < vaparmsp->pc_vaparmscnt; cnt++, vpp++) {
931 
932 		switch (vpp->pc_key) {
933 		case TS_KY_UPRILIM:
934 			if (limflag++)
935 				return (EINVAL);
936 			tsparmsp->ts_uprilim = (pri_t)vpp->pc_parm;
937 			if (tsparmsp->ts_uprilim > ts_maxupri ||
938 			    tsparmsp->ts_uprilim < -ts_maxupri)
939 				return (EINVAL);
940 			break;
941 
942 		case TS_KY_UPRI:
943 			if (priflag++)
944 				return (EINVAL);
945 			tsparmsp->ts_upri = (pri_t)vpp->pc_parm;
946 			if (tsparmsp->ts_upri > ts_maxupri ||
947 			    tsparmsp->ts_upri < -ts_maxupri)
948 				return (EINVAL);
949 			break;
950 
951 		default:
952 			return (EINVAL);
953 		}
954 	}
955 
956 	if (vaparmsp->pc_vaparmscnt == 0) {
957 		/*
958 		 * Use default parameters.
959 		 */
960 		tsparmsp->ts_upri = tsparmsp->ts_uprilim = 0;
961 	}
962 
963 	return (0);
964 }
965 
966 static int
967 ia_vaparmsin(void *parmsp, pc_vaparms_t *vaparmsp)
968 {
969 	iaparms_t	*iaparmsp = (iaparms_t *)parmsp;
970 	int		priflag = 0;
971 	int		limflag = 0;
972 	int		mflag = 0;
973 	uint_t		cnt;
974 	pc_vaparm_t	*vpp = &vaparmsp->pc_parms[0];
975 
976 	/*
977 	 * IA_NOCHANGE (-32768) is outside of the range of values for
978 	 * ia_uprilim, ia_upri and ia_mode. If the structure iaparms_t is
979 	 * changed, IA_NOCHANGE should be replaced by a flag word (in the
980 	 * same manner as in rt.c).
981 	 */
982 	iaparmsp->ia_uprilim = IA_NOCHANGE;
983 	iaparmsp->ia_upri = IA_NOCHANGE;
984 	iaparmsp->ia_mode = IA_NOCHANGE;
985 
986 	/*
987 	 * Get the varargs parameter and check validity of parameters.
988 	 */
989 	if (vaparmsp->pc_vaparmscnt > PC_VAPARMCNT)
990 		return (EINVAL);
991 
992 	for (cnt = 0; cnt < vaparmsp->pc_vaparmscnt; cnt++, vpp++) {
993 
994 		switch (vpp->pc_key) {
995 		case IA_KY_UPRILIM:
996 			if (limflag++)
997 				return (EINVAL);
998 			iaparmsp->ia_uprilim = (pri_t)vpp->pc_parm;
999 			if (iaparmsp->ia_uprilim > ia_maxupri ||
1000 			    iaparmsp->ia_uprilim < -ia_maxupri)
1001 				return (EINVAL);
1002 			break;
1003 
1004 		case IA_KY_UPRI:
1005 			if (priflag++)
1006 				return (EINVAL);
1007 			iaparmsp->ia_upri = (pri_t)vpp->pc_parm;
1008 			if (iaparmsp->ia_upri > ia_maxupri ||
1009 			    iaparmsp->ia_upri < -ia_maxupri)
1010 				return (EINVAL);
1011 			break;
1012 
1013 		case IA_KY_MODE:
1014 			if (mflag++)
1015 				return (EINVAL);
1016 			iaparmsp->ia_mode = (int)vpp->pc_parm;
1017 			if (iaparmsp->ia_mode != IA_SET_INTERACTIVE &&
1018 			    iaparmsp->ia_mode != IA_INTERACTIVE_OFF)
1019 				return (EINVAL);
1020 			break;
1021 
1022 		default:
1023 			return (EINVAL);
1024 		}
1025 	}
1026 
1027 	if (vaparmsp->pc_vaparmscnt == 0) {
1028 		/*
1029 		 * Use default parameters.
1030 		 */
1031 		iaparmsp->ia_upri = iaparmsp->ia_uprilim = 0;
1032 		iaparmsp->ia_mode = IA_SET_INTERACTIVE;
1033 	}
1034 
1035 	return (0);
1036 }
1037 
1038 /*
1039  * Nothing to do here but return success.
1040  */
1041 /* ARGSUSED */
1042 static int
1043 ts_parmsout(void *parmsp, pc_vaparms_t *vaparmsp)
1044 {
1045 	return (0);
1046 }
1047 
1048 
1049 /*
1050  * Copy all selected time-sharing class parameters to the user.
1051  * The parameters are specified by a key.
1052  */
1053 static int
1054 ts_vaparmsout(void *prmsp, pc_vaparms_t *vaparmsp)
1055 {
1056 	tsparms_t	*tsprmsp = (tsparms_t *)prmsp;
1057 	int		priflag = 0;
1058 	int		limflag = 0;
1059 	uint_t		cnt;
1060 	pc_vaparm_t	*vpp = &vaparmsp->pc_parms[0];
1061 
1062 	ASSERT(MUTEX_NOT_HELD(&curproc->p_lock));
1063 
1064 	if (vaparmsp->pc_vaparmscnt > PC_VAPARMCNT)
1065 		return (EINVAL);
1066 
1067 	for (cnt = 0; cnt < vaparmsp->pc_vaparmscnt; cnt++, vpp++) {
1068 
1069 		switch (vpp->pc_key) {
1070 		case TS_KY_UPRILIM:
1071 			if (limflag++)
1072 				return (EINVAL);
1073 			if (copyout(&tsprmsp->ts_uprilim,
1074 			    (caddr_t)(uintptr_t)vpp->pc_parm, sizeof (pri_t)))
1075 				return (EFAULT);
1076 			break;
1077 
1078 		case TS_KY_UPRI:
1079 			if (priflag++)
1080 				return (EINVAL);
1081 			if (copyout(&tsprmsp->ts_upri,
1082 			    (caddr_t)(uintptr_t)vpp->pc_parm, sizeof (pri_t)))
1083 				return (EFAULT);
1084 			break;
1085 
1086 		default:
1087 			return (EINVAL);
1088 		}
1089 	}
1090 
1091 	return (0);
1092 }
1093 
1094 
1095 /*
1096  * Copy all selected interactive class parameters to the user.
1097  * The parameters are specified by a key.
1098  */
1099 static int
1100 ia_vaparmsout(void *prmsp, pc_vaparms_t *vaparmsp)
1101 {
1102 	iaparms_t	*iaprmsp = (iaparms_t *)prmsp;
1103 	int		priflag = 0;
1104 	int		limflag = 0;
1105 	int		mflag = 0;
1106 	uint_t		cnt;
1107 	pc_vaparm_t	*vpp = &vaparmsp->pc_parms[0];
1108 
1109 	ASSERT(MUTEX_NOT_HELD(&curproc->p_lock));
1110 
1111 	if (vaparmsp->pc_vaparmscnt > PC_VAPARMCNT)
1112 		return (EINVAL);
1113 
1114 	for (cnt = 0; cnt < vaparmsp->pc_vaparmscnt; cnt++, vpp++) {
1115 
1116 		switch (vpp->pc_key) {
1117 		case IA_KY_UPRILIM:
1118 			if (limflag++)
1119 				return (EINVAL);
1120 			if (copyout(&iaprmsp->ia_uprilim,
1121 			    (caddr_t)(uintptr_t)vpp->pc_parm, sizeof (pri_t)))
1122 				return (EFAULT);
1123 			break;
1124 
1125 		case IA_KY_UPRI:
1126 			if (priflag++)
1127 				return (EINVAL);
1128 			if (copyout(&iaprmsp->ia_upri,
1129 			    (caddr_t)(uintptr_t)vpp->pc_parm, sizeof (pri_t)))
1130 				return (EFAULT);
1131 			break;
1132 
1133 		case IA_KY_MODE:
1134 			if (mflag++)
1135 				return (EINVAL);
1136 			if (copyout(&iaprmsp->ia_mode,
1137 			    (caddr_t)(uintptr_t)vpp->pc_parm, sizeof (int)))
1138 				return (EFAULT);
1139 			break;
1140 
1141 		default:
1142 			return (EINVAL);
1143 		}
1144 	}
1145 	return (0);
1146 }
1147 
1148 
1149 /*
1150  * Set the scheduling parameters of the thread pointed to by tsprocp
1151  * to those specified in the buffer pointed to by tsparmsp.
1152  * ts_parmsset() is called for TS threads, and ia_parmsset() is
1153  * called for IA threads.
1154  */
1155 /* ARGSUSED */
1156 static int
1157 ts_parmsset(kthread_t *tx, void *parmsp, id_t reqpcid, cred_t *reqpcredp)
1158 {
1159 	char		nice;
1160 	pri_t		reqtsuprilim;
1161 	pri_t		reqtsupri;
1162 	tsparms_t	*tsparmsp = (tsparms_t *)parmsp;
1163 	tsproc_t	*tspp = (tsproc_t *)tx->t_cldata;
1164 
1165 	ASSERT(MUTEX_HELD(&(ttoproc(tx))->p_lock));
1166 
1167 	if (tsparmsp->ts_uprilim == TS_NOCHANGE)
1168 		reqtsuprilim = tspp->ts_uprilim;
1169 	else
1170 		reqtsuprilim = tsparmsp->ts_uprilim;
1171 
1172 	if (tsparmsp->ts_upri == TS_NOCHANGE)
1173 		reqtsupri = tspp->ts_upri;
1174 	else
1175 		reqtsupri = tsparmsp->ts_upri;
1176 
1177 	/*
1178 	 * Make sure the user priority doesn't exceed the upri limit.
1179 	 */
1180 	if (reqtsupri > reqtsuprilim)
1181 		reqtsupri = reqtsuprilim;
1182 
1183 	/*
1184 	 * Basic permissions enforced by generic kernel code
1185 	 * for all classes require that a thread attempting
1186 	 * to change the scheduling parameters of a target
1187 	 * thread be privileged or have a real or effective
1188 	 * UID matching that of the target thread. We are not
1189 	 * called unless these basic permission checks have
1190 	 * already passed. The time-sharing class requires in
1191 	 * addition that the calling thread be privileged if it
1192 	 * is attempting to raise the upri limit above its current
1193 	 * value This may have been checked previously but if our
1194 	 * caller passed us a non-NULL credential pointer we assume
1195 	 * it hasn't and we check it here.
1196 	 */
1197 	if (reqpcredp != NULL &&
1198 	    reqtsuprilim > tspp->ts_uprilim &&
1199 	    secpolicy_raisepriority(reqpcredp) != 0)
1200 		return (EPERM);
1201 
1202 	/*
1203 	 * Set ts_nice to the nice value corresponding to the user
1204 	 * priority we are setting.  Note that setting the nice field
1205 	 * of the parameter struct won't affect upri or nice.
1206 	 */
1207 	nice = NZERO - (reqtsupri * NZERO) / ts_maxupri;
1208 	if (nice >= 2 * NZERO)
1209 		nice = 2 * NZERO - 1;
1210 
1211 	thread_lock(tx);
1212 
1213 	tspp->ts_uprilim = reqtsuprilim;
1214 	tspp->ts_upri = reqtsupri;
1215 	TS_NEWUMDPRI(tspp);
1216 	tspp->ts_nice = nice;
1217 
1218 	tspp->ts_dispwait = 0;
1219 	ts_change_priority(tx, tspp);
1220 	thread_unlock(tx);
1221 	return (0);
1222 }
1223 
1224 
1225 static int
1226 ia_parmsset(kthread_t *tx, void *parmsp, id_t reqpcid, cred_t *reqpcredp)
1227 {
1228 	tsproc_t	*tspp = (tsproc_t *)tx->t_cldata;
1229 	iaparms_t	*iaparmsp = (iaparms_t *)parmsp;
1230 	proc_t		*p;
1231 	pid_t		pid, pgid, sid;
1232 	pid_t		on, off;
1233 	struct stdata	*stp;
1234 	int		sess_held;
1235 
1236 	/*
1237 	 * Handle user priority changes
1238 	 */
1239 	if (iaparmsp->ia_mode == IA_NOCHANGE)
1240 		return (ts_parmsset(tx, parmsp, reqpcid, reqpcredp));
1241 
1242 	/*
1243 	 * Check permissions for changing modes.
1244 	 */
1245 
1246 	if (reqpcredp != NULL && !groupmember(IA_gid, reqpcredp) &&
1247 	    secpolicy_raisepriority(reqpcredp) != 0) {
1248 		/*
1249 		 * Silently fail in case this is just a priocntl
1250 		 * call with upri and uprilim set to IA_NOCHANGE.
1251 		 */
1252 		return (0);
1253 	}
1254 
1255 	ASSERT(MUTEX_HELD(&pidlock));
1256 	if ((p = ttoproc(tx)) == NULL) {
1257 		return (0);
1258 	}
1259 	ASSERT(MUTEX_HELD(&p->p_lock));
1260 	if (p->p_stat == SIDL) {
1261 		return (0);
1262 	}
1263 	pid = p->p_pid;
1264 	sid = p->p_sessp->s_sid;
1265 	pgid = p->p_pgrp;
1266 	if (iaparmsp->ia_mode == IA_SET_INTERACTIVE) {
1267 		/*
1268 		 * session leaders must be turned on now so all processes
1269 		 * in the group controlling the tty will be turned on or off.
1270 		 * if the ia_mode is off for the session leader,
1271 		 * ia_set_process_group will return without setting the
1272 		 * processes in the group controlling the tty on.
1273 		 */
1274 		thread_lock(tx);
1275 		tspp->ts_flags |= TSIASET;
1276 		thread_unlock(tx);
1277 	}
1278 	mutex_enter(&p->p_sessp->s_lock);
1279 	sess_held = 1;
1280 	if ((pid == sid) && (p->p_sessp->s_vp != NULL) &&
1281 	    ((stp = p->p_sessp->s_vp->v_stream) != NULL)) {
1282 		if ((stp->sd_pgidp != NULL) && (stp->sd_sidp != NULL)) {
1283 			pgid = stp->sd_pgidp->pid_id;
1284 			sess_held = 0;
1285 			mutex_exit(&p->p_sessp->s_lock);
1286 			if (iaparmsp->ia_mode ==
1287 			    IA_SET_INTERACTIVE) {
1288 				off = 0;
1289 				on = pgid;
1290 			} else {
1291 				off = pgid;
1292 				on = 0;
1293 			}
1294 			TRACE_3(TR_FAC_IA, TR_ACTIVE_CHAIN,
1295 			    "active chain:pid %d gid %d %p",
1296 			    pid, pgid, p);
1297 			ia_set_process_group(sid, off, on);
1298 		}
1299 	}
1300 	if (sess_held)
1301 		mutex_exit(&p->p_sessp->s_lock);
1302 
1303 	thread_lock(tx);
1304 
1305 	if (iaparmsp->ia_mode == IA_SET_INTERACTIVE) {
1306 		tspp->ts_flags |= TSIASET;
1307 		tspp->ts_boost = ia_boost;
1308 	} else {
1309 		tspp->ts_flags &= ~TSIASET;
1310 		tspp->ts_boost = -ia_boost;
1311 	}
1312 	thread_unlock(tx);
1313 
1314 	return (ts_parmsset(tx, parmsp, reqpcid, reqpcredp));
1315 }
1316 
1317 static void
1318 ts_exit(kthread_t *t)
1319 {
1320 	tsproc_t *tspp;
1321 
1322 	if (CPUCAPS_ON()) {
1323 		/*
1324 		 * A thread could be exiting in between clock ticks,
1325 		 * so we need to calculate how much CPU time it used
1326 		 * since it was charged last time.
1327 		 *
1328 		 * CPU caps are not enforced on exiting processes - it is
1329 		 * usually desirable to exit as soon as possible to free
1330 		 * resources.
1331 		 */
1332 		thread_lock(t);
1333 		tspp = (tsproc_t *)t->t_cldata;
1334 		(void) cpucaps_charge(t, &tspp->ts_caps, CPUCAPS_CHARGE_ONLY);
1335 		thread_unlock(t);
1336 	}
1337 }
1338 
1339 /*
1340  * Return the global scheduling priority that would be assigned
1341  * to a thread entering the time-sharing class with the ts_upri.
1342  */
1343 static pri_t
1344 ts_globpri(kthread_t *t)
1345 {
1346 	tsproc_t *tspp;
1347 	pri_t	tspri;
1348 
1349 	ASSERT(MUTEX_HELD(&ttoproc(t)->p_lock));
1350 	tspp = (tsproc_t *)t->t_cldata;
1351 	tspri = tsmedumdpri + tspp->ts_upri;
1352 	if (tspri > ts_maxumdpri)
1353 		tspri = ts_maxumdpri;
1354 	else if (tspri < 0)
1355 		tspri = 0;
1356 	return (ts_dptbl[tspri].ts_globpri);
1357 }
1358 
1359 /*
1360  * Arrange for thread to be placed in appropriate location
1361  * on dispatcher queue.
1362  *
1363  * This is called with the current thread in TS_ONPROC and locked.
1364  */
1365 static void
1366 ts_preempt(kthread_t *t)
1367 {
1368 	tsproc_t	*tspp = (tsproc_t *)(t->t_cldata);
1369 	klwp_t		*lwp = ttolwp(t);
1370 	pri_t		oldpri = t->t_pri;
1371 
1372 	ASSERT(t == curthread);
1373 	ASSERT(THREAD_LOCK_HELD(curthread));
1374 
1375 	/*
1376 	 * This thread may be placed on wait queue by CPU Caps. In this case we
1377 	 * do not need to do anything until it is removed from the wait queue.
1378 	 */
1379 	if (CPUCAPS_ON()) {
1380 		(void) cpucaps_charge(t, &tspp->ts_caps,
1381 		    CPUCAPS_CHARGE_ENFORCE);
1382 		if (CPUCAPS_ENFORCE(t))
1383 			return;
1384 	}
1385 
1386 	/*
1387 	 * If thread got preempted in the user-land then we know
1388 	 * it isn't holding any locks.  Mark it as swappable.
1389 	 */
1390 	ASSERT(t->t_schedflag & TS_DONT_SWAP);
1391 	if (lwp != NULL && lwp->lwp_state == LWP_USER)
1392 		t->t_schedflag &= ~TS_DONT_SWAP;
1393 
1394 	/*
1395 	 * Check to see if we're doing "preemption control" here.  If
1396 	 * we are, and if the user has requested that this thread not
1397 	 * be preempted, and if preemptions haven't been put off for
1398 	 * too long, let the preemption happen here but try to make
1399 	 * sure the thread is rescheduled as soon as possible.  We do
1400 	 * this by putting it on the front of the highest priority run
1401 	 * queue in the TS class.  If the preemption has been put off
1402 	 * for too long, clear the "nopreempt" bit and let the thread
1403 	 * be preempted.
1404 	 */
1405 	if (t->t_schedctl && schedctl_get_nopreempt(t)) {
1406 		if (tspp->ts_timeleft > -SC_MAX_TICKS) {
1407 			DTRACE_SCHED1(schedctl__nopreempt, kthread_t *, t);
1408 			/*
1409 			 * If not already remembered, remember current
1410 			 * priority for restoration in ts_yield().
1411 			 */
1412 			if (!(tspp->ts_flags & TSRESTORE)) {
1413 				tspp->ts_scpri = t->t_pri;
1414 				tspp->ts_flags |= TSRESTORE;
1415 			}
1416 			THREAD_CHANGE_PRI(t, ts_maxumdpri);
1417 			t->t_schedflag |= TS_DONT_SWAP;
1418 			schedctl_set_yield(t, 1);
1419 			setfrontdq(t);
1420 			goto done;
1421 		} else {
1422 			if (tspp->ts_flags & TSRESTORE) {
1423 				THREAD_CHANGE_PRI(t, tspp->ts_scpri);
1424 				tspp->ts_flags &= ~TSRESTORE;
1425 			}
1426 			schedctl_set_nopreempt(t, 0);
1427 			DTRACE_SCHED1(schedctl__preempt, kthread_t *, t);
1428 			TNF_PROBE_2(schedctl_preempt, "schedctl TS ts_preempt",
1429 			    /* CSTYLED */, tnf_pid, pid, ttoproc(t)->p_pid,
1430 			    tnf_lwpid, lwpid, t->t_tid);
1431 			/*
1432 			 * Fall through and be preempted below.
1433 			 */
1434 		}
1435 	}
1436 
1437 	if ((tspp->ts_flags & TSBACKQ) != 0) {
1438 		tspp->ts_timeleft = ts_dptbl[tspp->ts_cpupri].ts_quantum;
1439 		tspp->ts_dispwait = 0;
1440 		tspp->ts_flags &= ~TSBACKQ;
1441 		setbackdq(t);
1442 	} else {
1443 		setfrontdq(t);
1444 	}
1445 
1446 done:
1447 	TRACE_2(TR_FAC_DISP, TR_PREEMPT,
1448 	    "preempt:tid %p old pri %d", t, oldpri);
1449 }
1450 
1451 static void
1452 ts_setrun(kthread_t *t)
1453 {
1454 	tsproc_t *tspp = (tsproc_t *)(t->t_cldata);
1455 
1456 	ASSERT(THREAD_LOCK_HELD(t));	/* t should be in transition */
1457 
1458 	if (tspp->ts_dispwait > ts_dptbl[tspp->ts_umdpri].ts_maxwait) {
1459 		tspp->ts_cpupri = ts_dptbl[tspp->ts_cpupri].ts_slpret;
1460 		TS_NEWUMDPRI(tspp);
1461 		tspp->ts_timeleft = ts_dptbl[tspp->ts_cpupri].ts_quantum;
1462 		tspp->ts_dispwait = 0;
1463 		THREAD_CHANGE_PRI(t, ts_dptbl[tspp->ts_umdpri].ts_globpri);
1464 		ASSERT(t->t_pri >= 0 && t->t_pri <= ts_maxglobpri);
1465 	}
1466 
1467 	tspp->ts_flags &= ~TSBACKQ;
1468 
1469 	if (tspp->ts_flags & TSIA) {
1470 		if (tspp->ts_flags & TSIASET)
1471 			setfrontdq(t);
1472 		else
1473 			setbackdq(t);
1474 	} else {
1475 		if (t->t_disp_time != ddi_get_lbolt())
1476 			setbackdq(t);
1477 		else
1478 			setfrontdq(t);
1479 	}
1480 }
1481 
1482 
1483 /*
1484  * Prepare thread for sleep.
1485  */
1486 static void
1487 ts_sleep(kthread_t *t)
1488 {
1489 	tsproc_t	*tspp = (tsproc_t *)(t->t_cldata);
1490 	pri_t		old_pri = t->t_pri;
1491 
1492 	ASSERT(t == curthread);
1493 	ASSERT(THREAD_LOCK_HELD(t));
1494 
1495 	/*
1496 	 * Account for time spent on CPU before going to sleep.
1497 	 */
1498 	(void) CPUCAPS_CHARGE(t, &tspp->ts_caps, CPUCAPS_CHARGE_ENFORCE);
1499 
1500 	if (tspp->ts_dispwait > ts_dptbl[tspp->ts_umdpri].ts_maxwait) {
1501 		tspp->ts_cpupri = ts_dptbl[tspp->ts_cpupri].ts_slpret;
1502 		TS_NEWUMDPRI(tspp);
1503 		tspp->ts_timeleft = ts_dptbl[tspp->ts_cpupri].ts_quantum;
1504 		tspp->ts_dispwait = 0;
1505 
1506 		THREAD_CHANGE_PRI(curthread,
1507 		    ts_dptbl[tspp->ts_umdpri].ts_globpri);
1508 		ASSERT(curthread->t_pri >= 0 &&
1509 		    curthread->t_pri <= ts_maxglobpri);
1510 
1511 		if (DISP_MUST_SURRENDER(curthread))
1512 			cpu_surrender(curthread);
1513 	}
1514 	t->t_stime = ddi_get_lbolt();		/* time stamp for the swapper */
1515 	TRACE_2(TR_FAC_DISP, TR_SLEEP,
1516 	    "sleep:tid %p old pri %d", t, old_pri);
1517 }
1518 
1519 
1520 /*
1521  * Return Values:
1522  *
1523  *	-1 if the thread is loaded or is not eligible to be swapped in.
1524  *
1525  *	effective priority of the specified thread based on swapout time
1526  *		and size of process (epri >= 0 , epri <= SHRT_MAX).
1527  */
1528 /* ARGSUSED */
1529 static pri_t
1530 ts_swapin(kthread_t *t, int flags)
1531 {
1532 	tsproc_t	*tspp = (tsproc_t *)(t->t_cldata);
1533 	long		epri = -1;
1534 	proc_t		*pp = ttoproc(t);
1535 
1536 	ASSERT(THREAD_LOCK_HELD(t));
1537 
1538 	/*
1539 	 * We know that pri_t is a short.
1540 	 * Be sure not to overrun its range.
1541 	 */
1542 	if (t->t_state == TS_RUN && (t->t_schedflag & TS_LOAD) == 0) {
1543 		time_t swapout_time;
1544 
1545 		swapout_time = (ddi_get_lbolt() - t->t_stime) / hz;
1546 		if (INHERITED(t) || (tspp->ts_flags & TSIASET)) {
1547 			epri = (long)DISP_PRIO(t) + swapout_time;
1548 		} else {
1549 			/*
1550 			 * Threads which have been out for a long time,
1551 			 * have high user mode priority and are associated
1552 			 * with a small address space are more deserving
1553 			 */
1554 			epri = ts_dptbl[tspp->ts_umdpri].ts_globpri;
1555 			ASSERT(epri >= 0 && epri <= ts_maxumdpri);
1556 			epri += swapout_time - pp->p_swrss / nz(maxpgio)/2;
1557 		}
1558 		/*
1559 		 * Scale epri so SHRT_MAX/2 represents zero priority.
1560 		 */
1561 		epri += SHRT_MAX/2;
1562 		if (epri < 0)
1563 			epri = 0;
1564 		else if (epri > SHRT_MAX)
1565 			epri = SHRT_MAX;
1566 	}
1567 	return ((pri_t)epri);
1568 }
1569 
1570 /*
1571  * Return Values
1572  *	-1 if the thread isn't loaded or is not eligible to be swapped out.
1573  *
1574  *	effective priority of the specified thread based on if the swapper
1575  *		is in softswap or hardswap mode.
1576  *
1577  *		Softswap:  Return a low effective priority for threads
1578  *			   sleeping for more than maxslp secs.
1579  *
1580  *		Hardswap:  Return an effective priority such that threads
1581  *			   which have been in memory for a while and are
1582  *			   associated with a small address space are swapped
1583  *			   in before others.
1584  *
1585  *		(epri >= 0 , epri <= SHRT_MAX).
1586  */
1587 time_t	ts_minrun = 2;		/* XXX - t_pri becomes 59 within 2 secs */
1588 time_t	ts_minslp = 2;		/* min time on sleep queue for hardswap */
1589 
1590 static pri_t
1591 ts_swapout(kthread_t *t, int flags)
1592 {
1593 	tsproc_t	*tspp = (tsproc_t *)(t->t_cldata);
1594 	long		epri = -1;
1595 	proc_t		*pp = ttoproc(t);
1596 	time_t		swapin_time;
1597 
1598 	ASSERT(THREAD_LOCK_HELD(t));
1599 
1600 	if (INHERITED(t) || (tspp->ts_flags & TSIASET) ||
1601 	    (t->t_proc_flag & TP_LWPEXIT) ||
1602 	    (t->t_state & (TS_ZOMB | TS_FREE | TS_STOPPED |
1603 	    TS_ONPROC | TS_WAIT)) ||
1604 	    !(t->t_schedflag & TS_LOAD) || !SWAP_OK(t))
1605 		return (-1);
1606 
1607 	ASSERT(t->t_state & (TS_SLEEP | TS_RUN));
1608 
1609 	/*
1610 	 * We know that pri_t is a short.
1611 	 * Be sure not to overrun its range.
1612 	 */
1613 	swapin_time = (ddi_get_lbolt() - t->t_stime) / hz;
1614 	if (flags == SOFTSWAP) {
1615 		if (t->t_state == TS_SLEEP && swapin_time > maxslp) {
1616 			epri = 0;
1617 		} else {
1618 			return ((pri_t)epri);
1619 		}
1620 	} else {
1621 		pri_t pri;
1622 
1623 		if ((t->t_state == TS_SLEEP && swapin_time > ts_minslp) ||
1624 		    (t->t_state == TS_RUN && swapin_time > ts_minrun)) {
1625 			pri = ts_dptbl[tspp->ts_umdpri].ts_globpri;
1626 			ASSERT(pri >= 0 && pri <= ts_maxumdpri);
1627 			epri = swapin_time -
1628 			    (rm_asrss(pp->p_as) / nz(maxpgio)/2) - (long)pri;
1629 		} else {
1630 			return ((pri_t)epri);
1631 		}
1632 	}
1633 
1634 	/*
1635 	 * Scale epri so SHRT_MAX/2 represents zero priority.
1636 	 */
1637 	epri += SHRT_MAX/2;
1638 	if (epri < 0)
1639 		epri = 0;
1640 	else if (epri > SHRT_MAX)
1641 		epri = SHRT_MAX;
1642 
1643 	return ((pri_t)epri);
1644 }
1645 
1646 /*
1647  * Check for time slice expiration.  If time slice has expired
1648  * move thread to priority specified in tsdptbl for time slice expiration
1649  * and set runrun to cause preemption.
1650  */
1651 static void
1652 ts_tick(kthread_t *t)
1653 {
1654 	tsproc_t *tspp = (tsproc_t *)(t->t_cldata);
1655 	klwp_t *lwp;
1656 	boolean_t call_cpu_surrender = B_FALSE;
1657 	pri_t	oldpri = t->t_pri;
1658 
1659 	ASSERT(MUTEX_HELD(&(ttoproc(t))->p_lock));
1660 
1661 	thread_lock(t);
1662 
1663 	/*
1664 	 * Keep track of thread's project CPU usage.  Note that projects
1665 	 * get charged even when threads are running in the kernel.
1666 	 */
1667 	if (CPUCAPS_ON()) {
1668 		call_cpu_surrender = cpucaps_charge(t, &tspp->ts_caps,
1669 		    CPUCAPS_CHARGE_ENFORCE);
1670 	}
1671 
1672 	if (--tspp->ts_timeleft <= 0) {
1673 		pri_t	new_pri;
1674 
1675 		/*
1676 		 * If we're doing preemption control and trying to avoid
1677 		 * preempting this thread, just note that the thread should
1678 		 * yield soon and let it keep running (unless it's been a
1679 		 * while).
1680 		 */
1681 		if (t->t_schedctl && schedctl_get_nopreempt(t)) {
1682 			if (tspp->ts_timeleft > -SC_MAX_TICKS) {
1683 				DTRACE_SCHED1(schedctl__nopreempt,
1684 				    kthread_t *, t);
1685 				schedctl_set_yield(t, 1);
1686 				thread_unlock_nopreempt(t);
1687 				return;
1688 			}
1689 
1690 			TNF_PROBE_2(schedctl_failsafe,
1691 			    "schedctl TS ts_tick", /* CSTYLED */,
1692 			    tnf_pid, pid, ttoproc(t)->p_pid,
1693 			    tnf_lwpid, lwpid, t->t_tid);
1694 		}
1695 		tspp->ts_flags &= ~TSRESTORE;
1696 		tspp->ts_cpupri = ts_dptbl[tspp->ts_cpupri].ts_tqexp;
1697 		TS_NEWUMDPRI(tspp);
1698 		tspp->ts_dispwait = 0;
1699 		new_pri = ts_dptbl[tspp->ts_umdpri].ts_globpri;
1700 		ASSERT(new_pri >= 0 && new_pri <= ts_maxglobpri);
1701 		/*
1702 		 * When the priority of a thread is changed, it may be
1703 		 * necessary to adjust its position on a sleep queue or
1704 		 * dispatch queue.  The function thread_change_pri accomplishes
1705 		 * this.
1706 		 */
1707 		if (thread_change_pri(t, new_pri, 0)) {
1708 			if ((t->t_schedflag & TS_LOAD) &&
1709 			    (lwp = t->t_lwp) &&
1710 			    lwp->lwp_state == LWP_USER)
1711 				t->t_schedflag &= ~TS_DONT_SWAP;
1712 			tspp->ts_timeleft =
1713 			    ts_dptbl[tspp->ts_cpupri].ts_quantum;
1714 		} else {
1715 			call_cpu_surrender = B_TRUE;
1716 		}
1717 		TRACE_2(TR_FAC_DISP, TR_TICK,
1718 		    "tick:tid %p old pri %d", t, oldpri);
1719 	} else if (t->t_state == TS_ONPROC &&
1720 	    t->t_pri < t->t_disp_queue->disp_maxrunpri) {
1721 		call_cpu_surrender = B_TRUE;
1722 	}
1723 
1724 	if (call_cpu_surrender) {
1725 		tspp->ts_flags |= TSBACKQ;
1726 		cpu_surrender(t);
1727 	}
1728 
1729 	thread_unlock_nopreempt(t);	/* clock thread can't be preempted */
1730 }
1731 
1732 
1733 /*
1734  * If we are lowering the thread's priority below that of other runnable
1735  * threads we will normally set runrun via cpu_surrender() to cause preemption.
1736  */
1737 static void
1738 ts_trapret(kthread_t *t)
1739 {
1740 	tsproc_t	*tspp = (tsproc_t *)t->t_cldata;
1741 	cpu_t		*cp = CPU;
1742 	pri_t		old_pri = curthread->t_pri;
1743 
1744 	ASSERT(THREAD_LOCK_HELD(t));
1745 	ASSERT(t == curthread);
1746 	ASSERT(cp->cpu_dispthread == t);
1747 	ASSERT(t->t_state == TS_ONPROC);
1748 
1749 	if (tspp->ts_dispwait > ts_dptbl[tspp->ts_umdpri].ts_maxwait) {
1750 		tspp->ts_cpupri = ts_dptbl[tspp->ts_cpupri].ts_slpret;
1751 		TS_NEWUMDPRI(tspp);
1752 		tspp->ts_timeleft = ts_dptbl[tspp->ts_cpupri].ts_quantum;
1753 		tspp->ts_dispwait = 0;
1754 
1755 		/*
1756 		 * If thread has blocked in the kernel (as opposed to
1757 		 * being merely preempted), recompute the user mode priority.
1758 		 */
1759 		THREAD_CHANGE_PRI(t, ts_dptbl[tspp->ts_umdpri].ts_globpri);
1760 		cp->cpu_dispatch_pri = DISP_PRIO(t);
1761 		ASSERT(t->t_pri >= 0 && t->t_pri <= ts_maxglobpri);
1762 
1763 		if (DISP_MUST_SURRENDER(t))
1764 			cpu_surrender(t);
1765 	}
1766 
1767 	/*
1768 	 * Swapout lwp if the swapper is waiting for this thread to reach a
1769 	 * safe point.
1770 	 */
1771 	if ((t->t_schedflag & TS_SWAPENQ) && !(tspp->ts_flags & TSIASET)) {
1772 		thread_unlock(t);
1773 		swapout_lwp(ttolwp(t));
1774 		thread_lock(t);
1775 	}
1776 
1777 	TRACE_2(TR_FAC_DISP, TR_TRAPRET,
1778 	    "trapret:tid %p old pri %d", t, old_pri);
1779 }
1780 
1781 
1782 /*
1783  * Update the ts_dispwait values of all time sharing threads that
1784  * are currently runnable at a user mode priority and bump the priority
1785  * if ts_dispwait exceeds ts_maxwait.  Called once per second via
1786  * timeout which we reset here.
1787  *
1788  * There are several lists of time sharing threads broken up by a hash on
1789  * the thread pointer.  Each list has its own lock.  This avoids blocking
1790  * all ts_enterclass, ts_fork, and ts_exitclass operations while ts_update
1791  * runs.  ts_update traverses each list in turn.
1792  *
1793  * If multiple threads have their priorities updated to the same value,
1794  * the system implicitly favors the one that is updated first (since it
1795  * winds up first on the run queue).  To avoid this unfairness, the
1796  * traversal of threads starts at the list indicated by a marker.  When
1797  * threads in more than one list have their priorities updated, the marker
1798  * is moved.  This changes the order the threads will be placed on the run
1799  * queue the next time ts_update is called and preserves fairness over the
1800  * long run.  The marker doesn't need to be protected by a lock since it's
1801  * only accessed by ts_update, which is inherently single-threaded (only
1802  * one instance can be running at a time).
1803  */
1804 static void
1805 ts_update(void *arg)
1806 {
1807 	int		i;
1808 	int		new_marker = -1;
1809 	static int	ts_update_marker;
1810 
1811 	/*
1812 	 * Start with the ts_update_marker list, then do the rest.
1813 	 */
1814 	i = ts_update_marker;
1815 	do {
1816 		/*
1817 		 * If this is the first list after the current marker to
1818 		 * have threads with priorities updated, advance the marker
1819 		 * to this list for the next time ts_update runs.
1820 		 */
1821 		if (ts_update_list(i) && new_marker == -1 &&
1822 		    i != ts_update_marker) {
1823 			new_marker = i;
1824 		}
1825 	} while ((i = TS_LIST_NEXT(i)) != ts_update_marker);
1826 
1827 	/* advance marker for next ts_update call */
1828 	if (new_marker != -1)
1829 		ts_update_marker = new_marker;
1830 
1831 	(void) timeout(ts_update, arg, hz);
1832 }
1833 
1834 /*
1835  * Updates priority for a list of threads.  Returns 1 if the priority of
1836  * one of the threads was actually updated, 0 if none were for various
1837  * reasons (thread is no longer in the TS or IA class, isn't runnable,
1838  * hasn't waited long enough, has the preemption control no-preempt bit
1839  * set, etc.)
1840  */
1841 static int
1842 ts_update_list(int i)
1843 {
1844 	tsproc_t *tspp;
1845 	kthread_t *tx;
1846 	int updated = 0;
1847 
1848 	mutex_enter(&ts_list_lock[i]);
1849 	for (tspp = ts_plisthead[i].ts_next; tspp != &ts_plisthead[i];
1850 	    tspp = tspp->ts_next) {
1851 		tx = tspp->ts_tp;
1852 		/*
1853 		 * Lock the thread and verify state.
1854 		 */
1855 		thread_lock(tx);
1856 		/*
1857 		 * Skip the thread if it is no longer in the TS (or IA) class.
1858 		 */
1859 		if (tx->t_clfuncs != &ts_classfuncs.thread &&
1860 		    tx->t_clfuncs != &ia_classfuncs.thread)
1861 			goto next;
1862 		tspp->ts_dispwait++;
1863 		if (tspp->ts_dispwait <= ts_dptbl[tspp->ts_umdpri].ts_maxwait)
1864 			goto next;
1865 		if (tx->t_schedctl && schedctl_get_nopreempt(tx))
1866 			goto next;
1867 		if (tx->t_state != TS_RUN && tx->t_state != TS_WAIT &&
1868 		    (tx->t_state != TS_SLEEP || !ts_sleep_promote)) {
1869 			/* make next syscall/trap do CL_TRAPRET */
1870 			tx->t_trapret = 1;
1871 			aston(tx);
1872 			goto next;
1873 		}
1874 		tspp->ts_cpupri = ts_dptbl[tspp->ts_cpupri].ts_lwait;
1875 		TS_NEWUMDPRI(tspp);
1876 		tspp->ts_dispwait = 0;
1877 		updated = 1;
1878 
1879 		/*
1880 		 * Only dequeue it if needs to move; otherwise it should
1881 		 * just round-robin here.
1882 		 */
1883 		if (tx->t_pri != ts_dptbl[tspp->ts_umdpri].ts_globpri) {
1884 			pri_t oldpri = tx->t_pri;
1885 			ts_change_priority(tx, tspp);
1886 			TRACE_2(TR_FAC_DISP, TR_UPDATE,
1887 			    "update:tid %p old pri %d", tx, oldpri);
1888 		}
1889 next:
1890 		thread_unlock(tx);
1891 	}
1892 	mutex_exit(&ts_list_lock[i]);
1893 
1894 	return (updated);
1895 }
1896 
1897 /*
1898  * Processes waking up go to the back of their queue.
1899  */
1900 static void
1901 ts_wakeup(kthread_t *t)
1902 {
1903 	tsproc_t	*tspp = (tsproc_t *)(t->t_cldata);
1904 
1905 	ASSERT(THREAD_LOCK_HELD(t));
1906 
1907 	t->t_stime = ddi_get_lbolt();		/* time stamp for the swapper */
1908 
1909 	if (tspp->ts_dispwait > ts_dptbl[tspp->ts_umdpri].ts_maxwait) {
1910 		tspp->ts_cpupri = ts_dptbl[tspp->ts_cpupri].ts_slpret;
1911 		TS_NEWUMDPRI(tspp);
1912 		tspp->ts_timeleft = ts_dptbl[tspp->ts_cpupri].ts_quantum;
1913 		tspp->ts_dispwait = 0;
1914 		THREAD_CHANGE_PRI(t, ts_dptbl[tspp->ts_umdpri].ts_globpri);
1915 		ASSERT(t->t_pri >= 0 && t->t_pri <= ts_maxglobpri);
1916 	}
1917 
1918 	tspp->ts_flags &= ~TSBACKQ;
1919 
1920 	if (tspp->ts_flags & TSIA) {
1921 		if (tspp->ts_flags & TSIASET)
1922 			setfrontdq(t);
1923 		else
1924 			setbackdq(t);
1925 	} else {
1926 		if (t->t_disp_time != ddi_get_lbolt())
1927 			setbackdq(t);
1928 		else
1929 			setfrontdq(t);
1930 	}
1931 }
1932 
1933 
1934 /*
1935  * When a thread yields, put it on the back of the run queue.
1936  */
1937 static void
1938 ts_yield(kthread_t *t)
1939 {
1940 	tsproc_t	*tspp = (tsproc_t *)(t->t_cldata);
1941 
1942 	ASSERT(t == curthread);
1943 	ASSERT(THREAD_LOCK_HELD(t));
1944 
1945 	/*
1946 	 * Collect CPU usage spent before yielding
1947 	 */
1948 	(void) CPUCAPS_CHARGE(t, &tspp->ts_caps, CPUCAPS_CHARGE_ENFORCE);
1949 
1950 	/*
1951 	 * Clear the preemption control "yield" bit since the user is
1952 	 * doing a yield.
1953 	 */
1954 	if (t->t_schedctl)
1955 		schedctl_set_yield(t, 0);
1956 	/*
1957 	 * If ts_preempt() artifically increased the thread's priority
1958 	 * to avoid preemption, restore the original priority now.
1959 	 */
1960 	if (tspp->ts_flags & TSRESTORE) {
1961 		THREAD_CHANGE_PRI(t, tspp->ts_scpri);
1962 		tspp->ts_flags &= ~TSRESTORE;
1963 	}
1964 	if (tspp->ts_timeleft <= 0) {
1965 		/*
1966 		 * Time slice was artificially extended to avoid
1967 		 * preemption, so pretend we're preempting it now.
1968 		 */
1969 		DTRACE_SCHED1(schedctl__yield, int, -tspp->ts_timeleft);
1970 		tspp->ts_cpupri = ts_dptbl[tspp->ts_cpupri].ts_tqexp;
1971 		TS_NEWUMDPRI(tspp);
1972 		tspp->ts_timeleft = ts_dptbl[tspp->ts_cpupri].ts_quantum;
1973 		tspp->ts_dispwait = 0;
1974 		THREAD_CHANGE_PRI(t, ts_dptbl[tspp->ts_umdpri].ts_globpri);
1975 		ASSERT(t->t_pri >= 0 && t->t_pri <= ts_maxglobpri);
1976 	}
1977 	tspp->ts_flags &= ~TSBACKQ;
1978 	setbackdq(t);
1979 }
1980 
1981 
1982 /*
1983  * Increment the nice value of the specified thread by incr and
1984  * return the new value in *retvalp.
1985  */
1986 static int
1987 ts_donice(kthread_t *t, cred_t *cr, int incr, int *retvalp)
1988 {
1989 	int		newnice;
1990 	tsproc_t	*tspp = (tsproc_t *)(t->t_cldata);
1991 	tsparms_t	tsparms;
1992 
1993 	ASSERT(MUTEX_HELD(&(ttoproc(t))->p_lock));
1994 
1995 	/* If there's no change to priority, just return current setting */
1996 	if (incr == 0) {
1997 		if (retvalp) {
1998 			*retvalp = tspp->ts_nice - NZERO;
1999 		}
2000 		return (0);
2001 	}
2002 
2003 	if ((incr < 0 || incr > 2 * NZERO) &&
2004 	    secpolicy_raisepriority(cr) != 0)
2005 		return (EPERM);
2006 
2007 	/*
2008 	 * Specifying a nice increment greater than the upper limit of
2009 	 * 2 * NZERO - 1 will result in the thread's nice value being
2010 	 * set to the upper limit.  We check for this before computing
2011 	 * the new value because otherwise we could get overflow
2012 	 * if a privileged process specified some ridiculous increment.
2013 	 */
2014 	if (incr > 2 * NZERO - 1)
2015 		incr = 2 * NZERO - 1;
2016 
2017 	newnice = tspp->ts_nice + incr;
2018 	if (newnice >= 2 * NZERO)
2019 		newnice = 2 * NZERO - 1;
2020 	else if (newnice < 0)
2021 		newnice = 0;
2022 
2023 	tsparms.ts_uprilim = tsparms.ts_upri =
2024 	    -((newnice - NZERO) * ts_maxupri) / NZERO;
2025 	/*
2026 	 * Reset the uprilim and upri values of the thread.
2027 	 * Call ts_parmsset even if thread is interactive since we're
2028 	 * not changing mode.
2029 	 */
2030 	(void) ts_parmsset(t, (void *)&tsparms, (id_t)0, (cred_t *)NULL);
2031 
2032 	/*
2033 	 * Although ts_parmsset already reset ts_nice it may
2034 	 * not have been set to precisely the value calculated above
2035 	 * because ts_parmsset determines the nice value from the
2036 	 * user priority and we may have truncated during the integer
2037 	 * conversion from nice value to user priority and back.
2038 	 * We reset ts_nice to the value we calculated above.
2039 	 */
2040 	tspp->ts_nice = (char)newnice;
2041 
2042 	if (retvalp)
2043 		*retvalp = newnice - NZERO;
2044 	return (0);
2045 }
2046 
2047 /*
2048  * Increment the priority of the specified thread by incr and
2049  * return the new value in *retvalp.
2050  */
2051 static int
2052 ts_doprio(kthread_t *t, cred_t *cr, int incr, int *retvalp)
2053 {
2054 	int		newpri;
2055 	tsproc_t	*tspp = (tsproc_t *)(t->t_cldata);
2056 	tsparms_t	tsparms;
2057 
2058 	ASSERT(MUTEX_HELD(&(ttoproc(t))->p_lock));
2059 
2060 	/* If there's no change to the priority, just return current setting */
2061 	if (incr == 0) {
2062 		*retvalp = tspp->ts_upri;
2063 		return (0);
2064 	}
2065 
2066 	newpri = tspp->ts_upri + incr;
2067 	if (newpri > ts_maxupri || newpri < -ts_maxupri)
2068 		return (EINVAL);
2069 
2070 	*retvalp = newpri;
2071 	tsparms.ts_uprilim = tsparms.ts_upri = newpri;
2072 	/*
2073 	 * Reset the uprilim and upri values of the thread.
2074 	 * Call ts_parmsset even if thread is interactive since we're
2075 	 * not changing mode.
2076 	 */
2077 	return (ts_parmsset(t, &tsparms, 0, cr));
2078 }
2079 
2080 /*
2081  * ia_set_process_group marks foreground processes as interactive
2082  * and background processes as non-interactive iff the session
2083  * leader is interactive.  This routine is called from two places:
2084  *	strioctl:SPGRP when a new process group gets
2085  *		control of the tty.
2086  *	ia_parmsset-when the process in question is a session leader.
2087  * ia_set_process_group assumes that pidlock is held by the caller,
2088  * either strioctl or priocntlsys.  If the caller is priocntlsys
2089  * (via ia_parmsset) then the p_lock of the session leader is held
2090  * and the code needs to be careful about acquiring other p_locks.
2091  */
2092 static void
2093 ia_set_process_group(pid_t sid, pid_t bg_pgid, pid_t fg_pgid)
2094 {
2095 	proc_t		*leader, *fg, *bg;
2096 	tsproc_t	*tspp;
2097 	kthread_t	*tx;
2098 	int		plocked = 0;
2099 
2100 	ASSERT(MUTEX_HELD(&pidlock));
2101 
2102 	/*
2103 	 * see if the session leader is interactive AND
2104 	 * if it is currently "on" AND controlling a tty
2105 	 * iff it is then make the processes in the foreground
2106 	 * group interactive and the processes in the background
2107 	 * group non-interactive.
2108 	 */
2109 	if ((leader = (proc_t *)prfind(sid)) == NULL) {
2110 		return;
2111 	}
2112 	if (leader->p_stat == SIDL) {
2113 		return;
2114 	}
2115 	if ((tx = proctot(leader)) == NULL) {
2116 		return;
2117 	}
2118 	/*
2119 	 * XXX do all the threads in the leader
2120 	 */
2121 	if (tx->t_cid != ia_cid) {
2122 		return;
2123 	}
2124 	tspp = tx->t_cldata;
2125 	/*
2126 	 * session leaders that are not interactive need not have
2127 	 * any processing done for them.  They are typically shells
2128 	 * that do not have focus and are changing the process group
2129 	 * attatched to the tty, e.g. a process that is exiting
2130 	 */
2131 	mutex_enter(&leader->p_sessp->s_lock);
2132 	if (!(tspp->ts_flags & TSIASET) ||
2133 	    (leader->p_sessp->s_vp == NULL) ||
2134 	    (leader->p_sessp->s_vp->v_stream == NULL)) {
2135 		mutex_exit(&leader->p_sessp->s_lock);
2136 		return;
2137 	}
2138 	mutex_exit(&leader->p_sessp->s_lock);
2139 
2140 	/*
2141 	 * If we're already holding the leader's p_lock, we should use
2142 	 * mutex_tryenter instead of mutex_enter to avoid deadlocks from
2143 	 * lock ordering violations.
2144 	 */
2145 	if (mutex_owned(&leader->p_lock))
2146 		plocked = 1;
2147 
2148 	if (fg_pgid == 0)
2149 		goto skip;
2150 	/*
2151 	 * now look for all processes in the foreground group and
2152 	 * make them interactive
2153 	 */
2154 	for (fg = (proc_t *)pgfind(fg_pgid); fg != NULL; fg = fg->p_pglink) {
2155 		/*
2156 		 * if the process is SIDL it's begin forked, ignore it
2157 		 */
2158 		if (fg->p_stat == SIDL) {
2159 			continue;
2160 		}
2161 		/*
2162 		 * sesssion leaders must be turned on/off explicitly
2163 		 * not implicitly as happens to other members of
2164 		 * the process group.
2165 		 */
2166 		if (fg->p_pid  == fg->p_sessp->s_sid) {
2167 			continue;
2168 		}
2169 
2170 		TRACE_1(TR_FAC_IA, TR_GROUP_ON,
2171 		    "group on:proc %p", fg);
2172 
2173 		if (plocked) {
2174 			if (mutex_tryenter(&fg->p_lock) == 0)
2175 				continue;
2176 		} else {
2177 			mutex_enter(&fg->p_lock);
2178 		}
2179 
2180 		if ((tx = proctot(fg)) == NULL) {
2181 			mutex_exit(&fg->p_lock);
2182 			continue;
2183 		}
2184 		do {
2185 			thread_lock(tx);
2186 			/*
2187 			 * if this thread is not interactive continue
2188 			 */
2189 			if (tx->t_cid != ia_cid) {
2190 				thread_unlock(tx);
2191 				continue;
2192 			}
2193 			tspp = tx->t_cldata;
2194 			tspp->ts_flags |= TSIASET;
2195 			tspp->ts_boost = ia_boost;
2196 			TS_NEWUMDPRI(tspp);
2197 			tspp->ts_dispwait = 0;
2198 			ts_change_priority(tx, tspp);
2199 			thread_unlock(tx);
2200 		} while ((tx = tx->t_forw) != fg->p_tlist);
2201 		mutex_exit(&fg->p_lock);
2202 	}
2203 skip:
2204 	if (bg_pgid == 0)
2205 		return;
2206 	for (bg = (proc_t *)pgfind(bg_pgid); bg != NULL; bg = bg->p_pglink) {
2207 		if (bg->p_stat == SIDL) {
2208 			continue;
2209 		}
2210 		/*
2211 		 * sesssion leaders must be turned off explicitly
2212 		 * not implicitly as happens to other members of
2213 		 * the process group.
2214 		 */
2215 		if (bg->p_pid == bg->p_sessp->s_sid) {
2216 			continue;
2217 		}
2218 
2219 		TRACE_1(TR_FAC_IA, TR_GROUP_OFF,
2220 		    "group off:proc %p", bg);
2221 
2222 		if (plocked) {
2223 			if (mutex_tryenter(&bg->p_lock) == 0)
2224 				continue;
2225 		} else {
2226 			mutex_enter(&bg->p_lock);
2227 		}
2228 
2229 		if ((tx = proctot(bg)) == NULL) {
2230 			mutex_exit(&bg->p_lock);
2231 			continue;
2232 		}
2233 		do {
2234 			thread_lock(tx);
2235 			/*
2236 			 * if this thread is not interactive continue
2237 			 */
2238 			if (tx->t_cid != ia_cid) {
2239 				thread_unlock(tx);
2240 				continue;
2241 			}
2242 			tspp = tx->t_cldata;
2243 			tspp->ts_flags &= ~TSIASET;
2244 			tspp->ts_boost = -ia_boost;
2245 			TS_NEWUMDPRI(tspp);
2246 
2247 			tspp->ts_dispwait = 0;
2248 			ts_change_priority(tx, tspp);
2249 			thread_unlock(tx);
2250 		} while ((tx = tx->t_forw) != bg->p_tlist);
2251 		mutex_exit(&bg->p_lock);
2252 	}
2253 }
2254 
2255 
2256 static void
2257 ts_change_priority(kthread_t *t, tsproc_t *tspp)
2258 {
2259 	pri_t	new_pri;
2260 
2261 	ASSERT(THREAD_LOCK_HELD(t));
2262 	new_pri = ts_dptbl[tspp->ts_umdpri].ts_globpri;
2263 	ASSERT(new_pri >= 0 && new_pri <= ts_maxglobpri);
2264 	tspp->ts_flags &= ~TSRESTORE;
2265 	t->t_cpri = tspp->ts_upri;
2266 	if (t == curthread || t->t_state == TS_ONPROC) {
2267 		/* curthread is always onproc */
2268 		cpu_t	*cp = t->t_disp_queue->disp_cpu;
2269 		THREAD_CHANGE_PRI(t, new_pri);
2270 		if (t == cp->cpu_dispthread)
2271 			cp->cpu_dispatch_pri = DISP_PRIO(t);
2272 		if (DISP_MUST_SURRENDER(t)) {
2273 			tspp->ts_flags |= TSBACKQ;
2274 			cpu_surrender(t);
2275 		} else {
2276 			tspp->ts_timeleft =
2277 			    ts_dptbl[tspp->ts_cpupri].ts_quantum;
2278 		}
2279 	} else {
2280 		int	frontq;
2281 
2282 		frontq = (tspp->ts_flags & TSIASET) != 0;
2283 		/*
2284 		 * When the priority of a thread is changed,
2285 		 * it may be necessary to adjust its position
2286 		 * on a sleep queue or dispatch queue.
2287 		 * The function thread_change_pri accomplishes
2288 		 * this.
2289 		 */
2290 		if (thread_change_pri(t, new_pri, frontq)) {
2291 			/*
2292 			 * The thread was on a run queue. Reset
2293 			 * its CPU timeleft from the quantum
2294 			 * associated with the new priority.
2295 			 */
2296 			tspp->ts_timeleft =
2297 			    ts_dptbl[tspp->ts_cpupri].ts_quantum;
2298 		} else {
2299 			tspp->ts_flags |= TSBACKQ;
2300 		}
2301 	}
2302 }
2303 
2304 static int
2305 ts_alloc(void **p, int flag)
2306 {
2307 	void *bufp;
2308 	bufp = kmem_alloc(sizeof (tsproc_t), flag);
2309 	if (bufp == NULL) {
2310 		return (ENOMEM);
2311 	} else {
2312 		*p = bufp;
2313 		return (0);
2314 	}
2315 }
2316 
2317 static void
2318 ts_free(void *bufp)
2319 {
2320 	if (bufp)
2321 		kmem_free(bufp, sizeof (tsproc_t));
2322 }
2323