xref: /illumos-gate/usr/src/uts/common/disp/fx.c (revision 355b4669e025ff377602b6fc7caaf30dbc218371)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
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 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/sysmacros.h>
32 #include <sys/cred.h>
33 #include <sys/proc.h>
34 #include <sys/session.h>
35 #include <sys/strsubr.h>
36 #include <sys/user.h>
37 #include <sys/priocntl.h>
38 #include <sys/class.h>
39 #include <sys/disp.h>
40 #include <sys/procset.h>
41 #include <sys/debug.h>
42 #include <sys/kmem.h>
43 #include <sys/errno.h>
44 #include <sys/fx.h>
45 #include <sys/fxpriocntl.h>
46 #include <sys/cpuvar.h>
47 #include <sys/systm.h>
48 #include <sys/vtrace.h>
49 #include <sys/schedctl.h>
50 #include <sys/tnf_probe.h>
51 #include <sys/sunddi.h>
52 #include <sys/spl.h>
53 #include <sys/modctl.h>
54 #include <sys/policy.h>
55 #include <sys/sdt.h>
56 #include <sys/cpupart.h>
57 
58 static pri_t fx_init(id_t, int, classfuncs_t **);
59 
60 static struct sclass csw = {
61 	"FX",
62 	fx_init,
63 	0
64 };
65 
66 static struct modlsched modlsched = {
67 	&mod_schedops, "Fixed priority sched class", &csw
68 };
69 
70 static struct modlinkage modlinkage = {
71 	MODREV_1, (void *)&modlsched, NULL
72 };
73 
74 
75 /*
76  * control flags (kparms->fx_cflags).
77  */
78 #define	FX_DOUPRILIM	0x01    /* change user priority limit */
79 #define	FX_DOUPRI	0x02    /* change user priority */
80 #define	FX_DOTQ		0x04    /* change FX time quantum */
81 
82 
83 #define	FXMAXUPRI 60		/* maximum user priority setting */
84 
85 #define	FX_MAX_UNPRIV_PRI	0	/* maximum unpriviledge priority */
86 
87 /*
88  * The fxproc_t structures are kept in an array of circular doubly linked
89  * lists. A hash on the thread pointer is used to determine which list
90  * each fxproc structure should be placed. Each list has a dummy "head" which
91  * is never removed, so the list is never empty.
92  */
93 
94 #define	FX_LISTS 16		/* number of lists, must be power of 2 */
95 #define	FX_LIST_HASH(tp)	(((uintptr_t)(tp) >> 9) & (FX_LISTS - 1))
96 
97 #define	FX_LIST_INSERT(fxpp)						\
98 {									\
99 	int index = FX_LIST_HASH(fxpp->fx_tp);				\
100 	kmutex_t *lockp = &fx_list_lock[index];				\
101 	fxproc_t *headp = &fx_plisthead[index];				\
102 	mutex_enter(lockp);						\
103 	fxpp->fx_next = headp->fx_next;					\
104 	fxpp->fx_prev = headp;						\
105 	headp->fx_next->fx_prev = fxpp;					\
106 	headp->fx_next = fxpp;						\
107 	mutex_exit(lockp);						\
108 }
109 
110 #define	FX_LIST_DELETE(fxpp)						\
111 {									\
112 	int index = FX_LIST_HASH(fxpp->fx_tp);				\
113 	kmutex_t *lockp = &fx_list_lock[index];				\
114 	mutex_enter(lockp);						\
115 	fxpp->fx_prev->fx_next = fxpp->fx_next;				\
116 	fxpp->fx_next->fx_prev = fxpp->fx_prev;				\
117 	mutex_exit(lockp);						\
118 }
119 
120 
121 /*
122  * The fxproc_t structures that have a registered callback vector,
123  * are also kept in an array of circular doubly linked lists. A hash on
124  * the thread id (from ddi_get_kt_did()) is used to determine which list
125  * each of such fxproc structures should be placed. Each list has a dummy
126  * "head" which is never removed, so the list is never empty.
127  */
128 
129 #define	FX_CB_LISTS 16		/* number of lists, must be power of 2 */
130 #define	FX_CB_LIST_HASH(ktid)	((uint_t)ktid & (FX_CB_LISTS - 1))
131 
132 /* Insert fxproc into callback list */
133 #define	FX_CB_LIST_INSERT(fxpp)						\
134 {									\
135 	int index = FX_CB_LIST_HASH(fxpp->fx_ktid);			\
136 	kmutex_t *lockp = &fx_cb_list_lock[index];			\
137 	fxproc_t *headp = &fx_cb_plisthead[index];			\
138 	mutex_enter(lockp);						\
139 	fxpp->fx_cb_next = headp->fx_cb_next;				\
140 	fxpp->fx_cb_prev = headp;					\
141 	headp->fx_cb_next->fx_cb_prev = fxpp;				\
142 	headp->fx_cb_next = fxpp;					\
143 	mutex_exit(lockp);						\
144 }
145 
146 /*
147  * Remove thread from callback list.
148  */
149 #define	FX_CB_LIST_DELETE(fxpp)						\
150 {									\
151 	int index = FX_CB_LIST_HASH(fxpp->fx_ktid);			\
152 	kmutex_t *lockp = &fx_cb_list_lock[index];			\
153 	mutex_enter(lockp);						\
154 	fxpp->fx_cb_prev->fx_cb_next = fxpp->fx_cb_next;		\
155 	fxpp->fx_cb_next->fx_cb_prev = fxpp->fx_cb_prev;		\
156 	mutex_exit(lockp);						\
157 }
158 
159 #define	FX_HAS_CB(fxpp)	(fxpp->fx_callback != NULL)
160 
161 /* adjust x to be between 0 and fx_maxumdpri */
162 
163 #define	FX_ADJUST_PRI(pri)						\
164 {									\
165 	if (pri < 0)							\
166 		pri = 0;  						\
167 	else if (pri > fx_maxumdpri) 					\
168 		pri = fx_maxumdpri;  					\
169 }
170 
171 #define	FX_ADJUST_QUANTUM(q)						\
172 {									\
173 	if (q > INT_MAX)						\
174 		q = INT_MAX;						\
175 	else if (q <= 0)						\
176 		q = FX_TQINF;						\
177 }
178 
179 #define	FX_ISVALID(pri, quantum) \
180 	(((pri >= 0) || (pri == FX_CB_NOCHANGE)) &&			\
181 	    ((quantum >= 0) || (quantum == FX_NOCHANGE) ||		\
182 		(quantum == FX_TQDEF) || (quantum == FX_TQINF)))
183 
184 
185 static id_t	fx_cid;		/* fixed priority class ID */
186 static fxdpent_t *fx_dptbl;	/* fixed priority disp parameter table */
187 
188 static pri_t	fx_maxupri = FXMAXUPRI;
189 static pri_t	fx_maxumdpri;	/* max user mode fixed priority */
190 
191 static pri_t	fx_maxglobpri;	/* maximum global priority used by fx class */
192 static kmutex_t	fx_dptblock;	/* protects fixed priority dispatch table */
193 
194 
195 static kmutex_t	fx_list_lock[FX_LISTS];	/* protects fxproc lists */
196 static fxproc_t	fx_plisthead[FX_LISTS];	/* dummy fxproc at head of lists */
197 
198 
199 static kmutex_t	fx_cb_list_lock[FX_CB_LISTS];	/* protects list of fxprocs */
200 						/* that have callbacks */
201 static fxproc_t	fx_cb_plisthead[FX_CB_LISTS];	/* dummy fxproc at head of */
202 						/* list of fxprocs with */
203 						/* callbacks */
204 
205 static int	fx_admin(caddr_t, cred_t *);
206 static int	fx_getclinfo(void *);
207 static int	fx_parmsin(void *);
208 static int	fx_parmsout(void *, pc_vaparms_t *);
209 static int	fx_vaparmsin(void *, pc_vaparms_t *);
210 static int	fx_vaparmsout(void *, pc_vaparms_t *);
211 static int	fx_getclpri(pcpri_t *);
212 static int	fx_alloc(void **, int);
213 static void	fx_free(void *);
214 static int	fx_enterclass(kthread_t *, id_t, void *, cred_t *, void *);
215 static void	fx_exitclass(void *);
216 static int	fx_canexit(kthread_t *, cred_t *);
217 static int	fx_fork(kthread_t *, kthread_t *, void *);
218 static void	fx_forkret(kthread_t *, kthread_t *);
219 static void	fx_parmsget(kthread_t *, void *);
220 static int	fx_parmsset(kthread_t *, void *, id_t, cred_t *);
221 static void	fx_stop(kthread_t *, int, int);
222 static void	fx_exit(kthread_t *);
223 static pri_t	fx_swapin(kthread_t *, int);
224 static pri_t	fx_swapout(kthread_t *, int);
225 static void	fx_trapret(kthread_t *);
226 static void	fx_preempt(kthread_t *);
227 static void	fx_setrun(kthread_t *);
228 static void	fx_sleep(kthread_t *);
229 static void	fx_tick(kthread_t *);
230 static void	fx_wakeup(kthread_t *);
231 static int	fx_donice(kthread_t *, cred_t *, int, int *);
232 static pri_t	fx_globpri(kthread_t *);
233 static void	fx_yield(kthread_t *);
234 static void	fx_nullsys();
235 
236 extern fxdpent_t *fx_getdptbl(void);
237 
238 static void	fx_change_priority(kthread_t *, fxproc_t *);
239 static fxproc_t *fx_list_lookup(kt_did_t);
240 static void fx_list_release(fxproc_t *);
241 
242 
243 static struct classfuncs fx_classfuncs = {
244 	/* class functions */
245 	fx_admin,
246 	fx_getclinfo,
247 	fx_parmsin,
248 	fx_parmsout,
249 	fx_vaparmsin,
250 	fx_vaparmsout,
251 	fx_getclpri,
252 	fx_alloc,
253 	fx_free,
254 
255 	/* thread functions */
256 	fx_enterclass,
257 	fx_exitclass,
258 	fx_canexit,
259 	fx_fork,
260 	fx_forkret,
261 	fx_parmsget,
262 	fx_parmsset,
263 	fx_stop,
264 	fx_exit,
265 	fx_nullsys,	/* active */
266 	fx_nullsys,	/* inactive */
267 	fx_swapin,
268 	fx_swapout,
269 	fx_trapret,
270 	fx_preempt,
271 	fx_setrun,
272 	fx_sleep,
273 	fx_tick,
274 	fx_wakeup,
275 	fx_donice,
276 	fx_globpri,
277 	fx_nullsys,	/* set_process_group */
278 	fx_yield,
279 };
280 
281 
282 int
283 _init()
284 {
285 	return (mod_install(&modlinkage));
286 }
287 
288 int
289 _fini()
290 {
291 	return (EBUSY);
292 }
293 
294 int
295 _info(struct modinfo *modinfop)
296 {
297 	return (mod_info(&modlinkage, modinfop));
298 }
299 
300 /*
301  * Fixed priority class initialization. Called by dispinit() at boot time.
302  * We can ignore the clparmsz argument since we know that the smallest
303  * possible parameter buffer is big enough for us.
304  */
305 /* ARGSUSED */
306 static pri_t
307 fx_init(id_t cid, int clparmsz, classfuncs_t **clfuncspp)
308 {
309 	int i;
310 	extern pri_t fx_getmaxumdpri(void);
311 
312 	fx_dptbl = fx_getdptbl();
313 	fx_maxumdpri = fx_getmaxumdpri();
314 	fx_maxglobpri = fx_dptbl[fx_maxumdpri].fx_globpri;
315 
316 	fx_cid = cid;		/* Record our class ID */
317 
318 	/*
319 	 * Initialize the fxproc hash table
320 	 */
321 	for (i = 0; i < FX_LISTS; i++) {
322 		fx_plisthead[i].fx_next = fx_plisthead[i].fx_prev =
323 		    &fx_plisthead[i];
324 	}
325 
326 	/*
327 	 * Initialize the hash table for fxprocs with callbacks
328 	 */
329 	for (i = 0; i < FX_CB_LISTS; i++) {
330 		fx_cb_plisthead[i].fx_cb_next = fx_cb_plisthead[i].fx_cb_prev =
331 			&fx_cb_plisthead[i];
332 	}
333 
334 	/*
335 	 * We're required to return a pointer to our classfuncs
336 	 * structure and the highest global priority value we use.
337 	 */
338 	*clfuncspp = &fx_classfuncs;
339 	return (fx_maxglobpri);
340 }
341 
342 /*
343  * Get or reset the fx_dptbl values per the user's request.
344  */
345 static int
346 fx_admin(caddr_t uaddr, cred_t *reqpcredp)
347 {
348 	fxadmin_t	fxadmin;
349 	fxdpent_t	*tmpdpp;
350 	int		userdpsz;
351 	int		i;
352 	size_t		fxdpsz;
353 
354 	if (get_udatamodel() == DATAMODEL_NATIVE) {
355 		if (copyin(uaddr, &fxadmin, sizeof (fxadmin_t)))
356 			return (EFAULT);
357 	}
358 #ifdef _SYSCALL32_IMPL
359 	else {
360 		/* get fxadmin struct from ILP32 caller */
361 		fxadmin32_t fxadmin32;
362 		if (copyin(uaddr, &fxadmin32, sizeof (fxadmin32_t)))
363 			return (EFAULT);
364 		fxadmin.fx_dpents =
365 		    (struct fxdpent *)(uintptr_t)fxadmin32.fx_dpents;
366 		fxadmin.fx_ndpents = fxadmin32.fx_ndpents;
367 		fxadmin.fx_cmd = fxadmin32.fx_cmd;
368 	}
369 #endif /* _SYSCALL32_IMPL */
370 
371 	fxdpsz = (fx_maxumdpri + 1) * sizeof (fxdpent_t);
372 
373 	switch (fxadmin.fx_cmd) {
374 	case FX_GETDPSIZE:
375 		fxadmin.fx_ndpents = fx_maxumdpri + 1;
376 
377 		if (get_udatamodel() == DATAMODEL_NATIVE) {
378 			if (copyout(&fxadmin, uaddr, sizeof (fxadmin_t)))
379 				return (EFAULT);
380 		}
381 #ifdef _SYSCALL32_IMPL
382 		else {
383 			/* return fxadmin struct to ILP32 caller */
384 			fxadmin32_t fxadmin32;
385 			fxadmin32.fx_dpents =
386 			    (caddr32_t)(uintptr_t)fxadmin.fx_dpents;
387 			fxadmin32.fx_ndpents = fxadmin.fx_ndpents;
388 			fxadmin32.fx_cmd = fxadmin.fx_cmd;
389 			if (copyout(&fxadmin32, uaddr, sizeof (fxadmin32_t)))
390 				return (EFAULT);
391 		}
392 #endif /* _SYSCALL32_IMPL */
393 		break;
394 
395 	case FX_GETDPTBL:
396 		userdpsz = MIN(fxadmin.fx_ndpents * sizeof (fxdpent_t),
397 		    fxdpsz);
398 		if (copyout(fx_dptbl, fxadmin.fx_dpents, userdpsz))
399 			return (EFAULT);
400 
401 		fxadmin.fx_ndpents = userdpsz / sizeof (fxdpent_t);
402 
403 		if (get_udatamodel() == DATAMODEL_NATIVE) {
404 			if (copyout(&fxadmin, uaddr, sizeof (fxadmin_t)))
405 				return (EFAULT);
406 		}
407 #ifdef _SYSCALL32_IMPL
408 		else {
409 			/* return fxadmin struct to ILP32 callers */
410 			fxadmin32_t fxadmin32;
411 			fxadmin32.fx_dpents =
412 			    (caddr32_t)(uintptr_t)fxadmin.fx_dpents;
413 			fxadmin32.fx_ndpents = fxadmin.fx_ndpents;
414 			fxadmin32.fx_cmd = fxadmin.fx_cmd;
415 			if (copyout(&fxadmin32, uaddr, sizeof (fxadmin32_t)))
416 				return (EFAULT);
417 		}
418 #endif /* _SYSCALL32_IMPL */
419 		break;
420 
421 	case FX_SETDPTBL:
422 		/*
423 		 * We require that the requesting process has sufficient
424 		 * privileges. We also require that the table supplied by
425 		 * the user exactly match the current fx_dptbl in size.
426 		 */
427 		if (secpolicy_dispadm(reqpcredp) != 0) {
428 			return (EPERM);
429 		}
430 		if (fxadmin.fx_ndpents * sizeof (fxdpent_t) != fxdpsz) {
431 			return (EINVAL);
432 		}
433 
434 		/*
435 		 * We read the user supplied table into a temporary buffer
436 		 * where it is validated before being copied over the
437 		 * fx_dptbl.
438 		 */
439 		tmpdpp = kmem_alloc(fxdpsz, KM_SLEEP);
440 		if (copyin(fxadmin.fx_dpents, tmpdpp, fxdpsz)) {
441 			kmem_free(tmpdpp, fxdpsz);
442 			return (EFAULT);
443 		}
444 		for (i = 0; i < fxadmin.fx_ndpents; i++) {
445 
446 			/*
447 			 * Validate the user supplied values. All we are doing
448 			 * here is verifying that the values are within their
449 			 * allowable ranges and will not panic the system. We
450 			 * make no attempt to ensure that the resulting
451 			 * configuration makes sense or results in reasonable
452 			 * performance.
453 			 */
454 			if (tmpdpp[i].fx_quantum <= 0 &&
455 			    tmpdpp[i].fx_quantum != FX_TQINF) {
456 				kmem_free(tmpdpp, fxdpsz);
457 				return (EINVAL);
458 			}
459 		}
460 
461 		/*
462 		 * Copy the user supplied values over the current fx_dptbl
463 		 * values. The fx_globpri member is read-only so we don't
464 		 * overwrite it.
465 		 */
466 		mutex_enter(&fx_dptblock);
467 		for (i = 0; i < fxadmin.fx_ndpents; i++) {
468 			fx_dptbl[i].fx_quantum = tmpdpp[i].fx_quantum;
469 		}
470 		mutex_exit(&fx_dptblock);
471 		kmem_free(tmpdpp, fxdpsz);
472 		break;
473 
474 	default:
475 		return (EINVAL);
476 	}
477 	return (0);
478 }
479 
480 
481 /*
482  * Allocate a fixed priority class specific thread structure and
483  * initialize it with the parameters supplied. Also move the thread
484  * to specified priority.
485  */
486 static int
487 fx_enterclass(kthread_t *t, id_t cid, void *parmsp, cred_t *reqpcredp,
488     void *bufp)
489 {
490 	fxkparms_t	*fxkparmsp = (fxkparms_t *)parmsp;
491 	fxproc_t	*fxpp;
492 	pri_t		reqfxupri;
493 	pri_t		reqfxuprilim;
494 
495 	fxpp = (fxproc_t *)bufp;
496 	ASSERT(fxpp != NULL);
497 
498 	/*
499 	 * Initialize the fxproc structure.
500 	 */
501 	fxpp->fx_flags = 0;
502 	fxpp->fx_callback = NULL;
503 	fxpp->fx_cookie = NULL;
504 
505 	if (fxkparmsp == NULL) {
506 		/*
507 		 * Use default values.
508 		 */
509 		fxpp->fx_pri = fxpp->fx_uprilim = 0;
510 		fxpp->fx_pquantum = fx_dptbl[fxpp->fx_pri].fx_quantum;
511 		fxpp->fx_nice =  NZERO;
512 	} else {
513 		/*
514 		 * Use supplied values.
515 		 */
516 
517 		if ((fxkparmsp->fx_cflags & FX_DOUPRILIM) == 0) {
518 			reqfxuprilim = 0;
519 		} else {
520 			if (fxkparmsp->fx_uprilim > FX_MAX_UNPRIV_PRI &&
521 			    secpolicy_setpriority(reqpcredp) != 0)
522 				return (EPERM);
523 			reqfxuprilim = fxkparmsp->fx_uprilim;
524 			FX_ADJUST_PRI(reqfxuprilim);
525 		}
526 
527 		if ((fxkparmsp->fx_cflags & FX_DOUPRI) == 0) {
528 			reqfxupri = reqfxuprilim;
529 		} else {
530 			if (fxkparmsp->fx_upri > FX_MAX_UNPRIV_PRI &&
531 			    secpolicy_setpriority(reqpcredp) != 0)
532 				return (EPERM);
533 			/*
534 			 * Set the user priority to the requested value
535 			 * or the upri limit, whichever is lower.
536 			 */
537 			reqfxupri = fxkparmsp->fx_upri;
538 			FX_ADJUST_PRI(reqfxupri);
539 
540 			if (reqfxupri > reqfxuprilim)
541 				reqfxupri = reqfxuprilim;
542 		}
543 
544 
545 		fxpp->fx_uprilim = reqfxuprilim;
546 		fxpp->fx_pri = reqfxupri;
547 
548 		fxpp->fx_nice = NZERO - (NZERO * reqfxupri)
549 			/ fx_maxupri;
550 
551 		if (((fxkparmsp->fx_cflags & FX_DOTQ) == 0) ||
552 		    (fxkparmsp->fx_tqntm == FX_TQDEF)) {
553 			fxpp->fx_pquantum = fx_dptbl[fxpp->fx_pri].fx_quantum;
554 		} else {
555 			if (secpolicy_setpriority(reqpcredp) != 0)
556 				return (EPERM);
557 
558 			if (fxkparmsp->fx_tqntm == FX_TQINF)
559 				fxpp->fx_pquantum = FX_TQINF;
560 			else {
561 				fxpp->fx_pquantum = fxkparmsp->fx_tqntm;
562 			}
563 		}
564 
565 	}
566 
567 	fxpp->fx_timeleft = fxpp->fx_pquantum;
568 	fxpp->fx_tp = t;
569 
570 	thread_lock(t);			/* get dispatcher lock on thread */
571 	t->t_clfuncs = &(sclass[cid].cl_funcs->thread);
572 	t->t_cid = cid;
573 	t->t_cldata = (void *)fxpp;
574 	t->t_schedflag &= ~TS_RUNQMATCH;
575 	fx_change_priority(t, fxpp);
576 	thread_unlock(t);
577 
578 	FX_LIST_INSERT(fxpp);
579 
580 	return (0);
581 }
582 
583 /*
584  * The thread is exiting.
585  */
586 static void
587 fx_exit(kthread_t *t)
588 {
589 	fxproc_t *fxpp;
590 
591 	thread_lock(t);
592 	fxpp = (fxproc_t *)(t->t_cldata);
593 
594 	if (FX_HAS_CB(fxpp)) {
595 		FX_CB_EXIT(FX_CALLB(fxpp), fxpp->fx_cookie);
596 		fxpp->fx_callback = NULL;
597 		fxpp->fx_cookie = NULL;
598 		thread_unlock(t);
599 		FX_CB_LIST_DELETE(fxpp);
600 		return;
601 	}
602 	thread_unlock(t);
603 }
604 
605 /*
606  * Exiting the class. Free fxproc structure of thread.
607  */
608 static void
609 fx_exitclass(void *procp)
610 {
611 	fxproc_t *fxpp = (fxproc_t *)procp;
612 
613 	thread_lock(fxpp->fx_tp);
614 	if (FX_HAS_CB(fxpp)) {
615 
616 		FX_CB_EXIT(FX_CALLB(fxpp), fxpp->fx_cookie);
617 
618 		fxpp->fx_callback = NULL;
619 		fxpp->fx_cookie = NULL;
620 		thread_unlock(fxpp->fx_tp);
621 		FX_CB_LIST_DELETE(fxpp);
622 	} else
623 		thread_unlock(fxpp->fx_tp);
624 	FX_LIST_DELETE(fxpp);
625 
626 	kmem_free(fxpp, sizeof (fxproc_t));
627 }
628 
629 /* ARGSUSED */
630 static int
631 fx_canexit(kthread_t *t, cred_t *cred)
632 {
633 	/*
634 	 * A thread can always leave the FX class
635 	 */
636 	return (0);
637 }
638 
639 /*
640  * Initialize fixed-priority class specific proc structure for a child.
641  * callbacks are not inherited upon fork.
642  */
643 static int
644 fx_fork(kthread_t *t, kthread_t *ct, void *bufp)
645 {
646 	fxproc_t	*pfxpp;		/* ptr to parent's fxproc structure */
647 	fxproc_t	*cfxpp;		/* ptr to child's fxproc structure */
648 
649 	ASSERT(MUTEX_HELD(&ttoproc(t)->p_lock));
650 
651 	cfxpp = (fxproc_t *)bufp;
652 	ASSERT(cfxpp != NULL);
653 	thread_lock(t);
654 	pfxpp = (fxproc_t *)t->t_cldata;
655 	/*
656 	 * Initialize child's fxproc structure.
657 	 */
658 	cfxpp->fx_timeleft = cfxpp->fx_pquantum = pfxpp->fx_pquantum;
659 	cfxpp->fx_pri = pfxpp->fx_pri;
660 	cfxpp->fx_uprilim = pfxpp->fx_uprilim;
661 	cfxpp->fx_nice = pfxpp->fx_nice;
662 	cfxpp->fx_callback = NULL;
663 	cfxpp->fx_cookie = NULL;
664 	cfxpp->fx_flags = pfxpp->fx_flags & ~(FXBACKQ);
665 
666 	cfxpp->fx_tp = ct;
667 	ct->t_cldata = (void *)cfxpp;
668 	thread_unlock(t);
669 
670 	/*
671 	 * Link new structure into fxproc list.
672 	 */
673 	FX_LIST_INSERT(cfxpp);
674 	return (0);
675 }
676 
677 
678 /*
679  * Child is placed at back of dispatcher queue and parent gives
680  * up processor so that the child runs first after the fork.
681  * This allows the child immediately execing to break the multiple
682  * use of copy on write pages with no disk home. The parent will
683  * get to steal them back rather than uselessly copying them.
684  */
685 static void
686 fx_forkret(kthread_t *t, kthread_t *ct)
687 {
688 	proc_t	*pp = ttoproc(t);
689 	proc_t	*cp = ttoproc(ct);
690 	fxproc_t *fxpp;
691 
692 	ASSERT(t == curthread);
693 	ASSERT(MUTEX_HELD(&pidlock));
694 
695 	/*
696 	 * Grab the child's p_lock before dropping pidlock to ensure
697 	 * the process does not disappear before we set it running.
698 	 */
699 	mutex_enter(&cp->p_lock);
700 	mutex_exit(&pidlock);
701 	continuelwps(cp);
702 	mutex_exit(&cp->p_lock);
703 
704 	mutex_enter(&pp->p_lock);
705 	continuelwps(pp);
706 	mutex_exit(&pp->p_lock);
707 
708 	thread_lock(t);
709 	fxpp = (fxproc_t *)(t->t_cldata);
710 	t->t_pri = fx_dptbl[fxpp->fx_pri].fx_globpri;
711 	ASSERT(t->t_pri >= 0 && t->t_pri <= fx_maxglobpri);
712 	THREAD_TRANSITION(t);
713 	fx_setrun(t);
714 	thread_unlock(t);
715 
716 	swtch();
717 }
718 
719 
720 /*
721  * Get information about the fixed-priority class into the buffer
722  * pointed to by fxinfop. The maximum configured user priority
723  * is the only information we supply.
724  */
725 static int
726 fx_getclinfo(void *infop)
727 {
728 	fxinfo_t *fxinfop = (fxinfo_t *)infop;
729 	fxinfop->fx_maxupri = fx_maxupri;
730 	return (0);
731 }
732 
733 
734 
735 /*
736  * Return the global scheduling priority ranges for the fixed-priority
737  * class in pcpri_t structure.
738  */
739 static int
740 fx_getclpri(pcpri_t *pcprip)
741 {
742 	pcprip->pc_clpmax = fx_dptbl[fx_maxumdpri].fx_globpri;
743 	pcprip->pc_clpmin = fx_dptbl[0].fx_globpri;
744 	return (0);
745 }
746 
747 
748 static void
749 fx_nullsys()
750 {}
751 
752 
753 /*
754  * Get the fixed-priority parameters of the thread pointed to by
755  * fxprocp into the buffer pointed to by fxparmsp.
756  */
757 static void
758 fx_parmsget(kthread_t *t, void *parmsp)
759 {
760 	fxproc_t *fxpp = (fxproc_t *)t->t_cldata;
761 	fxkparms_t *fxkparmsp = (fxkparms_t *)parmsp;
762 
763 	fxkparmsp->fx_upri = fxpp->fx_pri;
764 	fxkparmsp->fx_uprilim = fxpp->fx_uprilim;
765 	fxkparmsp->fx_tqntm = fxpp->fx_pquantum;
766 }
767 
768 
769 
770 /*
771  * Check the validity of the fixed-priority parameters in the buffer
772  * pointed to by fxparmsp.
773  */
774 static int
775 fx_parmsin(void *parmsp)
776 {
777 	fxparms_t	*fxparmsp = (fxparms_t *)parmsp;
778 	uint_t		cflags;
779 	longlong_t	ticks;
780 	/*
781 	 * Check validity of parameters.
782 	 */
783 
784 	if ((fxparmsp->fx_uprilim > fx_maxupri ||
785 	    fxparmsp->fx_uprilim < 0) &&
786 	    fxparmsp->fx_uprilim != FX_NOCHANGE)
787 		return (EINVAL);
788 
789 	if ((fxparmsp->fx_upri > fx_maxupri ||
790 	    fxparmsp->fx_upri < 0) &&
791 	    fxparmsp->fx_upri != FX_NOCHANGE)
792 		return (EINVAL);
793 
794 	if ((fxparmsp->fx_tqsecs == 0 && fxparmsp->fx_tqnsecs == 0) ||
795 		fxparmsp->fx_tqnsecs >= NANOSEC)
796 		return (EINVAL);
797 
798 	cflags = (fxparmsp->fx_upri != FX_NOCHANGE ? FX_DOUPRI : 0);
799 
800 	if (fxparmsp->fx_uprilim != FX_NOCHANGE) {
801 		cflags |= FX_DOUPRILIM;
802 	}
803 
804 	if (fxparmsp->fx_tqnsecs != FX_NOCHANGE)
805 		cflags |= FX_DOTQ;
806 
807 	/*
808 	 * convert the buffer to kernel format.
809 	 */
810 
811 	if (fxparmsp->fx_tqnsecs >= 0) {
812 		if ((ticks = SEC_TO_TICK((longlong_t)fxparmsp->fx_tqsecs) +
813 		    NSEC_TO_TICK_ROUNDUP(fxparmsp->fx_tqnsecs)) > INT_MAX)
814 			return (ERANGE);
815 
816 		((fxkparms_t *)fxparmsp)->fx_tqntm = (int)ticks;
817 	} else {
818 		if ((fxparmsp->fx_tqnsecs != FX_NOCHANGE) &&
819 		    (fxparmsp->fx_tqnsecs != FX_TQINF) &&
820 		    (fxparmsp->fx_tqnsecs != FX_TQDEF))
821 			return (EINVAL);
822 		((fxkparms_t *)fxparmsp)->fx_tqntm = fxparmsp->fx_tqnsecs;
823 	}
824 
825 	((fxkparms_t *)fxparmsp)->fx_cflags = cflags;
826 
827 	return (0);
828 }
829 
830 
831 /*
832  * Check the validity of the fixed-priority parameters in the pc_vaparms_t
833  * structure vaparmsp and put them in the buffer pointed to by fxprmsp.
834  * pc_vaparms_t contains (key, value) pairs of parameter.
835  */
836 static int
837 fx_vaparmsin(void *prmsp, pc_vaparms_t *vaparmsp)
838 {
839 	uint_t		secs = 0;
840 	uint_t		cnt;
841 	int		nsecs = 0;
842 	int		priflag, secflag, nsecflag, limflag;
843 	longlong_t	ticks;
844 	fxkparms_t	*fxprmsp = (fxkparms_t *)prmsp;
845 	pc_vaparm_t	*vpp = &vaparmsp->pc_parms[0];
846 
847 
848 	/*
849 	 * First check the validity of parameters and convert them
850 	 * from the user supplied format to the internal format.
851 	 */
852 	priflag = secflag = nsecflag = limflag = 0;
853 
854 	fxprmsp->fx_cflags = 0;
855 
856 	if (vaparmsp->pc_vaparmscnt > PC_VAPARMCNT)
857 		return (EINVAL);
858 
859 	for (cnt = 0; cnt < vaparmsp->pc_vaparmscnt; cnt++, vpp++) {
860 
861 		switch (vpp->pc_key) {
862 		case FX_KY_UPRILIM:
863 			if (limflag++)
864 				return (EINVAL);
865 			fxprmsp->fx_cflags |= FX_DOUPRILIM;
866 			fxprmsp->fx_uprilim = (pri_t)vpp->pc_parm;
867 			if (fxprmsp->fx_uprilim > fx_maxupri ||
868 			    fxprmsp->fx_uprilim < 0)
869 				return (EINVAL);
870 			break;
871 
872 		case FX_KY_UPRI:
873 			if (priflag++)
874 				return (EINVAL);
875 			fxprmsp->fx_cflags |= FX_DOUPRI;
876 			fxprmsp->fx_upri = (pri_t)vpp->pc_parm;
877 			if (fxprmsp->fx_upri > fx_maxupri ||
878 			    fxprmsp->fx_upri < 0)
879 				return (EINVAL);
880 			break;
881 
882 		case FX_KY_TQSECS:
883 			if (secflag++)
884 				return (EINVAL);
885 			fxprmsp->fx_cflags |= FX_DOTQ;
886 			secs = (uint_t)vpp->pc_parm;
887 			break;
888 
889 		case FX_KY_TQNSECS:
890 			if (nsecflag++)
891 				return (EINVAL);
892 			fxprmsp->fx_cflags |= FX_DOTQ;
893 			nsecs = (int)vpp->pc_parm;
894 			break;
895 
896 		default:
897 			return (EINVAL);
898 		}
899 	}
900 
901 	if (vaparmsp->pc_vaparmscnt == 0) {
902 		/*
903 		 * Use default parameters.
904 		 */
905 		fxprmsp->fx_upri = 0;
906 		fxprmsp->fx_uprilim = 0;
907 		fxprmsp->fx_tqntm = FX_TQDEF;
908 		fxprmsp->fx_cflags = FX_DOUPRI | FX_DOUPRILIM | FX_DOTQ;
909 	} else if ((fxprmsp->fx_cflags & FX_DOTQ) != 0) {
910 		if ((secs == 0 && nsecs == 0) || nsecs >= NANOSEC)
911 			return (EINVAL);
912 
913 		if (nsecs >= 0) {
914 			if ((ticks = SEC_TO_TICK((longlong_t)secs) +
915 			    NSEC_TO_TICK_ROUNDUP(nsecs)) > INT_MAX)
916 				return (ERANGE);
917 
918 			fxprmsp->fx_tqntm = (int)ticks;
919 		} else {
920 			if (nsecs != FX_TQINF && nsecs != FX_TQDEF)
921 				return (EINVAL);
922 			fxprmsp->fx_tqntm = nsecs;
923 		}
924 	}
925 
926 	return (0);
927 }
928 
929 
930 /*
931  * Nothing to do here but return success.
932  */
933 /* ARGSUSED */
934 static int
935 fx_parmsout(void *parmsp, pc_vaparms_t *vaparmsp)
936 {
937 	register fxkparms_t	*fxkprmsp = (fxkparms_t *)parmsp;
938 
939 	if (vaparmsp != NULL)
940 		return (0);
941 
942 	if (fxkprmsp->fx_tqntm < 0) {
943 		/*
944 		 * Quantum field set to special value (e.g. FX_TQINF)
945 		 */
946 		((fxparms_t *)fxkprmsp)->fx_tqnsecs = fxkprmsp->fx_tqntm;
947 		((fxparms_t *)fxkprmsp)->fx_tqsecs = 0;
948 
949 	} else {
950 		/* Convert quantum from ticks to seconds-nanoseconds */
951 
952 		timestruc_t ts;
953 		TICK_TO_TIMESTRUC(fxkprmsp->fx_tqntm, &ts);
954 		((fxparms_t *)fxkprmsp)->fx_tqsecs = ts.tv_sec;
955 		((fxparms_t *)fxkprmsp)->fx_tqnsecs = ts.tv_nsec;
956 	}
957 
958 	return (0);
959 }
960 
961 
962 /*
963  * Copy all selected fixed-priority class parameters to the user.
964  * The parameters are specified by a key.
965  */
966 static int
967 fx_vaparmsout(void *prmsp, pc_vaparms_t *vaparmsp)
968 {
969 	fxkparms_t	*fxkprmsp = (fxkparms_t *)prmsp;
970 	timestruc_t	ts;
971 	uint_t		cnt;
972 	uint_t		secs;
973 	int		nsecs;
974 	int		priflag, secflag, nsecflag, limflag;
975 	pc_vaparm_t	*vpp = &vaparmsp->pc_parms[0];
976 
977 	ASSERT(MUTEX_NOT_HELD(&curproc->p_lock));
978 
979 	priflag = secflag = nsecflag = limflag = 0;
980 
981 	if (vaparmsp->pc_vaparmscnt > PC_VAPARMCNT)
982 		return (EINVAL);
983 
984 	if (fxkprmsp->fx_tqntm < 0) {
985 		/*
986 		 * Quantum field set to special value (e.g. FX_TQINF).
987 		 */
988 		secs = 0;
989 		nsecs = fxkprmsp->fx_tqntm;
990 	} else {
991 		/*
992 		 * Convert quantum from ticks to seconds-nanoseconds.
993 		 */
994 		TICK_TO_TIMESTRUC(fxkprmsp->fx_tqntm, &ts);
995 		secs = ts.tv_sec;
996 		nsecs = ts.tv_nsec;
997 	}
998 
999 
1000 	for (cnt = 0; cnt < vaparmsp->pc_vaparmscnt; cnt++, vpp++) {
1001 
1002 		switch (vpp->pc_key) {
1003 		case FX_KY_UPRILIM:
1004 			if (limflag++)
1005 				return (EINVAL);
1006 			if (copyout(&fxkprmsp->fx_uprilim,
1007 			    (void *)(uintptr_t)vpp->pc_parm, sizeof (pri_t)))
1008 				return (EFAULT);
1009 			break;
1010 
1011 		case FX_KY_UPRI:
1012 			if (priflag++)
1013 				return (EINVAL);
1014 			if (copyout(&fxkprmsp->fx_upri,
1015 			    (void *)(uintptr_t)vpp->pc_parm, sizeof (pri_t)))
1016 				return (EFAULT);
1017 			break;
1018 
1019 		case FX_KY_TQSECS:
1020 			if (secflag++)
1021 				return (EINVAL);
1022 			if (copyout(&secs,
1023 			    (void *)(uintptr_t)vpp->pc_parm, sizeof (uint_t)))
1024 				return (EFAULT);
1025 			break;
1026 
1027 		case FX_KY_TQNSECS:
1028 			if (nsecflag++)
1029 				return (EINVAL);
1030 			if (copyout(&nsecs,
1031 			    (void *)(uintptr_t)vpp->pc_parm, sizeof (int)))
1032 				return (EFAULT);
1033 			break;
1034 
1035 		default:
1036 			return (EINVAL);
1037 		}
1038 	}
1039 
1040 	return (0);
1041 }
1042 
1043 /*
1044  * Set the scheduling parameters of the thread pointed to by fxprocp
1045  * to those specified in the buffer pointed to by fxparmsp.
1046  */
1047 /* ARGSUSED */
1048 static int
1049 fx_parmsset(kthread_t *tx, void *parmsp, id_t reqpcid, cred_t *reqpcredp)
1050 {
1051 	char		nice;
1052 	pri_t		reqfxuprilim;
1053 	pri_t		reqfxupri;
1054 	fxkparms_t	*fxkparmsp = (fxkparms_t *)parmsp;
1055 	fxproc_t	*fxpp;
1056 
1057 
1058 	ASSERT(MUTEX_HELD(&(ttoproc(tx))->p_lock));
1059 
1060 	thread_lock(tx);
1061 	fxpp = (fxproc_t *)tx->t_cldata;
1062 
1063 	if ((fxkparmsp->fx_cflags & FX_DOUPRILIM) == 0)
1064 		reqfxuprilim = fxpp->fx_uprilim;
1065 	else
1066 		reqfxuprilim = fxkparmsp->fx_uprilim;
1067 
1068 	/*
1069 	 * Basic permissions enforced by generic kernel code
1070 	 * for all classes require that a thread attempting
1071 	 * to change the scheduling parameters of a target
1072 	 * thread be privileged or have a real or effective
1073 	 * UID matching that of the target thread. We are not
1074 	 * called unless these basic permission checks have
1075 	 * already passed. The fixed priority class requires in
1076 	 * addition that the calling thread be privileged if it
1077 	 * is attempting to raise the pri above its current
1078 	 * value This may have been checked previously but if our
1079 	 * caller passed us a non-NULL credential pointer we assume
1080 	 * it hasn't and we check it here.
1081 	 */
1082 
1083 	if ((reqpcredp != NULL) &&
1084 	    (reqfxuprilim > fxpp->fx_uprilim ||
1085 	    ((fxkparmsp->fx_cflags & FX_DOTQ) != 0)) &&
1086 	    secpolicy_setpriority(reqpcredp) != 0) {
1087 		thread_unlock(tx);
1088 		return (EPERM);
1089 	}
1090 
1091 	FX_ADJUST_PRI(reqfxuprilim);
1092 
1093 	if ((fxkparmsp->fx_cflags & FX_DOUPRI) == 0)
1094 		reqfxupri = fxpp->fx_pri;
1095 	else
1096 		reqfxupri = fxkparmsp->fx_upri;
1097 
1098 
1099 	/*
1100 	 * Make sure the user priority doesn't exceed the upri limit.
1101 	 */
1102 	if (reqfxupri > reqfxuprilim)
1103 		reqfxupri = reqfxuprilim;
1104 
1105 	/*
1106 	 * Set fx_nice to the nice value corresponding to the user
1107 	 * priority we are setting.  Note that setting the nice field
1108 	 * of the parameter struct won't affect upri or nice.
1109 	 */
1110 
1111 	nice = NZERO - (reqfxupri * NZERO) / fx_maxupri;
1112 
1113 	if (nice > NZERO)
1114 		nice = NZERO;
1115 
1116 	fxpp->fx_uprilim = reqfxuprilim;
1117 	fxpp->fx_pri = reqfxupri;
1118 
1119 	if (fxkparmsp->fx_tqntm == FX_TQINF)
1120 		fxpp->fx_pquantum = FX_TQINF;
1121 	else if (fxkparmsp->fx_tqntm == FX_TQDEF)
1122 		fxpp->fx_pquantum = fx_dptbl[fxpp->fx_pri].fx_quantum;
1123 	else if ((fxkparmsp->fx_cflags & FX_DOTQ) != 0)
1124 		fxpp->fx_pquantum = fxkparmsp->fx_tqntm;
1125 
1126 	fxpp->fx_nice = nice;
1127 
1128 	fx_change_priority(tx, fxpp);
1129 	thread_unlock(tx);
1130 	return (0);
1131 }
1132 
1133 
1134 /*
1135  * Return the global scheduling priority that would be assigned
1136  * to a thread entering the fixed-priority class with the fx_upri.
1137  */
1138 static pri_t
1139 fx_globpri(kthread_t *t)
1140 {
1141 	fxproc_t *fxpp;
1142 
1143 	ASSERT(MUTEX_HELD(&ttoproc(t)->p_lock));
1144 
1145 	fxpp = (fxproc_t *)t->t_cldata;
1146 	return (fx_dptbl[fxpp->fx_pri].fx_globpri);
1147 
1148 }
1149 
1150 /*
1151  * Arrange for thread to be placed in appropriate location
1152  * on dispatcher queue.
1153  *
1154  * This is called with the current thread in TS_ONPROC and locked.
1155  */
1156 static void
1157 fx_preempt(kthread_t *t)
1158 {
1159 	fxproc_t	*fxpp = (fxproc_t *)(t->t_cldata);
1160 #ifdef KSLICE
1161 	extern int	kslice;
1162 #endif
1163 
1164 	ASSERT(t == curthread);
1165 	ASSERT(THREAD_LOCK_HELD(curthread));
1166 
1167 	/*
1168 	 * Check to see if we're doing "preemption control" here.  If
1169 	 * we are, and if the user has requested that this thread not
1170 	 * be preempted, and if preemptions haven't been put off for
1171 	 * too long, let the preemption happen here but try to make
1172 	 * sure the thread is rescheduled as soon as possible.  We do
1173 	 * this by putting it on the front of the highest priority run
1174 	 * queue in the FX class.  If the preemption has been put off
1175 	 * for too long, clear the "nopreempt" bit and let the thread
1176 	 * be preempted.
1177 	 */
1178 	if (t->t_schedctl && schedctl_get_nopreempt(t)) {
1179 		if (fxpp->fx_pquantum == FX_TQINF ||
1180 		    fxpp->fx_timeleft > -SC_MAX_TICKS) {
1181 			DTRACE_SCHED1(schedctl__nopreempt, kthread_t *, t);
1182 			schedctl_set_yield(t, 1);
1183 			setfrontdq(t);
1184 			return;
1185 		} else {
1186 			schedctl_set_nopreempt(t, 0);
1187 			DTRACE_SCHED1(schedctl__preempt, kthread_t *, t);
1188 			TNF_PROBE_2(schedctl_preempt, "schedctl FX fx_preempt",
1189 			    /* CSTYLED */, tnf_pid, pid, ttoproc(t)->p_pid,
1190 			    tnf_lwpid, lwpid, t->t_tid);
1191 			/*
1192 			 * Fall through and be preempted below.
1193 			 */
1194 		}
1195 	}
1196 
1197 	if (FX_HAS_CB(fxpp)) {
1198 		clock_t new_quantum =  (clock_t)fxpp->fx_pquantum;
1199 		pri_t	newpri = fxpp->fx_pri;
1200 		FX_CB_PREEMPT(FX_CALLB(fxpp), fxpp->fx_cookie,
1201 			&new_quantum, &newpri);
1202 		FX_ADJUST_QUANTUM(new_quantum);
1203 		if ((int)new_quantum != fxpp->fx_pquantum) {
1204 			fxpp->fx_pquantum = (int)new_quantum;
1205 			fxpp->fx_timeleft = fxpp->fx_pquantum;
1206 		}
1207 		FX_ADJUST_PRI(newpri);
1208 		fxpp->fx_pri = newpri;
1209 		THREAD_CHANGE_PRI(t, fx_dptbl[fxpp->fx_pri].fx_globpri);
1210 	}
1211 
1212 	if ((fxpp->fx_flags & (FXBACKQ)) == FXBACKQ) {
1213 		fxpp->fx_timeleft = fxpp->fx_pquantum;
1214 		fxpp->fx_flags &= ~FXBACKQ;
1215 		setbackdq(t);
1216 	} else {
1217 #ifdef KSLICE
1218 		if (kslice)
1219 			setbackdq(t);
1220 		else
1221 #endif
1222 			setfrontdq(t);
1223 	}
1224 }
1225 
1226 static void
1227 fx_setrun(kthread_t *t)
1228 {
1229 	fxproc_t *fxpp = (fxproc_t *)(t->t_cldata);
1230 
1231 	ASSERT(THREAD_LOCK_HELD(t));	/* t should be in transition */
1232 	fxpp->fx_flags &= ~FXBACKQ;
1233 
1234 	if (t->t_disp_time != lbolt)
1235 		setbackdq(t);
1236 	else
1237 		setfrontdq(t);
1238 }
1239 
1240 
1241 /*
1242  * Prepare thread for sleep. We reset the thread priority so it will
1243  * run at the kernel priority level when it wakes up.
1244  */
1245 static void
1246 fx_sleep(kthread_t *t)
1247 {
1248 	fxproc_t	*fxpp = (fxproc_t *)(t->t_cldata);
1249 
1250 	ASSERT(t == curthread);
1251 	ASSERT(THREAD_LOCK_HELD(t));
1252 
1253 	if (FX_HAS_CB(fxpp)) {
1254 		FX_CB_SLEEP(FX_CALLB(fxpp), fxpp->fx_cookie);
1255 	}
1256 	t->t_stime = lbolt;		/* time stamp for the swapper */
1257 }
1258 
1259 
1260 /*
1261  * Return Values:
1262  *
1263  *	-1 if the thread is loaded or is not eligible to be swapped in.
1264  *
1265  * FX and RT threads are designed so that they don't swapout; however,
1266  * it is possible that while the thread is swapped out and in another class, it
1267  * can be changed to FX or RT.  Since these threads should be swapped in
1268  * as soon as they're runnable, rt_swapin returns SHRT_MAX, and fx_swapin
1269  * returns SHRT_MAX - 1, so that it gives deference to any swapped out
1270  * RT threads.
1271  */
1272 /* ARGSUSED */
1273 static pri_t
1274 fx_swapin(kthread_t *t, int flags)
1275 {
1276 	pri_t	tpri = -1;
1277 
1278 	ASSERT(THREAD_LOCK_HELD(t));
1279 
1280 	if (t->t_state == TS_RUN && (t->t_schedflag & TS_LOAD) == 0) {
1281 		tpri = (pri_t)SHRT_MAX - 1;
1282 	}
1283 
1284 	return (tpri);
1285 }
1286 
1287 /*
1288  * Return Values
1289  *	-1 if the thread isn't loaded or is not eligible to be swapped out.
1290  */
1291 /* ARGSUSED */
1292 static pri_t
1293 fx_swapout(kthread_t *t, int flags)
1294 {
1295 	ASSERT(THREAD_LOCK_HELD(t));
1296 
1297 	return (-1);
1298 
1299 }
1300 
1301 /* ARGSUSED */
1302 static void
1303 fx_stop(kthread_t *t, int why, int what)
1304 {
1305 	fxproc_t *fxpp = (fxproc_t *)(t->t_cldata);
1306 
1307 	ASSERT(THREAD_LOCK_HELD(t));
1308 
1309 	if (FX_HAS_CB(fxpp)) {
1310 		FX_CB_STOP(FX_CALLB(fxpp), fxpp->fx_cookie);
1311 	}
1312 }
1313 
1314 /*
1315  * Check for time slice expiration.  If time slice has expired
1316  * set runrun to cause preemption.
1317  */
1318 static void
1319 fx_tick(kthread_t *t)
1320 {
1321 	fxproc_t *fxpp;
1322 
1323 	ASSERT(MUTEX_HELD(&(ttoproc(t))->p_lock));
1324 
1325 	thread_lock(t);
1326 
1327 	fxpp = (fxproc_t *)(t->t_cldata);
1328 
1329 	if (FX_HAS_CB(fxpp)) {
1330 		clock_t new_quantum =  (clock_t)fxpp->fx_pquantum;
1331 		pri_t	newpri = fxpp->fx_pri;
1332 		FX_CB_TICK(FX_CALLB(fxpp), fxpp->fx_cookie,
1333 			&new_quantum, &newpri);
1334 		FX_ADJUST_QUANTUM(new_quantum);
1335 		if ((int)new_quantum != fxpp->fx_pquantum) {
1336 			fxpp->fx_pquantum = (int)new_quantum;
1337 			fxpp->fx_timeleft = fxpp->fx_pquantum;
1338 		}
1339 		FX_ADJUST_PRI(newpri);
1340 		if (newpri != fxpp->fx_pri) {
1341 			fxpp->fx_pri = newpri;
1342 			fx_change_priority(t, fxpp);
1343 		}
1344 	}
1345 	if ((fxpp->fx_pquantum != FX_TQINF) &&
1346 	    (--fxpp->fx_timeleft <= 0)) {
1347 		pri_t	new_pri;
1348 
1349 		/*
1350 		 * If we're doing preemption control and trying to
1351 		 * avoid preempting this thread, just note that
1352 		 * the thread should yield soon and let it keep
1353 		 * running (unless it's been a while).
1354 		 */
1355 		if (t->t_schedctl && schedctl_get_nopreempt(t)) {
1356 			if (fxpp->fx_timeleft > -SC_MAX_TICKS) {
1357 				DTRACE_SCHED1(schedctl__nopreempt,
1358 				    kthread_t *, t);
1359 				schedctl_set_yield(t, 1);
1360 				thread_unlock_nopreempt(t);
1361 				return;
1362 			}
1363 			TNF_PROBE_2(schedctl_failsafe,
1364 			    "schedctl FX fx_tick", /* CSTYLED */,
1365 			    tnf_pid, pid, ttoproc(t)->p_pid,
1366 			    tnf_lwpid, lwpid, t->t_tid);
1367 		}
1368 		new_pri = fx_dptbl[fxpp->fx_pri].fx_globpri;
1369 		ASSERT(new_pri >= 0 && new_pri <= fx_maxglobpri);
1370 		/*
1371 		 * When the priority of a thread is changed,
1372 		 * it may be necessary to adjust its position
1373 		 * on a sleep queue or dispatch queue. Even
1374 		 * when the priority is not changed, we need
1375 		 * to preserve round robin on dispatch queue.
1376 		 * The function thread_change_pri accomplishes
1377 		 * this.
1378 		 */
1379 		if (thread_change_pri(t, new_pri, 0)) {
1380 			fxpp->fx_timeleft = fxpp->fx_pquantum;
1381 		} else {
1382 			fxpp->fx_flags |= FXBACKQ;
1383 			cpu_surrender(t);
1384 		}
1385 	} else if (t->t_state == TS_ONPROC &&
1386 		    t->t_pri < t->t_disp_queue->disp_maxrunpri) {
1387 		fxpp->fx_flags |= FXBACKQ;
1388 		cpu_surrender(t);
1389 	}
1390 
1391 	thread_unlock_nopreempt(t);	/* clock thread can't be preempted */
1392 }
1393 
1394 
1395 static void
1396 fx_trapret(kthread_t *t)
1397 {
1398 	cpu_t		*cp = CPU;
1399 
1400 	ASSERT(THREAD_LOCK_HELD(t));
1401 	ASSERT(t == curthread);
1402 	ASSERT(cp->cpu_dispthread == t);
1403 	ASSERT(t->t_state == TS_ONPROC);
1404 }
1405 
1406 
1407 /*
1408  * Processes waking up go to the back of their queue.
1409  */
1410 static void
1411 fx_wakeup(kthread_t *t)
1412 {
1413 	fxproc_t	*fxpp = (fxproc_t *)(t->t_cldata);
1414 
1415 	ASSERT(THREAD_LOCK_HELD(t));
1416 
1417 	t->t_stime = lbolt;		/* time stamp for the swapper */
1418 	if (FX_HAS_CB(fxpp)) {
1419 		clock_t new_quantum =  (clock_t)fxpp->fx_pquantum;
1420 		pri_t	newpri = fxpp->fx_pri;
1421 		FX_CB_WAKEUP(FX_CALLB(fxpp), fxpp->fx_cookie,
1422 			&new_quantum, &newpri);
1423 		FX_ADJUST_QUANTUM(new_quantum);
1424 		if ((int)new_quantum != fxpp->fx_pquantum) {
1425 			fxpp->fx_pquantum = (int)new_quantum;
1426 			fxpp->fx_timeleft = fxpp->fx_pquantum;
1427 		}
1428 
1429 		FX_ADJUST_PRI(newpri);
1430 		if (newpri != fxpp->fx_pri) {
1431 			fxpp->fx_pri = newpri;
1432 			THREAD_CHANGE_PRI(t, fx_dptbl[fxpp->fx_pri].fx_globpri);
1433 		}
1434 	}
1435 
1436 	fxpp->fx_flags &= ~FXBACKQ;
1437 
1438 	if (t->t_disp_time != lbolt)
1439 		setbackdq(t);
1440 	else
1441 		setfrontdq(t);
1442 }
1443 
1444 
1445 /*
1446  * When a thread yields, put it on the back of the run queue.
1447  */
1448 static void
1449 fx_yield(kthread_t *t)
1450 {
1451 	fxproc_t	*fxpp = (fxproc_t *)(t->t_cldata);
1452 
1453 	ASSERT(t == curthread);
1454 	ASSERT(THREAD_LOCK_HELD(t));
1455 
1456 	if (FX_HAS_CB(fxpp))  {
1457 		clock_t new_quantum =  (clock_t)fxpp->fx_pquantum;
1458 		pri_t	newpri = fxpp->fx_pri;
1459 		FX_CB_PREEMPT(FX_CALLB(fxpp), fxpp->fx_cookie,
1460 				&new_quantum, &newpri);
1461 		FX_ADJUST_QUANTUM(new_quantum);
1462 		if ((int)new_quantum != fxpp->fx_pquantum) {
1463 			fxpp->fx_pquantum = (int)new_quantum;
1464 			fxpp->fx_timeleft = fxpp->fx_pquantum;
1465 		}
1466 		FX_ADJUST_PRI(newpri);
1467 		fxpp->fx_pri = newpri;
1468 		THREAD_CHANGE_PRI(t, fx_dptbl[fxpp->fx_pri].fx_globpri);
1469 	}
1470 
1471 	/*
1472 	 * Clear the preemption control "yield" bit since the user is
1473 	 * doing a yield.
1474 	 */
1475 	if (t->t_schedctl)
1476 		schedctl_set_yield(t, 0);
1477 
1478 	if (fxpp->fx_timeleft <= 0) {
1479 		/*
1480 		 * Time slice was artificially extended to avoid
1481 		 * preemption, so pretend we're preempting it now.
1482 		 */
1483 		DTRACE_SCHED1(schedctl__yield, int, -fxpp->fx_timeleft);
1484 		fxpp->fx_timeleft = fxpp->fx_pquantum;
1485 		THREAD_CHANGE_PRI(t, fx_dptbl[fxpp->fx_pri].fx_globpri);
1486 		ASSERT(t->t_pri >= 0 && t->t_pri <= fx_maxglobpri);
1487 	}
1488 
1489 	fxpp->fx_flags &= ~FXBACKQ;
1490 	setbackdq(t);
1491 }
1492 
1493 
1494 /*
1495  * Increment the nice value of the specified thread by incr and
1496  * return the new value in *retvalp.
1497  */
1498 static int
1499 fx_donice(kthread_t *t, cred_t *cr, int incr, int *retvalp)
1500 {
1501 	int		newnice;
1502 	fxproc_t	*fxpp = (fxproc_t *)(t->t_cldata);
1503 	fxkparms_t	fxkparms;
1504 
1505 	ASSERT(MUTEX_HELD(&(ttoproc(t))->p_lock));
1506 
1507 	/* If there's no change to priority, just return current setting */
1508 	if (incr == 0) {
1509 		if (retvalp) {
1510 			*retvalp = fxpp->fx_nice - NZERO;
1511 		}
1512 		return (0);
1513 	}
1514 
1515 	if ((incr < 0 || incr > 2 * NZERO) &&
1516 	    secpolicy_setpriority(cr) != 0)
1517 		return (EPERM);
1518 
1519 	/*
1520 	 * Specifying a nice increment greater than the upper limit of
1521 	 * 2 * NZERO - 1 will result in the thread's nice value being
1522 	 * set to the upper limit.  We check for this before computing
1523 	 * the new value because otherwise we could get overflow
1524 	 * if a privileged user specified some ridiculous increment.
1525 	 */
1526 	if (incr > 2 * NZERO - 1)
1527 		incr = 2 * NZERO - 1;
1528 
1529 	newnice = fxpp->fx_nice + incr;
1530 	if (newnice > NZERO)
1531 		newnice = NZERO;
1532 	else if (newnice < 0)
1533 		newnice = 0;
1534 
1535 	fxkparms.fx_uprilim = fxkparms.fx_upri =
1536 		-((newnice - NZERO) * fx_maxupri) / NZERO;
1537 
1538 	fxkparms.fx_cflags = FX_DOUPRILIM | FX_DOUPRI;
1539 
1540 	fxkparms.fx_tqntm = FX_TQDEF;
1541 
1542 	/*
1543 	 * Reset the uprilim and upri values of the thread. Adjust
1544 	 * time quantum accordingly.
1545 	 */
1546 
1547 	(void) fx_parmsset(t, (void *)&fxkparms, (id_t)0, (cred_t *)NULL);
1548 
1549 	/*
1550 	 * Although fx_parmsset already reset fx_nice it may
1551 	 * not have been set to precisely the value calculated above
1552 	 * because fx_parmsset determines the nice value from the
1553 	 * user priority and we may have truncated during the integer
1554 	 * conversion from nice value to user priority and back.
1555 	 * We reset fx_nice to the value we calculated above.
1556 	 */
1557 	fxpp->fx_nice = (char)newnice;
1558 
1559 	if (retvalp)
1560 		*retvalp = newnice - NZERO;
1561 
1562 	return (0);
1563 }
1564 
1565 static void
1566 fx_change_priority(kthread_t *t, fxproc_t *fxpp)
1567 {
1568 	pri_t	new_pri;
1569 
1570 	ASSERT(THREAD_LOCK_HELD(t));
1571 	new_pri = fx_dptbl[fxpp->fx_pri].fx_globpri;
1572 	ASSERT(new_pri >= 0 && new_pri <= fx_maxglobpri);
1573 	if (t == curthread || t->t_state == TS_ONPROC) {
1574 		/* curthread is always onproc */
1575 		cpu_t	*cp = t->t_disp_queue->disp_cpu;
1576 		THREAD_CHANGE_PRI(t, new_pri);
1577 		if (t == cp->cpu_dispthread)
1578 			cp->cpu_dispatch_pri = DISP_PRIO(t);
1579 		if (DISP_MUST_SURRENDER(t)) {
1580 			fxpp->fx_flags |= FXBACKQ;
1581 			cpu_surrender(t);
1582 		} else {
1583 			fxpp->fx_timeleft = fxpp->fx_pquantum;
1584 		}
1585 	} else {
1586 		/*
1587 		 * When the priority of a thread is changed,
1588 		 * it may be necessary to adjust its position
1589 		 * on a sleep queue or dispatch queue.
1590 		 * The function thread_change_pri accomplishes
1591 		 * this.
1592 		 */
1593 		if (thread_change_pri(t, new_pri, 0)) {
1594 			/*
1595 			 * The thread was on a run queue. Reset
1596 			 * its CPU timeleft from the quantum
1597 			 * associated with the new priority.
1598 			 */
1599 			fxpp->fx_timeleft = fxpp->fx_pquantum;
1600 		} else {
1601 			fxpp->fx_flags |= FXBACKQ;
1602 		}
1603 	}
1604 }
1605 
1606 static int
1607 fx_alloc(void **p, int flag)
1608 {
1609 	void *bufp;
1610 
1611 	bufp = kmem_alloc(sizeof (fxproc_t), flag);
1612 	if (bufp == NULL) {
1613 		return (ENOMEM);
1614 	} else {
1615 		*p = bufp;
1616 		return (0);
1617 	}
1618 }
1619 
1620 static void
1621 fx_free(void *bufp)
1622 {
1623 	if (bufp)
1624 		kmem_free(bufp, sizeof (fxproc_t));
1625 }
1626 
1627 /*
1628  * Release the callback list mutex after successful lookup
1629  */
1630 void
1631 fx_list_release(fxproc_t *fxpp)
1632 {
1633 	int index = FX_CB_LIST_HASH(fxpp->fx_ktid);
1634 	kmutex_t *lockp = &fx_cb_list_lock[index];
1635 	mutex_exit(lockp);
1636 }
1637 
1638 fxproc_t *
1639 fx_list_lookup(kt_did_t ktid)
1640 {
1641 	int index = FX_CB_LIST_HASH(ktid);
1642 	kmutex_t *lockp = &fx_cb_list_lock[index];
1643 	fxproc_t *fxpp;
1644 
1645 	mutex_enter(lockp);
1646 
1647 	for (fxpp = fx_cb_plisthead[index].fx_cb_next;
1648 	    fxpp != &fx_cb_plisthead[index]; fxpp = fxpp->fx_cb_next) {
1649 		if (fxpp->fx_tp->t_cid == fx_cid && fxpp->fx_ktid == ktid &&
1650 		    fxpp->fx_callback != NULL) {
1651 			/*
1652 			 * The caller is responsible for calling
1653 			 * fx_list_release to drop the lock upon
1654 			 * successful lookup
1655 			 */
1656 			return (fxpp);
1657 		}
1658 	}
1659 	mutex_exit(lockp);
1660 	return ((fxproc_t *)NULL);
1661 }
1662 
1663 
1664 /*
1665  * register a callback set of routines for current thread
1666  * thread should already be in FX class
1667  */
1668 int
1669 fx_register_callbacks(fx_callbacks_t *fx_callback, fx_cookie_t cookie,
1670 	pri_t pri, clock_t quantum)
1671 {
1672 
1673 	fxproc_t	*fxpp;
1674 
1675 	if (fx_callback == NULL)
1676 		return (EINVAL);
1677 
1678 	if (secpolicy_dispadm(CRED()) != 0)
1679 		return (EPERM);
1680 
1681 	if (FX_CB_VERSION(fx_callback) != FX_CALLB_REV)
1682 		return (EINVAL);
1683 
1684 	if (!FX_ISVALID(pri, quantum))
1685 		return (EINVAL);
1686 
1687 	thread_lock(curthread);		/* get dispatcher lock on thread */
1688 
1689 	if (curthread->t_cid != fx_cid) {
1690 		thread_unlock(curthread);
1691 		return (EINVAL);
1692 	}
1693 
1694 	fxpp = (fxproc_t *)(curthread->t_cldata);
1695 	ASSERT(fxpp != NULL);
1696 	if (FX_HAS_CB(fxpp)) {
1697 		thread_unlock(curthread);
1698 		return (EINVAL);
1699 	}
1700 
1701 	fxpp->fx_callback = fx_callback;
1702 	fxpp->fx_cookie = cookie;
1703 
1704 	if (pri != FX_CB_NOCHANGE) {
1705 		fxpp->fx_pri = pri;
1706 		FX_ADJUST_PRI(fxpp->fx_pri);
1707 		if (quantum == FX_TQDEF) {
1708 			fxpp->fx_pquantum = fx_dptbl[fxpp->fx_pri].fx_quantum;
1709 		} else if (quantum == FX_TQINF) {
1710 			fxpp->fx_pquantum = FX_TQINF;
1711 		} else if (quantum != FX_NOCHANGE) {
1712 			FX_ADJUST_QUANTUM(quantum);
1713 			fxpp->fx_pquantum = quantum;
1714 		}
1715 	} else if (quantum != FX_NOCHANGE && quantum != FX_TQDEF) {
1716 		if (quantum == FX_TQINF)
1717 			fxpp->fx_pquantum = FX_TQINF;
1718 		else {
1719 			FX_ADJUST_QUANTUM(quantum);
1720 			fxpp->fx_pquantum = quantum;
1721 		}
1722 	}
1723 
1724 	fxpp->fx_ktid = ddi_get_kt_did();
1725 
1726 	fx_change_priority(curthread, fxpp);
1727 
1728 	thread_unlock(curthread);
1729 
1730 	/*
1731 	 * Link new structure into fxproc list.
1732 	 */
1733 	FX_CB_LIST_INSERT(fxpp);
1734 	return (0);
1735 }
1736 
1737 /* unregister a callback set of routines for current thread */
1738 int
1739 fx_unregister_callbacks()
1740 {
1741 	fxproc_t	*fxpp;
1742 
1743 	if ((fxpp = fx_list_lookup(ddi_get_kt_did())) == NULL) {
1744 		/*
1745 		 * did not have a registered callback;
1746 		 */
1747 		return (EINVAL);
1748 	}
1749 
1750 	thread_lock(fxpp->fx_tp);
1751 	fxpp->fx_callback = NULL;
1752 	fxpp->fx_cookie = NULL;
1753 	thread_unlock(fxpp->fx_tp);
1754 	fx_list_release(fxpp);
1755 
1756 	FX_CB_LIST_DELETE(fxpp);
1757 	return (0);
1758 }
1759 
1760 /*
1761  * modify priority and/or quantum value of a thread with callback
1762  */
1763 int
1764 fx_modify_priority(kt_did_t ktid, clock_t quantum, pri_t pri)
1765 {
1766 	fxproc_t	*fxpp;
1767 
1768 	if (!FX_ISVALID(pri, quantum))
1769 		return (EINVAL);
1770 
1771 	if ((fxpp = fx_list_lookup(ktid)) == NULL) {
1772 		/*
1773 		 * either thread had exited or did not have a registered
1774 		 * callback;
1775 		 */
1776 		return (ESRCH);
1777 	}
1778 
1779 	thread_lock(fxpp->fx_tp);
1780 
1781 	if (pri != FX_CB_NOCHANGE) {
1782 		fxpp->fx_pri = pri;
1783 		FX_ADJUST_PRI(fxpp->fx_pri);
1784 		if (quantum == FX_TQDEF) {
1785 			fxpp->fx_pquantum = fx_dptbl[fxpp->fx_pri].fx_quantum;
1786 		} else if (quantum == FX_TQINF) {
1787 			fxpp->fx_pquantum = FX_TQINF;
1788 		} else if (quantum != FX_NOCHANGE) {
1789 			FX_ADJUST_QUANTUM(quantum);
1790 			fxpp->fx_pquantum = quantum;
1791 		}
1792 	} else if (quantum != FX_NOCHANGE && quantum != FX_TQDEF) {
1793 		if (quantum == FX_TQINF) {
1794 			fxpp->fx_pquantum = FX_TQINF;
1795 		} else {
1796 			FX_ADJUST_QUANTUM(quantum);
1797 			fxpp->fx_pquantum = quantum;
1798 		}
1799 	}
1800 
1801 	fx_change_priority(fxpp->fx_tp, fxpp);
1802 
1803 	thread_unlock(fxpp->fx_tp);
1804 	fx_list_release(fxpp);
1805 	return (0);
1806 }
1807 
1808 
1809 /*
1810  * return an iblock cookie for mutex initialization to be used in callbacks
1811  */
1812 void *
1813 fx_get_mutex_cookie()
1814 {
1815 	return ((void *)(uintptr_t)__ipltospl(DISP_LEVEL));
1816 }
1817 
1818 /*
1819  * return maximum relative priority
1820  */
1821 pri_t
1822 fx_get_maxpri()
1823 {
1824 	return (fx_maxumdpri);
1825 }
1826