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) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2018 Joyent, Inc.
25 */
26
27 #include <sys/types.h>
28 #include <sys/param.h>
29 #include <sys/sysmacros.h>
30 #include <sys/signal.h>
31 #include <sys/stack.h>
32 #include <sys/pcb.h>
33 #include <sys/user.h>
34 #include <sys/systm.h>
35 #include <sys/sysinfo.h>
36 #include <sys/errno.h>
37 #include <sys/cmn_err.h>
38 #include <sys/cred.h>
39 #include <sys/resource.h>
40 #include <sys/task.h>
41 #include <sys/project.h>
42 #include <sys/proc.h>
43 #include <sys/debug.h>
44 #include <sys/disp.h>
45 #include <sys/class.h>
46 #include <vm/seg_kmem.h>
47 #include <vm/seg_kp.h>
48 #include <sys/machlock.h>
49 #include <sys/kmem.h>
50 #include <sys/varargs.h>
51 #include <sys/turnstile.h>
52 #include <sys/poll.h>
53 #include <sys/vtrace.h>
54 #include <sys/callb.h>
55 #include <c2/audit.h>
56 #include <sys/tnf.h>
57 #include <sys/sobject.h>
58 #include <sys/cpupart.h>
59 #include <sys/pset.h>
60 #include <sys/door.h>
61 #include <sys/spl.h>
62 #include <sys/copyops.h>
63 #include <sys/rctl.h>
64 #include <sys/brand.h>
65 #include <sys/pool.h>
66 #include <sys/zone.h>
67 #include <sys/tsol/label.h>
68 #include <sys/tsol/tndb.h>
69 #include <sys/cpc_impl.h>
70 #include <sys/sdt.h>
71 #include <sys/reboot.h>
72 #include <sys/kdi.h>
73 #include <sys/schedctl.h>
74 #include <sys/waitq.h>
75 #include <sys/cpucaps.h>
76 #include <sys/kiconv.h>
77 #include <sys/ctype.h>
78
79 struct kmem_cache *thread_cache; /* cache of free threads */
80 struct kmem_cache *lwp_cache; /* cache of free lwps */
81 struct kmem_cache *turnstile_cache; /* cache of free turnstiles */
82
83 /*
84 * allthreads is only for use by kmem_readers. All kernel loops can use
85 * the current thread as a start/end point.
86 */
87 kthread_t *allthreads = &t0; /* circular list of all threads */
88
89 static kcondvar_t reaper_cv; /* synchronization var */
90 kthread_t *thread_deathrow; /* circular list of reapable threads */
91 kthread_t *lwp_deathrow; /* circular list of reapable threads */
92 kmutex_t reaplock; /* protects lwp and thread deathrows */
93 int thread_reapcnt = 0; /* number of threads on deathrow */
94 int lwp_reapcnt = 0; /* number of lwps on deathrow */
95 int reaplimit = 16; /* delay reaping until reaplimit */
96
97 thread_free_lock_t *thread_free_lock;
98 /* protects tick thread from reaper */
99
100 extern int nthread;
101
102 /* System Scheduling classes. */
103 id_t syscid; /* system scheduling class ID */
104 id_t sysdccid = CLASS_UNUSED; /* reset when SDC loads */
105
106 void *segkp_thread; /* cookie for segkp pool */
107
108 int lwp_cache_sz = 32;
109 int t_cache_sz = 8;
110 static kt_did_t next_t_id = 1;
111
112 /* Default mode for thread binding to CPUs and processor sets */
113 int default_binding_mode = TB_ALLHARD;
114
115 /*
116 * Min/Max stack sizes for stack size parameters
117 */
118 #define MAX_STKSIZE (32 * DEFAULTSTKSZ)
119 #define MIN_STKSIZE DEFAULTSTKSZ
120
121 /*
122 * default_stksize overrides lwp_default_stksize if it is set.
123 */
124 int default_stksize;
125 int lwp_default_stksize;
126
127 static zone_key_t zone_thread_key;
128
129 unsigned int kmem_stackinfo; /* stackinfo feature on-off */
130 kmem_stkinfo_t *kmem_stkinfo_log; /* stackinfo circular log */
131 static kmutex_t kmem_stkinfo_lock; /* protects kmem_stkinfo_log */
132
133 /*
134 * forward declarations for internal thread specific data (tsd)
135 */
136 static void *tsd_realloc(void *, size_t, size_t);
137
138 void thread_reaper(void);
139
140 /* forward declarations for stackinfo feature */
141 static void stkinfo_begin(kthread_t *);
142 static void stkinfo_end(kthread_t *);
143 static size_t stkinfo_percent(caddr_t, caddr_t, caddr_t);
144
145 /*ARGSUSED*/
146 static int
147 turnstile_constructor(void *buf, void *cdrarg, int kmflags)
148 {
149 bzero(buf, sizeof (turnstile_t));
150 return (0);
151 }
152
153 /*ARGSUSED*/
154 static void
155 turnstile_destructor(void *buf, void *cdrarg)
156 {
157 turnstile_t *ts = buf;
158
159 ASSERT(ts->ts_free == NULL);
160 ASSERT(ts->ts_waiters == 0);
161 ASSERT(ts->ts_inheritor == NULL);
162 ASSERT(ts->ts_sleepq[0].sq_first == NULL);
163 ASSERT(ts->ts_sleepq[1].sq_first == NULL);
164 }
165
166 void
167 thread_init(void)
168 {
169 kthread_t *tp;
170 extern char sys_name[];
171 extern void idle();
172 struct cpu *cpu = CPU;
173 int i;
174 kmutex_t *lp;
175
176 mutex_init(&reaplock, NULL, MUTEX_SPIN, (void *)ipltospl(DISP_LEVEL));
177 thread_free_lock =
178 kmem_alloc(sizeof (thread_free_lock_t) * THREAD_FREE_NUM, KM_SLEEP);
179 for (i = 0; i < THREAD_FREE_NUM; i++) {
180 lp = &thread_free_lock[i].tf_lock;
181 mutex_init(lp, NULL, MUTEX_DEFAULT, NULL);
182 }
183
184 #if defined(__i386) || defined(__amd64)
185 thread_cache = kmem_cache_create("thread_cache", sizeof (kthread_t),
186 PTR24_ALIGN, NULL, NULL, NULL, NULL, NULL, 0);
187
188 /*
189 * "struct _klwp" includes a "struct pcb", which includes a
190 * "struct fpu", which needs to be 64-byte aligned on amd64
191 * (and even on i386) for xsave/xrstor.
192 */
193 lwp_cache = kmem_cache_create("lwp_cache", sizeof (klwp_t),
194 64, NULL, NULL, NULL, NULL, NULL, 0);
195 #else
196 /*
197 * Allocate thread structures from static_arena. This prevents
198 * issues where a thread tries to relocate its own thread
199 * structure and touches it after the mapping has been suspended.
200 */
201 thread_cache = kmem_cache_create("thread_cache", sizeof (kthread_t),
202 PTR24_ALIGN, NULL, NULL, NULL, NULL, static_arena, 0);
203
204 lwp_stk_cache_init();
205
206 lwp_cache = kmem_cache_create("lwp_cache", sizeof (klwp_t),
207 0, NULL, NULL, NULL, NULL, NULL, 0);
208 #endif
209
210 turnstile_cache = kmem_cache_create("turnstile_cache",
211 sizeof (turnstile_t), 0,
212 turnstile_constructor, turnstile_destructor, NULL, NULL, NULL, 0);
213
214 label_init();
215 cred_init();
216
217 /*
218 * Initialize various resource management facilities.
219 */
220 rctl_init();
221 cpucaps_init();
222 /*
223 * Zone_init() should be called before project_init() so that project ID
224 * for the first project is initialized correctly.
225 */
226 zone_init();
227 project_init();
228 brand_init();
229 kiconv_init();
230 task_init();
231 tcache_init();
232 pool_init();
233
234 curthread->t_ts = kmem_cache_alloc(turnstile_cache, KM_SLEEP);
235
236 /*
237 * Originally, we had two parameters to set default stack
238 * size: one for lwp's (lwp_default_stksize), and one for
239 * kernel-only threads (DEFAULTSTKSZ, a.k.a. _defaultstksz).
240 * Now we have a third parameter that overrides both if it is
241 * set to a legal stack size, called default_stksize.
242 */
243
244 if (default_stksize == 0) {
245 default_stksize = DEFAULTSTKSZ;
246 } else if (default_stksize % PAGESIZE != 0 ||
247 default_stksize > MAX_STKSIZE ||
248 default_stksize < MIN_STKSIZE) {
249 cmn_err(CE_WARN, "Illegal stack size. Using %d",
250 (int)DEFAULTSTKSZ);
251 default_stksize = DEFAULTSTKSZ;
252 } else {
253 lwp_default_stksize = default_stksize;
254 }
255
256 if (lwp_default_stksize == 0) {
257 lwp_default_stksize = default_stksize;
258 } else if (lwp_default_stksize % PAGESIZE != 0 ||
259 lwp_default_stksize > MAX_STKSIZE ||
260 lwp_default_stksize < MIN_STKSIZE) {
261 cmn_err(CE_WARN, "Illegal stack size. Using %d",
262 default_stksize);
263 lwp_default_stksize = default_stksize;
264 }
265
266 segkp_lwp = segkp_cache_init(segkp, lwp_cache_sz,
267 lwp_default_stksize,
268 (KPD_NOWAIT | KPD_HASREDZONE | KPD_LOCKED));
269
270 segkp_thread = segkp_cache_init(segkp, t_cache_sz,
271 default_stksize, KPD_HASREDZONE | KPD_LOCKED | KPD_NO_ANON);
272
273 (void) getcid(sys_name, &syscid);
274 curthread->t_cid = syscid; /* current thread is t0 */
275
276 /*
277 * Set up the first CPU's idle thread.
278 * It runs whenever the CPU has nothing worthwhile to do.
279 */
280 tp = thread_create(NULL, 0, idle, NULL, 0, &p0, TS_STOPPED, -1);
281 cpu->cpu_idle_thread = tp;
282 tp->t_preempt = 1;
283 tp->t_disp_queue = cpu->cpu_disp;
284 ASSERT(tp->t_disp_queue != NULL);
285 tp->t_bound_cpu = cpu;
286 tp->t_affinitycnt = 1;
287
288 /*
289 * Registering a thread in the callback table is usually
290 * done in the initialization code of the thread. In this
291 * case, we do it right after thread creation to avoid
292 * blocking idle thread while registering itself. It also
293 * avoids the possibility of reregistration in case a CPU
294 * restarts its idle thread.
295 */
296 CALLB_CPR_INIT_SAFE(tp, "idle");
297
298 /*
299 * Create the thread_reaper daemon. From this point on, exited
300 * threads will get reaped.
301 */
302 (void) thread_create(NULL, 0, (void (*)())thread_reaper,
303 NULL, 0, &p0, TS_RUN, minclsyspri);
304
305 /*
306 * Finish initializing the kernel memory allocator now that
307 * thread_create() is available.
308 */
309 kmem_thread_init();
310
311 if (boothowto & RB_DEBUG)
312 kdi_dvec_thravail();
313 }
314
315 /*
316 * Create a thread.
317 *
318 * thread_create() blocks for memory if necessary. It never fails.
319 *
320 * If stk is NULL, the thread is created at the base of the stack
321 * and cannot be swapped.
322 */
323 kthread_t *
324 thread_create(
325 caddr_t stk,
326 size_t stksize,
327 void (*proc)(),
328 void *arg,
329 size_t len,
330 proc_t *pp,
331 int state,
332 pri_t pri)
333 {
334 kthread_t *t;
335 extern struct classfuncs sys_classfuncs;
336 turnstile_t *ts;
337
338 /*
339 * Every thread keeps a turnstile around in case it needs to block.
340 * The only reason the turnstile is not simply part of the thread
341 * structure is that we may have to break the association whenever
342 * more than one thread blocks on a given synchronization object.
343 * From a memory-management standpoint, turnstiles are like the
344 * "attached mblks" that hang off dblks in the streams allocator.
345 */
346 ts = kmem_cache_alloc(turnstile_cache, KM_SLEEP);
347
348 if (stk == NULL) {
349 /*
350 * alloc both thread and stack in segkp chunk
351 */
352
353 if (stksize < default_stksize)
354 stksize = default_stksize;
355
356 if (stksize == default_stksize) {
357 stk = (caddr_t)segkp_cache_get(segkp_thread);
358 } else {
359 stksize = roundup(stksize, PAGESIZE);
360 stk = (caddr_t)segkp_get(segkp, stksize,
361 (KPD_HASREDZONE | KPD_NO_ANON | KPD_LOCKED));
362 }
363
364 ASSERT(stk != NULL);
365
366 /*
367 * The machine-dependent mutex code may require that
368 * thread pointers (since they may be used for mutex owner
369 * fields) have certain alignment requirements.
370 * PTR24_ALIGN is the size of the alignment quanta.
371 * XXX - assumes stack grows toward low addresses.
372 */
373 if (stksize <= sizeof (kthread_t) + PTR24_ALIGN)
374 cmn_err(CE_PANIC, "thread_create: proposed stack size"
375 " too small to hold thread.");
376 #ifdef STACK_GROWTH_DOWN
377 stksize -= SA(sizeof (kthread_t) + PTR24_ALIGN - 1);
378 stksize &= -PTR24_ALIGN; /* make thread aligned */
379 t = (kthread_t *)(stk + stksize);
380 bzero(t, sizeof (kthread_t));
381 if (audit_active)
382 audit_thread_create(t);
383 t->t_stk = stk + stksize;
384 t->t_stkbase = stk;
385 #else /* stack grows to larger addresses */
386 stksize -= SA(sizeof (kthread_t));
387 t = (kthread_t *)(stk);
388 bzero(t, sizeof (kthread_t));
389 t->t_stk = stk + sizeof (kthread_t);
390 t->t_stkbase = stk + stksize + sizeof (kthread_t);
391 #endif /* STACK_GROWTH_DOWN */
392 t->t_flag |= T_TALLOCSTK;
393 t->t_swap = stk;
394 } else {
395 t = kmem_cache_alloc(thread_cache, KM_SLEEP);
396 bzero(t, sizeof (kthread_t));
397 ASSERT(((uintptr_t)t & (PTR24_ALIGN - 1)) == 0);
398 if (audit_active)
399 audit_thread_create(t);
400 /*
401 * Initialize t_stk to the kernel stack pointer to use
402 * upon entry to the kernel
403 */
404 #ifdef STACK_GROWTH_DOWN
405 t->t_stk = stk + stksize;
406 t->t_stkbase = stk;
407 #else
408 t->t_stk = stk; /* 3b2-like */
409 t->t_stkbase = stk + stksize;
410 #endif /* STACK_GROWTH_DOWN */
411 }
412
413 if (kmem_stackinfo != 0) {
414 stkinfo_begin(t);
415 }
416
417 t->t_ts = ts;
418
419 /*
420 * p_cred could be NULL if it thread_create is called before cred_init
421 * is called in main.
422 */
423 mutex_enter(&pp->p_crlock);
424 if (pp->p_cred)
425 crhold(t->t_cred = pp->p_cred);
426 mutex_exit(&pp->p_crlock);
427 t->t_start = gethrestime_sec();
428 t->t_startpc = proc;
429 t->t_procp = pp;
430 t->t_clfuncs = &sys_classfuncs.thread;
431 t->t_cid = syscid;
432 t->t_pri = pri;
433 t->t_stime = ddi_get_lbolt();
434 t->t_schedflag = TS_LOAD | TS_DONT_SWAP;
435 t->t_bind_cpu = PBIND_NONE;
436 t->t_bindflag = (uchar_t)default_binding_mode;
437 t->t_bind_pset = PS_NONE;
438 t->t_plockp = &pp->p_lock;
439 t->t_copyops = NULL;
440 t->t_taskq = NULL;
441 t->t_anttime = 0;
442 t->t_hatdepth = 0;
443
444 t->t_dtrace_vtime = 1; /* assure vtimestamp is always non-zero */
445
446 CPU_STATS_ADDQ(CPU, sys, nthreads, 1);
447 #ifndef NPROBE
448 /* Kernel probe */
449 tnf_thread_create(t);
450 #endif /* NPROBE */
451 LOCK_INIT_CLEAR(&t->t_lock);
452
453 /*
454 * Callers who give us a NULL proc must do their own
455 * stack initialization. e.g. lwp_create()
456 */
457 if (proc != NULL) {
458 t->t_stk = thread_stk_init(t->t_stk);
459 thread_load(t, proc, arg, len);
460 }
461
462 /*
463 * Put a hold on project0. If this thread is actually in a
464 * different project, then t_proj will be changed later in
465 * lwp_create(). All kernel-only threads must be in project 0.
466 */
467 t->t_proj = project_hold(proj0p);
468
469 lgrp_affinity_init(&t->t_lgrp_affinity);
470
471 mutex_enter(&pidlock);
472 nthread++;
473 t->t_did = next_t_id++;
474 t->t_prev = curthread->t_prev;
475 t->t_next = curthread;
476
477 /*
478 * Add the thread to the list of all threads, and initialize
479 * its t_cpu pointer. We need to block preemption since
480 * cpu_offline walks the thread list looking for threads
481 * with t_cpu pointing to the CPU being offlined. We want
482 * to make sure that the list is consistent and that if t_cpu
483 * is set, the thread is on the list.
484 */
485 kpreempt_disable();
486 curthread->t_prev->t_next = t;
487 curthread->t_prev = t;
488
489 /*
490 * Threads should never have a NULL t_cpu pointer so assign it
491 * here. If the thread is being created with state TS_RUN a
492 * better CPU may be chosen when it is placed on the run queue.
493 *
494 * We need to keep kernel preemption disabled when setting all
495 * three fields to keep them in sync. Also, always create in
496 * the default partition since that's where kernel threads go
497 * (if this isn't a kernel thread, t_cpupart will be changed
498 * in lwp_create before setting the thread runnable).
499 */
500 t->t_cpupart = &cp_default;
501
502 /*
503 * For now, affiliate this thread with the root lgroup.
504 * Since the kernel does not (presently) allocate its memory
505 * in a locality aware fashion, the root is an appropriate home.
506 * If this thread is later associated with an lwp, it will have
507 * it's lgroup re-assigned at that time.
508 */
509 lgrp_move_thread(t, &cp_default.cp_lgrploads[LGRP_ROOTID], 1);
510
511 /*
512 * Inherit the current cpu. If this cpu isn't part of the chosen
513 * lgroup, a new cpu will be chosen by cpu_choose when the thread
514 * is ready to run.
515 */
516 if (CPU->cpu_part == &cp_default)
517 t->t_cpu = CPU;
518 else
519 t->t_cpu = disp_lowpri_cpu(cp_default.cp_cpulist, t->t_lpl,
520 t->t_pri, NULL);
521
522 t->t_disp_queue = t->t_cpu->cpu_disp;
523 kpreempt_enable();
524
525 /*
526 * Initialize thread state and the dispatcher lock pointer.
527 * Need to hold onto pidlock to block allthreads walkers until
528 * the state is set.
529 */
530 switch (state) {
531 case TS_RUN:
532 curthread->t_oldspl = splhigh(); /* get dispatcher spl */
533 THREAD_SET_STATE(t, TS_STOPPED, &transition_lock);
534 CL_SETRUN(t);
535 thread_unlock(t);
536 break;
537
538 case TS_ONPROC:
539 THREAD_ONPROC(t, t->t_cpu);
540 break;
541
542 case TS_FREE:
543 /*
544 * Free state will be used for intr threads.
545 * The interrupt routine must set the thread dispatcher
546 * lock pointer (t_lockp) if starting on a CPU
547 * other than the current one.
548 */
549 THREAD_FREEINTR(t, CPU);
550 break;
551
552 case TS_STOPPED:
553 THREAD_SET_STATE(t, TS_STOPPED, &stop_lock);
554 break;
555
556 default: /* TS_SLEEP, TS_ZOMB or TS_TRANS */
557 cmn_err(CE_PANIC, "thread_create: invalid state %d", state);
558 }
559 mutex_exit(&pidlock);
560 return (t);
561 }
562
563 /*
564 * Move thread to project0 and take care of project reference counters.
565 */
566 void
567 thread_rele(kthread_t *t)
568 {
569 kproject_t *kpj;
570
571 thread_lock(t);
572
573 ASSERT(t == curthread || t->t_state == TS_FREE || t->t_procp == &p0);
574 kpj = ttoproj(t);
575 t->t_proj = proj0p;
576
577 thread_unlock(t);
578
579 if (kpj != proj0p) {
580 project_rele(kpj);
581 (void) project_hold(proj0p);
582 }
583 }
584
585 void
586 thread_exit(void)
587 {
588 kthread_t *t = curthread;
589
590 if ((t->t_proc_flag & TP_ZTHREAD) != 0)
591 cmn_err(CE_PANIC, "thread_exit: zthread_exit() not called");
592
593 tsd_exit(); /* Clean up this thread's TSD */
594
595 kcpc_passivate(); /* clean up performance counter state */
596
597 /*
598 * No kernel thread should have called poll() without arranging
599 * calling pollcleanup() here.
600 */
601 ASSERT(t->t_pollstate == NULL);
602 ASSERT(t->t_schedctl == NULL);
603 if (t->t_door)
604 door_slam(); /* in case thread did an upcall */
605
606 #ifndef NPROBE
607 /* Kernel probe */
608 if (t->t_tnf_tpdp)
609 tnf_thread_exit();
610 #endif /* NPROBE */
611
612 thread_rele(t);
613 t->t_preempt++;
614
615 /*
616 * remove thread from the all threads list so that
617 * death-row can use the same pointers.
618 */
619 mutex_enter(&pidlock);
620 t->t_next->t_prev = t->t_prev;
621 t->t_prev->t_next = t->t_next;
622 ASSERT(allthreads != t); /* t0 never exits */
623 cv_broadcast(&t->t_joincv); /* wake up anyone in thread_join */
624 mutex_exit(&pidlock);
625
626 if (t->t_ctx != NULL)
627 exitctx(t);
628 if (t->t_procp->p_pctx != NULL)
629 exitpctx(t->t_procp);
630
631 if (kmem_stackinfo != 0) {
632 stkinfo_end(t);
633 }
634
635 t->t_state = TS_ZOMB; /* set zombie thread */
636
637 swtch_from_zombie(); /* give up the CPU */
638 /* NOTREACHED */
639 }
640
641 /*
642 * Check to see if the specified thread is active (defined as being on
643 * the thread list). This is certainly a slow way to do this; if there's
644 * ever a reason to speed it up, we could maintain a hash table of active
645 * threads indexed by their t_did.
646 */
647 static kthread_t *
648 did_to_thread(kt_did_t tid)
649 {
650 kthread_t *t;
651
652 ASSERT(MUTEX_HELD(&pidlock));
653 for (t = curthread->t_next; t != curthread; t = t->t_next) {
654 if (t->t_did == tid)
655 break;
656 }
657 if (t->t_did == tid)
658 return (t);
659 else
660 return (NULL);
661 }
662
663 /*
664 * Wait for specified thread to exit. Returns immediately if the thread
665 * could not be found, meaning that it has either already exited or never
666 * existed.
667 */
668 void
669 thread_join(kt_did_t tid)
670 {
671 kthread_t *t;
672
673 ASSERT(tid != curthread->t_did);
674 ASSERT(tid != t0.t_did);
675
676 mutex_enter(&pidlock);
677 /*
678 * Make sure we check that the thread is on the thread list
679 * before blocking on it; otherwise we could end up blocking on
680 * a cv that's already been freed. In other words, don't cache
681 * the thread pointer across calls to cv_wait.
682 *
683 * The choice of loop invariant means that whenever a thread
684 * is taken off the allthreads list, a cv_broadcast must be
685 * performed on that thread's t_joincv to wake up any waiters.
686 * The broadcast doesn't have to happen right away, but it
687 * shouldn't be postponed indefinitely (e.g., by doing it in
688 * thread_free which may only be executed when the deathrow
689 * queue is processed.
690 */
691 while (t = did_to_thread(tid))
692 cv_wait(&t->t_joincv, &pidlock);
693 mutex_exit(&pidlock);
694 }
695
696 void
697 thread_free_prevent(kthread_t *t)
698 {
699 kmutex_t *lp;
700
701 lp = &thread_free_lock[THREAD_FREE_HASH(t)].tf_lock;
702 mutex_enter(lp);
703 }
704
705 void
706 thread_free_allow(kthread_t *t)
707 {
708 kmutex_t *lp;
709
710 lp = &thread_free_lock[THREAD_FREE_HASH(t)].tf_lock;
711 mutex_exit(lp);
712 }
713
714 static void
715 thread_free_barrier(kthread_t *t)
716 {
717 kmutex_t *lp;
718
719 lp = &thread_free_lock[THREAD_FREE_HASH(t)].tf_lock;
720 mutex_enter(lp);
721 mutex_exit(lp);
722 }
723
724 void
725 thread_free(kthread_t *t)
726 {
727 boolean_t allocstk = (t->t_flag & T_TALLOCSTK);
728 klwp_t *lwp = t->t_lwp;
729 caddr_t swap = t->t_swap;
730
731 ASSERT(t != &t0 && t->t_state == TS_FREE);
732 ASSERT(t->t_door == NULL);
733 ASSERT(t->t_schedctl == NULL);
734 ASSERT(t->t_pollstate == NULL);
735
736 t->t_pri = 0;
737 t->t_pc = 0;
738 t->t_sp = 0;
739 t->t_wchan0 = NULL;
740 t->t_wchan = NULL;
741 if (t->t_cred != NULL) {
742 crfree(t->t_cred);
743 t->t_cred = 0;
744 }
745 if (t->t_pdmsg) {
746 kmem_free(t->t_pdmsg, strlen(t->t_pdmsg) + 1);
747 t->t_pdmsg = NULL;
748 }
749 if (audit_active)
750 audit_thread_free(t);
751 #ifndef NPROBE
752 if (t->t_tnf_tpdp)
753 tnf_thread_free(t);
754 #endif /* NPROBE */
755 if (t->t_cldata) {
756 CL_EXITCLASS(t->t_cid, (caddr_t *)t->t_cldata);
757 }
758 if (t->t_rprof != NULL) {
759 kmem_free(t->t_rprof, sizeof (*t->t_rprof));
760 t->t_rprof = NULL;
761 }
762 t->t_lockp = NULL; /* nothing should try to lock this thread now */
763 if (lwp)
764 lwp_freeregs(lwp, 0);
765 if (t->t_ctx)
766 freectx(t, 0);
767 t->t_stk = NULL;
768 if (lwp)
769 lwp_stk_fini(lwp);
770 lock_clear(&t->t_lock);
771
772 if (t->t_ts->ts_waiters > 0)
773 panic("thread_free: turnstile still active");
774
775 kmem_cache_free(turnstile_cache, t->t_ts);
776
777 free_afd(&t->t_activefd);
778
779 /*
780 * Barrier for the tick accounting code. The tick accounting code
781 * holds this lock to keep the thread from going away while it's
782 * looking at it.
783 */
784 thread_free_barrier(t);
785
786 ASSERT(ttoproj(t) == proj0p);
787 project_rele(ttoproj(t));
788
789 lgrp_affinity_free(&t->t_lgrp_affinity);
790
791 mutex_enter(&pidlock);
792 nthread--;
793 mutex_exit(&pidlock);
794
795 if (t->t_name != NULL) {
796 kmem_free(t->t_name, THREAD_NAME_MAX);
797 t->t_name = NULL;
798 }
799
800 /*
801 * Free thread, lwp and stack. This needs to be done carefully, since
802 * if T_TALLOCSTK is set, the thread is part of the stack.
803 */
804 t->t_lwp = NULL;
805 t->t_swap = NULL;
806
807 if (swap) {
808 segkp_release(segkp, swap);
809 }
810 if (lwp) {
811 kmem_cache_free(lwp_cache, lwp);
812 }
813 if (!allocstk) {
814 kmem_cache_free(thread_cache, t);
815 }
816 }
817
818 /*
819 * Removes threads associated with the given zone from a deathrow queue.
820 * tp is a pointer to the head of the deathrow queue, and countp is a
821 * pointer to the current deathrow count. Returns a linked list of
822 * threads removed from the list.
823 */
824 static kthread_t *
825 thread_zone_cleanup(kthread_t **tp, int *countp, zoneid_t zoneid)
826 {
827 kthread_t *tmp, *list = NULL;
828 cred_t *cr;
829
830 ASSERT(MUTEX_HELD(&reaplock));
831 while (*tp != NULL) {
832 if ((cr = (*tp)->t_cred) != NULL && crgetzoneid(cr) == zoneid) {
833 tmp = *tp;
834 *tp = tmp->t_forw;
835 tmp->t_forw = list;
836 list = tmp;
837 (*countp)--;
838 } else {
839 tp = &(*tp)->t_forw;
840 }
841 }
842 return (list);
843 }
844
845 static void
846 thread_reap_list(kthread_t *t)
847 {
848 kthread_t *next;
849
850 while (t != NULL) {
851 next = t->t_forw;
852 thread_free(t);
853 t = next;
854 }
855 }
856
857 /* ARGSUSED */
858 static void
859 thread_zone_destroy(zoneid_t zoneid, void *unused)
860 {
861 kthread_t *t, *l;
862
863 mutex_enter(&reaplock);
864 /*
865 * Pull threads and lwps associated with zone off deathrow lists.
866 */
867 t = thread_zone_cleanup(&thread_deathrow, &thread_reapcnt, zoneid);
868 l = thread_zone_cleanup(&lwp_deathrow, &lwp_reapcnt, zoneid);
869 mutex_exit(&reaplock);
870
871 /*
872 * Guard against race condition in mutex_owner_running:
873 * thread=owner(mutex)
874 * <interrupt>
875 * thread exits mutex
876 * thread exits
877 * thread reaped
878 * thread struct freed
879 * cpu = thread->t_cpu <- BAD POINTER DEREFERENCE.
880 * A cross call to all cpus will cause the interrupt handler
881 * to reset the PC if it is in mutex_owner_running, refreshing
882 * stale thread pointers.
883 */
884 mutex_sync(); /* sync with mutex code */
885
886 /*
887 * Reap threads
888 */
889 thread_reap_list(t);
890
891 /*
892 * Reap lwps
893 */
894 thread_reap_list(l);
895 }
896
897 /*
898 * cleanup zombie threads that are on deathrow.
899 */
900 void
901 thread_reaper()
902 {
903 kthread_t *t, *l;
904 callb_cpr_t cprinfo;
905
906 /*
907 * Register callback to clean up threads when zone is destroyed.
908 */
909 zone_key_create(&zone_thread_key, NULL, NULL, thread_zone_destroy);
910
911 CALLB_CPR_INIT(&cprinfo, &reaplock, callb_generic_cpr, "t_reaper");
912 for (;;) {
913 mutex_enter(&reaplock);
914 while (thread_deathrow == NULL && lwp_deathrow == NULL) {
915 CALLB_CPR_SAFE_BEGIN(&cprinfo);
916 cv_wait(&reaper_cv, &reaplock);
917 CALLB_CPR_SAFE_END(&cprinfo, &reaplock);
918 }
919 /*
920 * mutex_sync() needs to be called when reaping, but
921 * not too often. We limit reaping rate to once
922 * per second. Reaplimit is max rate at which threads can
923 * be freed. Does not impact thread destruction/creation.
924 */
925 t = thread_deathrow;
926 l = lwp_deathrow;
927 thread_deathrow = NULL;
928 lwp_deathrow = NULL;
929 thread_reapcnt = 0;
930 lwp_reapcnt = 0;
931 mutex_exit(&reaplock);
932
933 /*
934 * Guard against race condition in mutex_owner_running:
935 * thread=owner(mutex)
936 * <interrupt>
937 * thread exits mutex
938 * thread exits
939 * thread reaped
940 * thread struct freed
941 * cpu = thread->t_cpu <- BAD POINTER DEREFERENCE.
942 * A cross call to all cpus will cause the interrupt handler
943 * to reset the PC if it is in mutex_owner_running, refreshing
944 * stale thread pointers.
945 */
946 mutex_sync(); /* sync with mutex code */
947 /*
948 * Reap threads
949 */
950 thread_reap_list(t);
951
952 /*
953 * Reap lwps
954 */
955 thread_reap_list(l);
956 delay(hz);
957 }
958 }
959
960 /*
961 * This is called by lwpcreate, etc.() to put a lwp_deathrow thread onto
962 * thread_deathrow. The thread's state is changed already TS_FREE to indicate
963 * that is reapable. The thread already holds the reaplock, and was already
964 * freed.
965 */
966 void
967 reapq_move_lq_to_tq(kthread_t *t)
968 {
969 ASSERT(t->t_state == TS_FREE);
970 ASSERT(MUTEX_HELD(&reaplock));
971 t->t_forw = thread_deathrow;
972 thread_deathrow = t;
973 thread_reapcnt++;
974 if (lwp_reapcnt + thread_reapcnt > reaplimit)
975 cv_signal(&reaper_cv); /* wake the reaper */
976 }
977
978 /*
979 * This is called by resume() to put a zombie thread onto deathrow.
980 * The thread's state is changed to TS_FREE to indicate that is reapable.
981 * This is called from the idle thread so it must not block - just spin.
982 */
983 void
984 reapq_add(kthread_t *t)
985 {
986 mutex_enter(&reaplock);
987
988 /*
989 * lwp_deathrow contains threads with lwp linkage and
990 * swappable thread stacks which have the default stacksize.
991 * These threads' lwps and stacks may be reused by lwp_create().
992 *
993 * Anything else goes on thread_deathrow(), where it will eventually
994 * be thread_free()d.
995 */
996 if (t->t_flag & T_LWPREUSE) {
997 ASSERT(ttolwp(t) != NULL);
998 t->t_forw = lwp_deathrow;
999 lwp_deathrow = t;
1000 lwp_reapcnt++;
1001 } else {
1002 t->t_forw = thread_deathrow;
1003 thread_deathrow = t;
1004 thread_reapcnt++;
1005 }
1006 if (lwp_reapcnt + thread_reapcnt > reaplimit)
1007 cv_signal(&reaper_cv); /* wake the reaper */
1008 t->t_state = TS_FREE;
1009 lock_clear(&t->t_lock);
1010
1011 /*
1012 * Before we return, we need to grab and drop the thread lock for
1013 * the dead thread. At this point, the current thread is the idle
1014 * thread, and the dead thread's CPU lock points to the current
1015 * CPU -- and we must grab and drop the lock to synchronize with
1016 * a racing thread walking a blocking chain that the zombie thread
1017 * was recently in. By this point, that blocking chain is (by
1018 * definition) stale: the dead thread is not holding any locks, and
1019 * is therefore not in any blocking chains -- but if we do not regrab
1020 * our lock before freeing the dead thread's data structures, the
1021 * thread walking the (stale) blocking chain will die on memory
1022 * corruption when it attempts to drop the dead thread's lock. We
1023 * only need do this once because there is no way for the dead thread
1024 * to ever again be on a blocking chain: once we have grabbed and
1025 * dropped the thread lock, we are guaranteed that anyone that could
1026 * have seen this thread in a blocking chain can no longer see it.
1027 */
1028 thread_lock(t);
1029 thread_unlock(t);
1030
1031 mutex_exit(&reaplock);
1032 }
1033
1034 /*
1035 * Install thread context ops for the current thread.
1036 */
1037 void
1038 installctx(
1039 kthread_t *t,
1040 void *arg,
1041 void (*save)(void *),
1042 void (*restore)(void *),
1043 void (*fork)(void *, void *),
1044 void (*lwp_create)(void *, void *),
1045 void (*exit)(void *),
1046 void (*free)(void *, int))
1047 {
1048 struct ctxop *ctx;
1049
1050 ctx = kmem_alloc(sizeof (struct ctxop), KM_SLEEP);
1051 ctx->save_op = save;
1052 ctx->restore_op = restore;
1053 ctx->fork_op = fork;
1054 ctx->lwp_create_op = lwp_create;
1055 ctx->exit_op = exit;
1056 ctx->free_op = free;
1057 ctx->arg = arg;
1058 ctx->next = t->t_ctx;
1059 t->t_ctx = ctx;
1060 }
1061
1062 /*
1063 * Remove the thread context ops from a thread.
1064 */
1065 int
1066 removectx(
1067 kthread_t *t,
1068 void *arg,
1069 void (*save)(void *),
1070 void (*restore)(void *),
1071 void (*fork)(void *, void *),
1072 void (*lwp_create)(void *, void *),
1073 void (*exit)(void *),
1074 void (*free)(void *, int))
1075 {
1076 struct ctxop *ctx, *prev_ctx;
1077
1078 /*
1079 * The incoming kthread_t (which is the thread for which the
1080 * context ops will be removed) should be one of the following:
1081 *
1082 * a) the current thread,
1083 *
1084 * b) a thread of a process that's being forked (SIDL),
1085 *
1086 * c) a thread that belongs to the same process as the current
1087 * thread and for which the current thread is the agent thread,
1088 *
1089 * d) a thread that is TS_STOPPED which is indicative of it
1090 * being (if curthread is not an agent) a thread being created
1091 * as part of an lwp creation.
1092 */
1093 ASSERT(t == curthread || ttoproc(t)->p_stat == SIDL ||
1094 ttoproc(t)->p_agenttp == curthread || t->t_state == TS_STOPPED);
1095
1096 /*
1097 * Serialize modifications to t->t_ctx to prevent the agent thread
1098 * and the target thread from racing with each other during lwp exit.
1099 */
1100 mutex_enter(&t->t_ctx_lock);
1101 prev_ctx = NULL;
1102 kpreempt_disable();
1103 for (ctx = t->t_ctx; ctx != NULL; ctx = ctx->next) {
1104 if (ctx->save_op == save && ctx->restore_op == restore &&
1105 ctx->fork_op == fork && ctx->lwp_create_op == lwp_create &&
1106 ctx->exit_op == exit && ctx->free_op == free &&
1107 ctx->arg == arg) {
1108 if (prev_ctx)
1109 prev_ctx->next = ctx->next;
1110 else
1111 t->t_ctx = ctx->next;
1112 mutex_exit(&t->t_ctx_lock);
1113 if (ctx->free_op != NULL)
1114 (ctx->free_op)(ctx->arg, 0);
1115 kmem_free(ctx, sizeof (struct ctxop));
1116 kpreempt_enable();
1117 return (1);
1118 }
1119 prev_ctx = ctx;
1120 }
1121 mutex_exit(&t->t_ctx_lock);
1122 kpreempt_enable();
1123
1124 return (0);
1125 }
1126
1127 void
1128 savectx(kthread_t *t)
1129 {
1130 struct ctxop *ctx;
1131
1132 ASSERT(t == curthread);
1133 for (ctx = t->t_ctx; ctx != 0; ctx = ctx->next)
1134 if (ctx->save_op != NULL)
1135 (ctx->save_op)(ctx->arg);
1136 }
1137
1138 void
1139 restorectx(kthread_t *t)
1140 {
1141 struct ctxop *ctx;
1142
1143 ASSERT(t == curthread);
1144 for (ctx = t->t_ctx; ctx != 0; ctx = ctx->next)
1145 if (ctx->restore_op != NULL)
1146 (ctx->restore_op)(ctx->arg);
1147 }
1148
1149 void
1150 forkctx(kthread_t *t, kthread_t *ct)
1151 {
1152 struct ctxop *ctx;
1153
1154 for (ctx = t->t_ctx; ctx != NULL; ctx = ctx->next)
1155 if (ctx->fork_op != NULL)
1156 (ctx->fork_op)(t, ct);
1157 }
1158
1159 /*
1160 * Note that this operator is only invoked via the _lwp_create
1161 * system call. The system may have other reasons to create lwps
1162 * e.g. the agent lwp or the doors unreferenced lwp.
1163 */
1164 void
1165 lwp_createctx(kthread_t *t, kthread_t *ct)
1166 {
1167 struct ctxop *ctx;
1168
1169 for (ctx = t->t_ctx; ctx != NULL; ctx = ctx->next)
1170 if (ctx->lwp_create_op != NULL)
1171 (ctx->lwp_create_op)(t, ct);
1172 }
1173
1174 /*
1175 * exitctx is called from thread_exit() and lwp_exit() to perform any actions
1176 * needed when the thread/LWP leaves the processor for the last time. This
1177 * routine is not intended to deal with freeing memory; freectx() is used for
1178 * that purpose during thread_free(). This routine is provided to allow for
1179 * clean-up that can't wait until thread_free().
1180 */
1181 void
1182 exitctx(kthread_t *t)
1183 {
1184 struct ctxop *ctx;
1185
1186 for (ctx = t->t_ctx; ctx != NULL; ctx = ctx->next)
1187 if (ctx->exit_op != NULL)
1188 (ctx->exit_op)(t);
1189 }
1190
1191 /*
1192 * freectx is called from thread_free() and exec() to get
1193 * rid of old thread context ops.
1194 */
1195 void
1196 freectx(kthread_t *t, int isexec)
1197 {
1198 struct ctxop *ctx;
1199
1200 kpreempt_disable();
1201 while ((ctx = t->t_ctx) != NULL) {
1202 t->t_ctx = ctx->next;
1203 if (ctx->free_op != NULL)
1204 (ctx->free_op)(ctx->arg, isexec);
1205 kmem_free(ctx, sizeof (struct ctxop));
1206 }
1207 kpreempt_enable();
1208 }
1209
1210 /*
1211 * freectx_ctx is called from lwp_create() when lwp is reused from
1212 * lwp_deathrow and its thread structure is added to thread_deathrow.
1213 * The thread structure to which this ctx was attached may be already
1214 * freed by the thread reaper so free_op implementations shouldn't rely
1215 * on thread structure to which this ctx was attached still being around.
1216 */
1217 void
1218 freectx_ctx(struct ctxop *ctx)
1219 {
1220 struct ctxop *nctx;
1221
1222 ASSERT(ctx != NULL);
1223
1224 kpreempt_disable();
1225 do {
1226 nctx = ctx->next;
1227 if (ctx->free_op != NULL)
1228 (ctx->free_op)(ctx->arg, 0);
1229 kmem_free(ctx, sizeof (struct ctxop));
1230 } while ((ctx = nctx) != NULL);
1231 kpreempt_enable();
1232 }
1233
1234 /*
1235 * Set the thread running; arrange for it to be swapped in if necessary.
1236 */
1237 void
1238 setrun_locked(kthread_t *t)
1239 {
1240 ASSERT(THREAD_LOCK_HELD(t));
1241 if (t->t_state == TS_SLEEP) {
1242 /*
1243 * Take off sleep queue.
1244 */
1245 SOBJ_UNSLEEP(t->t_sobj_ops, t);
1246 } else if (t->t_state & (TS_RUN | TS_ONPROC)) {
1247 /*
1248 * Already on dispatcher queue.
1249 */
1250 return;
1251 } else if (t->t_state == TS_WAIT) {
1252 waitq_setrun(t);
1253 } else if (t->t_state == TS_STOPPED) {
1254 /*
1255 * All of the sending of SIGCONT (TC_XSTART) and /proc
1256 * (TC_PSTART) and lwp_continue() (TC_CSTART) must have
1257 * requested that the thread be run.
1258 * Just calling setrun() is not sufficient to set a stopped
1259 * thread running. TP_TXSTART is always set if the thread
1260 * is not stopped by a jobcontrol stop signal.
1261 * TP_TPSTART is always set if /proc is not controlling it.
1262 * TP_TCSTART is always set if lwp_suspend() didn't stop it.
1263 * The thread won't be stopped unless one of these
1264 * three mechanisms did it.
1265 *
1266 * These flags must be set before calling setrun_locked(t).
1267 * They can't be passed as arguments because the streams
1268 * code calls setrun() indirectly and the mechanism for
1269 * doing so admits only one argument. Note that the
1270 * thread must be locked in order to change t_schedflags.
1271 */
1272 if ((t->t_schedflag & TS_ALLSTART) != TS_ALLSTART)
1273 return;
1274 /*
1275 * Process is no longer stopped (a thread is running).
1276 */
1277 t->t_whystop = 0;
1278 t->t_whatstop = 0;
1279 /*
1280 * Strictly speaking, we do not have to clear these
1281 * flags here; they are cleared on entry to stop().
1282 * However, they are confusing when doing kernel
1283 * debugging or when they are revealed by ps(1).
1284 */
1285 t->t_schedflag &= ~TS_ALLSTART;
1286 THREAD_TRANSITION(t); /* drop stopped-thread lock */
1287 ASSERT(t->t_lockp == &transition_lock);
1288 ASSERT(t->t_wchan0 == NULL && t->t_wchan == NULL);
1289 /*
1290 * Let the class put the process on the dispatcher queue.
1291 */
1292 CL_SETRUN(t);
1293 }
1294 }
1295
1296 void
1297 setrun(kthread_t *t)
1298 {
1299 thread_lock(t);
1300 setrun_locked(t);
1301 thread_unlock(t);
1302 }
1303
1304 /*
1305 * Unpin an interrupted thread.
1306 * When an interrupt occurs, the interrupt is handled on the stack
1307 * of an interrupt thread, taken from a pool linked to the CPU structure.
1308 *
1309 * When swtch() is switching away from an interrupt thread because it
1310 * blocked or was preempted, this routine is called to complete the
1311 * saving of the interrupted thread state, and returns the interrupted
1312 * thread pointer so it may be resumed.
1313 *
1314 * Called by swtch() only at high spl.
1315 */
1316 kthread_t *
1317 thread_unpin()
1318 {
1319 kthread_t *t = curthread; /* current thread */
1320 kthread_t *itp; /* interrupted thread */
1321 int i; /* interrupt level */
1322 extern int intr_passivate();
1323
1324 ASSERT(t->t_intr != NULL);
1325
1326 itp = t->t_intr; /* interrupted thread */
1327 t->t_intr = NULL; /* clear interrupt ptr */
1328
1329 /*
1330 * Get state from interrupt thread for the one
1331 * it interrupted.
1332 */
1333
1334 i = intr_passivate(t, itp);
1335
1336 TRACE_5(TR_FAC_INTR, TR_INTR_PASSIVATE,
1337 "intr_passivate:level %d curthread %p (%T) ithread %p (%T)",
1338 i, t, t, itp, itp);
1339
1340 /*
1341 * Dissociate the current thread from the interrupted thread's LWP.
1342 */
1343 t->t_lwp = NULL;
1344
1345 /*
1346 * Interrupt handlers above the level that spinlocks block must
1347 * not block.
1348 */
1349 #if DEBUG
1350 if (i < 0 || i > LOCK_LEVEL)
1351 cmn_err(CE_PANIC, "thread_unpin: ipl out of range %x", i);
1352 #endif
1353
1354 /*
1355 * Compute the CPU's base interrupt level based on the active
1356 * interrupts.
1357 */
1358 ASSERT(CPU->cpu_intr_actv & (1 << i));
1359 set_base_spl();
1360
1361 return (itp);
1362 }
1363
1364 /*
1365 * Create and initialize an interrupt thread.
1366 * Returns non-zero on error.
1367 * Called at spl7() or better.
1368 */
1369 void
1370 thread_create_intr(struct cpu *cp)
1371 {
1372 kthread_t *tp;
1373
1374 tp = thread_create(NULL, 0,
1375 (void (*)())thread_create_intr, NULL, 0, &p0, TS_ONPROC, 0);
1376
1377 /*
1378 * Set the thread in the TS_FREE state. The state will change
1379 * to TS_ONPROC only while the interrupt is active. Think of these
1380 * as being on a private free list for the CPU. Being TS_FREE keeps
1381 * inactive interrupt threads out of debugger thread lists.
1382 *
1383 * We cannot call thread_create with TS_FREE because of the current
1384 * checks there for ONPROC. Fix this when thread_create takes flags.
1385 */
1386 THREAD_FREEINTR(tp, cp);
1387
1388 /*
1389 * Nobody should ever reference the credentials of an interrupt
1390 * thread so make it NULL to catch any such references.
1391 */
1392 tp->t_cred = NULL;
1393 tp->t_flag |= T_INTR_THREAD;
1394 tp->t_cpu = cp;
1395 tp->t_bound_cpu = cp;
1396 tp->t_disp_queue = cp->cpu_disp;
1397 tp->t_affinitycnt = 1;
1398 tp->t_preempt = 1;
1399
1400 /*
1401 * Don't make a user-requested binding on this thread so that
1402 * the processor can be offlined.
1403 */
1404 tp->t_bind_cpu = PBIND_NONE; /* no USER-requested binding */
1405 tp->t_bind_pset = PS_NONE;
1406
1407 #if defined(__i386) || defined(__amd64)
1408 tp->t_stk -= STACK_ALIGN;
1409 *(tp->t_stk) = 0; /* terminate intr thread stack */
1410 #endif
1411
1412 /*
1413 * Link onto CPU's interrupt pool.
1414 */
1415 tp->t_link = cp->cpu_intr_thread;
1416 cp->cpu_intr_thread = tp;
1417 }
1418
1419 /*
1420 * TSD -- THREAD SPECIFIC DATA
1421 */
1422 static kmutex_t tsd_mutex; /* linked list spin lock */
1423 static uint_t tsd_nkeys; /* size of destructor array */
1424 /* per-key destructor funcs */
1425 static void (**tsd_destructor)(void *);
1426 /* list of tsd_thread's */
1427 static struct tsd_thread *tsd_list;
1428
1429 /*
1430 * Default destructor
1431 * Needed because NULL destructor means that the key is unused
1432 */
1433 /* ARGSUSED */
1434 void
1435 tsd_defaultdestructor(void *value)
1436 {}
1437
1438 /*
1439 * Create a key (index into per thread array)
1440 * Locks out tsd_create, tsd_destroy, and tsd_exit
1441 * May allocate memory with lock held
1442 */
1443 void
1444 tsd_create(uint_t *keyp, void (*destructor)(void *))
1445 {
1446 int i;
1447 uint_t nkeys;
1448
1449 /*
1450 * if key is allocated, do nothing
1451 */
1452 mutex_enter(&tsd_mutex);
1453 if (*keyp) {
1454 mutex_exit(&tsd_mutex);
1455 return;
1456 }
1457 /*
1458 * find an unused key
1459 */
1460 if (destructor == NULL)
1461 destructor = tsd_defaultdestructor;
1462
1463 for (i = 0; i < tsd_nkeys; ++i)
1464 if (tsd_destructor[i] == NULL)
1465 break;
1466
1467 /*
1468 * if no unused keys, increase the size of the destructor array
1469 */
1470 if (i == tsd_nkeys) {
1471 if ((nkeys = (tsd_nkeys << 1)) == 0)
1472 nkeys = 1;
1473 tsd_destructor =
1474 (void (**)(void *))tsd_realloc((void *)tsd_destructor,
1475 (size_t)(tsd_nkeys * sizeof (void (*)(void *))),
1476 (size_t)(nkeys * sizeof (void (*)(void *))));
1477 tsd_nkeys = nkeys;
1478 }
1479
1480 /*
1481 * allocate the next available unused key
1482 */
1483 tsd_destructor[i] = destructor;
1484 *keyp = i + 1;
1485 mutex_exit(&tsd_mutex);
1486 }
1487
1488 /*
1489 * Destroy a key -- this is for unloadable modules
1490 *
1491 * Assumes that the caller is preventing tsd_set and tsd_get
1492 * Locks out tsd_create, tsd_destroy, and tsd_exit
1493 * May free memory with lock held
1494 */
1495 void
1496 tsd_destroy(uint_t *keyp)
1497 {
1498 uint_t key;
1499 struct tsd_thread *tsd;
1500
1501 /*
1502 * protect the key namespace and our destructor lists
1503 */
1504 mutex_enter(&tsd_mutex);
1505 key = *keyp;
1506 *keyp = 0;
1507
1508 ASSERT(key <= tsd_nkeys);
1509
1510 /*
1511 * if the key is valid
1512 */
1513 if (key != 0) {
1514 uint_t k = key - 1;
1515 /*
1516 * for every thread with TSD, call key's destructor
1517 */
1518 for (tsd = tsd_list; tsd; tsd = tsd->ts_next) {
1519 /*
1520 * no TSD for key in this thread
1521 */
1522 if (key > tsd->ts_nkeys)
1523 continue;
1524 /*
1525 * call destructor for key
1526 */
1527 if (tsd->ts_value[k] && tsd_destructor[k])
1528 (*tsd_destructor[k])(tsd->ts_value[k]);
1529 /*
1530 * reset value for key
1531 */
1532 tsd->ts_value[k] = NULL;
1533 }
1534 /*
1535 * actually free the key (NULL destructor == unused)
1536 */
1537 tsd_destructor[k] = NULL;
1538 }
1539
1540 mutex_exit(&tsd_mutex);
1541 }
1542
1543 /*
1544 * Quickly return the per thread value that was stored with the specified key
1545 * Assumes the caller is protecting key from tsd_create and tsd_destroy
1546 */
1547 void *
1548 tsd_get(uint_t key)
1549 {
1550 return (tsd_agent_get(curthread, key));
1551 }
1552
1553 /*
1554 * Set a per thread value indexed with the specified key
1555 */
1556 int
1557 tsd_set(uint_t key, void *value)
1558 {
1559 return (tsd_agent_set(curthread, key, value));
1560 }
1561
1562 /*
1563 * Like tsd_get(), except that the agent lwp can get the tsd of
1564 * another thread in the same process (the agent thread only runs when the
1565 * process is completely stopped by /proc), or syslwp is creating a new lwp.
1566 */
1567 void *
1568 tsd_agent_get(kthread_t *t, uint_t key)
1569 {
1570 struct tsd_thread *tsd = t->t_tsd;
1571
1572 ASSERT(t == curthread ||
1573 ttoproc(t)->p_agenttp == curthread || t->t_state == TS_STOPPED);
1574
1575 if (key && tsd != NULL && key <= tsd->ts_nkeys)
1576 return (tsd->ts_value[key - 1]);
1577 return (NULL);
1578 }
1579
1580 /*
1581 * Like tsd_set(), except that the agent lwp can set the tsd of
1582 * another thread in the same process, or syslwp can set the tsd
1583 * of a thread it's in the middle of creating.
1584 *
1585 * Assumes the caller is protecting key from tsd_create and tsd_destroy
1586 * May lock out tsd_destroy (and tsd_create), may allocate memory with
1587 * lock held
1588 */
1589 int
1590 tsd_agent_set(kthread_t *t, uint_t key, void *value)
1591 {
1592 struct tsd_thread *tsd = t->t_tsd;
1593
1594 ASSERT(t == curthread ||
1595 ttoproc(t)->p_agenttp == curthread || t->t_state == TS_STOPPED);
1596
1597 if (key == 0)
1598 return (EINVAL);
1599 if (tsd == NULL)
1600 tsd = t->t_tsd = kmem_zalloc(sizeof (*tsd), KM_SLEEP);
1601 if (key <= tsd->ts_nkeys) {
1602 tsd->ts_value[key - 1] = value;
1603 return (0);
1604 }
1605
1606 ASSERT(key <= tsd_nkeys);
1607
1608 /*
1609 * lock out tsd_destroy()
1610 */
1611 mutex_enter(&tsd_mutex);
1612 if (tsd->ts_nkeys == 0) {
1613 /*
1614 * Link onto list of threads with TSD
1615 */
1616 if ((tsd->ts_next = tsd_list) != NULL)
1617 tsd_list->ts_prev = tsd;
1618 tsd_list = tsd;
1619 }
1620
1621 /*
1622 * Allocate thread local storage and set the value for key
1623 */
1624 tsd->ts_value = tsd_realloc(tsd->ts_value,
1625 tsd->ts_nkeys * sizeof (void *),
1626 key * sizeof (void *));
1627 tsd->ts_nkeys = key;
1628 tsd->ts_value[key - 1] = value;
1629 mutex_exit(&tsd_mutex);
1630
1631 return (0);
1632 }
1633
1634
1635 /*
1636 * Return the per thread value that was stored with the specified key
1637 * If necessary, create the key and the value
1638 * Assumes the caller is protecting *keyp from tsd_destroy
1639 */
1640 void *
1641 tsd_getcreate(uint_t *keyp, void (*destroy)(void *), void *(*allocate)(void))
1642 {
1643 void *value;
1644 uint_t key = *keyp;
1645 struct tsd_thread *tsd = curthread->t_tsd;
1646
1647 if (tsd == NULL)
1648 tsd = curthread->t_tsd = kmem_zalloc(sizeof (*tsd), KM_SLEEP);
1649 if (key && key <= tsd->ts_nkeys && (value = tsd->ts_value[key - 1]))
1650 return (value);
1651 if (key == 0)
1652 tsd_create(keyp, destroy);
1653 (void) tsd_set(*keyp, value = (*allocate)());
1654
1655 return (value);
1656 }
1657
1658 /*
1659 * Called from thread_exit() to run the destructor function for each tsd
1660 * Locks out tsd_create and tsd_destroy
1661 * Assumes that the destructor *DOES NOT* use tsd
1662 */
1663 void
1664 tsd_exit(void)
1665 {
1666 int i;
1667 struct tsd_thread *tsd = curthread->t_tsd;
1668
1669 if (tsd == NULL)
1670 return;
1671
1672 if (tsd->ts_nkeys == 0) {
1673 kmem_free(tsd, sizeof (*tsd));
1674 curthread->t_tsd = NULL;
1675 return;
1676 }
1677
1678 /*
1679 * lock out tsd_create and tsd_destroy, call
1680 * the destructor, and mark the value as destroyed.
1681 */
1682 mutex_enter(&tsd_mutex);
1683
1684 for (i = 0; i < tsd->ts_nkeys; i++) {
1685 if (tsd->ts_value[i] && tsd_destructor[i])
1686 (*tsd_destructor[i])(tsd->ts_value[i]);
1687 tsd->ts_value[i] = NULL;
1688 }
1689
1690 /*
1691 * remove from linked list of threads with TSD
1692 */
1693 if (tsd->ts_next)
1694 tsd->ts_next->ts_prev = tsd->ts_prev;
1695 if (tsd->ts_prev)
1696 tsd->ts_prev->ts_next = tsd->ts_next;
1697 if (tsd_list == tsd)
1698 tsd_list = tsd->ts_next;
1699
1700 mutex_exit(&tsd_mutex);
1701
1702 /*
1703 * free up the TSD
1704 */
1705 kmem_free(tsd->ts_value, tsd->ts_nkeys * sizeof (void *));
1706 kmem_free(tsd, sizeof (struct tsd_thread));
1707 curthread->t_tsd = NULL;
1708 }
1709
1710 /*
1711 * realloc
1712 */
1713 static void *
1714 tsd_realloc(void *old, size_t osize, size_t nsize)
1715 {
1716 void *new;
1717
1718 new = kmem_zalloc(nsize, KM_SLEEP);
1719 if (old) {
1720 bcopy(old, new, osize);
1721 kmem_free(old, osize);
1722 }
1723 return (new);
1724 }
1725
1726 /*
1727 * Return non-zero if an interrupt is being serviced.
1728 */
1729 int
1730 servicing_interrupt()
1731 {
1732 int onintr = 0;
1733
1734 /* Are we an interrupt thread */
1735 if (curthread->t_flag & T_INTR_THREAD)
1736 return (1);
1737 /* Are we servicing a high level interrupt? */
1738 if (CPU_ON_INTR(CPU)) {
1739 kpreempt_disable();
1740 onintr = CPU_ON_INTR(CPU);
1741 kpreempt_enable();
1742 }
1743 return (onintr);
1744 }
1745
1746
1747 /*
1748 * Change the dispatch priority of a thread in the system.
1749 * Used when raising or lowering a thread's priority.
1750 * (E.g., priority inheritance)
1751 *
1752 * Since threads are queued according to their priority, we
1753 * we must check the thread's state to determine whether it
1754 * is on a queue somewhere. If it is, we've got to:
1755 *
1756 * o Dequeue the thread.
1757 * o Change its effective priority.
1758 * o Enqueue the thread.
1759 *
1760 * Assumptions: The thread whose priority we wish to change
1761 * must be locked before we call thread_change_(e)pri().
1762 * The thread_change(e)pri() function doesn't drop the thread
1763 * lock--that must be done by its caller.
1764 */
1765 void
1766 thread_change_epri(kthread_t *t, pri_t disp_pri)
1767 {
1768 uint_t state;
1769
1770 ASSERT(THREAD_LOCK_HELD(t));
1771
1772 /*
1773 * If the inherited priority hasn't actually changed,
1774 * just return.
1775 */
1776 if (t->t_epri == disp_pri)
1777 return;
1778
1779 state = t->t_state;
1780
1781 /*
1782 * If it's not on a queue, change the priority with impunity.
1783 */
1784 if ((state & (TS_SLEEP | TS_RUN | TS_WAIT)) == 0) {
1785 t->t_epri = disp_pri;
1786 if (state == TS_ONPROC) {
1787 cpu_t *cp = t->t_disp_queue->disp_cpu;
1788
1789 if (t == cp->cpu_dispthread)
1790 cp->cpu_dispatch_pri = DISP_PRIO(t);
1791 }
1792 } else if (state == TS_SLEEP) {
1793 /*
1794 * Take the thread out of its sleep queue.
1795 * Change the inherited priority.
1796 * Re-enqueue the thread.
1797 * Each synchronization object exports a function
1798 * to do this in an appropriate manner.
1799 */
1800 SOBJ_CHANGE_EPRI(t->t_sobj_ops, t, disp_pri);
1801 } else if (state == TS_WAIT) {
1802 /*
1803 * Re-enqueue a thread on the wait queue if its
1804 * effective priority needs to change.
1805 */
1806 if (disp_pri != t->t_epri)
1807 waitq_change_pri(t, disp_pri);
1808 } else {
1809 /*
1810 * The thread is on a run queue.
1811 * Note: setbackdq() may not put the thread
1812 * back on the same run queue where it originally
1813 * resided.
1814 */
1815 (void) dispdeq(t);
1816 t->t_epri = disp_pri;
1817 setbackdq(t);
1818 }
1819 schedctl_set_cidpri(t);
1820 }
1821
1822 /*
1823 * Function: Change the t_pri field of a thread.
1824 * Side Effects: Adjust the thread ordering on a run queue
1825 * or sleep queue, if necessary.
1826 * Returns: 1 if the thread was on a run queue, else 0.
1827 */
1828 int
1829 thread_change_pri(kthread_t *t, pri_t disp_pri, int front)
1830 {
1831 uint_t state;
1832 int on_rq = 0;
1833
1834 ASSERT(THREAD_LOCK_HELD(t));
1835
1836 state = t->t_state;
1837 THREAD_WILLCHANGE_PRI(t, disp_pri);
1838
1839 /*
1840 * If it's not on a queue, change the priority with impunity.
1841 */
1842 if ((state & (TS_SLEEP | TS_RUN | TS_WAIT)) == 0) {
1843 t->t_pri = disp_pri;
1844
1845 if (state == TS_ONPROC) {
1846 cpu_t *cp = t->t_disp_queue->disp_cpu;
1847
1848 if (t == cp->cpu_dispthread)
1849 cp->cpu_dispatch_pri = DISP_PRIO(t);
1850 }
1851 } else if (state == TS_SLEEP) {
1852 /*
1853 * If the priority has changed, take the thread out of
1854 * its sleep queue and change the priority.
1855 * Re-enqueue the thread.
1856 * Each synchronization object exports a function
1857 * to do this in an appropriate manner.
1858 */
1859 if (disp_pri != t->t_pri)
1860 SOBJ_CHANGE_PRI(t->t_sobj_ops, t, disp_pri);
1861 } else if (state == TS_WAIT) {
1862 /*
1863 * Re-enqueue a thread on the wait queue if its
1864 * priority needs to change.
1865 */
1866 if (disp_pri != t->t_pri)
1867 waitq_change_pri(t, disp_pri);
1868 } else {
1869 /*
1870 * The thread is on a run queue.
1871 * Note: setbackdq() may not put the thread
1872 * back on the same run queue where it originally
1873 * resided.
1874 *
1875 * We still requeue the thread even if the priority
1876 * is unchanged to preserve round-robin (and other)
1877 * effects between threads of the same priority.
1878 */
1879 on_rq = dispdeq(t);
1880 ASSERT(on_rq);
1881 t->t_pri = disp_pri;
1882 if (front) {
1883 setfrontdq(t);
1884 } else {
1885 setbackdq(t);
1886 }
1887 }
1888 schedctl_set_cidpri(t);
1889 return (on_rq);
1890 }
1891
1892 /*
1893 * Tunable kmem_stackinfo is set, fill the kernel thread stack with a
1894 * specific pattern.
1895 */
1896 static void
1897 stkinfo_begin(kthread_t *t)
1898 {
1899 caddr_t start; /* stack start */
1900 caddr_t end; /* stack end */
1901 uint64_t *ptr; /* pattern pointer */
1902
1903 /*
1904 * Stack grows up or down, see thread_create(),
1905 * compute stack memory area start and end (start < end).
1906 */
1907 if (t->t_stk > t->t_stkbase) {
1908 /* stack grows down */
1909 start = t->t_stkbase;
1910 end = t->t_stk;
1911 } else {
1912 /* stack grows up */
1913 start = t->t_stk;
1914 end = t->t_stkbase;
1915 }
1916
1917 /*
1918 * Stackinfo pattern size is 8 bytes. Ensure proper 8 bytes
1919 * alignement for start and end in stack area boundaries
1920 * (protection against corrupt t_stkbase/t_stk data).
1921 */
1922 if ((((uintptr_t)start) & 0x7) != 0) {
1923 start = (caddr_t)((((uintptr_t)start) & (~0x7)) + 8);
1924 }
1925 end = (caddr_t)(((uintptr_t)end) & (~0x7));
1926
1927 if ((end <= start) || (end - start) > (1024 * 1024)) {
1928 /* negative or stack size > 1 meg, assume bogus */
1929 return;
1930 }
1931
1932 /* fill stack area with a pattern (instead of zeros) */
1933 ptr = (uint64_t *)((void *)start);
1934 while (ptr < (uint64_t *)((void *)end)) {
1935 *ptr++ = KMEM_STKINFO_PATTERN;
1936 }
1937 }
1938
1939
1940 /*
1941 * Tunable kmem_stackinfo is set, create stackinfo log if doesn't already exist,
1942 * compute the percentage of kernel stack really used, and set in the log
1943 * if it's the latest highest percentage.
1944 */
1945 static void
1946 stkinfo_end(kthread_t *t)
1947 {
1948 caddr_t start; /* stack start */
1949 caddr_t end; /* stack end */
1950 uint64_t *ptr; /* pattern pointer */
1951 size_t stksz; /* stack size */
1952 size_t smallest = 0;
1953 size_t percent = 0;
1954 uint_t index = 0;
1955 uint_t i;
1956 static size_t smallest_percent = (size_t)-1;
1957 static uint_t full = 0;
1958
1959 /* create the stackinfo log, if doesn't already exist */
1960 mutex_enter(&kmem_stkinfo_lock);
1961 if (kmem_stkinfo_log == NULL) {
1962 kmem_stkinfo_log = (kmem_stkinfo_t *)
1963 kmem_zalloc(KMEM_STKINFO_LOG_SIZE *
1964 (sizeof (kmem_stkinfo_t)), KM_NOSLEEP);
1965 if (kmem_stkinfo_log == NULL) {
1966 mutex_exit(&kmem_stkinfo_lock);
1967 return;
1968 }
1969 }
1970 mutex_exit(&kmem_stkinfo_lock);
1971
1972 /*
1973 * Stack grows up or down, see thread_create(),
1974 * compute stack memory area start and end (start < end).
1975 */
1976 if (t->t_stk > t->t_stkbase) {
1977 /* stack grows down */
1978 start = t->t_stkbase;
1979 end = t->t_stk;
1980 } else {
1981 /* stack grows up */
1982 start = t->t_stk;
1983 end = t->t_stkbase;
1984 }
1985
1986 /* stack size as found in kthread_t */
1987 stksz = end - start;
1988
1989 /*
1990 * Stackinfo pattern size is 8 bytes. Ensure proper 8 bytes
1991 * alignement for start and end in stack area boundaries
1992 * (protection against corrupt t_stkbase/t_stk data).
1993 */
1994 if ((((uintptr_t)start) & 0x7) != 0) {
1995 start = (caddr_t)((((uintptr_t)start) & (~0x7)) + 8);
1996 }
1997 end = (caddr_t)(((uintptr_t)end) & (~0x7));
1998
1999 if ((end <= start) || (end - start) > (1024 * 1024)) {
2000 /* negative or stack size > 1 meg, assume bogus */
2001 return;
2002 }
2003
2004 /* search until no pattern in the stack */
2005 if (t->t_stk > t->t_stkbase) {
2006 /* stack grows down */
2007 #if defined(__i386) || defined(__amd64)
2008 /*
2009 * 6 longs are pushed on stack, see thread_load(). Skip
2010 * them, so if kthread has never run, percent is zero.
2011 * 8 bytes alignement is preserved for a 32 bit kernel,
2012 * 6 x 4 = 24, 24 is a multiple of 8.
2013 *
2014 */
2015 end -= (6 * sizeof (long));
2016 #endif
2017 ptr = (uint64_t *)((void *)start);
2018 while (ptr < (uint64_t *)((void *)end)) {
2019 if (*ptr != KMEM_STKINFO_PATTERN) {
2020 percent = stkinfo_percent(end,
2021 start, (caddr_t)ptr);
2022 break;
2023 }
2024 ptr++;
2025 }
2026 } else {
2027 /* stack grows up */
2028 ptr = (uint64_t *)((void *)end);
2029 ptr--;
2030 while (ptr >= (uint64_t *)((void *)start)) {
2031 if (*ptr != KMEM_STKINFO_PATTERN) {
2032 percent = stkinfo_percent(start,
2033 end, (caddr_t)ptr);
2034 break;
2035 }
2036 ptr--;
2037 }
2038 }
2039
2040 DTRACE_PROBE3(stack__usage, kthread_t *, t,
2041 size_t, stksz, size_t, percent);
2042
2043 if (percent == 0) {
2044 return;
2045 }
2046
2047 mutex_enter(&kmem_stkinfo_lock);
2048 if (full == KMEM_STKINFO_LOG_SIZE && percent < smallest_percent) {
2049 /*
2050 * The log is full and already contains the highest values
2051 */
2052 mutex_exit(&kmem_stkinfo_lock);
2053 return;
2054 }
2055
2056 /* keep a log of the highest used stack */
2057 for (i = 0; i < KMEM_STKINFO_LOG_SIZE; i++) {
2058 if (kmem_stkinfo_log[i].percent == 0) {
2059 index = i;
2060 full++;
2061 break;
2062 }
2063 if (smallest == 0) {
2064 smallest = kmem_stkinfo_log[i].percent;
2065 index = i;
2066 continue;
2067 }
2068 if (kmem_stkinfo_log[i].percent < smallest) {
2069 smallest = kmem_stkinfo_log[i].percent;
2070 index = i;
2071 }
2072 }
2073
2074 if (percent >= kmem_stkinfo_log[index].percent) {
2075 kmem_stkinfo_log[index].kthread = (caddr_t)t;
2076 kmem_stkinfo_log[index].t_startpc = (caddr_t)t->t_startpc;
2077 kmem_stkinfo_log[index].start = start;
2078 kmem_stkinfo_log[index].stksz = stksz;
2079 kmem_stkinfo_log[index].percent = percent;
2080 kmem_stkinfo_log[index].t_tid = t->t_tid;
2081 kmem_stkinfo_log[index].cmd[0] = '\0';
2082 if (t->t_tid != 0) {
2083 stksz = strlen((t->t_procp)->p_user.u_comm);
2084 if (stksz >= KMEM_STKINFO_STR_SIZE) {
2085 stksz = KMEM_STKINFO_STR_SIZE - 1;
2086 kmem_stkinfo_log[index].cmd[stksz] = '\0';
2087 } else {
2088 stksz += 1;
2089 }
2090 (void) memcpy(kmem_stkinfo_log[index].cmd,
2091 (t->t_procp)->p_user.u_comm, stksz);
2092 }
2093 if (percent < smallest_percent) {
2094 smallest_percent = percent;
2095 }
2096 }
2097 mutex_exit(&kmem_stkinfo_lock);
2098 }
2099
2100 /*
2101 * Tunable kmem_stackinfo is set, compute stack utilization percentage.
2102 */
2103 static size_t
2104 stkinfo_percent(caddr_t t_stk, caddr_t t_stkbase, caddr_t sp)
2105 {
2106 size_t percent;
2107 size_t s;
2108
2109 if (t_stk > t_stkbase) {
2110 /* stack grows down */
2111 if (sp > t_stk) {
2112 return (0);
2113 }
2114 if (sp < t_stkbase) {
2115 return (100);
2116 }
2117 percent = t_stk - sp + 1;
2118 s = t_stk - t_stkbase + 1;
2119 } else {
2120 /* stack grows up */
2121 if (sp < t_stk) {
2122 return (0);
2123 }
2124 if (sp > t_stkbase) {
2125 return (100);
2126 }
2127 percent = sp - t_stk + 1;
2128 s = t_stkbase - t_stk + 1;
2129 }
2130 percent = ((100 * percent) / s) + 1;
2131 if (percent > 100) {
2132 percent = 100;
2133 }
2134 return (percent);
2135 }
2136
2137 /*
2138 * NOTE: This will silently truncate a name > THREAD_NAME_MAX - 1 characters
2139 * long. It is expected that callers (acting on behalf of userland clients)
2140 * will perform any required checks to return the correct error semantics.
2141 * It is also expected callers on behalf of userland clients have done
2142 * any necessary permission checks.
2143 */
2144 int
2145 thread_setname(kthread_t *t, const char *name)
2146 {
2147 char *buf = NULL;
2148
2149 /*
2150 * We optimistically assume that a thread's name will only be set
2151 * once and so allocate memory in preparation of setting t_name.
2152 * If it turns out a name has already been set, we just discard (free)
2153 * the buffer we just allocated and reuse the current buffer
2154 * (as all should be THREAD_NAME_MAX large).
2155 *
2156 * Such an arrangement means over the lifetime of a kthread_t, t_name
2157 * is either NULL or has one value (the address of the buffer holding
2158 * the current thread name). The assumption is that most kthread_t
2159 * instances will not have a name assigned, so dynamically allocating
2160 * the memory should minimize the footprint of this feature, but by
2161 * having the buffer persist for the life of the thread, it simplifies
2162 * usage in highly constrained situations (e.g. dtrace).
2163 */
2164 if (name != NULL && name[0] != '\0') {
2165 for (size_t i = 0; name[i] != '\0'; i++) {
2166 if (!isprint(name[i]))
2167 return (EINVAL);
2168 }
2169
2170 buf = kmem_zalloc(THREAD_NAME_MAX, KM_SLEEP);
2171 (void) strlcpy(buf, name, THREAD_NAME_MAX);
2172 }
2173
2174 mutex_enter(&ttoproc(t)->p_lock);
2175 if (t->t_name == NULL) {
2176 t->t_name = buf;
2177 } else {
2178 if (buf != NULL) {
2179 (void) strlcpy(t->t_name, name, THREAD_NAME_MAX);
2180 kmem_free(buf, THREAD_NAME_MAX);
2181 } else {
2182 bzero(t->t_name, THREAD_NAME_MAX);
2183 }
2184 }
2185 mutex_exit(&ttoproc(t)->p_lock);
2186 return (0);
2187 }
2188
2189 int
2190 thread_vsetname(kthread_t *t, const char *fmt, ...)
2191 {
2192 char name[THREAD_NAME_MAX];
2193 va_list va;
2194 int rc;
2195
2196 va_start(va, fmt);
2197 rc = vsnprintf(name, sizeof (name), fmt, va);
2198 va_end(va);
2199
2200 if (rc < 0)
2201 return (EINVAL);
2202
2203 if (rc >= sizeof (name))
2204 return (ENAMETOOLONG);
2205
2206 return (thread_setname(t, name));
2207 }