]> git.saurik.com Git - apple/xnu.git/blob - bsd/kern/pthread_synch.c
projects / apple / xnu.git / blob
? search:


9ccbc9a9e50a61d1829d32214f69d58aff844835
[apple/xnu.git] / bsd / kern / pthread_synch.c
1 /*
2 * Copyright (c) 2000-2007 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /* Copyright (c) 1995-2005 Apple Computer, Inc. All Rights Reserved */
29 /*
30 * pthread_synch.c
31 */
32
33 #define _PTHREAD_CONDATTR_T
34 #define _PTHREAD_COND_T
35 #define _PTHREAD_MUTEXATTR_T
36 #define _PTHREAD_MUTEX_T
37 #define _PTHREAD_RWLOCKATTR_T
38 #define _PTHREAD_RWLOCK_T
39
40 #undef pthread_mutexattr_t
41 #undef pthread_mutex_t
42 #undef pthread_condattr_t
43 #undef pthread_cond_t
44 #undef pthread_rwlockattr_t
45 #undef pthread_rwlock_t
46
47 #include <sys/param.h>
48 #include <sys/queue.h>
49 #include <sys/resourcevar.h>
50 #include <sys/proc_internal.h>
51 #include <sys/kauth.h>
52 #include <sys/systm.h>
53 #include <sys/timeb.h>
54 #include <sys/times.h>
55 #include <sys/acct.h>
56 #include <sys/kernel.h>
57 #include <sys/wait.h>
58 #include <sys/signalvar.h>
59 #include <sys/syslog.h>
60 #include <sys/stat.h>
61 #include <sys/lock.h>
62 #include <sys/kdebug.h>
63 #include <sys/sysproto.h>
64 #include <sys/pthread_internal.h>
65 #include <sys/vm.h>
66 #include <sys/user.h> /* for coredump */
67
68
69 #include <mach/mach_types.h>
70 #include <mach/vm_prot.h>
71 #include <mach/semaphore.h>
72 #include <mach/sync_policy.h>
73 #include <mach/task.h>
74 #include <kern/kern_types.h>
75 #include <kern/task.h>
76 #include <kern/clock.h>
77 #include <mach/kern_return.h>
78 #include <kern/thread.h>
79 #include <kern/sched_prim.h>
80 #include <kern/kalloc.h>
81 #include <kern/sched_prim.h> /* for thread_exception_return */
82 #include <kern/processor.h>
83 #include <kern/affinity.h>
84 #include <mach/mach_vm.h>
85 #include <mach/mach_param.h>
86 #include <mach/thread_status.h>
87 #include <mach/thread_policy.h>
88 #include <mach/message.h>
89 #include <mach/port.h>
90 #include <vm/vm_protos.h>
91 #include <vm/vm_map.h>` /* for current_map() */
92 #include <mach/thread_act.h> /* for thread_resume */
93 #include <machine/machine_routines.h>
94 #if defined(__i386__)
95 #include <i386/machine_routines.h>
96 #include <i386/eflags.h>
97 #include <i386/psl.h>
98 #include <i386/seg.h>
99 #endif
100
101 #include <libkern/OSAtomic.h>
102
103 #if 0
104 #undef KERNEL_DEBUG
105 #define KERNEL_DEBUG KERNEL_DEBUG_CONSTANT
106 #undef KERNEL_DEBUG1
107 #define KERNEL_DEBUG1 KERNEL_DEBUG_CONSTANT1
108 #endif
109
110
111 #if defined(__ppc__) || defined(__ppc64__)
112 #include <architecture/ppc/cframe.h>
113 #endif
114
115
116 lck_grp_attr_t *pthread_lck_grp_attr;
117 lck_grp_t *pthread_lck_grp;
118 lck_attr_t *pthread_lck_attr;
119 lck_mtx_t * pthread_list_mlock;
120 extern void pthread_init(void);
121
122 extern kern_return_t thread_getstatus(register thread_t act, int flavor,
123 thread_state_t tstate, mach_msg_type_number_t *count);
124 extern kern_return_t thread_setstatus(thread_t thread, int flavor,
125 thread_state_t tstate, mach_msg_type_number_t count);
126 extern void thread_set_cthreadself(thread_t thread, uint64_t pself, int isLP64);
127 extern kern_return_t mach_port_deallocate(ipc_space_t, mach_port_name_t);
128 extern kern_return_t semaphore_signal_internal_trap(mach_port_name_t);
129
130 static int workqueue_additem(struct workqueue *wq, int prio, user_addr_t item);
131 static int workqueue_removeitem(struct workqueue *wq, int prio, user_addr_t item);
132 static void workqueue_run_nextitem(proc_t p, thread_t th);
133 static void wq_runitem(proc_t p, user_addr_t item, thread_t th, struct threadlist *tl,
134 int reuse_thread, int wake_thread, int return_directly);
135 static int setup_wqthread(proc_t p, thread_t th, user_addr_t item, int reuse_thread, struct threadlist *tl);
136 static int workqueue_addnewthread(struct workqueue *wq);
137 static void workqueue_removethread(struct workqueue *wq);
138 static void workqueue_lock(proc_t);
139 static void workqueue_lock_spin(proc_t);
140 static void workqueue_unlock(proc_t);
141
142 #define C_32_STK_ALIGN 16
143 #define C_64_STK_ALIGN 16
144 #define C_64_REDZONE_LEN 128
145 #define TRUNC_DOWN32(a,c) ((((uint32_t)a)-(c)) & ((uint32_t)(-(c))))
146 #define TRUNC_DOWN64(a,c) ((((uint64_t)a)-(c)) & ((uint64_t)(-(c))))
147
148
149 /*
150 * Flags filed passed to bsdthread_create and back in pthread_start
151 31 <---------------------------------> 0
152 _________________________________________
153 | flags(8) | policy(8) | importance(16) |
154 -----------------------------------------
155 */
156 void _pthread_start(pthread_t self, mach_port_t kport, void *(*fun)(void *), void * funarg, size_t stacksize, unsigned int flags);
157
158 #define PTHREAD_START_CUSTOM 0x01000000
159 #define PTHREAD_START_SETSCHED 0x02000000
160 #define PTHREAD_START_DETACHED 0x04000000
161 #define PTHREAD_START_POLICY_BITSHIFT 16
162 #define PTHREAD_START_POLICY_MASK 0xff
163 #define PTHREAD_START_IMPORTANCE_MASK 0xffff
164
165 #define SCHED_OTHER POLICY_TIMESHARE
166 #define SCHED_FIFO POLICY_FIFO
167 #define SCHED_RR POLICY_RR
168
169 void
170 pthread_init(void)
171 {
172
173 pthread_lck_grp_attr = lck_grp_attr_alloc_init();
174 pthread_lck_grp = lck_grp_alloc_init("pthread", pthread_lck_grp_attr);
175
176 /*
177 * allocate the lock attribute for pthread synchronizers
178 */
179 pthread_lck_attr = lck_attr_alloc_init();
180
181 pthread_list_mlock = lck_mtx_alloc_init(pthread_lck_grp, pthread_lck_attr);
182
183 }
184
185 void
186 pthread_list_lock(void)
187 {
188 lck_mtx_lock(pthread_list_mlock);
189 }
190
191 void
192 pthread_list_unlock(void)
193 {
194 lck_mtx_unlock(pthread_list_mlock);
195 }
196
197
198 int
199 __pthread_mutex_destroy(__unused struct proc *p, struct __pthread_mutex_destroy_args *uap, __unused register_t *retval)
200 {
201 int res;
202 int mutexid = uap->mutexid;
203 pthread_mutex_t * mutex;
204 lck_mtx_t * lmtx;
205 lck_mtx_t * lmtx1;
206
207
208 mutex = pthread_id_to_mutex(mutexid);
209 if (mutex == 0)
210 return(EINVAL);
211
212 MTX_LOCK(mutex->lock);
213 if (mutex->sig == _PTHREAD_KERN_MUTEX_SIG)
214 {
215 if (mutex->owner == (thread_t)NULL &&
216 mutex->refcount == 1)
217 {
218 mutex->sig = _PTHREAD_NO_SIG;
219 lmtx = mutex->mutex;
220 lmtx1 = mutex->lock;
221 mutex->mutex = NULL;
222 pthread_id_mutex_remove(mutexid);
223 mutex->refcount --;
224 MTX_UNLOCK(mutex->lock);
225 lck_mtx_free(lmtx, pthread_lck_grp);
226 lck_mtx_free(lmtx1, pthread_lck_grp);
227 kfree((void *)mutex, sizeof(struct _pthread_mutex));
228 return(0);
229 }
230 else
231 res = EBUSY;
232 }
233 else
234 res = EINVAL;
235 MTX_UNLOCK(mutex->lock);
236 pthread_mutex_release(mutex);
237 return (res);
238 }
239
240 /*
241 * Initialize a mutex variable, possibly with additional attributes.
242 */
243 static void
244 pthread_mutex_init_internal(pthread_mutex_t *mutex, const pthread_mutexattr_t *attr)
245 {
246 mutex->prioceiling = attr->prioceiling;
247 mutex->protocol = attr->protocol;
248 mutex->type = attr->type;
249 mutex->pshared = attr->pshared;
250 mutex->refcount = 0;
251 mutex->owner = (thread_t)NULL;
252 mutex->owner_proc = current_proc();
253 mutex->sig = _PTHREAD_KERN_MUTEX_SIG;
254 mutex->lock = lck_mtx_alloc_init(pthread_lck_grp, pthread_lck_attr);
255 mutex->mutex = lck_mtx_alloc_init(pthread_lck_grp, pthread_lck_attr);
256 }
257
258 /*
259 * Initialize a mutex variable, possibly with additional attributes.
260 * Public interface - so don't trust the lock - initialize it first.
261 */
262 int
263 __pthread_mutex_init(__unused struct proc *p, struct __pthread_mutex_init_args *uap, __unused register_t *retval)
264 {
265 user_addr_t umutex = uap->mutex;
266 pthread_mutex_t * mutex;
267 user_addr_t uattr = uap->attr;
268 pthread_mutexattr_t attr;
269 unsigned int addr = (unsigned int)((uintptr_t)uap->mutex);
270 int pmutex_sig;
271 int mutexid;
272 int error = 0;
273
274 if ((umutex == 0) || (uattr == 0))
275 return(EINVAL);
276
277 if ((error = copyin(uattr, &attr, sizeof(pthread_mutexattr_t))))
278 return(error);
279
280 if (attr.sig != _PTHREAD_MUTEX_ATTR_SIG)
281 return (EINVAL);
282
283 if ((error = copyin(umutex, &pmutex_sig, sizeof(int))))
284 return(error);
285
286 if (pmutex_sig == _PTHREAD_KERN_MUTEX_SIG)
287 return(EBUSY);
288 mutex = (pthread_mutex_t *)kalloc(sizeof(pthread_mutex_t));
289
290 pthread_mutex_init_internal(mutex, &attr);
291
292
293 addr += 8;
294 mutexid = pthread_id_mutex_add(mutex);
295 if (mutexid) {
296 if ((error = copyout(&mutexid, ((user_addr_t)((uintptr_t)(addr))), 4)))
297 goto cleanup;
298 return(0);
299 } else
300 error = ENOMEM;
301 cleanup:
302 if(mutexid)
303 pthread_id_mutex_remove(mutexid);
304 lck_mtx_free(mutex->lock, pthread_lck_grp);
305 lck_mtx_free(mutex->mutex, pthread_lck_grp);
306 kfree(mutex, sizeof(struct _pthread_mutex));
307 return(error);
308 }
309
310 /*
311 * Lock a mutex.
312 * TODO: Priority inheritance stuff
313 */
314 int
315 __pthread_mutex_lock(struct proc *p, struct __pthread_mutex_lock_args *uap, __unused register_t *retval)
316 {
317 int mutexid = uap->mutexid;
318 pthread_mutex_t * mutex;
319 int error;
320
321 mutex = pthread_id_to_mutex(mutexid);
322 if (mutex == 0)
323 return(EINVAL);
324
325 MTX_LOCK(mutex->lock);
326
327 if (mutex->sig != _PTHREAD_KERN_MUTEX_SIG)
328 {
329 error = EINVAL;
330 goto out;
331 }
332
333 if ((p != mutex->owner_proc) && (mutex->pshared != PTHREAD_PROCESS_SHARED)) {
334 error = EINVAL;
335 goto out;
336 }
337
338 MTX_UNLOCK(mutex->lock);
339
340 lck_mtx_lock(mutex->mutex);
341
342 MTX_LOCK(mutex->lock);
343 mutex->owner = current_thread();
344 error = 0;
345 out:
346 MTX_UNLOCK(mutex->lock);
347 pthread_mutex_release(mutex);
348 return (error);
349 }
350
351 /*
352 * Attempt to lock a mutex, but don't block if this isn't possible.
353 */
354 int
355 __pthread_mutex_trylock(struct proc *p, struct __pthread_mutex_trylock_args *uap, __unused register_t *retval)
356 {
357 int mutexid = uap->mutexid;
358 pthread_mutex_t * mutex;
359 boolean_t state;
360 int error;
361
362 mutex = pthread_id_to_mutex(mutexid);
363 if (mutex == 0)
364 return(EINVAL);
365
366 MTX_LOCK(mutex->lock);
367
368 if (mutex->sig != _PTHREAD_KERN_MUTEX_SIG)
369 {
370 error = EINVAL;
371 goto out;
372 }
373
374 if ((p != mutex->owner_proc) && (mutex->pshared != PTHREAD_PROCESS_SHARED)) {
375 error = EINVAL;
376 goto out;
377 }
378
379 MTX_UNLOCK(mutex->lock);
380
381 state = lck_mtx_try_lock(mutex->mutex);
382 if (state) {
383 MTX_LOCK(mutex->lock);
384 mutex->owner = current_thread();
385 MTX_UNLOCK(mutex->lock);
386 error = 0;
387 } else
388 error = EBUSY;
389
390 pthread_mutex_release(mutex);
391 return (error);
392 out:
393 MTX_UNLOCK(mutex->lock);
394 pthread_mutex_release(mutex);
395 return (error);
396 }
397
398 /*
399 * Unlock a mutex.
400 * TODO: Priority inheritance stuff
401 */
402 int
403 __pthread_mutex_unlock(struct proc *p, struct __pthread_mutex_unlock_args *uap, __unused register_t *retval)
404 {
405 int mutexid = uap->mutexid;
406 pthread_mutex_t * mutex;
407 int error;
408
409 mutex = pthread_id_to_mutex(mutexid);
410 if (mutex == 0)
411 return(EINVAL);
412
413 MTX_LOCK(mutex->lock);
414
415 if (mutex->sig != _PTHREAD_KERN_MUTEX_SIG)
416 {
417 error = EINVAL;
418 goto out;
419 }
420
421 if ((p != mutex->owner_proc) && (mutex->pshared != PTHREAD_PROCESS_SHARED)) {
422 error = EINVAL;
423 goto out;
424 }
425
426 MTX_UNLOCK(mutex->lock);
427
428 lck_mtx_unlock(mutex->mutex);
429
430 MTX_LOCK(mutex->lock);
431 mutex->owner = NULL;
432 error = 0;
433 out:
434 MTX_UNLOCK(mutex->lock);
435 pthread_mutex_release(mutex);
436 return (error);
437 }
438
439
440 int
441 __pthread_cond_init(__unused struct proc *p, struct __pthread_cond_init_args *uap, __unused register_t *retval)
442 {
443 pthread_cond_t * cond;
444 pthread_condattr_t attr;
445 user_addr_t ucond = uap->cond;
446 user_addr_t uattr = uap->attr;
447 unsigned int addr = (unsigned int)((uintptr_t)uap->cond);
448 int condid, error, cond_sig;
449 semaphore_t sem;
450 kern_return_t kret;
451 int value = 0;
452
453 if ((ucond == 0) || (uattr == 0))
454 return(EINVAL);
455
456 if ((error = copyin(uattr, &attr, sizeof(pthread_condattr_t))))
457 return(error);
458
459 if (attr.sig != _PTHREAD_COND_ATTR_SIG)
460 return (EINVAL);
461
462 if ((error = copyin(ucond, &cond_sig, sizeof(int))))
463 return(error);
464
465 if (cond_sig == _PTHREAD_KERN_COND_SIG)
466 return(EBUSY);
467 kret = semaphore_create(kernel_task, &sem, SYNC_POLICY_FIFO, value);
468 if (kret != KERN_SUCCESS)
469 return(ENOMEM);
470
471 cond = (pthread_cond_t *)kalloc(sizeof(pthread_cond_t));
472
473 cond->lock = lck_mtx_alloc_init(pthread_lck_grp, pthread_lck_attr);
474 cond->pshared = attr.pshared;
475 cond->sig = _PTHREAD_KERN_COND_SIG;
476 cond->sigpending = 0;
477 cond->waiters = 0;
478 cond->refcount = 0;
479 cond->mutex = (pthread_mutex_t *)0;
480 cond->owner_proc = current_proc();
481 cond->sem = sem;
482
483 addr += 8;
484 condid = pthread_id_cond_add(cond);
485 if (condid) {
486 if ((error = copyout(&condid, ((user_addr_t)((uintptr_t)(addr))), 4)))
487 goto cleanup;
488 return(0);
489 } else
490 error = ENOMEM;
491 cleanup:
492 if(condid)
493 pthread_id_cond_remove(condid);
494 semaphore_destroy(kernel_task, cond->sem);
495 kfree(cond, sizeof(pthread_cond_t));
496 return(error);
497 }
498
499
500 /*
501 * Destroy a condition variable.
502 */
503 int
504 __pthread_cond_destroy(__unused struct proc *p, struct __pthread_cond_destroy_args *uap, __unused register_t *retval)
505 {
506 pthread_cond_t *cond;
507 int condid = uap->condid;
508 semaphore_t sem;
509 lck_mtx_t * lmtx;
510 int res;
511
512 cond = pthread_id_to_cond(condid);
513 if (cond == 0)
514 return(EINVAL);
515
516 COND_LOCK(cond->lock);
517 if (cond->sig == _PTHREAD_KERN_COND_SIG)
518 {
519 if (cond->refcount == 1)
520 {
521 cond->sig = _PTHREAD_NO_SIG;
522 sem = cond->sem;
523 cond->sem = NULL;
524 lmtx = cond->lock;
525 pthread_id_cond_remove(condid);
526 cond->refcount --;
527 COND_UNLOCK(cond->lock);
528 lck_mtx_free(lmtx, pthread_lck_grp);
529 (void)semaphore_destroy(kernel_task, sem);
530 kfree((void *)cond, sizeof(pthread_cond_t));
531 return(0);
532 }
533 else
534 res = EBUSY;
535 }
536 else
537 res = EINVAL;
538 COND_UNLOCK(cond->lock);
539 pthread_cond_release(cond);
540 return (res);
541 }
542
543
544 /*
545 * Signal a condition variable, waking up all threads waiting for it.
546 */
547 int
548 __pthread_cond_broadcast(__unused struct proc *p, struct __pthread_cond_broadcast_args *uap, __unused register_t *retval)
549 {
550 int condid = uap->condid;
551 pthread_cond_t * cond;
552 int error;
553 kern_return_t kret;
554
555 cond = pthread_id_to_cond(condid);
556 if (cond == 0)
557 return(EINVAL);
558
559 COND_LOCK(cond->lock);
560
561 if (cond->sig != _PTHREAD_KERN_COND_SIG)
562 {
563 error = EINVAL;
564 goto out;
565 }
566
567 if ((p != cond->owner_proc) && (cond->pshared != PTHREAD_PROCESS_SHARED)) {
568 error = EINVAL;
569 goto out;
570 }
571
572 COND_UNLOCK(cond->lock);
573
574 kret = semaphore_signal_all(cond->sem);
575 switch (kret) {
576 case KERN_INVALID_ADDRESS:
577 case KERN_PROTECTION_FAILURE:
578 error = EINVAL;
579 break;
580 case KERN_ABORTED:
581 case KERN_OPERATION_TIMED_OUT:
582 error = EINTR;
583 break;
584 case KERN_SUCCESS:
585 error = 0;
586 break;
587 default:
588 error = EINVAL;
589 break;
590 }
591
592 COND_LOCK(cond->lock);
593 out:
594 COND_UNLOCK(cond->lock);
595 pthread_cond_release(cond);
596 return (error);
597 }
598
599
600 /*
601 * Signal a condition variable, waking only one thread.
602 */
603 int
604 __pthread_cond_signal(__unused struct proc *p, struct __pthread_cond_signal_args *uap, __unused register_t *retval)
605 {
606 int condid = uap->condid;
607 pthread_cond_t * cond;
608 int error;
609 kern_return_t kret;
610
611 cond = pthread_id_to_cond(condid);
612 if (cond == 0)
613 return(EINVAL);
614
615 COND_LOCK(cond->lock);
616
617 if (cond->sig != _PTHREAD_KERN_COND_SIG)
618 {
619 error = EINVAL;
620 goto out;
621 }
622
623 if ((p != cond->owner_proc) && (cond->pshared != PTHREAD_PROCESS_SHARED)) {
624 error = EINVAL;
625 goto out;
626 }
627
628 COND_UNLOCK(cond->lock);
629
630 kret = semaphore_signal(cond->sem);
631 switch (kret) {
632 case KERN_INVALID_ADDRESS:
633 case KERN_PROTECTION_FAILURE:
634 error = EINVAL;
635 break;
636 case KERN_ABORTED:
637 case KERN_OPERATION_TIMED_OUT:
638 error = EINTR;
639 break;
640 case KERN_SUCCESS:
641 error = 0;
642 break;
643 default:
644 error = EINVAL;
645 break;
646 }
647
648 COND_LOCK(cond->lock);
649 out:
650 COND_UNLOCK(cond->lock);
651 pthread_cond_release(cond);
652 return (error);
653 }
654
655
656 int
657 __pthread_cond_wait(__unused struct proc *p, struct __pthread_cond_wait_args *uap, __unused register_t *retval)
658 {
659 int condid = uap->condid;
660 pthread_cond_t * cond;
661 int mutexid = uap->mutexid;
662 pthread_mutex_t * mutex;
663 int error;
664 kern_return_t kret;
665
666 cond = pthread_id_to_cond(condid);
667 if (cond == 0)
668 return(EINVAL);
669
670 mutex = pthread_id_to_mutex(mutexid);
671 if (mutex == 0) {
672 pthread_cond_release(cond);
673 return(EINVAL);
674 }
675 COND_LOCK(cond->lock);
676
677 if (cond->sig != _PTHREAD_KERN_COND_SIG)
678 {
679 error = EINVAL;
680 goto out;
681 }
682
683 if ((p != cond->owner_proc) && (cond->pshared != PTHREAD_PROCESS_SHARED)) {
684 error = EINVAL;
685 goto out;
686 }
687
688 COND_UNLOCK(cond->lock);
689
690 kret = semaphore_wait(cond->sem);
691 switch (kret) {
692 case KERN_INVALID_ADDRESS:
693 case KERN_PROTECTION_FAILURE:
694 error = EACCES;
695 break;
696 case KERN_ABORTED:
697 case KERN_OPERATION_TIMED_OUT:
698 error = EINTR;
699 break;
700 case KERN_SUCCESS:
701 error = 0;
702 break;
703 default:
704 error = EINVAL;
705 break;
706 }
707
708 COND_LOCK(cond->lock);
709 out:
710 COND_UNLOCK(cond->lock);
711 pthread_cond_release(cond);
712 pthread_mutex_release(mutex);
713 return (error);
714 }
715
716 int
717 __pthread_cond_timedwait(__unused struct proc *p, struct __pthread_cond_timedwait_args *uap, __unused register_t *retval)
718 {
719 int condid = uap->condid;
720 pthread_cond_t * cond;
721 int mutexid = uap->mutexid;
722 pthread_mutex_t * mutex;
723 mach_timespec_t absts;
724 int error;
725 kern_return_t kret;
726
727 absts.tv_sec = 0;
728 absts.tv_nsec = 0;
729
730 if (uap->abstime)
731 if ((error = copyin(uap->abstime, &absts, sizeof(mach_timespec_t ))))
732 return(error);
733 cond = pthread_id_to_cond(condid);
734 if (cond == 0)
735 return(EINVAL);
736
737 mutex = pthread_id_to_mutex(mutexid);
738 if (mutex == 0) {
739 pthread_cond_release(cond);
740 return(EINVAL);
741 }
742 COND_LOCK(cond->lock);
743
744 if (cond->sig != _PTHREAD_KERN_COND_SIG)
745 {
746 error = EINVAL;
747 goto out;
748 }
749
750 if ((p != cond->owner_proc) && (cond->pshared != PTHREAD_PROCESS_SHARED)) {
751 error = EINVAL;
752 goto out;
753 }
754
755 COND_UNLOCK(cond->lock);
756
757 kret = semaphore_timedwait(cond->sem, absts);
758 switch (kret) {
759 case KERN_INVALID_ADDRESS:
760 case KERN_PROTECTION_FAILURE:
761 error = EACCES;
762 break;
763 case KERN_ABORTED:
764 case KERN_OPERATION_TIMED_OUT:
765 error = EINTR;
766 break;
767 case KERN_SUCCESS:
768 error = 0;
769 break;
770 default:
771 error = EINVAL;
772 break;
773 }
774
775 COND_LOCK(cond->lock);
776 out:
777 COND_UNLOCK(cond->lock);
778 pthread_cond_release(cond);
779 pthread_mutex_release(mutex);
780 return (error);
781 }
782
783 int
784 bsdthread_create(__unused struct proc *p, struct bsdthread_create_args *uap, user_addr_t *retval)
785 {
786 kern_return_t kret;
787 void * sright;
788 int error = 0;
789 int allocated = 0;
790 mach_vm_offset_t stackaddr;
791 mach_vm_size_t th_allocsize = 0;
792 mach_vm_size_t user_stacksize;
793 mach_vm_size_t th_stacksize;
794 mach_vm_offset_t th_stackaddr;
795 mach_vm_offset_t th_stack;
796 mach_vm_offset_t th_pthread;
797 mach_port_t th_thport;
798 thread_t th;
799 user_addr_t user_func = uap->func;
800 user_addr_t user_funcarg = uap->func_arg;
801 user_addr_t user_stack = uap->stack;
802 user_addr_t user_pthread = uap->pthread;
803 unsigned int flags = (unsigned int)uap->flags;
804 vm_map_t vmap = current_map();
805 task_t ctask = current_task();
806 unsigned int policy, importance;
807
808 int isLP64 = 0;
809
810
811 #if 0
812 KERNEL_DEBUG_CONSTANT(0x9000080 | DBG_FUNC_START, flags, 0, 0, 0, 0);
813 #endif
814
815 isLP64 = IS_64BIT_PROCESS(p);
816
817
818 #if defined(__ppc__)
819 stackaddr = 0xF0000000;
820 #elif defined(__i386__)
821 stackaddr = 0xB0000000;
822 #else
823 #error Need to define a stack address hint for this architecture
824 #endif
825 kret = thread_create(ctask, &th);
826 if (kret != KERN_SUCCESS)
827 return(ENOMEM);
828 thread_reference(th);
829
830 sright = (void *) convert_thread_to_port(th);
831 th_thport = (void *)ipc_port_copyout_send(sright, get_task_ipcspace(ctask));
832
833 if ((flags & PTHREAD_START_CUSTOM) == 0) {
834 th_stacksize = (mach_vm_size_t)user_stack; /* if it is custom them it is stacksize */
835 th_allocsize = th_stacksize + PTH_DEFAULT_GUARDSIZE + p->p_pthsize;
836
837 kret = mach_vm_map(vmap, &stackaddr,
838 th_allocsize,
839 page_size-1,
840 VM_MAKE_TAG(VM_MEMORY_STACK)| VM_FLAGS_ANYWHERE , NULL,
841 0, FALSE, VM_PROT_DEFAULT, VM_PROT_ALL,
842 VM_INHERIT_DEFAULT);
843 if (kret != KERN_SUCCESS)
844 kret = mach_vm_allocate(vmap,
845 &stackaddr, th_allocsize,
846 VM_MAKE_TAG(VM_MEMORY_STACK)| VM_FLAGS_ANYWHERE);
847 if (kret != KERN_SUCCESS) {
848 error = ENOMEM;
849 goto out;
850 }
851 #if 0
852 KERNEL_DEBUG_CONSTANT(0x9000080 |DBG_FUNC_NONE, th_allocsize, stackaddr, 0, 2, 0);
853 #endif
854 th_stackaddr = stackaddr;
855 allocated = 1;
856 /*
857 * The guard page is at the lowest address
858 * The stack base is the highest address
859 */
860 kret = mach_vm_protect(vmap, stackaddr, PTH_DEFAULT_GUARDSIZE, FALSE, VM_PROT_NONE);
861
862 if (kret != KERN_SUCCESS) {
863 error = ENOMEM;
864 goto out1;
865 }
866 th_stack = (stackaddr + th_stacksize + PTH_DEFAULT_GUARDSIZE);
867 th_pthread = (stackaddr + th_stacksize + PTH_DEFAULT_GUARDSIZE);
868 user_stacksize = th_stacksize;
869 } else {
870 th_stack = user_stack;
871 user_stacksize = user_stack;
872 th_pthread = user_pthread;
873 #if 0
874 KERNEL_DEBUG_CONSTANT(0x9000080 |DBG_FUNC_NONE, 0, 0, 0, 3, 0);
875 #endif
876 }
877
878 #if defined(__ppc__)
879 /*
880 * Set up PowerPC registers...
881 * internally they are always kept as 64 bit and
882 * since the register set is the same between 32 and 64bit modes
883 * we don't need 2 different methods for setting the state
884 */
885 {
886 ppc_thread_state64_t state64;
887 ppc_thread_state64_t *ts64 = &state64;
888
889 ts64->srr0 = (uint64_t)p->p_threadstart;
890 ts64->r1 = (uint64_t)(th_stack - C_ARGSAVE_LEN - C_RED_ZONE);
891 ts64->r3 = (uint64_t)th_pthread;
892 ts64->r4 = (uint64_t)((unsigned int)th_thport);
893 ts64->r5 = (uint64_t)user_func;
894 ts64->r6 = (uint64_t)user_funcarg;
895 ts64->r7 = (uint64_t)user_stacksize;
896 ts64->r8 = (uint64_t)uap->flags;
897
898 thread_set_wq_state64(th, (thread_state_t)ts64);
899
900 thread_set_cthreadself(th, (uint64_t)th_pthread, isLP64);
901 }
902 #elif defined(__i386__)
903 {
904 /*
905 * Set up i386 registers & function call.
906 */
907 if (isLP64 == 0) {
908 x86_thread_state32_t state;
909 x86_thread_state32_t *ts = &state;
910
911 ts->eip = (int)p->p_threadstart;
912 ts->eax = (unsigned int)th_pthread;
913 ts->ebx = (unsigned int)th_thport;
914 ts->ecx = (unsigned int)user_func;
915 ts->edx = (unsigned int)user_funcarg;
916 ts->edi = (unsigned int)user_stacksize;
917 ts->esi = (unsigned int)uap->flags;
918 /*
919 * set stack pointer
920 */
921 ts->esp = (int)((vm_offset_t)(th_stack-C_32_STK_ALIGN));
922
923 thread_set_wq_state32(th, (thread_state_t)ts);
924
925 } else {
926 x86_thread_state64_t state64;
927 x86_thread_state64_t *ts64 = &state64;
928
929 ts64->rip = (uint64_t)p->p_threadstart;
930 ts64->rdi = (uint64_t)th_pthread;
931 ts64->rsi = (uint64_t)((unsigned int)(th_thport));
932 ts64->rdx = (uint64_t)user_func;
933 ts64->rcx = (uint64_t)user_funcarg;
934 ts64->r8 = (uint64_t)user_stacksize;
935 ts64->r9 = (uint64_t)uap->flags;
936 /*
937 * set stack pointer aligned to 16 byte boundary
938 */
939 ts64->rsp = (uint64_t)(th_stack - C_64_REDZONE_LEN);
940
941 thread_set_wq_state64(th, (thread_state_t)ts64);
942 }
943 }
944 #else
945 #error bsdthread_create not defined for this architecture
946 #endif
947 /* Set scheduling parameters if needed */
948 if ((flags & PTHREAD_START_SETSCHED) != 0) {
949 thread_extended_policy_data_t extinfo;
950 thread_precedence_policy_data_t precedinfo;
951
952 importance = (flags & PTHREAD_START_IMPORTANCE_MASK);
953 policy = (flags >> PTHREAD_START_POLICY_BITSHIFT) & PTHREAD_START_POLICY_MASK;
954
955 if (policy == SCHED_OTHER)
956 extinfo.timeshare = 1;
957 else
958 extinfo.timeshare = 0;
959 thread_policy_set(th, THREAD_EXTENDED_POLICY, (thread_policy_t)&extinfo, THREAD_EXTENDED_POLICY_COUNT);
960
961 #define BASEPRI_DEFAULT 31
962 precedinfo.importance = (importance - BASEPRI_DEFAULT);
963 thread_policy_set(th, THREAD_PRECEDENCE_POLICY, (thread_policy_t)&precedinfo, THREAD_PRECEDENCE_POLICY_COUNT);
964 }
965
966 kret = thread_resume(th);
967 if (kret != KERN_SUCCESS) {
968 error = EINVAL;
969 goto out1;
970 }
971 thread_deallocate(th); /* drop the creator reference */
972 #if 0
973 KERNEL_DEBUG_CONSTANT(0x9000080 |DBG_FUNC_END, error, (unsigned int)th_pthread, 0, 0, 0);
974 #endif
975 *retval = th_pthread;
976
977 return(0);
978
979 out1:
980 if (allocated != 0)
981 (void)mach_vm_deallocate(vmap, stackaddr, th_allocsize);
982 out:
983 (void)mach_port_deallocate(get_task_ipcspace(ctask), (mach_port_name_t)th_thport);
984 (void)thread_terminate(th);
985 (void)thread_deallocate(th);
986 return(error);
987 }
988
989 int
990 bsdthread_terminate(__unused struct proc *p, struct bsdthread_terminate_args *uap, __unused register_t *retval)
991 {
992 mach_vm_offset_t freeaddr;
993 mach_vm_size_t freesize;
994 kern_return_t kret;
995 mach_port_name_t kthport = (mach_port_name_t)uap->port;
996 mach_port_name_t sem = (mach_port_name_t)uap->sem;
997
998 freeaddr = (mach_vm_offset_t)uap->stackaddr;
999 freesize = uap->freesize;
1000
1001 #if 0
1002 KERNEL_DEBUG_CONSTANT(0x9000084 |DBG_FUNC_START, (unsigned int)freeaddr, (unsigned int)freesize, (unsigned int)kthport, 0xff, 0);
1003 #endif
1004 if (sem != MACH_PORT_NULL) {
1005 kret = semaphore_signal_internal_trap(sem);
1006 if (kret != KERN_SUCCESS) {
1007 return(EINVAL);
1008 }
1009 }
1010 if ((freesize != (mach_vm_size_t)0) && (freeaddr != (mach_vm_offset_t)0)) {
1011 kret = mach_vm_deallocate(current_map(), freeaddr, freesize);
1012 if (kret != KERN_SUCCESS) {
1013 return(EINVAL);
1014 }
1015 }
1016
1017 (void) thread_terminate(current_thread());
1018 if (kthport != MACH_PORT_NULL)
1019 mach_port_deallocate(get_task_ipcspace(current_task()), kthport);
1020 thread_exception_return();
1021 panic("bsdthread_terminate: still running\n");
1022 #if 0
1023 KERNEL_DEBUG_CONSTANT(0x9000084 |DBG_FUNC_END, 0, 0, 0, 0xff, 0);
1024 #endif
1025 return(0);
1026 }
1027
1028
1029 int
1030 bsdthread_register(struct proc *p, struct bsdthread_register_args *uap, __unused register_t *retval)
1031 {
1032 /* syscall randomizer test can pass bogus values */
1033 if (uap->pthsize > MAX_PTHREAD_SIZE) {
1034 return(EINVAL);
1035 }
1036 p->p_threadstart = uap->threadstart;
1037 p->p_wqthread = uap->wqthread;
1038 p->p_pthsize = uap->pthsize;
1039
1040 return(0);
1041 }
1042
1043
1044
1045
1046 int wq_stalled_window_usecs = WQ_STALLED_WINDOW_USECS;
1047 int wq_reduce_pool_window_usecs = WQ_REDUCE_POOL_WINDOW_USECS;
1048 int wq_max_run_latency_usecs = WQ_MAX_RUN_LATENCY_USECS;
1049 int wq_timer_interval_msecs = WQ_TIMER_INTERVAL_MSECS;
1050
1051
1052 SYSCTL_INT(_kern, OID_AUTO, wq_stalled_window_usecs, CTLFLAG_RW,
1053 &wq_stalled_window_usecs, 0, "");
1054
1055 SYSCTL_INT(_kern, OID_AUTO, wq_reduce_pool_window_usecs, CTLFLAG_RW,
1056 &wq_reduce_pool_window_usecs, 0, "");
1057
1058 SYSCTL_INT(_kern, OID_AUTO, wq_max_run_latency_usecs, CTLFLAG_RW,
1059 &wq_max_run_latency_usecs, 0, "");
1060
1061 SYSCTL_INT(_kern, OID_AUTO, wq_timer_interval_msecs, CTLFLAG_RW,
1062 &wq_timer_interval_msecs, 0, "");
1063
1064
1065
1066
1067 void
1068 workqueue_init_lock(proc_t p)
1069 {
1070 lck_mtx_init(&p->p_wqlock, pthread_lck_grp, pthread_lck_attr);
1071 }
1072
1073 void
1074 workqueue_destroy_lock(proc_t p)
1075 {
1076 lck_mtx_destroy(&p->p_wqlock, pthread_lck_grp);
1077 }
1078
1079 static void
1080 workqueue_lock(proc_t p)
1081 {
1082 lck_mtx_lock(&p->p_wqlock);
1083 }
1084
1085 static void
1086 workqueue_lock_spin(proc_t p)
1087 {
1088 lck_mtx_lock_spin(&p->p_wqlock);
1089 }
1090
1091 static void
1092 workqueue_unlock(proc_t p)
1093 {
1094 lck_mtx_unlock(&p->p_wqlock);
1095 }
1096
1097
1098
1099 static void
1100 workqueue_interval_timer_start(thread_call_t call, int interval_in_ms)
1101 {
1102 uint64_t deadline;
1103
1104 clock_interval_to_deadline(interval_in_ms, 1000 * 1000, &deadline);
1105
1106 thread_call_enter_delayed(call, deadline);
1107 }
1108
1109
1110 static void
1111 workqueue_timer(struct workqueue *wq, __unused int param1)
1112 {
1113 struct timeval tv, dtv;
1114 uint32_t i;
1115 boolean_t added_more_threads = FALSE;
1116 boolean_t reset_maxactive = FALSE;
1117 boolean_t restart_timer = FALSE;
1118
1119 microuptime(&tv);
1120
1121 KERNEL_DEBUG(0xefffd108, (int)wq, 0, 0, 0, 0);
1122
1123 /*
1124 * check to see if the stall frequency was beyond our tolerance
1125 * or we have work on the queue, but haven't scheduled any
1126 * new work within our acceptable time interval because
1127 * there were no idle threads left to schedule
1128 *
1129 * WQ_TIMER_WATCH will only be set if we have 1 or more affinity
1130 * groups that have stalled (no active threads and no idle threads)...
1131 * it will not be set if all affinity groups have at least 1 thread
1132 * that is currently runnable... if all processors have a runnable
1133 * thread, there is no need to add more threads even if we're not
1134 * scheduling new work within our allowed window... it just means
1135 * that the work items are taking a long time to complete.
1136 */
1137 if (wq->wq_flags & (WQ_ADD_TO_POOL | WQ_TIMER_WATCH)) {
1138
1139 if (wq->wq_flags & WQ_ADD_TO_POOL)
1140 added_more_threads = TRUE;
1141 else {
1142 timersub(&tv, &wq->wq_lastran_ts, &dtv);
1143
1144 if (((dtv.tv_sec * 1000000) + dtv.tv_usec) > wq_stalled_window_usecs)
1145 added_more_threads = TRUE;
1146 }
1147 if (added_more_threads == TRUE) {
1148 for (i = 0; i < wq->wq_affinity_max && wq->wq_nthreads < WORKQUEUE_MAXTHREADS; i++) {
1149 (void)workqueue_addnewthread(wq);
1150 }
1151 }
1152 }
1153 timersub(&tv, &wq->wq_reduce_ts, &dtv);
1154
1155 if (((dtv.tv_sec * 1000000) + dtv.tv_usec) > wq_reduce_pool_window_usecs)
1156 reset_maxactive = TRUE;
1157
1158 /*
1159 * if the pool size has grown beyond the minimum number
1160 * of threads needed to keep all of the processors busy, and
1161 * the maximum number of threads scheduled concurrently during
1162 * the last sample period didn't exceed half the current pool
1163 * size, then its time to trim the pool size back
1164 */
1165 if (added_more_threads == FALSE &&
1166 reset_maxactive == TRUE &&
1167 wq->wq_nthreads > wq->wq_affinity_max &&
1168 wq->wq_max_threads_scheduled <= (wq->wq_nthreads / 2)) {
1169 uint32_t nthreads_to_remove;
1170
1171 if ((nthreads_to_remove = (wq->wq_nthreads / 4)) == 0)
1172 nthreads_to_remove = 1;
1173
1174 for (i = 0; i < nthreads_to_remove && wq->wq_nthreads > wq->wq_affinity_max; i++)
1175 workqueue_removethread(wq);
1176 }
1177 workqueue_lock_spin(wq->wq_proc);
1178
1179 if (reset_maxactive == TRUE) {
1180 wq->wq_max_threads_scheduled = 0;
1181 microuptime(&wq->wq_reduce_ts);
1182 }
1183 if (added_more_threads) {
1184 wq->wq_flags &= ~(WQ_ADD_TO_POOL | WQ_TIMER_WATCH);
1185
1186 /*
1187 * since we added more threads, we should be
1188 * able to run some work if its still available
1189 */
1190 workqueue_run_nextitem(wq->wq_proc, THREAD_NULL);
1191 workqueue_lock_spin(wq->wq_proc);
1192 }
1193 if ((wq->wq_nthreads > wq->wq_affinity_max) ||
1194 (wq->wq_flags & WQ_TIMER_WATCH)) {
1195 restart_timer = TRUE;
1196 } else
1197 wq->wq_flags &= ~WQ_TIMER_RUNNING;
1198
1199 workqueue_unlock(wq->wq_proc);
1200
1201 /*
1202 * we needed to knock down the WQ_TIMER_RUNNING flag while behind
1203 * the workqueue lock... however, we don't want to hold the lock
1204 * while restarting the timer and we certainly don't want 2 or more
1205 * instances of the timer... so set a local to indicate the need
1206 * for a restart since the state of wq_flags may change once we
1207 * drop the workqueue lock...
1208 */
1209 if (restart_timer == TRUE)
1210 workqueue_interval_timer_start(wq->wq_timer_call, wq_timer_interval_msecs);
1211 }
1212
1213
1214 static void
1215 workqueue_callback(
1216 int type,
1217 thread_t thread)
1218 {
1219 struct uthread *uth;
1220 struct threadlist *tl;
1221 struct workqueue *wq;
1222
1223 uth = get_bsdthread_info(thread);
1224 tl = uth->uu_threadlist;
1225 wq = tl->th_workq;
1226
1227 switch (type) {
1228
1229 case SCHED_CALL_BLOCK:
1230 {
1231 uint32_t old_activecount;
1232
1233 old_activecount = OSAddAtomic(-1, (SInt32 *)&wq->wq_thactivecount[tl->th_affinity_tag]);
1234
1235 if (old_activecount == 1 && wq->wq_itemcount) {
1236 /*
1237 * we were the last active thread on this affinity set
1238 * and we've got work to do
1239 */
1240 workqueue_lock_spin(wq->wq_proc);
1241 /*
1242 * if this thread is blocking (not parking)
1243 * and the idle list is empty for this affinity group
1244 * we'll count it as a 'stall'
1245 */
1246 if ((tl->th_flags & TH_LIST_RUNNING) &&
1247 TAILQ_EMPTY(&wq->wq_thidlelist[tl->th_affinity_tag]))
1248 wq->wq_stalled_count++;
1249
1250 workqueue_run_nextitem(wq->wq_proc, THREAD_NULL);
1251 /*
1252 * workqueue_run_nextitem will drop the workqueue
1253 * lock before it returns
1254 */
1255 }
1256 KERNEL_DEBUG(0xefffd020, (int)thread, wq->wq_threads_scheduled, tl->th_affinity_tag, 0, 0);
1257 }
1258 break;
1259
1260 case SCHED_CALL_UNBLOCK:
1261 /*
1262 * we cannot take the workqueue_lock here...
1263 * an UNBLOCK can occur from a timer event which
1264 * is run from an interrupt context... if the workqueue_lock
1265 * is already held by this processor, we'll deadlock...
1266 * the thread lock for the thread being UNBLOCKED
1267 * is also held
1268 */
1269 if (tl->th_unparked)
1270 OSAddAtomic(-1, (SInt32 *)&tl->th_unparked);
1271 else
1272 OSAddAtomic(1, (SInt32 *)&wq->wq_thactivecount[tl->th_affinity_tag]);
1273
1274 KERNEL_DEBUG(0xefffd024, (int)thread, wq->wq_threads_scheduled, tl->th_affinity_tag, 0, 0);
1275 break;
1276 }
1277 }
1278
1279 static void
1280 workqueue_removethread(struct workqueue *wq)
1281 {
1282 struct threadlist *tl;
1283 uint32_t i, affinity_tag = 0;
1284
1285 tl = NULL;
1286
1287 workqueue_lock_spin(wq->wq_proc);
1288
1289 for (i = 0; i < wq->wq_affinity_max; i++) {
1290
1291 affinity_tag = wq->wq_nextaffinitytag;
1292
1293 if (affinity_tag == 0)
1294 affinity_tag = wq->wq_affinity_max - 1;
1295 else
1296 affinity_tag--;
1297 wq->wq_nextaffinitytag = affinity_tag;
1298
1299 /*
1300 * look for an idle thread to steal from this affinity group
1301 * but don't grab the only thread associated with it
1302 */
1303 if (!TAILQ_EMPTY(&wq->wq_thidlelist[affinity_tag]) && wq->wq_thcount[affinity_tag] > 1) {
1304 tl = TAILQ_FIRST(&wq->wq_thidlelist[affinity_tag]);
1305 TAILQ_REMOVE(&wq->wq_thidlelist[affinity_tag], tl, th_entry);
1306
1307 wq->wq_nthreads--;
1308 wq->wq_thcount[affinity_tag]--;
1309
1310 break;
1311 }
1312 }
1313 workqueue_unlock(wq->wq_proc);
1314
1315 if (tl != NULL) {
1316 thread_sched_call(tl->th_thread, NULL);
1317
1318 if ( (tl->th_flags & TH_LIST_BLOCKED) )
1319 wakeup(tl);
1320 else {
1321 /*
1322 * thread was created, but never used...
1323 * need to clean up the stack and port ourselves
1324 * since we're not going to spin up through the
1325 * normal exit path triggered from Libc
1326 */
1327 (void)mach_vm_deallocate(wq->wq_map, tl->th_stackaddr, tl->th_allocsize);
1328 (void)mach_port_deallocate(get_task_ipcspace(wq->wq_task), (mach_port_name_t)tl->th_thport);
1329
1330 thread_terminate(tl->th_thread);
1331 }
1332 KERNEL_DEBUG(0xefffd030, (int)tl->th_thread, wq->wq_nthreads, tl->th_flags & TH_LIST_BLOCKED, 0, 0);
1333 /*
1334 * drop our ref on the thread
1335 */
1336 thread_deallocate(tl->th_thread);
1337
1338 kfree(tl, sizeof(struct threadlist));
1339 }
1340 }
1341
1342
1343 static int
1344 workqueue_addnewthread(struct workqueue *wq)
1345 {
1346 struct threadlist *tl;
1347 struct uthread *uth;
1348 kern_return_t kret;
1349 thread_t th;
1350 proc_t p;
1351 void *sright;
1352 mach_vm_offset_t stackaddr;
1353 uint32_t affinity_tag;
1354
1355 p = wq->wq_proc;
1356
1357 kret = thread_create(wq->wq_task, &th);
1358
1359 if (kret != KERN_SUCCESS)
1360 return(EINVAL);
1361
1362 tl = kalloc(sizeof(struct threadlist));
1363 bzero(tl, sizeof(struct threadlist));
1364
1365 #if defined(__ppc__)
1366 stackaddr = 0xF0000000;
1367 #elif defined(__i386__)
1368 stackaddr = 0xB0000000;
1369 #else
1370 #error Need to define a stack address hint for this architecture
1371 #endif
1372 tl->th_allocsize = PTH_DEFAULT_STACKSIZE + PTH_DEFAULT_GUARDSIZE + p->p_pthsize;
1373
1374 kret = mach_vm_map(wq->wq_map, &stackaddr,
1375 tl->th_allocsize,
1376 page_size-1,
1377 VM_MAKE_TAG(VM_MEMORY_STACK)| VM_FLAGS_ANYWHERE , NULL,
1378 0, FALSE, VM_PROT_DEFAULT, VM_PROT_ALL,
1379 VM_INHERIT_DEFAULT);
1380
1381 if (kret != KERN_SUCCESS) {
1382 kret = mach_vm_allocate(wq->wq_map,
1383 &stackaddr, tl->th_allocsize,
1384 VM_MAKE_TAG(VM_MEMORY_STACK) | VM_FLAGS_ANYWHERE);
1385 }
1386 if (kret == KERN_SUCCESS) {
1387 /*
1388 * The guard page is at the lowest address
1389 * The stack base is the highest address
1390 */
1391 kret = mach_vm_protect(wq->wq_map, stackaddr, PTH_DEFAULT_GUARDSIZE, FALSE, VM_PROT_NONE);
1392
1393 if (kret != KERN_SUCCESS)
1394 (void) mach_vm_deallocate(wq->wq_map, stackaddr, tl->th_allocsize);
1395 }
1396 if (kret != KERN_SUCCESS) {
1397 (void) thread_terminate(th);
1398
1399 kfree(tl, sizeof(struct threadlist));
1400
1401 return(EINVAL);
1402 }
1403 thread_reference(th);
1404
1405 sright = (void *) convert_thread_to_port(th);
1406 tl->th_thport = (void *)ipc_port_copyout_send(sright, get_task_ipcspace(wq->wq_task));
1407
1408 thread_static_param(th, TRUE);
1409
1410 workqueue_lock_spin(p);
1411
1412 affinity_tag = wq->wq_nextaffinitytag;
1413 wq->wq_nextaffinitytag = (affinity_tag + 1) % wq->wq_affinity_max;
1414
1415 workqueue_unlock(p);
1416
1417 tl->th_flags = TH_LIST_INITED | TH_LIST_SUSPENDED;
1418
1419 tl->th_thread = th;
1420 tl->th_workq = wq;
1421 tl->th_stackaddr = stackaddr;
1422 tl->th_affinity_tag = affinity_tag;
1423
1424 #if defined(__ppc__)
1425 //ml_fp_setvalid(FALSE);
1426 thread_set_cthreadself(th, (uint64_t)(tl->th_stackaddr + PTH_DEFAULT_STACKSIZE + PTH_DEFAULT_GUARDSIZE), IS_64BIT_PROCESS(p));
1427 #endif /* __ppc__ */
1428 /*
1429 * affinity tag of 0 means no affinity...
1430 * but we want our tags to be 0 based because they
1431 * are used to index arrays, so...
1432 * keep it 0 based internally and bump by 1 when
1433 * calling out to set it
1434 */
1435 (void)thread_affinity_set(th, affinity_tag + 1);
1436 thread_sched_call(th, workqueue_callback);
1437
1438 uth = get_bsdthread_info(tl->th_thread);
1439 uth->uu_threadlist = (void *)tl;
1440
1441 workqueue_lock_spin(p);
1442
1443 TAILQ_INSERT_TAIL(&wq->wq_thidlelist[tl->th_affinity_tag], tl, th_entry);
1444 wq->wq_nthreads++;
1445 wq->wq_thcount[affinity_tag]++;
1446
1447 KERNEL_DEBUG1(0xefffd014 | DBG_FUNC_START, (int)current_thread(), affinity_tag, wq->wq_nthreads, 0, (int)tl->th_thread);
1448
1449 /*
1450 * work may have come into the queue while
1451 * no threads were available to run... since
1452 * we're adding a new thread, go evaluate the
1453 * current state
1454 */
1455 workqueue_run_nextitem(p, THREAD_NULL);
1456 /*
1457 * workqueue_run_nextitem is responsible for
1458 * dropping the workqueue lock in all cases
1459 */
1460
1461 return(0);
1462 }
1463
1464 int
1465 workq_open(__unused struct proc *p, __unused struct workq_open_args *uap, __unused register_t *retval)
1466 {
1467 struct workqueue * wq;
1468 int size;
1469 char * ptr;
1470 int j;
1471 uint32_t i;
1472 int error = 0;
1473 int num_cpus;
1474 struct workitem * witem;
1475 struct workitemlist *wl;
1476
1477 workqueue_lock(p);
1478
1479 if (p->p_wqptr == NULL) {
1480 num_cpus = ml_get_max_cpus();
1481
1482 size = (sizeof(struct workqueue)) +
1483 (num_cpus * sizeof(int *)) +
1484 (num_cpus * sizeof(TAILQ_HEAD(, threadlist)));
1485
1486 ptr = (char *)kalloc(size);
1487 bzero(ptr, size);
1488
1489 wq = (struct workqueue *)ptr;
1490 wq->wq_flags = WQ_LIST_INITED;
1491 wq->wq_proc = p;
1492 wq->wq_affinity_max = num_cpus;
1493 wq->wq_task = current_task();
1494 wq->wq_map = current_map();
1495
1496 for (i = 0; i < WORKQUEUE_NUMPRIOS; i++) {
1497 wl = (struct workitemlist *)&wq->wq_list[i];
1498 TAILQ_INIT(&wl->wl_itemlist);
1499 TAILQ_INIT(&wl->wl_freelist);
1500
1501 for (j = 0; j < WORKITEM_SIZE; j++) {
1502 witem = &wq->wq_array[(i*WORKITEM_SIZE) + j];
1503 TAILQ_INSERT_TAIL(&wl->wl_freelist, witem, wi_entry);
1504 }
1505 }
1506 wq->wq_thactivecount = (uint32_t *)((char *)ptr + sizeof(struct workqueue));
1507 wq->wq_thcount = (uint32_t *)&wq->wq_thactivecount[wq->wq_affinity_max];
1508 wq->wq_thidlelist = (struct wq_thidlelist *)&wq->wq_thcount[wq->wq_affinity_max];
1509
1510 for (i = 0; i < wq->wq_affinity_max; i++)
1511 TAILQ_INIT(&wq->wq_thidlelist[i]);
1512
1513 TAILQ_INIT(&wq->wq_thrunlist);
1514
1515 p->p_wqptr = (void *)wq;
1516 p->p_wqsize = size;
1517
1518 workqueue_unlock(p);
1519
1520 wq->wq_timer_call = thread_call_allocate((thread_call_func_t)workqueue_timer, (thread_call_param_t)wq);
1521
1522 for (i = 0; i < wq->wq_affinity_max; i++) {
1523 (void)workqueue_addnewthread(wq);
1524 }
1525 /* If unable to create any threads, return error */
1526 if (wq->wq_nthreads == 0)
1527 error = EINVAL;
1528 workqueue_lock_spin(p);
1529
1530 microuptime(&wq->wq_reduce_ts);
1531 microuptime(&wq->wq_lastran_ts);
1532 wq->wq_max_threads_scheduled = 0;
1533 wq->wq_stalled_count = 0;
1534 }
1535 workqueue_unlock(p);
1536
1537 return(error);
1538 }
1539
1540 int
1541 workq_ops(struct proc *p, struct workq_ops_args *uap, __unused register_t *retval)
1542 {
1543 int options = uap->options;
1544 int prio = uap->prio; /* should be used to find the right workqueue */
1545 user_addr_t item = uap->item;
1546 int error = 0;
1547 thread_t th = THREAD_NULL;
1548 struct workqueue *wq;
1549
1550 prio += 2; /* normalize prio -2 to +2 to 0 -4 */
1551
1552 switch (options) {
1553
1554 case WQOPS_QUEUE_ADD: {
1555
1556 KERNEL_DEBUG(0xefffd008 | DBG_FUNC_NONE, (int)item, 0, 0, 0, 0);
1557
1558 workqueue_lock_spin(p);
1559
1560 if ((wq = (struct workqueue *)p->p_wqptr) == NULL) {
1561 workqueue_unlock(p);
1562 return (EINVAL);
1563 }
1564 error = workqueue_additem(wq, prio, item);
1565
1566 }
1567 break;
1568 case WQOPS_QUEUE_REMOVE: {
1569
1570 workqueue_lock_spin(p);
1571
1572 if ((wq = (struct workqueue *)p->p_wqptr) == NULL) {
1573 workqueue_unlock(p);
1574 return (EINVAL);
1575 }
1576 error = workqueue_removeitem(wq, prio, item);
1577 }
1578 break;
1579 case WQOPS_THREAD_RETURN: {
1580
1581 th = current_thread();
1582
1583 KERNEL_DEBUG(0xefffd004 | DBG_FUNC_END, 0, 0, 0, 0, 0);
1584
1585 workqueue_lock_spin(p);
1586
1587 if ((wq = (struct workqueue *)p->p_wqptr) == NULL) {
1588 workqueue_unlock(p);
1589 return (EINVAL);
1590 }
1591 }
1592 break;
1593 default:
1594 return (EINVAL);
1595 }
1596 workqueue_run_nextitem(p, th);
1597 /*
1598 * workqueue_run_nextitem is responsible for
1599 * dropping the workqueue lock in all cases
1600 */
1601 return(error);
1602 }
1603
1604 void
1605 workqueue_exit(struct proc *p)
1606 {
1607 struct workqueue * wq;
1608 struct threadlist * tl, *tlist;
1609 uint32_t i;
1610
1611 if (p->p_wqptr != NULL) {
1612
1613 workqueue_lock_spin(p);
1614
1615 wq = (struct workqueue *)p->p_wqptr;
1616 p->p_wqptr = NULL;
1617
1618 workqueue_unlock(p);
1619
1620 if (wq == NULL)
1621 return;
1622
1623 if (wq->wq_flags & WQ_TIMER_RUNNING)
1624 thread_call_cancel(wq->wq_timer_call);
1625 thread_call_free(wq->wq_timer_call);
1626
1627 TAILQ_FOREACH_SAFE(tl, &wq->wq_thrunlist, th_entry, tlist) {
1628 /*
1629 * drop our last ref on the thread
1630 */
1631 thread_sched_call(tl->th_thread, NULL);
1632 thread_deallocate(tl->th_thread);
1633
1634 TAILQ_REMOVE(&wq->wq_thrunlist, tl, th_entry);
1635 kfree(tl, sizeof(struct threadlist));
1636 }
1637 for (i = 0; i < wq->wq_affinity_max; i++) {
1638 TAILQ_FOREACH_SAFE(tl, &wq->wq_thidlelist[i], th_entry, tlist) {
1639 /*
1640 * drop our last ref on the thread
1641 */
1642 thread_sched_call(tl->th_thread, NULL);
1643 thread_deallocate(tl->th_thread);
1644
1645 TAILQ_REMOVE(&wq->wq_thidlelist[i], tl, th_entry);
1646 kfree(tl, sizeof(struct threadlist));
1647 }
1648 }
1649 kfree(wq, p->p_wqsize);
1650 }
1651 }
1652
1653 static int
1654 workqueue_additem(struct workqueue *wq, int prio, user_addr_t item)
1655 {
1656 struct workitem *witem;
1657 struct workitemlist *wl;
1658
1659 wl = (struct workitemlist *)&wq->wq_list[prio];
1660
1661 if (TAILQ_EMPTY(&wl->wl_freelist))
1662 return (ENOMEM);
1663
1664 witem = (struct workitem *)TAILQ_FIRST(&wl->wl_freelist);
1665 TAILQ_REMOVE(&wl->wl_freelist, witem, wi_entry);
1666
1667 witem->wi_item = item;
1668 TAILQ_INSERT_TAIL(&wl->wl_itemlist, witem, wi_entry);
1669
1670 if (wq->wq_itemcount == 0) {
1671 microuptime(&wq->wq_lastran_ts);
1672 wq->wq_stalled_count = 0;
1673 }
1674 wq->wq_itemcount++;
1675
1676 return (0);
1677 }
1678
1679 static int
1680 workqueue_removeitem(struct workqueue *wq, int prio, user_addr_t item)
1681 {
1682 struct workitem *witem;
1683 struct workitemlist *wl;
1684 int error = ESRCH;
1685
1686 wl = (struct workitemlist *)&wq->wq_list[prio];
1687
1688 TAILQ_FOREACH(witem, &wl->wl_itemlist, wi_entry) {
1689 if (witem->wi_item == item) {
1690 TAILQ_REMOVE(&wl->wl_itemlist, witem, wi_entry);
1691 wq->wq_itemcount--;
1692
1693 witem->wi_item = (user_addr_t)0;
1694 TAILQ_INSERT_HEAD(&wl->wl_freelist, witem, wi_entry);
1695
1696 error = 0;
1697 break;
1698 }
1699 }
1700 if (wq->wq_itemcount == 0)
1701 wq->wq_flags &= ~(WQ_ADD_TO_POOL | WQ_TIMER_WATCH);
1702
1703 return (error);
1704 }
1705
1706 /*
1707 * workqueue_run_nextitem:
1708 * called with the workqueue lock held...
1709 * responsible for dropping it in all cases
1710 */
1711 static void
1712 workqueue_run_nextitem(proc_t p, thread_t thread)
1713 {
1714 struct workqueue *wq;
1715 struct workitem *witem = NULL;
1716 user_addr_t item = 0;
1717 thread_t th_to_run = THREAD_NULL;
1718 thread_t th_to_park = THREAD_NULL;
1719 int wake_thread = 0;
1720 int reuse_thread = 1;
1721 uint32_t stalled_affinity_count = 0;
1722 int i;
1723 uint32_t affinity_tag;
1724 struct threadlist *tl = NULL;
1725 struct uthread *uth = NULL;
1726 struct workitemlist *wl;
1727 boolean_t start_timer = FALSE;
1728 struct timeval tv, lat_tv;
1729
1730 wq = (struct workqueue *)p->p_wqptr;
1731
1732 KERNEL_DEBUG(0xefffd000 | DBG_FUNC_START, (int)thread, wq->wq_threads_scheduled, wq->wq_stalled_count, 0, 0);
1733
1734 if (wq->wq_itemcount == 0) {
1735 if ((th_to_park = thread) == THREAD_NULL)
1736 goto out;
1737 goto parkit;
1738 }
1739 if (thread != THREAD_NULL) {
1740 /*
1741 * we're a worker thread from the pool... currently we
1742 * are considered 'active' which means we're counted
1743 * in "wq_thactivecount"
1744 */
1745 uth = get_bsdthread_info(thread);
1746 tl = uth->uu_threadlist;
1747
1748 if (wq->wq_thactivecount[tl->th_affinity_tag] == 1) {
1749 /*
1750 * we're the only active thread associated with our
1751 * affinity group, so pick up some work and keep going
1752 */
1753 th_to_run = thread;
1754 goto pick_up_work;
1755 }
1756 }
1757 for (affinity_tag = 0; affinity_tag < wq->wq_affinity_max; affinity_tag++) {
1758 /*
1759 * look for first affinity group that is currently not active
1760 * and has at least 1 idle thread
1761 */
1762 if (wq->wq_thactivecount[affinity_tag] == 0) {
1763 if (!TAILQ_EMPTY(&wq->wq_thidlelist[affinity_tag]))
1764 break;
1765 stalled_affinity_count++;
1766 }
1767 }
1768 if (thread == THREAD_NULL) {
1769 /*
1770 * we're not one of the 'worker' threads
1771 */
1772 if (affinity_tag >= wq->wq_affinity_max) {
1773 /*
1774 * we've already got at least 1 thread per
1775 * affinity group in the active state... or
1776 * we've got no idle threads to play with
1777 */
1778 if (stalled_affinity_count) {
1779
1780 if ( !(wq->wq_flags & WQ_TIMER_RUNNING) ) {
1781 wq->wq_flags |= WQ_TIMER_RUNNING;
1782 start_timer = TRUE;
1783 }
1784 wq->wq_flags |= WQ_TIMER_WATCH;
1785 }
1786 goto out;
1787 }
1788 } else {
1789 /*
1790 * we're overbooked on the affinity group we're associated with,
1791 * so park this thread
1792 */
1793 th_to_park = thread;
1794
1795 if (affinity_tag >= wq->wq_affinity_max) {
1796 /*
1797 * all the affinity groups have active threads
1798 * running, or there are no idle threads to
1799 * schedule
1800 */
1801 if (stalled_affinity_count) {
1802
1803 if ( !(wq->wq_flags & WQ_TIMER_RUNNING) ) {
1804 wq->wq_flags |= WQ_TIMER_RUNNING;
1805 start_timer = TRUE;
1806 }
1807 wq->wq_flags |= WQ_TIMER_WATCH;
1808 }
1809 goto parkit;
1810 }
1811 /*
1812 * we've got a candidate (affinity group with no currently
1813 * active threads) to start a new thread on...
1814 * we already know there is both work available
1815 * and an idle thread with the correct affinity tag, so
1816 * fall into the code that pulls a new thread and workitem...
1817 * once we've kicked that thread off, we'll park this one
1818 */
1819 }
1820 tl = TAILQ_FIRST(&wq->wq_thidlelist[affinity_tag]);
1821 TAILQ_REMOVE(&wq->wq_thidlelist[affinity_tag], tl, th_entry);
1822
1823 th_to_run = tl->th_thread;
1824 TAILQ_INSERT_TAIL(&wq->wq_thrunlist, tl, th_entry);
1825
1826 if ((tl->th_flags & TH_LIST_SUSPENDED) == TH_LIST_SUSPENDED) {
1827 tl->th_flags &= ~TH_LIST_SUSPENDED;
1828 reuse_thread = 0;
1829 } else if ((tl->th_flags & TH_LIST_BLOCKED) == TH_LIST_BLOCKED) {
1830 tl->th_flags &= ~TH_LIST_BLOCKED;
1831 wake_thread = 1;
1832 }
1833 tl->th_flags |= TH_LIST_RUNNING;
1834
1835 wq->wq_threads_scheduled++;
1836
1837 if (wq->wq_threads_scheduled > wq->wq_max_threads_scheduled)
1838 wq->wq_max_threads_scheduled = wq->wq_threads_scheduled;
1839
1840 pick_up_work:
1841 for (i = 0; i < WORKQUEUE_NUMPRIOS; i++) {
1842 wl = (struct workitemlist *)&wq->wq_list[i];
1843
1844 if (!(TAILQ_EMPTY(&wl->wl_itemlist))) {
1845
1846 witem = TAILQ_FIRST(&wl->wl_itemlist);
1847 TAILQ_REMOVE(&wl->wl_itemlist, witem, wi_entry);
1848 wq->wq_itemcount--;
1849
1850 item = witem->wi_item;
1851 witem->wi_item = (user_addr_t)0;
1852 TAILQ_INSERT_HEAD(&wl->wl_freelist, witem, wi_entry);
1853
1854 break;
1855 }
1856 }
1857 if (witem == NULL)
1858 panic("workq_run_nextitem: NULL witem");
1859
1860 if (thread != th_to_run) {
1861 /*
1862 * we're starting up a thread from a parked/suspended condition
1863 */
1864 OSAddAtomic(1, (SInt32 *)&wq->wq_thactivecount[tl->th_affinity_tag]);
1865 OSAddAtomic(1, (SInt32 *)&tl->th_unparked);
1866 }
1867 if (wq->wq_itemcount == 0)
1868 wq->wq_flags &= ~WQ_TIMER_WATCH;
1869 else {
1870 microuptime(&tv);
1871 /*
1872 * if we had any affinity groups stall (no threads runnable)
1873 * since we last scheduled an item... and
1874 * the elapsed time since we last scheduled an item
1875 * exceeds the latency tolerance...
1876 * we ask the timer thread (which should already be running)
1877 * to add some more threads to the pool
1878 */
1879 if (wq->wq_stalled_count && !(wq->wq_flags & WQ_ADD_TO_POOL)) {
1880 timersub(&tv, &wq->wq_lastran_ts, &lat_tv);
1881
1882 if (((lat_tv.tv_sec * 1000000) + lat_tv.tv_usec) > wq_max_run_latency_usecs)
1883 wq->wq_flags |= WQ_ADD_TO_POOL;
1884
1885 KERNEL_DEBUG(0xefffd10c, wq->wq_stalled_count, lat_tv.tv_sec, lat_tv.tv_usec, wq->wq_flags, 0);
1886 }
1887 wq->wq_lastran_ts = tv;
1888 }
1889 wq->wq_stalled_count = 0;
1890 workqueue_unlock(p);
1891
1892 KERNEL_DEBUG(0xefffd02c, wq->wq_thactivecount[0], wq->wq_thactivecount[1],
1893 wq->wq_thactivecount[2], wq->wq_thactivecount[3], 0);
1894
1895 KERNEL_DEBUG(0xefffd02c, wq->wq_thactivecount[4], wq->wq_thactivecount[5],
1896 wq->wq_thactivecount[6], wq->wq_thactivecount[7], 0);
1897
1898 /*
1899 * if current thread is reused for workitem, does not return via unix_syscall
1900 */
1901 wq_runitem(p, item, th_to_run, tl, reuse_thread, wake_thread, (thread == th_to_run));
1902
1903 if (th_to_park == THREAD_NULL) {
1904
1905 KERNEL_DEBUG(0xefffd000 | DBG_FUNC_END, (int)thread, (int)item, wq->wq_flags, 1, 0);
1906
1907 return;
1908 }
1909 workqueue_lock_spin(p);
1910
1911 parkit:
1912 wq->wq_threads_scheduled--;
1913 /*
1914 * this is a workqueue thread with no more
1915 * work to do... park it for now
1916 */
1917 uth = get_bsdthread_info(th_to_park);
1918 tl = uth->uu_threadlist;
1919 if (tl == 0)
1920 panic("wq thread with no threadlist ");
1921
1922 TAILQ_REMOVE(&wq->wq_thrunlist, tl, th_entry);
1923 tl->th_flags &= ~TH_LIST_RUNNING;
1924
1925 tl->th_flags |= TH_LIST_BLOCKED;
1926 TAILQ_INSERT_HEAD(&wq->wq_thidlelist[tl->th_affinity_tag], tl, th_entry);
1927
1928 assert_wait((caddr_t)tl, (THREAD_INTERRUPTIBLE));
1929
1930 workqueue_unlock(p);
1931
1932 if (start_timer)
1933 workqueue_interval_timer_start(wq->wq_timer_call, wq_timer_interval_msecs);
1934
1935 KERNEL_DEBUG1(0xefffd018 | DBG_FUNC_START, (int)current_thread(), wq->wq_threads_scheduled, 0, 0, (int)th_to_park);
1936
1937 thread_block((thread_continue_t)thread_exception_return);
1938
1939 panic("unexpected return from thread_block");
1940
1941 out:
1942 workqueue_unlock(p);
1943
1944 if (start_timer)
1945 workqueue_interval_timer_start(wq->wq_timer_call, wq_timer_interval_msecs);
1946
1947 KERNEL_DEBUG(0xefffd000 | DBG_FUNC_END, (int)thread, 0, wq->wq_flags, 2, 0);
1948
1949 return;
1950 }
1951
1952 static void
1953 wq_runitem(proc_t p, user_addr_t item, thread_t th, struct threadlist *tl,
1954 int reuse_thread, int wake_thread, int return_directly)
1955 {
1956 int ret = 0;
1957
1958 KERNEL_DEBUG1(0xefffd004 | DBG_FUNC_START, (int)current_thread(), (int)item, wake_thread, tl->th_affinity_tag, (int)th);
1959
1960 ret = setup_wqthread(p, th, item, reuse_thread, tl);
1961
1962 if (ret != 0)
1963 panic("setup_wqthread failed %x\n", ret);
1964
1965 if (return_directly) {
1966 thread_exception_return();
1967
1968 panic("wq_runitem: thread_exception_return returned ...\n");
1969 }
1970 if (wake_thread) {
1971 KERNEL_DEBUG1(0xefffd018 | DBG_FUNC_END, (int)current_thread(), 0, 0, 0, (int)th);
1972
1973 wakeup(tl);
1974 } else {
1975 KERNEL_DEBUG1(0xefffd014 | DBG_FUNC_END, (int)current_thread(), 0, 0, 0, (int)th);
1976
1977 thread_resume(th);
1978 }
1979 }
1980
1981
1982 int
1983 setup_wqthread(proc_t p, thread_t th, user_addr_t item, int reuse_thread, struct threadlist *tl)
1984 {
1985
1986 #if defined(__ppc__)
1987 /*
1988 * Set up PowerPC registers...
1989 * internally they are always kept as 64 bit and
1990 * since the register set is the same between 32 and 64bit modes
1991 * we don't need 2 different methods for setting the state
1992 */
1993 {
1994 ppc_thread_state64_t state64;
1995 ppc_thread_state64_t *ts64 = &state64;
1996
1997 ts64->srr0 = (uint64_t)p->p_wqthread;
1998 ts64->r1 = (uint64_t)((tl->th_stackaddr + PTH_DEFAULT_STACKSIZE + PTH_DEFAULT_GUARDSIZE) - C_ARGSAVE_LEN - C_RED_ZONE);
1999 ts64->r3 = (uint64_t)(tl->th_stackaddr + PTH_DEFAULT_STACKSIZE + PTH_DEFAULT_GUARDSIZE);
2000 ts64->r4 = (uint64_t)((unsigned int)tl->th_thport);
2001 ts64->r5 = (uint64_t)(tl->th_stackaddr + PTH_DEFAULT_GUARDSIZE);
2002 ts64->r6 = (uint64_t)item;
2003 ts64->r7 = (uint64_t)reuse_thread;
2004 ts64->r8 = (uint64_t)0;
2005
2006 thread_set_wq_state64(th, (thread_state_t)ts64);
2007 }
2008 #elif defined(__i386__)
2009 int isLP64 = 0;
2010
2011 isLP64 = IS_64BIT_PROCESS(p);
2012 /*
2013 * Set up i386 registers & function call.
2014 */
2015 if (isLP64 == 0) {
2016 x86_thread_state32_t state;
2017 x86_thread_state32_t *ts = &state;
2018
2019 ts->eip = (int)p->p_wqthread;
2020 ts->eax = (unsigned int)(tl->th_stackaddr + PTH_DEFAULT_STACKSIZE + PTH_DEFAULT_GUARDSIZE);
2021 ts->ebx = (unsigned int)tl->th_thport;
2022 ts->ecx = (unsigned int)(tl->th_stackaddr + PTH_DEFAULT_GUARDSIZE);
2023 ts->edx = (unsigned int)item;
2024 ts->edi = (unsigned int)reuse_thread;
2025 ts->esi = (unsigned int)0;
2026 /*
2027 * set stack pointer
2028 */
2029 ts->esp = (int)((vm_offset_t)((tl->th_stackaddr + PTH_DEFAULT_STACKSIZE + PTH_DEFAULT_GUARDSIZE) - C_32_STK_ALIGN));
2030
2031 thread_set_wq_state32(th, (thread_state_t)ts);
2032
2033 } else {
2034 x86_thread_state64_t state64;
2035 x86_thread_state64_t *ts64 = &state64;
2036
2037 ts64->rip = (uint64_t)p->p_wqthread;
2038 ts64->rdi = (uint64_t)(tl->th_stackaddr + PTH_DEFAULT_STACKSIZE + PTH_DEFAULT_GUARDSIZE);
2039 ts64->rsi = (uint64_t)((unsigned int)(tl->th_thport));
2040 ts64->rdx = (uint64_t)(tl->th_stackaddr + PTH_DEFAULT_GUARDSIZE);
2041 ts64->rcx = (uint64_t)item;
2042 ts64->r8 = (uint64_t)reuse_thread;
2043 ts64->r9 = (uint64_t)0;
2044
2045 /*
2046 * set stack pointer aligned to 16 byte boundary
2047 */
2048 ts64->rsp = (uint64_t)((tl->th_stackaddr + PTH_DEFAULT_STACKSIZE + PTH_DEFAULT_GUARDSIZE) - C_64_REDZONE_LEN);
2049
2050 thread_set_wq_state64(th, (thread_state_t)ts64);
2051 }
2052 #else
2053 #error setup_wqthread not defined for this architecture
2054 #endif
2055 return(0);
2056 }
2057
mirror of XNU