2 * Copyright (c) 1994 by Xerox Corporation. All rights reserved.
3 * Copyright (c) 1996 by Silicon Graphics. All rights reserved.
4 * Copyright (c) 1998 by Fergus Henderson. All rights reserved.
5 * Copyright (c) 2000-2004 by Hewlett-Packard Company. All rights reserved.
7 * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
8 * OR IMPLIED. ANY USE IS AT YOUR OWN RISK.
10 * Permission is hereby granted to use or copy this program
11 * for any purpose, provided the above notices are retained on all copies.
12 * Permission to modify the code and to distribute modified code is granted,
13 * provided the above notices are retained, and a notice that the code was
14 * modified is included with the above copyright notice.
17 * Support code for LinuxThreads, the clone()-based kernel
18 * thread package for Linux which is included in libc6.
20 * This code relies on implementation details of LinuxThreads,
21 * (i.e. properties not guaranteed by the Pthread standard),
22 * though this version now does less of that than the other Pthreads
25 * Note that there is a lot of code duplication between linux_threads.c
26 * and thread support for some of the other Posix platforms; any changes
27 * made here may need to be reflected there too.
29 /* DG/UX ix86 support <takis@xfree86.org> */
31 * Linux_threads.c now also includes some code to support HPUX and
32 * OSF1 (Compaq Tru64 Unix, really). The OSF1 support is based on Eric Benson's
35 * Eric also suggested an alternate basis for a lock implementation in
37 * + #elif defined(OSF1)
38 * + unsigned long GC_allocate_lock = 0;
39 * + msemaphore GC_allocate_semaphore;
40 * + # define GC_TRY_LOCK() \
41 * + ((msem_lock(&GC_allocate_semaphore, MSEM_IF_NOWAIT) == 0) \
42 * + ? (GC_allocate_lock = 1) \
44 * + # define GC_LOCK_TAKEN GC_allocate_lock
47 /*#define DEBUG_THREADS 1*/
48 /*#define GC_ASSERTIONS*/
50 # include "private/pthread_support.h"
52 # if defined(GC_PTHREADS) && !defined(GC_SOLARIS_THREADS) \
53 && !defined(GC_WIN32_THREADS)
55 # if defined(GC_HPUX_THREADS) && !defined(USE_PTHREAD_SPECIFIC) \
56 && !defined(USE_COMPILER_TLS)
58 # define USE_PTHREAD_SPECIFIC
59 /* Empirically, as of gcc 3.3, USE_COMPILER_TLS doesn't work. */
61 # define USE_COMPILER_TLS
65 # if defined USE_HPUX_TLS
66 --> Macro replaced by USE_COMPILER_TLS
69 # if (defined(GC_DGUX386_THREADS) || defined(GC_OSF1_THREADS) || \
70 defined(GC_DARWIN_THREADS) || defined(GC_AIX_THREADS)) \
71 && !defined(USE_PTHREAD_SPECIFIC)
72 # define USE_PTHREAD_SPECIFIC
75 # if defined(GC_DGUX386_THREADS) && !defined(_POSIX4A_DRAFT10_SOURCE)
76 # define _POSIX4A_DRAFT10_SOURCE 1
79 # if defined(GC_DGUX386_THREADS) && !defined(_USING_POSIX4A_DRAFT10)
80 # define _USING_POSIX4A_DRAFT10 1
83 # ifdef THREAD_LOCAL_ALLOC
84 # if !defined(USE_PTHREAD_SPECIFIC) && !defined(USE_COMPILER_TLS)
85 # include "private/specific.h"
88 /* Note that these macros should be used only to get/set the GC_thread pointer.
89 * We need to use both tls and pthread because we use the pthread_create function hook to
90 * free the data for foreign threads. When that doesn't happen, libgc could have old
91 * pthread_t that get reused...
93 # if defined(USE_PTHREAD_SPECIFIC)
94 # define GC_getspecific pthread_getspecific
95 # define GC_setspecific pthread_setspecific
96 # define GC_key_create pthread_key_create
97 typedef pthread_key_t GC_key_t;
99 # if defined(USE_COMPILER_TLS)
100 # define GC_getspecific(x) (GC_thread_tls)
101 # define GC_setspecific(key, v) (GC_thread_tls = (v), pthread_setspecific ((key), (v)))
102 # define GC_key_create pthread_key_create
103 typedef void * GC_key_t;
107 # include <pthread.h>
112 # include <sys/mman.h>
113 # include <sys/time.h>
114 # include <sys/types.h>
115 # include <sys/stat.h>
119 #if defined(GC_DARWIN_THREADS)
120 # include "private/darwin_semaphore.h"
122 # include <semaphore.h>
123 #endif /* !GC_DARWIN_THREADS */
125 #if defined(GC_DARWIN_THREADS) || defined(GC_FREEBSD_THREADS)
126 # include <sys/sysctl.h>
127 #endif /* GC_DARWIN_THREADS */
131 #if defined(GC_DGUX386_THREADS)
132 # include <sys/dg_sys_info.h>
133 # include <sys/_int_psem.h>
134 /* sem_t is an uint in DG/UX */
135 typedef unsigned int sem_t;
136 #endif /* GC_DGUX386_THREADS */
142 #ifdef GC_USE_LD_WRAP
143 # define WRAP_FUNC(f) __wrap_##f
144 # define REAL_FUNC(f) __real_##f
146 # define WRAP_FUNC(f) GC_##f
147 # if !defined(GC_DGUX386_THREADS)
148 # define REAL_FUNC(f) f
149 # else /* GC_DGUX386_THREADS */
150 # define REAL_FUNC(f) __d10_##f
151 # endif /* GC_DGUX386_THREADS */
152 # undef pthread_create
153 # if !defined(GC_DARWIN_THREADS)
154 # undef pthread_sigmask
157 # undef pthread_detach
158 # if defined(GC_OSF1_THREADS) && defined(_PTHREAD_USE_MANGLED_NAMES_) \
159 && !defined(_PTHREAD_USE_PTDNAM_)
160 /* Restore the original mangled names on Tru64 UNIX. */
161 # define pthread_create __pthread_create
162 # define pthread_join __pthread_join
163 # define pthread_detach __pthread_detach
169 static GC_bool parallel_initialized = FALSE;
171 void GC_init_parallel();
173 # if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
175 /* We don't really support thread-local allocation with DBG_HDRS_ALL */
177 /* work around a dlopen issue (bug #75390), undefs to avoid warnings with redefinitions */
178 #undef PACKAGE_BUGREPORT
180 #undef PACKAGE_STRING
181 #undef PACKAGE_TARNAME
182 #undef PACKAGE_VERSION
183 #include "mono/utils/mono-compiler.h"
186 GC_key_t GC_thread_key;
188 #ifdef USE_COMPILER_TLS
189 static __thread MONO_TLS_FAST void* GC_thread_tls;
192 static GC_bool keys_initialized;
194 /* Recover the contents of the freelist array fl into the global one gfl.*/
195 /* Note that the indexing scheme differs, in that gfl has finer size */
196 /* resolution, even if not all entries are used. */
197 /* We hold the allocator lock. */
198 static void return_freelists(ptr_t *fl, ptr_t *gfl)
204 for (i = 1; i < NFREELISTS; ++i) {
205 nwords = i * (GRANULARITY/sizeof(word));
208 if ((word)q >= HBLKSIZE) {
209 if (gfl[nwords] == 0) {
213 for (; (word)q >= HBLKSIZE; qptr = &(obj_link(q)), q = *qptr);
219 /* Clear fl[i], since the thread structure may hang around. */
220 /* Do it in a way that is likely to trap if we access it. */
221 fl[i] = (ptr_t)HBLKSIZE;
225 /* We statically allocate a single "size 0" object. It is linked to */
226 /* itself, and is thus repeatedly reused for all size 0 allocation */
227 /* requests. (Size 0 gcj allocation requests are incorrect, and */
228 /* we arrange for those to fault asap.) */
229 static ptr_t size_zero_object = (ptr_t)(&size_zero_object);
231 void GC_delete_gc_thread(pthread_t id, GC_thread gct);
233 void GC_thread_deregister_foreign (void *data)
235 GC_thread me = (GC_thread)data;
236 /* GC_fprintf1( "\n\n\n\n --- Deregister %x ---\n\n\n\n\n", me->flags ); */
237 if (me -> flags & FOREIGN_THREAD) {
239 /* GC_fprintf0( "\n\n\n\n --- FOO ---\n\n\n\n\n" ); */
240 #if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
241 GC_destroy_thread_local (me);
243 GC_delete_gc_thread(me->id, me);
248 /* Each thread structure must be initialized. */
249 /* This call must be made from the new thread. */
250 /* Caller holds allocation lock. */
251 void GC_init_thread_local(GC_thread p)
255 if (!keys_initialized) {
256 if (0 != GC_key_create(&GC_thread_key, GC_thread_deregister_foreign)) {
257 ABORT("Failed to create key for local allocator");
259 keys_initialized = TRUE;
261 if (0 != GC_setspecific(GC_thread_key, p)) {
262 ABORT("Failed to set thread specific allocation pointers");
264 for (i = 1; i < NFREELISTS; ++i) {
265 p -> ptrfree_freelists[i] = (ptr_t)1;
266 p -> normal_freelists[i] = (ptr_t)1;
267 # ifdef GC_GCJ_SUPPORT
268 p -> gcj_freelists[i] = (ptr_t)1;
271 /* Set up the size 0 free lists. */
272 p -> ptrfree_freelists[0] = (ptr_t)(&size_zero_object);
273 p -> normal_freelists[0] = (ptr_t)(&size_zero_object);
274 # ifdef GC_GCJ_SUPPORT
275 p -> gcj_freelists[0] = (ptr_t)(-1);
279 #ifdef GC_GCJ_SUPPORT
280 extern ptr_t * GC_gcjobjfreelist;
283 /* We hold the allocator lock. */
284 void GC_destroy_thread_local(GC_thread p)
286 /* We currently only do this from the thread itself or from */
287 /* the fork handler for a child process. */
289 GC_ASSERT(GC_getspecific(GC_thread_key) == (void *)p);
291 return_freelists(p -> ptrfree_freelists, GC_aobjfreelist);
292 return_freelists(p -> normal_freelists, GC_objfreelist);
293 # ifdef GC_GCJ_SUPPORT
294 return_freelists(p -> gcj_freelists, GC_gcjobjfreelist);
298 extern GC_PTR GC_generic_malloc_many();
300 GC_PTR GC_local_malloc(size_t bytes)
302 if (EXPECT(!SMALL_ENOUGH(bytes),0)) {
303 return(GC_malloc(bytes));
305 int index = INDEX_FROM_BYTES(bytes);
308 # if defined(REDIRECT_MALLOC) && !defined(USE_PTHREAD_SPECIFIC)
309 GC_key_t k = GC_thread_key;
313 # if defined(REDIRECT_MALLOC) && !defined(USE_PTHREAD_SPECIFIC)
314 if (EXPECT(0 == k, 0)) {
315 /* This can happen if we get called when the world is */
316 /* being initialized. Whether we can actually complete */
317 /* the initialization then is unclear. */
322 tsd = GC_getspecific(GC_thread_key);
323 # ifdef GC_ASSERTIONS
325 GC_ASSERT(tsd == (void *)GC_lookup_thread(pthread_self()));
328 my_fl = ((GC_thread)tsd) -> normal_freelists + index;
330 if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
331 ptr_t next = obj_link(my_entry);
332 GC_PTR result = (GC_PTR)my_entry;
334 obj_link(my_entry) = 0;
335 PREFETCH_FOR_WRITE(next);
337 } else if ((word)my_entry - 1 < DIRECT_GRANULES) {
338 *my_fl = my_entry + index + 1;
339 return GC_malloc(bytes);
341 GC_generic_malloc_many(BYTES_FROM_INDEX(index), NORMAL, my_fl);
342 if (*my_fl == 0) return GC_oom_fn(bytes);
343 return GC_local_malloc(bytes);
348 GC_PTR GC_local_malloc_atomic(size_t bytes)
350 if (EXPECT(!SMALL_ENOUGH(bytes), 0)) {
351 return(GC_malloc_atomic(bytes));
353 int index = INDEX_FROM_BYTES(bytes);
354 ptr_t * my_fl = ((GC_thread)GC_getspecific(GC_thread_key))
355 -> ptrfree_freelists + index;
356 ptr_t my_entry = *my_fl;
358 if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
359 GC_PTR result = (GC_PTR)my_entry;
360 *my_fl = obj_link(my_entry);
362 } else if ((word)my_entry - 1 < DIRECT_GRANULES) {
363 *my_fl = my_entry + index + 1;
364 return GC_malloc_atomic(bytes);
366 GC_generic_malloc_many(BYTES_FROM_INDEX(index), PTRFREE, my_fl);
367 /* *my_fl is updated while the collector is excluded; */
368 /* the free list is always visible to the collector as */
370 if (*my_fl == 0) return GC_oom_fn(bytes);
371 return GC_local_malloc_atomic(bytes);
376 #ifdef GC_GCJ_SUPPORT
378 #include "include/gc_gcj.h"
381 extern GC_bool GC_gcj_malloc_initialized;
384 extern int GC_gcj_kind;
386 GC_PTR GC_local_gcj_malloc(size_t bytes,
387 void * ptr_to_struct_containing_descr)
389 GC_ASSERT(GC_gcj_malloc_initialized);
390 if (EXPECT(!SMALL_ENOUGH(bytes), 0)) {
391 return GC_gcj_malloc(bytes, ptr_to_struct_containing_descr);
393 int index = INDEX_FROM_BYTES(bytes);
394 ptr_t * my_fl = ((GC_thread)GC_getspecific(GC_thread_key))
395 -> gcj_freelists + index;
396 ptr_t my_entry = *my_fl;
397 if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
398 GC_PTR result = (GC_PTR)my_entry;
399 GC_ASSERT(!GC_incremental);
400 /* We assert that any concurrent marker will stop us. */
401 /* Thus it is impossible for a mark procedure to see the */
402 /* allocation of the next object, but to see this object */
403 /* still containing a free list pointer. Otherwise the */
404 /* marker might find a random "mark descriptor". */
405 *(volatile ptr_t *)my_fl = obj_link(my_entry);
406 /* We must update the freelist before we store the pointer. */
407 /* Otherwise a GC at this point would see a corrupted */
409 /* A memory barrier is probably never needed, since the */
410 /* action of stopping this thread will cause prior writes */
412 GC_ASSERT(((void * volatile *)result)[1] == 0);
413 *(void * volatile *)result = ptr_to_struct_containing_descr;
415 } else if ((word)my_entry - 1 < DIRECT_GRANULES) {
416 if (!GC_incremental) *my_fl = my_entry + index + 1;
417 /* In the incremental case, we always have to take this */
418 /* path. Thus we leave the counter alone. */
419 return GC_gcj_malloc(bytes, ptr_to_struct_containing_descr);
421 GC_generic_malloc_many(BYTES_FROM_INDEX(index), GC_gcj_kind, my_fl);
422 if (*my_fl == 0) return GC_oom_fn(bytes);
423 return GC_local_gcj_malloc(bytes, ptr_to_struct_containing_descr);
428 /* Similar to GC_local_gcj_malloc, but the size is in words, and we don't */
429 /* adjust it. The size is assumed to be such that it can be */
430 /* allocated as a small object. */
431 void * GC_local_gcj_fast_malloc(size_t lw, void * ptr_to_struct_containing_descr)
433 ptr_t * my_fl = ((GC_thread)GC_getspecific(GC_thread_key))
434 -> gcj_freelists + lw;
435 ptr_t my_entry = *my_fl;
437 GC_ASSERT(GC_gcj_malloc_initialized);
439 if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
440 GC_PTR result = (GC_PTR)my_entry;
441 GC_ASSERT(!GC_incremental);
442 /* We assert that any concurrent marker will stop us. */
443 /* Thus it is impossible for a mark procedure to see the */
444 /* allocation of the next object, but to see this object */
445 /* still containing a free list pointer. Otherwise the */
446 /* marker might find a random "mark descriptor". */
447 *(volatile ptr_t *)my_fl = obj_link(my_entry);
448 /* We must update the freelist before we store the pointer. */
449 /* Otherwise a GC at this point would see a corrupted */
451 /* A memory barrier is probably never needed, since the */
452 /* action of stopping this thread will cause prior writes */
454 GC_ASSERT(((void * volatile *)result)[1] == 0);
455 *(void * volatile *)result = ptr_to_struct_containing_descr;
457 } else if ((word)my_entry - 1 < DIRECT_GRANULES) {
458 if (!GC_incremental) *my_fl = my_entry + lw + 1;
459 /* In the incremental case, we always have to take this */
460 /* path. Thus we leave the counter alone. */
461 return GC_gcj_fast_malloc(lw, ptr_to_struct_containing_descr);
463 GC_generic_malloc_many(BYTES_FROM_INDEX(lw), GC_gcj_kind, my_fl);
464 if (*my_fl == 0) return GC_oom_fn(BYTES_FROM_INDEX(lw));
465 return GC_local_gcj_fast_malloc(lw, ptr_to_struct_containing_descr);
469 #endif /* GC_GCJ_SUPPORT */
471 # else /* !THREAD_LOCAL_ALLOC && !DBG_HDRS_ALL */
473 # define GC_destroy_thread_local(t)
475 # endif /* !THREAD_LOCAL_ALLOC */
479 To make sure that we're using LinuxThreads and not some other thread
480 package, we generate a dummy reference to `pthread_kill_other_threads_np'
481 (was `__pthread_initial_thread_bos' but that disappeared),
482 which is a symbol defined in LinuxThreads, but (hopefully) not in other
485 We no longer do this, since this code is now portable enough that it might
486 actually work for something else.
488 void (*dummy_var_to_force_linux_threads)() = pthread_kill_other_threads_np;
491 long GC_nprocs = 1; /* Number of processors. We may not have */
492 /* access to all of them, but this is as good */
493 /* a guess as any ... */
498 # define MAX_MARKERS 16
501 static ptr_t marker_sp[MAX_MARKERS] = {0};
503 void * GC_mark_thread(void * id)
507 marker_sp[(word)id] = GC_approx_sp();
508 for (;; ++my_mark_no) {
509 /* GC_mark_no is passed only to allow GC_help_marker to terminate */
510 /* promptly. This is important if it were called from the signal */
511 /* handler or from the GC lock acquisition code. Under Linux, it's */
512 /* not safe to call it from a signal handler, since it uses mutexes */
513 /* and condition variables. Since it is called only here, the */
514 /* argument is unnecessary. */
515 if (my_mark_no < GC_mark_no || my_mark_no > GC_mark_no + 2) {
516 /* resynchronize if we get far off, e.g. because GC_mark_no */
518 my_mark_no = GC_mark_no;
520 # ifdef DEBUG_THREADS
521 GC_printf1("Starting mark helper for mark number %ld\n", my_mark_no);
523 GC_help_marker(my_mark_no);
527 extern long GC_markers; /* Number of mark threads we would */
528 /* like to have. Includes the */
529 /* initiating thread. */
531 pthread_t GC_mark_threads[MAX_MARKERS];
533 #define PTHREAD_CREATE REAL_FUNC(pthread_create)
535 static void start_mark_threads()
540 if (GC_markers > MAX_MARKERS) {
541 WARN("Limiting number of mark threads\n", 0);
542 GC_markers = MAX_MARKERS;
544 if (0 != pthread_attr_init(&attr)) ABORT("pthread_attr_init failed");
546 if (0 != pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED))
547 ABORT("pthread_attr_setdetachstate failed");
549 # if defined(HPUX) || defined(GC_DGUX386_THREADS)
550 /* Default stack size is usually too small: fix it. */
551 /* Otherwise marker threads or GC may run out of */
553 # define MIN_STACK_SIZE (8*HBLKSIZE*sizeof(word))
558 if (pthread_attr_getstacksize(&attr, &old_size) != 0)
559 ABORT("pthread_attr_getstacksize failed\n");
560 if (old_size < MIN_STACK_SIZE) {
561 if (pthread_attr_setstacksize(&attr, MIN_STACK_SIZE) != 0)
562 ABORT("pthread_attr_setstacksize failed\n");
565 # endif /* HPUX || GC_DGUX386_THREADS */
567 if (GC_print_stats) {
568 GC_printf1("Starting %ld marker threads\n", GC_markers - 1);
571 for (i = 0; i < GC_markers - 1; ++i) {
572 if (0 != PTHREAD_CREATE(GC_mark_threads + i, &attr,
573 GC_mark_thread, (void *)(word)i)) {
574 WARN("Marker thread creation failed, errno = %ld.\n", errno);
579 #else /* !PARALLEL_MARK */
581 static __inline__ void start_mark_threads()
585 #endif /* !PARALLEL_MARK */
587 GC_bool GC_thr_initialized = FALSE;
589 volatile GC_thread GC_threads[THREAD_TABLE_SZ];
592 * gcc-3.3.6 miscompiles the &GC_thread_key+sizeof(&GC_thread_key) expression so
593 * put it into a separate function.
595 # if defined(__GNUC__) && defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
596 static __attribute__((noinline)) unsigned char* get_gc_thread_key_addr GC_PROTO((void))
598 return (unsigned char*)&GC_thread_key;
601 void GC_push_thread_structures GC_PROTO((void))
603 GC_push_all((ptr_t)(GC_threads), (ptr_t)(GC_threads)+sizeof(GC_threads));
604 # if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
605 GC_push_all((ptr_t)get_gc_thread_key_addr(),
606 (ptr_t)(get_gc_thread_key_addr())+sizeof(&GC_thread_key));
612 void GC_push_thread_structures GC_PROTO((void))
614 GC_push_all((ptr_t)(GC_threads), (ptr_t)(GC_threads)+sizeof(GC_threads));
615 # if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
616 GC_push_all((ptr_t)(&GC_thread_key),
617 (ptr_t)(&GC_thread_key)+sizeof(&GC_thread_key));
623 #ifdef THREAD_LOCAL_ALLOC
624 /* We must explicitly mark ptrfree and gcj free lists, since the free */
625 /* list links wouldn't otherwise be found. We also set them in the */
626 /* normal free lists, since that involves touching less memory than if */
627 /* we scanned them normally. */
628 void GC_mark_thread_local_free_lists(void)
634 for (i = 0; i < THREAD_TABLE_SZ; ++i) {
635 for (p = GC_threads[i]; 0 != p; p = p -> next) {
636 for (j = 1; j < NFREELISTS; ++j) {
637 q = p -> ptrfree_freelists[j];
638 if ((word)q > HBLKSIZE) GC_set_fl_marks(q);
639 q = p -> normal_freelists[j];
640 if ((word)q > HBLKSIZE) GC_set_fl_marks(q);
641 # ifdef GC_GCJ_SUPPORT
642 q = p -> gcj_freelists[j];
643 if ((word)q > HBLKSIZE) GC_set_fl_marks(q);
644 # endif /* GC_GCJ_SUPPORT */
649 #endif /* THREAD_LOCAL_ALLOC */
651 static struct GC_Thread_Rep first_thread;
653 /* Add a thread to GC_threads. We assume it wasn't already there. */
654 /* Caller holds allocation lock. */
655 GC_thread GC_new_thread(pthread_t id)
657 int hv = ((word)id) % THREAD_TABLE_SZ;
659 static GC_bool first_thread_used = FALSE;
661 if (!first_thread_used) {
662 result = &first_thread;
663 first_thread_used = TRUE;
665 result = (struct GC_Thread_Rep *)
666 GC_INTERNAL_MALLOC(sizeof(struct GC_Thread_Rep), NORMAL);
668 if (result == 0) return(0);
670 result -> next = GC_threads[hv];
671 GC_threads[hv] = result;
672 GC_ASSERT(result -> flags == 0 && result -> thread_blocked == 0);
676 /* Delete a thread from GC_threads. We assume it is there. */
677 /* (The code intentionally traps if it wasn't.) */
678 /* Caller holds allocation lock. */
679 void GC_delete_thread(pthread_t id)
681 int hv = ((word)id) % THREAD_TABLE_SZ;
682 register GC_thread p = GC_threads[hv];
683 register GC_thread prev = 0;
685 while (!pthread_equal(p -> id, id)) {
690 GC_threads[hv] = p -> next;
692 prev -> next = p -> next;
697 /* If a thread has been joined, but we have not yet */
698 /* been notified, then there may be more than one thread */
699 /* in the table with the same pthread id. */
700 /* This is OK, but we need a way to delete a specific one. */
701 void GC_delete_gc_thread(pthread_t id, GC_thread gc_id)
703 int hv = ((word)id) % THREAD_TABLE_SZ;
704 register GC_thread p = GC_threads[hv];
705 register GC_thread prev = 0;
712 GC_threads[hv] = p -> next;
714 prev -> next = p -> next;
719 /* Return a GC_thread corresponding to a given pthread_t. */
720 /* Returns 0 if it's not there. */
721 /* Caller holds allocation lock or otherwise inhibits */
723 /* If there is more than one thread with the given id we */
724 /* return the most recent one. */
725 GC_thread GC_lookup_thread(pthread_t id)
727 int hv = ((word)id) % THREAD_TABLE_SZ;
728 register GC_thread p = GC_threads[hv];
730 while (p != 0 && !pthread_equal(p -> id, id)) p = p -> next;
734 int GC_thread_is_registered (void)
739 ptr = (void *)GC_lookup_thread(pthread_self());
746 /* Remove all entries from the GC_threads table, except the */
747 /* one for the current thread. We need to do this in the child */
748 /* process after a fork(), since only the current thread */
749 /* survives in the child. */
750 void GC_remove_all_threads_but_me(void)
752 pthread_t self = pthread_self();
754 GC_thread p, next, me;
756 for (hv = 0; hv < THREAD_TABLE_SZ; ++hv) {
758 for (p = GC_threads[hv]; 0 != p; p = next) {
760 if (p -> id == self) {
764 # ifdef THREAD_LOCAL_ALLOC
765 if (!(p -> flags & FINISHED)) {
766 GC_destroy_thread_local(p);
768 # endif /* THREAD_LOCAL_ALLOC */
769 if (p != &first_thread) GC_INTERNAL_FREE(p);
775 #endif /* HANDLE_FORK */
777 #ifdef USE_PROC_FOR_LIBRARIES
778 int GC_segment_is_thread_stack(ptr_t lo, ptr_t hi)
783 # ifdef PARALLEL_MARK
784 for (i = 0; i < GC_markers; ++i) {
785 if (marker_sp[i] > lo & marker_sp[i] < hi) return 1;
788 for (i = 0; i < THREAD_TABLE_SZ; i++) {
789 for (p = GC_threads[i]; p != 0; p = p -> next) {
790 if (0 != p -> stack_end) {
791 # ifdef STACK_GROWS_UP
792 if (p -> stack_end >= lo && p -> stack_end < hi) return 1;
793 # else /* STACK_GROWS_DOWN */
794 if (p -> stack_end > lo && p -> stack_end <= hi) return 1;
801 #endif /* USE_PROC_FOR_LIBRARIES */
803 #ifdef GC_LINUX_THREADS
804 /* Return the number of processors, or i<= 0 if it can't be determined. */
807 /* Should be "return sysconf(_SC_NPROCESSORS_ONLN);" but that */
808 /* appears to be buggy in many cases. */
809 /* We look for lines "cpu<n>" in /proc/stat. */
810 # define STAT_BUF_SIZE 4096
811 # define STAT_READ read
812 /* If read is wrapped, this may need to be redefined to call */
814 char stat_buf[STAT_BUF_SIZE];
817 /* Some old kernels only have a single "cpu nnnn ..." */
818 /* entry in /proc/stat. We identify those as */
822 f = open("/proc/stat", O_RDONLY);
823 if (f < 0 || (len = STAT_READ(f, stat_buf, STAT_BUF_SIZE)) < 100) {
824 WARN("Couldn't read /proc/stat\n", 0);
827 for (i = 0; i < len - 100; ++i) {
828 if (stat_buf[i] == '\n' && stat_buf[i+1] == 'c'
829 && stat_buf[i+2] == 'p' && stat_buf[i+3] == 'u') {
830 int cpu_no = atoi(stat_buf + i + 4);
831 if (cpu_no >= result) result = cpu_no + 1;
837 #endif /* GC_LINUX_THREADS */
839 /* We hold the GC lock. Wait until an in-progress GC has finished. */
840 /* Repeatedly RELEASES GC LOCK in order to wait. */
841 /* If wait_for_all is true, then we exit with the GC lock held and no */
842 /* collection in progress; otherwise we just wait for the current GC */
844 extern GC_bool GC_collection_in_progress();
845 void GC_wait_for_gc_completion(GC_bool wait_for_all)
847 if (GC_incremental && GC_collection_in_progress()) {
848 int old_gc_no = GC_gc_no;
850 /* Make sure that no part of our stack is still on the mark stack, */
851 /* since it's about to be unmapped. */
852 while (GC_incremental && GC_collection_in_progress()
853 && (wait_for_all || old_gc_no == GC_gc_no)) {
855 GC_in_thread_creation = TRUE;
856 GC_collect_a_little_inner(1);
857 GC_in_thread_creation = FALSE;
867 /* Procedures called before and after a fork. The goal here is to make */
868 /* it safe to call GC_malloc() in a forked child. It's unclear that is */
869 /* attainable, since the single UNIX spec seems to imply that one */
870 /* should only call async-signal-safe functions, and we probably can't */
871 /* quite guarantee that. But we give it our best shot. (That same */
872 /* spec also implies that it's not safe to call the system malloc */
873 /* between fork() and exec(). Thus we're doing no worse than it. */
875 /* Called before a fork() */
876 void GC_fork_prepare_proc(void)
878 /* Acquire all relevant locks, so that after releasing the locks */
879 /* the child will see a consistent state in which monitor */
880 /* invariants hold. Unfortunately, we can't acquire libc locks */
881 /* we might need, and there seems to be no guarantee that libc */
882 /* must install a suitable fork handler. */
883 /* Wait for an ongoing GC to finish, since we can't finish it in */
884 /* the (one remaining thread in) the child. */
886 # if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
887 GC_wait_for_reclaim();
889 GC_wait_for_gc_completion(TRUE);
890 # if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
891 GC_acquire_mark_lock();
895 /* Called in parent after a fork() */
896 void GC_fork_parent_proc(void)
898 # if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
899 GC_release_mark_lock();
904 /* Called in child after a fork() */
905 void GC_fork_child_proc(void)
907 /* Clean up the thread table, so that just our thread is left. */
908 # if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
909 GC_release_mark_lock();
911 GC_remove_all_threads_but_me();
912 # ifdef PARALLEL_MARK
913 /* Turn off parallel marking in the child, since we are probably */
914 /* just going to exec, and we would have to restart mark threads. */
917 # endif /* PARALLEL_MARK */
920 #endif /* HANDLE_FORK */
922 #if defined(GC_DGUX386_THREADS)
923 /* Return the number of processors, or i<= 0 if it can't be determined. */
926 /* <takis@XFree86.Org> */
928 struct dg_sys_info_pm_info pm_sysinfo;
931 status = dg_sys_info((long int *) &pm_sysinfo,
932 DG_SYS_INFO_PM_INFO_TYPE, DG_SYS_INFO_PM_CURRENT_VERSION);
934 /* set -1 for error */
938 numCpus = pm_sysinfo.idle_vp_count;
940 # ifdef DEBUG_THREADS
941 GC_printf1("Number of active CPUs in this system: %d\n", numCpus);
945 #endif /* GC_DGUX386_THREADS */
947 /* We hold the allocation lock. */
950 # ifndef GC_DARWIN_THREADS
955 if (GC_thr_initialized) return;
956 GC_thr_initialized = TRUE;
959 /* Prepare for a possible fork. */
960 pthread_atfork(GC_fork_prepare_proc, GC_fork_parent_proc,
962 # endif /* HANDLE_FORK */
963 /* Add the initial thread, so we can stop it. */
964 t = GC_new_thread(pthread_self());
965 # ifdef GC_DARWIN_THREADS
966 t -> stop_info.mach_thread = mach_thread_self();
968 t -> stop_info.stack_ptr = (ptr_t)(&dummy);
970 t -> flags = DETACHED | MAIN_THREAD;
976 char * nprocs_string = GETENV("GC_NPROCS");
978 if (nprocs_string != NULL) GC_nprocs = atoi(nprocs_string);
980 if (GC_nprocs <= 0) {
981 # if defined(GC_HPUX_THREADS)
982 GC_nprocs = pthread_num_processors_np();
984 # if defined(GC_OSF1_THREADS) || defined(GC_AIX_THREADS)
985 GC_nprocs = sysconf(_SC_NPROCESSORS_ONLN);
986 if (GC_nprocs <= 0) GC_nprocs = 1;
988 # if defined(GC_IRIX_THREADS)
989 GC_nprocs = sysconf(_SC_NPROC_ONLN);
990 if (GC_nprocs <= 0) GC_nprocs = 1;
992 # if defined(GC_DARWIN_THREADS) || defined(GC_FREEBSD_THREADS)
994 size_t len = sizeof(ncpus);
995 sysctl((int[2]) {CTL_HW, HW_NCPU}, 2, &ncpus, &len, NULL, 0);
998 # if defined(GC_LINUX_THREADS) || defined(GC_DGUX386_THREADS)
999 GC_nprocs = GC_get_nprocs();
1002 if (GC_nprocs <= 0) {
1003 WARN("GC_get_nprocs() returned %ld\n", GC_nprocs);
1005 # ifdef PARALLEL_MARK
1009 # ifdef PARALLEL_MARK
1011 char * markers_string = GETENV("GC_MARKERS");
1012 if (markers_string != NULL) {
1013 GC_markers = atoi(markers_string);
1015 GC_markers = GC_nprocs;
1020 # ifdef PARALLEL_MARK
1022 if (GC_print_stats) {
1023 GC_printf2("Number of processors = %ld, "
1024 "number of marker threads = %ld\n", GC_nprocs, GC_markers);
1027 if (GC_markers == 1) {
1028 GC_parallel = FALSE;
1030 if (GC_print_stats) {
1031 GC_printf0("Single marker thread, turning off parallel marking\n");
1036 /* Disable true incremental collection, but generational is OK. */
1037 GC_time_limit = GC_TIME_UNLIMITED;
1039 /* If we are using a parallel marker, actually start helper threads. */
1040 if (GC_parallel) start_mark_threads();
1045 /* Perform all initializations, including those that */
1046 /* may require allocation. */
1047 /* Called without allocation lock. */
1048 /* Must be called before a second thread is created. */
1049 /* Called without allocation lock. */
1050 void GC_init_parallel()
1052 if (parallel_initialized) return;
1053 parallel_initialized = TRUE;
1055 /* GC_init() calls us back, so set flag first. */
1056 if (!GC_is_initialized) GC_init();
1057 /* Initialize thread local free lists if used. */
1058 # if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
1060 GC_init_thread_local(GC_lookup_thread(pthread_self()));
1066 #if !defined(GC_DARWIN_THREADS)
1067 int WRAP_FUNC(pthread_sigmask)(int how, const sigset_t *set, sigset_t *oset)
1069 sigset_t fudged_set;
1071 if (set != NULL && (how == SIG_BLOCK || how == SIG_SETMASK)) {
1073 sigdelset(&fudged_set, SIG_SUSPEND);
1076 return(REAL_FUNC(pthread_sigmask)(how, set, oset));
1078 #endif /* !GC_DARWIN_THREADS */
1080 /* Wrappers for functions that are likely to block for an appreciable */
1081 /* length of time. Must be called in pairs, if at all. */
1082 /* Nothing much beyond the system call itself should be executed */
1083 /* between these. */
1085 void GC_start_blocking(void) {
1086 # define SP_SLOP 128
1089 me = GC_lookup_thread(pthread_self());
1090 GC_ASSERT(!(me -> thread_blocked));
1092 me -> stop_info.stack_ptr = (ptr_t)GC_save_regs_in_stack();
1094 # ifndef GC_DARWIN_THREADS
1095 me -> stop_info.stack_ptr = (ptr_t)GC_approx_sp();
1099 me -> backing_store_ptr = (ptr_t)GC_save_regs_in_stack() + SP_SLOP;
1101 /* Add some slop to the stack pointer, since the wrapped call may */
1102 /* end up pushing more callee-save registers. */
1103 # ifndef GC_DARWIN_THREADS
1104 # ifdef STACK_GROWS_UP
1105 me -> stop_info.stack_ptr += SP_SLOP;
1107 me -> stop_info.stack_ptr -= SP_SLOP;
1110 me -> thread_blocked = TRUE;
1114 void GC_end_blocking(void) {
1116 LOCK(); /* This will block if the world is stopped. */
1117 me = GC_lookup_thread(pthread_self());
1118 GC_ASSERT(me -> thread_blocked);
1119 me -> thread_blocked = FALSE;
1123 #if defined(GC_DGUX386_THREADS)
1124 #define __d10_sleep sleep
1125 #endif /* GC_DGUX386_THREADS */
1127 /* A wrapper for the standard C sleep function */
1128 int WRAP_FUNC(sleep) (unsigned int seconds)
1132 GC_start_blocking();
1133 result = REAL_FUNC(sleep)(seconds);
1139 void *(*start_routine)(void *);
1142 sem_t registered; /* 1 ==> in our thread table, but */
1143 /* parent hasn't yet noticed. */
1146 /* Called at thread exit. */
1147 /* Never called for main thread. That's OK, since it */
1148 /* results in at most a tiny one-time leak. And */
1149 /* linuxthreads doesn't reclaim the main threads */
1150 /* resources or id anyway. */
1151 void GC_thread_exit_proc(void *arg)
1156 me = GC_lookup_thread(pthread_self());
1157 GC_destroy_thread_local(me);
1158 if (me -> flags & DETACHED) {
1159 GC_delete_thread(pthread_self());
1161 me -> flags |= FINISHED;
1163 # if defined(THREAD_LOCAL_ALLOC) && !defined(USE_PTHREAD_SPECIFIC) \
1164 && !defined(USE_COMPILER_TLS) && !defined(DBG_HDRS_ALL)
1165 GC_remove_specific(GC_thread_key);
1167 /* The following may run the GC from "nonexistent" thread. */
1168 GC_wait_for_gc_completion(FALSE);
1172 int WRAP_FUNC(pthread_join)(pthread_t thread, void **retval)
1175 GC_thread thread_gc_id;
1178 thread_gc_id = GC_lookup_thread(thread);
1179 /* This is guaranteed to be the intended one, since the thread id */
1180 /* cant have been recycled by pthreads. */
1182 result = REAL_FUNC(pthread_join)(thread, retval);
1183 # if defined (GC_FREEBSD_THREADS)
1184 /* On FreeBSD, the wrapped pthread_join() sometimes returns (what
1185 appears to be) a spurious EINTR which caused the test and real code
1186 to gratuitously fail. Having looked at system pthread library source
1187 code, I see how this return code may be generated. In one path of
1188 code, pthread_join() just returns the errno setting of the thread
1189 being joined. This does not match the POSIX specification or the
1190 local man pages thus I have taken the liberty to catch this one
1191 spurious return value properly conditionalized on GC_FREEBSD_THREADS. */
1192 if (result == EINTR) result = 0;
1196 /* Here the pthread thread id may have been recycled. */
1197 GC_delete_gc_thread(thread, thread_gc_id);
1204 WRAP_FUNC(pthread_detach)(pthread_t thread)
1207 GC_thread thread_gc_id;
1210 thread_gc_id = GC_lookup_thread(thread);
1212 result = REAL_FUNC(pthread_detach)(thread);
1215 thread_gc_id -> flags |= DETACHED;
1216 /* Here the pthread thread id may have been recycled. */
1217 if (thread_gc_id -> flags & FINISHED) {
1218 GC_delete_gc_thread(thread, thread_gc_id);
1225 GC_bool GC_in_thread_creation = FALSE;
1227 typedef void *(*ThreadStartFn)(void *);
1228 void * GC_start_routine_head(void * arg, void *base_addr,
1229 ThreadStartFn *start, void **start_arg )
1231 struct start_info * si = arg;
1234 pthread_t my_pthread;
1236 my_pthread = pthread_self();
1237 # ifdef DEBUG_THREADS
1238 GC_printf1("Starting thread 0x%lx\n", my_pthread);
1239 GC_printf1("pid = %ld\n", (long) getpid());
1240 GC_printf1("sp = 0x%lx\n", (long) &arg);
1243 GC_in_thread_creation = TRUE;
1244 me = GC_new_thread(my_pthread);
1245 GC_in_thread_creation = FALSE;
1246 #ifdef GC_DARWIN_THREADS
1247 me -> stop_info.mach_thread = mach_thread_self();
1249 me -> stop_info.stack_ptr = 0;
1251 me -> flags = si -> flags;
1252 /* me -> stack_end = GC_linux_stack_base(); -- currently (11/99) */
1253 /* doesn't work because the stack base in /proc/self/stat is the */
1254 /* one for the main thread. There is a strong argument that that's */
1255 /* a kernel bug, but a pervasive one. */
1256 # ifdef STACK_GROWS_DOWN
1257 me -> stack_end = (ptr_t)(((word)(base_addr) + (GC_page_size - 1))
1258 & ~(GC_page_size - 1));
1259 # ifndef GC_DARWIN_THREADS
1260 me -> stop_info.stack_ptr = me -> stack_end - 0x10;
1262 /* Needs to be plausible, since an asynchronous stack mark */
1263 /* should not crash. */
1265 me -> stack_end = (ptr_t)((word)(base_addr) & ~(GC_page_size - 1));
1266 me -> stop_info.stack_ptr = me -> stack_end + 0x10;
1268 /* This is dubious, since we may be more than a page into the stack, */
1269 /* and hence skip some of it, though it's not clear that matters. */
1271 me -> backing_store_end = (ptr_t)
1272 (GC_save_regs_in_stack() & ~(GC_page_size - 1));
1273 /* This is also < 100% convincing. We should also read this */
1274 /* from /proc, but the hook to do so isn't there yet. */
1278 if (start) *start = si -> start_routine;
1279 if (start_arg) *start_arg = si -> arg;
1281 sem_post(&(si -> registered)); /* Last action on si. */
1282 /* OK to deallocate. */
1283 # if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
1285 GC_init_thread_local(me);
1292 int GC_thread_register_foreign (void *base_addr)
1294 struct start_info si = { 0, }; /* stacked for legibility & locking */
1297 # ifdef DEBUG_THREADS
1298 GC_printf1( "GC_thread_register_foreign %p\n", &si );
1301 si.flags = FOREIGN_THREAD;
1303 if (!parallel_initialized) GC_init_parallel();
1305 if (!GC_thr_initialized) GC_thr_init();
1309 me = GC_start_routine_head(&si, base_addr, NULL, NULL);
1314 void * GC_start_routine(void * arg)
1317 struct start_info * si = arg;
1320 ThreadStartFn start;
1323 me = GC_start_routine_head (arg, &dummy, &start, &start_arg);
1325 pthread_cleanup_push(GC_thread_exit_proc, 0);
1326 # ifdef DEBUG_THREADS
1327 GC_printf1("start_routine = 0x%lx\n", start);
1329 result = (*start)(start_arg);
1331 GC_printf1("Finishing thread 0x%x\n", pthread_self());
1333 me -> status = result;
1334 pthread_cleanup_pop(1);
1335 /* Cleanup acquires lock, ensuring that we can't exit */
1336 /* while a collection that thinks we're alive is trying to stop */
1342 WRAP_FUNC(pthread_create)(pthread_t *new_thread,
1343 const pthread_attr_t *attr,
1344 void *(*start_routine)(void *), void *arg)
1349 struct start_info * si;
1350 /* This is otherwise saved only in an area mmapped by the thread */
1351 /* library, which isn't visible to the collector. */
1353 /* We resist the temptation to muck with the stack size here, */
1354 /* even if the default is unreasonably small. That's the client's */
1355 /* responsibility. */
1358 si = (struct start_info *)GC_INTERNAL_MALLOC(sizeof(struct start_info),
1361 if (!parallel_initialized) GC_init_parallel();
1362 if (0 == si) return(ENOMEM);
1363 sem_init(&(si -> registered), 0, 0);
1364 si -> start_routine = start_routine;
1367 if (!GC_thr_initialized) GC_thr_init();
1368 # ifdef GC_ASSERTIONS
1372 pthread_attr_t my_attr;
1373 pthread_attr_init(&my_attr);
1374 pthread_attr_getstacksize(&my_attr, &stack_size);
1376 pthread_attr_getstacksize(attr, &stack_size);
1378 # ifdef PARALLEL_MARK
1379 GC_ASSERT(stack_size >= (8*HBLKSIZE*sizeof(word)));
1381 /* FreeBSD-5.3/Alpha: default pthread stack is 64K, */
1382 /* HBLKSIZE=8192, sizeof(word)=8 */
1383 GC_ASSERT(stack_size >= 65536);
1385 /* Our threads may need to do some work for the GC. */
1386 /* Ridiculously small threads won't work, and they */
1387 /* probably wouldn't work anyway. */
1391 detachstate = PTHREAD_CREATE_JOINABLE;
1393 pthread_attr_getdetachstate(attr, &detachstate);
1395 if (PTHREAD_CREATE_DETACHED == detachstate) my_flags |= DETACHED;
1396 si -> flags = my_flags;
1398 # ifdef DEBUG_THREADS
1399 GC_printf1("About to start new thread from thread 0x%X\n",
1403 result = REAL_FUNC(pthread_create)(new_thread, attr, GC_start_routine, si);
1405 # ifdef DEBUG_THREADS
1406 GC_printf1("Started thread 0x%X\n", *new_thread);
1408 /* Wait until child has been added to the thread table. */
1409 /* This also ensures that we hold onto si until the child is done */
1410 /* with it. Thus it doesn't matter whether it is otherwise */
1411 /* visible to the collector. */
1413 while (0 != sem_wait(&(si -> registered))) {
1414 if (EINTR != errno) ABORT("sem_wait failed");
1417 sem_destroy(&(si -> registered));
1419 GC_INTERNAL_FREE(si);
1425 #ifdef GENERIC_COMPARE_AND_SWAP
1426 pthread_mutex_t GC_compare_and_swap_lock = PTHREAD_MUTEX_INITIALIZER;
1428 GC_bool GC_compare_and_exchange(volatile GC_word *addr,
1429 GC_word old, GC_word new_val)
1432 pthread_mutex_lock(&GC_compare_and_swap_lock);
1439 pthread_mutex_unlock(&GC_compare_and_swap_lock);
1443 GC_word GC_atomic_add(volatile GC_word *addr, GC_word how_much)
1446 pthread_mutex_lock(&GC_compare_and_swap_lock);
1448 *addr = old + how_much;
1449 pthread_mutex_unlock(&GC_compare_and_swap_lock);
1453 #endif /* GENERIC_COMPARE_AND_SWAP */
1454 /* Spend a few cycles in a way that can't introduce contention with */
1455 /* othre threads. */
1459 # if !defined(__GNUC__) || defined(__INTEL_COMPILER)
1460 volatile word dummy = 0;
1463 for (i = 0; i < 10; ++i) {
1464 # if defined(__GNUC__) && !defined(__INTEL_COMPILER)
1465 __asm__ __volatile__ (" " : : : "memory");
1467 /* Something that's unlikely to be optimized away. */
1473 #define SPIN_MAX 128 /* Maximum number of calls to GC_pause before */
1476 VOLATILE GC_bool GC_collecting = 0;
1477 /* A hint that we're in the collector and */
1478 /* holding the allocation lock for an */
1479 /* extended period. */
1481 #if !defined(USE_SPIN_LOCK) || defined(PARALLEL_MARK)
1482 /* If we don't want to use the below spinlock implementation, either */
1483 /* because we don't have a GC_test_and_set implementation, or because */
1484 /* we don't want to risk sleeping, we can still try spinning on */
1485 /* pthread_mutex_trylock for a while. This appears to be very */
1486 /* beneficial in many cases. */
1487 /* I suspect that under high contention this is nearly always better */
1488 /* than the spin lock. But it's a bit slower on a uniprocessor. */
1489 /* Hence we still default to the spin lock. */
1490 /* This is also used to acquire the mark lock for the parallel */
1493 /* Here we use a strict exponential backoff scheme. I don't know */
1494 /* whether that's better or worse than the above. We eventually */
1495 /* yield by calling pthread_mutex_lock(); it never makes sense to */
1496 /* explicitly sleep. */
1500 unsigned long GC_spin_count = 0;
1501 unsigned long GC_block_count = 0;
1502 unsigned long GC_unlocked_count = 0;
1505 void GC_generic_lock(pthread_mutex_t * lock)
1507 #ifndef NO_PTHREAD_TRYLOCK
1508 unsigned pause_length = 1;
1511 if (0 == pthread_mutex_trylock(lock)) {
1513 ++GC_unlocked_count;
1517 for (; pause_length <= SPIN_MAX; pause_length <<= 1) {
1518 for (i = 0; i < pause_length; ++i) {
1521 switch(pthread_mutex_trylock(lock)) {
1530 ABORT("Unexpected error from pthread_mutex_trylock");
1533 #endif /* !NO_PTHREAD_TRYLOCK */
1537 pthread_mutex_lock(lock);
1540 #endif /* !USE_SPIN_LOCK || PARALLEL_MARK */
1542 #if defined(USE_SPIN_LOCK)
1544 /* Reasonably fast spin locks. Basically the same implementation */
1545 /* as STL alloc.h. This isn't really the right way to do this. */
1546 /* but until the POSIX scheduling mess gets straightened out ... */
1548 volatile unsigned int GC_allocate_lock = 0;
1553 # define low_spin_max 30 /* spin cycles if we suspect uniprocessor */
1554 # define high_spin_max SPIN_MAX /* spin cycles for multiprocessor */
1555 static unsigned spin_max = low_spin_max;
1556 unsigned my_spin_max;
1557 static unsigned last_spins = 0;
1558 unsigned my_last_spins;
1561 if (!GC_test_and_set(&GC_allocate_lock)) {
1564 my_spin_max = spin_max;
1565 my_last_spins = last_spins;
1566 for (i = 0; i < my_spin_max; i++) {
1567 if (GC_collecting || GC_nprocs == 1) goto yield;
1568 if (i < my_last_spins/2 || GC_allocate_lock) {
1572 if (!GC_test_and_set(&GC_allocate_lock)) {
1575 * Spinning worked. Thus we're probably not being scheduled
1576 * against the other process with which we were contending.
1577 * Thus it makes sense to spin longer the next time.
1580 spin_max = high_spin_max;
1584 /* We are probably being scheduled against the other process. Sleep. */
1585 spin_max = low_spin_max;
1588 if (!GC_test_and_set(&GC_allocate_lock)) {
1591 # define SLEEP_THRESHOLD 12
1592 /* Under Linux very short sleeps tend to wait until */
1593 /* the current time quantum expires. On old Linux */
1594 /* kernels nanosleep(<= 2ms) just spins under Linux. */
1595 /* (Under 2.4, this happens only for real-time */
1596 /* processes.) We want to minimize both behaviors */
1598 if (i < SLEEP_THRESHOLD) {
1604 /* Don't wait for more than about 15msecs, even */
1605 /* under extreme contention. */
1607 ts.tv_nsec = 1 << i;
1613 #else /* !USE_SPINLOCK */
1616 #ifndef NO_PTHREAD_TRYLOCK
1617 if (1 == GC_nprocs || GC_collecting) {
1618 pthread_mutex_lock(&GC_allocate_ml);
1620 GC_generic_lock(&GC_allocate_ml);
1622 #else /* !NO_PTHREAD_TRYLOCK */
1623 pthread_mutex_lock(&GC_allocate_ml);
1624 #endif /* !NO_PTHREAD_TRYLOCK */
1627 #endif /* !USE_SPINLOCK */
1629 #if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
1631 #ifdef GC_ASSERTIONS
1632 pthread_t GC_mark_lock_holder = NO_THREAD;
1636 /* Ugly workaround for a linux threads bug in the final versions */
1637 /* of glibc2.1. Pthread_mutex_trylock sets the mutex owner */
1638 /* field even when it fails to acquire the mutex. This causes */
1639 /* pthread_cond_wait to die. Remove for glibc2.2. */
1640 /* According to the man page, we should use */
1641 /* PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP, but that isn't actually */
1643 static pthread_mutex_t mark_mutex =
1644 {0, 0, 0, PTHREAD_MUTEX_ERRORCHECK_NP, {0, 0}};
1646 static pthread_mutex_t mark_mutex = PTHREAD_MUTEX_INITIALIZER;
1649 static pthread_cond_t builder_cv = PTHREAD_COND_INITIALIZER;
1651 void GC_acquire_mark_lock()
1654 if (pthread_mutex_lock(&mark_mutex) != 0) {
1655 ABORT("pthread_mutex_lock failed");
1658 GC_generic_lock(&mark_mutex);
1659 # ifdef GC_ASSERTIONS
1660 GC_mark_lock_holder = pthread_self();
1664 void GC_release_mark_lock()
1666 GC_ASSERT(GC_mark_lock_holder == pthread_self());
1667 # ifdef GC_ASSERTIONS
1668 GC_mark_lock_holder = NO_THREAD;
1670 if (pthread_mutex_unlock(&mark_mutex) != 0) {
1671 ABORT("pthread_mutex_unlock failed");
1675 /* Collector must wait for a freelist builders for 2 reasons: */
1676 /* 1) Mark bits may still be getting examined without lock. */
1677 /* 2) Partial free lists referenced only by locals may not be scanned */
1678 /* correctly, e.g. if they contain "pointer-free" objects, since the */
1679 /* free-list link may be ignored. */
1680 void GC_wait_builder()
1682 GC_ASSERT(GC_mark_lock_holder == pthread_self());
1683 # ifdef GC_ASSERTIONS
1684 GC_mark_lock_holder = NO_THREAD;
1686 if (pthread_cond_wait(&builder_cv, &mark_mutex) != 0) {
1687 ABORT("pthread_cond_wait failed");
1689 GC_ASSERT(GC_mark_lock_holder == NO_THREAD);
1690 # ifdef GC_ASSERTIONS
1691 GC_mark_lock_holder = pthread_self();
1695 void GC_wait_for_reclaim()
1697 GC_acquire_mark_lock();
1698 while (GC_fl_builder_count > 0) {
1701 GC_release_mark_lock();
1704 void GC_notify_all_builder()
1706 GC_ASSERT(GC_mark_lock_holder == pthread_self());
1707 if (pthread_cond_broadcast(&builder_cv) != 0) {
1708 ABORT("pthread_cond_broadcast failed");
1712 #endif /* PARALLEL_MARK || THREAD_LOCAL_ALLOC */
1714 #ifdef PARALLEL_MARK
1716 static pthread_cond_t mark_cv = PTHREAD_COND_INITIALIZER;
1718 void GC_wait_marker()
1720 GC_ASSERT(GC_mark_lock_holder == pthread_self());
1721 # ifdef GC_ASSERTIONS
1722 GC_mark_lock_holder = NO_THREAD;
1724 if (pthread_cond_wait(&mark_cv, &mark_mutex) != 0) {
1725 ABORT("pthread_cond_wait failed");
1727 GC_ASSERT(GC_mark_lock_holder == NO_THREAD);
1728 # ifdef GC_ASSERTIONS
1729 GC_mark_lock_holder = pthread_self();
1733 void GC_notify_all_marker()
1735 if (pthread_cond_broadcast(&mark_cv) != 0) {
1736 ABORT("pthread_cond_broadcast failed");
1740 #endif /* PARALLEL_MARK */
1742 # endif /* GC_LINUX_THREADS and friends */