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) 2000-2001 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_IRIX_THREADS) && !defined(GC_WIN32_THREADS) \
54 && !defined(GC_AIX_THREADS)
56 # if defined(GC_HPUX_THREADS) && !defined(USE_PTHREAD_SPECIFIC) \
57 && !defined(USE_COMPILER_TLS)
59 # define USE_PTHREAD_SPECIFIC
60 /* Empirically, as of gcc 3.3, USE_COMPILER_TLS doesn't work. */
62 # define USE_COMPILER_TLS
66 # if defined USE_HPUX_TLS
67 --> Macro replaced by USE_COMPILER_TLS
70 # if (defined(GC_DGUX386_THREADS) || defined(GC_OSF1_THREADS) || \
71 defined(GC_DARWIN_THREADS)) && !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"
87 # if defined(USE_PTHREAD_SPECIFIC)
88 # define GC_getspecific pthread_getspecific
89 # define GC_setspecific pthread_setspecific
90 # define GC_key_create pthread_key_create
91 typedef pthread_key_t GC_key_t;
93 # if defined(USE_COMPILER_TLS)
94 # define GC_getspecific(x) (x)
95 # define GC_setspecific(key, v) ((key) = (v), 0)
96 # define GC_key_create(key, d) 0
97 typedef void * GC_key_t;
101 # include <pthread.h>
106 # include <sys/mman.h>
107 # include <sys/time.h>
108 # include <sys/types.h>
109 # include <sys/stat.h>
113 #if defined(GC_DARWIN_THREADS)
114 # include "private/darwin_semaphore.h"
116 # include <semaphore.h>
117 #endif /* !GC_DARWIN_THREADS */
119 #if defined(GC_DARWIN_THREADS)
120 # include <sys/sysctl.h>
121 #endif /* GC_DARWIN_THREADS */
125 #if defined(GC_DGUX386_THREADS)
126 # include <sys/dg_sys_info.h>
127 # include <sys/_int_psem.h>
128 /* sem_t is an uint in DG/UX */
129 typedef unsigned int sem_t;
130 #endif /* GC_DGUX386_THREADS */
136 #ifdef GC_USE_LD_WRAP
137 # define WRAP_FUNC(f) __wrap_##f
138 # define REAL_FUNC(f) __real_##f
140 # define WRAP_FUNC(f) GC_##f
141 # if !defined(GC_DGUX386_THREADS)
142 # define REAL_FUNC(f) f
143 # else /* GC_DGUX386_THREADS */
144 # define REAL_FUNC(f) __d10_##f
145 # endif /* GC_DGUX386_THREADS */
146 # undef pthread_create
147 # if !defined(GC_DARWIN_THREADS)
148 # undef pthread_sigmask
151 # undef pthread_detach
152 # if defined(GC_OSF1_THREADS) && defined(_PTHREAD_USE_MANGLED_NAMES_) \
153 && !defined(_PTHREAD_USE_PTDNAM_)
154 /* Restore the original mangled names on Tru64 UNIX. */
155 # define pthread_create __pthread_create
156 # define pthread_join __pthread_join
157 # define pthread_detach __pthread_detach
163 static GC_bool parallel_initialized = FALSE;
165 void GC_init_parallel();
167 # if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
169 /* We don't really support thread-local allocation with DBG_HDRS_ALL */
172 #ifdef USE_COMPILER_TLS
175 GC_key_t GC_thread_key;
177 static GC_bool keys_initialized;
179 /* Recover the contents of the freelist array fl into the global one gfl.*/
180 /* Note that the indexing scheme differs, in that gfl has finer size */
181 /* resolution, even if not all entries are used. */
182 /* We hold the allocator lock. */
183 static void return_freelists(ptr_t *fl, ptr_t *gfl)
189 for (i = 1; i < NFREELISTS; ++i) {
190 nwords = i * (GRANULARITY/sizeof(word));
193 if ((word)q >= HBLKSIZE) {
194 if (gfl[nwords] == 0) {
198 for (; (word)q >= HBLKSIZE; qptr = &(obj_link(q)), q = *qptr);
204 /* Clear fl[i], since the thread structure may hang around. */
205 /* Do it in a way that is likely to trap if we access it. */
206 fl[i] = (ptr_t)HBLKSIZE;
210 /* We statically allocate a single "size 0" object. It is linked to */
211 /* itself, and is thus repeatedly reused for all size 0 allocation */
212 /* requests. (Size 0 gcj allocation requests are incorrect, and */
213 /* we arrange for those to fault asap.) */
214 static ptr_t size_zero_object = (ptr_t)(&size_zero_object);
216 /* Each thread structure must be initialized. */
217 /* This call must be made from the new thread. */
218 /* Caller holds allocation lock. */
219 void GC_init_thread_local(GC_thread p)
223 if (!keys_initialized) {
224 if (0 != GC_key_create(&GC_thread_key, 0)) {
225 ABORT("Failed to create key for local allocator");
227 keys_initialized = TRUE;
229 if (0 != GC_setspecific(GC_thread_key, p)) {
230 ABORT("Failed to set thread specific allocation pointers");
232 for (i = 1; i < NFREELISTS; ++i) {
233 p -> ptrfree_freelists[i] = (ptr_t)1;
234 p -> normal_freelists[i] = (ptr_t)1;
235 # ifdef GC_GCJ_SUPPORT
236 p -> gcj_freelists[i] = (ptr_t)1;
239 /* Set up the size 0 free lists. */
240 p -> ptrfree_freelists[0] = (ptr_t)(&size_zero_object);
241 p -> normal_freelists[0] = (ptr_t)(&size_zero_object);
242 # ifdef GC_GCJ_SUPPORT
243 p -> gcj_freelists[0] = (ptr_t)(-1);
247 #ifdef GC_GCJ_SUPPORT
248 extern ptr_t * GC_gcjobjfreelist;
251 /* We hold the allocator lock. */
252 void GC_destroy_thread_local(GC_thread p)
254 /* We currently only do this from the thread itself or from */
255 /* the fork handler for a child process. */
257 GC_ASSERT(GC_getspecific(GC_thread_key) == (void *)p);
259 return_freelists(p -> ptrfree_freelists, GC_aobjfreelist);
260 return_freelists(p -> normal_freelists, GC_objfreelist);
261 # ifdef GC_GCJ_SUPPORT
262 return_freelists(p -> gcj_freelists, GC_gcjobjfreelist);
266 extern GC_PTR GC_generic_malloc_many();
268 GC_PTR GC_local_malloc(size_t bytes)
270 if (EXPECT(!SMALL_ENOUGH(bytes),0)) {
271 return(GC_malloc(bytes));
273 int index = INDEX_FROM_BYTES(bytes);
276 # if defined(REDIRECT_MALLOC) && !defined(USE_PTHREAD_SPECIFIC)
277 GC_key_t k = GC_thread_key;
281 # if defined(REDIRECT_MALLOC) && !defined(USE_PTHREAD_SPECIFIC)
282 if (EXPECT(0 == k, 0)) {
283 /* This can happen if we get called when the world is */
284 /* being initialized. Whether we can actually complete */
285 /* the initialization then is unclear. */
290 tsd = GC_getspecific(GC_thread_key);
291 # ifdef GC_ASSERTIONS
293 GC_ASSERT(tsd == (void *)GC_lookup_thread(pthread_self()));
296 my_fl = ((GC_thread)tsd) -> normal_freelists + index;
298 if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
299 ptr_t next = obj_link(my_entry);
300 GC_PTR result = (GC_PTR)my_entry;
302 obj_link(my_entry) = 0;
303 PREFETCH_FOR_WRITE(next);
305 } else if ((word)my_entry - 1 < DIRECT_GRANULES) {
306 *my_fl = my_entry + index + 1;
307 return GC_malloc(bytes);
309 GC_generic_malloc_many(BYTES_FROM_INDEX(index), NORMAL, my_fl);
310 if (*my_fl == 0) return GC_oom_fn(bytes);
311 return GC_local_malloc(bytes);
316 GC_PTR GC_local_malloc_atomic(size_t bytes)
318 if (EXPECT(!SMALL_ENOUGH(bytes), 0)) {
319 return(GC_malloc_atomic(bytes));
321 int index = INDEX_FROM_BYTES(bytes);
322 ptr_t * my_fl = ((GC_thread)GC_getspecific(GC_thread_key))
323 -> ptrfree_freelists + index;
324 ptr_t my_entry = *my_fl;
326 if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
327 GC_PTR result = (GC_PTR)my_entry;
328 *my_fl = obj_link(my_entry);
330 } else if ((word)my_entry - 1 < DIRECT_GRANULES) {
331 *my_fl = my_entry + index + 1;
332 return GC_malloc_atomic(bytes);
334 GC_generic_malloc_many(BYTES_FROM_INDEX(index), PTRFREE, my_fl);
335 /* *my_fl is updated while the collector is excluded; */
336 /* the free list is always visible to the collector as */
338 if (*my_fl == 0) return GC_oom_fn(bytes);
339 return GC_local_malloc_atomic(bytes);
344 #ifdef GC_GCJ_SUPPORT
346 #include "include/gc_gcj.h"
349 extern GC_bool GC_gcj_malloc_initialized;
352 extern int GC_gcj_kind;
354 GC_PTR GC_local_gcj_malloc(size_t bytes,
355 void * ptr_to_struct_containing_descr)
357 GC_ASSERT(GC_gcj_malloc_initialized);
358 if (EXPECT(!SMALL_ENOUGH(bytes), 0)) {
359 return GC_gcj_malloc(bytes, ptr_to_struct_containing_descr);
361 int index = INDEX_FROM_BYTES(bytes);
362 ptr_t * my_fl = ((GC_thread)GC_getspecific(GC_thread_key))
363 -> gcj_freelists + index;
364 ptr_t my_entry = *my_fl;
365 if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
366 GC_PTR result = (GC_PTR)my_entry;
367 GC_ASSERT(!GC_incremental);
368 /* We assert that any concurrent marker will stop us. */
369 /* Thus it is impossible for a mark procedure to see the */
370 /* allocation of the next object, but to see this object */
371 /* still containing a free list pointer. Otherwise the */
372 /* marker might find a random "mark descriptor". */
373 *(volatile ptr_t *)my_fl = obj_link(my_entry);
374 /* We must update the freelist before we store the pointer. */
375 /* Otherwise a GC at this point would see a corrupted */
377 /* A memory barrier is probably never needed, since the */
378 /* action of stopping this thread will cause prior writes */
380 GC_ASSERT(((void * volatile *)result)[1] == 0);
381 *(void * volatile *)result = ptr_to_struct_containing_descr;
383 } else if ((word)my_entry - 1 < DIRECT_GRANULES) {
384 if (!GC_incremental) *my_fl = my_entry + index + 1;
385 /* In the incremental case, we always have to take this */
386 /* path. Thus we leave the counter alone. */
387 return GC_gcj_malloc(bytes, ptr_to_struct_containing_descr);
389 GC_generic_malloc_many(BYTES_FROM_INDEX(index), GC_gcj_kind, my_fl);
390 if (*my_fl == 0) return GC_oom_fn(bytes);
391 return GC_local_gcj_malloc(bytes, ptr_to_struct_containing_descr);
396 /* Similar to GC_local_gcj_malloc, but the size is in words, and we don't */
397 /* adjust it. The size is assumed to be such that it can be */
398 /* allocated as a small object. */
399 void * GC_local_gcj_fast_malloc(size_t lw, void * ptr_to_struct_containing_descr)
401 ptr_t * my_fl = ((GC_thread)GC_getspecific(GC_thread_key))
402 -> gcj_freelists + lw;
403 ptr_t my_entry = *my_fl;
405 GC_ASSERT(GC_gcj_malloc_initialized);
407 if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
408 GC_PTR result = (GC_PTR)my_entry;
409 GC_ASSERT(!GC_incremental);
410 /* We assert that any concurrent marker will stop us. */
411 /* Thus it is impossible for a mark procedure to see the */
412 /* allocation of the next object, but to see this object */
413 /* still containing a free list pointer. Otherwise the */
414 /* marker might find a random "mark descriptor". */
415 *(volatile ptr_t *)my_fl = obj_link(my_entry);
416 /* We must update the freelist before we store the pointer. */
417 /* Otherwise a GC at this point would see a corrupted */
419 /* A memory barrier is probably never needed, since the */
420 /* action of stopping this thread will cause prior writes */
422 GC_ASSERT(((void * volatile *)result)[1] == 0);
423 *(void * volatile *)result = ptr_to_struct_containing_descr;
425 } else if ((word)my_entry - 1 < DIRECT_GRANULES) {
426 if (!GC_incremental) *my_fl = my_entry + lw + 1;
427 /* In the incremental case, we always have to take this */
428 /* path. Thus we leave the counter alone. */
429 return GC_gcj_fast_malloc(lw, ptr_to_struct_containing_descr);
431 GC_generic_malloc_many(BYTES_FROM_INDEX(lw), GC_gcj_kind, my_fl);
432 if (*my_fl == 0) return GC_oom_fn(BYTES_FROM_INDEX(lw));
433 return GC_local_gcj_fast_malloc(lw, ptr_to_struct_containing_descr);
437 #endif /* GC_GCJ_SUPPORT */
439 # else /* !THREAD_LOCAL_ALLOC && !DBG_HDRS_ALL */
441 # define GC_destroy_thread_local(t)
443 # endif /* !THREAD_LOCAL_ALLOC */
447 To make sure that we're using LinuxThreads and not some other thread
448 package, we generate a dummy reference to `pthread_kill_other_threads_np'
449 (was `__pthread_initial_thread_bos' but that disappeared),
450 which is a symbol defined in LinuxThreads, but (hopefully) not in other
453 We no longer do this, since this code is now portable enough that it might
454 actually work for something else.
456 void (*dummy_var_to_force_linux_threads)() = pthread_kill_other_threads_np;
459 long GC_nprocs = 1; /* Number of processors. We may not have */
460 /* access to all of them, but this is as good */
461 /* a guess as any ... */
466 # define MAX_MARKERS 16
469 static ptr_t marker_sp[MAX_MARKERS] = {0};
471 void * GC_mark_thread(void * id)
475 marker_sp[(word)id] = GC_approx_sp();
476 for (;; ++my_mark_no) {
477 /* GC_mark_no is passed only to allow GC_help_marker to terminate */
478 /* promptly. This is important if it were called from the signal */
479 /* handler or from the GC lock acquisition code. Under Linux, it's */
480 /* not safe to call it from a signal handler, since it uses mutexes */
481 /* and condition variables. Since it is called only here, the */
482 /* argument is unnecessary. */
483 if (my_mark_no < GC_mark_no || my_mark_no > GC_mark_no + 2) {
484 /* resynchronize if we get far off, e.g. because GC_mark_no */
486 my_mark_no = GC_mark_no;
488 # ifdef DEBUG_THREADS
489 GC_printf1("Starting mark helper for mark number %ld\n", my_mark_no);
491 GC_help_marker(my_mark_no);
495 extern long GC_markers; /* Number of mark threads we would */
496 /* like to have. Includes the */
497 /* initiating thread. */
499 pthread_t GC_mark_threads[MAX_MARKERS];
501 #define PTHREAD_CREATE REAL_FUNC(pthread_create)
503 static void start_mark_threads()
508 if (GC_markers > MAX_MARKERS) {
509 WARN("Limiting number of mark threads\n", 0);
510 GC_markers = MAX_MARKERS;
512 if (0 != pthread_attr_init(&attr)) ABORT("pthread_attr_init failed");
514 if (0 != pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED))
515 ABORT("pthread_attr_setdetachstate failed");
517 # if defined(HPUX) || defined(GC_DGUX386_THREADS)
518 /* Default stack size is usually too small: fix it. */
519 /* Otherwise marker threads or GC may run out of */
521 # define MIN_STACK_SIZE (8*HBLKSIZE*sizeof(word))
526 if (pthread_attr_getstacksize(&attr, &old_size) != 0)
527 ABORT("pthread_attr_getstacksize failed\n");
528 if (old_size < MIN_STACK_SIZE) {
529 if (pthread_attr_setstacksize(&attr, MIN_STACK_SIZE) != 0)
530 ABORT("pthread_attr_setstacksize failed\n");
533 # endif /* HPUX || GC_DGUX386_THREADS */
535 if (GC_print_stats) {
536 GC_printf1("Starting %ld marker threads\n", GC_markers - 1);
539 for (i = 0; i < GC_markers - 1; ++i) {
540 if (0 != PTHREAD_CREATE(GC_mark_threads + i, &attr,
541 GC_mark_thread, (void *)(word)i)) {
542 WARN("Marker thread creation failed, errno = %ld.\n", errno);
547 #else /* !PARALLEL_MARK */
549 static __inline__ void start_mark_threads()
553 #endif /* !PARALLEL_MARK */
555 GC_bool GC_thr_initialized = FALSE;
557 volatile GC_thread GC_threads[THREAD_TABLE_SZ];
559 void GC_push_thread_structures GC_PROTO((void))
561 GC_push_all((ptr_t)(GC_threads), (ptr_t)(GC_threads)+sizeof(GC_threads));
562 # if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
563 GC_push_all((ptr_t)(&GC_thread_key),
564 (ptr_t)(&GC_thread_key)+sizeof(&GC_thread_key));
568 #ifdef THREAD_LOCAL_ALLOC
569 /* We must explicitly mark ptrfree and gcj free lists, since the free */
570 /* list links wouldn't otherwise be found. We also set them in the */
571 /* normal free lists, since that involves touching less memory than if */
572 /* we scanned them normally. */
573 void GC_mark_thread_local_free_lists(void)
579 for (i = 0; i < THREAD_TABLE_SZ; ++i) {
580 for (p = GC_threads[i]; 0 != p; p = p -> next) {
581 for (j = 1; j < NFREELISTS; ++j) {
582 q = p -> ptrfree_freelists[j];
583 if ((word)q > HBLKSIZE) GC_set_fl_marks(q);
584 q = p -> normal_freelists[j];
585 if ((word)q > HBLKSIZE) GC_set_fl_marks(q);
586 # ifdef GC_GCJ_SUPPORT
587 q = p -> gcj_freelists[j];
588 if ((word)q > HBLKSIZE) GC_set_fl_marks(q);
589 # endif /* GC_GCJ_SUPPORT */
594 #endif /* THREAD_LOCAL_ALLOC */
596 static struct GC_Thread_Rep first_thread;
598 /* Add a thread to GC_threads. We assume it wasn't already there. */
599 /* Caller holds allocation lock. */
600 GC_thread GC_new_thread(pthread_t id)
602 int hv = ((word)id) % THREAD_TABLE_SZ;
604 static GC_bool first_thread_used = FALSE;
606 if (!first_thread_used) {
607 result = &first_thread;
608 first_thread_used = TRUE;
610 result = (struct GC_Thread_Rep *)
611 GC_INTERNAL_MALLOC(sizeof(struct GC_Thread_Rep), NORMAL);
613 if (result == 0) return(0);
615 result -> next = GC_threads[hv];
616 GC_threads[hv] = result;
617 GC_ASSERT(result -> flags == 0 && result -> thread_blocked == 0);
621 /* Delete a thread from GC_threads. We assume it is there. */
622 /* (The code intentionally traps if it wasn't.) */
623 /* Caller holds allocation lock. */
624 void GC_delete_thread(pthread_t id)
626 int hv = ((word)id) % THREAD_TABLE_SZ;
627 register GC_thread p = GC_threads[hv];
628 register GC_thread prev = 0;
630 while (!pthread_equal(p -> id, id)) {
635 GC_threads[hv] = p -> next;
637 prev -> next = p -> next;
642 /* If a thread has been joined, but we have not yet */
643 /* been notified, then there may be more than one thread */
644 /* in the table with the same pthread id. */
645 /* This is OK, but we need a way to delete a specific one. */
646 void GC_delete_gc_thread(pthread_t id, GC_thread gc_id)
648 int hv = ((word)id) % THREAD_TABLE_SZ;
649 register GC_thread p = GC_threads[hv];
650 register GC_thread prev = 0;
657 GC_threads[hv] = p -> next;
659 prev -> next = p -> next;
664 /* Return a GC_thread corresponding to a given pthread_t. */
665 /* Returns 0 if it's not there. */
666 /* Caller holds allocation lock or otherwise inhibits */
668 /* If there is more than one thread with the given id we */
669 /* return the most recent one. */
670 GC_thread GC_lookup_thread(pthread_t id)
672 int hv = ((word)id) % THREAD_TABLE_SZ;
673 register GC_thread p = GC_threads[hv];
675 while (p != 0 && !pthread_equal(p -> id, id)) p = p -> next;
679 int GC_thread_is_registered (void)
684 ptr = (void *)GC_lookup_thread(pthread_self());
691 /* Remove all entries from the GC_threads table, except the */
692 /* one for the current thread. We need to do this in the child */
693 /* process after a fork(), since only the current thread */
694 /* survives in the child. */
695 void GC_remove_all_threads_but_me(void)
697 pthread_t self = pthread_self();
699 GC_thread p, next, me;
701 for (hv = 0; hv < THREAD_TABLE_SZ; ++hv) {
703 for (p = GC_threads[hv]; 0 != p; p = next) {
705 if (p -> id == self) {
709 # ifdef THREAD_LOCAL_ALLOC
710 if (!(p -> flags & FINISHED)) {
711 GC_destroy_thread_local(p);
713 # endif /* THREAD_LOCAL_ALLOC */
714 if (p != &first_thread) GC_INTERNAL_FREE(p);
720 #endif /* HANDLE_FORK */
722 #ifdef USE_PROC_FOR_LIBRARIES
723 int GC_segment_is_thread_stack(ptr_t lo, ptr_t hi)
728 # ifdef PARALLEL_MARK
729 for (i = 0; i < GC_markers; ++i) {
730 if (marker_sp[i] > lo & marker_sp[i] < hi) return 1;
733 for (i = 0; i < THREAD_TABLE_SZ; i++) {
734 for (p = GC_threads[i]; p != 0; p = p -> next) {
735 if (0 != p -> stack_end) {
736 # ifdef STACK_GROWS_UP
737 if (p -> stack_end >= lo && p -> stack_end < hi) return 1;
738 # else /* STACK_GROWS_DOWN */
739 if (p -> stack_end > lo && p -> stack_end <= hi) return 1;
746 #endif /* USE_PROC_FOR_LIBRARIES */
748 #ifdef GC_LINUX_THREADS
749 /* Return the number of processors, or i<= 0 if it can't be determined. */
752 /* Should be "return sysconf(_SC_NPROCESSORS_ONLN);" but that */
753 /* appears to be buggy in many cases. */
754 /* We look for lines "cpu<n>" in /proc/stat. */
755 # define STAT_BUF_SIZE 4096
756 # define STAT_READ read
757 /* If read is wrapped, this may need to be redefined to call */
759 char stat_buf[STAT_BUF_SIZE];
762 /* Some old kernels only have a single "cpu nnnn ..." */
763 /* entry in /proc/stat. We identify those as */
767 f = open("/proc/stat", O_RDONLY);
768 if (f < 0 || (len = STAT_READ(f, stat_buf, STAT_BUF_SIZE)) < 100) {
769 WARN("Couldn't read /proc/stat\n", 0);
772 for (i = 0; i < len - 100; ++i) {
773 if (stat_buf[i] == '\n' && stat_buf[i+1] == 'c'
774 && stat_buf[i+2] == 'p' && stat_buf[i+3] == 'u') {
775 int cpu_no = atoi(stat_buf + i + 4);
776 if (cpu_no >= result) result = cpu_no + 1;
782 #endif /* GC_LINUX_THREADS */
784 /* We hold the GC lock. Wait until an in-progress GC has finished. */
785 /* Repeatedly RELEASES GC LOCK in order to wait. */
786 /* If wait_for_all is true, then we exit with the GC lock held and no */
787 /* collection in progress; otherwise we just wait for the current GC */
789 extern GC_bool GC_collection_in_progress();
790 void GC_wait_for_gc_completion(GC_bool wait_for_all)
792 if (GC_incremental && GC_collection_in_progress()) {
793 int old_gc_no = GC_gc_no;
795 /* Make sure that no part of our stack is still on the mark stack, */
796 /* since it's about to be unmapped. */
797 while (GC_incremental && GC_collection_in_progress()
798 && (wait_for_all || old_gc_no == GC_gc_no)) {
800 GC_in_thread_creation = TRUE;
801 GC_collect_a_little_inner(1);
802 GC_in_thread_creation = FALSE;
812 /* Procedures called before and after a fork. The goal here is to make */
813 /* it safe to call GC_malloc() in a forked child. It's unclear that is */
814 /* attainable, since the single UNIX spec seems to imply that one */
815 /* should only call async-signal-safe functions, and we probably can't */
816 /* quite guarantee that. But we give it our best shot. (That same */
817 /* spec also implies that it's not safe to call the system malloc */
818 /* between fork() and exec(). Thus we're doing no worse than it. */
820 /* Called before a fork() */
821 void GC_fork_prepare_proc(void)
823 /* Acquire all relevant locks, so that after releasing the locks */
824 /* the child will see a consistent state in which monitor */
825 /* invariants hold. Unfortunately, we can't acquire libc locks */
826 /* we might need, and there seems to be no guarantee that libc */
827 /* must install a suitable fork handler. */
828 /* Wait for an ongoing GC to finish, since we can't finish it in */
829 /* the (one remaining thread in) the child. */
831 # if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
832 GC_wait_for_reclaim();
834 GC_wait_for_gc_completion(TRUE);
835 # if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
836 GC_acquire_mark_lock();
840 /* Called in parent after a fork() */
841 void GC_fork_parent_proc(void)
843 # if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
844 GC_release_mark_lock();
849 /* Called in child after a fork() */
850 void GC_fork_child_proc(void)
852 /* Clean up the thread table, so that just our thread is left. */
853 # if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
854 GC_release_mark_lock();
856 GC_remove_all_threads_but_me();
857 # ifdef PARALLEL_MARK
858 /* Turn off parallel marking in the child, since we are probably */
859 /* just going to exec, and we would have to restart mark threads. */
862 # endif /* PARALLEL_MARK */
865 #endif /* HANDLE_FORK */
867 #if defined(GC_DGUX386_THREADS)
868 /* Return the number of processors, or i<= 0 if it can't be determined. */
871 /* <takis@XFree86.Org> */
873 struct dg_sys_info_pm_info pm_sysinfo;
876 status = dg_sys_info((long int *) &pm_sysinfo,
877 DG_SYS_INFO_PM_INFO_TYPE, DG_SYS_INFO_PM_CURRENT_VERSION);
879 /* set -1 for error */
883 numCpus = pm_sysinfo.idle_vp_count;
885 # ifdef DEBUG_THREADS
886 GC_printf1("Number of active CPUs in this system: %d\n", numCpus);
890 #endif /* GC_DGUX386_THREADS */
892 /* We hold the allocation lock. */
895 # ifndef GC_DARWIN_THREADS
900 if (GC_thr_initialized) return;
901 GC_thr_initialized = TRUE;
904 /* Prepare for a possible fork. */
905 pthread_atfork(GC_fork_prepare_proc, GC_fork_parent_proc,
907 # endif /* HANDLE_FORK */
908 /* Add the initial thread, so we can stop it. */
909 t = GC_new_thread(pthread_self());
910 # ifdef GC_DARWIN_THREADS
911 t -> stop_info.mach_thread = mach_thread_self();
913 t -> stop_info.stack_ptr = (ptr_t)(&dummy);
915 t -> flags = DETACHED | MAIN_THREAD;
921 char * nprocs_string = GETENV("GC_NPROCS");
923 if (nprocs_string != NULL) GC_nprocs = atoi(nprocs_string);
925 if (GC_nprocs <= 0) {
926 # if defined(GC_HPUX_THREADS)
927 GC_nprocs = pthread_num_processors_np();
929 # if defined(GC_OSF1_THREADS)
930 GC_nprocs = sysconf(_SC_NPROCESSORS_ONLN);
931 if (GC_nprocs <= 0) GC_nprocs = 1;
933 # if defined(GC_FREEBSD_THREADS)
936 # if defined(GC_DARWIN_THREADS)
938 size_t len = sizeof(ncpus);
939 sysctl((int[2]) {CTL_HW, HW_NCPU}, 2, &ncpus, &len, NULL, 0);
942 # if defined(GC_LINUX_THREADS) || defined(GC_DGUX386_THREADS)
943 GC_nprocs = GC_get_nprocs();
946 if (GC_nprocs <= 0) {
947 WARN("GC_get_nprocs() returned %ld\n", GC_nprocs);
949 # ifdef PARALLEL_MARK
953 # ifdef PARALLEL_MARK
955 char * markers_string = GETENV("GC_MARKERS");
956 if (markers_string != NULL) {
957 GC_markers = atoi(markers_string);
959 GC_markers = GC_nprocs;
964 # ifdef PARALLEL_MARK
966 if (GC_print_stats) {
967 GC_printf2("Number of processors = %ld, "
968 "number of marker threads = %ld\n", GC_nprocs, GC_markers);
971 if (GC_markers == 1) {
974 if (GC_print_stats) {
975 GC_printf0("Single marker thread, turning off parallel marking\n");
980 /* Disable true incremental collection, but generational is OK. */
981 GC_time_limit = GC_TIME_UNLIMITED;
987 /* Perform all initializations, including those that */
988 /* may require allocation. */
989 /* Called without allocation lock. */
990 /* Must be called before a second thread is created. */
991 /* Called without allocation lock. */
992 void GC_init_parallel()
994 if (parallel_initialized) return;
995 parallel_initialized = TRUE;
997 /* GC_init() calls us back, so set flag first. */
998 if (!GC_is_initialized) GC_init();
999 /* If we are using a parallel marker, start the helper threads. */
1000 # ifdef PARALLEL_MARK
1001 if (GC_parallel) start_mark_threads();
1003 /* Initialize thread local free lists if used. */
1004 # if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
1006 GC_init_thread_local(GC_lookup_thread(pthread_self()));
1012 #if !defined(GC_DARWIN_THREADS)
1013 int WRAP_FUNC(pthread_sigmask)(int how, const sigset_t *set, sigset_t *oset)
1015 sigset_t fudged_set;
1017 if (set != NULL && (how == SIG_BLOCK || how == SIG_SETMASK)) {
1019 sigdelset(&fudged_set, SIG_SUSPEND);
1022 return(REAL_FUNC(pthread_sigmask)(how, set, oset));
1024 #endif /* !GC_DARWIN_THREADS */
1026 /* Wrappers for functions that are likely to block for an appreciable */
1027 /* length of time. Must be called in pairs, if at all. */
1028 /* Nothing much beyond the system call itself should be executed */
1029 /* between these. */
1031 void GC_start_blocking(void) {
1032 # define SP_SLOP 128
1035 me = GC_lookup_thread(pthread_self());
1036 GC_ASSERT(!(me -> thread_blocked));
1038 me -> stop_info.stack_ptr = (ptr_t)GC_save_regs_in_stack();
1040 # ifndef GC_DARWIN_THREADS
1041 me -> stop_info.stack_ptr = (ptr_t)GC_approx_sp();
1045 me -> backing_store_ptr = (ptr_t)GC_save_regs_in_stack() + SP_SLOP;
1047 /* Add some slop to the stack pointer, since the wrapped call may */
1048 /* end up pushing more callee-save registers. */
1049 # ifndef GC_DARWIN_THREADS
1050 # ifdef STACK_GROWS_UP
1051 me -> stop_info.stack_ptr += SP_SLOP;
1053 me -> stop_info.stack_ptr -= SP_SLOP;
1056 me -> thread_blocked = TRUE;
1060 void GC_end_blocking(void) {
1062 LOCK(); /* This will block if the world is stopped. */
1063 me = GC_lookup_thread(pthread_self());
1064 GC_ASSERT(me -> thread_blocked);
1065 me -> thread_blocked = FALSE;
1069 #if defined(GC_DGUX386_THREADS)
1070 #define __d10_sleep sleep
1071 #endif /* GC_DGUX386_THREADS */
1073 /* A wrapper for the standard C sleep function */
1074 int WRAP_FUNC(sleep) (unsigned int seconds)
1078 GC_start_blocking();
1079 result = REAL_FUNC(sleep)(seconds);
1085 void *(*start_routine)(void *);
1088 sem_t registered; /* 1 ==> in our thread table, but */
1089 /* parent hasn't yet noticed. */
1092 /* Called at thread exit. */
1093 /* Never called for main thread. That's OK, since it */
1094 /* results in at most a tiny one-time leak. And */
1095 /* linuxthreads doesn't reclaim the main threads */
1096 /* resources or id anyway. */
1097 void GC_thread_exit_proc(void *arg)
1102 me = GC_lookup_thread(pthread_self());
1103 GC_destroy_thread_local(me);
1104 if (me -> flags & DETACHED) {
1105 GC_delete_thread(pthread_self());
1107 me -> flags |= FINISHED;
1109 # if defined(THREAD_LOCAL_ALLOC) && !defined(USE_PTHREAD_SPECIFIC) \
1110 && !defined(USE_COMPILER_TLS) && !defined(DBG_HDRS_ALL)
1111 GC_remove_specific(GC_thread_key);
1113 /* The following may run the GC from "nonexistent" thread. */
1114 GC_wait_for_gc_completion(FALSE);
1118 int WRAP_FUNC(pthread_join)(pthread_t thread, void **retval)
1121 GC_thread thread_gc_id;
1124 thread_gc_id = GC_lookup_thread(thread);
1125 /* This is guaranteed to be the intended one, since the thread id */
1126 /* cant have been recycled by pthreads. */
1128 result = REAL_FUNC(pthread_join)(thread, retval);
1129 # if defined (GC_FREEBSD_THREADS)
1130 /* On FreeBSD, the wrapped pthread_join() sometimes returns (what
1131 appears to be) a spurious EINTR which caused the test and real code
1132 to gratuitously fail. Having looked at system pthread library source
1133 code, I see how this return code may be generated. In one path of
1134 code, pthread_join() just returns the errno setting of the thread
1135 being joined. This does not match the POSIX specification or the
1136 local man pages thus I have taken the liberty to catch this one
1137 spurious return value properly conditionalized on GC_FREEBSD_THREADS. */
1138 if (result == EINTR) result = 0;
1142 /* Here the pthread thread id may have been recycled. */
1143 GC_delete_gc_thread(thread, thread_gc_id);
1150 WRAP_FUNC(pthread_detach)(pthread_t thread)
1153 GC_thread thread_gc_id;
1156 thread_gc_id = GC_lookup_thread(thread);
1158 result = REAL_FUNC(pthread_detach)(thread);
1161 thread_gc_id -> flags |= DETACHED;
1162 /* Here the pthread thread id may have been recycled. */
1163 if (thread_gc_id -> flags & FINISHED) {
1164 GC_delete_gc_thread(thread, thread_gc_id);
1171 GC_bool GC_in_thread_creation = FALSE;
1173 void * GC_start_routine(void * arg)
1176 struct start_info * si = arg;
1179 pthread_t my_pthread;
1180 void *(*start)(void *);
1183 my_pthread = pthread_self();
1184 # ifdef DEBUG_THREADS
1185 GC_printf1("Starting thread 0x%lx\n", my_pthread);
1186 GC_printf1("pid = %ld\n", (long) getpid());
1187 GC_printf1("sp = 0x%lx\n", (long) &arg);
1190 GC_in_thread_creation = TRUE;
1191 me = GC_new_thread(my_pthread);
1192 GC_in_thread_creation = FALSE;
1193 #ifdef GC_DARWIN_THREADS
1194 me -> stop_info.mach_thread = mach_thread_self();
1196 me -> stop_info.stack_ptr = 0;
1198 me -> flags = si -> flags;
1199 /* me -> stack_end = GC_linux_stack_base(); -- currently (11/99) */
1200 /* doesn't work because the stack base in /proc/self/stat is the */
1201 /* one for the main thread. There is a strong argument that that's */
1202 /* a kernel bug, but a pervasive one. */
1203 # ifdef STACK_GROWS_DOWN
1204 me -> stack_end = (ptr_t)(((word)(&dummy) + (GC_page_size - 1))
1205 & ~(GC_page_size - 1));
1206 # ifndef GC_DARWIN_THREADS
1207 me -> stop_info.stack_ptr = me -> stack_end - 0x10;
1209 /* Needs to be plausible, since an asynchronous stack mark */
1210 /* should not crash. */
1212 me -> stack_end = (ptr_t)((word)(&dummy) & ~(GC_page_size - 1));
1213 me -> stop_info.stack_ptr = me -> stack_end + 0x10;
1215 /* This is dubious, since we may be more than a page into the stack, */
1216 /* and hence skip some of it, though it's not clear that matters. */
1218 me -> backing_store_end = (ptr_t)
1219 (GC_save_regs_in_stack() & ~(GC_page_size - 1));
1220 /* This is also < 100% convincing. We should also read this */
1221 /* from /proc, but the hook to do so isn't there yet. */
1224 start = si -> start_routine;
1225 # ifdef DEBUG_THREADS
1226 GC_printf1("start_routine = 0x%lx\n", start);
1228 start_arg = si -> arg;
1229 sem_post(&(si -> registered)); /* Last action on si. */
1230 /* OK to deallocate. */
1231 pthread_cleanup_push(GC_thread_exit_proc, 0);
1232 # if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
1234 GC_init_thread_local(me);
1237 result = (*start)(start_arg);
1239 GC_printf1("Finishing thread 0x%x\n", pthread_self());
1241 me -> status = result;
1242 pthread_cleanup_pop(1);
1243 /* Cleanup acquires lock, ensuring that we can't exit */
1244 /* while a collection that thinks we're alive is trying to stop */
1250 WRAP_FUNC(pthread_create)(pthread_t *new_thread,
1251 const pthread_attr_t *attr,
1252 void *(*start_routine)(void *), void *arg)
1257 struct start_info * si;
1258 /* This is otherwise saved only in an area mmapped by the thread */
1259 /* library, which isn't visible to the collector. */
1261 /* We resist the temptation to muck with the stack size here, */
1262 /* even if the default is unreasonably small. That's the client's */
1263 /* responsibility. */
1266 si = (struct start_info *)GC_INTERNAL_MALLOC(sizeof(struct start_info),
1269 if (!parallel_initialized) GC_init_parallel();
1270 if (0 == si) return(ENOMEM);
1271 sem_init(&(si -> registered), 0, 0);
1272 si -> start_routine = start_routine;
1275 if (!GC_thr_initialized) GC_thr_init();
1276 # ifdef GC_ASSERTIONS
1280 pthread_attr_t my_attr;
1281 pthread_attr_init(&my_attr);
1282 pthread_attr_getstacksize(&my_attr, &stack_size);
1284 pthread_attr_getstacksize(attr, &stack_size);
1286 GC_ASSERT(stack_size >= (8*HBLKSIZE*sizeof(word)));
1287 /* Our threads may need to do some work for the GC. */
1288 /* Ridiculously small threads won't work, and they */
1289 /* probably wouldn't work anyway. */
1293 detachstate = PTHREAD_CREATE_JOINABLE;
1295 pthread_attr_getdetachstate(attr, &detachstate);
1297 if (PTHREAD_CREATE_DETACHED == detachstate) my_flags |= DETACHED;
1298 si -> flags = my_flags;
1300 # ifdef DEBUG_THREADS
1301 GC_printf1("About to start new thread from thread 0x%X\n",
1305 result = REAL_FUNC(pthread_create)(new_thread, attr, GC_start_routine, si);
1307 # ifdef DEBUG_THREADS
1308 GC_printf1("Started thread 0x%X\n", *new_thread);
1310 /* Wait until child has been added to the thread table. */
1311 /* This also ensures that we hold onto si until the child is done */
1312 /* with it. Thus it doesn't matter whether it is otherwise */
1313 /* visible to the collector. */
1315 while (0 != sem_wait(&(si -> registered))) {
1316 if (EINTR != errno) ABORT("sem_wait failed");
1319 sem_destroy(&(si -> registered));
1321 GC_INTERNAL_FREE(si);
1327 #ifdef GENERIC_COMPARE_AND_SWAP
1328 pthread_mutex_t GC_compare_and_swap_lock = PTHREAD_MUTEX_INITIALIZER;
1330 GC_bool GC_compare_and_exchange(volatile GC_word *addr,
1331 GC_word old, GC_word new_val)
1334 pthread_mutex_lock(&GC_compare_and_swap_lock);
1341 pthread_mutex_unlock(&GC_compare_and_swap_lock);
1345 GC_word GC_atomic_add(volatile GC_word *addr, GC_word how_much)
1348 pthread_mutex_lock(&GC_compare_and_swap_lock);
1350 *addr = old + how_much;
1351 pthread_mutex_unlock(&GC_compare_and_swap_lock);
1355 #endif /* GENERIC_COMPARE_AND_SWAP */
1356 /* Spend a few cycles in a way that can't introduce contention with */
1357 /* othre threads. */
1361 # if !defined(__GNUC__) || defined(__INTEL_COMPILER)
1362 volatile word dummy = 0;
1365 for (i = 0; i < 10; ++i) {
1366 # if defined(__GNUC__) && !defined(__INTEL_COMPILER)
1367 __asm__ __volatile__ (" " : : : "memory");
1369 /* Something that's unlikely to be optimized away. */
1375 #define SPIN_MAX 128 /* Maximum number of calls to GC_pause before */
1378 VOLATILE GC_bool GC_collecting = 0;
1379 /* A hint that we're in the collector and */
1380 /* holding the allocation lock for an */
1381 /* extended period. */
1383 #if !defined(USE_SPIN_LOCK) || defined(PARALLEL_MARK)
1384 /* If we don't want to use the below spinlock implementation, either */
1385 /* because we don't have a GC_test_and_set implementation, or because */
1386 /* we don't want to risk sleeping, we can still try spinning on */
1387 /* pthread_mutex_trylock for a while. This appears to be very */
1388 /* beneficial in many cases. */
1389 /* I suspect that under high contention this is nearly always better */
1390 /* than the spin lock. But it's a bit slower on a uniprocessor. */
1391 /* Hence we still default to the spin lock. */
1392 /* This is also used to acquire the mark lock for the parallel */
1395 /* Here we use a strict exponential backoff scheme. I don't know */
1396 /* whether that's better or worse than the above. We eventually */
1397 /* yield by calling pthread_mutex_lock(); it never makes sense to */
1398 /* explicitly sleep. */
1402 unsigned long GC_spin_count = 0;
1403 unsigned long GC_block_count = 0;
1404 unsigned long GC_unlocked_count = 0;
1407 void GC_generic_lock(pthread_mutex_t * lock)
1409 #ifndef NO_PTHREAD_TRYLOCK
1410 unsigned pause_length = 1;
1413 if (0 == pthread_mutex_trylock(lock)) {
1415 ++GC_unlocked_count;
1419 for (; pause_length <= SPIN_MAX; pause_length <<= 1) {
1420 for (i = 0; i < pause_length; ++i) {
1423 switch(pthread_mutex_trylock(lock)) {
1432 ABORT("Unexpected error from pthread_mutex_trylock");
1435 #endif /* !NO_PTHREAD_TRYLOCK */
1439 pthread_mutex_lock(lock);
1442 #endif /* !USE_SPIN_LOCK || PARALLEL_MARK */
1444 #if defined(USE_SPIN_LOCK)
1446 /* Reasonably fast spin locks. Basically the same implementation */
1447 /* as STL alloc.h. This isn't really the right way to do this. */
1448 /* but until the POSIX scheduling mess gets straightened out ... */
1450 volatile unsigned int GC_allocate_lock = 0;
1455 # define low_spin_max 30 /* spin cycles if we suspect uniprocessor */
1456 # define high_spin_max SPIN_MAX /* spin cycles for multiprocessor */
1457 static unsigned spin_max = low_spin_max;
1458 unsigned my_spin_max;
1459 static unsigned last_spins = 0;
1460 unsigned my_last_spins;
1463 if (!GC_test_and_set(&GC_allocate_lock)) {
1466 my_spin_max = spin_max;
1467 my_last_spins = last_spins;
1468 for (i = 0; i < my_spin_max; i++) {
1469 if (GC_collecting || GC_nprocs == 1) goto yield;
1470 if (i < my_last_spins/2 || GC_allocate_lock) {
1474 if (!GC_test_and_set(&GC_allocate_lock)) {
1477 * Spinning worked. Thus we're probably not being scheduled
1478 * against the other process with which we were contending.
1479 * Thus it makes sense to spin longer the next time.
1482 spin_max = high_spin_max;
1486 /* We are probably being scheduled against the other process. Sleep. */
1487 spin_max = low_spin_max;
1490 if (!GC_test_and_set(&GC_allocate_lock)) {
1493 # define SLEEP_THRESHOLD 12
1494 /* Under Linux very short sleeps tend to wait until */
1495 /* the current time quantum expires. On old Linux */
1496 /* kernels nanosleep(<= 2ms) just spins under Linux. */
1497 /* (Under 2.4, this happens only for real-time */
1498 /* processes.) We want to minimize both behaviors */
1500 if (i < SLEEP_THRESHOLD) {
1506 /* Don't wait for more than about 15msecs, even */
1507 /* under extreme contention. */
1509 ts.tv_nsec = 1 << i;
1515 #else /* !USE_SPINLOCK */
1518 #ifndef NO_PTHREAD_TRYLOCK
1519 if (1 == GC_nprocs || GC_collecting) {
1520 pthread_mutex_lock(&GC_allocate_ml);
1522 GC_generic_lock(&GC_allocate_ml);
1524 #else /* !NO_PTHREAD_TRYLOCK */
1525 pthread_mutex_lock(&GC_allocate_ml);
1526 #endif /* !NO_PTHREAD_TRYLOCK */
1529 #endif /* !USE_SPINLOCK */
1531 #if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
1533 #ifdef GC_ASSERTIONS
1534 pthread_t GC_mark_lock_holder = NO_THREAD;
1538 /* Ugly workaround for a linux threads bug in the final versions */
1539 /* of glibc2.1. Pthread_mutex_trylock sets the mutex owner */
1540 /* field even when it fails to acquire the mutex. This causes */
1541 /* pthread_cond_wait to die. Remove for glibc2.2. */
1542 /* According to the man page, we should use */
1543 /* PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP, but that isn't actually */
1545 static pthread_mutex_t mark_mutex =
1546 {0, 0, 0, PTHREAD_MUTEX_ERRORCHECK_NP, {0, 0}};
1548 static pthread_mutex_t mark_mutex = PTHREAD_MUTEX_INITIALIZER;
1551 static pthread_cond_t builder_cv = PTHREAD_COND_INITIALIZER;
1553 void GC_acquire_mark_lock()
1556 if (pthread_mutex_lock(&mark_mutex) != 0) {
1557 ABORT("pthread_mutex_lock failed");
1560 GC_generic_lock(&mark_mutex);
1561 # ifdef GC_ASSERTIONS
1562 GC_mark_lock_holder = pthread_self();
1566 void GC_release_mark_lock()
1568 GC_ASSERT(GC_mark_lock_holder == pthread_self());
1569 # ifdef GC_ASSERTIONS
1570 GC_mark_lock_holder = NO_THREAD;
1572 if (pthread_mutex_unlock(&mark_mutex) != 0) {
1573 ABORT("pthread_mutex_unlock failed");
1577 /* Collector must wait for a freelist builders for 2 reasons: */
1578 /* 1) Mark bits may still be getting examined without lock. */
1579 /* 2) Partial free lists referenced only by locals may not be scanned */
1580 /* correctly, e.g. if they contain "pointer-free" objects, since the */
1581 /* free-list link may be ignored. */
1582 void GC_wait_builder()
1584 GC_ASSERT(GC_mark_lock_holder == pthread_self());
1585 # ifdef GC_ASSERTIONS
1586 GC_mark_lock_holder = NO_THREAD;
1588 if (pthread_cond_wait(&builder_cv, &mark_mutex) != 0) {
1589 ABORT("pthread_cond_wait failed");
1591 GC_ASSERT(GC_mark_lock_holder == NO_THREAD);
1592 # ifdef GC_ASSERTIONS
1593 GC_mark_lock_holder = pthread_self();
1597 void GC_wait_for_reclaim()
1599 GC_acquire_mark_lock();
1600 while (GC_fl_builder_count > 0) {
1603 GC_release_mark_lock();
1606 void GC_notify_all_builder()
1608 GC_ASSERT(GC_mark_lock_holder == pthread_self());
1609 if (pthread_cond_broadcast(&builder_cv) != 0) {
1610 ABORT("pthread_cond_broadcast failed");
1614 #endif /* PARALLEL_MARK || THREAD_LOCAL_ALLOC */
1616 #ifdef PARALLEL_MARK
1618 static pthread_cond_t mark_cv = PTHREAD_COND_INITIALIZER;
1620 void GC_wait_marker()
1622 GC_ASSERT(GC_mark_lock_holder == pthread_self());
1623 # ifdef GC_ASSERTIONS
1624 GC_mark_lock_holder = NO_THREAD;
1626 if (pthread_cond_wait(&mark_cv, &mark_mutex) != 0) {
1627 ABORT("pthread_cond_wait failed");
1629 GC_ASSERT(GC_mark_lock_holder == NO_THREAD);
1630 # ifdef GC_ASSERTIONS
1631 GC_mark_lock_holder = pthread_self();
1635 void GC_notify_all_marker()
1637 if (pthread_cond_broadcast(&mark_cv) != 0) {
1638 ABORT("pthread_cond_broadcast failed");
1642 #endif /* PARALLEL_MARK */
1644 # endif /* GC_LINUX_THREADS and friends */