[BACK]Return to linux_threads.c CVS log [TXT][DIR] Up to [local] / OpenXM_contrib2 / asir2000 / gc

Diff for /OpenXM_contrib2/asir2000/gc/Attic/linux_threads.c between version 1.3 and 1.4

version 1.3, 2000/12/01 09:26:11 version 1.4, 2001/04/20 07:39:18
Line 2 
Line 2 
  * Copyright (c) 1994 by Xerox Corporation.  All rights reserved.   * Copyright (c) 1994 by Xerox Corporation.  All rights reserved.
  * Copyright (c) 1996 by Silicon Graphics.  All rights reserved.   * Copyright (c) 1996 by Silicon Graphics.  All rights reserved.
  * Copyright (c) 1998 by Fergus Henderson.  All rights reserved.   * Copyright (c) 1998 by Fergus Henderson.  All rights reserved.
    * Copyright (c) 2000-2001 by Hewlett-Packard Company.  All rights reserved.
  *   *
  * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED   * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
  * OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.   * OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
Line 17 
Line 18 
  * thread package for Linux which is included in libc6.   * thread package for Linux which is included in libc6.
  *   *
  * This code relies on implementation details of LinuxThreads,   * This code relies on implementation details of LinuxThreads,
  * (i.e. properties not guaranteed by the Pthread standard):   * (i.e. properties not guaranteed by the Pthread standard),
    * though this version now does less of that than the other Pthreads
    * support code.
  *   *
  *      - the function GC_linux_thread_top_of_stack(void)  
  *        relies on the way LinuxThreads lays out thread stacks  
  *        in the address space.  
  *  
  * Note that there is a lot of code duplication between linux_threads.c   * Note that there is a lot of code duplication between linux_threads.c
  * and irix_threads.c; any changes made here may need to be reflected   * and thread support for some of the other Posix platforms; any changes
  * there too.   * made here may need to be reflected there too.
  */   */
   /*
    * Linux_threads.c now also includes some code to support HPUX and
    * OSF1 (Compaq Tru64 Unix, really).  The OSF1 support is not yet
    * functional.  The OSF1 code is based on Eric Benson's
    * patch, though that was originally against hpux_irix_threads.  The code
    * here is completely untested.  With 0.0000001% probability, it might
    * actually work.
    *
    * Eric also suggested an alternate basis for a lock implementation in
    * his code:
    * + #elif defined(OSF1)
    * +    unsigned long GC_allocate_lock = 0;
    * +    msemaphore GC_allocate_semaphore;
    * + #  define GC_TRY_LOCK() \
    * +    ((msem_lock(&GC_allocate_semaphore, MSEM_IF_NOWAIT) == 0) \
    * +     ? (GC_allocate_lock = 1) \
    * +     : 0)
    * + #  define GC_LOCK_TAKEN GC_allocate_lock
    */
   
 /* #define DEBUG_THREADS 1 */  /* #define DEBUG_THREADS 1 */
   
 /* ANSI C requires that a compilation unit contains something */  /* ANSI C requires that a compilation unit contains something */
 # include "gc_priv.h"  
   
 # if defined(LINUX_THREADS)  # if defined(GC_LINUX_THREADS) || defined(LINUX_THREADS) \
        || defined(GC_HPUX_THREADS) || defined(HPUX_THREADS) \
        || defined(GC_OSF1_THREADS) || defined(OSF1_THREADS) \
   
   # include "private/gc_priv.h"
   # ifdef THREAD_LOCAL_ALLOC
   #   if !defined(USE_PTHREAD_SPECIFIC) && !defined(USE_HPUX_TLS)
   #     include "private/specific.h"
   #   endif
   #   if defined(USE_PTHREAD_SPECIFIC)
   #     define GC_getspecific pthread_getspecific
   #     define GC_setspecific pthread_setspecific
   #     define GC_key_create pthread_key_create
         typedef pthread_key_t GC_key_t;
   #   endif
   #   if defined(USE_HPUX_TLS)
   #     define GC_getspecific(x) (x)
   #     define GC_setspecific(key, v) ((key) = (v), 0)
   #     define GC_key_create(key, d) 0
         typedef void * GC_key_t;
   #   endif
   # endif
   # include <stdlib.h>
 # include <pthread.h>  # include <pthread.h>
 # include <sched.h>  # include <sched.h>
 # include <time.h>  # include <time.h>
Line 44 
Line 82 
 # include <sys/time.h>  # include <sys/time.h>
 # include <semaphore.h>  # include <semaphore.h>
 # include <signal.h>  # include <signal.h>
   # include <sys/types.h>
   # include <sys/stat.h>
   # include <fcntl.h>
   
 #ifdef USE_LD_WRAP  #ifndef __GNUC__
   #   define __inline__
   #endif
   
   #ifdef GC_USE_LD_WRAP
 #   define WRAP_FUNC(f) __wrap_##f  #   define WRAP_FUNC(f) __wrap_##f
 #   define REAL_FUNC(f) __real_##f  #   define REAL_FUNC(f) __real_##f
 #else  #else
Line 54 
Line 99 
 #   undef pthread_create  #   undef pthread_create
 #   undef pthread_sigmask  #   undef pthread_sigmask
 #   undef pthread_join  #   undef pthread_join
   #   undef pthread_detach
 #endif  #endif
   
   
Line 75  void GC_print_sig_mask()
Line 121  void GC_print_sig_mask()
 }  }
 #endif  #endif
   
   
 /* We use the allocation lock to protect thread-related data structures. */  /* We use the allocation lock to protect thread-related data structures. */
   
 /* The set of all known threads.  We intercept thread creation and      */  /* The set of all known threads.  We intercept thread creation and      */
 /* joins.  We never actually create detached threads.  We allocate all  */  /* joins.                                                               */
 /* new thread stacks ourselves.  These allow us to maintain this        */  /* Protected by allocation/GC lock.                                     */
 /* data structure.                                                      */  /* Some of this should be declared volatile, but that's inconsistent    */
 /* Protected by GC_thr_lock.                                            */  
 /* Some of this should be declared volatile, but that's incosnsistent   */  
 /* with some library routine declarations.                              */  /* with some library routine declarations.                              */
 typedef struct GC_Thread_Rep {  typedef struct GC_Thread_Rep {
     struct GC_Thread_Rep * next;  /* More recently allocated threads    */      struct GC_Thread_Rep * next;  /* More recently allocated threads    */
Line 91  typedef struct GC_Thread_Rep {
Line 136  typedef struct GC_Thread_Rep {
                                   /* guaranteed to be dead, but we may  */                                    /* guaranteed to be dead, but we may  */
                                   /* not yet have registered the join.) */                                    /* not yet have registered the join.) */
     pthread_t id;      pthread_t id;
     word flags;      short flags;
 #       define FINISHED 1       /* Thread has exited.   */  #       define FINISHED 1       /* Thread has exited.   */
 #       define DETACHED 2       /* Thread is intended to be detached.   */  #       define DETACHED 2       /* Thread is intended to be detached.   */
 #       define MAIN_THREAD 4    /* True for the original thread only.   */  #       define MAIN_THREAD 4    /* True for the original thread only.   */
       short thread_blocked;       /* Protected by GC lock.                */
                                   /* Treated as a boolean value.  If set, */
                                   /* thread will acquire GC lock before   */
                                   /* doing any pointer manipulations, and */
                                   /* has set its sp value.  Thus it does  */
                                   /* not need to be sent a signal to stop */
                                   /* it.                                  */
     ptr_t stack_end;            /* Cold end of the stack.               */      ptr_t stack_end;            /* Cold end of the stack.               */
     ptr_t stack_ptr;            /* Valid only when stopped.             */      ptr_t stack_ptr;            /* Valid only when stopped.             */
 #   ifdef IA64  #   ifdef IA64
Line 107  typedef struct GC_Thread_Rep {
Line 158  typedef struct GC_Thread_Rep {
                                 /* Used only to avoid premature         */                                  /* Used only to avoid premature         */
                                 /* reclamation of any data it might     */                                  /* reclamation of any data it might     */
                                 /* reference.                           */                                  /* reference.                           */
   #   ifdef THREAD_LOCAL_ALLOC
   #       if CPP_WORDSZ == 64 && defined(ALIGN_DOUBLE)
   #           define GRANULARITY 16
   #           define NFREELISTS 48
   #       else
   #           define GRANULARITY 8
   #           define NFREELISTS 64
   #       endif
           /* The ith free list corresponds to size (i+1)*GRANULARITY */
   #       define INDEX_FROM_BYTES(n) (ADD_SLOP(n) - 1)/GRANULARITY
   #       define BYTES_FROM_INDEX(i) (((i) + 1) * GRANULARITY - EXTRA_BYTES)
   #       define SMALL_ENOUGH(bytes) (ADD_SLOP(bytes) <= NFREELISTS*GRANULARITY)
           ptr_t ptrfree_freelists[NFREELISTS];
           ptr_t normal_freelists[NFREELISTS];
   #       ifdef GC_GCJ_SUPPORT
             ptr_t gcj_freelists[NFREELISTS];
   #       endif
                   /* Free lists contain either a pointer or a small count */
                   /* reflecting the number of granules allocated at that  */
                   /* size.                                                */
                   /* 0 ==> thread-local allocation in use, free list      */
                   /*       empty.                                         */
                   /* > 0, <= DIRECT_GRANULES ==> Using global allocation, */
                   /*       too few objects of this size have been         */
                   /*       allocated by this thread.                      */
                   /* >= HBLKSIZE  => pointer to nonempty free list.       */
                   /* > DIRECT_GRANULES, < HBLKSIZE ==> transition to      */
                   /*    local alloc, equivalent to 0.                     */
   #       define DIRECT_GRANULES (HBLKSIZE/GRANULARITY)
                   /* Don't use local free lists for up to this much       */
                   /* allocation.                                          */
   #   endif
 } * GC_thread;  } * GC_thread;
   
 GC_thread GC_lookup_thread(pthread_t id);  GC_thread GC_lookup_thread(pthread_t id);
   
   static GC_bool fully_initialized = FALSE;
   
   # if defined(__GNUC__)
       void GC_full_init() __attribute__ ((constructor));
   # else
       void GC_full_init();
   # endif
   
   # if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
   
   /* We don't really support thread-local allocation with DBG_HDRS_ALL */
   
   #ifdef USE_HPUX_TLS
     __thread
   #endif
   GC_key_t GC_thread_key;
   
   static GC_bool keys_initialized;
   
   /* Recover the contents of the freelist array fl into the global one gfl.*/
   /* Note that the indexing scheme differs, in that gfl has finer size    */
   /* resolution, even if not all entries are used.                        */
   /* We hold the allocator lock.                                          */
   static void return_freelists(ptr_t *fl, ptr_t *gfl)
   {
       int i;
       ptr_t q, *qptr;
       size_t nwords;
   
       for (i = 0; i < NFREELISTS; ++i) {
           nwords = (i + 1) * (GRANULARITY/sizeof(word));
           qptr = fl + i;
           q = *qptr;
           if ((word)q < HBLKSIZE) continue;
           if (gfl[nwords] == 0) {
               gfl[nwords] = q;
           } else {
               /* Concatenate: */
               for (; (word)q >= HBLKSIZE; qptr = &(obj_link(q)), q = *qptr);
               GC_ASSERT(0 == q);
               *qptr = gfl[nwords];
               gfl[nwords] = fl[i];
           }
           /* Clear fl[i], since the thread structure may hang around.     */
           /* Do it in a way that is likely to trap if we access it.       */
           fl[i] = (ptr_t)HBLKSIZE;
       }
   }
   
   /* Each thread structure must be initialized.   */
   /* This call must be made from the new thread.  */
   /* Caller holds allocation lock.                */
   void GC_init_thread_local(GC_thread p)
   {
       int i;
   
       if (!keys_initialized) {
           if (0 != GC_key_create(&GC_thread_key, 0)) {
               ABORT("Failed to create key for local allocator");
           }
           keys_initialized = TRUE;
       }
       if (0 != GC_setspecific(GC_thread_key, p)) {
           ABORT("Failed to set thread specific allocation pointers");
       }
       for (i = 0; i < NFREELISTS; ++i) {
           p -> ptrfree_freelists[i] = (ptr_t)1;
           p -> normal_freelists[i] = (ptr_t)1;
   #       ifdef GC_GCJ_SUPPORT
             p -> gcj_freelists[i] = (ptr_t)1;
   #       endif
       }
   }
   
   #ifdef GC_GCJ_SUPPORT
     extern ptr_t * GC_gcjobjfreelist;
   #endif
   
   /* We hold the allocator lock.  */
   void GC_destroy_thread_local(GC_thread p)
   {
       /* We currently only do this from the thread itself.        */
           GC_ASSERT(GC_getspecific(GC_thread_key) == (void *)p);
       return_freelists(p -> ptrfree_freelists, GC_aobjfreelist);
       return_freelists(p -> normal_freelists, GC_objfreelist);
   #   ifdef GC_GCJ_SUPPORT
           return_freelists(p -> gcj_freelists, GC_gcjobjfreelist);
   #   endif
   }
   
   extern GC_PTR GC_generic_malloc_many();
   
   GC_PTR GC_local_malloc(size_t bytes)
   {
       if (EXPECT(!SMALL_ENOUGH(bytes),0)) {
           return(GC_malloc(bytes));
       } else {
           int index = INDEX_FROM_BYTES(bytes);
           ptr_t * my_fl;
           ptr_t my_entry;
           GC_key_t k = GC_thread_key;
           void * tsd;
   
   #       if defined(REDIRECT_MALLOC) && !defined(USE_PTHREAD_SPECIFIC) \
              || !defined(__GNUC__)
               if (EXPECT(0 == k, 0)) {
                   /* This can happen if we get called when the world is   */
                   /* being initialized.  Whether we can actually complete */
                   /* the initialization then is unclear.                  */
                   GC_full_init();
                   k = GC_thread_key;
               }
   #       endif
           tsd = GC_getspecific(GC_thread_key);
   #       ifdef GC_ASSERTIONS
             LOCK();
             GC_ASSERT(tsd == (void *)GC_lookup_thread(pthread_self()));
             UNLOCK();
   #       endif
           my_fl = ((GC_thread)tsd) -> normal_freelists + index;
           my_entry = *my_fl;
           if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
               ptr_t next = obj_link(my_entry);
               GC_PTR result = (GC_PTR)my_entry;
               *my_fl = next;
               obj_link(my_entry) = 0;
               PREFETCH_FOR_WRITE(next);
               return result;
           } else if ((word)my_entry - 1 < DIRECT_GRANULES) {
               *my_fl = my_entry + index + 1;
               return GC_malloc(bytes);
           } else {
               my_entry = GC_generic_malloc_many(BYTES_FROM_INDEX(index),
                                                 NORMAL);
               *my_fl = my_entry;
               if (my_entry == 0) return GC_oom_fn(bytes);
               return GC_local_malloc(bytes);
           }
       }
   }
   
   GC_PTR GC_local_malloc_atomic(size_t bytes)
   {
       if (EXPECT(!SMALL_ENOUGH(bytes), 0)) {
           return(GC_malloc_atomic(bytes));
       } else {
           int index = INDEX_FROM_BYTES(bytes);
           ptr_t * my_fl = ((GC_thread)GC_getspecific(GC_thread_key))
                           -> ptrfree_freelists + index;
           ptr_t my_entry = *my_fl;
           if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
               GC_PTR result = (GC_PTR)my_entry;
               *my_fl = obj_link(my_entry);
               return result;
           } else if ((word)my_entry - 1 < DIRECT_GRANULES) {
               *my_fl = my_entry + index + 1;
               return GC_malloc_atomic(bytes);
           } else {
               my_entry = GC_generic_malloc_many(BYTES_FROM_INDEX(index),
                                                 PTRFREE);
               *my_fl = my_entry;
               if (my_entry == 0) return GC_oom_fn(bytes);
               return GC_local_malloc_atomic(bytes);
           }
       }
   }
   
   #ifdef GC_GCJ_SUPPORT
   
   #include "include/gc_gcj.h"
   
   #ifdef GC_ASSERTIONS
     extern GC_bool GC_gcj_malloc_initialized;
   #endif
   
   extern int GC_gcj_kind;
   
   GC_PTR GC_local_gcj_malloc(size_t bytes,
                              void * ptr_to_struct_containing_descr)
   {
       GC_ASSERT(GC_gcj_malloc_initialized);
       if (EXPECT(!SMALL_ENOUGH(bytes), 0)) {
           return GC_gcj_malloc(bytes, ptr_to_struct_containing_descr);
       } else {
           int index = INDEX_FROM_BYTES(bytes);
           ptr_t * my_fl = ((GC_thread)GC_getspecific(GC_thread_key))
                           -> gcj_freelists + index;
           ptr_t my_entry = *my_fl;
           if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
               GC_PTR result = (GC_PTR)my_entry;
               GC_ASSERT(!GC_incremental);
               /* We assert that any concurrent marker will stop us.       */
               /* Thus it is impossible for a mark procedure to see the    */
               /* allocation of the next object, but to see this object    */
               /* still containing a free list pointer.  Otherwise the     */
               /* marker might find a random "mark descriptor".            */
               *my_fl = obj_link(my_entry);
               *(void **)result = ptr_to_struct_containing_descr;
               return result;
           } else if ((word)my_entry - 1 < DIRECT_GRANULES) {
               *my_fl = my_entry + index + 1;
               return GC_gcj_malloc(bytes, ptr_to_struct_containing_descr);
           } else {
               my_entry = GC_generic_malloc_many(BYTES_FROM_INDEX(index),
                                                 GC_gcj_kind);
               *my_fl = my_entry;
               if (my_entry == 0) return GC_oom_fn(bytes);
               return GC_gcj_malloc(bytes, ptr_to_struct_containing_descr);
           }
       }
   }
   
   #endif /* GC_GCJ_SUPPORT */
   
   # else  /* !THREAD_LOCAL_ALLOC  && !DBG_HDRS_ALL */
   
   #   define GC_destroy_thread_local(t)
   
   # endif /* !THREAD_LOCAL_ALLOC */
   
 /*  /*
  * The only way to suspend threads given the pthread interface is to send   * The only way to suspend threads given the pthread interface is to send
  * signals.  We can't use SIGSTOP directly, because we need to get the   * signals.  We can't use SIGSTOP directly, because we need to get the
Line 121  GC_thread GC_lookup_thread(pthread_t id);
Line 424  GC_thread GC_lookup_thread(pthread_t id);
  * so we need to reuse something else.  I chose SIGPWR.   * so we need to reuse something else.  I chose SIGPWR.
  * (Perhaps SIGUNUSED would be a better choice.)   * (Perhaps SIGUNUSED would be a better choice.)
  */   */
 #define SIG_SUSPEND SIGPWR  #ifndef SIG_SUSPEND
   #  if defined(HPUX_THREADS) || defined(GC_OSF1_THREADS)
   #   define SIG_SUSPEND _SIGRTMIN + 6
   #  else
   #   define SIG_SUSPEND SIGPWR
   #  endif
   #endif
   
 #define SIG_RESTART SIGXCPU  #ifndef SIG_THR_RESTART
   #  if defined(HPUX_THREADS) || defined(GC_OSF1_THREADS)
   #   define SIG_THR_RESTART _SIGRTMIN + 5
   #  else
   #   define SIG_THR_RESTART SIGXCPU
   #  endif
   #endif
   
   /* SPARC/Linux doesn't properly define SIGPWR in <signal.h>.
    * It is aliased to SIGLOST in asm/signal.h, though.            */
   #if defined(SPARC) && !defined(SIGPWR)
   #   define SIGPWR SIGLOST
   #endif
   
 sem_t GC_suspend_ack_sem;  sem_t GC_suspend_ack_sem;
   
   #if !defined(HPUX_THREADS) && !defined(GC_OSF1_THREADS)
 /*  /*
 GC_linux_thread_top_of_stack() relies on implementation details of  
 LinuxThreads, namely that thread stacks are allocated on 2M boundaries  
 and grow to no more than 2M.  
 To make sure that we're using LinuxThreads and not some other thread  To make sure that we're using LinuxThreads and not some other thread
 package, we generate a dummy reference to `pthread_kill_other_threads_np'  package, we generate a dummy reference to `pthread_kill_other_threads_np'
 (was `__pthread_initial_thread_bos' but that disappeared),  (was `__pthread_initial_thread_bos' but that disappeared),
Line 138  which is a symbol defined in LinuxThreads, but (hopefu
Line 457  which is a symbol defined in LinuxThreads, but (hopefu
 thread packages.  thread packages.
 */  */
 void (*dummy_var_to_force_linux_threads)() = pthread_kill_other_threads_np;  void (*dummy_var_to_force_linux_threads)() = pthread_kill_other_threads_np;
   #endif /* !HPUX_THREADS */
   
 #define LINUX_THREADS_STACK_SIZE  (2 * 1024 * 1024)  #if defined(SPARC) || defined(IA64)
     extern word GC_save_regs_in_stack();
   #endif
   
 static inline ptr_t GC_linux_thread_top_of_stack(void)  long GC_nprocs = 1;     /* Number of processors.  We may not have       */
                           /* access to all of them, but this is as good   */
                           /* a guess as any ...                           */
   
   #ifdef PARALLEL_MARK
   
   # ifndef MAX_MARKERS
   #   define MAX_MARKERS 16
   # endif
   
   static ptr_t marker_sp[MAX_MARKERS] = {0};
   
   void * GC_mark_thread(void * id)
 {  {
   char *sp = GC_approx_sp();    word my_mark_no = 0;
   ptr_t tos = (ptr_t) (((unsigned long)sp | (LINUX_THREADS_STACK_SIZE - 1)) + 1);  
 #if DEBUG_THREADS    marker_sp[(word)id] = GC_approx_sp();
   GC_printf1("SP = %lx\n", (unsigned long)sp);    for (;; ++my_mark_no) {
   GC_printf1("TOS = %lx\n", (unsigned long)tos);      /* GC_mark_no is passed only to allow GC_help_marker to terminate   */
 #endif      /* promptly.  This is important if it were called from the signal   */
   return tos;      /* handler or from the GC lock acquisition code.  Under Linux, it's */
       /* not safe to call it from a signal handler, since it uses mutexes */
       /* and condition variables.  Since it is called only here, the      */
       /* argument is unnecessary.                                         */
       if (my_mark_no < GC_mark_no || my_mark_no > GC_mark_no + 2) {
           /* resynchronize if we get far off, e.g. because GC_mark_no     */
           /* wrapped.                                                     */
           my_mark_no = GC_mark_no;
       }
   #   ifdef DEBUG_THREADS
           GC_printf1("Starting mark helper for mark number %ld\n", my_mark_no);
   #   endif
       GC_help_marker(my_mark_no);
     }
 }  }
   
 #ifdef IA64  extern long GC_markers;         /* Number of mark threads we would      */
   extern word GC_save_regs_in_stack();                                  /* like to have.  Includes the          */
 #endif                                  /* initiating thread.                   */
   
   pthread_t GC_mark_threads[MAX_MARKERS];
   
   #define PTHREAD_CREATE REAL_FUNC(pthread_create)
   
   static void start_mark_threads()
   {
       unsigned i;
       pthread_attr_t attr;
   
       if (GC_markers > MAX_MARKERS) {
           WARN("Limiting number of mark threads\n", 0);
           GC_markers = MAX_MARKERS;
       }
       if (0 != pthread_attr_init(&attr)) ABORT("pthread_attr_init failed");
   
       if (0 != pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED))
           ABORT("pthread_attr_setdetachstate failed");
   #   ifdef CONDPRINT
         if (GC_print_stats) {
           GC_printf1("Starting %ld marker threads\n", GC_markers - 1);
         }
   #   endif
       for (i = 0; i < GC_markers - 1; ++i) {
         if (0 != PTHREAD_CREATE(GC_mark_threads + i, &attr,
                                 GC_mark_thread, (void *)(word)i)) {
           WARN("Marker thread creation failed, errno = %ld.\n", errno);
         }
       }
   }
   
   #else  /* !PARALLEL_MARK */
   
   static __inline__ void start_mark_threads()
   {
   }
   
   #endif /* !PARALLEL_MARK */
   
 void GC_suspend_handler(int sig)  void GC_suspend_handler(int sig)
 {  {
     int dummy;      int dummy;
Line 165  void GC_suspend_handler(int sig)
Line 550  void GC_suspend_handler(int sig)
     sigset_t old_sigs;      sigset_t old_sigs;
     int i;      int i;
     sigset_t mask;      sigset_t mask;
   #   ifdef PARALLEL_MARK
           word my_mark_no = GC_mark_no;
           /* Marker can't proceed until we acknowledge.  Thus this is     */
           /* guaranteed to be the mark_no correspending to our            */
           /* suspension, i.e. the marker can't have incremented it yet.   */
   #   endif
   
     if (sig != SIG_SUSPEND) ABORT("Bad signal in suspend_handler");      if (sig != SIG_SUSPEND) ABORT("Bad signal in suspend_handler");
   
Line 177  void GC_suspend_handler(int sig)
Line 568  void GC_suspend_handler(int sig)
     /* of a thread which holds the allocation lock in order     */      /* of a thread which holds the allocation lock in order     */
     /* to stop the world.  Thus concurrent modification of the  */      /* to stop the world.  Thus concurrent modification of the  */
     /* data structure is impossible.                            */      /* data structure is impossible.                            */
     me -> stack_ptr = (ptr_t)(&dummy);  #   ifdef SPARC
           me -> stack_ptr = (ptr_t)GC_save_regs_in_stack();
   #   else
           me -> stack_ptr = (ptr_t)(&dummy);
   #   endif
 #   ifdef IA64  #   ifdef IA64
         me -> backing_store_ptr = (ptr_t)GC_save_regs_in_stack();          me -> backing_store_ptr = (ptr_t)GC_save_regs_in_stack();
 #   endif  #   endif
Line 188  void GC_suspend_handler(int sig)
Line 583  void GC_suspend_handler(int sig)
     sem_post(&GC_suspend_ack_sem);      sem_post(&GC_suspend_ack_sem);
   
     /* Wait until that thread tells us to restart by sending    */      /* Wait until that thread tells us to restart by sending    */
     /* this thread a SIG_RESTART signal.                        */      /* this thread a SIG_THR_RESTART signal.                    */
     /* SIG_RESTART should be masked at this point.  Thus there  */      /* SIG_THR_RESTART should be masked at this point.  Thus there      */
     /* is no race.                                              */      /* is no race.                                              */
     if (sigfillset(&mask) != 0) ABORT("sigfillset() failed");      if (sigfillset(&mask) != 0) ABORT("sigfillset() failed");
     if (sigdelset(&mask, SIG_RESTART) != 0) ABORT("sigdelset() failed");      if (sigdelset(&mask, SIG_THR_RESTART) != 0) ABORT("sigdelset() failed");
 #   ifdef NO_SIGNALS  #   ifdef NO_SIGNALS
       if (sigdelset(&mask, SIGINT) != 0) ABORT("sigdelset() failed");        if (sigdelset(&mask, SIGINT) != 0) ABORT("sigdelset() failed");
       if (sigdelset(&mask, SIGQUIT) != 0) ABORT("sigdelset() failed");        if (sigdelset(&mask, SIGQUIT) != 0) ABORT("sigdelset() failed");
       if (sigdelset(&mask, SIGTERM) != 0) ABORT("sigdelset() failed");        if (sigdelset(&mask, SIGTERM) != 0) ABORT("sigdelset() failed");
         if (sigdelset(&mask, SIGABRT) != 0) ABORT("sigdelset() failed");
 #   endif  #   endif
     do {      do {
             me->signal = 0;              me->signal = 0;
             sigsuspend(&mask);             /* Wait for signal */              sigsuspend(&mask);             /* Wait for signal */
     } while (me->signal != SIG_RESTART);      } while (me->signal != SIG_THR_RESTART);
   
 #if DEBUG_THREADS  #if DEBUG_THREADS
     GC_printf1("Continuing 0x%x\n", my_thread);      GC_printf1("Continuing 0x%x\n", my_thread);
Line 212  void GC_restart_handler(int sig)
Line 608  void GC_restart_handler(int sig)
 {  {
     GC_thread me;      GC_thread me;
   
     if (sig != SIG_RESTART) ABORT("Bad signal in suspend_handler");      if (sig != SIG_THR_RESTART) ABORT("Bad signal in suspend_handler");
   
     /* Let the GC_suspend_handler() know that we got a SIG_RESTART. */      /* Let the GC_suspend_handler() know that we got a SIG_THR_RESTART. */
     /* The lookup here is safe, since I'm doing this on behalf  */      /* The lookup here is safe, since I'm doing this on behalf  */
     /* of a thread which holds the allocation lock in order     */      /* of a thread which holds the allocation lock in order     */
     /* to stop the world.  Thus concurrent modification of the  */      /* to stop the world.  Thus concurrent modification of the  */
     /* data structure is impossible.                            */      /* data structure is impossible.                            */
     me = GC_lookup_thread(pthread_self());      me = GC_lookup_thread(pthread_self());
     me->signal = SIG_RESTART;      me->signal = SIG_THR_RESTART;
   
     /*      /*
     ** Note: even if we didn't do anything useful here,      ** Note: even if we didn't do anything useful here,
Line 235  void GC_restart_handler(int sig)
Line 631  void GC_restart_handler(int sig)
 #endif  #endif
 }  }
   
   /* Defining INSTALL_LOOPING_SEGV_HANDLER causes SIGSEGV and SIGBUS to   */
   /* result in an infinite loop in a signal handler.  This can be very    */
   /* useful for debugging, since (as of RH7) gdb still seems to have      */
   /* serious problems with threads.                                       */
   #ifdef INSTALL_LOOPING_SEGV_HANDLER
   void GC_looping_handler(int sig)
   {
       GC_printf3("Signal %ld in thread %lx, pid %ld\n",
                  sig, pthread_self(), getpid());
       for (;;);
   }
   #endif
   
 GC_bool GC_thr_initialized = FALSE;  GC_bool GC_thr_initialized = FALSE;
   
 # define THREAD_TABLE_SZ 128    /* Must be power of 2   */  # define THREAD_TABLE_SZ 128    /* Must be power of 2   */
 volatile GC_thread GC_threads[THREAD_TABLE_SZ];  volatile GC_thread GC_threads[THREAD_TABLE_SZ];
   
   void GC_push_thread_structures GC_PROTO((void))
   {
       GC_push_all((ptr_t)(GC_threads), (ptr_t)(GC_threads)+sizeof(GC_threads));
   }
   
 /* Add a thread to GC_threads.  We assume it wasn't already there.      */  /* Add a thread to GC_threads.  We assume it wasn't already there.      */
 /* Caller holds allocation lock.                                        */  /* Caller holds allocation lock.                                        */
 GC_thread GC_new_thread(pthread_t id)  GC_thread GC_new_thread(pthread_t id)
Line 252  GC_thread GC_new_thread(pthread_t id)
Line 666  GC_thread GC_new_thread(pthread_t id)
     if (!first_thread_used) {      if (!first_thread_used) {
         result = &first_thread;          result = &first_thread;
         first_thread_used = TRUE;          first_thread_used = TRUE;
         /* Dont acquire allocation lock, since we may already hold it. */  
     } else {      } else {
         result = (struct GC_Thread_Rep *)          result = (struct GC_Thread_Rep *)
                  GC_generic_malloc_inner(sizeof(struct GC_Thread_Rep), NORMAL);                   GC_INTERNAL_MALLOC(sizeof(struct GC_Thread_Rep), NORMAL);
     }      }
     if (result == 0) return(0);      if (result == 0) return(0);
     result -> id = id;      result -> id = id;
     result -> next = GC_threads[hv];      result -> next = GC_threads[hv];
     GC_threads[hv] = result;      GC_threads[hv] = result;
     /* result -> flags = 0; */      GC_ASSERT(result -> flags == 0 && result -> thread_blocked == 0);
     return(result);      return(result);
 }  }
   
Line 283  void GC_delete_thread(pthread_t id)
Line 696  void GC_delete_thread(pthread_t id)
     } else {      } else {
         prev -> next = p -> next;          prev -> next = p -> next;
     }      }
       GC_INTERNAL_FREE(p);
 }  }
   
 /* If a thread has been joined, but we have not yet             */  /* If a thread has been joined, but we have not yet             */
Line 304  void GC_delete_gc_thread(pthread_t id, GC_thread gc_id
Line 718  void GC_delete_gc_thread(pthread_t id, GC_thread gc_id
     } else {      } else {
         prev -> next = p -> next;          prev -> next = p -> next;
     }      }
       GC_INTERNAL_FREE(p);
 }  }
   
 /* Return a GC_thread corresponding to a given thread_t.        */  /* Return a GC_thread corresponding to a given thread_t.        */
Line 330  void GC_stop_world()
Line 745  void GC_stop_world()
     register int n_live_threads = 0;      register int n_live_threads = 0;
     register int result;      register int result;
   
       /* Make sure all free list construction has stopped before we start. */
       /* No new construction can start, since free list construction is   */
       /* required to acquire and release the GC lock before it starts,    */
       /* and we have the lock.                                            */
   #   ifdef PARALLEL_MARK
         GC_acquire_mark_lock();
         GC_ASSERT(GC_fl_builder_count == 0);
         /* We should have previously waited for it to become zero. */
   #   endif /* PARALLEL_MARK */
     for (i = 0; i < THREAD_TABLE_SZ; i++) {      for (i = 0; i < THREAD_TABLE_SZ; i++) {
       for (p = GC_threads[i]; p != 0; p = p -> next) {        for (p = GC_threads[i]; p != 0; p = p -> next) {
         if (p -> id != my_thread) {          if (p -> id != my_thread) {
             if (p -> flags & FINISHED) continue;              if (p -> flags & FINISHED) continue;
               if (p -> thread_blocked) /* Will wait */ continue;
             n_live_threads++;              n_live_threads++;
             #if DEBUG_THREADS              #if DEBUG_THREADS
               GC_printf1("Sending suspend signal to 0x%x\n", p -> id);                GC_printf1("Sending suspend signal to 0x%x\n", p -> id);
Line 353  void GC_stop_world()
Line 778  void GC_stop_world()
       }        }
     }      }
     for (i = 0; i < n_live_threads; i++) {      for (i = 0; i < n_live_threads; i++) {
         sem_wait(&GC_suspend_ack_sem);          if (0 != sem_wait(&GC_suspend_ack_sem))
               ABORT("sem_wait in handler failed");
     }      }
   #   ifdef PARALLEL_MARK
         GC_release_mark_lock();
   #   endif
     #if DEBUG_THREADS      #if DEBUG_THREADS
     GC_printf1("World stopped 0x%x\n", pthread_self());        GC_printf1("World stopped 0x%x\n", pthread_self());
     #endif      #endif
 }  }
   
 /* Caller holds allocation lock.        */  /* Caller holds allocation lock, and has held it continuously since     */
   /* the world stopped.                                                   */
 void GC_start_world()  void GC_start_world()
 {  {
     pthread_t my_thread = pthread_self();      pthread_t my_thread = pthread_self();
Line 377  void GC_start_world()
Line 807  void GC_start_world()
       for (p = GC_threads[i]; p != 0; p = p -> next) {        for (p = GC_threads[i]; p != 0; p = p -> next) {
         if (p -> id != my_thread) {          if (p -> id != my_thread) {
             if (p -> flags & FINISHED) continue;              if (p -> flags & FINISHED) continue;
               if (p -> thread_blocked) continue;
             n_live_threads++;              n_live_threads++;
             #if DEBUG_THREADS              #if DEBUG_THREADS
               GC_printf1("Sending restart signal to 0x%x\n", p -> id);                GC_printf1("Sending restart signal to 0x%x\n", p -> id);
             #endif              #endif
             result = pthread_kill(p -> id, SIG_RESTART);              result = pthread_kill(p -> id, SIG_THR_RESTART);
             switch(result) {              switch(result) {
                 case ESRCH:                  case ESRCH:
                     /* Not really there anymore.  Possible? */                      /* Not really there anymore.  Possible? */
Line 425  void GC_push_all_stacks()
Line 856  void GC_push_all_stacks()
       for (p = GC_threads[i]; p != 0; p = p -> next) {        for (p = GC_threads[i]; p != 0; p = p -> next) {
         if (p -> flags & FINISHED) continue;          if (p -> flags & FINISHED) continue;
         if (pthread_equal(p -> id, me)) {          if (pthread_equal(p -> id, me)) {
             lo = GC_approx_sp();  #           ifdef SPARC
                   lo = (ptr_t)GC_save_regs_in_stack();
   #           else
                   lo = GC_approx_sp();
   #           endif
             IF_IA64(bs_hi = (ptr_t)GC_save_regs_in_stack();)              IF_IA64(bs_hi = (ptr_t)GC_save_regs_in_stack();)
         } else {          } else {
             lo = p -> stack_ptr;              lo = p -> stack_ptr;
Line 445  void GC_push_all_stacks()
Line 880  void GC_push_all_stacks()
                 (unsigned long) lo, (unsigned long) hi);                  (unsigned long) lo, (unsigned long) hi);
         #endif          #endif
         if (0 == lo) ABORT("GC_push_all_stacks: sp not set!\n");          if (0 == lo) ABORT("GC_push_all_stacks: sp not set!\n");
         GC_push_all_stack(lo, hi);  #       ifdef STACK_GROWS_UP
             /* We got them backwards! */
             GC_push_all_stack(hi, lo);
   #       else
             GC_push_all_stack(lo, hi);
   #       endif
 #       ifdef IA64  #       ifdef IA64
           if (pthread_equal(p -> id, me)) {            if (pthread_equal(p -> id, me)) {
             GC_push_all_eager(bs_lo, bs_hi);              GC_push_all_eager(bs_lo, bs_hi);
Line 457  void GC_push_all_stacks()
Line 897  void GC_push_all_stacks()
     }      }
 }  }
   
   #ifdef USE_PROC_FOR_LIBRARIES
   int GC_segment_is_thread_stack(ptr_t lo, ptr_t hi)
   {
       int i;
       GC_thread p;
   
   #   ifdef PARALLEL_MARK
         for (i = 0; i < GC_markers; ++i) {
           if (marker_sp[i] > lo & marker_sp[i] < hi) return 1;
         }
   #   endif
       for (i = 0; i < THREAD_TABLE_SZ; i++) {
         for (p = GC_threads[i]; p != 0; p = p -> next) {
           if (0 != p -> stack_end) {
   #         ifdef STACK_GROWS_UP
               if (p -> stack_end >= lo && p -> stack_end < hi) return 1;
   #         else /* STACK_GROWS_DOWN */
               if (p -> stack_end > lo && p -> stack_end <= hi) return 1;
   #         endif
           }
         }
       }
       return 0;
   }
   #endif /* USE_PROC_FOR_LIBRARIES */
   
   #ifdef LINUX_THREADS
   /* Return the number of processors, or i<= 0 if it can't be determined. */
   int GC_get_nprocs()
   {
       /* Should be "return sysconf(_SC_NPROCESSORS_ONLN);" but that       */
       /* appears to be buggy in many cases.                               */
       /* We look for lines "cpu<n>" in /proc/stat.                        */
   #   define STAT_BUF_SIZE 4096
   #   if defined(GC_USE_LD_WRAP)
   #       define STAT_READ __real_read
   #   else
   #       define STAT_READ read
   #   endif
       char stat_buf[STAT_BUF_SIZE];
       int f;
       char c;
       word result = 1;
           /* Some old kernels only have a single "cpu nnnn ..."   */
           /* entry in /proc/stat.  We identify those as           */
           /* uniprocessors.                                       */
       size_t i, len = 0;
   
       f = open("/proc/stat", O_RDONLY);
       if (f < 0 || (len = STAT_READ(f, stat_buf, STAT_BUF_SIZE)) < 100) {
           WARN("Couldn't read /proc/stat\n", 0);
           return -1;
       }
       for (i = 0; i < len - 100; ++i) {
           if (stat_buf[i] == '\n' && stat_buf[i+1] == 'c'
               && stat_buf[i+2] == 'p' && stat_buf[i+3] == 'u') {
               int cpu_no = atoi(stat_buf + i + 4);
               if (cpu_no >= result) result = cpu_no + 1;
           }
       }
       return result;
   }
   #endif /* LINUX_THREADS */
   
 /* We hold the allocation lock. */  /* We hold the allocation lock. */
 void GC_thr_init()  void GC_thr_init()
 {  {
Line 478  void GC_thr_init()
Line 981  void GC_thr_init()
 #   ifdef NO_SIGNALS  #   ifdef NO_SIGNALS
       if (sigdelset(&act.sa_mask, SIGINT) != 0        if (sigdelset(&act.sa_mask, SIGINT) != 0
           || sigdelset(&act.sa_mask, SIGQUIT != 0)            || sigdelset(&act.sa_mask, SIGQUIT != 0)
             || sigdelset(&act.sa_mask, SIGABRT != 0)
           || sigdelset(&act.sa_mask, SIGTERM != 0)) {            || sigdelset(&act.sa_mask, SIGTERM != 0)) {
         ABORT("sigdelset() failed");          ABORT("sigdelset() failed");
       }        }
 #   endif  #   endif
   
     /* SIG_RESTART is unmasked by the handler when necessary.   */      /* SIG_THR_RESTART is unmasked by the handler when necessary.       */
     act.sa_handler = GC_suspend_handler;      act.sa_handler = GC_suspend_handler;
     if (sigaction(SIG_SUSPEND, &act, NULL) != 0) {      if (sigaction(SIG_SUSPEND, &act, NULL) != 0) {
         ABORT("Cannot set SIG_SUSPEND handler");          ABORT("Cannot set SIG_SUSPEND handler");
     }      }
   
     act.sa_handler = GC_restart_handler;      act.sa_handler = GC_restart_handler;
     if (sigaction(SIG_RESTART, &act, NULL) != 0) {      if (sigaction(SIG_THR_RESTART, &act, NULL) != 0) {
         ABORT("Cannot set SIG_SUSPEND handler");          ABORT("Cannot set SIG_THR_RESTART handler");
     }      }
   #   ifdef INSTALL_LOOPING_SEGV_HANDLER
           act.sa_handler = GC_looping_handler;
           if (sigaction(SIGSEGV, &act, NULL) != 0
               || sigaction(SIGBUS, &act, NULL) != 0) {
               ABORT("Cannot set SIGSEGV or SIGBUS looping handler");
           }
   #   endif  /* INSTALL_LOOPING_SEGV_HANDLER */
   
     /* Add the initial thread, so we can stop it.       */      /* Add the initial thread, so we can stop it.       */
       t = GC_new_thread(pthread_self());        t = GC_new_thread(pthread_self());
       t -> stack_ptr = (ptr_t)(&dummy);        t -> stack_ptr = (ptr_t)(&dummy);
       t -> flags = DETACHED | MAIN_THREAD;        t -> flags = DETACHED | MAIN_THREAD;
   
       /* Set GC_nprocs.  */
         {
           char * nprocs_string = GETENV("GC_NPROCS");
           GC_nprocs = -1;
           if (nprocs_string != NULL) GC_nprocs = atoi(nprocs_string);
         }
         if (GC_nprocs <= 0) {
   #       if defined(HPUX_THREADS)
             GC_nprocs = pthread_num_processors_np();
   #       endif
   #       if defined(OSF1_THREADS)
             GC_nprocs = 1;
   #       endif
   #       ifdef LINUX_THREADS
             GC_nprocs = GC_get_nprocs();
   #       endif
         }
         if (GC_nprocs <= 0) {
           WARN("GC_get_nprocs() returned %ld\n", GC_nprocs);
           GC_nprocs = 2;
   #       ifdef PARALLEL_MARK
             GC_markers = 1;
   #       endif
         } else {
   #       ifdef PARALLEL_MARK
             GC_markers = GC_nprocs;
   #       endif
         }
   #   ifdef PARALLEL_MARK
   #     ifdef CONDPRINT
           if (GC_print_stats) {
             GC_printf2("Number of processors = %ld, "
                    "number of marker threads = %ld\n", GC_nprocs, GC_markers);
           }
   #     endif
         if (GC_markers == 1) {
           GC_parallel = FALSE;
   #       ifdef CONDPRINT
             if (GC_print_stats) {
               GC_printf0("Single marker thread, turning off parallel marking\n");
             }
   #       endif
         } else {
           GC_parallel = TRUE;
         }
   #   endif
 }  }
   
   
   /* Perform all initializations, including those that    */
   /* may require allocation.                              */
   /* Called as constructor without allocation lock.       */
   /* Must be called before a second thread is created.    */
   void GC_full_init()
   {
       if (fully_initialized) return;
       if (!GC_is_initialized) GC_init();
       /* If we are using a parallel marker, start the helper threads.  */
   #     ifdef PARALLEL_MARK
           if (GC_parallel) start_mark_threads();
   #     endif
       /* Initialize thread local free lists if used.      */
   #   if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
         LOCK();
         GC_init_thread_local(GC_lookup_thread(pthread_self()));
         UNLOCK();
   #   endif
       fully_initialized = TRUE;
   }
   
   
 int WRAP_FUNC(pthread_sigmask)(int how, const sigset_t *set, sigset_t *oset)  int WRAP_FUNC(pthread_sigmask)(int how, const sigset_t *set, sigset_t *oset)
 {  {
     sigset_t fudged_set;      sigset_t fudged_set;
Line 512  int WRAP_FUNC(pthread_sigmask)(int how, const sigset_t
Line 1093  int WRAP_FUNC(pthread_sigmask)(int how, const sigset_t
     return(REAL_FUNC(pthread_sigmask)(how, set, oset));      return(REAL_FUNC(pthread_sigmask)(how, set, oset));
 }  }
   
   /* Wrappers for functions that are likely to block for an appreciable   */
   /* length of time.  Must be called in pairs, if at all.                 */
   /* Nothing much beyond the system call itself should be executed        */
   /* between these.                                                       */
   
   void GC_start_blocking(void) {
   #   define SP_SLOP 128
       GC_thread me;
       LOCK();
       me = GC_lookup_thread(pthread_self());
       GC_ASSERT(!(me -> thread_blocked));
   #   ifdef SPARC
           me -> stack_ptr = (ptr_t)GC_save_regs_in_stack();
   #   else
           me -> stack_ptr = (ptr_t)GC_approx_sp();
   #   endif
   #   ifdef IA64
           me -> backing_store_ptr = (ptr_t)GC_save_regs_in_stack() + SP_SLOP;
   #   endif
       /* Add some slop to the stack pointer, since the wrapped call may   */
       /* end up pushing more callee-save registers.                       */
   #   ifdef STACK_GROWS_UP
           me -> stack_ptr += SP_SLOP;
   #   else
           me -> stack_ptr -= SP_SLOP;
   #   endif
       me -> thread_blocked = TRUE;
       UNLOCK();
   }
   
   GC_end_blocking(void) {
       GC_thread me;
       LOCK();   /* This will block if the world is stopped.       */
       me = GC_lookup_thread(pthread_self());
       GC_ASSERT(me -> thread_blocked);
       me -> thread_blocked = FALSE;
       UNLOCK();
   }
   
   /* A wrapper for the standard C sleep function  */
   int WRAP_FUNC(sleep) (unsigned int seconds)
   {
       int result;
   
       GC_start_blocking();
       result = REAL_FUNC(sleep)(seconds);
       GC_end_blocking();
       return result;
   }
   
 struct start_info {  struct start_info {
     void *(*start_routine)(void *);      void *(*start_routine)(void *);
     void *arg;      void *arg;
Line 520  struct start_info {
Line 1151  struct start_info {
                                 /* parent hasn't yet noticed.           */                                  /* parent hasn't yet noticed.           */
 };  };
   
   /* Called at thread exit.                               */
   /* Never called for main thread.  That's OK, since it   */
   /* results in at most a tiny one-time leak.  And        */
   /* linuxthreads doesn't reclaim the main threads        */
   /* resources or id anyway.                              */
 void GC_thread_exit_proc(void *arg)  void GC_thread_exit_proc(void *arg)
 {  {
     GC_thread me;      GC_thread me;
     struct start_info * si = arg;  
   
     LOCK();      LOCK();
     me = GC_lookup_thread(pthread_self());      me = GC_lookup_thread(pthread_self());
       GC_destroy_thread_local(me);
     if (me -> flags & DETACHED) {      if (me -> flags & DETACHED) {
         GC_delete_thread(pthread_self());          GC_delete_thread(pthread_self());
     } else {      } else {
         me -> flags |= FINISHED;          me -> flags |= FINISHED;
     }      }
   #   if defined(THREAD_LOCAL_ALLOC) && !defined(USE_PTHREAD_SPECIFIC) \
          && !defined(USE_HPUX_TLS) && !defined(DBG_HDRS_ALL)
         GC_remove_specific(GC_thread_key);
   #   endif
     if (GC_incremental && GC_collection_in_progress()) {      if (GC_incremental && GC_collection_in_progress()) {
         int old_gc_no = GC_gc_no;          int old_gc_no = GC_gc_no;
   
Line 562  int WRAP_FUNC(pthread_join)(pthread_t thread, void **r
Line 1201  int WRAP_FUNC(pthread_join)(pthread_t thread, void **r
     /* cant have been recycled by pthreads.                             */      /* cant have been recycled by pthreads.                             */
     UNLOCK();      UNLOCK();
     result = REAL_FUNC(pthread_join)(thread, retval);      result = REAL_FUNC(pthread_join)(thread, retval);
       if (result == 0) {
           LOCK();
           /* Here the pthread thread id may have been recycled. */
           GC_delete_gc_thread(thread, thread_gc_id);
           UNLOCK();
       }
       return result;
   }
   
   int
   WRAP_FUNC(pthread_detach)(pthread_t thread)
   {
       int result;
       GC_thread thread_gc_id;
   
     LOCK();      LOCK();
     /* Here the pthread thread id may have been recycled. */      thread_gc_id = GC_lookup_thread(thread);
     GC_delete_gc_thread(thread, thread_gc_id);  
     UNLOCK();      UNLOCK();
       result = REAL_FUNC(pthread_detach)(thread);
       if (result == 0) {
         LOCK();
         thread_gc_id -> flags |= DETACHED;
         /* Here the pthread thread id may have been recycled. */
         if (thread_gc_id -> flags & FINISHED) {
           GC_delete_gc_thread(thread, thread_gc_id);
         }
         UNLOCK();
       }
     return result;      return result;
 }  }
   
Line 600  void * GC_start_routine(void * arg)
Line 1263  void * GC_start_routine(void * arg)
         /* Needs to be plausible, since an asynchronous stack mark      */          /* Needs to be plausible, since an asynchronous stack mark      */
         /* should not crash.                                            */          /* should not crash.                                            */
 #   else  #   else
       me -> stack_end = (ptr_t)(((word)(&dummy) & ~(GC_page_size - 1));        me -> stack_end = (ptr_t)((word)(&dummy) & ~(GC_page_size - 1));
       me -> stack_ptr = me -> stack_end + 0x10;        me -> stack_ptr = me -> stack_end + 0x10;
 #   endif  #   endif
     /* This is dubious, since we may be more than a page into the stack, */      /* This is dubious, since we may be more than a page into the stack, */
Line 617  void * GC_start_routine(void * arg)
Line 1280  void * GC_start_routine(void * arg)
         GC_printf1("start_routine = 0x%lx\n", start);          GC_printf1("start_routine = 0x%lx\n", start);
 #   endif  #   endif
     start_arg = si -> arg;      start_arg = si -> arg;
     sem_post(&(si -> registered));      sem_post(&(si -> registered));      /* Last action on si.   */
     pthread_cleanup_push(GC_thread_exit_proc, si);                                          /* OK to deallocate.    */
       pthread_cleanup_push(GC_thread_exit_proc, 0);
   #   if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
           LOCK();
           GC_init_thread_local(me);
           UNLOCK();
   #   endif
     result = (*start)(start_arg);      result = (*start)(start_arg);
 #if DEBUG_THREADS  #if DEBUG_THREADS
         GC_printf1("Finishing thread 0x%x\n", pthread_self());          GC_printf1("Finishing thread 0x%x\n", pthread_self());
Line 640  WRAP_FUNC(pthread_create)(pthread_t *new_thread,
Line 1309  WRAP_FUNC(pthread_create)(pthread_t *new_thread,
     int result;      int result;
     GC_thread t;      GC_thread t;
     pthread_t my_new_thread;      pthread_t my_new_thread;
     void * stack;  
     size_t stacksize;  
     pthread_attr_t new_attr;  
     int detachstate;      int detachstate;
     word my_flags = 0;      word my_flags = 0;
     struct start_info * si = GC_malloc(sizeof(struct start_info));      struct start_info * si;
         /* This is otherwise saved only in an area mmapped by the thread */          /* This is otherwise saved only in an area mmapped by the thread */
         /* library, which isn't visible to the collector.                */          /* library, which isn't visible to the collector.                */
   
       LOCK();
       si = (struct start_info *)GC_INTERNAL_MALLOC(sizeof(struct start_info), NORMAL);
       UNLOCK();
       if (!fully_initialized) GC_full_init();
     if (0 == si) return(ENOMEM);      if (0 == si) return(ENOMEM);
     sem_init(&(si -> registered), 0, 0);      sem_init(&(si -> registered), 0, 0);
     si -> start_routine = start_routine;      si -> start_routine = start_routine;
Line 656  WRAP_FUNC(pthread_create)(pthread_t *new_thread,
Line 1326  WRAP_FUNC(pthread_create)(pthread_t *new_thread,
     LOCK();      LOCK();
     if (!GC_thr_initialized) GC_thr_init();      if (!GC_thr_initialized) GC_thr_init();
     if (NULL == attr) {      if (NULL == attr) {
         stack = 0;          detachstate = PTHREAD_CREATE_JOINABLE;
         (void) pthread_attr_init(&new_attr);  
     } else {      } else {
         new_attr = *attr;          pthread_attr_getdetachstate(attr, &detachstate);
     }      }
     pthread_attr_getdetachstate(&new_attr, &detachstate);  
     if (PTHREAD_CREATE_DETACHED == detachstate) my_flags |= DETACHED;      if (PTHREAD_CREATE_DETACHED == detachstate) my_flags |= DETACHED;
     si -> flags = my_flags;      si -> flags = my_flags;
     UNLOCK();      UNLOCK();
Line 669  WRAP_FUNC(pthread_create)(pthread_t *new_thread,
Line 1337  WRAP_FUNC(pthread_create)(pthread_t *new_thread,
         GC_printf1("About to start new thread from thread 0x%X\n",          GC_printf1("About to start new thread from thread 0x%X\n",
                    pthread_self());                     pthread_self());
 #   endif  #   endif
     result = REAL_FUNC(pthread_create)(new_thread, &new_attr, GC_start_routine, si);      result = REAL_FUNC(pthread_create)(new_thread, attr, GC_start_routine, si);
 #   ifdef DEBUG_THREADS  #   ifdef DEBUG_THREADS
         GC_printf1("Started thread 0x%X\n", *new_thread);          GC_printf1("Started thread 0x%X\n", *new_thread);
 #   endif  #   endif
Line 677  WRAP_FUNC(pthread_create)(pthread_t *new_thread,
Line 1345  WRAP_FUNC(pthread_create)(pthread_t *new_thread,
     /* This also ensures that we hold onto si until the child is done   */      /* This also ensures that we hold onto si until the child is done   */
     /* with it.  Thus it doesn't matter whether it is otherwise         */      /* with it.  Thus it doesn't matter whether it is otherwise         */
     /* visible to the collector.                                        */      /* visible to the collector.                                        */
         if (0 != sem_wait(&(si -> registered))) ABORT("sem_wait failed");          while (0 != sem_wait(&(si -> registered))) {
               if (EINTR != errno) ABORT("sem_wait failed");
           }
         sem_destroy(&(si -> registered));          sem_destroy(&(si -> registered));
     /* pthread_attr_destroy(&new_attr); */          LOCK();
     /* pthread_attr_destroy(&new_attr); */          GC_INTERNAL_FREE(si);
           UNLOCK();
     return(result);      return(result);
 }  }
   
 #if defined(USE_SPIN_LOCK)  #ifdef GENERIC_COMPARE_AND_SWAP
     pthread_mutex_t GC_compare_and_swap_lock = PTHREAD_MUTEX_INITIALIZER;
   
     GC_bool GC_compare_and_exchange(volatile GC_word *addr,
                                     GC_word old, GC_word new_val)
     {
       GC_bool result;
       pthread_mutex_lock(&GC_compare_and_swap_lock);
       if (*addr == old) {
         *addr = new_val;
         result = TRUE;
       } else {
         result = FALSE;
       }
       pthread_mutex_unlock(&GC_compare_and_swap_lock);
       return result;
     }
   
     GC_word GC_atomic_add(volatile GC_word *addr, GC_word how_much)
     {
       GC_word old;
       pthread_mutex_lock(&GC_compare_and_swap_lock);
       old = *addr;
       *addr = old + how_much;
       pthread_mutex_unlock(&GC_compare_and_swap_lock);
       return old;
     }
   
   #endif /* GENERIC_COMPARE_AND_SWAP */
   /* Spend a few cycles in a way that can't introduce contention with     */
   /* othre threads.                                                       */
   void GC_pause()
   {
       int i;
       volatile word dummy = 0;
   
       for (i = 0; i < 10; ++i) {
   #     ifdef __GNUC__
           __asm__ __volatile__ (" " : : : "memory");
   #     else
           /* Something that's unlikely to be optimized away. */
           GC_noop(++dummy);
   #     endif
       }
   }
   
   #define SPIN_MAX 1024   /* Maximum number of calls to GC_pause before   */
                           /* give up.                                     */
   
 VOLATILE GC_bool GC_collecting = 0;  VOLATILE GC_bool GC_collecting = 0;
                         /* A hint that we're in the collector and       */                          /* A hint that we're in the collector and       */
                         /* holding the allocation lock for an           */                          /* holding the allocation lock for an           */
                         /* extended period.                             */                          /* extended period.                             */
   
   #if !defined(USE_SPIN_LOCK) || defined(PARALLEL_MARK)
   /* If we don't want to use the below spinlock implementation, either    */
   /* because we don't have a GC_test_and_set implementation, or because   */
   /* we don't want to risk sleeping, we can still try spinning on         */
   /* pthread_mutex_trylock for a while.  This appears to be very          */
   /* beneficial in many cases.                                            */
   /* I suspect that under high contention this is nearly always better    */
   /* than the spin lock.  But it's a bit slower on a uniprocessor.        */
   /* Hence we still default to the spin lock.                             */
   /* This is also used to acquire the mark lock for the parallel          */
   /* marker.                                                              */
   
   /* Here we use a strict exponential backoff scheme.  I don't know       */
   /* whether that's better or worse than the above.  We eventually        */
   /* yield by calling pthread_mutex_lock(); it never makes sense to       */
   /* explicitly sleep.                                                    */
   
   void GC_generic_lock(pthread_mutex_t * lock)
   {
       unsigned pause_length = 1;
       unsigned i;
   
       if (0 == pthread_mutex_trylock(lock)) return;
       for (; pause_length <= SPIN_MAX; pause_length <<= 1) {
           for (i = 0; i < pause_length; ++i) {
               GC_pause();
           }
           switch(pthread_mutex_trylock(lock)) {
               case 0:
                   return;
               case EBUSY:
                   break;
               default:
                   ABORT("Unexpected error from pthread_mutex_trylock");
           }
       }
       pthread_mutex_lock(lock);
   }
   
   #endif /* !USE_SPIN_LOCK || PARALLEL_MARK */
   
   #if defined(USE_SPIN_LOCK)
   
 /* Reasonably fast spin locks.  Basically the same implementation */  /* Reasonably fast spin locks.  Basically the same implementation */
 /* as STL alloc.h.  This isn't really the right way to do this.   */  /* as STL alloc.h.  This isn't really the right way to do this.   */
 /* but until the POSIX scheduling mess gets straightened out ...  */  /* but until the POSIX scheduling mess gets straightened out ...  */
Line 701  volatile unsigned int GC_allocate_lock = 0;
Line 1462  volatile unsigned int GC_allocate_lock = 0;
 void GC_lock()  void GC_lock()
 {  {
 #   define low_spin_max 30  /* spin cycles if we suspect uniprocessor */  #   define low_spin_max 30  /* spin cycles if we suspect uniprocessor */
 #   define high_spin_max 1000 /* spin cycles for multiprocessor */  #   define high_spin_max SPIN_MAX /* spin cycles for multiprocessor */
     static unsigned spin_max = low_spin_max;      static unsigned spin_max = low_spin_max;
     unsigned my_spin_max;      unsigned my_spin_max;
     static unsigned last_spins = 0;      static unsigned last_spins = 0;
     unsigned my_last_spins;      unsigned my_last_spins;
     volatile unsigned junk;  
 #   define PAUSE junk *= junk; junk *= junk; junk *= junk; junk *= junk  
     int i;      int i;
   
     if (!GC_test_and_set(&GC_allocate_lock)) {      if (!GC_test_and_set(&GC_allocate_lock)) {
         return;          return;
     }      }
     junk = 0;  
     my_spin_max = spin_max;      my_spin_max = spin_max;
     my_last_spins = last_spins;      my_last_spins = last_spins;
     for (i = 0; i < my_spin_max; i++) {      for (i = 0; i < my_spin_max; i++) {
         if (GC_collecting) goto yield;          if (GC_collecting || GC_nprocs == 1) goto yield;
         if (i < my_last_spins/2 || GC_allocate_lock) {          if (i < my_last_spins/2 || GC_allocate_lock) {
             PAUSE;              GC_pause();
             continue;              continue;
         }          }
         if (!GC_test_and_set(&GC_allocate_lock)) {          if (!GC_test_and_set(&GC_allocate_lock)) {
Line 749  yield:
Line 1507  yield:
         } else {          } else {
             struct timespec ts;              struct timespec ts;
   
             if (i > 26) i = 26;              if (i > 24) i = 24;
                         /* Don't wait for more than about 60msecs, even */                          /* Don't wait for more than about 15msecs, even */
                         /* under extreme contention.                    */                          /* under extreme contention.                    */
             ts.tv_sec = 0;              ts.tv_sec = 0;
             ts.tv_nsec = 1 << i;              ts.tv_nsec = 1 << i;
Line 759  yield:
Line 1517  yield:
     }      }
 }  }
   
 #endif /* known architecture */  #else  /* !USE_SPINLOCK */
   
   void GC_lock()
   {
       if (1 == GC_nprocs || GC_collecting) {
           pthread_mutex_lock(&GC_allocate_ml);
       } else {
           GC_generic_lock(&GC_allocate_ml);
       }
   }
   
   #endif /* !USE_SPINLOCK */
   
   #ifdef PARALLEL_MARK
   
   #ifdef GC_ASSERTIONS
     pthread_t GC_mark_lock_holder = NO_THREAD;
   #endif
   
   #ifdef IA64
     /* Ugly workaround for a linux threads bug in the final versions      */
     /* of glibc2.1.  Pthread_mutex_trylock sets the mutex owner           */
     /* field even when it fails to acquire the mutex.  This causes        */
     /* pthread_cond_wait to die.  Remove for glibc2.2.                    */
     /* According to the man page, we should use                           */
     /* PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP, but that isn't actually   */
     /* defined.                                                           */
     static pthread_mutex_t mark_mutex =
           {0, 0, 0, PTHREAD_MUTEX_ERRORCHECK_NP, {0, 0}};
   #else
     static pthread_mutex_t mark_mutex = PTHREAD_MUTEX_INITIALIZER;
   #endif
   
   static pthread_cond_t mark_cv = PTHREAD_COND_INITIALIZER;
   
   static pthread_cond_t builder_cv = PTHREAD_COND_INITIALIZER;
   
   void GC_acquire_mark_lock()
   {
   /*
       if (pthread_mutex_lock(&mark_mutex) != 0) {
           ABORT("pthread_mutex_lock failed");
       }
   */
       GC_generic_lock(&mark_mutex);
   #   ifdef GC_ASSERTIONS
           GC_mark_lock_holder = pthread_self();
   #   endif
   }
   
   void GC_release_mark_lock()
   {
       GC_ASSERT(GC_mark_lock_holder == pthread_self());
   #   ifdef GC_ASSERTIONS
           GC_mark_lock_holder = NO_THREAD;
   #   endif
       if (pthread_mutex_unlock(&mark_mutex) != 0) {
           ABORT("pthread_mutex_unlock failed");
       }
   }
   
   void GC_wait_marker()
   {
       GC_ASSERT(GC_mark_lock_holder == pthread_self());
   #   ifdef GC_ASSERTIONS
           GC_mark_lock_holder = NO_THREAD;
   #   endif
       if (pthread_cond_wait(&mark_cv, &mark_mutex) != 0) {
           ABORT("pthread_cond_wait failed");
       }
       GC_ASSERT(GC_mark_lock_holder == NO_THREAD);
   #   ifdef GC_ASSERTIONS
           GC_mark_lock_holder = pthread_self();
   #   endif
   }
   
   void GC_wait_builder()
   {
       GC_ASSERT(GC_mark_lock_holder == pthread_self());
   #   ifdef GC_ASSERTIONS
           GC_mark_lock_holder = NO_THREAD;
   #   endif
       if (pthread_cond_wait(&builder_cv, &mark_mutex) != 0) {
           ABORT("pthread_cond_wait failed");
       }
       GC_ASSERT(GC_mark_lock_holder == NO_THREAD);
   #   ifdef GC_ASSERTIONS
           GC_mark_lock_holder = pthread_self();
   #   endif
   }
   
   void GC_notify_all_marker()
   {
       if (pthread_cond_broadcast(&mark_cv) != 0) {
           ABORT("pthread_cond_broadcast failed");
       }
   }
   
   void GC_notify_all_builder()
   {
       GC_ASSERT(GC_mark_lock_holder == pthread_self());
       if (pthread_cond_broadcast(&builder_cv) != 0) {
           ABORT("pthread_cond_broadcast failed");
       }
   }
   
   void GC_wait_for_reclaim()
   {
       GC_acquire_mark_lock();
       while (GC_fl_builder_count > 0) {
           GC_wait_builder();
       }
       GC_release_mark_lock();
   }
   #endif /* PARALLEL_MARK */
   
 # endif /* LINUX_THREADS */  # endif /* LINUX_THREADS */
   
Legend:
Removed from v.1.3  
changed lines
  Added in v.1.4
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>