/*
* Copyright 1988, 1989 Hans-J. Boehm, Alan J. Demers
* Copyright (c) 1991-1994 by Xerox Corporation.  All rights reserved.
* Copyright (c) 1999-2004 Hewlett-Packard Development Company, L.P.
*
* THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
* OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
*
* Permission is hereby granted to use or copy this program
* for any purpose,  provided the above notices are retained on all copies.
* Permission to modify the code and to distribute modified code is granted,
* provided the above notices are retained, and a notice that the code was
* modified is included with the above copyright notice.
*/

#include "private/gc_priv.h"
#include "gc_inline.h" /* for GC_malloc_kind */

#include <stdio.h>
#include <string.h>

/* Allocate reclaim list for kind:      */
/* Return TRUE on success               */
STATIC GC_bool GC_alloc_reclaim_list(struct obj_kind *kind)
{
   struct hblk ** result = (struct hblk **)
               GC_scratch_alloc((MAXOBJGRANULES+1) * sizeof(struct hblk *));
   if (result == 0) return(FALSE);
   BZERO(result, (MAXOBJGRANULES+1)*sizeof(struct hblk *));
   kind -> ok_reclaim_list = result;
   return(TRUE);
}

/* Allocate a large block of size lb bytes.  The block is not cleared.  */
/* flags argument should be 0 or IGNORE_OFF_PAGE.  EXTRA_BYTES value    */
/* was already added to lb.                                             */
GC_INNER ptr_t GC_alloc_large(size_t lb, int k, unsigned flags)
{
   struct hblk * h;
   word n_blocks;
   ptr_t result;
   GC_bool retry = FALSE;

   GC_ASSERT(I_HOLD_LOCK());
   lb = ROUNDUP_GRANULE_SIZE(lb);
   n_blocks = OBJ_SZ_TO_BLOCKS_CHECKED(lb);
   if (!EXPECT(GC_is_initialized, TRUE)) {
     DCL_LOCK_STATE;
     UNLOCK(); /* just to unset GC_lock_holder */
     GC_init();
     LOCK();
   }
   /* Do our share of marking work */
       if (GC_incremental && !GC_dont_gc) {
           ENTER_GC();
           GC_collect_a_little_inner((int)n_blocks);
           EXIT_GC();
       }
   h = GC_allochblk(lb, k, flags);
#   ifdef USE_MUNMAP
       if (0 == h) {
           GC_merge_unmapped();
           h = GC_allochblk(lb, k, flags);
       }
#   endif
   while (0 == h && GC_collect_or_expand(n_blocks, flags != 0, retry)) {
       h = GC_allochblk(lb, k, flags);
       retry = TRUE;
   }
   if (h == 0) {
       result = 0;
   } else {
       size_t total_bytes = n_blocks * HBLKSIZE;
       if (n_blocks > 1) {
           GC_large_allocd_bytes += total_bytes;
           if (GC_large_allocd_bytes > GC_max_large_allocd_bytes)
               GC_max_large_allocd_bytes = GC_large_allocd_bytes;
       }
       /* FIXME: Do we need some way to reset GC_max_large_allocd_bytes? */
       result = h -> hb_body;
   }
   return result;
}

/* Allocate a large block of size lb bytes.  Clear if appropriate.      */
/* EXTRA_BYTES were already added to lb.                                */
STATIC ptr_t GC_alloc_large_and_clear(size_t lb, int k, unsigned flags)
{
   ptr_t result;

   GC_ASSERT(I_HOLD_LOCK());
   result = GC_alloc_large(lb, k, flags);
   if (result != NULL
         && (GC_debugging_started || GC_obj_kinds[k].ok_init)) {
       word n_blocks = OBJ_SZ_TO_BLOCKS(lb);

       /* Clear the whole block, in case of GC_realloc call. */
       BZERO(result, n_blocks * HBLKSIZE);
   }
   return result;
}

/* Fill in additional entries in GC_size_map, including the i-th one.   */
/* Note that a filled in section of the array ending at n always        */
/* has the length of at least n/4.                                      */
STATIC void GC_extend_size_map(size_t i)
{
 size_t orig_granule_sz = ROUNDED_UP_GRANULES(i);
 size_t granule_sz;
 size_t byte_sz = GRANULES_TO_BYTES(orig_granule_sz);
                       /* The size we try to preserve.         */
                       /* Close to i, unless this would        */
                       /* introduce too many distinct sizes.   */
 size_t smaller_than_i = byte_sz - (byte_sz >> 3);
 size_t low_limit; /* The lowest indexed entry we initialize.  */
 size_t number_of_objs;

 GC_ASSERT(I_HOLD_LOCK());
 GC_ASSERT(0 == GC_size_map[i]);
 if (0 == GC_size_map[smaller_than_i]) {
   low_limit = byte_sz - (byte_sz >> 2); /* much smaller than i */
   granule_sz = orig_granule_sz;
   while (GC_size_map[low_limit] != 0)
     low_limit++;
 } else {
   low_limit = smaller_than_i + 1;
   while (GC_size_map[low_limit] != 0)
     low_limit++;

   granule_sz = ROUNDED_UP_GRANULES(low_limit);
   granule_sz += granule_sz >> 3;
   if (granule_sz < orig_granule_sz)
     granule_sz = orig_granule_sz;
 }

 /* For these larger sizes, we use an even number of granules.         */
 /* This makes it easier to, e.g., construct a 16-byte-aligned         */
 /* allocator even if GRANULE_BYTES is 8.                              */
 granule_sz = (granule_sz + 1) & ~(size_t)1;
 if (granule_sz > MAXOBJGRANULES)
   granule_sz = MAXOBJGRANULES;

 /* If we can fit the same number of larger objects in a block, do so. */
 number_of_objs = HBLK_GRANULES / granule_sz;
 GC_ASSERT(number_of_objs != 0);
 granule_sz = (HBLK_GRANULES / number_of_objs) & ~(size_t)1;

 byte_sz = GRANULES_TO_BYTES(granule_sz) - EXTRA_BYTES;
                       /* We may need one extra byte; do not always    */
                       /* fill in GC_size_map[byte_sz].                */

 for (; low_limit <= byte_sz; low_limit++)
   GC_size_map[low_limit] = granule_sz;
}

/* Allocate lb bytes for an object of kind k.           */
/* Should not be used to directly allocate objects      */
/* that require special handling on allocation.         */
GC_INNER void * GC_generic_malloc_inner(size_t lb, int k)
{
   void *op;

   GC_ASSERT(I_HOLD_LOCK());
   GC_ASSERT(k < MAXOBJKINDS);
   if (SMALL_OBJ(lb)) {
       struct obj_kind * kind = GC_obj_kinds + k;
       size_t lg = GC_size_map[lb];
       void ** opp = &(kind -> ok_freelist[lg]);

       op = *opp;
       if (EXPECT(0 == op, FALSE)) {
         if (lg == 0) {
           if (!EXPECT(GC_is_initialized, TRUE)) {
             DCL_LOCK_STATE;
             UNLOCK(); /* just to unset GC_lock_holder */
             GC_init();
             LOCK();
             lg = GC_size_map[lb];
           }
           if (0 == lg) {
             GC_extend_size_map(lb);
             lg = GC_size_map[lb];
             GC_ASSERT(lg != 0);
           }
           /* Retry */
           opp = &(kind -> ok_freelist[lg]);
           op = *opp;
         }
         if (0 == op) {
           if (0 == kind -> ok_reclaim_list &&
               !GC_alloc_reclaim_list(kind))
             return NULL;
           op = GC_allocobj(lg, k);
           if (0 == op)
             return NULL;
         }
       }
       *opp = obj_link(op);
       obj_link(op) = 0;
       GC_bytes_allocd += GRANULES_TO_BYTES((word)lg);
   } else {
       size_t lb_adjusted = ADD_SLOP(lb);

       op = (ptr_t)GC_alloc_large_and_clear(lb_adjusted, k, 0 /* flags */);
       if (op != NULL)
           GC_bytes_allocd += lb_adjusted;
   }

   return op;
}

#if defined(DBG_HDRS_ALL) || defined(GC_GCJ_SUPPORT) \
   || !defined(GC_NO_FINALIZATION)
 /* Allocate a composite object of size n bytes.  The caller           */
 /* guarantees that pointers past the first hblk are not relevant.     */
 GC_INNER void * GC_generic_malloc_inner_ignore_off_page(size_t lb, int k)
 {
   size_t lb_adjusted;
   void * op;

   GC_ASSERT(I_HOLD_LOCK());
   if (lb <= HBLKSIZE)
       return GC_generic_malloc_inner(lb, k);
   GC_ASSERT(k < MAXOBJKINDS);
   lb_adjusted = ADD_SLOP(lb);
   op = GC_alloc_large_and_clear(lb_adjusted, k, IGNORE_OFF_PAGE);
   if (op != NULL)
       GC_bytes_allocd += lb_adjusted;
   return op;
 }
#endif

#ifdef GC_COLLECT_AT_MALLOC
 /* Parameter to force GC at every malloc of size greater or equal to  */
 /* the given value.  This might be handy during debugging.            */
# if defined(CPPCHECK)
   size_t GC_dbg_collect_at_malloc_min_lb = 16*1024; /* e.g. */
# else
   size_t GC_dbg_collect_at_malloc_min_lb = (GC_COLLECT_AT_MALLOC);
# endif
#endif

GC_API GC_ATTR_MALLOC void * GC_CALL GC_generic_malloc(size_t lb, int k)
{
   void * result;
   DCL_LOCK_STATE;

   GC_ASSERT(k < MAXOBJKINDS);
   if (EXPECT(get_have_errors(), FALSE))
     GC_print_all_errors();
   GC_INVOKE_FINALIZERS();
   GC_DBG_COLLECT_AT_MALLOC(lb);
   if (SMALL_OBJ(lb)) {
       LOCK();
       result = GC_generic_malloc_inner(lb, k);
       UNLOCK();
   } else {
       size_t lg;
       size_t lb_rounded;
       word n_blocks;
       GC_bool init;

       lg = ROUNDED_UP_GRANULES(lb);
       lb_rounded = GRANULES_TO_BYTES(lg);
       n_blocks = OBJ_SZ_TO_BLOCKS(lb_rounded);
       init = GC_obj_kinds[k].ok_init;
       LOCK();
       result = (ptr_t)GC_alloc_large(lb_rounded, k, 0);
       if (0 != result) {
         if (GC_debugging_started
#             ifndef THREADS
               || init
#             endif
            ) {
           BZERO(result, n_blocks * HBLKSIZE);
         } else {
#           ifdef THREADS
             /* Clear any memory that might be used for GC descriptors */
             /* before we release the lock.                            */
               ((word *)result)[0] = 0;
               ((word *)result)[1] = 0;
               ((word *)result)[GRANULES_TO_WORDS(lg)-1] = 0;
               ((word *)result)[GRANULES_TO_WORDS(lg)-2] = 0;
#           endif
         }
         GC_bytes_allocd += lb_rounded;
       }
       UNLOCK();
#       ifdef THREADS
         if (init && !GC_debugging_started && result != NULL) {
           /* Clear the rest (i.e. excluding the initial 2 words). */
           BZERO((word *)result + 2,
                 n_blocks * HBLKSIZE - 2 * sizeof(word));
         }
#       endif
   }
   if (0 == result) {
       return((*GC_get_oom_fn())(lb));
   } else {
       return(result);
   }
}

GC_API GC_ATTR_MALLOC void * GC_CALL GC_malloc_kind_global(size_t lb, int k)
{
   GC_ASSERT(k < MAXOBJKINDS);
   if (SMALL_OBJ(lb)) {
       void *op;
       void **opp;
       size_t lg;
       DCL_LOCK_STATE;

       GC_DBG_COLLECT_AT_MALLOC(lb);
       LOCK();
       lg = GC_size_map[lb];
       opp = &GC_obj_kinds[k].ok_freelist[lg];
       op = *opp;
       if (EXPECT(op != NULL, TRUE)) {
           if (k == PTRFREE) {
               *opp = obj_link(op);
           } else {
               GC_ASSERT(0 == obj_link(op)
                         || ((word)obj_link(op)
                               <= (word)GC_greatest_plausible_heap_addr
                             && (word)obj_link(op)
                               >= (word)GC_least_plausible_heap_addr));
               *opp = obj_link(op);
               obj_link(op) = 0;
           }
           GC_bytes_allocd += GRANULES_TO_BYTES((word)lg);
           UNLOCK();
           return op;
       }
       UNLOCK();
   }

   /* We make the GC_clear_stack() call a tail one, hoping to get more */
   /* of the stack.                                                    */
   return GC_clear_stack(GC_generic_malloc(lb, k));
}

#if defined(THREADS) && !defined(THREAD_LOCAL_ALLOC)
 GC_API GC_ATTR_MALLOC void * GC_CALL GC_malloc_kind(size_t lb, int k)
 {
   return GC_malloc_kind_global(lb, k);
 }
#endif

/* Allocate lb bytes of atomic (pointer-free) data.     */
GC_API GC_ATTR_MALLOC void * GC_CALL GC_malloc_atomic(size_t lb)
{
   return GC_malloc_kind(lb, PTRFREE);
}

/* Allocate lb bytes of composite (pointerful) data.    */
GC_API GC_ATTR_MALLOC void * GC_CALL GC_malloc(size_t lb)
{
   return GC_malloc_kind(lb, NORMAL);
}

GC_API GC_ATTR_MALLOC void * GC_CALL GC_generic_malloc_uncollectable(
                                                       size_t lb, int k)
{
   void *op;
   DCL_LOCK_STATE;

   GC_ASSERT(k < MAXOBJKINDS);
   if (SMALL_OBJ(lb)) {
       void **opp;
       size_t lg;

       GC_DBG_COLLECT_AT_MALLOC(lb);
       if (EXTRA_BYTES != 0 && lb != 0) lb--;
                 /* We don't need the extra byte, since this won't be  */
                 /* collected anyway.                                  */
       LOCK();
       lg = GC_size_map[lb];
       opp = &GC_obj_kinds[k].ok_freelist[lg];
       op = *opp;
       if (EXPECT(op != NULL, TRUE)) {
           *opp = obj_link(op);
           obj_link(op) = 0;
           GC_bytes_allocd += GRANULES_TO_BYTES((word)lg);
           /* Mark bit was already set on free list.  It will be       */
           /* cleared only temporarily during a collection, as a       */
           /* result of the normal free list mark bit clearing.        */
           GC_non_gc_bytes += GRANULES_TO_BYTES((word)lg);
           UNLOCK();
       } else {
           UNLOCK();
           op = GC_generic_malloc(lb, k);
           /* For small objects, the free lists are completely marked. */
       }
       GC_ASSERT(0 == op || GC_is_marked(op));
   } else {
     op = GC_generic_malloc(lb, k);
     if (op /* != NULL */) { /* CPPCHECK */
       hdr * hhdr = HDR(op);

       GC_ASSERT(((word)op & (HBLKSIZE - 1)) == 0); /* large block */
       /* We don't need the lock here, since we have an undisguised    */
       /* pointer.  We do need to hold the lock while we adjust        */
       /* mark bits.                                                   */
       LOCK();
       set_mark_bit_from_hdr(hhdr, 0); /* Only object. */
#       ifndef THREADS
         GC_ASSERT(hhdr -> hb_n_marks == 0);
               /* This is not guaranteed in the multi-threaded case    */
               /* because the counter could be updated before locking. */
#       endif
       hhdr -> hb_n_marks = 1;
       UNLOCK();
     }
   }
   return op;
}

/* Allocate lb bytes of pointerful, traced, but not collectible data.   */
GC_API GC_ATTR_MALLOC void * GC_CALL GC_malloc_uncollectable(size_t lb)
{
 return GC_generic_malloc_uncollectable(lb, UNCOLLECTABLE);
}

#ifdef GC_ATOMIC_UNCOLLECTABLE
 /* Allocate lb bytes of pointer-free, untraced, uncollectible data    */
 /* This is normally roughly equivalent to the system malloc.          */
 /* But it may be useful if malloc is redefined.                       */
 GC_API GC_ATTR_MALLOC void * GC_CALL
       GC_malloc_atomic_uncollectable(size_t lb)
 {
   return GC_generic_malloc_uncollectable(lb, AUNCOLLECTABLE);
 }
#endif /* GC_ATOMIC_UNCOLLECTABLE */

#if defined(REDIRECT_MALLOC) && !defined(REDIRECT_MALLOC_IN_HEADER)

# ifndef MSWINCE
#  include <errno.h>
# endif

 /* Avoid unnecessary nested procedure calls here, by #defining some   */
 /* malloc replacements.  Otherwise we end up saving a meaningless     */
 /* return address in the object.  It also speeds things up, but it is */
 /* admittedly quite ugly.                                             */
# define GC_debug_malloc_replacement(lb) GC_debug_malloc(lb, GC_DBG_EXTRAS)

# if defined(CPPCHECK)
#   define REDIRECT_MALLOC_F GC_malloc /* e.g. */
# else
#   define REDIRECT_MALLOC_F REDIRECT_MALLOC
# endif

 void * malloc(size_t lb)
 {
   /* It might help to manually inline the GC_malloc call here.        */
   /* But any decent compiler should reduce the extra procedure call   */
   /* to at most a jump instruction in this case.                      */
#   if defined(I386) && defined(GC_SOLARIS_THREADS)
     /* Thread initialization can call malloc before we are ready for. */
     /* It is not clear that this is enough to help matters.           */
     /* The thread implementation may well call malloc at other        */
     /* inopportune times.                                             */
     if (!EXPECT(GC_is_initialized, TRUE)) return sbrk(lb);
#   endif
   return (void *)REDIRECT_MALLOC_F(lb);
 }

# if defined(GC_LINUX_THREADS)
   STATIC ptr_t GC_libpthread_start = 0;
   STATIC ptr_t GC_libpthread_end = 0;
   STATIC ptr_t GC_libld_start = 0;
   STATIC ptr_t GC_libld_end = 0;
   static GC_bool lib_bounds_set = FALSE;

   GC_INNER void GC_init_lib_bounds(void)
   {
     IF_CANCEL(int cancel_state;)
     DCL_LOCK_STATE;

     /* This test does not need to ensure memory visibility, since     */
     /* the bounds will be set when/if we create another thread.       */
     if (EXPECT(lib_bounds_set, TRUE)) return;

     DISABLE_CANCEL(cancel_state);
     GC_init(); /* if not called yet */
#     if defined(GC_ASSERTIONS) && defined(GC_ALWAYS_MULTITHREADED)
       LOCK(); /* just to set GC_lock_holder */
#     endif
     if (!GC_text_mapping("libpthread-",
                          &GC_libpthread_start, &GC_libpthread_end)) {
       /* Some libc implementations like bionic, musl and glibc 2.34   */
       /* do not have libpthread.so because the pthreads-related code  */
       /* is located in libc.so, thus potential calloc calls from such */
       /* code are forwarded to real (libc) calloc without any special */
       /* handling on the libgc side.  Checking glibc version at       */
       /* compile time to turn off the warning seems to be fine.       */
       /* TODO: Remove GC_text_mapping() call for this case.           */
#       if defined(__GLIBC__) \
          && (__GLIBC__ < 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ < 34))
         WARN("Failed to find libpthread.so text mapping: Expect crash\n", 0);
         /* This might still work with some versions of libpthread,      */
         /* so we do not abort.                                          */
#       endif
     }
     if (!GC_text_mapping("ld-", &GC_libld_start, &GC_libld_end)) {
         WARN("Failed to find ld.so text mapping: Expect crash\n", 0);
     }
#     if defined(GC_ASSERTIONS) && defined(GC_ALWAYS_MULTITHREADED)
       UNLOCK();
#     endif
     RESTORE_CANCEL(cancel_state);
     lib_bounds_set = TRUE;
   }
# endif /* GC_LINUX_THREADS */

 void * calloc(size_t n, size_t lb)
 {
   if ((lb | n) > GC_SQRT_SIZE_MAX /* fast initial test */
       && lb && n > GC_SIZE_MAX / lb)
     return (*GC_get_oom_fn())(GC_SIZE_MAX); /* n*lb overflow */
#   if defined(GC_LINUX_THREADS)
     /* libpthread allocated some memory that is only pointed to by    */
     /* mmapped thread stacks.  Make sure it is not collectible.       */
     {
       ptr_t caller = (ptr_t)__builtin_return_address(0);

       GC_init_lib_bounds();
       if (((word)caller >= (word)GC_libpthread_start
            && (word)caller < (word)GC_libpthread_end)
           || ((word)caller >= (word)GC_libld_start
               && (word)caller < (word)GC_libld_end))
         return GC_generic_malloc_uncollectable(n * lb, UNCOLLECTABLE);
       /* The two ranges are actually usually adjacent, so there may   */
       /* be a way to speed this up.                                   */
     }
#   endif
   return (void *)REDIRECT_MALLOC_F(n * lb);
 }

# ifndef strdup
   char *strdup(const char *s)
   {
     size_t lb = strlen(s) + 1;
     char *result = (char *)REDIRECT_MALLOC_F(lb);
     if (result == 0) {
       errno = ENOMEM;
       return 0;
     }
     BCOPY(s, result, lb);
     return result;
   }
# endif /* !defined(strdup) */
/* If strdup is macro defined, we assume that it actually calls malloc, */
/* and thus the right thing will happen even without overriding it.     */
/* This seems to be true on most Linux systems.                         */

# ifndef strndup
   /* This is similar to strdup().     */
   char *strndup(const char *str, size_t size)
   {
     char *copy;
     size_t len = strlen(str);
     if (len > size)
       len = size;
     copy = (char *)REDIRECT_MALLOC_F(len + 1);
     if (copy == NULL) {
       errno = ENOMEM;
       return NULL;
     }
     if (EXPECT(len > 0, TRUE))
       BCOPY(str, copy, len);
     copy[len] = '\0';
     return copy;
   }
# endif /* !strndup */

# undef GC_debug_malloc_replacement

#endif /* REDIRECT_MALLOC */

/* Explicitly deallocate an object p.                           */
GC_API void GC_CALL GC_free(void * p)
{
   struct hblk *h;
   hdr *hhdr;
   size_t sz; /* In bytes */
   size_t ngranules;   /* sz in granules */
   int knd;
   struct obj_kind * ok;
   DCL_LOCK_STATE;

   if (p /* != NULL */) {
       /* CPPCHECK */
   } else {
       /* Required by ANSI.  It's not my fault ...     */
       return;
   }

#   ifdef LOG_ALLOCS
     GC_log_printf("GC_free(%p) after GC #%lu\n",
                   p, (unsigned long)GC_gc_no);
#   endif
   h = HBLKPTR(p);
   hhdr = HDR(h);
#   if defined(REDIRECT_MALLOC) && \
       ((defined(NEED_CALLINFO) && defined(GC_HAVE_BUILTIN_BACKTRACE)) \
        || defined(GC_SOLARIS_THREADS) || defined(GC_LINUX_THREADS) \
        || defined(MSWIN32))
       /* This might be called indirectly by GC_print_callers to free  */
       /* the result of backtrace_symbols.                             */
       /* For Solaris, we have to redirect malloc calls during         */
       /* initialization.  For the others, this seems to happen        */
       /* implicitly.                                                  */
       /* Don't try to deallocate that memory.                         */
       if (0 == hhdr) return;
#   endif
   GC_ASSERT(GC_base(p) == p);
   sz = (size_t)hhdr->hb_sz;
   ngranules = BYTES_TO_GRANULES(sz);
   knd = hhdr -> hb_obj_kind;
   ok = &GC_obj_kinds[knd];
   if (EXPECT(ngranules <= MAXOBJGRANULES, TRUE)) {
       void **flh;

       LOCK();
       GC_bytes_freed += sz;
       if (IS_UNCOLLECTABLE(knd)) GC_non_gc_bytes -= sz;
               /* It's unnecessary to clear the mark bit.  If the      */
               /* object is reallocated, it doesn't matter.  O.w. the  */
               /* collector will do it, since it's on a free list.     */
       if (ok -> ok_init && EXPECT(sz > sizeof(word), TRUE)) {
           BZERO((word *)p + 1, sz-sizeof(word));
       }
       flh = &(ok -> ok_freelist[ngranules]);
       obj_link(p) = *flh;
       *flh = (ptr_t)p;
       UNLOCK();
   } else {
       size_t nblocks = OBJ_SZ_TO_BLOCKS(sz);

       LOCK();
       GC_bytes_freed += sz;
       if (IS_UNCOLLECTABLE(knd)) GC_non_gc_bytes -= sz;
       if (nblocks > 1) {
         GC_large_allocd_bytes -= nblocks * HBLKSIZE;
       }
       GC_freehblk(h);
       UNLOCK();
   }
}

/* Explicitly deallocate an object p when we already hold lock.         */
/* Only used for internally allocated objects, so we can take some      */
/* shortcuts.                                                           */
#ifdef THREADS
 GC_INNER void GC_free_inner(void * p)
 {
   struct hblk *h;
   hdr *hhdr;
   size_t sz; /* bytes */
   size_t ngranules;  /* sz in granules */
   int knd;
   struct obj_kind * ok;

   h = HBLKPTR(p);
   hhdr = HDR(h);
   knd = hhdr -> hb_obj_kind;
   sz = (size_t)hhdr->hb_sz;
   ngranules = BYTES_TO_GRANULES(sz);
   ok = &GC_obj_kinds[knd];
   if (ngranules <= MAXOBJGRANULES) {
       void ** flh;

       GC_bytes_freed += sz;
       if (IS_UNCOLLECTABLE(knd)) GC_non_gc_bytes -= sz;
       if (ok -> ok_init && EXPECT(sz > sizeof(word), TRUE)) {
           BZERO((word *)p + 1, sz-sizeof(word));
       }
       flh = &(ok -> ok_freelist[ngranules]);
       obj_link(p) = *flh;
       *flh = (ptr_t)p;
   } else {
       size_t nblocks = OBJ_SZ_TO_BLOCKS(sz);
       GC_bytes_freed += sz;
       if (IS_UNCOLLECTABLE(knd)) GC_non_gc_bytes -= sz;
       if (nblocks > 1) {
         GC_large_allocd_bytes -= nblocks * HBLKSIZE;
       }
       GC_freehblk(h);
   }
 }
#endif /* THREADS */

#if defined(REDIRECT_MALLOC) && !defined(REDIRECT_FREE)
# define REDIRECT_FREE GC_free
#endif

#if defined(REDIRECT_FREE) && !defined(REDIRECT_MALLOC_IN_HEADER)

# if defined(CPPCHECK)
#   define REDIRECT_FREE_F GC_free /* e.g. */
# else
#   define REDIRECT_FREE_F REDIRECT_FREE
# endif

 void free(void * p GC_ATTR_UNUSED)
 {
#   ifndef IGNORE_FREE
#     if defined(GC_LINUX_THREADS) && !defined(USE_PROC_FOR_LIBRARIES)
       /* Don't bother with initialization checks.  If nothing         */
       /* has been initialized, the check fails, and that's safe,      */
       /* since we have not allocated uncollectible objects neither.   */
       ptr_t caller = (ptr_t)__builtin_return_address(0);
       /* This test does not need to ensure memory visibility, since   */
       /* the bounds will be set when/if we create another thread.     */
       if (((word)caller >= (word)GC_libpthread_start
            && (word)caller < (word)GC_libpthread_end)
           || ((word)caller >= (word)GC_libld_start
               && (word)caller < (word)GC_libld_end)) {
         GC_free(p);
         return;
       }
#     endif
     REDIRECT_FREE_F(p);
#   endif
 }
#endif /* REDIRECT_FREE */