* properly, we need to make the align function publicly visible, too */
static void initialize(xbt_mheap_t mdp)
{
+ int i;
+ malloc_info mi; /* to compute the offset of the swag hook */
+
mdp->heapsize = HEAP / BLOCKSIZE;
mdp->heapinfo = (malloc_info *)
align(mdp, mdp->heapsize * sizeof(malloc_info));
mdp->heapindex = 0;
mdp->heapbase = (void *) mdp->heapinfo;
mdp->flags |= MMALLOC_INITIALIZED;
+
+ for (i=0;i<BLOCKLOG;i++) {
+ xbt_swag_init(&(mdp->fraghead[i]),
+ xbt_swag_offset(mi, freehook));
+ }
}
+#define update_hook(a,offset) do { if (a) { a = ((char*)a +(offset));} }while(0)
+
/* Get neatly aligned memory from the low level layers, and register it
* into the heap info table as necessary. */
static void *register_morecore(struct mdesc *mdp, size_t size)
{
+ int i;
void *result;
malloc_info *newinfo, *oldinfo;
size_t newsize;
newinfo = (malloc_info *) align(mdp, newsize * sizeof(malloc_info));
memset(newinfo, 0, newsize * sizeof(malloc_info));
memcpy(newinfo, oldinfo, mdp->heapsize * sizeof(malloc_info));
+
+ /* Update the swag of busy blocks containing free fragments by applying the offset to all swag_hooks. Yeah. My hand is right in the fan and I still type */
+ size_t offset=((char*)newinfo)-((char*)oldinfo);
+
+ for (i=1/*first element of heapinfo describes the mdesc area*/;
+ i<mdp->heaplimit;
+ i++) {
+ update_hook(newinfo[i].freehook.next,offset);
+ update_hook(newinfo[i].freehook.prev,offset);
+ }
+ // also update the starting points of the swag
+ for (i=0;i<BLOCKLOG;i++) {
+ update_hook(mdp->fraghead[i].head,offset);
+ update_hook(mdp->fraghead[i].tail,offset);
+ }
mdp->heapinfo = newinfo;
/* mark the space previously occupied by the block info as free by first marking it
mdp->heaplimit = BLOCK((char *) result + size);
return (result);
}
+#undef update_hook
/* Allocate memory from the heap. */
-
-void *mmalloc(xbt_mheap_t mdp, size_t size)
+void *mmalloc(xbt_mheap_t mdp, size_t size) {
+ void *res= mmalloc_no_memset(mdp,size);
+// fprintf(stderr,"malloc(%zu)~>%p\n",size,res);
+ memset(res,0,size);
+ return res;
+}
+/* Spliting mmalloc this way is mandated by a trick in mrealloc, that gives
+ back the memory of big blocks to the system before reallocating them: we don't
+ want to loose the beginning of the area when this happens */
+void *mmalloc_no_memset(xbt_mheap_t mdp, size_t size)
{
void *result;
size_t block, blocks, lastblocks, start;
register size_t i;
- struct list *next;
register size_t log;
int it;
if (!(mdp->flags & MMALLOC_INITIALIZED))
initialize(mdp);
+ mmalloc_paranoia(mdp);
+
/* Determine the allocation policy based on the request size. */
if (size <= BLOCKSIZE / 2) {
/* Small allocation to receive a fragment of a block.
++log;
}
- /* Look in the fragment lists for a
- free fragment of the desired size. */
- next = mdp->fraghead[log].next;
- if (next != NULL) {
- /* There are free fragments of this size.
- Pop a fragment out of the fragment list and return it.
- Update the block's nfree and first counters. */
- int frag_nb;
- result = (void *) next;
- block = BLOCK(result);
+ /* Look in the fragment lists for a free fragment of the desired size. */
+ if (xbt_swag_size(&mdp->fraghead[log])>0) {
+ /* There are free fragments of this size; Get one of them and prepare to return it.
+ Update the block's nfree and if no other free fragment, get out of the swag. */
+
+ /* search a fragment that I could return as a result */
+ malloc_info *candidate_info = xbt_swag_getFirst(&mdp->fraghead[log]);
+ size_t candidate_block = (candidate_info - &(mdp->heapinfo[0]));
+ size_t candidate_frag;
+ for (candidate_frag=0;candidate_frag<(size_t) (BLOCKSIZE >> log);candidate_frag++)
+ if (candidate_info->busy_frag.frag_size[candidate_frag] == -1)
+ break;
+ xbt_assert(candidate_frag < (size_t) (BLOCKSIZE >> log),
+ "Block %zu was registered as containing free fragments of type %zu, but I can't find any",candidate_block,log);
+
+ result = (void*) (((char*)ADDRESS(candidate_block)) + (candidate_frag << log));
+
+ /* Remove this fragment from the list of free guys */
+ candidate_info->busy_frag.nfree--;
+ if (candidate_info->busy_frag.nfree == 0) {
+ xbt_swag_remove(candidate_info,&mdp->fraghead[log]);
+ }
- frag_nb = RESIDUAL(result, BLOCKSIZE) >> log;
- mdp->heapinfo[block].busy_frag.frag_size[frag_nb] = requested_size;
- //FIXME setup backtrace
+ /* Update our metadata about this fragment */
+ candidate_info->busy_frag.frag_size[candidate_frag] = requested_size;
+ xbt_backtrace_no_malloc(candidate_info->busy_frag.bt[candidate_frag],XBT_BACKTRACE_SIZE);
- next->prev->next = next->next;
- if (next->next != NULL) {
- next->next->prev = next->prev;
- }
- if (--mdp->heapinfo[block].busy_frag.nfree != 0) {
- mdp->heapinfo[block].busy_frag.first =
- RESIDUAL(next->next, BLOCKSIZE) >> log;
- }
+ /* Update the statistics. */
+ mdp -> heapstats.chunks_used++;
+ mdp -> heapstats.bytes_used += 1 << log;
+ mdp -> heapstats.chunks_free--;
+ mdp -> heapstats.bytes_free -= 1 << log;
} else {
/* No free fragments of the desired size, so get a new block
and break it into fragments, returning the first. */
- //printf("(%s) No free fragment...",xbt_thread_self_name());
result = mmalloc(mdp, BLOCKSIZE); // does not return NULL
-
- /* Link all fragments but the first into the free list, and mark their requested size to 0. */
block = BLOCK(result);
- for (i = 1; i < (size_t) (BLOCKSIZE >> log); ++i) {
- mdp->heapinfo[block].busy_frag.frag_size[i] = 0;
- next = (struct list *) ((char *) result + (i << log));
- next->next = mdp->fraghead[log].next;
- next->prev = &mdp->fraghead[log];
- next->prev->next = next;
- if (next->next != NULL) {
- next->next->prev = next;
- }
- }
- mdp->heapinfo[block].busy_frag.frag_size[0] = requested_size;
- // FIXME: setup backtrace
- /* Initialize the nfree and first counters for this block. */
mdp->heapinfo[block].type = log;
+ /* Link all fragments but the first as free, and add the block to the swag of blocks containing free frags */
+ for (i = 1; i < (size_t) (BLOCKSIZE >> log); ++i) {
+ mdp->heapinfo[block].busy_frag.frag_size[i] = -1;
+ }
mdp->heapinfo[block].busy_frag.nfree = i - 1;
- mdp->heapinfo[block].busy_frag.first = i - 1;
+ mdp->heapinfo[block].freehook.prev = NULL;
+ mdp->heapinfo[block].freehook.next = NULL;
+
+ xbt_swag_insert(&mdp->heapinfo[block], &(mdp->fraghead[log]));
+
+ /* mark the fragment returned as busy */
+ mdp->heapinfo[block].busy_frag.frag_size[0] = requested_size;
+ xbt_backtrace_no_malloc(mdp->heapinfo[block].busy_frag.bt[0],XBT_BACKTRACE_SIZE);
+
+ /* update stats */
+ mdp -> heapstats.chunks_free += (BLOCKSIZE >> log) - 1;
+ mdp -> heapstats.bytes_free += BLOCKSIZE - (1 << log);
+ mdp -> heapstats.bytes_used -= BLOCKSIZE - (1 << log);
}
} else {
/* Large allocation to receive one or more blocks.
start = block = MALLOC_SEARCH_START;
while (mdp->heapinfo[block].free_block.size < blocks) {
if (mdp->heapinfo[block].type >=0) { // Don't trust xbt_die and friends in malloc-level library, you fool!
- fprintf(stderr,"Internal error: found a free block not marked as such (block=%lu type=%lu). Please report this bug.\n",(unsigned long)block,(unsigned long)mdp->heapinfo[block].type);
- abort();
+ fprintf(stderr,"Internal error: found a free block not marked as such (block=%lu type=%lu). Please report this bug.\n",(unsigned long)block,(unsigned long)mdp->heapinfo[block].type);
+ abort();
}
block = mdp->heapinfo[block].free_block.next;
result = register_morecore(mdp, blocks * BLOCKSIZE);
block = BLOCK(result);
- for (it=0;it<blocks;it++)
- mdp->heapinfo[block+it].type = 0;
+ for (it=0;it<blocks;it++){
+ mdp->heapinfo[block+it].type = 0;
+ }
mdp->heapinfo[block].busy_block.size = blocks;
mdp->heapinfo[block].busy_block.busy_size = requested_size;
mdp->heapinfo[block].busy_block.bt_size=xbt_backtrace_no_malloc(mdp->heapinfo[block].busy_block.bt,XBT_BACKTRACE_SIZE);
+ mdp -> heapstats.chunks_used++;
+ mdp -> heapstats.bytes_used += blocks * BLOCKSIZE;
return result;
}
/* The block we found has a bit left over,
so relink the tail end back into the free list. */
mdp->heapinfo[block + blocks].free_block.size
- = mdp->heapinfo[block].free_block.size - blocks;
+ = mdp->heapinfo[block].free_block.size - blocks;
mdp->heapinfo[block + blocks].free_block.next
- = mdp->heapinfo[block].free_block.next;
+ = mdp->heapinfo[block].free_block.next;
mdp->heapinfo[block + blocks].free_block.prev
- = mdp->heapinfo[block].free_block.prev;
+ = mdp->heapinfo[block].free_block.prev;
mdp->heapinfo[mdp->heapinfo[block].free_block.prev].free_block.next
- = mdp->heapinfo[mdp->heapinfo[block].free_block.next].free_block.prev
- = mdp->heapindex = block + blocks;
+ = mdp->heapinfo[mdp->heapinfo[block].free_block.next].free_block.prev
+ = mdp->heapindex = block + blocks;
} else {
/* The block exactly matches our requirements,
so just remove it from the list. */
mdp->heapinfo[mdp->heapinfo[block].free_block.next].free_block.prev
- = mdp->heapinfo[block].free_block.prev;
+ = mdp->heapinfo[block].free_block.prev;
mdp->heapinfo[mdp->heapinfo[block].free_block.prev].free_block.next
- = mdp->heapindex = mdp->heapinfo[block].free_block.next;
+ = mdp->heapindex = mdp->heapinfo[block].free_block.next;
}
- for (it=0;it<blocks;it++)
+ for (it=0;it<blocks;it++){
mdp->heapinfo[block+it].type = 0;
+ }
mdp->heapinfo[block].busy_block.size = blocks;
mdp->heapinfo[block].busy_block.busy_size = requested_size;
//mdp->heapinfo[block].busy_block.bt_size = 0;
mdp->heapinfo[block].busy_block.bt_size = xbt_backtrace_no_malloc(mdp->heapinfo[block].busy_block.bt,XBT_BACKTRACE_SIZE);
+
+ mdp -> heapstats.chunks_used++;
+ mdp -> heapstats.bytes_used += blocks * BLOCKSIZE;
+ mdp -> heapstats.bytes_free -= blocks * BLOCKSIZE;
+
}
//printf("(%s) Done mallocing. Result is %p\n",xbt_thread_self_name(),result);fflush(stdout);
+
return (result);
}
