[wg@malloc.de] [PATCH] Correction for new malloc behaviour

[Ulrich Drepper]

To : drepper at redhat dot com
Subject : [PATCH] Correction for new malloc behaviour
>From : wg at malloc dot de
Date : 21 Jul 2000 17:28:21 -0000
Cc : wg at malloc dot de

[My usual internet connection is failing today and over the weekend,
 please forward to libc-hacker if that's OK]

Hello,

Here is the promised patch for malloc, such that mmap() isn't used
excessively when it's not really necessary.  With this, it should be
possible to run an application with `ulimit -d <small value>'
(simulating `running out of addres space'), and not pay the mmap()
overhead for small chunks.  Please test.

Regards,
Wolfram.

2000-07-21  Wolfram Gloger  <wg@malloc.de>

	* malloc/malloc.c (chunk_alloc): Try mmap_chunk() for smaller
	allocations only if a non-main arena cannot be extended.

--- malloc.c	2000/07/17 11:33:40	1.1.1.18
+++ malloc.c	2000/07/21 17:10:26
@@ -2938,9 +2938,12 @@
     if ((remainder_size = chunksize(top(ar_ptr)) - nb) < (long)MINSIZE)
     {
 #if HAVE_MMAP
-      /* A last attempt:  when we are out of address space in the arena,
-         try mmap anyway, as long as it is allowed at all.  */
-      if (n_mmaps_max > 0 && (victim = mmap_chunk(nb)) != 0)
+      /* A last attempt: when we are out of address space in a
+         non-main arena, try mmap anyway, as long as it is allowed at
+         all.  */
+      if (ar_ptr != &main_arena &&
+          n_mmaps_max > 0 &&
+          (victim = mmap_chunk(nb)) != 0)
         return victim;
 #endif
       return 0; /* propagate failure */