A few days ago, a colleague asked me for help with some malloc problem.
I applied my usual solution, efence, and found the bug. Then she ran a
larger simulation, and it promptly failed with ENOMEM. After a bit of
searching, I found that MAX_MAP_COUNT was the culprit, and that in
order to raise the limit, I had to rebuild the kernel and reboot the
machine. Reboot a Linux system to change one stupid little integer !
I felt very embarrassed ...
Okay, here's a patch for 2.4.1 that should prevent further humiliations
of that type from happening. (I'm now hitting a limit at around 229309
maps (~0.87GB), but that's something else. Test program to exercise
max_map_count at ftp://icaftp.epfl.ch/pub/people/almesber/junk/mm.c)
BTW, I've noticed that mm/mmap.c:sys_brk and do_brk both check
vm_enough_memory. Do we really need this ? Removing one of them may
make sys_brk about 2% faster ;-)
- Werner
------------------------------------ patch ------------------------------------
--- linux.orig/include/linux/sched.h Tue Jan 30 08:24:56 2001
+++ linux/include/linux/sched.h Sat Feb 3 14:29:41 2001
@@ -195,7 +195,9 @@
}
/* Maximum number of active map areas.. This is a random (large) number */
-#define MAX_MAP_COUNT (65536)
+#define DEFAULT_MAX_MAP_COUNT (65536)
+
+extern int max_map_count;
/* Number of map areas at which the AVL tree is activated. This is arbitrary. */
#define AVL_MIN_MAP_COUNT 32
--- linux.orig/include/linux/sysctl.h Tue Jan 30 08:24:55 2001
+++ linux/include/linux/sysctl.h Sat Feb 3 14:29:49 2001
@@ -132,7 +132,8 @@
VM_PAGECACHE=7, /* struct: Set cache memory thresholds */
VM_PAGERDAEMON=8, /* struct: Control kswapd behaviour */
VM_PGT_CACHE=9, /* struct: Set page table cache parameters */
- VM_PAGE_CLUSTER=10 /* int: set number of pages to swap together */
+ VM_PAGE_CLUSTER=10, /* int: set number of pages to swap together */
+ VM_MAX_MAP_COUNT=11, /* int: Maximum number of active map areas */
};
--- linux.orig/kernel/sysctl.c Fri Dec 29 23:07:24 2000
+++ linux/kernel/sysctl.c Sat Feb 3 14:28:05 2001
@@ -257,6 +257,8 @@
&pgt_cache_water, 2*sizeof(int), 0644, NULL, &proc_dointvec},
{VM_PAGE_CLUSTER, "page-cluster",
&page_cluster, sizeof(int), 0644, NULL, &proc_dointvec},
+ {VM_MAX_MAP_COUNT, "max_map_count",
+ &max_map_count, sizeof(int), 0644, NULL, &proc_dointvec},
{0}
};
--- linux.orig/mm/mmap.c Mon Jan 29 17:10:41 2001
+++ linux/mm/mmap.c Sat Feb 3 14:28:49 2001
@@ -37,6 +37,7 @@
};
int sysctl_overcommit_memory;
+int max_map_count = DEFAULT_MAX_MAP_COUNT;
/* Check that a process has enough memory to allocate a
* new virtual mapping.
@@ -207,7 +208,7 @@
return -EINVAL;
/* Too many mappings? */
- if (mm->map_count > MAX_MAP_COUNT)
+ if (mm->map_count > max_map_count)
return -ENOMEM;
/* mlock MCL_FUTURE? */
@@ -691,7 +692,7 @@
/* If we'll make "hole", check the vm areas limit */
if ((mpnt->vm_start < addr && mpnt->vm_end > addr+len)
- && mm->map_count >= MAX_MAP_COUNT)
+ && mm->map_count >= max_map_count)
return -ENOMEM;
/*
@@ -809,7 +810,7 @@
> current->rlim[RLIMIT_AS].rlim_cur)
return -ENOMEM;
- if (mm->map_count > MAX_MAP_COUNT)
+ if (mm->map_count > max_map_count)
return -ENOMEM;
if (!vm_enough_memory(len >> PAGE_SHIFT))
--- linux.orig/mm/filemap.c Tue Jan 16 02:14:41 2001
+++ linux/mm/filemap.c Sat Feb 3 14:18:02 2001
@@ -1923,7 +1923,7 @@
int error = 0;
/* This caps the number of vma's this process can own */
- if (vma->vm_mm->map_count > MAX_MAP_COUNT)
+ if (vma->vm_mm->map_count > max_map_count)
return -ENOMEM;
if (start == vma->vm_start) {
--- linux.orig/Documentation/sysctl/vm.txt Tue Aug 8 08:01:34 2000
+++ linux/Documentation/sysctl/vm.txt Sat Feb 3 14:33:14 2001
@@ -21,6 +21,7 @@
- buffermem
- freepages
- kswapd
+- max_map_count
- overcommit_memory
- page-cluster
- pagecache
@@ -171,6 +172,19 @@
and don't use much of it.
Look at: mm/mmap.c::vm_enough_memory() for more information.
+
+==============================================================
+
+max_map_count:
+
+This file contains the maximum number of memory map areas a
+process may have. Memory map areas are used as a side-effect
+of calling malloc, directly by mmap and mprotect, and also
+when loading shared libraries.
+
+While most applications need less than a thousand maps,
+certain programs, particularly malloc debuggers, may consume
+lots of them, e.g. up to one or two maps per allocation.
==============================================================
--
_________________________________________________________________________
/ Werner Almesberger, ICA, EPFL, CH [email protected] /
/_IN_N_032__Tel_+41_21_693_6621__Fax_+41_21_693_6610_____________________/