2.4.16: Out of memory - when still more than 100MB free
I've got VMWare killed a couple of times mysteriously.
I've got 256MB memory and no swap on my laptop running 2.4.16 and for some reason VMWare has got killed with the following syslog information:
Dec 17 23:33:23 puck kernel: Out of Memory: Killed process 28803 (vmware). Dec 17 23:33:35 puck kernel: Out of Memory: Killed process 28804 (vmware). Dec 17 23:33:37 puck kernel: /dev/vmmon: Vmx86_ReleaseVM: unlocked pages: 75286, unlocked dirty pages: 51084
What I find odd is, that I am quite certain this machine did _not_ run out of memory when this happened. Just a few minutes ago I had an idle VMWare session and started an XEmacs to edit a large file. Just being curious, I happened to say 'free' a few seconds before VMWare got killed:
$ free
total used free shared buffers cached
Mem: 256224 219748 36476 0 12956 48816
-/+ buffers/cache: 157976 98248
Swap: 0 0 0Boom, it died about the same time I exit from XEmacs. After that, I ran 'free' again:
n$ free
total used free shared buffers cached
Mem: 256224 203644 52580 0 13148 81620
-/+ buffers/cache: 108876 147348
Swap: 0 0 0
I may be missing something obvious herfe, but I just can't figure out why kernel killes VMWare in this situation.
If anyone's interested, I think I can reproduce this and - if someone will kindly instruct me a bit - produce some more information. I _think_ this is the place where experienced kernel hackers start speaking about running 'vmstat'. And where I usually start having problems undertanding what it is that people are talking about...
Suonpää...
Quoted 9 lines · Samuli Suonpaa
I've got VMWare killed a couple of times mysteriously.
I've got 256MB memory and no swap on my laptop running 2.4.16 and for some reason VMWare has got killed with the following syslog information:
Dec 17 23:33:23 puck kernel: Out of Memory: Killed process 28803 (vmware). Dec 17 23:33:35 puck kernel: Out of Memory: Killed process 28804 (vmware). Dec 17 23:33:37 puck kernel: /dev/vmmon: Vmx86_ReleaseVM: unlocked pages: 75286, unlocked dirty pages: 51084
Samuli,
The problem is that buffer/cache/{i,d}cache pages are not getting freed easily, and instead the kernel swapouts anonymous memory.
Could you please try 2.4.17-rc1 and tell me if it makes a difference for you ?
Thanks
On Mon, Dec 17, 2001 at 07:10:54PM -0200, you [Marcelo Tosatti] claimed:
Quoted 14 lines · Samuli Suonpaa
Quoted 10 lines
I've got VMWare killed a couple of times mysteriously.
I've got 256MB memory and no swap on my laptop running 2.4.16 and for some reason VMWare has got killed with the following syslog information:
Dec 17 23:33:23 puck kernel: Out of Memory: Killed process 28803 (vmware). Dec 17 23:33:35 puck kernel: Out of Memory: Killed process 28804 (vmware). Dec 17 23:33:37 puck kernel: /dev/vmmon: Vmx86_ReleaseVM: unlocked pages: 75286, unlocked dirty pages: 51084
Quoted 7 lines
Samuli,
The problem is that buffer/cache/{i,d}cache pages are not getting freed easily, and instead the kernel swapouts anonymous memory.
Could you please try 2.4.17-rc1 and tell me if it makes a difference for you ?
See my report on what happens on a 2GB box with .16 or .17rc1. Buffers are still not released as they should.
http://marc.theaimsgroup.com/?l=linux-kernel&m=100849985518543&w=2
http://marc.theaimsgroup.com/?l=linux-kernel&m=100857274818037&w=2
Perhaps someone could test on x86 with less memory (I can do that later, but right now I don't have any throw-away box with a recent kernel on it). On ia64 with 2GB+256MB swap this results in OOM when trying to allocate and use 1.7GB, albeit the real mem usage (-buffers) is less than 200MB.
Basically the test is
fill cache
find / -type f -exec cat {} \; > /dev/null
updatedb
then run this with suitable argument:
#include <stdio.h>
#include <stdlib.h>
#define BKSP "\010\010\010\010\010\010"
int main(int argc, char** argv)
{
unsigned long megs = 512;
unsigned long size, i;
unsigned char* buf;
if (argc > 1) megs = atol(argv[1]);
size = megs * 1024 * 1024;
fprintf(stderr, "Allocating %lu megs...\n\n ", megs);
buf = malloc(size);
if (!buf)
{
fprintf(stderr, "malloc(%lu", size);
perror(")");
exit(-1);
}
for (i = 0; i < size; i++)
{
buf[i] = 42;
if ((i + 1) % (1024 * 1024) == 0)
fprintf(stderr, BKSP "%4uMB", (i + 1) / 1024 / 1024);
}
fprintf(stderr, "\n Success.\n");
return 1;
}-- v --