Hi guys,

I'm having an issue with swapoff which gets killed by OOM. I was able to reproduce it with UML guests running different kernel versions, starting from 2.6.26 to  3.1.3. The only difference is with latest kernels sometimes it crashes instead of being killed. I haven't seen it on 'native' kernels so I guess it's UML specific.

Richard, I know I mentioned on the IRC yesterday that it was fine with 2.6.37, but I was wrong.

Basically this is the command line to start UML:

/usr/local/sbin/linux32.uml mem=160m elevator=noop ubda=/home/uml/machines/static-fe02/root_cow.img,/opt/uml/shared/ia32_squeeze/images/root01.img ubdb=swap.img root=/dev/ubda con0=fd:0,fd:1 con=pty eth0=daemon,de:00:00:00:00:08,unix,/var/run/uml-utilities/uml_switch.ctl umid=static-fe02 uml_dir=/root/.uml hostfs=/var/uml

swap.img is a solid file, 360M in size. Then inside the VM I ran the following perl script:

#!/usr/bin/perl

my @a;

print "Filling...\n";
my $i = 0;
while ($i < 20000000) {
    push @a, 1;
    $i++;
}
print "Done.\n";


There is nothing magical about this script, it just makes sure some swap is being used after it finishes.

static-fe02:~# free
             total       used       free     shared    buffers     cached
Mem:        158468       8524     149944          0        136       4016
-/+ buffers/cache:       4372     154096
Swap:       327672       2344     325328

There should be enough RAM to clear the swap, right? However if I run swapoff -a, I get this:

swapoff invoked oom-killer: gfp_mask=0x1200d2, order=0, oomkilladj=0
1294cd74:  [<081ea1a9>] dump_stack+0x1c/0x20
1294cd8c:  [<08095712>] oom_kill_process+0x4b/0x166
1294cdb8:  [<08095bcb>] out_of_memory+0x151/0x17e
1294cdf0:  [<08097abf>] __alloc_pages_internal+0x299/0x33f
1294ce48:  [<08097b8e>] __alloc_pages+0x13/0x15
1294ce54:  [<080a506a>] read_swap_cache_async+0x40/0xab
1294ce7c:  [<080a6583>] sys_swapoff+0x1ea/0x774
1294cf28:  [<0805b017>] handle_syscall+0x7b/0x98
1294cf78:  [<08068f72>] userspace+0x2cd/0x374
1294cfe0:  [<08058d2b>] fork_handler+0x53/0x5b
1294cffc:  [<00000000>] 0x0

Mem-info:
Normal per-cpu:
CPU    0: hi:   42, btch:   7 usd:  35
Active:200 inactive:1 dirty:0 writeback:0 unstable:0
 free:456 slab:38427 mapped:0 pagetables:49 bounce:0
Normal free:1824kB min:1680kB low:2100kB high:2520kB active:800kB inactive:4kB present:177112kB pages_scanned:422 all_unreclaimable? no
lowmem_reserve[]: 0 0
Normal: 42*4kB 1*8kB 1*16kB 1*32kB 1*64kB 0*128kB 0*256kB 1*512kB 1*1024kB 0*2048kB 0*4096kB = 1824kB
22 total pagecache pages
Swap cache: add 802082, delete 802082, find 30288/37816
Free swap  = 4294965488kB
Total swap = 0kB
Free swap:        -1808kB
44627 pages of RAM
0 pages of HIGHMEM
5010 reserved pages
20 pages shared
0 pages swap cached
Out of memory: kill process 646 (swapoff) score -1 or a child
Killed process 646 (swapoff)

Can anybody reproduce this?


Best regards,
-- 
Dmitry Panov