lkml.org 
[lkml]   [2020]   [Jun]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: mm: mkfs.ext4 invoked oom-killer on i386 - pagecache_get_page
Naresh Kamboju writes:
>mkfs -t ext4 /dev/disk/by-id/ata-TOSHIBA_MG04ACA100N_Y8RQK14KF6XF
>mke2fs 1.43.8 (1-Jan-2018)
>Creating filesystem with 244190646 4k blocks and 61054976 inodes
>Filesystem UUID: 7c380766-0ed8-41ba-a0de-3c08e78f1891
>Superblock backups stored on blocks:
>32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208,
>4096000, 7962624, 11239424, 20480000, 23887872, 71663616, 78675968,
>102400000, 214990848
>Allocating group tables: 0/7453 done
>Writing inode tables: 0/7453 done
>Creating journal (262144 blocks): [ 51.544525] under min:0 emin:0
>[ 51.845304] under min:0 emin:0
>[ 51.848738] under min:0 emin:0
>[ 51.858147] under min:0 emin:0
>[ 51.861333] under min:0 emin:0
>[ 51.862034] under min:0 emin:0
>[ 51.862442] under min:0 emin:0
>[ 51.862763] under min:0 emin:0

Thanks, this helps a lot. Somehow we're entering mem_cgroup_below_min even when
min/emin is 0 (which should indeed be the case if you haven't set them in the
hierarchy).

My guess is that page_counter_read(&memcg->memory) is 0, which means
mem_cgroup_below_min will return 1.

However, I don't know for sure why that should then result in the OOM killer
coming along. My guess is that since this memcg has 0 pages to scan anyway, we
enter premature OOM under some conditions. I don't know why we wouldn't have
hit that with the old version of mem_cgroup_protected that returned
MEMCG_PROT_* members, though.

Can you please try the patch with the `>=` checks in mem_cgroup_below_min and
mem_cgroup_below_low changed to `>`? If that fixes it, then that gives a strong
hint about what's going on here.

Thanks for your help!

\
 
 \ /
  Last update: 2020-06-17 15:58    [W:0.167 / U:0.688 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site