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
    [Our emails have crossed]

    On Wed 17-06-20 14:57:58, Chris Down wrote:
    > 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.

    Yes this is the case because this is likely the root memcg which skips
    all charges.

    > 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.

    Not really. There is likely no other memcg to reclaim from and assuming
    min limit protection will result in no reclaimable memory and thus the
    OOM killer.

    > 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.

    This would work but I believe an explicit check for the root memcg would
    be easier to spot the reasoning.

    --
    Michal Hocko
    SUSE Labs

    \
     
     \ /
      Last update: 2020-06-17 16:13    [W:4.233 / U:0.084 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site