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!