On Wed, 17 Jun 2020 at 19:41, Michal Hocko <mho...@kernel.org> wrote: > > [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.
May I request you to send debugging or proposed fix patches here. I am happy to do more testing. FYI, Here is my repository for testing. git: https://github.com/nareshkamboju/linux/tree/printk branch: printk - Naresh
