linux-kernel - Re: Abnormal values show up in /proc/allocinfo

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAJuCfpHho8se-f4cnvk0g1YLNzhvG3q8QTYmvMmweUnGAhtA=g@mail.gmail.com>
Date: Mon, 25 Nov 2024 09:31:28 -0800
From: Suren Baghdasaryan <surenb@...gle.com>
To: David Wang <00107082@....com>
Cc: kent.overstreet@...ux.dev, linux-kernel@...r.kernel.org
Subject: Re: Abnormal values show up in /proc/allocinfo

On Mon, Nov 25, 2024 at 2:10 AM David Wang <00107082@....com> wrote:
>
>
> Hi,
>
> Some update
>
> I reproduce the abnormal values of "func:compaction_alloc" today, just once. But I have not found a deterministic procedure yet.
> Seems to me, it happens when kcompactd starts to work
>
>     ret_from_fork(100.000% 57/57)
>         kthread(100.000% 57/57)
>             kcompactd(100.000% 57/57)
>                 compact_node(100.000% 57/57)
>                     compact_zone(100.000% 57/57)
>                         migrate_pages(100.000% 57/57)
>                             migrate_pages_batch(100.000% 57/57)
>                                 compaction_alloc(100.000% 57/57)
>
>
> Maybe, kcompactd mess up information needed by memory tracking? Just a wild guess.
> And those negative signed values, and underflowed unsigned values could also be the side-effect of memory compaction.
> A wilder guess....

Ok, thanks! I think that's enough for me to start digging. Will post
an update once I find something.
Thanks,
Suren.

>
>
>
> FYI
> David
>
>
>
>
> At 2024-11-25 08:35:54, "Suren Baghdasaryan" <surenb@...gle.com> wrote:
> >On Sat, Nov 23, 2024 at 11:43 PM David Wang <00107082@....com> wrote:
> >>
> >> Hi,
> >>
> >> I am running 6.12.0 for a week, and today I notice several strange
> >> items in /proc/allocinfo:
> >>
> >>        -4096 18446744073709551615 mm/filemap.c:3787 func:do_read_cache_folio
> >>  -1946730496 18446744073709076340 mm/filemap.c:1952 func:__filemap_get_folio
> >>   -903294976 18446744073709331085 mm/readahead.c:263 func:page_cache_ra_unbounded
> >>   -353054720 18446744073709465421 mm/shmem.c:1769 func:shmem_alloc_folio
> >>  10547565210        0 mm/compaction.c:1880 func:compaction_alloc
> >>   -156487680 18446744073709513411 mm/memory.c:1064 func:folio_prealloc
> >>  -2422685696 18446744073708960140 mm/memory.c:1062 func:folio_prealloc
> >>  -2332479488 18446744073708982163 fs/btrfs/extent_io.c:635 [btrfs] func:btrfs_alloc_page_array
> >>
> >> some values are way too large, seems like corrupted/uninitialized, and values for compaction_alloc
> >> are inconsistent: non-zero size with zero count.
> >>
> >> I do not know when those data became this strange, and I have not reboot my system yet.
> >> Do you guys need extra information before I reboot my system and started to try reproducing?
> >
> >Hi David,
> >Thanks for reporting. Can you share your .config file? Also, do you
> >see these abnormal values shortly after boot or does it take time for
> >them to get into abnormal state?
> >I'll take a look on Monday and see if there is an obvious issue and if
> >I can reproduce this.
> >Thanks,
> >Suren.
> >
> >>
> >>
> >> Thanks
> >> David
> >>
> >>
> >>
> >>