Re: Occasional 'WARNING: mm/backing-dev.c:636 at cgwb_release_workfn' when running mdadm
Luca Boccassi <[email protected]>
| Newsgroups | dev.linux.lists.regressions,org.kernel.vger.linux-fsdevel,org.kvack.linux-mm |
|---|---|
| Message-ID | <CAMw=ZnR4TaZvobAE5Pr+9MUGBswTmHdus5uPJ_Ae+dGpDuNoMw@mail.gmail.com> |
On Tue, 26 May 2026 at 15:14, Jan Kara <[email protected]> wrote: > > Hi! > > On Tue 26-05-26 13:01:11, Luca Boccassi wrote: > > Hi, I've been told to report this here. > > > > In our CI test that runs some mdadm commands we occasionally (maybe 1 > > in 100 runs, not reproducible manually) see the following kernel warn: > > > > [ 15.769393] TEST-64-UDEV-STORAGE.sh[245]: + mdadm -v --stop /dev/md/mdparity > > [ 15.786033] ------------[ cut here ]------------ > > [ 15.788632] !list_empty(&wb->b_attached) > > [ 15.788634] WARNING: mm/backing-dev.c:636 at > > Thanks for report! Interesting. I haven't seen this yet but I suspect my > rework of cgwb inode switching (happened in 6.18) could be responsible for > this. > > Since the 7.0.11 kernel we've got a couple of fixups to this code. In > particular 6689f01d6740 ("writeback: Fix use after free in > inode_switch_wbs_work_fn()") and now also fixes sitting in > https://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git/log/?h=vfs-7.2.writeback > > Your script seems to be exercising mount/umount so the above fixes could > indeed be related to your problem. If you can run with a kernel built from > the above branch, it would be great if you can verify the problem got > indeed fixed. Thanks, it's very occasional so can't reproduce it manually, but once that new kernel version starts appearing in arch/rawhide, I'll report back if I see the same warn again