Re: [syzbot] [mm?] kernel BUG in collapse_scan_file

[Lorenzo Stoakes (Oracle)]

On Thu, Mar 19, 2026 at 11:59:57AM +0100, David Hildenbrand (Arm) wrote:
> > We shouldn't, really, be just ignoring splatting kernels like that. At least
> > that's my personal point of view on it.
>
> Yes, we shouldn't.
>
> >
> > But anyway I really don't have the time or energy to try to track this down or
> > push on this further.
>
> The problem I'm having is that reverting the commit might hide a real
> problem, and understanding weather there is a real problem requires ...
> real work.
>
> Willy said
>
> "It *might* still be safe in this instance; I'll look
> at this carefully for a bit and decide how best to fix it."
>
> That likely didn't happen.

Yep, and in that case given the assert is suspect, the correct thing is to drop
the introduction of the potentially-buggy asserting logic rather than tear our
hair out as to whether there is something real or not.

>
> Later he said:
>
> "But this is a long and complicated function (over 400 lines!) and I
> don't know if fixing this one way or the other would serve to make other
> bugs more likely or expose some future problem to the debugging code."
>
> with
>
> "This isn't the kind of thing where you can just jump in with a one line
> patch and actually be helpful, sorry."


Yup, so again really I don't think the kernel should take the xarray debug patch
until it is discussed and verified as correct on-list.

See the discussion with Vlastimil.

>
> So I am clueless if there is a real problem there or whether it's just a
> false positive. And whether (if it's a real problem) the simple fix
> would actually be a good temporary fix ("make other bugs more likely").
>
> I can try finding some time to look into that, but pagecache code is not
> particularly the code I'm familiar with ...

I don't think we should waste our time on it, willy's series should be yoinked
until it can prove itself for-sure well behaved.

>
> --
> Cheers,
>
> David

Thanks, Lorenzo