Re: mm: kernel BUG at mm/huge_memory.c:2785!

[Kirill A. Shutemov]

On Thu, Mar 06, 2014 at 08:32:48PM -0500, Sasha Levin wrote:
> On 03/04/2014 10:53 PM, Sasha Levin wrote:
> >On 03/04/2014 10:16 PM, Bob Liu wrote:
> >>On Thu, Feb 27, 2014 at 11:03 PM, Kirill A. Shutemov
> >><kirill.shutemov@xxxxxxxxxxxxxxx> wrote:
> >>>Sasha Levin wrote:
> >>>>Hi all,
> >>>>
> >>>>While fuzzing with trinity inside a KVM tools guest running latest -next kernel I've stumbled on the
> >>>>following spew:
> >>>>
> >>>>[ 1428.146261] kernel BUG at mm/huge_memory.c:2785!
> >>>
> >>>Hm, interesting.
> >>>
> >>>It seems we either failed to split huge page on vma split or it
> >>>materialized from under us. I don't see how it can happen:
> >>>
> >>>   - it seems we do the right thing with vma_adjust_trans_huge() in
> >>>     __split_vma();
> >>>   - we hold ->mmap_sem all the way from vm_munmap(). At least I don't see
> >>>     a place where we could drop it;
> >>>
> >>
> >>Enable CONFIG_DEBUG_VM may show some useful information, at least we
> >>can confirm weather rwsem_is_locked(&tlb->mm->mmap_sem) before
> >>split_huge_page_pmd().
> >
> >I have CONFIG_DEBUG_VM enabled and that code you're talking is not triggering, so mmap_sem
> >is locked.
> 
> Guess what. I've just hit it.

I think this particular traceback is not a real problem: by time of
exit_mm() we shouldn't race with anybody for the mm_struct.

We probably could drop ->mmap_sem later in mmput() rather then in
exit_mm() to fix this false positive.

> It's worth keeping in mind that this is the first time I see it.

Hm. That's strange exit_mmap() is called without holding ->mmap_sem.

-- 
 Kirill A. Shutemov
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/