Re: mm: BUG in expand_downwards

From: Dmitry Vyukov
Date: Wed Jan 27 2016 - 16:12:15 EST


On Wed, Jan 27, 2016 at 8:41 PM, Oleg Nesterov <oleg@xxxxxxxxxx> wrote:
> On 01/27, Dmitry Vyukov wrote:
>>
>> On Wed, Jan 27, 2016 at 1:24 PM, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
>> > On Wed, Jan 27, 2016 at 12:49 PM, Konstantin Khlebnikov
>> > <koct9i@xxxxxxxxx> wrote:
>> >> It seems anon_vma appeared between lock and unlock.
>> >>
>> >> This should fix the bug and make code faster (write lock isn't required here)
>> >>
>> >> --- a/mm/mmap.c
>> >> +++ b/mm/mmap.c
>> >> @@ -453,12 +453,16 @@ static void validate_mm(struct mm_struct *mm)
>> >> struct vm_area_struct *vma = mm->mmap;
>> >>
>> >> while (vma) {
>> >> + struct anon_vma *anon_vma = vma->anon_vma;
>> >> struct anon_vma_chain *avc;
>> >>
>> >> - vma_lock_anon_vma(vma);
>> >> - list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
>> >> - anon_vma_interval_tree_verify(avc);
>> >> - vma_unlock_anon_vma(vma);
>> >> + if (anon_vma) {
>> >> + anon_vma_lock_read(anon_vma);
>> >> + list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
>> >> + anon_vma_interval_tree_verify(avc);
>> >> + anon_vma_unlock_read(anon_vma);
>> >> + }
>> >> +
>> >> highest_address = vma->vm_end;
>> >> vma = vma->vm_next;
>> >> i++;
>> >
>> >
>> > Now testing with this patch. Thanks for quick fix!
>>
>>
>> Hit the same BUG with this patch.
>
> Do you mean the same "bad unlock balance detected" BUG? this should be "obviously"
> fixed by the patch above...
>
> Or you mean the 2nd VM_BUG_ON_MM() ?
>
>> Please try to reproduce it locally and test.
>
> tried to reproduce, doesn't work.


Sorry, I meant only the second once. The mm bug.
I guess you need at least CONFIG_DEBUG_VM. Run it in a tight parallel
loop with CPU oversubscription (e.g. 32 parallel processes on 2 cores)
for at least an hour.