Re: [PATCH] [BUGFIX] update mm->owner even if no next owner.

[Johannes Weiner]

On Sat, Jun 11, 2011 at 06:39:43PM +0200, Andrea Arcangeli wrote:
> On Sat, Jun 11, 2011 at 09:04:14AM -0700, Hugh Dickins wrote:
> > I had another go at reproducing it, 2 hours that time, then a try with
> > 692e0b35427a reverted: it ran overnight for 9 hours when I stopped it.
> > 
> > Andrea, please would you ask Linus to revert that commit before -rc3?
> > Or is there something else you'd like us to try instead?  I admit that
> > I've not actually taken the time to think through exactly how it goes
> > wrong, but it does look dangerous.
> 
> Here I was asked if the mem_cgroup_newpage_charge need the mmap_sem at
> all. And if not why not to release the mmap_sem early.
> 
> https://lkml.org/lkml/2011/3/14/276
> 
> So I didn't see why mmap_sem was needed, I also asked confirmation and
> who answered agreed it was safe without mmap_sem even if it's the only
> place doing that. Maybe that assumption was wrong and we need
> mmap_sem after all if this commit is causing problems.
> 
> Or did you find something wrong in the actual patch?
> 
> Do I understand right that the bug just that we must run
> alloc_hugepage_vma+mem_cgroup_newpage_charge within the same critical
> section protected by the mmap_sem read mode? Do we know why?

The problem is that mm->owner points to a stale task structure if the
last possible owner is exiting.  The mmap_sem just prevented the task
from actually exiting through write-acquiring the mmap_sem in
khugepaged_exit().

I think enforcing lifetime of an object through locks is not the
nicest thing to do, so I stand by what I wrote in the mail you linked
to above :) and agree with Kame that mm->owner should just not point
to a stale task struct.  The memcg code can handle it going NULL.

--
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/