Re: [PATCH 2/2] mm, oom: fix potential data corruption when oom_reaper races with writer

[Michal Hocko <mhocko@kernel.org>]

On Thu 10-08-17 10:21:18, Michal Hocko wrote:
> On Tue 08-08-17 19:48:55, Andrea Arcangeli wrote:
> [...]
> > The bug corrected by this patch 1/2 I pointed it out last week while
> > reviewing other oom reaper fixes so that looks fine.
> > 
> > However I'd prefer to dump MMF_UNSTABLE for good instead of adding
> > more of it. It can be replaced with unmap_page_range in
> > __oom_reap_task_mm with a function that arms a special migration entry
> > so that no branchs are added to the fast paths and it's all hidden
> > inside is_migration_entry slow paths.
> 
> This sounds like an interesting idea but I would like to address the
> _correctness_ issue first and optimize on top of it. If for nothing else
> backporting a follow up fix sounds easier than a complete rework. There
> are quite some callers of is_migration_entry and the patch won't be
> trivial either. So can we focus on the fix first please?

Btw, if the overhead is a concern then we can add a jump label and only
make the code active only while the OOM is in progress. We already do
count all oom victims so we have a clear entry and exit points. This
would still sound easier to do than teach every is_migration_entry a new
migration entry type and handle it properly, not to mention make
everybody aware of this for future callers of is_migration_entry.
-- 
Michal Hocko
SUSE Labs