On Sun 31-07-16 11:35:30, Michal Hocko wrote:

On Sat 30-07-16 17:20:30, Tetsuo Handa wrote:

[...]

quoted

But safer way is to get rid of TIF_MEMDIE's triple meanings. The first
one which prevents the OOM killer from selecting next OOM victim was
removed by replacing TIF_MEMDIE test in oom_scan_process_thread() with
tsk_is_oom_victim(). The second one which allows the OOM victims to
deplete 100% of memory reserves wants some changes in order not to
block memory allocations by non OOM victims (e.g. GFP_ATOMIC allocations
by interrupt handlers, GFP_NOIO / GFP_NOFS allocations by subsystems
which are needed for making forward progress of threads in do_exit())
by consuming too much of memory reserves. The third one which blocks
oom_killer_disable() can be removed by replacing TIF_MEMDIE test in
exit_oom_victim() with PFA_OOM_WAITING test like below patch.

I plan to remove TIF_MEMDIE dependency for this as well but I would like
to finish this pile first. We actually do not need any flag for that. We
just need to detect last exiting thread and tsk_is_oom_victim. I have
some preliminary code for that.

That being said. If you _really_ consider this patch to be controversial
I can drop it and handle it with other patches which should handle also
TIF_MEMDIE removal. The rest of the series doesn't really depend on it
in any way. I just though this would be easy enough to carry it with
this pile already. I do not insist on it.
-- 
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>