On Fri, 26 Aug 2011, Konstantin Khlebnikov wrote:
A much better solution would be to lower the badness score that the oom
killer uses for PF_FROZEN threads so that they aren't considered a
priority for kill unless there's nothing else left to kill.
Anyway, oom killer shouldn't loop endlessly if it see TIF_MEMDIE on frozen
task,
it must go on and try to kill somebody else. We cannot wait for thawing this
task.
Did you read my suggestion? I quoted it above again for you. The badness
heuristic would only select those tasks to kill as a last resort in the
hopes they will eventually be thawed and may exit. Panicking the entire
machine for what could be isolated by a cgroup is insanity.