According to my traces this 43ms could drop to the average of 11ms and
worst case 25ms if throttle_direct_reclaim would return true when
fatal signal is pending but I would like to hear your opinion about
throttle_direct_reclaim logic.
Digging some more into this I realize my last statement might be incorrect. Throttling in this situation might not help with the signal handling delay because of the logic in __alloc_pages_slowpath. I'll have to experiment with this first, please disregard that last statement for now.
