The first 2 patches were silently ignored, resend. I don't really understand
who should be cc'ed on this.
The third one is an example. Obviously we can fix the problem without changing
use_mm/PF_BORROWED_MM, but in that case we must do task_lock/unlock for each
process in the system, not good and annoying.
Oleg.
I like all three of these changes. The 1/3 and 2/3 together constitute a
fix for the coredump vs aio race that is nice and clean. The 2/2 is all
cleanup or possibly fixes more arcane cases we haven't enumerated yet.
(I would have made that the last patch of the three.) It merits some
follow-up looking at all (!task->mm) sorts of tests and making them use
a helper for what they really mean to express (e.g. is_kernel_thread()
if that's what they intend).
Thanks,
Roland