Rather a lot of wordwrapping here, but the patch applied OK nonetheless.
This means that all the code was within 80 columns ;)
I assume this trick is here to minimise runtime overhead on systems which
never use cgroups?
It'd be nice if the changelog were to describe what that overhead is.
Then I'd know whether to ask whether we should look at clearing
need_mm_owner_callback when the system stops using cgroups. Somehow.
But this comment rather duplicates the above one.
Potentially-long tasklist_lock hold times are a concern. I don't suppose
rcu can save us?
Some additional commentary fleshing out "We should not get here often"
might set minds at ease. How not-often? Under which circumstances?
afaict no callbacks are implemented in this patch? What's the plan here?
Should be covered in the changelog, please.
The mem_cgroup_zoneinfo() will immediately and reliably oops if mem_cont is
NULL, won't it? If so, the BUG_ON() is redundant?