Hello All,Forgot to mention the ida-based index allocation, instead of keeping our own bitmap.
This is my new take for the memcg kmem accounting.
At this point, I consider the series pretty mature - although of course,
bugs are always there...
As a disclaimer, however, I must say that the slub code is much more stressed
by me, since I know it better. If you have no more objections to the concepts
presented, the remaining edges can probably be polished in a rc cycle,
at the maintainers discretion, of course.
Otherwise, I'll be happy to address any concerns of yours.
Since last submission:
* memcgs can be properly removed.
* We are not charging based on current->mm->owner instead of current
* kmem_large allocations for slub got some fixes, specially for the free case
* A cache that is registered can be properly removed (common module case)
even if it spans memcg children. Slab had some code for that, now it works
well with both
* A new mechanism for skipping allocations is proposed (patch posted
separately already). Now instead of having kmalloc_no_account, we mark
a region as non-accountable for memcg.