CCing memcg maintainers.

On Wed, Aug 06, 2025 at 03:00:43PM +0100, Matthew Wilcox wrote:

On Tue, Aug 05, 2025 at 05:11:46PM -0700, Boris Burkov wrote:

quoted

I would like to revisit Qu's proposal to not charge btrfs extent_buffer
allocations to the user's cgroup.

https://lore.kernel.org/linux-mm/b5fef5372ae454a7b6da4f2f75c427aeab6a07d6.1727498749.git.wqu@suse.com/ (local)

I prefer Qu's suggestion to add a flag to the address_space.  This really
is a property of the address_space, not a property of the call-site.

I think Michal wanted call-site (explicit interface) for easy
searchability. I don't have a strong opinion either way. However I
wonder if having this information in address_space might be more useful.
At the moment, this series is using !folio_memcg(folio) to detect if the
folio has skipped memcg charging on the free path to decrement the stat.
Having information in address_space would be another way to extract that
information. I need to think more on this.