On Wed, Jan 6, 2021 at 4:14 PM Roman Gushchin <guro@xxxxxx> wrote: > > On Tue, Jan 05, 2021 at 02:58:08PM -0800, Yang Shi wrote: > > The shrinker map management is not really memcg specific, it's just allocation > > In the current form it doesn't look so, especially because each name > has a memcg_ prefix and each function takes a memcg argument. That statement from commit log might be ambiguous and confusing. "Not really memcg specific" doesn't mean it has nothing to do with memcg. It is the intersection between memcg and shrinker. So, I don't think of why it can't take a memcg argument. There are plenty of functions from vmscan.c that take memcg as argument. The direct reason for this consolidation is actually the following patch which uses shrinker_rwsem to protect shrinker_maps allocation. With this code consolidation we could keep the use of shrinker_rwsem in one single file. And it also makes some sense to have shrinker related code in vmscan.c, just like lruvec. > > It begs for some refactorings (Kirill suggested some) and renamings. I apologize that I can't remember what specific suggestions from Kirill you mean. Removing the "memcg_" prefix makes some sense to me, we don't have "memcg_" prefix for lruvec either. > > Thanks!
