On Thu, 2018-07-12 at 09:49 -0700, Matthew Wilcox wrote: > On Thu, Jul 12, 2018 at 09:04:54AM -0700, James Bottomley wrote: [...] > > The question I'm trying to get an answer to is why does the dentry > > cache need special limits when the mm handling of the page cache > > (and other mm caches) just works? > > I don't know that it does work. Or that it works well. I'm not claiming the general heuristics are perfect (in fact I know we still have a lot of problems with dirty reclaim and writeback). I am willing to bet that any discussion of the heuristics will get a lot of opposition if we try to introduce per-object limits for every object. Our clean cache heuristics are simple: clean caches are easy to reclaim and are thus treated like free memory (there's little cost to filling them or reclaiming them again). There is speculation that this equivalence is problematic because the shrinkers reclaim objects but mm is looking to reclaim pages and thus you can end up with a few objects pinning many pages even if the shrinker freed a lot of them. However, we haven't even reached that level yet ... I'm still struggling to establish that we have a problem with the behaviour of the dentry cache under current mm heuristics. James
