2020년 8월 26일 (수) 오전 9:42, Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>님이 작성: > > On Tue, 25 Aug 2020 14:34:32 +0900 Joonsoo Kim <js1304@xxxxxxxxx> wrote: > > > > > > > That's a bunch more code on a very hot path to serve an obscure feature > > > which has a single obscure callsite. > > > > > > Can we instead put the burden on that callsite rather than upon > > > everyone? For (dumb) example, teach __gup_longterm_locked() to put the > > > page back if it's CMA and go get another one? > > > > Hmm... Unfortunately, it cannot ensure that we eventually get the non-CMA page. > > I think that the only way to ensure it is to implement the > > functionality here. We can > > use 'unlikely' or 'static branch' to reduce the overhead for a really > > rare case but > > for now I have no idea how to completely remove the overhead. > > Gee, there must be something? Provide the gup code with a special > entry point which takes the page straight from __rmqueue() and bypasses > the pcp lists? Hmm... it seems not possible. It's allocation context API and maybe actual allocation happens in handle_mm_fault() or it's successor. We cannot use a special entry point for allocation there since it's a general function. And, IMHO, making a special allocation function that bypasses the pcp list would not be a good practice. Thanks.
