On Mon 31-10-22 16:51:11, Huang, Ying wrote: > Michal Hocko <mhocko@xxxxxxxx> writes: > > > On Fri 28-10-22 07:22:27, Huang, Ying wrote: > >> Michal Hocko <mhocko@xxxxxxxx> writes: > >> > >> > On Thu 27-10-22 17:31:35, Huang, Ying wrote: > > [...] > >> >> I think that it's possible for different processes have different > >> >> requirements. > >> >> > >> >> - Some processes don't care about where the memory is placed, prefer > >> >> local, then fall back to remote if no free space. > >> >> > >> >> - Some processes want to avoid cross-socket traffic, bind to nodes of > >> >> local socket. > >> >> > >> >> - Some processes want to avoid to use slow memory, bind to fast memory > >> >> node only. > >> > > >> > Yes, I do understand that. Do you have any specific examples in mind? > >> > [...] > >> > >> Sorry, I don't have specific examples. > > > > OK, then let's stop any complicated solution right here then. Let's > > start simple with a per-mm flag to disable demotion of an address > > space. > > I'm not a big fan of per-mm flag. Because we don't have users for that > too and it needs to add ABI too. OK, if there are no users for opt-out then let's jus document the current limitations and be done with it. > > Should there ever be a real demand for a more fine grained solution > > let's go further but I do not think we want a half baked solution > > without real usecases. > > I'm OK to ignore per-task (and missing per-process) memory policy > support for now. I am against such a half baked solution. -- Michal Hocko SUSE Labs
