On Tue, 30 Mar 2010 15:29:58 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote: > On Tue, 30 Mar 2010 14:44:58 +0900 > KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote: > > > On Tue, 30 Mar 2010 14:30:38 +0900 > > Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx> wrote: > > > > > On Tue, 30 Mar 2010 10:30:50 +0530, Balbir Singh <balbir@xxxxxxxxxxxxxxxxxx> wrote: > > > > * KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> [2010-03-30 13:51:59]: > > > > Yep, I tend to agree, but I need to take a closer look again at the > > > > patches. > > > > > > > I agree it would be more simple. I selected the current policy because > > > I was not sure whether we should move file caches(!tmpfs) with mapcount > 1, > > > and, IMHO, shared memory and file caches are different for users. > > > But it's O.K. for me to change current policy. > > > > > > > To explain what I think of, I wrote a patch onto yours. (Maybe overkill for explaination ;) > > > > Summary. > > > > + adding move_anon, move_file, move_shmem information to move_charge_struct. > > + adding hanlders for each pte types. > > + checking # of referer should be divided to each type. > > It's complicated to catch all cases in one "if" sentense. > > + FILE pages will be moved if it's charged against "from". no mapcount check. > > i.e. FILE pages should be moved even if it's not page-faulted. > > + ANON pages will be moved if it's really private. > > > > For widely shared FILE, "if it's charged against "from"" is enough good limitation. > > > > > > Hmm....how about changing meanings of new flags ? > > 1 : a charge of page caches are moved. Page cache means cache of regular files > and shared memory. But only privately mapped pages (mapcount==1) are moved. > > 2 : a charge of page caches are moved. Page cache means cache of regular files > and shared memory. They are moved even if it's shared among processes. > > When both of 1 and 2 are specified, "2" is used. Anonymous pages will not be > moved if it's shared. > > Then, total view of user interface will be simple and I think this will allow > what you want. > Thank you for your suggestion. It would be simple. I'll try in that way. Thanks, Daisuke Nishimura. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>