Messages in this thread | | | Date | Fri, 17 Dec 2010 09:01:03 +0900 | From | KAMEZAWA Hiroyuki <> | Subject | Re: [PATCH] mm: add replace_page_cache_page() function |
| |
On Thu, 16 Dec 2010 13:05:44 +0100 Miklos Szeredi <miklos@szeredi.hu> wrote:
> On Thu, 16 Dec 2010, KAMEZAWA Hiroyuki wrote:
> > Hmm, then, the page will be recharged to "current" instead of the memcg > > where "old" was under control. Is this design ? If so, why ? > > No, I just haven't thought about it. > > Porbably charging "new" to where "old" was charged is the logical > thing to do here. > > > > > In mm/migrate.c, following is called. > > > > charge = mem_cgroup_prepare_migration(page, newpage, &mem); > > ....do migration.... > > if (!charge) > > mem_cgroup_end_migration(mem, page, newpage); > > > > BTW, off topic, in fuse/dev.c > > > > add_to_page_cache_locked(page) > > This is the call which the above patch replaces with > replace_page_cache_page(). So if I fix replace_page_cache_page() to > charge "newpage" to the correct memory cgroup, that should solve all > problems, no? > No. memory cgroup expects all pages should be found on LRU. But, IIUC, pages on this radix-tree will not be on LRU. So, memory cgroup can't find it at destroying cgroup and can't reduce "usage" of resource to be 0. This makes rmdir() returns -EBUSY.
I'm sorry if this page will be on LRU, somewhere.
Thanks, -Kame
| |