| Message ID | 20120423082835.GA32359@lizard |
|---|---|
| State | New |
| Headers | show |
(2012/04/23 17:28), Anton Vorontsov wrote: > ...otherwise the we're getting the wrong numbers in usage_in_bytes. > > On Mon, Mar 05, 2012 at 09:19:34AM +0900, KAMEZAWA Hiroyuki wrote: > [...] >>> diff --git a/mm/memcontrol.c b/mm/memcontrol.c >>> index 228d646..c8abdc5 100644 >>> --- a/mm/memcontrol.c >>> +++ b/mm/memcontrol.c >>> @@ -3812,6 +3812,9 @@ static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) >>> >>> val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); >>> val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); >>> + val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); >>> >>> 1. Is there any particular reason we don't currently account file mapped >>> memory in usage_in_bytes? >>> >>> To me, MEM_CGROUP_STAT_FILE_MAPPED hunk seems logical even if we >>> don't use it for lowmemory notifications. >>> >>> Plus, it seems that FILE_MAPPED _is_ accounted for the non-root >>> cgroups, so I guess it's clearly a bug for the root memcg? >> >> CACHE includes all file caches. Why do you think FILE_MAPPED is not included in CACHE ? > > There were tons of changes in the memcg lately, but I believe the issue > is still there. > > For example, looking into this code flow: > > -> page_add_file_rmap() (mm/rmap.c) > -> mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED) (include/linux/memcontrol.h) > -> void mem_cgroup_update_page_stat(page, MEMCG_NR_FILE_MAPPED, 1) (mm/memcontrol.c) > > And then: > > void mem_cgroup_update_page_stat(struct page *page, > enum mem_cgroup_page_stat_item idx, int val) > { > ... > switch (idx) { > case MEMCG_NR_FILE_MAPPED: > idx = MEM_CGROUP_STAT_FILE_MAPPED; > break; > default: > BUG(); > } > > this_cpu_add(memcg->stat->count[idx], val); > ... > } > > So, clearly, this function only bothers updating _FILE_MAPPED only, > leaving _CACHE alone. > > If you're saying that _CACHE meant to include _FILE_MAPPED, then > I guess the patch down below would be a proper fix then... Otherwise > we need to be consistent on stats reporting, and either fall-back > to my original fix (in mem_cgroup_usage()), or think about doing it > some other way... > NACK. CACHE is updated at charge()/uncharge()...inserting/removing page cache to radix-tree. Thanks, -Kame > Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org> > --- > > The patch is against current -next. > > Thanks, > > mm/memcontrol.c | 2 ++ > 1 file changed, 2 insertions(+) > > diff --git a/mm/memcontrol.c b/mm/memcontrol.c > index 884e936..760ecf5 100644 > --- a/mm/memcontrol.c > +++ b/mm/memcontrol.c > @@ -1958,6 +1958,8 @@ void mem_cgroup_update_page_stat(struct page *page, > > switch (idx) { > case MEMCG_NR_FILE_MAPPED: > + idx = MEM_CGROUP_STAT_CACHE; > + this_cpu_add(memcg->stat->count[idx], val); > idx = MEM_CGROUP_STAT_FILE_MAPPED; > break; > default:
On Mon, Apr 23, 2012 at 05:35:12PM +0900, KAMEZAWA Hiroyuki wrote: [...] > > For example, looking into this code flow: > > > > -> page_add_file_rmap() (mm/rmap.c) > > -> mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED) (include/linux/memcontrol.h) > > -> void mem_cgroup_update_page_stat(page, MEMCG_NR_FILE_MAPPED, 1) (mm/memcontrol.c) > > > > And then: > > > > void mem_cgroup_update_page_stat(struct page *page, > > enum mem_cgroup_page_stat_item idx, int val) > > { > > ... > > switch (idx) { > > case MEMCG_NR_FILE_MAPPED: > > idx = MEM_CGROUP_STAT_FILE_MAPPED; > > break; > > default: > > BUG(); > > } > > > > this_cpu_add(memcg->stat->count[idx], val); > > ... > > } > > > > So, clearly, this function only bothers updating _FILE_MAPPED only, > > leaving _CACHE alone. [...] > > NACK. > CACHE is updated at charge()/uncharge()...inserting/removing page cache to radix-tree. Interesting; true, we have charge/uncharge in __do_fault()/do_wp_page and friends. So, we seem to update FILE_MAPPED in the rmap via cgroup_dec/inc_page_stat, and CACHE is updated via charge/uncharge. Hm. The code in memory.c is full of if/else ifs, and I wonder if there's some discrepancy in there, but briefly looking it looks fine. The code looks correct indeed, but I'm getting the wrong stats. :-/ Thanks,
diff --git a/mm/memcontrol.c b/mm/memcontrol.c index 884e936..760ecf5 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -1958,6 +1958,8 @@ void mem_cgroup_update_page_stat(struct page *page, switch (idx) { case MEMCG_NR_FILE_MAPPED: + idx = MEM_CGROUP_STAT_CACHE; + this_cpu_add(memcg->stat->count[idx], val); idx = MEM_CGROUP_STAT_FILE_MAPPED; break; default: