On Thu, 29 Jul 2010 11:03:26 -0700
Greg Thelen <gthelen@xxxxxxxxxx> wrote:
> KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> writes:
>
> > From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
> >
> > Now, memory cgroup has an ID per cgroup and make use of it at
> > - hierarchy walk,
> > - swap recording.
> >
> > This patch is for making more use of it. The final purpose is
> > to replace page_cgroup->mem_cgroup's pointer to an unsigned short.
> >
> > This patch caches a pointer of memcg in an array. By this, we
> > don't have to call css_lookup() which requires radix-hash walk.
> > This saves some amount of memory footprint at lookup memcg via id.
> > It's called in very fast path and need to be quick AMAP.
> >
> > Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
> > ---
> > init/Kconfig | 11 +++++++++++
> > mm/memcontrol.c | 46 ++++++++++++++++++++++++++++++++--------------
> > 2 files changed, 43 insertions(+), 14 deletions(-)
> >
> > Index: mmotm-0727/mm/memcontrol.c
> > ===================================================================
> > --- mmotm-0727.orig/mm/memcontrol.c
> > +++ mmotm-0727/mm/memcontrol.c
> > @@ -292,6 +292,27 @@ static bool move_file(void)
> > &mc.to->move_charge_at_immigrate);
> > }
> >
> > +atomic_t mem_cgroup_num;
>
> Maybe static and init?
> + static atomic_t mem_cgroup_num = ATOMIC_INIT(0);
>
IIUC, "0" is not required to be initialized. but ok, this is static.
> > +struct mem_cgroup *mem_cgroups[CONFIG_MEM_CGROUP_MAX_GROUPS] __read_mostly;
>
> Make this static?
>
> Because value [0] is reserved, maybe this should be:
> +struct mem_cgroup *mem_cgroups[CONFIG_MEM_CGROUP_MAX_GROUPS+1] __read_mostly;
>
Hmm ? I don't like this. I'll write "0" is unused in CONFIG.
> > +
> > +static struct mem_cgroup* id_to_memcg(unsigned short id)
> > +{
> > + /*
> > + * This array is set to NULL when mem_cgroup is freed.
> > + * IOW, there are no more references && rcu_synchronized().
> > + * This lookup-caching is safe.
> > + */
> > + if (unlikely(!mem_cgroups[id])) {
> > + struct cgroup_subsys_state *css;
> > + rcu_read_lock();
> > + css = css_lookup(&mem_cgroup_subsys, id);
> > + if (!css)
> > + return NULL;
> > + rcu_read_unlock();
>
> I think we should move rcu_read_unlock() above to just before "if
> (!css)" to unlock rcu when returning NULL.
>
yes.
> > + mem_cgroups[id] = container_of(css, struct mem_cgroup, css);
> > + }
> > + return mem_cgroups[id];
> > +}
> > /*
> > * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
> > * limit reclaim to prevent infinite loops, if they ever occur.
> > @@ -1824,18 +1845,7 @@ static void mem_cgroup_cancel_charge(str
> > * it's concern. (dropping refcnt from swap can be called against removed
> > * memcg.)
> > */
> > -static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
> > -{
> > - struct cgroup_subsys_state *css;
> >
> > - /* ID 0 is unused ID */
> > - if (!id)
> > - return NULL;
> > - css = css_lookup(&mem_cgroup_subsys, id);
> > - if (!css)
> > - return NULL;
> > - return container_of(css, struct mem_cgroup, css);
> > -}
> >
> > struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
> > {
> > @@ -1856,7 +1866,7 @@ struct mem_cgroup *try_get_mem_cgroup_fr
> > ent.val = page_private(page);
> > id = lookup_swap_cgroup(ent);
> > rcu_read_lock();
> > - mem = mem_cgroup_lookup(id);
> > + mem = id_to_memcg(id);
> > if (mem && !css_tryget(&mem->css))
> > mem = NULL;
> > rcu_read_unlock();
> > @@ -2208,7 +2218,7 @@ __mem_cgroup_commit_charge_swapin(struct
> >
> > id = swap_cgroup_record(ent, 0);
> > rcu_read_lock();
> > - memcg = mem_cgroup_lookup(id);
> > + memcg = id_to_memcg(id);
> > if (memcg) {
> > /*
> > * This recorded memcg can be obsolete one. So, avoid
> > @@ -2472,7 +2482,7 @@ void mem_cgroup_uncharge_swap(swp_entry_
> >
> > id = swap_cgroup_record(ent, 0);
> > rcu_read_lock();
> > - memcg = mem_cgroup_lookup(id);
> > + memcg = id_to_memcg(id);
> > if (memcg) {
> > /*
> > * We uncharge this because swap is freed.
> > @@ -3988,6 +3998,10 @@ static struct mem_cgroup *mem_cgroup_all
> > struct mem_cgroup *mem;
> > int size = sizeof(struct mem_cgroup);
> >
> > + /* 0 is unused */
> > + if (atomic_read(&mem_cgroup_num) == CONFIG_MEM_CGROUP_MAX_GROUPS-1)
> > + return NULL;
> > +
> > /* Can be very big if MAX_NUMNODES is very big */
> > if (size < PAGE_SIZE)
> > mem = kmalloc(size, GFP_KERNEL);
> > @@ -4025,7 +4039,10 @@ static void __mem_cgroup_free(struct mem
> > int node;
> >
> > mem_cgroup_remove_from_trees(mem);
> > + /* No more lookup against this ID */
> > + mem_cgroups[css_id(&mem->css)] = NULL;
>
> Are css_id() values tightly packed? If there are 4 memcg allocated, are
> we guaranteed that all of their id's have value 1..4?
>
No. It can be sparse.
Thanks,
-Kame
--
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>