[CC Dave]
On Tue 27-03-18 15:59:04, Li RongQing wrote:
> when reclaim memory, shink_slab will take lots of time even if
> no memory is reclaimed, since list_lru_count_one called by it
> needs to take a spinlock
>
> try to optimize it by replacing spinlock with RCU in
> __list_lru_count_one
Isn't the RCU overkill here? Why cannot we simply do an optimistic
lockless check for nr_items? It would be racy but does it actually
matter? We should be able to tolerate occasional 0 to non-zero and vice
versa transitions AFAICS.
>
> $dd if=aaa of=bbb bs=1k count=3886080
> $rm -f bbb
> $time echo 100000000 >/cgroup/memory/test/memory.limit_in_bytes
>
> Before: 0m0.415s ===> after: 0m0.395s
>
> Signed-off-by: Li RongQing <lirongq...@baidu.com>
> ---
> include/linux/list_lru.h | 2 ++
> mm/list_lru.c | 69
> ++++++++++++++++++++++++++++++++++--------------
> 2 files changed, 51 insertions(+), 20 deletions(-)
>
> diff --git a/include/linux/list_lru.h b/include/linux/list_lru.h
> index bb8129a3474d..ae472538038e 100644
> --- a/include/linux/list_lru.h
> +++ b/include/linux/list_lru.h
> @@ -29,6 +29,7 @@ struct list_lru_one {
> struct list_head list;
> /* may become negative during memcg reparenting */
> long nr_items;
> + struct rcu_head rcu;
> };
>
> struct list_lru_memcg {
> @@ -46,6 +47,7 @@ struct list_lru_node {
> struct list_lru_memcg *memcg_lrus;
> #endif
> long nr_items;
> + struct rcu_head rcu;
> } ____cacheline_aligned_in_smp;
>
> struct list_lru {
> diff --git a/mm/list_lru.c b/mm/list_lru.c
> index fd41e969ede5..4c58ed861729 100644
> --- a/mm/list_lru.c
> +++ b/mm/list_lru.c
> @@ -52,13 +52,13 @@ static inline bool list_lru_memcg_aware(struct list_lru
> *lru)
> static inline struct list_lru_one *
> list_lru_from_memcg_idx(struct list_lru_node *nlru, int idx)
> {
> - /*
> - * The lock protects the array of per cgroup lists from relocation
> - * (see memcg_update_list_lru_node).
> - */
> - lockdep_assert_held(&nlru->lock);
> - if (nlru->memcg_lrus && idx >= 0)
> - return nlru->memcg_lrus->lru[idx];
> + struct list_lru_memcg *tmp;
> +
> + WARN_ON_ONCE(!rcu_read_lock_held());
> +
> + tmp = rcu_dereference(nlru->memcg_lrus);
> + if (tmp && idx >= 0)
> + return rcu_dereference(tmp->lru[idx]);
>
> return &nlru->lru;
> }
> @@ -113,14 +113,17 @@ bool list_lru_add(struct list_lru *lru, struct
> list_head *item)
> struct list_lru_one *l;
>
> spin_lock(&nlru->lock);
> + rcu_read_lock();
> if (list_empty(item)) {
> l = list_lru_from_kmem(nlru, item);
> list_add_tail(item, &l->list);
> l->nr_items++;
> nlru->nr_items++;
> + rcu_read_unlock();
> spin_unlock(&nlru->lock);
> return true;
> }
> + rcu_read_unlock();
> spin_unlock(&nlru->lock);
> return false;
> }
> @@ -133,14 +136,17 @@ bool list_lru_del(struct list_lru *lru, struct
> list_head *item)
> struct list_lru_one *l;
>
> spin_lock(&nlru->lock);
> + rcu_read_lock();
> if (!list_empty(item)) {
> l = list_lru_from_kmem(nlru, item);
> list_del_init(item);
> l->nr_items--;
> nlru->nr_items--;
> + rcu_read_unlock();
> spin_unlock(&nlru->lock);
> return true;
> }
> + rcu_read_unlock();
> spin_unlock(&nlru->lock);
> return false;
> }
> @@ -166,12 +172,13 @@ static unsigned long __list_lru_count_one(struct
> list_lru *lru,
> {
> struct list_lru_node *nlru = &lru->node[nid];
> struct list_lru_one *l;
> - unsigned long count;
> + unsigned long count = 0;
>
> - spin_lock(&nlru->lock);
> + rcu_read_lock();
> l = list_lru_from_memcg_idx(nlru, memcg_idx);
> - count = l->nr_items;
> - spin_unlock(&nlru->lock);
> + if (l)
> + count = l->nr_items;
> + rcu_read_unlock();
>
> return count;
> }
> @@ -204,6 +211,7 @@ __list_lru_walk_one(struct list_lru *lru, int nid, int
> memcg_idx,
> unsigned long isolated = 0;
>
> spin_lock(&nlru->lock);
> + rcu_read_lock();
> l = list_lru_from_memcg_idx(nlru, memcg_idx);
> restart:
> list_for_each_safe(item, n, &l->list) {
> @@ -250,6 +258,7 @@ __list_lru_walk_one(struct list_lru *lru, int nid, int
> memcg_idx,
> }
> }
>
> + rcu_read_unlock();
> spin_unlock(&nlru->lock);
> return isolated;
> }
> @@ -296,9 +305,14 @@ static void __memcg_destroy_list_lru_node(struct
> list_lru_memcg *memcg_lrus,
> int begin, int end)
> {
> int i;
> + struct list_lru_one *tmp;
>
> - for (i = begin; i < end; i++)
> - kfree(memcg_lrus->lru[i]);
> + for (i = begin; i < end; i++) {
> + tmp = memcg_lrus->lru[i];
> + rcu_assign_pointer(memcg_lrus->lru[i], NULL);
> + if (tmp)
> + kfree_rcu(tmp, rcu);
> + }
> }
>
> static int __memcg_init_list_lru_node(struct list_lru_memcg *memcg_lrus,
> @@ -314,7 +328,7 @@ static int __memcg_init_list_lru_node(struct
> list_lru_memcg *memcg_lrus,
> goto fail;
>
> init_one_lru(l);
> - memcg_lrus->lru[i] = l;
> + rcu_assign_pointer(memcg_lrus->lru[i], l);
> }
> return 0;
> fail:
> @@ -325,25 +339,37 @@ static int __memcg_init_list_lru_node(struct
> list_lru_memcg *memcg_lrus,
> static int memcg_init_list_lru_node(struct list_lru_node *nlru)
> {
> int size = memcg_nr_cache_ids;
> + struct list_lru_memcg *tmp;
>
> - nlru->memcg_lrus = kvmalloc(size * sizeof(void *), GFP_KERNEL);
> - if (!nlru->memcg_lrus)
> + tmp = kvmalloc(size * sizeof(void *), GFP_KERNEL);
> + if (!tmp)
> return -ENOMEM;
>
> - if (__memcg_init_list_lru_node(nlru->memcg_lrus, 0, size)) {
> - kvfree(nlru->memcg_lrus);
> + if (__memcg_init_list_lru_node(tmp, 0, size)) {
> + kvfree(tmp);
> return -ENOMEM;
> }
>
> + rcu_assign_pointer(nlru->memcg_lrus, tmp);
> +
> return 0;
> }
>
> -static void memcg_destroy_list_lru_node(struct list_lru_node *nlru)
> +static void memcg_destroy_list_lru_node_rcu(struct rcu_head *rcu)
> {
> + struct list_lru_node *nlru;
> +
> + nlru = container_of(rcu, struct list_lru_node, rcu);
> +
> __memcg_destroy_list_lru_node(nlru->memcg_lrus, 0, memcg_nr_cache_ids);
> kvfree(nlru->memcg_lrus);
> }
>
> +static void memcg_destroy_list_lru_node(struct list_lru_node *nlru)
> +{
> + call_rcu(&nlru->rcu, memcg_destroy_list_lru_node_rcu);
> +}
> +
> static int memcg_update_list_lru_node(struct list_lru_node *nlru,
> int old_size, int new_size)
> {
> @@ -371,9 +397,10 @@ static int memcg_update_list_lru_node(struct
> list_lru_node *nlru,
> * we have to use IRQ-safe primitives here to avoid deadlock.
> */
> spin_lock_irq(&nlru->lock);
> - nlru->memcg_lrus = new;
> + rcu_assign_pointer(nlru->memcg_lrus, new);
> spin_unlock_irq(&nlru->lock);
>
> + synchronize_rcu();
> kvfree(old);
> return 0;
> }
> @@ -487,6 +514,7 @@ static void memcg_drain_list_lru_node(struct
> list_lru_node *nlru,
> * we have to use IRQ-safe primitives here to avoid deadlock.
> */
> spin_lock_irq(&nlru->lock);
> + rcu_read_lock();
>
> src = list_lru_from_memcg_idx(nlru, src_idx);
> dst = list_lru_from_memcg_idx(nlru, dst_idx);
> @@ -495,6 +523,7 @@ static void memcg_drain_list_lru_node(struct
> list_lru_node *nlru,
> dst->nr_items += src->nr_items;
> src->nr_items = 0;
>
> + rcu_read_unlock();
> spin_unlock_irq(&nlru->lock);
> }
>
> --
> 2.11.0
--
Michal Hocko
SUSE Labs
