Actually, when using a boot time kernel option "cgroup.memory=nokmem",
all lru items are inserted to list_lru_node. But for those users who
invoke list_lru_init_memcg() to initialize list_lru, list_lru_memcg_aware()
returns true. And this brings unneeded operations related to memcg.
To make things more convenient, let's disable memcg_aware when
cgroup.memory is set to "nokmem".
Signed-off-by: Haifeng Xu <[email protected]>
---
mm/list_lru.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/mm/list_lru.c b/mm/list_lru.c
index a05e5bef3b40..86f63c92fd88 100644
--- a/mm/list_lru.c
+++ b/mm/list_lru.c
@@ -566,6 +566,9 @@ int __list_lru_init(struct list_lru *lru, bool memcg_aware,
lru->shrinker_id = shrinker->id;
else
lru->shrinker_id = -1;
+
+ if (mem_cgroup_kmem_disabled())
+ memcg_aware = false;
#endif
lru->node = kcalloc(nr_node_ids, sizeof(*lru->node), GFP_KERNEL);
--
2.25.1
On Thu, Dec 28, 2023 at 06:27:14AM +0000, Haifeng Xu wrote:
> Actually, when using a boot time kernel option "cgroup.memory=nokmem",
> all lru items are inserted to list_lru_node. But for those users who
> invoke list_lru_init_memcg() to initialize list_lru, list_lru_memcg_aware()
> returns true. And this brings unneeded operations related to memcg.
>
> To make things more convenient, let's disable memcg_aware when
> cgroup.memory is set to "nokmem".
I guess it's not only more convenient, but also should save a bit of memory.
But _probably_ not enough to justify messing up with stable backports.
Do you have any data on memory savings?
>
> Signed-off-by: Haifeng Xu <[email protected]>
Acked-by: Roman Gushchin <[email protected]>
Thanks!
On 2024/1/3 09:53, Roman Gushchin wrote:
> On Thu, Dec 28, 2023 at 06:27:14AM +0000, Haifeng Xu wrote:
>> Actually, when using a boot time kernel option "cgroup.memory=nokmem",
>> all lru items are inserted to list_lru_node. But for those users who
>> invoke list_lru_init_memcg() to initialize list_lru, list_lru_memcg_aware()
>> returns true. And this brings unneeded operations related to memcg.
>>
>> To make things more convenient, let's disable memcg_aware when
>> cgroup.memory is set to "nokmem".
>
> I guess it's not only more convenient, but also should save a bit of memory.
> But _probably_ not enough to justify messing up with stable backports.
> Do you have any data on memory savings?
Do you mean that it will remove the allocation of 'list_lru_memcg' in memcg_init_list_lru_one()?
In fact, even without this patch, if we set cgroup.memory set to "nokmem", all memcg's kmemcg_id is -1, so
memcg_list_lru_allocated() always returns true and we won't invoke memcg_init_list_lru_one().
>
>>
>> Signed-off-by: Haifeng Xu <[email protected]>
>
> Acked-by: Roman Gushchin <[email protected]>
>
> Thanks!