There are three places, vmpressure_prio(), shrink_node_memcgs() and
shrink_node(), which invoke vmpressure(). But only shrink_node_memcgs()
sets tree to false and the memcg used in it is not NULL, so we don't
check it again in vmpressure().
Signed-off-by: Haifeng Xu <[email protected]>
---
mm/vmpressure.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/mm/vmpressure.c b/mm/vmpressure.c
index b52644771cc4..3e4251a55e56 100644
--- a/mm/vmpressure.c
+++ b/mm/vmpressure.c
@@ -284,7 +284,7 @@ void vmpressure(gfp_t gfp, struct mem_cgroup *memcg, bool tree,
enum vmpressure_levels level;
/* For now, no users for root-level efficiency */
- if (!memcg || mem_cgroup_is_root(memcg))
+ if (mem_cgroup_is_root(memcg))
return;
spin_lock(&vmpr->sr_lock);
--
2.25.1
On Wed, 19 Apr 2023 09:20:07 +0000 Haifeng Xu <[email protected]> wrote:
> There are three places, vmpressure_prio(), shrink_node_memcgs() and
> shrink_node(), which invoke vmpressure(). But only shrink_node_memcgs()
> sets tree to false and the memcg used in it is not NULL, so we don't
> check it again in vmpressure().
>
> ...
>
> --- a/mm/vmpressure.c
> +++ b/mm/vmpressure.c
> @@ -284,7 +284,7 @@ void vmpressure(gfp_t gfp, struct mem_cgroup *memcg, bool tree,
> enum vmpressure_levels level;
>
> /* For now, no users for root-level efficiency */
> - if (!memcg || mem_cgroup_is_root(memcg))
> + if (mem_cgroup_is_root(memcg))
> return;
>
> spin_lock(&vmpr->sr_lock);
try_to_free_pages()->
do_try_to_free_pages()->
vmpressure_prio()->
vmpressure()->
crash
what am I missing here?
It does appear that vmpressure() could be simplified with
if (!memcg)
memcg = root_mem_cgroup;
so the test you identified goes away and the memcg_to_vmpressure() call
becomes simpler. But that's such a small change it doesn't seem worth
the effort.
On 2023/4/20 04:03, Andrew Morton wrote:
> On Wed, 19 Apr 2023 09:20:07 +0000 Haifeng Xu <[email protected]> wrote:
>
>> There are three places, vmpressure_prio(), shrink_node_memcgs() and
>> shrink_node(), which invoke vmpressure(). But only shrink_node_memcgs()
>> sets tree to false and the memcg used in it is not NULL, so we don't
>> check it again in vmpressure().
>>
>> ...
>>
>> --- a/mm/vmpressure.c
>> +++ b/mm/vmpressure.c
>> @@ -284,7 +284,7 @@ void vmpressure(gfp_t gfp, struct mem_cgroup *memcg, bool tree,
>> enum vmpressure_levels level;
>>
>> /* For now, no users for root-level efficiency */
>> - if (!memcg || mem_cgroup_is_root(memcg))
>> + if (mem_cgroup_is_root(memcg))
>> return;
>>
>> spin_lock(&vmpr->sr_lock);
>
> try_to_free_pages()->
> do_try_to_free_pages()->
> vmpressure_prio()->
> vmpressure()->
> crash
>
> what am I missing here?
>
>
vmpressure_prio() set tree to true, so the crash won't happen.
> It does appear that vmpressure() could be simplified with
>
> if (!memcg)
> memcg = root_mem_cgroup;
>
> so the test you identified goes away and the memcg_to_vmpressure() call
> becomes simpler. But that's such a small change it doesn't seem worth
> the effort.
memcg_to_vmpressure() has taken this case that the memcg is NULL into account and
we do not need to change it.
The test I identified is in the branch that the tree is set to false. In this branch,
memcg can never be NULL because only shrink_node_memcgs() will step into it.