Re: [PATCH 1/4] mm/memcg: use NUMA_NO_NODE to indicate allocation from unspecified node

[Michal Hocko <mhocko@suse.com>]

On Fri 14-01-22 00:29:37, Wei Yang wrote:
> On Wed, Jan 12, 2022 at 09:56:15AM +0100, Michal Hocko wrote:
> >On Wed 12-01-22 00:46:34, Wei Yang wrote:
> >> On Tue, Jan 11, 2022 at 09:40:20AM +0100, Michal Hocko wrote:
> >> >On Tue 11-01-22 01:02:59, Wei Yang wrote:
> >> >> Instead of use "-1", let's use NUMA_NO_NODE for consistency.
> >> >> 
> >> >> Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
> >> >
> >> >I am not really sure this is worth it. After the merge window I plan to
> >> >post http://lkml.kernel.org/r/20211214100732.26335-1-mhocko@kernel.org.
> >> 
> >> Give me some time to understand it :-)
> >
> >Just for the record, here is what I have put on top of that series:
> 
> Ok, I got what you try to resolve. I am ok with the following change except
> one point.
> 
> >--- 
> >>From b7195eba02fe6308a6927450f4630057c05e808e Mon Sep 17 00:00:00 2001
> >From: Wei Yang <richard.weiyang@gmail.com>
> >Date: Tue, 11 Jan 2022 09:45:25 +0100
> >Subject: [PATCH] memcg: do not tweak node in alloc_mem_cgroup_per_node_info
> >
> >alloc_mem_cgroup_per_node_info is allocated for each possible node and
> >this used to be a problem because not !node_online nodes didn't have
> >appropriate data structure allocated. This has changed by "mm: handle
> >uninitialized numa nodes gracefully" so we can drop the special casing
> >here.
> >
> >Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
> >Signed-off-by: Michal Hocko <mhocko@suse.com>
> >---
> > mm/memcontrol.c | 14 ++------------
> > 1 file changed, 2 insertions(+), 12 deletions(-)
> >
> >diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> >index 781605e92015..ed19a21ee14e 100644
> >--- a/mm/memcontrol.c
> >+++ b/mm/memcontrol.c
> >@@ -5044,18 +5044,8 @@ struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
> > static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
> > {
> > 	struct mem_cgroup_per_node *pn;
> >-	int tmp = node;
> >-	/*
> >-	 * This routine is called against possible nodes.
> >-	 * But it's BUG to call kmalloc() against offline node.
> >-	 *
> >-	 * TODO: this routine can waste much memory for nodes which will
> >-	 *       never be onlined. It's better to use memory hotplug callback
> >-	 *       function.
> >-	 */
> 
> Do you think this TODO is not related to this change?

It is not really related but I am not sure how useful it is. Essentially
any allocation that is per-possible node is in the same situation and if
we really need to deal with large and sparse possible nodes masks.

If you want me to keep the TODO I will do it though.

-- 
Michal Hocko
SUSE Labs