Re: [PATCH mm-unstable v1 0/2] Track pages allocated for struct
From: Andrew Morton
Date: Thu Oct 31 2024 - 19:06:13 EST
hm.
On Thu, 31 Oct 2024 22:45:49 +0000 Kinsey Ho <kinseyho@xxxxxxxxxx> wrote:
> We noticed high overhead for pages allocated for struct swap_cgroup in
> our fleet.
This is scanty. Please describe the problem further.
> This patchset adds the number of pages allocated for struct
> swap_cgroup to vmstat. This can be a useful metric for identifying
> unneeded overhead on systems which configure swap.
Possibly dumb question: can we switch swap_cgroup_prepare to kmalloc()
(or kmem-cache_alloc()) and use slab's accounting to satisfy this
requirement?
> Before adding the new stat, Patch 1 introduces a generic system-wide
> counting interface.
I don't really understand this one and would like to hear much more
about the motivation.
And: "the existing use case" is OK with a global counter, but what about
future use cases?
And: what are the future use cases?
Thanks.