[Patch v4] mm/vmscan.c: remove cpu online notification for now

[Patch v4] mm/vmscan.c: remove cpu online notification for now

[Wei Yang]

kswapd kernel thread starts either with a CPU affinity set to the full
cpu mask of its target node or without any affinity at all if the node
is CPUless. There is a cpu hotplug callback (kswapd_cpu_online) that
implements an elaborate way to update this mask when a cpu is onlined.

It is not really clear whether there is any actual benefit from this
scheme. Completely CPU-less NUMA nodes rarely gain a new CPU during
runtime. Drop the code for that reason. If there is a real usecase then
we can resurrect and simplify the code.

[mhocko@suse.com rewrite changelog]

Suggested-by: Michal Hocko <mhocko@suse.org>
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Acked-by: Michal Hocko <mhocko@suse.com>

---
v4:
  * adjust changelog suggested by Michal
v3:
  * remove the cpu online notification suggested by Michal
v2:
  * rephrase the changelog
---
 mm/vmscan.c | 27 +--------------------------
 1 file changed, 1 insertion(+), 26 deletions(-)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 665f33258cd7..a4fdf3dc8887 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -4023,27 +4023,6 @@ unsigned long shrink_all_memory(unsigned long nr_to_reclaim)
 }
 #endif /* CONFIG_HIBERNATION */
 
-/* It's optimal to keep kswapds on the same CPUs as their memory, but
-   not required for correctness.  So if the last cpu in a node goes
-   away, we get changed to run anywhere: as the first one comes back,
-   restore their cpu bindings. */
-static int kswapd_cpu_online(unsigned int cpu)
-{
-	int nid;
-
-	for_each_node_state(nid, N_MEMORY) {
-		pg_data_t *pgdat = NODE_DATA(nid);
-		const struct cpumask *mask;
-
-		mask = cpumask_of_node(pgdat->node_id);
-
-		if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids)
-			/* One of our CPUs online: restore mask */
-			set_cpus_allowed_ptr(pgdat->kswapd, mask);
-	}
-	return 0;
-}
-
 /*
  * This kswapd start function will be called by init and node-hot-add.
  * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added.
@@ -4083,15 +4062,11 @@ void kswapd_stop(int nid)
 
 static int __init kswapd_init(void)
 {
-	int nid, ret;
+	int nid;
 
 	swap_setup();
 	for_each_node_state(nid, N_MEMORY)
  		kswapd_run(nid);
-	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
-					"mm/vmscan:online", kswapd_cpu_online,
-					NULL);
-	WARN_ON(ret < 0);
 	return 0;
 }
 
-- 
2.17.1

Re: [Patch v4] mm/vmscan.c: remove cpu online notification for now

[Andrew Morton]

On Wed, 19 Feb 2020 06:44:22 +0800 Wei Yang <richardw.yang@linux.intel.com> wrote:

> kswapd kernel thread starts either with a CPU affinity set to the full
> cpu mask of its target node or without any affinity at all if the node
> is CPUless. There is a cpu hotplug callback (kswapd_cpu_online) that
> implements an elaborate way to update this mask when a cpu is onlined.
> 
> It is not really clear whether there is any actual benefit from this
> scheme. Completely CPU-less NUMA nodes rarely gain a new CPU during
> runtime.

This is the case across all platforms, all architectures, all users for
the next N years?  I'm surprised that we know this with sufficient
confidence.  Can you explain how you came to make this assertion?

> Drop the code for that reason. If there is a real usecase then
> we can resurrect and simplify the code.

Re: [Patch v4] mm/vmscan.c: remove cpu online notification for now

[Michal Hocko]

On Wed 19-02-20 12:08:10, Andrew Morton wrote:
> On Wed, 19 Feb 2020 06:44:22 +0800 Wei Yang <richardw.yang@linux.intel.com> wrote:
> 
> > kswapd kernel thread starts either with a CPU affinity set to the full
> > cpu mask of its target node or without any affinity at all if the node
> > is CPUless. There is a cpu hotplug callback (kswapd_cpu_online) that
> > implements an elaborate way to update this mask when a cpu is onlined.
> > 
> > It is not really clear whether there is any actual benefit from this
> > scheme. Completely CPU-less NUMA nodes rarely gain a new CPU during
> > runtime.
> 
> This is the case across all platforms, all architectures, all users for
> the next N years?  I'm surprised that we know this with sufficient
> confidence.  Can you explain how you came to make this assertion?

CPUless NUMA nodes are quite rare - mostly ppc with crippled LPARs.
I am not aware those would dynamically get CPUs for those nodes later in
the runtime. Maybe they do but we would like to learn about that. A
missing cpu mask is not going cause any fatal problems anyway.

As the changelog states the callback can be reintroduced with a sign of
testing and usecase description. I prefer we drop this code in the mean
time as the benefit is not really clear or testable.

> > Drop the code for that reason. If there is a real usecase then
> > we can resurrect and simplify the code.

-- 
Michal Hocko
SUSE Labs

Re: [Patch v4] mm/vmscan.c: remove cpu online notification for now

[Yang Shi]

On Wed, Feb 19, 2020 at 11:52 PM Michal Hocko <mhocko@kernel.org> wrote:
>
> On Wed 19-02-20 12:08:10, Andrew Morton wrote:
> > On Wed, 19 Feb 2020 06:44:22 +0800 Wei Yang <richardw.yang@linux.intel.com> wrote:
> >
> > > kswapd kernel thread starts either with a CPU affinity set to the full
> > > cpu mask of its target node or without any affinity at all if the node
> > > is CPUless. There is a cpu hotplug callback (kswapd_cpu_online) that
> > > implements an elaborate way to update this mask when a cpu is onlined.
> > >
> > > It is not really clear whether there is any actual benefit from this
> > > scheme. Completely CPU-less NUMA nodes rarely gain a new CPU during
> > > runtime.
> >
> > This is the case across all platforms, all architectures, all users for
> > the next N years?  I'm surprised that we know this with sufficient
> > confidence.  Can you explain how you came to make this assertion?
>
> CPUless NUMA nodes are quite rare - mostly ppc with crippled LPARs.
> I am not aware those would dynamically get CPUs for those nodes later in
> the runtime. Maybe they do but we would like to learn about that. A
> missing cpu mask is not going cause any fatal problems anyway.

Persistent memory nodes are CPUless nodes. But, I don't think they
would get any CPU online later in the runtime.

>
> As the changelog states the callback can be reintroduced with a sign of
> testing and usecase description. I prefer we drop this code in the mean
> time as the benefit is not really clear or testable.
>
> > > Drop the code for that reason. If there is a real usecase then
> > > we can resurrect and simplify the code.
>
> --
> Michal Hocko
> SUSE Labs
>