On Fri, Apr 22, 2022 at 12:55 PM Jagdish Gediya <jvgediya@xxxxxxxxxxxxx> wrote:
>
> Kernel prepares per-node demotion target list based on
> node_states[N_DEMOTION_TARGETS], If enabled through sysfs,
> demotion kicks in during reclaim, and pages get migrated
> according to demotion target list prepared by kernel.
>
> It is helpful to know demotion target list prepared by
> kernel to understand the demotion behaviour, so add
> interface /sys/devices/system/node/nodeX/demotion_targets
> to view per-node demotion targets via sysfs.
>
> Signed-off-by: Jagdish Gediya <jvgediya@xxxxxxxxxxxxx>
> ---
> drivers/base/node.c | 10 ++++++++++
> include/linux/migrate.h | 1 +
> mm/migrate.c | 17 +++++++++++++++++
> 3 files changed, 28 insertions(+)
>
> diff --git a/drivers/base/node.c b/drivers/base/node.c
> index e03eedbc421b..92326219aac2 100644
> --- a/drivers/base/node.c
> +++ b/drivers/base/node.c
> @@ -561,11 +561,21 @@ static ssize_t node_read_distance(struct device *dev,
> }
> static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
>
> +static ssize_t demotion_targets_show(struct device *dev,
> + struct device_attribute *attr, char *buf)
> +{
> + nodemask_t demotion_targets = node_get_demotion_targets(dev->id);
> +
> + return sysfs_emit(buf, "%*pbl\n", nodemask_pr_args(&demotion_targets));
> +}
> +static DEVICE_ATTR_RO(demotion_targets);
> +
> static struct attribute *node_dev_attrs[] = {
> &dev_attr_meminfo.attr,
> &dev_attr_numastat.attr,
> &dev_attr_distance.attr,
> &dev_attr_vmstat.attr,
> + &dev_attr_demotion_targets.attr,
> NULL
> };
>
> diff --git a/include/linux/migrate.h b/include/linux/migrate.h
> index 90e75d5a54d6..072019441a24 100644
> --- a/include/linux/migrate.h
> +++ b/include/linux/migrate.h
> @@ -173,6 +173,7 @@ int migrate_vma_setup(struct migrate_vma *args);
> void migrate_vma_pages(struct migrate_vma *migrate);
> void migrate_vma_finalize(struct migrate_vma *migrate);
> int next_demotion_node(int node);
> +nodemask_t node_get_demotion_targets(int node);
I think a stub implementation for !CONFIG_MIGRATION is also needed.
> #else /* CONFIG_MIGRATION disabled: */
>
> diff --git a/mm/migrate.c b/mm/migrate.c
> index 5b92a09fbe4a..da864831bc0c 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -2187,6 +2187,23 @@ struct demotion_nodes {
>
> static struct demotion_nodes *node_demotion __read_mostly;
>
> +nodemask_t node_get_demotion_targets(int node)
> +{
> + nodemask_t demotion_targets = NODE_MASK_NONE;
> + unsigned short target_nr;
> +
> + if (!node_demotion)
> + return NODE_MASK_NONE;
> +
> + rcu_read_lock();
> + target_nr = READ_ONCE(node_demotion[node].nr);
> + for (int i = 0; i < target_nr; i++)
> + node_set(READ_ONCE(node_demotion[node].nodes[i]), demotion_targets);
> + rcu_read_unlock();
> +
> + return demotion_targets;
> +}
> +
> /**
> * next_demotion_node() - Get the next node in the demotion path
> * @node: The starting node to lookup the next node
> --
> 2.35.1
>
