Received-SPF: softfail (google.com: domain of transitioning linux-kernel-owner@vger.kernel.org does not designate 23.128.96.19 as permitted sender) client-ip=23.128.96.19;
MIME-Version: 1.0
References: <20220413092206.73974-1-jvgediya@linux.ibm.com> <20220413092206.73974-4-jvgediya@linux.ibm.com>
In-Reply-To: <20220413092206.73974-4-jvgediya@linux.ibm.com>
From:   Wei Xu <weixugc@google.com>
Date:   Wed, 20 Apr 2022 21:26:34 -0700
Message-ID: <CAAPL-u9OHuRWXvc0RAPvEtLQLy0AkLbAwjZtGERhmhn-yFZgrw@mail.gmail.com>
Subject: Re: [PATCH v2 3/5] mm: demotion: Add support to set targets from userspace
To:     Jagdish Gediya <jvgediya@linux.ibm.com>
Cc:     Linux MM <linux-mm@kvack.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        aneesh.kumar@linux.ibm.com,
        Baolin Wang <baolin.wang@linux.alibaba.com>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        Huang Ying <ying.huang@intel.com>,
        Greg Thelen <gthelen@google.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Wed, Apr 13, 2022 at 2:22 AM Jagdish Gediya <jvgediya@linux.ibm.com> wrote:
>
> Add support to set node_states[N_DEMOTION_TARGETS] from
> user space to override the default demotion targets.
>
> Restrict demotion targets to memory only numa nodes
> while setting them from user space.

Why should we restrict demotion targets to memory only nodes if they
get set explicitly from user space? For example, if we use NUMA
emulation to test demotion without actual hardware, these emulated
NUMA nodes can have CPUs, but we still want some of them to serve as
demotion targets.

> Demotion targets can be set from userspace using command,
> echo <nodelist> > /sys/kernel/mm/numa/demotion_target
>
> Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
> Signed-off-by: Jagdish Gediya <jvgediya@linux.ibm.com>
> ---
>  .../ABI/testing/sysfs-kernel-mm-numa          | 12 +++++++
>  mm/migrate.c                                  | 35 +++++++++++++++++++
>  2 files changed, 47 insertions(+)
>
> diff --git a/Documentation/ABI/testing/sysfs-kernel-mm-numa b/Documentation/ABI/testing/sysfs-kernel-mm-numa
> index 77e559d4ed80..10e9e643845c 100644
> --- a/Documentation/ABI/testing/sysfs-kernel-mm-numa
> +++ b/Documentation/ABI/testing/sysfs-kernel-mm-numa
> @@ -22,3 +22,15 @@ Description: Enable/disable demoting pages during reclaim
>                 the guarantees of cpusets.  This should not be enabled
>                 on systems which need strict cpuset location
>                 guarantees.
> +
> +What:          /sys/kernel/mm/numa/demotion_target
> +Date:          April 2022
> +Contact:       Linux memory management mailing list <linux-mm@kvack.org>
> +Description:   Configure demotion target nodes
> +
> +               Page migration during reclaim is intended for systems
> +               with tiered memory configurations. Preferred migration target
> +               nodes can be configured in a system using this interface, based
> +               on which demotion table is prepared in kernel. If demotion is
> +               enabled then pages will be migrated to set demotion targets
> +               during reclaim.
> diff --git a/mm/migrate.c b/mm/migrate.c
> index 516f4e1348c1..4d3d80ca0a7f 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -2564,12 +2564,47 @@ static ssize_t numa_demotion_enabled_store(struct kobject *kobj,
>         return count;
>  }
>
> +
> +static ssize_t numa_demotion_target_show(struct kobject *kobj,
> +                                         struct kobj_attribute *attr, char *buf)
> +{
> +       return sysfs_emit(buf, "%*pbl\n",
> +                         nodemask_pr_args(&node_states[N_DEMOTION_TARGETS]));
> +}
> +
> +static ssize_t numa_demotion_target_store(struct kobject *kobj,
> +                                         struct kobj_attribute *attr,
> +                                         const char *nodelist, size_t count)
> +{
> +       nodemask_t nodes;
> +
> +       if (nodelist_parse(nodelist, nodes))
> +               return -EINVAL;
> +
> +       if (!nodes_subset(nodes, node_states[N_MEMORY]))
> +               return -EINVAL;
> +
> +       if (nodes_intersects(nodes, node_states[N_CPU]))
> +               return -EINVAL;
> +
> +       node_states[N_DEMOTION_TARGETS] = nodes;
> +
> +       set_migration_target_nodes();
> +
> +       return count;
> +}
> +
>  static struct kobj_attribute numa_demotion_enabled_attr =
>         __ATTR(demotion_enabled, 0644, numa_demotion_enabled_show,
>                numa_demotion_enabled_store);
>
> +static struct kobj_attribute numa_demotion_target_attr =
> +       __ATTR(demotion_target, 0644, numa_demotion_target_show,
> +              numa_demotion_target_store);
> +
>  static struct attribute *numa_attrs[] = {
>         &numa_demotion_enabled_attr.attr,
> +       &numa_demotion_target_attr.attr,
>         NULL,
>  };
>
> --
> 2.35.1
>
>