Re: [PATCH RFC] reboot: hotplug cpus in migrate_to_reboot_cpu()

[Hsin-Yi Wang]

On Wed, Oct 2, 2019 at 7:41 PM Hsin-Yi Wang <hsinyi@xxxxxxxxxxxx> wrote:
>
> Currently system reboots use arch specific codes (eg. smp_send_stop) to
> offline non reboot cpus. Some arch like arm64, arm, and x86... set offline
> masks to cpu without really offlining them. Thus it causes some race
> condition and kernel warning comes out sometimes when system reboots. We
> can do cpu hotplug in migrate_to_reboot_cpu() to avoid this issue.
>
> Signed-off-by: Hsin-Yi Wang <hsinyi@xxxxxxxxxxxx>
> ---
> kernel warnings at reboot:
> [1] https://lore.kernel.org/lkml/20190820100843.3028-1-hsinyi@xxxxxxxxxxxx/
> [2] https://lore.kernel.org/lkml/20190727164450.GA11726@xxxxxxxxxxxx/
> ---
>  kernel/cpu.c    | 35 +++++++++++++++++++++++++++++++++++
>  kernel/reboot.c | 18 ------------------
>  2 files changed, 35 insertions(+), 18 deletions(-)
>
> diff --git a/kernel/cpu.c b/kernel/cpu.c
> index fc28e17940e0..2f4d51fe91e3 100644
> --- a/kernel/cpu.c
> +++ b/kernel/cpu.c
> @@ -31,6 +31,7 @@
>  #include <linux/relay.h>
>  #include <linux/slab.h>
>  #include <linux/percpu-rwsem.h>
> +#include <linux/reboot.h>
>
>  #include <trace/events/power.h>
>  #define CREATE_TRACE_POINTS
> @@ -1366,6 +1367,40 @@ int __boot_cpu_id;
>
>  #endif /* CONFIG_SMP */
>
> +void migrate_to_reboot_cpu(void)
> +{
> +       /* The boot cpu is always logical cpu 0 */
> +       int cpu = reboot_cpu;
> +
> +       /* Make certain the cpu I'm about to reboot on is online */
> +       if (!cpu_online(cpu))
> +               cpu = cpumask_first(cpu_online_mask);
> +
> +       /* Prevent races with other tasks migrating this task */
> +       current->flags |= PF_NO_SETAFFINITY;
> +
> +       /* Make certain I only run on the appropriate processor */
> +       set_cpus_allowed_ptr(current, cpumask_of(cpu));
> +
> +       /* Hotplug other cpus if possible */
> +       if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) {

Should use #ifdef CONFIG_HOTPLUG_CPU here. Will fix in the next
version if this patch is reasonable.
(Reported-by: kbuild test robot <lkp@xxxxxxxxx>)
> +               int i, err;
> +
> +               cpu_maps_update_begin();
> +
> +               for_each_online_cpu(i) {
> +                       if (i == cpu)
> +                               continue;
> +                       err = _cpu_down(i, 0, CPUHP_OFFLINE);
> +                       if (err)
> +                               pr_info("Failed to offline cpu %d\n", i);
> +               }
> +               cpu_hotplug_disabled++;
> +
> +               cpu_maps_update_done();
> +       }
> +}
> +
>  /* Boot processor state steps */
>  static struct cpuhp_step cpuhp_hp_states[] = {
>         [CPUHP_OFFLINE] = {
> diff --git a/kernel/reboot.c b/kernel/reboot.c
> index c4d472b7f1b4..f0046be34a60 100644
> --- a/kernel/reboot.c
> +++ b/kernel/reboot.c
> @@ -215,24 +215,6 @@ void do_kernel_restart(char *cmd)
>         atomic_notifier_call_chain(&restart_handler_list, reboot_mode, cmd);
>  }
>
> -void migrate_to_reboot_cpu(void)
> -{
> -       /* The boot cpu is always logical cpu 0 */
> -       int cpu = reboot_cpu;
> -
> -       cpu_hotplug_disable();
> -
> -       /* Make certain the cpu I'm about to reboot on is online */
> -       if (!cpu_online(cpu))
> -               cpu = cpumask_first(cpu_online_mask);
> -
> -       /* Prevent races with other tasks migrating this task */
> -       current->flags |= PF_NO_SETAFFINITY;
> -
> -       /* Make certain I only run on the appropriate processor */
> -       set_cpus_allowed_ptr(current, cpumask_of(cpu));
> -}
> -
>  /**
>   *     kernel_restart - reboot the system
>   *     @cmd: pointer to buffer containing command to execute for restart
> --
> 2.23.0.444.g18eeb5a265-goog
>