Re: [PATCH] PM: domains: Fix sleep-in-atomic bug caused by genpd_debug_remove()

[Ulf Hansson]

On Fri, 25 Feb 2022 at 07:48, Shawn Guo <shawnguo@xxxxxxxxxx> wrote:
>
> From: Shawn Guo <shawn.guo@xxxxxxxxxx>
>
> When a genpd with GENPD_FLAG_IRQ_SAFE gets removed, the following
> sleep-in-atomic bug will be seen, as genpd_debug_remove() will be called
> with a spinlock being held.
>
> [    0.029183] BUG: sleeping function called from invalid context at kernel/locking/rwsem.c:1460
> [    0.029204] in_atomic(): 1, irqs_disabled(): 128, non_block: 0, pid: 1, name: swapper/0
> [    0.029219] preempt_count: 1, expected: 0
> [    0.029230] CPU: 1 PID: 1 Comm: swapper/0 Not tainted 5.17.0-rc4+ #489
> [    0.029245] Hardware name: Thundercomm TurboX CM2290 (DT)
> [    0.029256] Call trace:
> [    0.029265]  dump_backtrace.part.0+0xbc/0xd0
> [    0.029285]  show_stack+0x3c/0xa0
> [    0.029298]  dump_stack_lvl+0x7c/0xa0
> [    0.029311]  dump_stack+0x18/0x34
> [    0.029323]  __might_resched+0x10c/0x13c
> [    0.029338]  __might_sleep+0x4c/0x80
> [    0.029351]  down_read+0x24/0xd0
> [    0.029363]  lookup_one_len_unlocked+0x9c/0xcc
> [    0.029379]  lookup_positive_unlocked+0x10/0x50
> [    0.029392]  debugfs_lookup+0x68/0xac
> [    0.029406]  genpd_remove.part.0+0x12c/0x1b4
> [    0.029419]  of_genpd_remove_last+0xa8/0xd4
> [    0.029434]  psci_cpuidle_domain_probe+0x174/0x53c
> [    0.029449]  platform_probe+0x68/0xe0
> [    0.029462]  really_probe+0x190/0x430
> [    0.029473]  __driver_probe_device+0x90/0x18c
> [    0.029485]  driver_probe_device+0x40/0xe0
> [    0.029497]  __driver_attach+0xf4/0x1d0
> [    0.029508]  bus_for_each_dev+0x70/0xd0
> [    0.029523]  driver_attach+0x24/0x30
> [    0.029534]  bus_add_driver+0x164/0x22c
> [    0.029545]  driver_register+0x78/0x130
> [    0.029556]  __platform_driver_register+0x28/0x34
> [    0.029569]  psci_idle_init_domains+0x1c/0x28
> [    0.029583]  do_one_initcall+0x50/0x1b0
> [    0.029595]  kernel_init_freeable+0x214/0x280
> [    0.029609]  kernel_init+0x2c/0x13c
> [    0.029622]  ret_from_fork+0x10/0x20
>
> It doesn't seem necessary to call genpd_debug_remove() with the lock, so
> move it out from locking to fix the problem.
>
> Fixes: 718072ceb211 ("PM: domains: create debugfs nodes when adding power domains")
> Signed-off-by: Shawn Guo <shawn.guo@xxxxxxxxxx>

Thanks for fixing this!

Reviewed-by: Ulf Hansson <ulf.hansson@xxxxxxxxxx>

Rafael, I think we should tag this for stable kernels too.

Kind regards
Uffe

> ---
>  drivers/base/power/domain.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/base/power/domain.c b/drivers/base/power/domain.c
> index 5db704f02e71..7e8039d1884c 100644
> --- a/drivers/base/power/domain.c
> +++ b/drivers/base/power/domain.c
> @@ -2058,9 +2058,9 @@ static int genpd_remove(struct generic_pm_domain *genpd)
>                 kfree(link);
>         }
>
> -       genpd_debug_remove(genpd);
>         list_del(&genpd->gpd_list_node);
>         genpd_unlock(genpd);
> +       genpd_debug_remove(genpd);
>         cancel_work_sync(&genpd->power_off_work);
>         if (genpd_is_cpu_domain(genpd))
>                 free_cpumask_var(genpd->cpus);
> --
> 2.25.1
>