Re: [PATCH 1/2] mfd: arizona: Fix race between runtime suspend and IRQs

[Lee Jones <lee.jones@linaro.org>]

On Sun, 14 Jun 2015, Charles Keepax wrote:

> The function arizona_irq_thread (the threaded handler for the arizona
> IRQs) calls pm_runtime_get_sync at the start to ensure that the chip is
> active as we handle the IRQ. If the chip is part way through a runtime
> suspend when an IRQ arrives the PM core will wait for the suspend to
> complete, before resuming. However, since commit 4f0216409f7c
> ("mfd: arizona: Add better support for system suspend") the runtime
> suspend function may call disable_irq, if the chip is going to fully
> power off, which will try to wait for any outstanding IRQs to complete.
> This results in deadlock as the IRQ thread is waiting for the PM
> operation to complete and the PM thread is waiting for the IRQ to
> complete.
> 
> To avoid this situation we use disable_irq_nosync, which allows the
> suspending thread to finish the suspend without waiting for the IRQ to
> complete. This is safe because if an IRQ is being processed it can only
> be blocked at the pm_runtime_get_sync at the start of the handler
> otherwise it wouldn't be possible to suspend.
> 
> Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
> ---
>  drivers/mfd/arizona-core.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)

Applied, thanks.

> diff --git a/drivers/mfd/arizona-core.c b/drivers/mfd/arizona-core.c
> index bebf58a..e60bcd9 100644
> --- a/drivers/mfd/arizona-core.c
> +++ b/drivers/mfd/arizona-core.c
> @@ -651,7 +651,7 @@ static int arizona_runtime_suspend(struct device *dev)
>  
>  		arizona->has_fully_powered_off = true;
>  
> -		disable_irq(arizona->irq);
> +		disable_irq_nosync(arizona->irq);
>  		arizona_enable_reset(arizona);
>  		regulator_bulk_disable(arizona->num_core_supplies,
>  				       arizona->core_supplies);

-- 
Lee Jones
Linaro STMicroelectronics Landing Team Lead
Linaro.org │ Open source software for ARM SoCs
Follow Linaro: Facebook | Twitter | Blog