Re: [PATCH] pinctrl: renesas: rzg2l: Execute atomically the interrupt configuration

[claudiu beznea <claudiu.beznea@tuxon.dev>]

Hi, Geert,

On 14.03.2024 15:21, Geert Uytterhoeven wrote:
> Hi Claudiu,
> 
> Thanks for your patch!
> 
> On Thu, Mar 7, 2024 at 12:25 PM Claudiu <claudiu.beznea@tuxon.dev> wrote:
>> From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com>
>>
>> Lockdep detects a possible deadlock as listed below. This is because it
>> detects the IA55 interrupt controller .irq_eoi() API is called from
>> interrupt context while configuration-specific API (e.g., .irq_enable())
>> could be called from process context on resume path (by calling
>> rzg2l_gpio_irq_restore()). To avoid this, protect the call of
>> rzg2l_gpio_irq_enable() with spin_lock_irqsave()/spin_unlock_irqrestore().
>> With this the same approach that is available in __setup_irq() is mimicked
>> to pinctrl IRQ resume function.
> 
> You mean __setup_irq() in kernel/irq/manage.c?

Yes!

> That one uses the raw spinlock methods?

Yes! Would you prefer to have raw spinlock here, too?

Thank you,
Claudiu Beznea

> 
>> --- a/drivers/pinctrl/renesas/pinctrl-rzg2l.c
>> +++ b/drivers/pinctrl/renesas/pinctrl-rzg2l.c
>> @@ -2063,8 +2063,17 @@ static void rzg2l_gpio_irq_restore(struct rzg2l_pinctrl *pctrl)
>>                         continue;
>>                 }
>>
>> -               if (!irqd_irq_disabled(data))
>> +               if (!irqd_irq_disabled(data)) {
>> +                       unsigned long flags;
>> +
>> +                       /*
>> +                        * This has to be atomically executed to protect against a concurrent
>> +                        * interrupt.
>> +                        */
>> +                       spin_lock_irqsave(&pctrl->lock, flags);
>>                         rzg2l_gpio_irq_enable(data);
>> +                       spin_unlock_irqrestore(&pctrl->lock, flags);
>> +               }
>>         }
>>  }
> 
> Gr{oetje,eeting}s,
> 
>                         Geert
>