Re: unexpected side effect of "gpiolib: override irq_enable/disable"

[Hans Verkuil <hverkuil@xxxxxxxxx>]

Hi Ludovic,

On 09/12/2018 04:58 PM, Ludovic Desroches wrote:
> Hi,
> 
> Using next-20180912, my kernel hangs during the boot. Git bisect tell me
> that the cause of my issue is the commit "gpiolib: override
> irq_enable/disable"
> 
> I dug further and this patch can have some side effects. When booting, I
> have an infinite loop when trying to enable a gpio irq. I don't know if
> the pinctrl-at91 driver is the only one concerned or not.
> 
> The pattern leading to this issue is quite simple: we have several gpio
> controllers sharing the same irq_chip structure. Installing the
> irq_enable/irq_disable hook works well the first time. The second time,
> since the irq_enable has been altered to use gpiochip_irq_enable,
> this latest function will call itself again and again by calling
> irq_enable.
> 
> I think it should be better to have one irq_chip structure per gpio
> controller. I am going to do a patch for pinctrl-at91. Excepting if you
> think it has to be solved in a different way.
> 
> 
> Ludovic
> 

Can you try this patch first?

Thanks!

	Hans

Signed-off-by: Hans Verkuil <hans.verkuil@xxxxxxxxx>
---
diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
index efce534a269b..e09e4e439928 100644
--- a/drivers/gpio/gpiolib.c
+++ b/drivers/gpio/gpiolib.c
@@ -1859,6 +1859,8 @@ static void gpiochip_set_irq_hooks(struct gpio_chip *gpiochip)
 	}
 	if (WARN_ON(gpiochip->irq.irq_enable))
 		return;
+	if (irqchip->irq_enable == gpiochip_irq_enable)
+		return;
 	gpiochip->irq.irq_enable = irqchip->irq_enable;
 	gpiochip->irq.irq_disable = irqchip->irq_disable;
 	irqchip->irq_enable = gpiochip_irq_enable;