Re: [PATCH] IRQ : bug fix in setup_irq

[Thomas Gleixner <tglx@linutronix.de>]

On Fri, 18 Jul 2008, sasa sasa wrote:
> diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
> index 1f31422..ca92f97 100644
> --- a/kernel/irq/manage.c
> +++ b/kernel/irq/manage.c
> @@ -339,6 +339,7 @@ int setup_irq(unsigned int irq, struct irqaction *new)
>  		desc->status |= IRQ_NO_BALANCING;
> 
>  	if (!shared) {
> +		int trigger_type = 0;

No need to initialize to zero.

>  		irq_chip_set_defaults(desc->chip);
> 
>  #if defined(CONFIG_IRQ_PER_CPU)
> @@ -346,11 +347,22 @@ int setup_irq(unsigned int irq, struct irqaction *new)
>  			desc->status |= IRQ_PER_CPU;
>  #endif
> 
> +		trigger_type = new->flags & IRQF_TRIGGER_MASK;
> +
>  		/* Setup the type (level, edge polarity) if configured: */
> -		if (new->flags & IRQF_TRIGGER_MASK) {
> -			if (desc->chip && desc->chip->set_type)
> -				desc->chip->set_type(irq,
> -						new->flags & IRQF_TRIGGER_MASK);
> +		if (trigger_type) {
> +			if (desc->chip && desc->chip->set_type)	{
> +				if ((desc->chip->set_type(irq, trigger_type))) {
> +					/* Action handler gets freed but
> +					 * desc->action also needs to be
> +					 * set to NULL
> +					 */

		That's not desc->action. That's the end of the desc->action list

> +					*p = NULL;

Can we have a printk here, which tells us what went wrong ?

> +					spin_unlock_irqrestore(&desc->lock,
> +								flags);
> +					return -EBUSY;

  Please use the already existing error exit via "goto ..." instead

> +				}
> +			}
>  			else
>  				/*
>  				 * IRQF_TRIGGER_* but the PIC does not support
> @@ -571,8 +583,10 @@ int request_irq(unsigned int irq, irq_handler_t handler,
>  #endif
> 
>  	retval = setup_irq(irq, action);
> -	if (retval)
> +	if (retval) {
> +		printk(KERN_WARNING"Failed to allocate interrupt %d\n", retval);
>  		kfree(action);

We dont want to know that the setup function returned -EBUSY. We
rather want to have a printk in the path above.

Thanks,
	tglx