Re: [-mm patch] aic7xxx: check irq validity (was Re: 2.6.18-mm2)

[Matthew Wilcox <matthew@wil.cx>]

On Sat, Sep 30, 2006 at 07:58:18PM -0400, Jeff Garzik wrote:
> Actually, rather than adding this check to every driver, I would rather 
> do something like the attached patch:  create a pci_request_irq(), and 
> pass a struct pci_device to it.  Then the driver author doesn't have to 
> worry about such details.

I like pci_request_irq(), but pci_valid_irq is bad.

> +#ifndef ARCH_VALIDATE_PCI_IRQ
> +int pci_valid_irq(struct pci_dev *pdev)
> +{
> +	if (pdev->irq == 0)
> +		return -EINVAL;
> +	
> +	return 0;
> +}
> +EXPORT_SYMBOL(pci_valid_irq);
> +#endif /* ARCH_VALIDATE_PCI_IRQ */

Better would be:

#ifndef ARCH_VALIDATE_IRQ
static inline int valid_irq(unsigned int irq)
{
	return irq ? 1 : 0;
}
#endif

in linux/interrupt.h (around request_irq).

And it doesn't need to be a __must_check.  There's no point -- it has
no side-effects.  The only reason to call it is if you want the answer
to the question.  You had the sense of the return code wrong too; you
want to use it as:

int pci_request_irq(struct pci_dev *pdev, irq_handler_t handler,
			unsigned long flags, const char *name, void *data)
{
	if (!valid_irq(pdev->irq)) {
		dev_printk(KERN_ERR, &pdev->dev, "invalid irq\n");
		return -EINVAL;
	}

	return request_irq(pdev->irq, handler, flags | IRQF_SHARED, name, data);
}