Re: [PATCH v4] gpio: driver for PrimeCell PL061 GPIO controller

[Baruch Siach <baruch@tkos.co.il>]

On Wed, Jun 10, 2009 at 12:44:47AM -0700, Andrew Morton wrote:
> On Wed, 10 Jun 2009 10:22:31 +0300 Baruch Siach <baruch@tkos.co.il> wrote:
> 
> > Hi Andrew,
> > 
> > > >  static unsigned int pl061_irq_startup(unsigned irq)
> > > >  {
> > > > -	int ret;
> > > > -
> > > > -	ret = gpio_request(irq_to_gpio(irq), "IRQ");
> > > > -	if (ret < 0) {
> > > > -		pr_warning("%s: warning: gpio_request(%d) returned %d\n",
> > > > -				__func__, irq_to_gpio(irq), ret);
> > > > -		return 0;
> > > > -	}
> > > > +	if (gpio_request(irq_to_gpio(irq), "IRQ") == 0)
> > > > +		pr_warning("%s: warning: GPIO%d has not been requested\n",
> > > > +				__func__, irq_to_gpio(irq));
> > > 
> > > This is wrong, isn't it?  gpio_request() returns 0 on success.
> > 
> > Russell said that gpio configuration is the responsibility of the platform 
> > code. Here I just warn when the gpio has not been requested, and thus 
> > gpio_request() succeeds. I'll add a comment.
> 
> OK.
> 
> If the gpio_request() accidentally succeeded, should we gpio_free() the
> result here?

I don't think so. This is just a warning. We do use this GPIO here, so I guess 
we should keep it requested.

> Should the gpio core provide a primitive to check that a gpio has been
> properly requested rathe rthan open-coding it here?

Probably. The author (maintainer?) of gpiolib, David Brownell, is not 
responsive at the moment. (I have an SPI master driver waiting for his 
review/ack).

> > > >  static void pl061_irq_handler(unsigned irq, struct irq_desc *desc)
> > > >  {
> > > > +	struct list_head *chip_list = get_irq_chip_data(irq);
> > > > +	struct list_head *ptr;
> > > > +	struct pl061_gpio *chip;
> > > > +
> > > >  	desc->chip->ack(irq);
> > > > -	while (1) {
> > > > +	list_for_each(ptr, chip_list) {
> > > 
> > > What locking protects the newly-added list?
> > 
> > Do we need locking even though we list_add() only at probe time?
> 
> Nope.  I guess.  It depends on the driver.  hotplug/hot-remove needs to
> beconsidered often.
> 
> > (Compiling as 
> > a module is not supported, so this only happens at boot time).
> 
> The probe handler is probably serialised against everything else even if
> the driver _is_ a module.

baruch

-- 
                                                     ~. .~   Tk Open Systems
=}------------------------------------------------ooO--U--Ooo------------{=
   - baruch@tkos.co.il - tel: +972.2.679.5364, http://www.tkos.co.il -