[RFC PATCH] arm/imx/gpio: add spinlock protection

[Baruch Siach]

On Tue, Jul 06, 2010 at 12:07:48PM +0200, Sascha Hauer wrote:
> On Tue, Jul 06, 2010 at 10:40:43AM +0300, Baruch Siach wrote:
> > On Tue, Jul 06, 2010 at 09:17:02AM +0200, Sascha Hauer wrote:
> > > On Tue, Jul 06, 2010 at 08:00:34AM +0300, Baruch Siach wrote:
> > > > On Mon, Jul 05, 2010 at 09:52:18AM +0200, Sascha Hauer wrote:
> > > > > On Sun, Jul 04, 2010 at 10:15:13AM +0300, Baruch Siach wrote:
> > > > > > The GPIO and IRQ/GPIO registers need protection from concurrent access for
> > > > > > operations that are not atomic.
> > > > > Apart from this other architectures do not use locking here aswell.
> > > > 
> > > > The Nomadic gpio driver does use a spinlock for mask/unmask operations.
> > > > 
> > > > What about the _set_gpio_direction, and mxc_gpio_set? These functions may be 
> > > > called from a process context (e.g., via sysfs). A context switch between 
> > > > __raw_readl and __raw_writel will cause corruption.
> > > 
> > > The gpio_chip functions are protected by a single spinlock in
> > > gpiolib.
> > 
> > gpio_direction_input uses the gpio_lock for its own internal sanity check, and 
> > releases it before calling chip->direction_input. The same goes for 
> > gpio_direction_output.
> 
> Ok, true.
> 
> > 
> > The __gpio_set_value function seems not acquire any lock before calling 
> > chip->set.
> > 
> > > The gpio related registers and the irq related regsiters are
> > > totally orthogonal, so we need no locking between these registers.
> > 
> > True.
> 
> This means we need locking for the gpio functions but not for the irq
> functions.

Agreed. I'll post an updated patch shortly.

baruch

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