Re: [PATCH v2 2/4] gpio: add viperboard gpio driver

[Lars Poeschel <poeschel@lemonage.de>]

On Tuesday 16 October 2012 at 12:00:13, Linus Walleij wrote:
> On Tue, Oct 16, 2012 at 8:51 AM, Lars Poeschel <poeschel@lemonage.de> wrote:
> > On Monday 15 October 2012 at 15:00:12, Linus Walleij wrote:
> >> > +       /* if io is set to output, just return the saved value */
> >> > +       if (gpio->gpioa_out & (1 << offset))
> >> > +               return gpio->gpioa_val & (1 << offset);
> >> 
> >> That's not going to work if the hardware changes state
> >> behind the back of the driver right? Oh well, maybe
> >> it doesn't matter.
> > 
> > I thought about that and then did this "cache" only in case the gpio is a
> > output to save to usb ping-pong that is needed otherwise. I thought that
> > nothing can change to state of the output but the driver itself.
> 
> On a second note there is already a standard mechanism for caching
> registers in the kernel, and that is regmap. Sadly it's a bit
> undocumented but there are several examples and the code
> lives in drivers/base/regmap and include/linux/regmap.h

I had a look at regmap. This is interesting. But there is no regmap_bus 
implementation for usb. Are you pointing me in this direction ? ;-) I don't 
think you want me as a kernel newbie write this code.
And another thing: I don't think this can be implemented in such a general way 
as it is done already for i2c or mmio. As you see in the example of my 
(viperboard-)adapter, there is a bunch bytes sent, just to set the direction 
bit in a register. And it is not guaranteed, that this "address" and "value" 
are at the same position for other usb adapters. And this is getting even 
worse, if I think of reading values out of the viperboard...
I would stay at the "cache" implementation I presented in the patch - this is 
too special.

Lars