
Hi Stephen,
On 30 September 2014 19:40, Stephen Warren swarren@wwwdotorg.org wrote:
On 09/22/2014 05:30 PM, Simon Glass wrote:
Convert the BCM2835 GPIO driver to use driver model, and switch over Raspberry Pi to use this, since it is the only board.
diff --git a/drivers/gpio/bcm2835_gpio.c b/drivers/gpio/bcm2835_gpio.c
+static inline bool gpio_is_requested(struct bcm2835_gpios *gpios, int offset) {
return *gpios->label[offset] != '\0';
}
+static int check_requested(struct udevice *dev, unsigned offset,
const char *func)
{
struct bcm2835_gpios *gpios = dev_get_priv(dev);
struct gpio_dev_priv *uc_priv = dev->uclass_priv;
if (!gpio_is_requested(gpios, offset)) {
printf("omap_gpio: %s: error: gpio %s%d not requested\n",
func, uc_priv->bank_name, offset);
return -EPERM;
}
return 0;
}
It seems like the core GPIO layer should be doing the high-level management of GPIO requesting, so each driver doesn't have to duplicate the code?
Yes, agreed. I figured this was coming but took a conservative approach with the uclass to start with. I will be taking a look at moving this logic into the uclass soon. I've done enough GPIO drivers to prove this out.
Regards, Simon