On 10/23/2015 6:02 AM, Laurent Pinchart wrote:

Hi Rob,

On Friday 23 October 2015 06:51:28 Rob Herring wrote:

quoted

On Fri, Oct 23, 2015 at 4:08 AM, Laurent Pinchart wrote:

quoted

On Thursday 22 October 2015 18:41:05 Rob Herring wrote:

quoted

On Thu, Oct 22, 2015 at 1:52 PM, Ray Jui wrote:

quoted

On 10/22/2015 11:43 AM, Rob Herring wrote:

quoted

On Mon, Oct 19, 2015 at 12:43 AM, Pramod Kumar wrote:

quoted

Add ngpios property to the gpio controller's DT node so that
controller driver extracts total number of gpio lines present in
controller from DT and removes dependency on driver.

Signed-off-by: Pramod Kumar <redacted>
Reviewed-by: Ray Jui <rjui@broadcom.com>
Reviewed-by: Scott Branden <sbranden@broadcom.com>
---

  Documentation/devicetree/bindings/pinctrl/brcm,cygnus-gpio.txt | 5
  +++
  1 file changed, 5 insertions(+)

diff --git
a/Documentation/devicetree/bindings/pinctrl/brcm,cygnus-gpio.txt
b/Documentation/devicetree/bindings/pinctrl/brcm,cygnus-gpio.txt
index f92b833..655a8d7 100644

--- a/Documentation/devicetree/bindings/pinctrl/brcm,cygnus-gpio.txt
+++ b/Documentation/devicetree/bindings/pinctrl/brcm,cygnus-gpio.txt

@@ -10,6 +10,9 @@ Required properties:
       Define the base and range of the I/O address space that
       contains

the Cygnus

   GPIO/PINCONF controller registers

+- ngpios:

+    Total number of GPIOs the controller provides

This must be optional for compatibility and the driver needs to handle
it not present.

You meant to be compatible with existing Cygnus devices, correct?

Just to clarify, here you suggest we still leave the existing hard
coded ngpios in the driver, in order to be compatible with all existing
Cygnus devices (while the Cygnus device tree changes to use ngpio is
still being merged and through different maintainer), and have all new
iProc SoCs switch to use ngpios from device tree, right?

Yes, an existing dtb should continue to work with a new kernel. You
can add the DT property to the older devices too and then eventually
remove the hard coded values some time in the future. That could be
immediately (don't care about compatibility at all), a couple of
kernel cycles, never... It all depends on users of the impacted
platforms.

But shouldn't the property still be documented as required to ensure that
new DTs always include it ?

Good point. If the intent is to eventually remove it from the driver,
then yes. We probably need "required for new designs" as a category or
maybe "recommended"? The wording is not so important here, but I'm
thinking about as we try to standardize the naming.

Required for new designs sounds better than recommended. Or maybe something
like "Required (optional for backward compatibility)".

Yes, it should be phrased as "required for new designs" because all new 
iProc SoCs using this GPIO driver need to have this in DT.

Thanks,

Ray