On Sunday, October 05, 2014 07:36:04 PM Alexandre Courbot wrote:

On Wed, Oct 1, 2014 at 5:03 PM, Arnd Bergmann [off-list ref] wrote:

quoted

On Wednesday 01 October 2014 04:12:41 Rafael J. Wysocki wrote:

quoted

+       static const char * const suffixes[] = { "gpios", "gpio" };
+       struct acpi_device *adev = ACPI_COMPANION(dev);
        struct acpi_gpio_info info;
        struct gpio_desc *desc;
+       char propname[32];
+       int i;

-       desc = acpi_get_gpiod_by_index(dev, idx, &info);
-       if (IS_ERR(desc))
-               return desc;
+       /* Try first from _DSD */
+       for (i = 0; i < ARRAY_SIZE(suffixes); i++) {
+               if (con_id && strcmp(con_id, "gpios")) {
+                       snprintf(propname, sizeof(propname), "%s-%s",
+                                con_id, suffixes[i]);
+               } else {
+                       snprintf(propname, sizeof(propname), "%s",
+                                suffixes[i]);
+               }

The general interface seems fine, but I'd be happier if you didn't
try to support all four of the possible syntaxes we have in DT.
It would be much better to have only "gpios" and not "gpio", and
the "foo-gpios" syntax should be replaced with whatever method you
use to name other subsystem specific links. For most subsystems
we now use something like "gpio-names", but unfortunately the GPIO
binding goes back to the time before we had come to that agreement.
The same applies to regulators.

Wouldn't restricting the naming scheme cause problems for ACPI drivers
that want to reuse existing DT bindings? Since DT bindings are set in
stone, this means we would have to use different properties for DT and
ACPI.

Which we would like to avoid.

Generally speaking, there are drivers (or upper-layer code) that will use
separate code paths for ACPI and DTs anyway (like the PCI root bridge
code) and in those cases having ACPI-specific properties or device data
in general is not really a problem.

However, for pieces of code that we'd like to use the unified properties
API going forward, we very much need the properties to be the same for
DTs and ACPI.

I agree that there are too much ways to define GPIOs, but I'm afraid
we will have to carry them over for the sake of consistency.

Well, I suppose so.

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.