wt., 23 lip 2019 o 17:32 Andy Shevchenko
[off-list ref] napisał(a):

On Tue, Jul 23, 2019 at 08:28:00AM +0200, Bartosz Golaszewski wrote:

quoted

pon., 22 lip 2019 o 18:06 Andy Shevchenko
[off-list ref] napisał(a):

quoted

On Mon, Jul 22, 2019 at 05:02:57PM +0200, Bartosz Golaszewski wrote:

quoted

From: Bartosz Golaszewski <redacted>

Now that the last user of platform data (sh ecovec24) defines a proper
GPIO lookup and sets the 'default-on' device property, we can drop the
platform_data-specific GPIO handling and unify a big chunk of code.

The only field used from the platform data is now the fbdev pointer.

quoted

-static int gpio_backlight_probe_dt(struct platform_device *pdev,
-                                struct gpio_backlight *gbl)
-{
-     struct device *dev = &pdev->dev;
-     enum gpiod_flags flags;
-     int ret;
-
-     gbl->def_value = device_property_read_bool(dev, "default-on");
-     flags = gbl->def_value ? GPIOD_OUT_HIGH : GPIOD_OUT_LOW;
-
-     gbl->gpiod = devm_gpiod_get(dev, NULL, flags);
-     if (IS_ERR(gbl->gpiod)) {
-             ret = PTR_ERR(gbl->gpiod);
-
-             if (ret != -EPROBE_DEFER) {
-                     dev_err(dev,
-                             "Error: The gpios parameter is missing or invalid.\n");
-             }
-             return ret;
-     }
-
-     return 0;
-}

Why not leave this function (perhaps with different name)?

Why would we do that if the entire probe() function is now less than
50 lines long? Also: it gets inlined by the compiler anyway. It
doesn't make sense IMO.

I'm not against this, perhaps, dropping and moving can be split to two changes.

This really is unnecessary - we can do it in a single patch alright.

Bart