Quoting Eric Anholt (2015-08-12 11:04:05)

Michael Turquette [off-list ref] writes:

quoted

Hi Eric,

Quoting Eric Anholt (2015-07-20 12:33:01)

quoted

+void __init rpi_firmware_init_clock_provider(struct device_node *node)
+{
+       /* We delay construction of our struct clks until get time,
+        * because we need to be able to return -EPROBE_DEFER if the
+        * firmware driver isn't up yet.  clk core doesn't support
+        * re-probing on -EPROBE_DEFER, but callers of clk_get can.
+        */
+       of_clk_add_provider(node, rpi_firmware_delayed_get_clk, node);
+}
+
+CLK_OF_DECLARE(rpi_firmware_clocks, "raspberrypi,bcm2835-firmware-clocks",
+              rpi_firmware_init_clock_provider);

Do you require CLK_OF_DECLARE here? Could this be a platform driver
instead?

I'm not actually sure.  The common pattern seemed to be using
CLK_OF_DECLARE (130 files using it versus declaring a struct
platform_driver), and it seems to avoid a whole lot of boilerplate.
What would the advantage be?

Correctly using the Linux driver model, having a nice struct device and
things like supsend and resume handlers, etc.

There is a lot of CLK_OF_DECLARE going on in drivers/clk, but I am
hoping we can fix that. The main reason for CLK_OF_DECLARE is when you
need to register clocks very early (e.g. timers). This is often not the
case.

Regards,
Mike