On Thursday 02 August 2012, Chao Xie wrote:

quoted

#define APBC_RTC       0x28

apbc_clks[rtc_clk] = mmp_clk_register_apbc(rtc_clk, clk32k, APBC_RTC, 10, APBC_POWER_CTRL, mmp_clk_lock);
clk_register_clkdev(apbc_clks[rtc_clk], NULL, "sa1100-rtc");

        Arnd

hi
I would like to keep the mmp_clk_register_apbc to receive the "reg
base" not "reg offset".
It will be aligned with other kind of clock register APIs.
To read out APBC base register from device tree can be added at the
clock-pxa168.c, and it can map the registers and pass to the
mmp_clk_register_apbc.

Right, my mistake.

The above should have been something like

#define APBC_RTC       0x28
apbc_clks[rtc_clk] = mmp_clk_register_apbc(rtc_clk, clk32k, clock_base + APBC_RTC, 10, APBC_POWER_CTRL, mmp_clk_lock);
clk_register_clkdev(apbc_clks[rtc_clk], NULL, "sa1100-rtc");

instead, with clock_base pointing to the __iomem token for the clock controller.

Now, i have talked to Haojian who is doing device tree maintainer in
pxa/mmp. This kind of support is not added.
I suggest that after device tree support in clock can be added later
after other functionality of the clock framework is fine.

You can do device tree support as a second step, but in this first step, you
should already start using ioremap to get the virtual address of the
clock controller, rather than hardcoding it.

	Arnd