On Fri, Jun 29, 2012 at 08:14:38PM +0800, Dong Aisheng wrote:

quoted

I'm with Lothar on this.  At least for imx28 case, we do not even need
dummy clock, because NULL con_id in clkdev lookup will have all those
clocks match the only one gpmi-nand clock on imx28, as I already said.

I'm not quite understand. Can you clarify a bit more?
For extra clocks need by mx6q like "gpmi_bch", they're indeed have con_id.
Getting them will not fail on imx28 without dummy clock?

You need to look at how imx28 registers clkdev for gpmi-nand.

static struct clk_lookup gpmi_lookups[] __initdata = {
        { .dev_id = "imx28-gpmi-nand", },
        { .dev_id = "8000c000.gpmi", },
};

clk_register_clkdevs(clks[gpmi], gpmi_lookups, ARRAY_SIZE(gpmi_lookups));

As you can see, con_id is NULL for imx28 case.  As NULL is actually
a wildcard for looking up, all the gpmi-nand clocks match it.

-- 
Regards,
Shawn