[PATCH v5] clk: sunxi: Add CLK_OF_DECLARE support for sun8i-a23-apb0-clk driver

[Chen-Yu Tsai]

On Mon, Dec 7, 2015 at 5:36 PM, Maxime Ripard
<maxime.ripard at free-electrons.com> wrote:
> Hi,
>
> On Thu, Dec 03, 2015 at 03:05:30PM +0800, Chen-Yu Tsai wrote:
>> The APBS clock on sun9i is the same as the APB0 clock on sun8i. With
>> sun9i we are supporting the PRCM clocks by using CLK_OF_DECLARE,
>> instead of through a PRCM mfd device and subdevices for each clock
>> and reset control. As such we need a CLK_OF_DECLARE version of
>> the sun8i-a23-apb0-clk driver.
>>
>> Also, build it for sun9i/A80, and not just for configurations with
>> MFD_SUN6I_PRCM enabled.
>>
>> Signed-off-by: Chen-Yu Tsai <wens at csie.org>
>> ---
>>
>> Changes since v4:
>>
>>   - Keep building clk-sun8i-apb0 for SUN6I_MFD_PRCM.
>>
>>   - Add an error message and comment for when of_io_request_and_map()
>>     fails. of_io_request_and_map() merges a bunch of errors into -EINVAL,
>>     so this might not be the best approach. But I think having an error
>>     message when we know something is wrong (-EBUSY, -ENOMEM) is better.
>>
>> ---
>>  drivers/clk/sunxi/Makefile         |  1 +
>>  drivers/clk/sunxi/clk-sun8i-apb0.c | 80 ++++++++++++++++++++++++++++++++------
>>  2 files changed, 69 insertions(+), 12 deletions(-)
>>
>> diff --git a/drivers/clk/sunxi/Makefile b/drivers/clk/sunxi/Makefile
>> index 103efab05ca8..ccf21ba3b6b0 100644
>> --- a/drivers/clk/sunxi/Makefile
>> +++ b/drivers/clk/sunxi/Makefile
>> @@ -15,6 +15,7 @@ obj-y += clk-sun9i-core.o
>>  obj-y += clk-sun9i-mmc.o
>>  obj-y += clk-usb.o
>>
>> +obj-$(CONFIG_MACH_SUN9I) += clk-sun8i-apb0.o
>>  obj-$(CONFIG_MACH_SUN9I) += clk-sun9i-cpus.o
>>
>>  obj-$(CONFIG_MFD_SUN6I_PRCM) += \
>> diff --git a/drivers/clk/sunxi/clk-sun8i-apb0.c b/drivers/clk/sunxi/clk-sun8i-apb0.c
>> index 7ae5d2c2cde1..7ba61103a6f5 100644
>> --- a/drivers/clk/sunxi/clk-sun8i-apb0.c
>> +++ b/drivers/clk/sunxi/clk-sun8i-apb0.c
>> @@ -17,13 +17,77 @@
>>  #include <linux/clk-provider.h>
>>  #include <linux/module.h>
>>  #include <linux/of.h>
>> +#include <linux/of_address.h>
>>  #include <linux/platform_device.h>
>>
>> +static struct clk *sun8i_a23_apb0_register(struct device_node *node,
>> +                                        void __iomem *reg)
>> +{
>> +     const char *clk_name = node->name;
>> +     const char *clk_parent;
>> +     struct clk *clk;
>> +     int ret;
>> +
>> +     clk_parent = of_clk_get_parent_name(node, 0);
>> +     if (!clk_parent)
>> +             return ERR_PTR(-EINVAL);
>> +
>> +     of_property_read_string(node, "clock-output-names", &clk_name);
>> +
>> +     /* The A23 APB0 clock is a standard 2 bit wide divider clock */
>> +     clk = clk_register_divider(NULL, clk_name, clk_parent, 0, reg,
>> +                                0, 2, CLK_DIVIDER_POWER_OF_TWO, NULL);
>> +     if (IS_ERR(clk))
>> +             return clk;
>> +
>> +     ret = of_clk_add_provider(node, of_clk_src_simple_get, clk);
>> +     if (ret)
>> +             goto err_unregister;
>> +
>> +     return clk;
>> +
>> +err_unregister:
>> +     clk_unregister_divider(clk);
>> +
>> +     return ERR_PTR(ret);
>> +}
>> +
>> +static void sun8i_a23_apb0_setup(struct device_node *node)
>> +{
>> +     void __iomem *reg;
>> +     struct resource res;
>> +     struct clk *clk;
>> +
>> +     reg = of_io_request_and_map(node, 0, of_node_full_name(node));
>> +     if (IS_ERR(reg)) {
>> +             /*
>> +              * This happens with clk nodes instantiated through mfd,
>> +              * as those do not have their resources assigned in the
>> +              * device tree. Do not print an error in this case.
>> +              */
>> +             if (PTR_ERR(reg) != -EINVAL)
>> +                     pr_err("Could not get registers for a23-apb0-clk\n");
>
> This is not the only case you have to take into account.
>
> There's also the case when you have a regular clock (and by regular I
> mean that is not in the PRCM) that will be probed by the
> CLK_OF_DECLARE mechanism and then later by the device model.
>
> In such a case, the second of_io_request_and_map will fail, and you
> will have an error returned that you do not ignore at the moment.

Right. It will return -EBUSY. But ignoring it and returning 0 is telling
the driver core that the device successfully binded. I think this is
wrong.

Normal clocks should be in the "clocks" node, and wouldn't be probed a
second time through the device model, would it? Am I missing something?

ChenYu