On Thu, Jul 2, 2020 at 8:40 AM Dong Aisheng <donga...@gmail.com> wrote:
> On Thu, Jul 2, 2020 at 2:11 PM Anson Huang <anson.hu...@nxp.com> wrote:
> > > Subject: Re: [PATCH V4 3/5] clk: imx: Support building i.MX common clock
> >
> > I am fine of adding the '#ifndef MODULE' to imx_clk_disable_uart() and
> > imx_keep_uart_clocks_param()
> > as well in next patch series. Others like ' imx_keep_uart_clocks ' and
> > imx_register_uart_clocks() need to
> > be kept always built, since they are used by each clock driver no matter
> > built-in or module build.
> >
> > So that means I have to add another 'ifndef MODULE' or I need to adjust
> > some code sequence to make
> > those code can be built-out in same block and just use single 'ifndef
> > MODULE', I think adjust the code
> > sequence should be better, will go with this way.
>
> What if we condionally compile it in clk.h? Will that be easiser?
Yes, that's what I expected to see in v4 after the previous discussion. If
imx_register_uart_clocks() is an empty inline function, then the arrays
passed to it are unused and also get dropped by the compiler.
The question is whether the #ifdef check in the header to test for MODULE
(only calling it if this particular soc has a built-in clk driver,
which is sufficient)
or for IS_MODULE(CONFIG_MXC_CLK) (call it if _any_ clk driver is built-in
and the function exists, which leaves extra code in the driver but is a more
conventional check).
Arnd
