On Tue, 9 Aug 2016, Paul Gortmaker wrote:

[Re: [PATCH v18 6/6] ARM: socfpga: fpga bridge driver support] On 08/08/2016 (Mon 13:44) Moritz Fischer wrote:

quoted

Hi Alan,

On Mon, Aug 8, 2016 at 12:18 PM, atull [off-list ref] wrote:

quoted

quoted

Please don't use module.h in drivers controlled by a bool
Kconfig setting.

THanks,
Paul.

Thanks for the feedback.  Can you provide an example of what you
would consider to be proper usage in the kernel?

I think Paul is suggesting to use

static int __init alt_fpga_bridge_init(void)
{
        platform_driver_register(&alt_fpga_bridge_driver);
}

device_initcall(alt_fpga_bridge_init);

or better:

builtin_platform_driver(&alt_fpga_bridge_driver);

Like for example in: drivers/cpuidle/cpuidle-mvebu-v7.c

Yes, pretty much that -- if you have a bool Kconfig, you should be using
builtin registration functions, and have no need for module.h or
anything MODULE_<xyz> or any module_init/module_exit calls.

An empty file containing nothing but #include <linux/module.h> will
cause cpp to emit about 750k of goop, so we really should only be using
it for drivers that are genuinely modular; i.e. tristate Kconfig.

Thanks,
Paul.
--

quoted

Cheers,

Moritz

Thanks for the feedback and explanations!

I've retested my stuff with it all built as modules (mgr, bridged,
and fpga-region) and it all works that way as well as built in.
So I'll fix up the Kconfig as tristates for everybody.  Also
I'll add some dependencies as FPGA_REGION should be dependent
on FPGA_BRIDGE.

Alan