Re: [PATCH RFC] power: reset: restart-poweroff: convert to module

[Sverdlin, Alexander]

On Wed, 2024-02-21 at 22:56 +0100, Andrew Lunn wrote:
> > @@ -47,15 +45,8 @@ static const struct of_device_id of_restart_poweroff_match[] = {
> >   };
> >   MODULE_DEVICE_TABLE(of, of_restart_poweroff_match);
> >   
> > -static struct platform_driver restart_poweroff_driver = {
> > -       .probe = restart_poweroff_probe,
> > -       .remove = restart_poweroff_remove,
> > -       .driver = {
> > -               .name = "poweroff-restart",
> > -               .of_match_table = of_restart_poweroff_match,
> > -       },
> > -};
> 
> of_restart_poweroff_match now seems to be disconnected from the
> driver.
> 
> kirkwood-linkstation.dtsi:              compatible = "restart-poweroff";
> kirkwood-lsxl.dtsi:             compatible = "restart-poweroff";
> orion5x-linkstation.dtsi:               compatible = "restart-poweroff";
> orion5x-lswsgl.dts:             compatible = "restart-poweroff";
> 
> How do these devices get this driver loaded?
> 
> This appears to be another reason to NACK it.

That's why MODULE_DEVICE_TABLE() was preserved for backwards compatibility,
because *loading* happens via MODULE_DEVICE_TABLE(). But I didn't realize
it was never buildable as module as Sebastian pointed out, because of
machine_restart().

For your use case it would continue to work as before I believe, just
the callback would be installed because of the fact the code
was compiled-in, not because there was a fake platform device.

I also didn't understand what is so special about bootloader support
for this functionality if no data is passed to the bootloader.
After ARM-specifics was removed from the code quite some time ago
any platform could re-use the code for the deployments which meant
to be "always on".

But If the resistance is so serious, so be it.

-- 
Alexander Sverdlin
Siemens AG
www.siemens.com