[PATCH 02/17] ARM: at91: use machine specific hook for late init

Jean-Christophe PLAGNIOL-VILLARD plagnioj at jcrosoft.com
Fri Apr 27 10:26:01 EDT 2012

On 22:07 Fri 27 Apr     , Shawn Guo wrote:
> On Thu, Apr 26, 2012 at 05:29:06PM +0200, Jean-Christophe PLAGNIOL-VILLARD wrote:
> > I like it but I think it's more time to put soc init that machine init
> > 
> Yes, just like other hooks in struct machine_desc, when there is no
> machine specific but only soc specific setup need to get done, the hook
> is pretty much like a soc specific one.  But when you get some machine
> specific setup besides soc one, you will be happy to have it.  Taking
> board "Carmeva" as exmaple, it will be something like the following.
> static void __init carmeva_init_late(void)
> {
> 	at91_init_late();
> 	carmeva_specific_late_init();
> }
>         /* Maintainer: Conitec Datasystems */
>         .timer          = &at91rm9200_timer,
>         .map_io         = at91_map_io,
>         .init_early     = carmeva_init_early,
>         .init_irq       = at91_init_irq_default,
>         .init_machine   = carmeva_board_init,
>         .init_late      = carmeva_init_late,
no really don't like this idea to duplicate code
I didn't like either for timer, or any other common init

we discuss with Arnd we came back with 2 ideas one is to use a machine
specific initcall or compatible specifc initcall or the second one is to
resurect Marc patch series to introduce soc_desc

and on at91 Ill not touch it as the old style board can NOT be compiled with
other soc (only one at91 soc at a time).

And I'll not fix it as we move to the DT which I fix to be able to be compiled
in the multiarch kernel.

Best Regards,

More information about the linux-arm-kernel mailing list