[En-Nut-Discussion] Nut/OS Initialization

[Thiago A. Corrêa]

On Wed, Sep 17, 2008 at 5:59 AM, Ole Reinhardt
<ole.reinhardt at embedded-it.de> wrote:
> Hi Harald,
>
>> > For example in linux this is done by filling up a struct like this one
>> >
>> > MACHINE_START(TRIZEPS4, "Keith und Koep Trizeps IV module")
>> >         /* MAINTAINER("Jürgen Schindele") */
>> >         .phys_io        = 0x40000000,
>> >         .io_pg_offst    = (io_p2v(0x40000000) >> 18) & 0xfffc,
>> >         .boot_params    = TRIZEPS4_SDRAM_BASE + 0x100,
>> >         .init_machine   = trizeps4_init,
>> >         .map_io         = trizeps4_map_io,
>> >         .init_irq       = pxa_init_irq,
>> >         .timer          = &pxa_timer,
>> > MACHINE_END
>> >
>> > and linking it with the application. Right now I don't know the _exact_
>> > mechanism, but I should not be that complicate.
>>
>> I understand the data structure and how it forces the linker to include
>> the driver. But I do not understand, how the program code becomes aware
>> of this structure to make use of the specified routines. In other words:
>> How does the initialization code know that TRIZEPS4 exists and that it
>> can call trizeps4_init() via (*init_machine)()?
>
> I'll examine this mechanism futher on friday. Today and tomorrow I won't
> be in office and will have to do some service work at some customers...
>
> But in general this kind of glue logic is used in several places in the
> linux kernel and allows you quite nice driver design.
>

I'm no kernel hacker, but I guess this is more for modules. At least
ARMs and other embedded platforms have a board specific code under
arch that does PIN and IRQ reservations and registers hardware,
passing structs. In that regards, it's no different from Ethernut.

Regards,
     Thiago A. Correa