Re: Bogus section mismatch errors?

[Sam Ravnborg <sam@ravnborg.org>]

On Tue, May 08, 2007 at 12:17:59PM +0100, Russell King wrote:
> 
> WARNING: arch/arm/mach-realview/built-in.o - Section mismatch:
>  reference to .init.text: from .data between 'realview_timer' (at offset 0x54)
>  and 'clcd_plat_data'
> 
> Weird, these two data structures don't interact with each other.  Probably
> bogus.
modpsot does it best to guess what symbol are represented by an address.
In this case I would assume that realview_timer has a function pointer to a function
that is marked __init.

> 
> WARNING: init/built-in.o - Section mismatch:
>  reference to .init.text: from .text between 'rest_init' (at offset 0x2c)
>  and 'run_init_process'
This is due to the reference to kernel_init - but binutils hardcode the offset
so modpost does not recognize the symbol.
A workaround (that I do not like) it to remove static declaration of kernel_init,
then modpost will recognize the symbol and see this is OK.


> WARNING: mm/built-in.o - Section mismatch:
>  reference to .init.text: from .text between 'kmem_cache_create' (at offset 0x1bd94)
>  and 'cache_reap'
> WARNING: mm/built-in.o - Section mismatch:
>  reference to .init.text: from .text between 'kmem_cache_create' (at offset 0x1bdcc)
>  and 'cache_reap'
There are in slab a function that is not marked __init that refer to a function marked
__init. But the logic surrounding makes it OK.

I have no fix/workaround for this.
Andres has suggested that we should mark functions/data that may refer to __init{data}
with something like: __nowarn and __nowarndata

I am planning to implement something so we can mark functions/data with:
__init_ref_ok and _init_data_ref_ok
Which is indeed more ugly but expalins what is actually OK.

I hope to take a look during the weekend.

	Sam