[U-Boot] [PATCH RFC] armv7: fixloop: don't fixup if location is NULL

[Minkyu Kang]

Dear Wolfgang Denk,

On 10 January 2011 21:14, Wolfgang Denk <wd at denx.de> wrote:
> Dear Minkyu Kang,
>
> In message <AANLkTimgBjOSE8c+_6AsUon5KnNR1_UKNqc=Wf_UD2+J at mail.gmail.com> you wrote:
>>
>> >> int (*test_func)(void);
>> >
>> > This results in a symbol in bss segment, right?
>> >
>> >> And then, set to NULL at arch_cpu_init()
>> >
>> > Such an assignment is illegal then. Bss has not been initalized before
>> > relocation, and must not be accessed (neither read nor write).
>>
>> Illegal? as a result, yes.
>
> No, illegal as an action.  You MUST NOT access any symbols in BSS
> before relocation (more precisely, before bss has been initialized).
>
> And you MUST NOT write any symbols in data segment before relocation,
> either.
>
> In both cases, the result of such actions is undefined behaviour.
>
>> But we do many things before the reloaction as arch init, board init and so on.
>
> Of course, but as mentioned we must not read or write to symbols in
> bss, and we must not write to symbols in data segment.
>
>> How about lcd_setmem function?
>> panel_info is located at bss area, but lcd_setmem access this structure.
>> Is it illegal?
>
> This must not be done before relocation.
>

No, please see 360 line of arch/arm/lib/board.c
This function is called before relocation.

And how about init_func_i2c()?
This function is called twice, before the relocation and after relocation.
When we use board_i2c_init function then, there is possibility that
use symbols in bss because of this function is called after
relocation.

If we ignore this exception, it will be a big constraint.

btw, there are any side effects on my patch?
I think.. It is just a little safety feature.

Thanks
Minkyu Kang
-- 
from. prom.
www.promsoft.net