[U-Boot] arm: timer_init() and bss (relocation)

[U-Boot] arm: timer_init() and bss (relocation)

[Alexander Holler]

Hello,

because I still have problems where I think the problem is the 
relocation, I've had a look at some timer code.

E.g. in arch/arm/cpu/arm926ejs/kirkwood/timer.c  on top there is

struct kwtmr_registers *kwtmr_regs = (struct kwtmr_registers 
*)KW_TIMER_BASE;

which is used in timer_init().

In arch/arm/cpu/armv7/omap-common/timer.c we have

static struct gptimer *timer_base = (struct gptimer *)CONFIG_SYS_TIMERBASE;

which is used in timer_init() too.

(I assume similiar code is used in more files, I've just had a look at 
those two.)

timer_init() is called in board_init_f() through init_sequence, that 
means before relocation.

Did I miss something, or I'm right that this is a problem because those 
two variables are stored in bss but are accessed before relocation?

Regards,

Alexander

[U-Boot] arm: timer_init() and bss (relocation)

[John Rigby]

On Fri, Jan 21, 2011 at 11:07 PM, Alexander Holler <holler@ahsoftware.de> wrote:
> Hello,
>
> because I still have problems where I think the problem is the
> relocation, I've had a look at some timer code.
>
> E.g. in arch/arm/cpu/arm926ejs/kirkwood/timer.c ?on top there is
>
> struct kwtmr_registers *kwtmr_regs = (struct kwtmr_registers
> *)KW_TIMER_BASE;
>
> which is used in timer_init().
>
> In arch/arm/cpu/armv7/omap-common/timer.c we have
>
> static struct gptimer *timer_base = (struct gptimer *)CONFIG_SYS_TIMERBASE;
>
> which is used in timer_init() too.
>
> (I assume similiar code is used in more files, I've just had a look at
> those two.)
>
> timer_init() is called in board_init_f() through init_sequence, that
> means before relocation.
>
> Did I miss something, or I'm right that this is a problem because those
> two variables are stored in bss but are accessed before relocation?
>
These are both initialized data so they do not go in bss.
> Regards,
>
> Alexander
> _______________________________________________
> U-Boot mailing list
> U-Boot at lists.denx.de
> http://lists.denx.de/mailman/listinfo/u-boot
>

[U-Boot] arm: timer_init() and bss (relocation)

[Alexander Holler]

Am 22.01.2011 07:38, schrieb John Rigby:
> On Fri, Jan 21, 2011 at 11:07 PM, Alexander Holler<holler@ahsoftware.de>  wrote:
>> Hello,
>>
>> because I still have problems where I think the problem is the
>> relocation, I've had a look at some timer code.
>>
>> E.g. in arch/arm/cpu/arm926ejs/kirkwood/timer.c  on top there is
>>
>> struct kwtmr_registers *kwtmr_regs = (struct kwtmr_registers
>> *)KW_TIMER_BASE;
>>
>> which is used in timer_init().
>>
>> In arch/arm/cpu/armv7/omap-common/timer.c we have
>>
>> static struct gptimer *timer_base = (struct gptimer *)CONFIG_SYS_TIMERBASE;
>>
>> which is used in timer_init() too.
>>
>> (I assume similiar code is used in more files, I've just had a look at
>> those two.)
>>
>> timer_init() is called in board_init_f() through init_sequence, that
>> means before relocation.
>>
>> Did I miss something, or I'm right that this is a problem because those
>> two variables are stored in bss but are accessed before relocation?
>>
> These are both initialized data so they do not go in bss.

Correct, I should have a look into u-boot.map before writing this mail. 
Thanks for the pointer.

But this leads me to the question what happens with the stuff in the 
data section. Seems I should have again a look at the lowlevel stuff to 
understand that.

Thanks,

Alexander

[U-Boot] arm: timer_init() and bss (relocation)

[Reinhard Meyer]

Dear Alexander Holler,

> Am 22.01.2011 07:38, schrieb John Rigby:
>> On Fri, Jan 21, 2011 at 11:07 PM, Alexander Holler<holler@ahsoftware.de>   wrote:
>>> Hello,
>>>
>>> because I still have problems where I think the problem is the
>>> relocation, I've had a look at some timer code.
>>>
>>> E.g. in arch/arm/cpu/arm926ejs/kirkwood/timer.c  on top there is
>>>
>>> struct kwtmr_registers *kwtmr_regs = (struct kwtmr_registers
>>> *)KW_TIMER_BASE;
>>>
>>> which is used in timer_init().
>>>
>>> In arch/arm/cpu/armv7/omap-common/timer.c we have
>>>
>>> static struct gptimer *timer_base = (struct gptimer *)CONFIG_SYS_TIMERBASE;
>>>
>>> which is used in timer_init() too.
>>>
>>> (I assume similiar code is used in more files, I've just had a look at
>>> those two.)
>>>
>>> timer_init() is called in board_init_f() through init_sequence, that
>>> means before relocation.
>>>
>>> Did I miss something, or I'm right that this is a problem because those
>>> two variables are stored in bss but are accessed before relocation?
>>>
>> These are both initialized data so they do not go in bss.
>
> Correct, I should have a look into u-boot.map before writing this mail.
> Thanks for the pointer.
>
> But this leads me to the question what happens with the stuff in the
> data section. Seems I should have again a look at the lowlevel stuff to
> understand that.

Intuitively (without looking at the code) I would think that
CONST DATA is ok,
but non const, initialized data cannot work because it is not
initialized before relocation...

Best Regards,
Reinhard

[U-Boot] arm: timer_init() and bss (relocation)

[Albert ARIBAUD]

Le 22/01/2011 08:21, Reinhard Meyer a ?crit :

>> But this leads me to the question what happens with the stuff in the
>> data section. Seems I should have again a look at the lowlevel stuff to
>> understand that.
>
> Intuitively (without looking at the code) I would think that
> CONST DATA is ok,
> but non const, initialized data cannot work because it is not
> initialized before relocation...

To be precise:

1. const data are readable before and after relocation.

2. non-const initialized data are readable only until relocated, and 
then writable as well.

3. BSS data cannot and must not be used at all until relocated, and are 
then zeroed out before board_init_r() starts.

> Best Regards,
> Reinhard

Amicalement,
-- 
Albert.