[U-Boot] [RFC] ARM: mx31pdk: Use the new relocation scheme

[Albert ARIBAUD]

Le 11/02/2011 11:51, Aneesh V a écrit :
> Hi Albert,
>
> On Wednesday 09 February 2011 02:20 AM, Albert ARIBAUD wrote:
>> Le 08/02/2011 21:18, stefano babic a écrit :
>>> Am 08.02.2011 20:26, schrieb Magnus Lilja:
>>>> Patch reposted as a separate mail a couple of minutes ago.
>>>>
>>>> As I mention in the patch I think Fabio's patch has to be applied
>>>> first.
>>>
>>> I think your patch is ok - Fabio fixed the syntax error as you do. We
>>> need only one of them.
>>>
>>>> Another solution would be to change my patch somewhat to apply it first
>>>> and then update Fabios patch to only touch the i.MX31-PDK specific
>>>> files.
>>>
>>> IMHO this is the preferred way, because the two issues are orthogonal.
>>> Your patch fixes booting from NAND for ARM11, and Fabio's patch fix the
>>> mx31pdk board only.
>>
>> Agreed.
>>
>> Note also that there was a recent patch to ARM926's start.S (replacing
>> 'adr r1, _start' with 'ldr r1, _TEXT_BASE' at line 284). The same should
>> be done on arm1136.
>
> Is this going to happen for armv7 too? What is the real reason behind
> this proposal. What is the case when _start is not same as _TEXT_BASE(I
> looked at the archives but couldn't filter out the original discussion
> on this)

The difference is that _TEXT_BASE always contains the link-time address 
of _start, whereas references to _start may contain a different value if 
the code is executed somewhere else than at the link-time address.

/Normally/, u-boot should always execute first at the link-time address 
-- that's a base constraint.

/But/ this change makes it more resilient to out-of-link-time-address 
execution, and I want, at some time in the future, to find a way for 
u-boot to be able to start anywhere -- within reasonable limits: 
anywhere in NOR for a NOR-based U-boot, anywhere in RAM for a RAM-based 
U-boot, but I am not talking about a generic, 
run-in-RAM-or-NOR-or-anywhere, binary.

Yet. :)

> I see a problem with that. _TEXT_BASE is based on
> CONFIG_SYS_TEXT_BASE. In our SPL's case CONFIG_SYS_TEXT_BASE indicates
> the TEXT_BASE for u-boot and *CONFIG_SYS_SPL_TEXT_BASE* indicates the
> TEXT_BASE for SPL. Both are defined and useful in SPL because one is
> used for linking SPL while the other is used while loading u-boot from
> MMC. So, CONFIG_SYS_TEXT_BASE used in the start.S of SPL will not be
> correct.

The change I indicate is under the #else of a #ifdef CONFIG_NAND_SPL, so 
it will not apply to SPL. Does that still cause an issue with armv7?

> In the worst case we need to define yet another label in the linker
> scripts like __text_base. But I was wondering if we could maintain the
> status quo for armv7: that is 'adr r1, _start'

As long as you run the u-boot start code at the link-time address, there 
will be no difference except the code is more correct with respect to 
what it should do; and if you run it elsewhere, which you should not, 
you have slightly better chances that it manages to survive.

> Best regards,
> Aneesh

Amicalement,
-- 
Albert.