[U-Boot] [RFC PATCH 1/5] armv7: move boot hook before '_start'

[Dr. Philipp Tomsich]

> On 13 Jun 2017, at 11:47, Marek Vasut <marex at denx.de> wrote:
> 
> On 06/13/2017 11:44 AM, Dr. Philipp Tomsich wrote:
>> Marek,
>> 
>>> On 13 Jun 2017, at 10:42, Marek Vasut <marex at denx.de
>>> <mailto:marex at denx.de>> wrote:
>>> 
>>> On 06/13/2017 03:51 AM, Kever Yang wrote:
>>>> Hi Marek,
>>> 
>>> Hi,
>>> 
>>>> On 06/07/2017 02:28 PM, Marek Vasut wrote:
>>>>> On 06/07/2017 04:28 AM, Kever Yang wrote:
>>>>>> Hi Andre, Steve, Marek,
>>>>>> 
>>>>>>    Could you help to check how to make it work with this patch on
>>>>>> sunxi, bcm and socfpga platform?
>>>>> The socfpga expects the hook at that exact position (0x40 I think) , so
>>>>> if you moved it somewhere, you broke socfpga.
>>>> 
>>>> I know this break socfpga, and that's why I cc you for help, could you
>>>> help the take a look if we can have a solution on socfpga to "make it
>>>> work with this patch"?
>>> 
>>> Can you generate the same u-boot binary with this patch ? If so, then it
>>> will work on socfpga. SoCFPGA expects that small piece of stuff at
>>> offset 0x40 , so if you move this boot0 hook, it will break. HTH
>> 
>> So the SoCFPGA-implementation relies on the boot0 hook to always start
>> at 0x40?  Then this needs fixing anyway… or it will break once someone
>> else touches the armv7 startup code.
> 
> I think it's not just the SoCFPGA.

What I am saying is that the boot0-hook is intended to add a header
before the actual U-Boot code (as Andre had originally stated in his
commit message and as various comments around the boot0 hook
also suggest): 

    arm/arm64: implement a boot header capability
    
    Some SPL loaders (like Allwinner's boot0, and Broadcom's boot0)
    require a header before the actual U-Boot binary to both check its
    validity and to find other data to load. Sometimes this header may
    only be a few bytes of information, and sometimes this might simply
    be space that needs to be reserved for a post-processing tool.

However, the actual start of the boot0-hook in the U-Boot today is right
after the vector-table (which is 0x40 long), which has a few interesting
consequences for various architectures:
- sunxi: doesn’t matter
- socfpga: boot0 hook relies on something being in front of it, as it uses
	a ".balign 64, …” to advance to 0x40
- rockchip: comes too late to allow inserting the word-sized field at the
	very start (and to move the vectors back by a word).

Now having looked at the implementation in
	arch/arm/mach-socfpga/include/mach/boot0.h
it is clear why Kever’s change breaks this code: .balign will not forward
to 0x40 (after all, 0x0 is also 64-byte aligned).

Once the boot0 hook is at the beginning of the binary (as always intended),
this will need to become a .space or .fill directive.

>> That said, I read your comment to mean that the following changes would
>> work for SoCFPGA:
>> 1.arch/arm/lib/vectors.S:
>> #if CONFIG_IS_ENABLED(BOOT0_HOOK)
>> … pull in the boot0 hook, which needs to contain the vectors
>> #else
>> b reset
>> … remaining vectors … 
>> #endif
>> 2.boot0.h (for socfpga):
>> b reset
>> … remaining vectors …
>> /* now at 0x40 */
>> … whatever currently lives in the boot0.h for socfpga ...
>> 
>> Ok?
> 
> I'm not sure I understand what you're trying to say. IIRC the SoCFPGA
> bootrom checks the content at 0x40-0x48 in SRAM and then jumps to 0x50
> if the content is valid. Otherwise it reloads the content from
> SD/SPI/NAND…

Thanks for the background info. With this, it should be possible to create
a new version of the patch that also correctly handles socfpga.

Regards,
Philipp.