[U-Boot] AM335x not working anymore

[Jeroen Hofstee]

Hi,

On 16-09-14 16:11, Jeroen Hofstee wrote:
> Hello Hannes, Tom,
>
> On di, 2014-09-16 at 09:56 -0400, Tom Rini wrote:
>> On Tue, Sep 16, 2014 at 02:17:01PM +0200, Hannes Petermaier wrote:
>>
>>> Hi,
>>>
>>> i've just pulled newest u-boot and tried it on my custom AM335x
>>> board without success. It hangs after (successful) running init_s
>>> function.
>>>
>>> ....
>>> Enable clock module - 44e0001c
>>> Enable clock module - 44e000d0
>>> Enable clock module - 44e00028
>>>
>>> is the last debug output.
>>> After some time the watchdog is doing his job and restarts again.
>>>
>>> Okay ... thinking maybe something had become incompatible with my
>>> customization ... so let's try the beaglebone board.
>>> Same behaviour :-(
>>>
>>> I debugged around and found finally this commit, which breaks the support:
>>> /
>>> //commit id: f0c3a6c4ad09210d5d4aeafe87685ee75e5683d6//
>>> //
>>> //    ARM: SPL: do not set gd again//
>>> ////
>>> //    Just before calling board_init_f, crt0.S has already//
>>> //    reserved space for the initial gd on the stack. There//
>>> //    should be no need to allocate it again.//
>>> ////
>>> //    cc: Albert ARIBAUD <albert.u.boot at aribaud.net>//
>>> //    Signed-off-by: Jeroen Hofstee <jeroen at myspectrum.nl>/
>>>
>>> I have reverted this commit in my branch and everything works fine.
>>> Finally i have to admit, that i do not fully understand whats going
>>> on there in detail.
>>>
>>> What i can see, is that there live more than one instance of the
>>> "gd". One is setup in arch/arm/lib/cpu/armv7/lowlevel_init.S,
>>> another one is setup in arch/arm/lib/crt0.S.
>>> Within the "gd-ptr" is loaded to &gdata (gdata __attribute__
>>> ((section(".data")))
>>>
>>> I am a bit confused all about this :-)
>> I just finished bisecting down to this being a problem commit as well.
>> As far as I can see, the commit in question doesn't make sense.  Or
>> perhaps rather, the comments in crt0.S aren't quite right in the SPL
>> case where we don't place the initial gd on the stack, it's in the .data
>> section.
>>
> As far as I can tell there are two gdata's is the SPL case.
> One on the stack and one in gdata. Can't the later not be
> removed, so we end up with a single gd init in crt0.S?
> Before this patch the code switches back and forth between
> the two, which is at least a bit weird.
>
> I will have a closer look at the call path, especially why
> it used to work...
>

While in my opinion the omap code is to blame for doing
more then needed in cpu_init_crit and eventually in s_init,
(and hence needing its own gdata) I am more then fine
with reverting this patch for now, but I would like to
have only a single gdata eventually, initialized only once.

Can you try this patch [1], just for educational purposes. It
will scare Tom a bit since dram is setup before the console,
but I still have to find an user of that.

Regards,
Jeroen

[1] http://patchwork.ozlabs.org/patch/354535/