[okl4-developer] Debugging tips for a new platform

[Joshua Root]

Geoffrey Lee wrote:
> On Thu, Jun 19, 2008 at 07:33:39PM +1000, Joshua Root wrote:
> 
> /* snip */
> 
>>> I got a fresh 2.1 tarball and skyeye 1.2.5 off skyeye.org and did
>>> ./configure; make, built OKL4 with gumstix and ran that and worked fine.
>>> To take a step back, can you at least confirm that it works on skyeye
>>> 1.2.5 for gumstix?
>> No, that's what I'm saying. I can't get a gumstix image to work in  
>> anything other than the 1.2.1n build linked from the wiki. Same result  
>> with 2.1 and 2.1.1-fix.7.
> 
> 
> Ah, I noticed that you are using image.sim.  Please re-try with
> image.boot.

Gumstix doesn't seem to build an image.boot.

>>> There are several issues that I have identified with your patch.
>>> First, your toolchain setup appears to be incorrect.  Please add
>>> the following to the ixp class:
>>>
>>>     c_flags = arm.c_flags + ["-mbig-endian"]
>>>     as_flags = ["-mbig-endian"]
>>>     link_flags = ["-EB"]
>>>
>>> As well, the default toolchain is gnu_arm_toolchain, which uses the
>>> libgcc included in libgcc.  If you are using the pre-built toolchain,
>>> it uses the gcc provided libgcc.  Since that is little endian it will
>>> not link.  You can use the OKL4 provided libgcc instead.
>> The ixp4xx platform is set up to use a toolchain that targets big-endian  
>> by default:
>> gnu_armv5b_toolchain = generic_gcc("armv5b-softfloat-linux-")
>>
> 
> 
> I don't think your toolchain is getting called, as the resulting
> binary is little endian, as printed out by the skyeye output.  You
> can see whether your toolchain is getting called by specifying
> verbose_str=True on the build command line.

The little-endian binary is for gumstix. I'm trying it first because I 
know it's a working image. If I can't get gumstix to work in the 
simulator, nslu2 has no chance.

Building with machine=nslu2 definitely uses the correct toolchain and 
produces a big-endian binary.

>>> Finally, the trace log will be of use in determining where you are
>>> stalling.  After those fixes I tried to boot it on Skyeye 1.2.5 and
>>> found that it was spinning on trying to write out to the serial.
>> Thanks! That's very helpful to know.
> 
> Just to ensure that we are on the same page, can you confirm that booting 
> image.boot when compiled in MSB mode will yield a binary that spins
> on trying to write out via serial?

I haven't yet been able to boot a big-endian image in the simulator.

- Josh