[powerstation-owners] questions about boot debugging

[Tejun Heo]

Hello,

Alexander Graf wrote:
>> I'm messing with early boot code in Linux and the kernel dies after
>> prom access goes away but before printk comes online.  early_setup()
>> finishes fine but gets stuck after rfid enables RI/DI.  The machine
>> simply hangs and I have no way of finding out where the problem
>> occurred (my last resort on x86 is combination of infinite loop and
>> reset by doublefault but dunno how to do something like that on
>> powerpc).
> 
> From my limited understanding you'd have to poke the BMC to do the
> reset. No idea how to do that though. Can't you just dump random
> characters on the serial console + infinite loops?

The serial was my second option as I didn't know which address to poke
on PPC to reach the uart as there's no separate io space.  My first
workaround was disabling address translation and then call
prom_printf() which for some strange reason didn't quite work.

>> The udbg stuff looks like the equivalent of early_printk() on x86 and
>> RTAS_PANEL/CONSOLE seems to be the right one but the machine hangs if
>> I enable it.  I ordered rj45 to db9 gender hoping that I could get
>> more information from the debug port, so my questions are...
>>
>> 1. The rj45 gender needs to be wired.  There seems to be several
>>   different rj45 to db9 mappings.  How should I wire it?
>>
>> 2. Is the debug port gonna help at all?
> 
> The debug port is a simple serial console interface. If you only need
> the machine temporarily I have one you could access with debug port
> attached.

Yeap, I know it's a serial connection to BMC and was wondering whether
the early console stuff ends up there.  I guess I'll find out tomorrow
one way or the other.  If I fail to get the serial wiring correct,
I'll ping you.

>> 3. Any ideas on how to get early debug output?  At this point, I'll be
>>   _VERY_ happy if I can just get the a double word value written in
>>   some nv storage and then retrieve it after hard reset.
> 
> Do you need to do this on a powerstation?

Not at all!

> I improved qemu recently to boot pretty far into Linux. The only
> missing part are external interrupts, but as early as you are that
> shouldn't really matter.  Ping me if you need help getting that to
> work.

Ah... that's sweet.  If I can get the part upto printk working in
qemu, I can test the rest on the real machine.  I'll give a shot at
the latest qemu and ping you if it doesn't work.

Thanks.

-- 
tejun