[Milkymist-devel] Single-Thread AEMB

[Shawn Tan]

Hi Guys,

I'm just guessing here, but that may have to do with the fact that both 
hardware threads actually share the same software stack. This is manually 
split in the 'hook.hh' file. 

However, there is no stack overflow checking. So, it is possible to encounter 
this problem. For example, the stack size may be defined as 2K and GCC thinks 
it has 2K but the threads only have 1K each. 

Assuming that this is the problem, the simplest solution for this is to 
override the 'hook.hh' routines so that they do not split the stack equally. 
It is also possible to increase the stack size if there is enough RAM.

However, I'm just guessing here. It may be something totally different, may 
even be a bug in the processor! d=

On Monday 01 June 2009 02:05:49 pm you wrote:
> Both the uClinux kernel and a simple test program doing context
> save/restore on the stack crash the CPU without more information. I did not
> spend much time doing this in simulation, which I still plan to in order to
> narrow down the problem.
>
> Will try to come up with the standalone program that did trigger this bug.

-- 
with metta,
Shawn Tan.