Gilles Chanteperdrix wrote:
> Jan Kiszka wrote:
>> Stefan Kisdaroczi wrote:
>>> Hi Jan,
>>>
>>> Only user space using the native skin. Some additional notes:
>>> I'm using the freepascal compiler and link against libnative, i dont use gcc.
>>> There is one process starting the other ten processes using fork() & execl().
>>> The tasks do a lot of mode switching between primary/secondary.
>>> The stacksize for the tasks is 1MB. We only use between 4K-16K of stack in our
>>> applications. So there should be still a bit less than 1 MB for socket calls
>>> and printf, is this to low ?
>> Also stack overflows must not crash the kernel, they may only cause
>> application crashes.
>>
>>> I rebuilt the kernel without arcnet and isdn and disabled smp too, using 2.5.1.
>>> It still crashes every time on termination. If I pull the network connector,
>>> I could start and shutdown the applications 5 times without problem. If the
>>> network is connected, as soon as there is some network activity like ssh/samba
>>> on termination the box crashes. Sometimes it gets even back to the prompt before
>>> crashing. Its a rtl8139 network card, but happens in a virtualbox-vm with a pcnet
>>> card too. In the traces is always a "do_IRQ+0x...", so I think the problem is
>>> related to IRQ-handling.
>> That sounds promising! If you can provide your test case to us (or me
>> privately), we could try to reproduce - inside a proper hypervisor which
>> has proper debugging facilities. :)
> 
> I do not know if this could be related, but you have two drivers we 8139
> in their name. Could there be a conflict?
> 

I would say only the 8139cp gets hold of all the hardware - unless there
cards plugged that only the 8139too supported. Otherwise, that 8139too
should simply remain unused.

Jan