[OpenAFS-devel] 1.3.79 on AIX 5.2, system dump when using token

[Niklas Edmundsson]

On Fri, 25 Feb 2005, Horst Birthelmer wrote:

> That's an old more or less known problem.
>
> The problem is the kernel allocation. The xmalloc call ... you can see that 
> in the backtrace above.
> I know that there's a problem but never had either the time to fix it nor any 
> clue how to fix that since it looks pretty correct to me when a pioctl 
> allocates some kernel memory.

I posted a patch a week ago that increases the stack which makes it 
work a bit better. Does that give any hint on what's wrong? I have 
recieved no comments at all on the issue...

Michael: You might want to try my patch and see how much things 
improve for you. Drop me a mail if you haven't recieved it on the 
list.

Would it help to enable memory overlay detection or some other debug 
feature found in bosdebug and reproduce the crash? Although I have 
compiled quite a few programs on AIX I'm rather new both to the 
OpenAFS codebase and to debugging AIX kernel space stuff, so any hints 
would be welcome.

Given that the thing works better with larger stack I'd assume that 
some assumption that's valid on AIX4 isn't valid now which causes the 
thing to overwrite memory in kernel space and things goes downhill 
from there.

I'm beginning to get rather keen on getting this issue fixed since the 
1.4 release is getting closer...

/Nikke
-- 
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
  Niklas Edmundsson, Admin @ {acc,hpc2n}.umu.se     |    nikke@hpc2n.umu.se
---------------------------------------------------------------------------
  A waist is a terrible thing to mind.
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=