Re: Weird GNUMach crash

[Marcus Brinkmann]

At Thu, 19 Feb 2004 19:24:22 +0100,
marco_g wrote:
> 
> Marcus Brinkmann <marcus.brinkmann@ruhr-uni-bochum.de> writes:
> 
> > At Mon, 16 Feb 2004 18:06:59 +0100 (MET),
> > ams wrote:
> >> 
> >>    panic: zalloc: zone kalloc.8192 exhausted
> >
> >> Silly question maybe, but, how much RAM and swap do you have?  zalloc
> >> messages tend to pop up when you have to little swap or RAM.
> >
> > Could also be a resource leak.  For example a port leak.  I have a
> > vague recollection that the 8192 zone is used for port data, but I
> > could remember it wrong.  If it is happening under low load,
> > reproducibly, the first thing I would do is look for a port leak.
> 
> Yes, I can easily reproduce it under low load.  How can I find out if
> the 8192 zone is used for port data?  What would leak the ports?  The
> crash server?

You could check which zones the IPC code in GNU Mach allocates for
port related data structures.  Somewhere in gnumach/ipc/.  If you
don't find it, get back to me, and I will have a look.

Of course, the origin of a port leak is as easy or hard to find as the
origin of a memory leak.  Use portinfo to spot the task which a
suspiciously high number of ports.  If you can trigger the leak
without crashing (ie, just leak a couple of dozen ports, then stop
whatever you are doing that leaks), you can investigate the ports more
with gdb and portinfo (what type of ports are they, etc).

Thanks,
Marcus