> 
> I can only suspect the ptrace code, since the program works fine outside
> 

There appears to be a problem when the debuggee's pages get paged out to
the swap device. The symptoms I noticed were that breakpoints set by the
debugger were suddenly gone by the time the program woke up to run again.

It is probably another VM bug. As far as I remember the old "context
switching" ptrace implementation did not show this.

-pk

------------------------------------------------------------------------------