[Help-smalltalk] Memory management

[Paolo Bonzini]

I did try not using MAP_FIXED.  But the problem is that I cannot map 256
MB at startup, that would be quite wasteful...  My usage of MAP_FIXED
(together with the mmap.m4 test) tries to find a place where I can assume
that no mmaps will be done.  Anyway the bug happened before I switched to
mmap.

Under Windows I can ask the OS to reserve 256 MB and leave them unmapped
via VirtualAlloc and VirtualFree (see `#ifdef _WIN32' code).  An idea
could be to mmap 256 MBs to a dummy file, then when I need memory I unmap
them and map them to /dev/zero.  I am told that under Linux mmap-ed files
don't consume swap space; I never wrote code to test this claim and cannot
test other OSes -- if I dare use 256 MBs of memory on the university's
HP-UX workstations, the root won't be happy...

Try writing a small stupid program like

  main(argc)
      unsigned int argc;
  { return argc*argc; }

and tracing through it with gdb... maybe that helps figuring out if it is
normal that .umul or _PROCEDURE_LINKAGE_TABLE_ end up executing that
address.

|_  _  _  ___
|_)(_)| )  ,'
--------- '-._.