Hmm just to drop a quick followup on this topic:

As Goran mentioned, I experienced the problem on OpenBSD (3.5) on x86
[not sure if we've ascertained whether the problem yields itself on
other architectures, either under Linux 2.6 or the *BSDs]. We _did_ seem
to conclude that the problem wasn't experienced on Win32's VM, though I
have no idea how it handles it's allocation --- so this might not be
that pertinent.

Therefore, I'd just like to ask anyone out there running a NetBSD,
Linux, OpenBSD or other to check out the Unix VM running without the
additional -memory option on non-x86 architecture; could prove to be an
interesting test. Perhaps not - just a thought.

But the important point I'm trying to make is this isn't just Linux
specific.

Regards,
Marc [zpg]

PS - this is my first post, in case you were wondering who the hell I am
;-) Just signed up on Squeak People yesterday as "mds", and I'm on
freenode as "zpg". Cheers guys.

On Fri, Jul 02, 2004 at 09:20:43AM -0700, John M McIntosh wrote:
> Well if linux 2.6 loaded the image over the 2  GB boundary you would  
> crash right away when you swizzle the oops at start up time.
> What you should do is put a break point in the memory allocation logic  
> that is responsible for growing the image and see when
> the pointer goes negative as it's being adjusted after a grow/shrink.
> 
> You could also check what the start of memory location is, since I  
> recall the unix VM wants to allocate 1GB, so does start of memory plus
> 1GB > 2GB?
> 
> On Jul 2, 2004, at 8:13 AM, goran.krampe at bluefish.se wrote:
> 
> >Hi all!
> >
> >Well, now I have come closer to the answer. I patched my interp.c so
> >that when it normally failed - after oopHasOKclass(anOop) returns 0, I
> >called oopHasOKclass() once more, so that I could step through it line
> >by line.
> >
> >It appears to me in insight/gdb that foo->endOfMemory is negative (sign
> >bit gotten set?) so the interval check that looks like this:
> >
> >        if (!((0 < oop) && (oop < foo->endOfMemory))) {
> >                return 0;
> >        }
> >
> >...fails.
> >
> >Hmmm, ok - so has this worked until now just because of luck? Has  
> >memory
> >normally been mapped below 32 bits, and now with kernel 2.6 (for
> >example) Linux decides to place Squeak above it sometimes? And is the
> >problem simply that we can't treat endOfMemory as a signed int?! And  
> >did
> >Greece really go to the Euro championship soccer final yesterday? It
> >seems so.
> >
> >Here are the shocking evidence (in the form of png images) from my
> >insight expedition into VM low level country, with assembler and all:
> >
> >	http://anakin.bluefish.se/gohu/uploads/4/bughunt.zip
> >
> >Over and out from VM-expeditioner-newbie Göran Krampe, who is now
> >jumping on a bus going home to eat a nice dinner and drink some wine
> >with his wonderful wife hoping that someone else can bring order into
> >this situation. :)
> >
> >regards, Göran
> >
> >
> >
> --
> ======================================================================== 
> ===
> John M. McIntosh <johnmci at smalltalkconsulting.com> 1-800-477-2659
> Corporate Smalltalk Consulting Ltd.  http://www.smalltalkconsulting.com
> ======================================================================== 
> ===
> 
> 

--