On Mon, 28 May 2001, Lord Isildur wrote:

> I thought that the contiguous memory was still an issue in the DS5k2xx
> whereby the system examines what's in slot 0, and then decides that all
> slots are of that size in terms of physical space. you can't mix 8's and
> 32's except in one case: all 32's except the last, which you can put an
> 8 in, and the system will see it as a 32 with 24 megs of failed memory. 
> otherwise, the system expects that your 8s are worth 32, and will barf 
> most royally when it tries to access some of the memory that's not 
> there.. 

 The firmware in the PROM memory checks what's in the first slot and sets
up the memory controller for either a 32MB or an 8MB stride.  It then
probes all slots for memory and stores results in a bitmap of available
512-byte chunks: a "1" denotes memory is present and a "0" means it's
absent; 24MB holes are marked appropriately.  The bitmap can be queried by
an OS.  Of course an OS can probe for memory itself -- an exception is
generated on a non-existent memory access -- this is actually necessary
for pre-REX firmware (e.g. on a pmax/pmin).

 What you wrote reminds me an Ultrix FAQ entry -- Ultrix does not seem to 
support mixed memory configurations well, indeed.  It just doesn't see any
memory above the first hole.

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +