Re: [cobalt-developers] Need to build a RaQXTR/550 kernel

[josh <josh@xxxxxxxxxxxxxxxxxx>]

On Fri, Dec 19, 2003 at 12:25:08PM -0800, Tim Hockin wrote:
> On Tue, Dec 16, 2003 at 10:20:24AM -0500, josh wrote:
> > If I read the FAQ right, the 550 has only two memory slots while the
> > XTR has four (all filled in my case). The 550 has space for two disks
> > and the XTR has four (all filled in my case). I'm running romrev
> > 2.9.34. How does the rom get the information on memory and disks? It
> 
> The ROM can detect which board you are running, and can probe for memory and
> disks based on that.
> 
> > is not setable in my version and the rom reports only 1/4 the memory
> > that is really there. (based on the results of memtest which reads the
> 
> on an XTR?  Are you sure the memory is Registered ECC memory?  XTR is *very*
> picky about memory.

Yes, I am very sure. Can I be sure that it is good registered ECC
memory. There the answer is, no I can't, because the version of
memtest86 that is downloadable from Sun, does not actually verify ECC
and also gets the amount of memory from the ROM. If I could run the
latest version or even figure out how to compile and run the current
tester so that I could override its reading the amount of memory from
ROM then I would. I did try to override the settings without luck, it
still checks the ROM. ROM values override the command line.

Tonight, I am rolling back to using a RaQ3i/550 for
production tonight - so starting next week I can just test the hell of
my RaQXTR/550. My first task will be verifying that the ROM is the ROM
I think it is. My memory is that after I had problems with the first
OSRCD I used. I spoke with Gerald, who did say somethings like, "I've
never seen that before" and "that's not supposed to happen"; Perhaps,
I should have regarded that alone as an indication I might have
problems. And of course I can pull out the two sticks I added (or even
three) and let the memtest run on each 512 stick by itself.

The original memory and the two sticks I have do differ in size and
one digit on their number -- I assume this was simply because memory
keeps be engineered slightly differently.

> 
> > info from the ROM using the standard linux methods). The cmos commands
> > seem limited -- are there multiple versions of the cmos program for
> > the different romrevs?
> 
> yes and no. The cmos/nvram tools get revised wioth rom revisions, but are
> backwards compatible.


-- 
Josh Kuperman                       
josh@xxxxxxxxxxxxxxxxxx