Subject: Re: Netbooting with OF 1.0.5 again...
To: None <mw@blobulent.com, mw@costello.cnf.cornell.edu>
From: None <Havard.Eidnes@runit.sintef.no>
List: port-macppc
Date: 10/21/2000 19:29:44
> Did you see my post a few days ago about a G3 upgrade in my
> 7500?

Admittedly, not until now.  It does sound familiar, though.

> Depending on the brand of G3 upgrade, it might or might not
> turn of speculative processing in OF (some early versions of G3
> upgrades required MacOS to load an extension before it was
> turned off).  Try downgrading to a 604 and see if you still
> can't boot.

OK, that's an option; I'll try downgrading, though it's more than
a bit disappointing having to do that.  The CPU upgrade I have is
a Sonnet 350MHz board, BTW.

> Yes, this is almost identical behavior to what I've seen with
> my G3 upgraded 7500.  The one thing I didn't try was (again)
> downgrading to a 604.  Right now, I don't have access to a
> netboot server so I can't try this.

OK, I'll retry this too once I have downgraded the CPU.

However, if someone can supply me with an ofwboot.xcf of recent
lineage (1.5 branch?) compiled with NETIF_DEBUG and DEBUG, that'd
be really great.  I'm interested in finding out where exactly the
boot loader loses during netboot, and I'm currently trying to
read the boot code, apparently without being able to understand
where it's getting those "TFTP timeout"s from, and I also haven't
found where the "file: " printout is coming from.

> As I understand it, speculative processing means that the G3
> executes instructions before it is certain that they will be
> necessary.  This becomes a problem with older devices (such as
> SCSI controllers, video, and network) when their state changes
> when information is read from them.  Perhaps mc0 is getting
> corrupted from this?

I thought that speculative execution as a technique to speed up
processors was mostly something which was reserved for CPU-
internal implementation, and that it would be Exceedingly Bad if
any results of that would end up being externally visible, ref.
the bugs reportedly present in (early revisions of?) MIPS R10000
CPUs (where the effects of speculative memory operations can end
up becoming externally visible).  Well, maybe I'm naive to think
that...

Regards,

- H=E5vard