> > If anyone would care to send me the docs I'd need (eg, the MMU
> > interface, which I know little about), I could probably undertake to
> > write a simulator.  I've already done one for the PDP-11.  (It crawls
> > "like a high spiral snail", to quote I think the sendmail docs, but it
> > does work.)
> 
> Gordon Irlam wrote a simulator called `spa' intended for performance
> analysis on sparcs. Maybe it's possible to expand on that.

Yes!  Todd Clayton and I did this.  We extended Spa so that it
simulates the privileged v8 instructions, added support for sun4d
hardware, added code to write instruction traces to disk, and added a
remote gdb interface.  The new program is called Sky.  "Simulator"
might actually be the wrong name, the software might be closer to a
virtual machine in the IBM VM conception.  We have put the
documentation online at

    http://www.base.com/gordoni/sky.html

*** Alas, the source and binaries are not publically available. ***

Spa is copylefted, so before working on Sky we got informal permission
from our manager and director to share our final results.  When coding
we were also careful not to include any Sun code.  We took the obp.h
from NetBSD and an sprintf.c Linux rather than any Sun internal code.
In addition we planned to share the instruction traces we gathered
with the academic community.

Once we had something working we attempted to get formal permission to
distribute it.  We discovered Sun didn't have a policy for giving
stuff away.  The legal department decided they needed to formulate a
policy and process.  They have now done this.  The details are
complex.  We have also now got different directors which makes things
even harder.

Larry McVoy's golden rule: do good by those that do good by you.  My
director was a very nice guy and was trying to do good by us by
letting us give Sky away.  So when it turns out this is more complex
than anticipated we have had to accept this and not simply give Sky
away regardless.  We don't want him to get into any sort of trouble
for attempting to be nice to us.

Internally Sun has been using Sky to gather multi-billion instruction
traces on benchmarks such as TPC.  Uncompressed some of the traces
weigh in at around 100Gb.  Gack!

I would still like to get permission to distribute Sky and some of the
instruction traces, however it is difficult for me because I went on
leave from Sun roughly 12 months ago to devote my time to hacking on
the Web.  It is possible I still might be able to free it up, but
please don't hold your breath.  Perhaps Todd or Brad will be able to
help me.


                                      very sorry,
                                                  gordoni


[For Brad and anyone else within Sun engineering you can grab a copy
of Sky as

    /home/gordoni/work/sky/sky-1.0.tar.gz

Contact me if you plan to run it on Solaris 2.5.  A small patch needs
to be made because the 2.5 ufsboot assumes ET is set to 1 whereas we
initially have it set to 0.]