[rescue] iAPX-432, was Re: sparc10 cpu - what to do.

[Jonathan Patschke]

On Mon, 19 Dec 2016, Dave McGuire wrote:

>> My personal opinion is that most of what '432 really interesting
>> belongs either in a virtual machine or operating system if only because
>> they're radical enough that Intel could hardly have gotten the balance
>> right the first time.
>
>  That's what later AS/400s do.  They implement capability-based
> addressing (IBM's official stance on it is that it had been removed
> post-System/38, but most of it is still there) in a layer between
> OS/400 and the hardware.

TIMI has always been effectively a virtual machine, even back before
iSeries was PowerPC underneath.  That (and, well, obvious analogues to the
day-job at an x86 design firm) was what made me look in the documentation
from that perspective.

> The hardware is nearly-standard POWER, but it runs the "LIC" (Licensed
> Internal Code) which it boots from a hard drive, and on top of that sits
> OS/400.

Except for a few early/odd RISC machines, it is standard POWER.

A fair number of RS64/POWER[34]-era iSeries and pSeries machines differ
only in their firmware (which IBM calls LIC or microcode, depending on
which piece it is and whether the machine "natively" runs OS/400 or AIX).
Starting with POWER5, the differences were even less significant; most
POWER5 (and later) i-series machines will boot AIX LPARs without the
operator even having to ask nicely[0].  OS/400 rights were locked to the
capacity-on-demand token, which you (IIRC) could have retrofitted to a
pSeries machine.

>  As I understand it, the iAPX-432 also implements capability-based
> addressing.

Some part capability, some part ACL, depending on which perspective you
interpret the bitfield from.  It's not as amazing as legend had led me to
believe.

>  (Nobody seems to know this, but AS/400s are actually phenomenally
> well-designed, well-architected, and generally fantastic machines.)

It's not widely appreciated because IBM never targeted the machine towards
hackers.  Also, a fair bit of that magic was done in the RPG and COBOL
compilers.  Since there weren't a whole lot of ways to load arbitrary
executable code onto a 400, the "hardware platform" could extend far up
into what we'd otherwise think of as application libraries.

That makes it fantastic in all the ways you can only get with iron-clad
vendor lock-in, which most hackers are allergic to.  Users mostly only
remember the awful 5250 user-experience.  So of course "everyone" thinks
they're crap, apart from the analysts who relied upon them for decades
upon decades of line-of-business continuity.


[0] And, if you're running micropartition-style LPARs, you're already
     running AIX to virtualize I/O.
-- 
Jonathan Patschke
Austin, TX
USA