Mark Green wrote:
>
> > "Zane H. Healy" wrote:
> > >
> > > >Um. I thought, given the context, that this person was talking about the
> > > >Sony Play Station 2, which is also frequently refered to as the PS/2.
> > >
> > > Exactly! PS2 .eq. Playstation 2. OTOH the PS/2 is a failed computer
> > > platfrom from IBM. Even the Playstation two box says PS2.
> >
> > Ah, well if we're talking about the PS2, I can say with some authority
> > that at least LInux ran on it before the TX5900 core ever hit fab. It was
> > part of the verification suite we ran on the behavioral and gate level
> > simulators as well as on the god-awful Quickturn...
> >
>
> The PS2 development system runs Linux, but the PS2 itself doesn't,
> at least at the present time. By the way, some of the development
> software was produced by Red Hat.
Let me try that again. Linux was ported by one of my teams to the
verification environment for what was then known as the R5900. Included
as part of this environment were high fidelity software implementation
of most of the PS2 peripheral devices; in the case of Quickturn this
was a hybrid solution with physical peripherals implemented by a
PC and the mapping between the R5900 and the PC performed by a
slug of software and a couple of gate arrays; essentially the same
hardware that Toshiba used for development of its first DVD
decoder.
As for Red Hat producing some of the development software, it's only
by act of purchase. Since the R5900 is a modified MIPS architecture
the decision was made to use gnu as the basis for the toolchain.
Cygnus had the contract to produce the production toolchain for
SCE, but they were so incredibly slow that we ended up building
our own to support the verification effort. The statement that
Red Hat had anything to do with it is only valid in the context
that Red Hat eventually purchased Cygnus.
> There are two problems with porting Linux to the PS2:
>
> 1) Actually getting Linux on the PS2. Currently the only way
> to get software on the PS2 is by DVD, and Sony uses its own
> format. Apparently the format is rather difficult to break, but
> I wouldn't be surprised if someone does crack it in the future.
Okay, so I'll admit that some things are easier when you have
the keys to the kingdom ;-)
> 2) The real killer is the highly non-standard peripherals. The
> graphics hardware is totally non-standard. It would be quite
> difficult to produce graphics drivers for it, and and X server
> would be a real pain. I'm not sure X would perform well on this
> configuration.
GPUII is decidedly an -- architecturally interesting -- bit of work.
Two groups of FPU units, its own assembler and weird DMA porting
between it and the processor (of course the processor itself can
be a DMA _target_ -- there's 1K of on-board scratch=pad ram
that's referenced by creating a PTE with what's normally a reserved
bit in the MIPS ISA PTE spec set -- that can be loaded via DMA
while the processor's pipe contnue execution (and for that matter,
while the BIU continues to fetch instructions and operands).
While it wouldn't materially affect a _port_, getting some apps to
run would be interesting as well. The TX5900 uses the same floating
point format as the GPU, which is best described as single-precision
IEEE without truncate or rounding modes.
> I think the Microsoft Xbox would be a better choice for Linux, at
> least it has a standard graphics chip.
I'd expect that to be the case. The PS2 was hardly designed as
a general purpose machine; there's all sorts of architectural brain
damage that makes sense only in the context of gaming.
Still, it's interesting how many problems we found running apps
on top of Linux on top of the simulations and emulations vs. what
was found with the purpose-build verification suite and RTG tools.
--
Chris Kennedy
chris_at_mainecoon.com
http://www.mainecoon.com
PGP fingerprint: 4E99 10B6 7253 B048 6685 6CBC 55E1 20A3 108D AB97
Received on Sun Jan 14 2001 - 10:06:18 GMT