Re: Phase 5 Statement on Linux for PowerPC

Subject: Re: Phase 5 Statement on Linux for PowerPC
From: Matt Porter (
Date: Thu Aug 19 1999 - 08:59:23 MDT

On Thu, 19 Aug 1999, Adrian Cox wrote:

> Jeramy B Smith wrote:
> > I emailed Phase 5 ( about possibly making CHRP motherboards
> > for the purpose of running Linux for PowePC and recieved a pleasant
> > response. Phase 5 is a well known maker of PPC accelerators for Macs and
> > Amigas.
> What the IBM reference design doesn't address is actually the hardest
> part: building a complete workstation with a modern graphics system.
> There are three problems:
> 1) If you use plug-in graphics cards, you get a lot more choice with
> AGP. The IBM chipset doesn't have AGP.
> 2) Most graphics boards require you to execute x86 BIOS code. This makes
> your boot firmware even more complicated, as it probably has to contain
> Gabriel Paubert's x86 emulator. I don't think IBM even supply any
> reference firmware, certainly not any that can execute x86 code.
> 2a) Or you need a deal with a graphics chip vendor under NDA. If you are
> not Apple, you are probably too small to keep their attention.
> 3) If you integrate a PCI graphics chipset, you're probably stuck with a
> low performance C&T or similar chipset, which will not give you
> 1600x1200/32-bit/OpenGl/etc.
> 3a) 2a applies to 3a.
> Most of the VME/CPCI PowerPC systems use serial consoles for this
> reason. Even MCG have stuck with supporting one or two low-performance
> graphics chip.

MCG chose Cirrus Logic parts because:

1) They are so easy to interface to PCI and PowerPC
2) They haven't historically cared about graphics and performance. (it is
   embedded hardware afterall). Serial console is far more convenient for
   racks of equipment, that's why it's in demand.

At this point MCG's PPCBUG only has support for obsolete graphics chips.
I've been pushing for a year now to get our firmware group to put a
Gabriel-like x86 real mode emulator in the firmware so we don't have to
have vga-core enabling code for every chipset out there.

Of course, that is a deadend considering that vgacore will go away and so
will real-mode BIOSes. What's worse, fewer and fewer chipsets are going
to be available as separate parts. Soon most solutions will be core IP
logic dropped into the host bridge LSI.

> If I'm too pessimistic, tell me the graphics chip vendor who's actually
> useful.

I'm pessimistic too, but I do know that MCG has a solution since they make
their own host bridges in house. Adding AGP to the Raven or Hawk parts
will be easy when it is necessary.

Matt Porter
This is Linux Country. On a quiet night, you can hear Windows reboot.

This archive was generated by hypermail 2a24 : Sun Sep 05 1999 - 13:46:32 MDT