Radeon X1550 in an SGI O2?
In a thread on the SGI Enthusiast site, Nekochan.net, about Linux compatibility with SGI's MIPS based line of workstations, a user, tillin9, suggested the possibility of installing modern, off the shelf graphics cards in the SGI O2 under Linux. The O2 is equipped with a single PCI slot for expansion, so physically installing a new graphics card is no problem. Unfortunately, graphics cards require special initialization code, stored in the card's ROM, to be run at boot time. Since the Radeon's initialization routine has to be able to execute on the processor architecture in the computer in which it's installed, different code is required for x86, PowerPC, PA-RISC, SPARC, and MIPS. Since there weren't any PCI Radeon's made for MIPS powered workstations, surely ten years later, a modern Radeon would never work in an old SGI workstation...
Linux support for SGI's MIPS systems isn't on par with other architectures such as AMD64 or even Alpha currently, mainly due to the lack of man power and programming documentation. For instance, SGI has never released any programming documentation on their IP35 line of workstations which includes the Fuel and Tezro. On top of this, there's the problem of being unable to initialize the Radeon. Given this information, I incorrectly thought this meant there was no chance of installing a PCI Radeon in an old O2.
tillin9 seemed to mention that the inclusion of Radeon kernel modesetting would alleviate the problem of initializing the card. I was skeptical, so I asked around on the #dri-devel and #xorg-devel channels on Freenode about the status of the Radeon kernel modesetting code. Davie Airlie of Red Hat, one of the developers responsible for R500 support among many other things, responded quickly.
<airlied> mattst88: we haven't got radeon modeseting written yet..
<airlied> it might make .28, but it more likely will be Fedora first.
<airlied> I'm writing it as fast as I can.
So, the code's not done. It is obviously a high priority from his response. I've never seen a developer say anything similar to I'm writing it as fast as I can. Hopefully it'll be included in the kernel for the 2.6.28 release which should be about 6 months away. If not, it should be included in 2.6.29 a few months later.
Now, this code isn't being written in order to allow alternative architecture enthusiasts to use nice 3D graphics cards in SGI workstations. It certainly would be a great side-effect though. Again, I asked on the appropriate IRC channels. Alex Deucher of AMD, one of the main developers of X.org's xf86-video-ati driver responded quickly to my query.
<mattst88> will radeon kernel modesetting allow the usage of a x86 PCI radeon in a sparc, hppa, or mips box even without a special ROM?
<agd5f> mattst88: it should
What an excellent unintended benefit of Radeon kernel modesetting. I mailed the xorg mailing list just to be sure. From this, Alex Deucher responded to me again and clarified. In short:
- If the Radeon has AtomBIOS available, the card will be initialized by AtomBIOS (>= R400)
- If the Radeon doesn't have AtomBIOS, the card will be initialized by parsing initialization tables in the card's BIOS (<=R300)
- If the card isn't a Radeon, either an x86 emulator will be required to run the card's initialization routine, the driver will need to bang the registers to initialize it, or the card will need to provide a system similar to AtomBIOS
This feature, whose main purpose is to remove an unnecessary obstacle and provide a flicker-free boot up, will also allow many users of SPARC, PA-RISC, and MIPS computers to enjoy the features provided by a modern Radeon graphics card.
To tillin9: I'm sorry for doubting you. Thanks a lot for bringing the possibility of a Radeon X1550 equipped SGI O2 workstation to my attention.