Anyone Get LINUX Working On An Octane?

[barutanseijin]

I use on my Octanes. I just yank the linux disk and move it around. I have a bit of a problem in that i was netbooting off my old ibook, but i don't have that machine anyone near the Octanes at the moment.

[mattst88]

As far as the kernel mode setting, I haven't gotten that working on an SGI. Actually, the only really working support (as in testable with a released distro) right now is Intel's integrated graphics. If Intel actually releases Larabee, that should work too. Open source radeon support is going to be included in kernel 2.6.27 or .28 (I know the devs. that are working on the ATI side - and have told them about my plans to put a card in an O2) but the current stable linux release is 2.6.25 so we're a few months off. When I get it working, I'll let everyone here know.

You're a little overconfident, to say the least. There are no definitive plans for Radeon kernel modesetting. It'll happen, yes. But currently, there is no timeline. Jerome Glisse is the person to talk to about kernel modesetting for radeon. I suggest you talk with him, as what you say about .27 and .28 is simply not the case as far as I can see. His modesetting code is available in his personal git repository (http://gitweb.freedesktop.org/?p=users/ ... ;a=summary) but it hasn't been touched since December 2007.

Also, kernel modesetting is useless if the card cannot be initialized (POST'd) first. It's not as easy as finding a PCI Radeon and throwing it in a MIPS box, or an HPPA box, or SPARC, or PPC. They all have different ROMs with the code for the appropriate architecture's initialization. This is why ATI produces FireGL cards for HP's PA-RISC workstations and Radeons for PPC Macs. They contain code specific to the platform for which they're intended. The HP C8000 can be equipped with FireGL cards. Unfortunately though, there is still no DRI support for them under Linux/HPPA. I'm not sure X11 works on the C8000 under Linux in any case, actually.

Now, assuming you could get an X1550 (or even a Radeon 9100) with the appropriate MIPS initialization ROM, which, mind you, wouldn't be possible since there were never any Radeons made for MIPS boxes, Linux isn't guaranteed to work. For instance, I've got an Alphaserver DS20L. The good engineers at Digital were kind enough to implement an x86 BIOS emulator into SRM so that x86 add-on cards could easily work in Alpha systems even without the appropriate ROMs. The card works with xf86-video-radeonhd and crashes with xf86-video-ati. It's quite possible that whatever causes xf86-video-ati to crash has been fixed in more recent versions. Unfortunately, more recent versions of X11 require libpciaccess which in turn requires the kernel to allow access to PCI resources through sysfs. This feature has never been needed on Alpha, and is therefore not implemented. (Kernel Bug Report)

All in all, it's not as easy to get a graphics card to work with strange hardware as it is to plug on into the strange hardware. I'm also not sure you've got any of your facts straight.

The only way I can see to use an x86 graphics card in a MIPS box is to find one that doesn't have an initialization routine. Matrox and Voodoo 3s come to mind.

[skywriter]

i put creamed corn on my Octane, and.... it didn't work.

[mattst88]

As a follow up, David Airlie said just a minute ago on #xorg-devel on Freenode

<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.

[bplaa.yai]

Now, assuming you could get an X1550 (or even a Radeon 9100) with the appropriate MIPS initialization ROM, which, mind you, wouldn't be possible since there were never any Radeons made for MIPS boxes

I think the correct statement should be "there were never any PCI Radeons made for MIPS boxes" as the Onyx4 G2 brick was basically made of two AGP FireGL X1/X2
But for sure, this does not change anything to our modesetting matter on O2 / Octane

[mattst88]

I asked around on #dri-devel. Good news! I was wrong. Well, that's not really the good news. The good news is that what you said appears to be right!

<mattst88|laptop> 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|laptop: it should

So maybe there is hope for us Linux/AltArch users of enjoying nice, fast, hardware accelerated X11 on our O2s and PA-RISC workstations alike.

[D-EJ915]

I can't wait for it to use with my HPPA boxes, my FX10s are bitchin but I've got no software

also the sparc support will be cool, but in the 25 and 45 the PCI-E is whacked out so you can't use it anyway :/ Not like any of us have those to begin with but O2 + new graphics would be sweet, man, turn on "show window contents" and it really shows the performance hit.

[mattst88]

I wrote up a little blog post about this stuff if anyone is interested. http://mattst88.com/blog/?dd=18&mm=07&yyyy=2008

Also, D-EJ915: I love your signature

1 working 1 dead, and they're my friends

[tillin9]

You're not the first person to doubt my statements, but a big thanks for bugging the devs yourself and making a nice article for me to link to in the future. After your first responses I was going to go dig up all the old irc / forum posts where I hashed it out the first time, but now I don't have to.

What really surprises me is that more alt-arch people don't know / care about kernel modesetting. I'd think there'd be tons of embedded MIPS / ARM people who program devices with PCI buses (you'd be surprised that this is the majority, and most of the rest have PCIe 1x) who would love this. Just stick in a PC video card and have a low-end desktop environment for debugging / development instead of messing around with serial ports and JTAG.

[Gray Fox]

Why do people want to install Linux on SGI Systems for anyway? Same with Sun, Mac, HPPA, and etc. Why now save some money and hours by just building a PC and just install Linux on it. It looks the same on any other system.

[zizban]

Why do people want to install Linux on SGI Systems for anyway? Same with Sun, Mac, HPPA, and etc. Why now save some money and hours by just building a PC and just install Linux on it. It looks the same on any other system.

Some people want to use Free Software, which Irix aint. You and I may not agree with putting it on a SGI but it's his choice.

No flame wars and now back on topic.

[mattst88]

What really surprises me is that more alt-arch people don't know / care about kernel modesetting. I'd think there'd be tons of embedded MIPS / ARM people who program devices with PCI buses (you'd be surprised that this is the majority, and most of the rest have PCIe 1x) who would love this. Just stick in a PC video card and have a low-end desktop environment for debugging / development instead of messing around with serial ports and JTAG.

In all fairness, for non-Radeons, the driver will have to initialize the card itself. So kernel modesetting won't magically cause all graphics cards to work in MIPS boxes.

It does, however, look good for AtomBIOS enabled Radeons though.

Why do people want to install Linux on SGI Systems for anyway? Same with Sun, Mac, HPPA, and etc. Why now save some money and hours by just building a PC and just install Linux on it. It looks the same on any other system.

For me, it's a fascination with architectures. I can't totally explain it. It's interesting to me as an assembly programmer for obvious reasons. Otherwise, it's just interesting because it's different. If you've ever done x86 assembly programming, you'll definitely understand my attraction to Alpha, MIPS, and PA-RISC.

This, coupled with the fact that getting a copy of commercial UNIX is both difficult and expensive, makes Linux an obvious choice.

Say I buy an Alpha. I can't easily, or cheaply, get Tru64. Even if I can, I've got to have a separate license to use hardware 3D acceleration offered only by a Radeon 7500 in my configuration. I've got to have another license to use the C++ compiler and another license to access certain features of the Advanced Filesystem.

Compare to Linux: Freely available. Maintained (that is, releases are more than just scheduled maintenance before decommissioning). C++ Compiler is freely available. Only cost involved in 3D acceleration is the time to setup. Plus, I can use my X1550 or even my 9100, both much newer than the 7500.

The situation is quite similar with that of MIPS/IRIX and PA-RISC/HP-UX I'd think.

[tillin9]

While AtomBIOS and AMD/ATI's new open source policy makes this a lot easier for radeons, there is no reason why kernel modesetting won't work for all cards. Intel does it themselves, and I'd actually think older cards would have a lot simpler sequence than even an AtomBIOS parsing radeon routine. That leaves Nvidia to be reverse-engineered, and the Nouveau devs seem to think they can do it. http://www.phoronix.com/scan.php?page=n ... &px=NjU0Mg

Also note that the radeonhd camp doesn't like AtomBIOS and bit bangs all the registers manually (although the higher ups got a little annoyed since it took much longer to support cards than the radeon folks which used AtomBIOS). AtomBIOS is just a nice interface for register table lookup allowing the same initialization code to reused. It doesn't actually initialize the card itself.

As far as to why, I got into Linux on SGIs since I needed to teach and intro assembly programming course. Most courses either use embedded MIPS devices which make learning very hard because they don't have enough power / memory to compile things themselves or try teaching x86 and confuse the hell out of everyone. In the first case, 9/10 times the problems aren't with understanding the assembly but with getting the program on the embedded device (cross-compiling, bricking the device while flashing) or being able to debug it (JTAG) when something goes wrong. I could just take the proper time to teach everyone how to do all that, but the department and university really look down on skills based classes. One intro to UNIX class was canceled for and I quote "Being too practical." Thus I got a couple of O2s, installed gcc, and was able to stick to Patterson and Hennessy.

[mattst88]

While AtomBIOS and AMD/ATI's new open source policy makes this a lot easier for radeons, there is no reason why kernel modesetting won't work for all cards. Intel does it themselves, and I'd actually think older cards would have a lot simpler sequence than even an AtomBIOS parsing radeon routine. That leaves Nvidia to be reverse-engineered, and the Nouveau devs seem to think they can do it. http://www.phoronix.com/scan.php?page=n ... &px=NjU0Mg

Yep, probably right.

Also note that the radeonhd camp doesn't like AtomBIOS and bit bangs all the registers manually

Well, used to. http://www.phoronix.com/vr.php?view=12563

It's silly to have two drivers for the same cards, especially when a significant amount of the developers' time is spent copying code from the other driver. Once radeon kernel modesetting hits the kernel, one is going to die off. The kernel developers aren't going to be interested in having two drivers in the kernel for the same hardware that use 80% of the same code.

[Zerolapse]

I've had Debian Sarge (back when that was new) running on one of my Octanes with Dual 300 and ESI graphics. Surprisingly, the speed wasn't bad and it was very usable. I built a cross compile toolchain on one of my other linux boxes and custom built a kernel for it based on the patches from linux-mips.org, then net-booted it. I was able to install Debian-MIPS via NFS. After that, I compiled Xorg with the patches from here: http://www.linux-mips.org/~skylark/ and was able to get Gnome/firefox/etc going with no trouble after that. It was basically a fully functional linux system. The previous link also contains the status of the Octane Drivers, etc. It actually ran fairly well and I didn't have much trouble with it at all. Let it be known that I do very much prefer IRIX though

-Jesse