[dihhuit]

Hi all, this is my first post here, but I'm running into a frustrating issue with my Octane. I've decided to wipe my system and do a fresh IRIX install (the hard disc came with whatever IRIX the previous owner had). Following the instructions for a netinstall mostly from here:
https://wiki.irixnet.org/network_install_ubuntu
using an Ubuntu VM as my network install host, I'm able to consistently get to the miniroot install, then it asks me to create a new file system. When I do, and choose the 4096 block size per guidance here:
http://www.sgidepot.co.uk/6.5inst.html
(my drive is 70GB), I get the following error screen:

Code:

WARNING: odsy_earlyinit re-entered. Init code probably paniced!
PANIC: CPU 0: TLBMISS: KERNEL FAULT
PC: 0xa80000002018cbe0 ep: 0xa800000020ba3ba8
EXC code: 8, 'Read TLB Miss '
Bad addr: 0xc0000fc000000000, cause: 0x8<CE=0
NOTICE - cpu 1 didn't dump TLB, may be hung
Reboot started from CPU 0
CPU 0 rebooting

Then nothing--just hangs there.

I've tried reseating all the boards, reseating the ram, and I'm still getting this crash (third attempt now). System info:

Octane, older mobo revision, 2x195MHz R10k, 768MB ram, MXE+SE dual head. 

I'm using the 6.5.22 netinstall files from the irixnet FTP.

Any thoughts on next steps I could take to figure out what's up here? I guess I could remove my dual head and go down to a basic graphics setup. I'm not sure why it's saying "odsy_earlyinit" I hope it doesn't think I have Odyssey graphics for some reason. The only other clue I have is the fact that CPU 1 "didn't dump TLB, may be hung"--could I have a bum CPU? Never had any issues running the system before the wipe.

Any help appreciated.

[Raion]

That error indicates a problem with your CPU. It could be overheating, it could be having a cache problem or your motherboard might not be seated correctly.

[dihhuit]

Thanks for the quick reply and guidance, Raion. I'll try reseating the mobo again, and if the issue persists, then I guess the next thing to consider is heat. Last resort would be a new CPU/mobo. I hope that doesn't end up being the case...

[Raion]

Could be something else but I doubt it's a software problem.

[jan-jaap]

I'm sorry, but an Octane with IMPACT graphics crashing in the initialization of the VPro graphics? That looks exactly like a software problem.

There should be an ide.IP30 file in your installation media. It's a standalone program, like sash and fx, so you can netboot it, without a hard disk installed in the system even. It will test your CPUs, memory etc. If it finds something you know there's a problem rather than speculating about it, but most likely everything will be OK.

There's a problem with the miniroot installation or IRIX that can easily bite you: if there are remains of an earlier installation on the disk, the miniroot will not format (mkfs) the disk and try to integrate the new installation into whatever is on that disk. That's a recipe for unexpected surprises. Also, a hard disk is probably the most likely part to fail in any computer so you want to make sure it's still good before you go through the effort of a fresh IRIX installation.

That's why I usually do this:

1. Boot the standalone fx utility.
2. Run a write-only disk exercise to zero out the entire hard disk. (exe > seq > wr-o -> select all sectors > go). This will take a while.
3. If step two produces complaints about remapped or failed sectors: toss the disk. Get another disk, rinse and repeat.
4. Before you exit fx, create a new disk label. Default options are fine for a workstation.

If you get this far, the results of 'fx' and 'ide' have verified that everything works (only the graphics have not yet been exercised). Now it the time to do an IRIX installation. The miniroot will create a new filesystem. If this results in an installation that doesn't work, you know the graphics are suspect because it's the only part you haven't exercised. But since the system has an SE and an MXE head you can pull one and see where the problem is.

NB: if the MXE head produces artifacts in texture demos, it's a good idea to re-seat the texture modules. The texture modules themselves are pretty durable, but they are only attached with some nylon screws on the side so they can get loose if a system is moved around.

[dihhuit]

"This will take a while" is probably the understatement of the century. The last time I did this, it was like 5+ hours. If anything is going to drive me to sell this thing, it will be waiting for the disc to exercise again...

That being said, there're a lot of good suggestions in there, jan-jaap, so thank you. I'm running the ide.IP30 now, but one update I do want to provide is that last night when I pulled the mobo to reseat it again for the n+1th time, I noticed that the Heart heatsink was loose. :/ All four screws needed to be tightened. That may not be THE issue, but I'm sure an overheating Heart would cause all manner of fun problems.

As ide.IP30 runs, I notice that it does say "This test takes approximately 12 minutes on a good dual processor machine with 128MB memory and MXI graphics." It also detects both of my cards correctly. Are you sure it doesn't exercise graphics? If it didn't why would the graphics card be called out in the printed time estimate?

If this doesn't get me anywhere, I'll pull one of the graphics cards, exercise the whole disc (I guess I could use a warm noisy office for the rest of the day), and try again.

---

Ok, it's done. And I get some weird results that point to maybe it is a graphics issue after all. Before the test starts, it detects both graphics cards, but at the end of the test, it says "No graphics boards detected" which is hilarious, because that's printed to the screen. Also my mouse is now frozen. Keyboard still works though, so poking around in the ide program I noticed that there are extended tests. Running those now. Says they'll take 25 minutes (2x as long, but it only took maybe 5 minutes for the regular ones).

[jan-jaap]

A 70GB disk taking 5 hours to exercise equals 4MB/s. That's slow. You probably didn't do a sequential write-only pass, but a write-compare or butterfly test. At 30MB/s it should be done in 40 minutes.

You might be right, the 'ide' doesn't do VPRO graphics, but it might do the IMPACT graphics. If it starts to throw triangles on the screen in psychedelic colors it does. In general it's a better idea to run diagnostics involving graphics from the serial console rather than the monitor screen, otherwise you loose most of the test output.

[dihhuit]

Interesting, I followed these instructions when doing the disc exercise:
https://software.majix.org/irix/install-fx.shtml

which is to do an "exercise/sequential"; modifier = wr-o (write only?); 1 scan(s), and it definitely took many hours (maybe 5 was an exaggeration, but it was definitely more than 2). Maybe I have a dying disc or a problem with the scsi controller?

Do you have any more info about running graphics tests from the serial console? So far I've been relying on my monitor for this whole process. If there's better diagnostic info I can get from the serial console then I'll go find my usb/serial adapter and null modem cable.

Other question: ide said tests were completed, but I didn't see a way to view results. Does it produce any kind of report, or only report if problems are found?

I super appreciate all the help, by the way. You guys are awesome.

---

Well, the extended tests are still going, now 45 minutes later, so maybe the machine has hung. I'll give it a full hour, and then I think maybe it's time to try pulling out graphics cards and see if some permutation of them gives me a stable setup. That will take some time, as I have:

1 SSI board
1 SSE board
1 SE board
2 SE compatible texture modules.

That gives 6 single headed permutations, and 4 dual head permutations...

[Jacques]

Nope, a large scsi disk will take a few hours to carry out sequential writes to. My 36GB disk took a few hours.

[jan-jaap]

Interesting, I followed these instructions when doing the disc exercise:
https://software.majix.org/irix/install-fx.shtml

which is to do an "exercise/sequential"; modifier = wr-o (write only?); 1 scan(s), and it definitely took many hours (maybe 5 was an exaggeration, but it was definitely more than 2). Maybe I have a dying disc or a problem with the scsi controller?

The instruction are correct, but if you wipe the entire disk you need to re-create a new disc label before you leave fx, because the miniroot installer cannot do this.

Do you have any more info about running graphics tests from the serial console? So far I've been relying on my monitor for this whole process. If there's better diagnostic info I can get from the serial console then I'll go find my usb/serial adapter and null modem cable.

You need a serial null-modem cable and another computer with a serial terminal program, e.g. a Windows PC with 'putty'. If your PC doesn't have a 'real' serial port, USB-to-serial dongles work too.

Connect with the following parameters: 9600-n-8-1. Hardware handshaking settings are mostly irrelevant at these speeds.

If you disconnect the keyboard and mouse, the Octane will default to the serial console at power-on, and print a 'running power-on diagnostics' message followed by the [1]...[5] menu you know from the monitor. You can take it from there.

Other question: ide said tests were completed, but I didn't see a way to view results. Does it produce any kind of report, or only report if problems are found?

I *think* it's supposed to print something along the lines of "all tests passed" and show an 'ide>>' prompt. The problem is that the graphics console doesn't really work when the diagnostics take over the graphics board, so you're missing most output until the test is finished. That's why the serial console is a good idea.