The collected works of jan-jaap - Page 20

http://techpubs.sgi.com/library/tpl/cgi ... /ch18.html

DevDriver_PG wrote: EISA Address Mapping
The pages of EISA I/O address space are mapped to physical addresses 0x0001 0000 (slot 1) through 0x0004 0000 (slot 4). The 112 MB of EISA memory address space is mapped to physical addresses between 0x000A 0000 and 0x06FF FFFF. Addresses in these ranges can be mapped into the kernel address space for PIO or for DMA (see “Kernel Functions for EISA Support”).

I have written IRIX device drivers professionally. The device driver programmer's guide is some 1100 pages and even then there are some things which are simply not documented or unexpected and/or which require some assistance from and SGI engineer. That is a bit of a problem, these days since SGI doesn't support IRIX anymore (unless maybe if you're related to the US defense apparatus and have serious money to spend). But the basics (initialization, memory mapped IO, interrupts) work mostly like your average Linux/UNIX.

Onyx2 / Origin 2000 memory can the same kind of sandwich contruction, and ditto for Octane memory.

That last one is a GIGAchannel Expansion Chassis, but it's missing the interface board. Also, judging from the photos, it uses a different form factor power supply.

FWIW, the Origin 200 is purely a server: it is headless and can hold a decent number of disks. It also some 15 years old and it's pretty loud. I have one and use it as my install server. This it does just fine but it's not a system I want to run for prolonged periods of time due to the noise and the power consumption.

I think I paid €250 for the system with 2x400MHz CPUs, 4GB RAM, V12 graphics and PCI cage. That was probably in 2008. It came with various bits and pieces to run a Discreet Flame as well. I think I paid some more for some fibre channel hardware (a SUN A5200??). The catch was that the seller had back problems and needed someone to come over and remove it from his place. I prefer to do business in person, but for the seller it limits the audience (thus lowers the price ).

The extra carrier with V12, DCD, DM2 and FC adapter was $200 ... $250 shipped from Israel. I don't know anymore what I paid for the CPU, I think it was $350 + shipping from the US. You can buy the 3c996bT ethernet cards and the QLA2342 FC cards for a couple of bucks.

Lesson to learn here: you can get a pretty decent machine for a fair price, but if the upgrade fever hits and you buy the top spec parts individually you can easily double or triple your total cost. Even if you patiently wait for 2 years for a favorable deal like I did. Of course now I have a spare V12, and the 2x400MHz CPU ended up in a system I sold on so I reclaimed some of my money again.

Back then the exchange rate USD:EUR was also more favorable and they didn't really bother with import tax on items worth less than ~ EUR 250 or so.

Oh, and I did a fresh install on the system as usual. And then I managed to loose the backup of the Flame installation

3D print it!

Edit: had a look at that auction, and it looks like someone pulled the mainboard out (to get to the PSU probably) and didn't reconnect anything when (s)he put it back in. System controller and SCSI appear to be unconnected. And the 3rd chassis fan is missing so the MSC will probably raise a fault condition and run the other two at max speed. Well, after you find a PSU, of course.

mopar5150 wrote: Well, not so lucky. Both side panels got broken in shipping and bent the rear fan housing.

That sucks.

mopar5150 wrote: cards I have pulled from origins. One is marked "missile emulator" and there are several Xilinx boards as well.

IIRC there used to be FPGA 'accelerator' cards for the Origin series. Sort of like RASC for the Altix / Altix UV series. Typically the sort of hardware out for each for just about anybody except those with unlimited budget. Which might explain the DOD stickers and the fact you rarely if ever see this gear on eBay.

nekonoko wrote: I have one exactly like it, but my power supply died a while back

Well, mine still works but it's the last one I have. My 4D/440VGX has a dead PSU and I have two dead PowerOne spares. My desk has been littered with PowerOne bits for months but I don't have the time to fix it ...

foetz wrote: 6000 bucks?

Just imagine what mine (with 150MHz R4400 CPU, 256MB RAM and 100Mbit FDDI networking) must be worth

That's an Elan graphics card, in an PowerPath-GIO32 adapter. My Crimson had Elan graphics too when I got it. If you look carefully you'll be able to spot the slot bracket and everything. It lacks only one thing wrt. the Indigo Elan board: the two plastic handles to insert it into the Indigo chassis.

robespierre wrote: If that's an Extreme, does it mean Extreme gfx exists for Indigo GIO32?

No, but a similar contraption existed to add Extreme graphics to a Challenge deskside.

Elan graphics for Personal Iris existed also: that required yet another adapter.

Axatax_ wrote: Are there any available schematics for this PSU?

I wish If anybody has anything please get in touch with me.

First: there are two different types of PSUs used in the Crimson / PowerSeries.

Originally, these systems shipped with a PowerOne PSU. This is a silver box. The Crimson in the eBay auction has this PSU.
Later (mostly Crimsons), these systems shipped with a Cherokee PSU. This is a black box.

Both PSU's provide the same output power: 1050W. If you search the comp.sys.sgi archives you'll find references to 1500W PSU's (presumably RE ready), but this is a mistake. 1500W is the max input power of the PowerOne design. These PSUs are connected to the backplane of the system with laminated steel rods (!), and the position of the terminals of the Cherokee is different from the PowerOne, so you can't simply transplant a 'bare' PSU into a chassis made for the other design, you need the corresponding wire harness.

I've always disliked the Cherokee design because it gives a much cheaper impression than the PowerOne design. It is a mess inside, like a PC PSU on steroids. The PowerOne design was marketed in nuclear and medical markets so they probably weren't cheap, and it shows. Also, I've had both designs fail on me: the Cherokee went up in smoke (literally: scorched PCBs), the PowerOne PSUs shut down and refuse to start the system.

I have given up on the Cherokee, and have PowerOne PSUs in my 4D/380 (works), 4D/440 (failed), Crimson (works) + at least 1 failed spare.

I have some info available here:
http://www.vdheijden-messerli.net/sgist ... wersupply/
http://www.vdheijden-messerli.net/sgist ... -powerone/
PowerOne has been acquired by another company in the mean time and I think most of the links are MIA now. I have copies of these documents.

I have also reverse engineered the schematics of the bias inverter on the mainboard. The mainboard is where (in my case) the problems are. I have enough of these things that I can swap modules around to make sure of that.

This is a work in progress but I rarely have time for it. I hope to fully understand these PSUs before the last working specimen kicks it.

FWIW: these PSUs are modular. There's always a rectifier board, a main (controller) board, and some modules which generate the low voltage / high current output. Different configurations of these power supplies were used in Teradyne testers. If you're sure the problem is in the mainboard you could try to find one on eBay (search:"power one spm5") and transplant the output modules used by SGI. I never tried that (these things are too expensive to ship for me to try). It may not work. It may even explode and/or kill you, you're on your own .

PS: there's an even bigger version of the PowerOne design around. It's 220VAC only and puts out 250A @ 5V. Seriously screwdriver melting stuff. You find them in PowerSeries 'SkyWriter' and 'Predator' racks, like my 4D/380.

foetz wrote: there're still some declarations missing and CLOCK_MONOTONIC doesn't exist on irix

CLOCK_SGI_CYCLE can be used as a replacement

vishnu wrote: a million years of googleing.

Google, meh. What's wrong with going straight to the source

CLOCK_SGI_CYCLE and CLOCK_MONOTONIC may have different corner cases, especially on 32bit systems (wrapping).

If you want a linear counter rather than a struct timespec, check the documentation of the SGI_QUERY_CYCLECNTR request to syssgi .

IIRC, there were people who placed really large CRT monitors on top of their pizzaboxes and cracked the case. Ever seen one of those 24" widescreen Sony FW900's? It weighs nearly 50kg.

Indigo2 MaxImpact is definitely a very nice Indigo2. Congrats.

julesr wrote: Did you see that I got the Origin loaded up as far as possible (it doesn't have the right router boards to take four nodes)? It does indeed have a snafu front-panel switch as you suspected, but I think that I can fix it.

I have both the router board(s) and the front panel switch but I'm on the wrong side of the Atlantic.

Asking price on eBay is $10K less .

hamei wrote: O350's are garbage. I would suggest avoiding them.

I guess I was lucky. I got mine in 2010 and it's been running 24/7 since then with only a single issue ( a PSU failed ).

hamei wrote: btw, ask j-j if his Fuel is running

Nope I could have fixed that long ago, but technically he ownership of the system is a bit undefined so I'm not terribly motivated.

I can name many things that are broken or unreliable in many SGI systems, but the O350 just isn't high on my list of crap designs. I can think of:

* PSUs of big iron. All of them. At least the old ones (4D series) are generic enough that they can be repaired or replaced. The ones in the Onyx1 can set your house on fire. The ones in O2K are impossible to diagnose because I can't figure out how to fire them up outside a system and you can't do anything once they're in the system. The O350 PSU was flawed but at least they are generic enough that you can buy a new one on eBay for as little as 1EUR (really. did that).

* RMs of big iron. Especially 1st gen SMD fabrication -- Reality Engine. There's enough flex in these boards that, not they're old and soldering joints are brittle, they will be 100% DOA if you ship them. Also affects IMPACT TRAMs.

* In my experience, power supplies are a very weak link once they get older. Which makes the Onyx1/Challenge, with it's countless DC/DC converters all over the place a complete reliability nightmare. Add to this the RM and PSU issues I just mentioned and you'll understand why I got rid of my Onyx RE.

* Anything with compression connectors. Even in a controlled environment there's dust, it's just finer, black dust. Which builds a 'cake' layer on your PCBs after a decade or so. Damn the poor soul who takes apart an Onyx2 or Octane in this condition because that crap will ruin your compression connectors and/or make them fail intermittently.

I can think of many other blunders, like how flashing the CPU speed of a mainboard works (sic) on many systems. This still kind of works for a Fuel as long as you install a faster CPU than what was in originally, but the same applies to O2000 and, more severely, O200. In an O2K you can flash one node from the other, on an O200 this means you need a second O200 system to do the job. Really, the L1 should have been able to take care of this. Or maybe it can and nobody told us?

If you have as many SGI systems as I have, there's always something broken. That's to be expected, many of them are 20 or 25 years old. It's sometimes a challenge to keep up with the breakage and get things running again.

I usually don't keep PC hardware around long enough to compare it to my SGI's in terms of reliability. It's easy to complain about the (un)reliability of the PSU of a 1990 vintage 4D PowerSeries, but the truth is I have no idea whether a PC from that era would still work because I just don't have them. My experience with 'PC' server hardware is quite good as far as reliability goes, but I use Tyan or Supermicro boards, Xeon CPUs and ECC memory, and I retire them after ~ 5 years, before they fail. My current PC workstation is an HP Z600 with two hexacore Xeons and 48GB RAM -- best PC I ever saw. Side panels in solid aluminum, sounds like a vault door when you close them.

ivelegacy wrote: so the IP30 is a bad system from this point of view because disassembling the the gfx out of the chassis is a trouble due to the dust, and you'd better put a protective cap on the compression connector.

I think the problem is more that a dirty system is best left undisturbed because once you pull the thing apart the dust goes everywhere, including the compression connectors.

ivelegacy wrote: I had dust troubles with the compression connector on the pci cartridge, solved blowing air compress.

Don't forget that you need to clean the opposite side (the backplane) as well, not just the compression connector. Properly cleaning an Onyx2 backplane is no walk in the park, you need to remove *a lot* of sheet metal before it comes out.

ivelegacy wrote: i have to remove a gfx, when can i get a few protective caps ?

If you have a 3D printer you can make your own . A tictac box works too

Good to see it all worked out!

Duck is right, the Indy/Indigo2 'slab' keyboard is basically the same as the Indigo keyboard. To type on, that is, the internals are completely different. The Indigo2 'slab' keyboard exists in a beige version too, if that's important to you.

Every component (PCB) should have a part# in the form 030-xxxx-yyy. Google will then tell you exactly what it is.

pentium wrote: Failing that, think there is a practical way to fully image the disk, regardless of errors?

ddrescue

Good luck!

kickstarter != startup company .

I worked for a couple of startups and if you think working at a startup is matter of taking it easy while you burn though someone else's money you've got it all wrong. Those who work for a startup usually work very long hours for a reduced salary (if any) -- the reward is that you will be (part) owner of a successful company if things work out, the risk is of course that you will be left with nothing if the startup doesn't take off.

If you think VCs are clueless idiots looking to give free money to any idiot, think again. They are sharks. Those who provide seed capital for startup companies will demand shares in the company and you really want to create income *fast* because every funding round will dilute your own shares.

Things were a little different in the late nineties during the dot-com bubble. Back then you could be part of a startup *and* be treated like a rockstar, nice hotels, limos etc. No kidding, been there, done that There's a reason that bubble didn't last...

Things maybe different for a kickstarter -- I'm not an expert on those.

Update: it took me a couple of years but I managed to fix the power supply and this baby runs once again!

foetz wrote: a true classic. what kind of disk(s) do you have in there?

One is a 1.2GB full height 5.25" Seagate, the second a plain old 3.5" 4GB IBM DDRS in a half height 5.25" adapter. They are both set to SCSI ID 1. The Seagate 'brick' has IRIX 4.05 installed, the IBM disk has IRIX 5.3. This system has an IO3 dual channel SCSI controller and both channels are wired to the drive bays. So by re-arranging the disks I can dual boot the system

Otherwise, this system has the v4.x PROM chips so it will recognize a CD-ROM for what it is:

I think it will boot any CDROM which is 512byte block capable, but this system has the infamous P6-CDROM first gen "really, I'm a harddisk" CD-ROM.

NB: that 1.2GB disk is rather unusual. Most of these things are 760MB. Deluxe

I was kind of interested in that 4D/70 product literature, but I'm not going to pay $70 for it .. that's more than I paid for my actual 4D/70 system ... Winning bidder seems to have a bidding history in Vintage Merchandise, not Computers & Networking (what you'd expect from the crowd on this board). Maybe there's a market for this kind of stuff regardless of the brand (SGI)?

I bought an unusual IRIX CD from this seller for a couple of bucks.

Never seen it mentioned here, thought I'd share. There's a good number of SGI brochures: http://www.1000bit.it/ad/bro/brochures.asp?id=119 but also other retro / game / 8-bit systems etc etc.

I found back my first computer: http://www.1000bit.it/scheda.asp?id=322

Yeah I saw that too. Same person won the 4D/70 leaflets also.

There's some info on the 4D/80 and 4D/85 on Gerhard Lenerz site: http://www.sgistuff.net/hardware/system ... ional.html

I guess it's a single tower 'Diehard' chassis with the VME backplane of a Professional IRIS inside? It was introduced the same year (1988) as the PowerSeries, so it must have been quickly overtaken by the 4D/210GTX (also single CPU limit, but better specs) and the MP Power Series at the high end.

I have a couple of binders with similar brochures, technical reports and other artifacts (price lists, invoices, warranty contacts). Apparently Ian Mapleson has a good number of them as well. Since people apparently willing to pay big bucks for them, maybe we should scanning this material so that the information is available to computer hobbyists and not just collectors of vintage merch with fat wallets.

Nice Indigo2!

If you want a 195MHz CPU there's one on eBay.de for 20buck o.b.o.

BTW: you mention Maximum Impact-AA, did you mean High Impact-AA?

No, it doesn't. I tried once but the build system was sufficiently Linuxified that I lost interest.

While it does have a MIPS code generator these days, it's not from MIPSpro.

The IRIX/MIPSpro compiler works like this:
* Front end: licensed from 3rd party, the Edinburgh Compiler Group (EDG). Many other commerial UNIX vendors used this front end, including the DEC/Compaq compilers for Tru64, and I'm pretty sure the Intel Compilers use it.
* Intermediate (optimizer): built by SGI
* Back end (code generator for MIPS CPUs): built by SGI.

Later, SGI built a new back end for IA64 CPUs, and open sourced the compiler as Open64. They couldn't open source parts they didn't own (EDG) so they replaced it with bits from GCC. The MIPS back end code was not open sourced.

Later on, other people implemented new back ends from scratch for x86-64, and MIPS.

ivelegacy wrote: I'd like to see an ADA compiler on Irix.

It has existed for every version of IRIX since at least 5.x. It's based on GNAT and was supplied by SGI. IIRC you can even find it on ftp.sgi.com.

rwengerter wrote: (you know compilers, don't you? ).

Well, they are the 'tools of my trade', so I better
Also, (many years ago), I used to contribute to GNU Pascal and MINGW , to name a few. Now I have kids and a full time job writing code so I don't spend too much of my free time writing/porting code anymore.

There's an article about the Reality Engine in the same issue.

Kumba wrote: I honestly cringe to know what odd security bugs still exist in the final release of IRIX that no one knows about.

A new 'out of the box' IRIX installation is a soft target. It will allow root login using telnet, 'nuff said ...

But any properly firewalled system is as secure as the services exposed to the network. If you limit that to SSH you should be reasonably safe, even if the source of random on IRIX probably can't compete with current implementations.

That leaves privilege escalation by local users. For IRIX systems operated by hobbyists that shouldn't be an issue -- the owner probably doesn't allow access to users (s)he doesn't trust.

Kumba wrote: What's the OpenSSH version in IRIX?

Last I have is v5.8p1 from patchSG0007246, April 2011.

A couple of CVE's apply , but nothing too alarming (nothing that will allow an unauthorized login over the network). Ironically, more severe CVEs apply to newer versions. So when you compile from source you have to keep up with the race.

My IRIX systems are never exposed to the internet.

robespierre wrote: that is not a good evaluation of security vulnerabilities... there are many attacks that do not rely on open ports or malicious local users. Heartbleed being only the most well-known.

You're right, I didn't cover the case of a user acting in good faith being abused by a malicious server. An outgoing firewall could be used to limit what an 'unsafe' system can do. I've got an old XP system which gets that treatment.

I've got a known good MMSC here, plus the touch screen unit which goes on the front of the O2K / Onyx2 rack. I've tested it with a deskside Onyx2 and it appears to work (it makes pretty bar graphs). One of the arrow buttons requires a bit of force. Can probably be cleaned.

I'll throw in some cables to hook it up, and a second MMSC with bad PSU.

Asking €50 + shipping (from The Netherlands).

TeamBlackFox wrote: I've learned how to pack systems much more wisely... its a shame this information I learned is just not widely discussed or taught.

http://www.sgidepot.co.uk/postingadvice.html

SoftWindows was not a pure VM like e.g. VMware. It's more Wine meets DosBOX. It included native versions of several key system files for performance reasons. Because of this you cannot simply install another OS on top of SoftWindows.

I think the predecessor to SoftWindows was SoftPC and ran DOS. We're now talking IRIX 4 days, though.

I don't understand what you mean with 'overheating'. There's no thermal shutdown on an Indigo2 -- it will run until it burns down, so to say.

I have a MaxImpact Indigo2, with the TRAM option cards and a G160 network card. It runs hot, surely. I used to have a cat who loved to bask in the exhaust heat. But overheat? Nope.

Beware that the Indigo2 chassis is divided into two sections. The part where the CPU and memory are is ventilated by the PSU and the airflow is front to back. The part where the graphics are has a dedicated fan and the airflow is from the back to the left side (or bottom, if it sits on it's feet). If you place the Indigo2 in a cramped space it may end up circulating the air!

The TRAMs get very hot, but that's not why they die. It's only the TRAM option cards that die, you don't hear much of HighImpact of MaxImpact boards where the 1MB 'base' TRAM goes bad. The TRAM option cards die because of mechanical and thermal stress in combination with aging soldering joints. I repaired a couple of them , you can find more details there.

There's firmware in just about every corner of any computer these days. Ever heard of BadUSB ?

robespierre wrote: that's a surprising amount of garbage on the console, I usually only see one or two chars of noise before it settles down on a sun 3/80 or 4/25. I never used the twin tower on serial though.

Solved it. My IRIS console cable has the hardware handshake lines wired, but PUTTY wasn't set to use flow control. I disassembled a good bit of the IP7 PROM many years ago, and I remember that the very first UART output used different routines than the rest. I've long retried the PC that ran the IDA software so I can't look at these listings right now, but I guess that's what bit me here.

The panel with the hex display and the mode switches is now connected, so DIAG is enabled. Now I get :

That's more like it

PS: this thing is 25+ years old, and unlike an Indigo it doesn't hang on an empty battery. Who knows, maybe it still works