SGI IRIS 4D Troubleshooting Guide

A Professional IRIS twin-tower workstation (Seiya Ishibashi, CC BY 2.0)

Fault-finding on the SGI IRIS 4D twin-tower Professional IRIS machines (4D/50, 4D/60, 4D/70, 4D/80 and 4D/85), from owners' restorations of a 4D/70 GT and a 4D/50 G. No SGI troubleshooting documentation for these machines was found. The other IRIS 4D lines have their own guides: SGI Personal Iris Troubleshooting Guide, SGI Indigo Troubleshooting Guide and SGI Crimson Troubleshooting Guide; for the Power Series see SGI Power Series.

Work with a serial terminal on the first serial port at 9600 baud. Most of the messages below appear only there.

Symptoms and causes

Faults found on two Professional IRIS machines

Symptom	Cause found	Section
Supply does not start on the bench	No load on the 5 V rail	Power supply
Loud whine or a steady beep from the supply	Worn chopper transistors; oscillation through the sense wires	Power supply
Exception during the first memory test	Faulty IP4 processor board	Power-on tests
"gfx: no working graphics hardware"	Processor board not making contact through broken insertion handles	No graphics
Kernel panic, bus error, when the graphics start	Wrong VME backplane jumpers	Kernel panic
"Graphics timeout (FIFO still > half full)"	Faulty GM1 board	Graphics timeout
Blue vertical lines in some windows	Display board fault, cleared by cleaning the VRAM contacts	Lines on the screen
IRIX 4 hangs when changing run level	No network interface fitted	IRIX hangs
PROM settings lost at every power-on	Flat NVRAM batteries on the IP4	General Maintenance

Power supply

• If the supply does not start, check that it has a load on 5 V. jan-jaap's 4D/50 supply would not start with only its built-in load resistors; it started with 10 A from an electronic load and supplied 40 A without trouble.
• The supply needs an external fuse in the machine's wiring harness. When bench-testing, the fuse has to be bridged or wired in.
• A whine from the supply under load was traced on jan-jaap's 4D/70 unit to its four 2N6678 chopper transistors, whose base-emitter voltage drop was higher than expected. After they were replaced, the +5 V rail showed about 0.6 V of ripple at 2500 Hz with a matching beep. That oscillation stopped when the sense wires to the backplane were disconnected. With new old-stock Motorola 2N6678s the supply ran without oscillation on the bench at 40 A, but in the machine, with the sense wires connected and more load, it oscillated at about 2 kHz again, so he left them disconnected.
• With the sense wires off, set the output so that the backplane, not the supply terminals, reads 5.00 V. jan-jaap measured 50 mV of drop in the cables.

Power-on tests

The IP4 runs its power-on tests on the serial console: console DUART, memory walking bit and memory address uniqueness. The early 4D1-2.1 PROM of November 1987 has no boot menu and no hinv command and drops straight to a >> prompt; the 4D1-3.1 PROM of December 1988 has hinv. The part numbers of the IP4 boards are listed on the IRIS 4D page.

An exception during the memory walking-bit test, before main memory is sized, came from a faulty IP4 on jan-jaap's 4D/50: the board failed its first write to memory, with an exception code for a data bus error. Cleaning the SIMM sockets and fitting known-good SIMMs made no difference, and the machine started with another IP4. His reasoning is that if the processor can run code from the PROM and drive the serial port, its data and address buses are working, and the memory addressing logic is more likely at fault.

No graphics

"gfx: no working graphics hardware" and "can't open gfx(0)" at power-on, with no picture and the graphics board LEDs solid red, appeared on the 4D/50 when its processor board was not making proper contact with the backplane because its VME insertion handles were broken. With a working IP4 properly seated, the graphics were found. Check the backplane jumpers as well; see General Maintenance.

Kernel panic when the graphics start

On the 4D/70 GT, starting the window system gave "Bus Error: Parity Error" followed by "PANIC: KERNEL FAULT" and a data bus error. A debugger trace of the crash dump showed the fault in the VME interrupt handler. The cause was wrongly set IACK and bus grant jumpers on the VME backplane.

Graphics timeout

"WARNING: Graphics timeout (FIFO still > half full)!" on the 4D/70 GT, with the screen going blank when the graphics login should start, was cured by replacing the GM1 Geometry Manager board. The GM1 has its own diagnostics serial port. At power-on its monitor reports PASS lines for its UART, DRAM, PP SRAM and SRAM word count register and then gives a gm> prompt; ? lists the commands and t ff runs a continuous memory test.

Lines on the screen

On the 4D/50 G, blue vertical lines every fifth pixel column appeared in some windows when several GL programs ran at once, and SGI's diagnostics reported errors on the TB2 and DE3 boards. jan-jaap linked the five-column spacing to the five-way parallel display subsystem. Cleaning the DE3's VRAM modules and sockets with alcohol, and treating the module contacts and the DE3 to TB2 interconnect with DeoxIT Gold, cleared the lines and the TB2 test errors, although the DE3 stripe test still added errors to its report.

IRIX hangs

On a machine with no network interface, IRIX 4 hung on every change of run level, including shutdown and a normal boot after the file systems were mounted, while single-user mode worked. Configuring the network interface hung the system. Fitting a VME ENP-10 Ethernet board cured it.

Diagnostics

• The ide diagnostics (the PROM's dbootfile was /stand/ide) passed on the 4D/70 but do not exercise the graphics.
• SGI's diagnostics tape for IRIX 4.0.1 includes the Clover 1 graphics tests, which jan-jaap describes as close to the manufacturing acceptance tests. Run straight after stopping the X server, they reported errors on two of the seventeen geometry engine chips that did not recur after a reboot.
• jan-jaap tested the GT graphics boards from his 4D/70 in a Power Series. The GE4, RM1 and RV boards work there with the Power Series' own GM2 board in place of the GM1, making a GTX system, and passed the field diagnostics. That pointed to the GM1.

Related pages

• SGI IRIS 4D
• SGI IRIS 4D General Maintenance
• SGI IRIS 4D Capacitor Replacement Guide
• SGI Personal Iris Troubleshooting Guide
• SGI Indigo Troubleshooting Guide
• SGI Crimson Troubleshooting Guide