Problem
The system freezes when the console is idle. All running processes
will suspend. Installing the Release Supplement rs506a and OSS648a
will not fix the problem.The system will wake up when pressing any key on the console
keyboard or by moving the mouse. All frozen users and jobs will
resume without any errors. Solution
NOTE: PLEASE MAKE SURE YOU FOLLOW THESE STEPS WITH EXTREME CAUTION.
This problem has been reported to SCO engineering and will be fixed in the next release of the product.
For OpenServer 5.0.6 you can use the following fix:
# cd /etc/conf/pack.d/clock
# cp -p Driver.o Driver.o.orig
# /etc/_fst -w Driver.o -
* tsc_hclock_init+9c?h 4
tsc_hclock_init+0x9c: 0x2 = 0x4
* $q
# cd /etc/conf/cf.d
# ./link_unixNOTE: You type everything except the "#" and "*" prompts, and the line that starts with "tsc", which is output from /etc/_fst. The tsc line shows that we've patched the old value 0x2 to new value 0x4; if the old value doesn't show as 2 then back out the change and contact SCO Support.
To back out the change, copy the /etc/conf/pack.d/clock/Driver.o.orig back to Driver.o. Relink the kernel and reboot the system.
# cp -p Driver.o.orig Driver.o
# cd /etc/conf/cf.d
# ./link_unixIn more detail, the value being patched is the absolute minimum
countdown value with which the clock driver will program the 8254
Programmable Interrupt Timer. The Openserver 5.0.6 clock driver
dynamically programs the PIT with different countdowns in order to
provide microsecond resolution timing.With a countdown of 2, the PIT toggles its IRQ pin so quickly that
by the time the CPU tries to acknowledge the interrupt, it finds
no interrupt pending. So it raises a "spurious interrupt".
Before that finishes being serviced, the CPU will try to find
another phantom clock interrupt. When receiving a higher priority
interrupt (such as the console keyboard, a NIC or a host adapter
interrupt) the cycle will break.Any PIT countdown value could cause an occasional spurious interrupt if interrupts were disabled until just before the PIT turned off its IRQ. An occasional spurious interrupt causes no problem. The problem happens if the countdown is so short that another_ spurious interrupt is consistently generated before the spurious interrupt handler finishes with the previous one.