Problem:
The 64GB option causes processes to spin in the kernel.
Setup:
Linux 2.4.2
Slack current 20-feb-2001
ServerWorks III HE
4GB main memory
dual 933 Piii: SL47Q (stepping 3), SL4KK (stepping 6)
gcc 2.95.2 19991024 (release)
Test procedure:
On /dev/tty1, type
cat /dev/sda >/dev/null # or any fast disk
On /dev/tty2, type
while sleep 0; do date; done
On /dev/tty3, type
top
The purpose of the disk read is to provide some nondeterminism,
which triggers the bug. `top' lets you watch the debacle.
Observation:
After a short time, dates will appear with increasing
irregularity. `top' will show `sleep' and `date' processes
racking up many seconds of CPU time. At this point, terminate
the disk read, and one of these processes will go completely
CPU-bound, and uninterruptible. In this state, the system is
unusable, since any exec'd process has a good chance of being
bushwhacked in the kernel.
Variations:
noapic problem remains
mem=1200m problem remains
nosmp noapic problem fixed
mem=800m problem fixed
4GB option problem fixed
Hypothesis:
Code to handle PAE has buggy spinlock management.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/