Using pilot-xfer causes an "unexpected IRQ vecor 216 and 219 on CPU #0 I'm using a Supermicro DBE (BX chipset) mobo uname -a Linux blue 2.2.16-22smp #1 SMP Tue Aug 22 16:39:21 EDT 2000 i686 unknown the following is a listing of ./log/messages I changed pilot-xfer to use ttyS1 and got the same results --cut-- . . . Nov 9 13:00:04 blue kernel: tty_io.c: process 3080 (pilot-xfer) used obsolete /dev/cua1 - update software to use /dev/ttyS1 Nov 9 13:00:35 blue kernel: unexpected IRQ vector 216 on CPU#0! Nov 9 13:00:35 blue kernel: unexpected IRQ vector 220 on CPU#0! Nov 9 13:00:35 blue kernel: tty_io.c: process 2474 (gpilotd) used obsolete /dev/cua1 - update software to use /dev/ttyS1 Nov 9 13:00:35 blue kernel: unexpected IRQ vector 216 on CPU#0! Nov 9 13:00:37 blue last message repeated 215 times Nov 9 13:00:37 blue kernel: unexpected IRQ vector 219 on CPU#0! Nov 9 13:00:37 blue kernel: unexpected IRQ vector 216 on CPU#0! . . . --cut---
This is a defect in the IO-APIC of the i840 chipset. It shows up under heavy i/o, or even rapid mount and umount of any filesystem. Intel needs to find out where the i840 defect is and help with this issue. The only workaround appears to be kernel option "noapic", which is a severe handicap to performance. At one point I tried to talk to Intel about this, but not being a manufacturer, they were uninterested. FYI, this same chipset appears to have stability problems in other SMP-capable o/s's as well. Many inquiries have not resulted in a solution, nor even specific identification of where the IO-APIC problems occur. This error is fatal on all i840 chipsets with the IO-APIC enabled. In my case the unexpected IRQ vector was always 217, to a non-existent handler.
My error, the earlier report was the Supermicro BX chipset, my intent was for i840 only. The description is accurate, but applies only the the i840 chipset. This is probably why the irq was other than 217. Perhaps the comments could be moved to a more valid bug report for i840 SMP (the motherboards that I have seen the reports on were Supermicro, as well as my own; the OR840 will have this problem also).