On Sat, Jun 10, 2006 at 06:45:22AM +1200, kjcsb wrote: > I am running Fedora Core 4 2.6.16-1.2111 smp kernel on an aging 2 CPU Dell > PowerEdge 2300. I am attempting to resolve some issues I'm having with a > PCI device. I have changed many things (see posts to Asterisk users list > entitled "Audio problems on Zap & SIP, local network, not IRQ related?") > and am now trying to change the SMP affinity so that the PCI device (which > is using IRQ20) is handled solely by CPU1: > cat /proc/interrupts > CPU0 CPU1 > 0: 54233 51707 IO-APIC-edge timer > 1: 1 7 IO-APIC-edge i8042 > 8: 0 1 IO-APIC-edge rtc > 9: 0 0 IO-APIC-level acpi > 16: 47254 46402 IO-APIC-level megaraid > 17: 2169 1325 IO-APIC-level aic7xxx, aic7xxx > 18: 305 13 IO-APIC-level eth0 > 19: 0 0 IO-APIC-level uhci_hcd:usb1 > 20: 25694 20011 IO-APIC-level wctdm > NMI: 0 0 > LOC: 105840 105584 > ERR: 0 > MIS: 0 > > echo 2 > /proc/irq/20/smp_affinity > cat /proc/irq/20/smp_affinity > 00000002 > > However I can see the interrupts for IRQ 20 are still incrementing on both > CPUs. Indeed, after about 10 seconds... > cat /proc/irq/20/smp_affinity > 00000001 > > On further investigation, the SMP affinity on ALL of the IRQs is set to > 00000001. This implies that everything is handled by CPU0, which it clearly > is not! > > I wonder if anyone has experienced anything similar or could offer some > suggestions? Your interrupts are being round-robin'd across CPUs by the irqbalance service. Dave -- http://www.codemonkey.org.uk
