ICH5 irq fails in 100% native mode

While recently setting up a new machine with 2.6.3 and then 2.6.7-rc2 I discovered this over and over again causing system stalls:

disabling IRQ18

syslog showed:

Jun  2 00:31:28 localhost kernel: irq 18: nobody cared!
Jun  2 00:31:28 localhost kernel:  [__report_bad_irq+42/121] 

[note_interrupt+145/175] [do_IRQ+279/321] [common_interrupt+24/32] [__do_softirq+66/177]
[do_softirq+45/47] [do_IRQ+286/321] [common_interrupt+24/32] [default_idle+0/44] [default_idle+41/44] [cpu_idle+46/60] [start_kernel+407/467] [unkno
wn_bootoption+0/294]

dmesg revealed this:
ICH5: 100%% native mode on irq 18

Adding "noapic" to boot options simply moved the error to IRQ5 and occurred routinely during boot at the ICH5 definition.

A quick google revealed the IDE settings in BIOS might be related so I disabled the "ENHANCED" option for the ide controller which is both P-ATA and S-ATA and chose the "COMPATIBLE" option.

The error went away, and now dmesg shows this:

ICH5: not 100%% native mode: will probe irqs later

While I am not able to determine whether there is any performance penalty for this it seems that the probing of irqs at this point is responsible.

The hardware is an ASUS P4P800S motherboard with i848 chipset + ICH5 and a 2.8C P4HT.

While this is a workaround I wonder what I/we need to do to make it work in native mode.

Advice?

Con

what was your driver setup?

Ideally, there should be two PCI devices ("combined mode" disabled), one for PATA and one for SATA. The PATA interface should be configured in legacy mode, and the SATA interface in native mode.

That seems to be the most stable configuration.

(that also presumes you use the IDE driver for PATA and libata for SATA)

Jeff