Re: IRQ remapping problem on Macbook Air 5,1

From: Seth Forshee
Date: Tue Aug 07 2012 - 16:55:21 EST


On Tue, Aug 07, 2012 at 01:47:52PM -0700, Suresh Siddha wrote:
> On Tue, 2012-08-07 at 22:10 +0200, Joerg Roedel wrote:
> > Hi Seth,
> >
> > On Tue, Aug 07, 2012 at 02:48:37PM -0500, Seth Forshee wrote:
> > > This machine has a bug in its ACPI tables that's causing it to fail to
> > > boot unless intremap=off is passed to the kernel. The MADT defines a
> > > single IOAPIC with id 2, but the remapping unit defined in DMAR matches
> > > id 0. Thus interrupt remapping fails, and the kernel panics with the
> > > message "timer doesn't work through Interrupt-remapped IO-APIC." If I
> > > force the use of the non-matching ir hardware for the IOAPIC everything
> > > seems to works fine. dmesg for the working boot is below.
> >
> > There are AMD systems with similar problems. For example I had a system
> > with 2 IO-APICs but only one was described in the IOMMU ACPI table.
>
> This issue should be caught by the current check in
> parse_ioapics_under_ir()
>
> if (ir_supported && ir_ioapic_num != nr_ioapics) {
> ...

No, there's one IOMMU and one IOAPIC, so this check doesn't catch it.
The problem is that the ids in the ACPI tables for the devices don't
match.

> > > I'm looking at how to make this machine able to boot by default. OS X
> > > obviously boots, and Windows presumably boots as well since Apple
> > > officially supports running Windows on its hardware.
> >
> > These operating systems probably don't support interrupt remapping, or
> > the IOMMU at all.
>
> Some of them use interrupt-remapping only if the bios has pre-enabled
> x2apic/interrupt-remapping etc. But Linux tries to enable
> x2apic/interrupt-remapping even if bios doesn't.
>
> >
> > > I've got a patch to leave IRQ remapping disabled whenever any IOAPIC
> > > does not have a matching ir hardware unit. This gets it to boot, but I
> > > thought I'd check and see whether anyone had any better ideas for how to
> > > fix this.
> >
> > This is exactly how I fixed this issue on the AMD side too. Mind to
> > submit the patch?
>
> yes, we should be able to extend the check in parse_ioapics_under_ir()
> to check for this condition.

I already sent the patch, but where I added it is in
intel_enable_irq_remapping(), just after the call to
parse_ioapics_under_ir(). But it would be no problem to move it if that
makes more sense.

Seth

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/