On Fri, Mar 10, 2017 at 02:47:06PM +0100, Stefan Assmann wrote:
> On 08.03.2017 23:34, Bjorn Helgaas wrote:
> > Hi Stefan,
>
> Hi Bjorn,
>
> > On Fri, Feb 10, 2017 at 12:45:17PM +0100, Stefan Assmann wrote:
> >> When checking for boot interrupts and PRT index is zero avoid doing
> >> any interrupt rerouting since the code currently does not handle
> >> determining the IRQ via _CRS method.
> >
> > I don't understand this at all. Can you educate me a little?
>
> Sure, I'll try to explain my chain of thoughts.
>
> > This bug occurs on the M2N-LR system. I think that board contains an
> > Intel 6702PXH PCIe-to-PCI-X bridge to support its PCI-X slots. A
> > PCI-X 3ware 9550SX is plugged in below the 6702PXH, and the problem is
> > with the wired INTx interrupts from the 9550SX.
> >
> > Can we tell where those PCI-X INTx wires are connected? There's an
> > IOAPIC in the 6702PXH; if they're connected there, the Intel 6700PXH
> > 64-bit PCI Hub datasheet [1], sec 2.15.2, says it should generate
> > upstream PCIe INTx messages ("boot interrupts") if the PCI-X INTx
> > interrupt is masked in the APIC.
>
> Correct, if you take a look at bridge_has_boot_interrupt_variant() we
> actually walk up all the bridges to the root bridge and check for the
> 6702PXH via irq_reroute_variant, which gets set via PCI quirks in
> quirk_reroute_to_boot_interrupts_intel().
>
> > If the PCI-X INTx wires are connected elsewhere, e.g., to some other
> > IOAPIC, the 6702PXH should never see its IOAPIC inputs wiggle and it
> > should never generate a boot interrupt.
> >
> > The fact that the 6702PXH IOAPIC doesn't appear in the lspci output
> > and that "pci=noioapicquirk" is a workaround makes me think that we
> > aren't using the 6702PXH IOAPIC and therefore we don't see boot
> > interrupts on this system.
>
> Oh, I did not notice that dmesg only reports 1 IO-APIC. Though walking
> the bridges reports the 6702PXH IO-APIC being involved, otherwise the
> quirk wouldn't hit.
The 6702PXH is definitely involved as a PCIe-to-PCI-X bridge, but that
doesn't mean the 6702PXH IOAPIC is involved. The PCI-X bus has INTx
wires that *could* be connected to the 6702PXH IOAPIC, but on this
system I think they're likely connected to an IOAPIC in the MCP55.
Bjorn
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
