Interrupt routing issue on BCM1480.

Interrupt routing issue on BCM1480.

[Kaz Kylheku]

Hi guys,

Ralf's fixes in arch/mips/pci/pci-bcm1480.c break on our board. Multiple
devices somehow report the same pin number (or perhaps some invalid pin
number that is mapped to 1).  The mapping function then assigns IRQ 8 to
multiple devices.

The old ``return dev->irq'' in pcibios_map_irq worked fine.

What was the issue: is it that CFE sets up these IRQ numbers, such that
they might not correspond to pins?

Interrupt routing issue on BCM1480.

[Kaz Kylheku]

Hi guys,

Ralf's fixes in arch/mips/pci/pci-bcm1480.c break on our board. Multiple
devices somehow report the same pin number (or perhaps some invalid pin
number that is mapped to 1).  The mapping function then assigns IRQ 8 to
multiple devices.

The old ``return dev->irq'' in pcibios_map_irq worked fine.

What was the issue: is it that CFE sets up these IRQ numbers, such that
they might not correspond to pins?

Re: Interrupt routing issue on BCM1480.

[Ralf Baechle]

On Tue, Nov 18, 2008 at 02:00:08PM -0800, Kaz Kylheku wrote:

> Ralf's fixes in arch/mips/pci/pci-bcm1480.c break on our board. Multiple
> devices somehow report the same pin number (or perhaps some invalid pin
> number that is mapped to 1).  The mapping function then assigns IRQ 8 to
> multiple devices.

That's fine.  A simple PCI device that uses an interrupt is supposed to
use INTA that is report a 1 in the interrupt pin register.  After going
through any PCI-to-PCI bridges (if they exist), INTA..INTD eventually
will be wired in a system specific way to the host.  For the BCM91480
that wiring seems to be to simply wire (or "or") all INTA pins together
and connect that signal to the SOC, same for the remaining INTB..INTD.
In practice this means it is expected to see most PCI devices to use
K_BCM1480_INT_PCI_INTA which is 8.  I suppose your system may wire up
things in a different way and CFE knows about that but not pcibios_map_irq().

> The old ``return dev->irq'' in pcibios_map_irq worked fine.
> 
> What was the issue: is it that CFE sets up these IRQ numbers, such that
> they might not correspond to pins?

Linux interrupt number is not necessarily the same as that of any
firmware, so relying on firmware is generally wrong or at least
something only to be done if you know what you do ...

  Ralf

Re: Interrupt routing issue on BCM1480.

[Maciej W. Rozycki]

On Tue, 18 Nov 2008, Ralf Baechle wrote:

> will be wired in a system specific way to the host.  For the BCM91480
> that wiring seems to be to simply wire (or "or") all INTA pins together
> and connect that signal to the SOC, same for the remaining INTB..INTD.

 Interesting -- swizzling is the usual approach to spread the load.  
Though this platform's software-driven IRQ dispatcher we have is probably 
not smart enough to take advantage of such an arrangement anyway, so 
perhaps they didn't bother to optimise hardware.  The 91250A did use 
swizzling.

  Maciej