IRQ-Tables once again

[Eric W. Biederman]

ron minnich <rminnich at lanl.gov> writes:

> On Tue, 20 Jan 2004, Stefan Reinauer wrote:
> 
> > Reading the Linux kernel sources it seems that Linux _does_ expect the
> > devices to have interrupts assigned already, which obviously LinuxBIOS
> > does not do. This has to be "fixed" if LinuxBIOS shall be open to
> > anybody not patching special kernels that do the right thing.
> 
> no. In 2.4, in the right circumstances, the kernel can assign interrupts 
> just fine. 

However the statement is still true that it is desirable that we
assign irqs to support non linux operating systems.

We should have enough information to do so when we can generate
the irq tables from the device tree.
 
> > If we provide tables at all, can we maybe drop the irq table and only
> > use the mptable?
> 
> no, sadly the kernel needs an irq table for pci bus scanning.

In particular when the kernel does not see a pci-pci bridge
going to a pci bus, it will not scan it unless it sees it in pirq
table.  This is only true of the i386 kernel.  The x86-64 kernel uses
different logic to accomplish the same thing.

Regardless if we can our house it order it will be easy to provide
kernels with what they need.

Eric