On Mar 22, 2007, at 4:01 PM, Benjamin Herrenschmidt wrote:

quoted

If you look at what mpic_init() does you will see that the registers
would be setup twice.

Also, you don't need to walk the device-tree at all.  As Olof pointed
out there maybe interrupts that are not described in the device
tree.  We can setup this information when you do a request_irq() ->
set_irq_type().

If you aren't doing a request_irq() for the interrupt than it doesn't
need to be setup.  Additionally, you should only doing a request_irq
() for an interrupt that is "owned" by that core.

The need to have an MPIC shared by more than one cores is something  
that
will be coming elsewhere too (can't give details...)

I think we should invent an MPIC specific property indicating that the
MPIC is to be shared and indicating on one of the users that it's the
"master" (the one doing the initial full setup, the others can then  
spin
on some bit somewhere before going on, unless we decide such a shared
setup, the PIC has to be pre-initialized by some firmware and we skip
all inits in mpic) and indicating which sources belong to which users.

I don't see why we need to do as much work as we do in mpic_init.  We  
already set IRQ_VECTOR_PRI in mpic_set_irq_type() we could easily add  
setting IRQ_DESTINATION as well.

- k