Can you give me a DTS example or code segment? I'm ambiguous about
what you said.

-----Original Message-----
From: Benjamin Herrenschmidt [mailto:benh@kernel.crashing.org]=20
Sent: Tuesday, July 24, 2007 7:36 AM
To: michael@ellerman.id.au
Cc: Chen Gong-B11801; linuxppc-dev@ozlabs.org; paulus@samba.org
Subject: Re: [PATCH 1/3] Add a new member name to structure irq_host
=20
On Tue, 2007-07-24 at 09:22 +1000, Michael Ellerman wrote:

quoted

On Tue, 2007-07-24 at 08:04 +1000, Benjamin Herrenschmidt wrote:

quoted

On Mon, 2007-07-23 at 19:13 +0800, Chen Gong wrote:

quoted

This patch adds a new member 'name' for irq host's name to=20
structure irq_host and modifies the definition of the function=20
irq_alloc_host(). The assignment to the irq host's name is also=20
added to irq_alloc_host().

=20
I would instead prefer adding a irq_name callback to host, which=20
would return a per-irq name. In some cases, they could=20

return the OF=20

quoted

quoted

path to the host for example.

=20
The pity is, for the 75% [1] of irq_hosts that have a device_node,=20
they will all have to implement the name callback, because they all=20
store the of node differently.
=20
Which makes me think (again) that we should have an optional=20
device_node pointer in irq_host. I know you said you wanted the irq=20
stuff to be OF agnostic, but the reality is most of the=20
implementations do have an OF node. And all of the newer irq_host=20
implementations do, with the exception of PS3 and celleb -=20

which are special.
=20
Won't work well in practice due to things like:
=20
 - Cell iic -> one node per core, but one global host
 - xics -> one node per source controller (lots of them in=20
the tree) but one global host
=20
Ben.
=20
=20
=20