RE: [PATCH 1/3] Add a new member name to structure irq_host

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Tue, 2007-07-24 at 17:25 +0800, Zhang Wei-r63237 wrote:
> Hi, Ben, 
> 
> Thanks! Please see my inline comments.
> 
> > 
> > Well, you would have to look at a cell blade or pSeries machine
> > device-tree, I don't have one at hand in DTS form... but basically,
> > those don't have a 1:1 relationship between hosts and nodes.
> 
> Is there any problem when the relationship between hosts and nodes is
> not 1:1?
> 
> The mapping method is defined in irq_alloc_host(). If there is only one
> host for many nodes,
> then these nodes will use the same mapping. Such as iic_host, the
> IRQ_HOST_MAP_LINEAR
> is used for its all nodes. We do _not_ care which node, we just care
> which virq is mapped from which host's hw irq.

Sure and it all works fine, I was replying to the suggestion of adding a
device_node pointer in there.

> > 
> > Anyway, I discussed with Michael today and it might be a good idea to
> > provide an optional device node field in the host struct that can be
> > used by the majority of the PICs that have one host == one node, that
> > would provide a good default for the name() function.
> > 
> 
> But we have some irq host which has not OF-node bound.

Sure, hence the word 'optional' which would provide a default name.

Ben.