[PATCH v3 00/25] irq_domain generalization and refinement

[Russell King - ARM Linux]

On Sat, Feb 04, 2012 at 05:38:53PM -0800, Tony Lindgren wrote:
> * Russell King - ARM Linux <linux at arm.linux.org.uk> [120204 14:00]:
> > 
> > Actually, it turns out to be not that hard, because twl doesn't actually
> > make use of the IRQ domain stuff:
> > 
> > commit aeb5032b3f8b9ab69daa545777433fa94b3494c4
> > Author:     Benoit Cousson <b-cousson at ti.com>
> > AuthorDate: Mon Aug 29 16:20:23 2011 +0200
> > Commit:     Samuel Ortiz <sameo at linux.intel.com>
> > CommitDate: Mon Jan 9 00:37:40 2012 +0100
> > 
> >     mfd: twl-core: Add initial DT support for twl4030/twl6030
> > 
> >     [grant.likely at secretlab.ca: Fix IRQ_DOMAIN dependency in kconfig]
> > 
> > Adding any dependency - especially one which wouldn't be enabled - for
> > a new feature which wasn't required before is going to break existing
> > users, so this shouldn't have been done in the first place.
> > 
> > A better fix to preserve existing users would've been as below - yes
> > it means more ifdefs, but if irq domain is to remain a DT only thing
> > then we're going to end up with _lots_ of this stuff.
> > 
> > I'd much prefer to see irq domain become more widely available so it
> > doesn't require these ifdefs everywhere.
> 
> Your patch below looks like a correct fix to me to the problem
> you and Grazvydas are seeing:
> 
> Acked-by: Tony Lindgren <tony at atomide.com>

It's not quite correct, because OMAP4 has issues in this area as well
(which does select IRQ_DOMAIN but can be without OF.)  The result is
an oops from irq_domain_add() because domain->ops is NULL.

The right solution is three fold:

1. Wrap the bits of code in CONFIG_IRQ_DOMAIN
2. Get rid of the #ifdef CONFIG_OF there, so the 'ops' member can be
   initialized.
3. Fix the OMAP vp code not to oops when voltdm->pmic is NULL

which I have in my combined patch for fixing OMAP so far.