Re: [PATCH 1/3] clocksource: atmel-st: Remove irq handler when clock event is unused

[Boris Brezillon]

On Mon, 3 Aug 2015 16:36:29 +0200 (CEST)
Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:

> On Mon, 3 Aug 2015, Boris Brezillon wrote:
> > On Sun, 2 Aug 2015 11:40:28 +0200
> > Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > 
> > > On Sun, Aug 02, 2015 at 11:10:21AM +0200, Thomas Gleixner wrote:
> > > > I think Boris Brezillon had implemented it at some point, but it was
> > > > shot down for reasons I can't remember. 
> > > 
> > > You weren't around at the time.. DT people didn't like it, said they
> > > didn't like having to make up fake hardware in their DT crap.
> > 
> > I don't know who was right, but the fact is they won't be inclined to
> > take such an approach unless the virtual demuxer is not exposed in the
> > DT, which is almost impossible since irq users are identifying their
> > irq lines with a phandle to the interrupt controller and an interrupt
> > id (usually extracted from the datasheet).
> 
> I really have no idea why DT folks think that virtual devices are not
> suitable for DT entries. Marks working assumption from the old thread:
> 
>  > This sounds like a DT-workaround for a Linux implementation problem,
>  > and I don't think this the right way to solve your problem.
> 
> is simply wrong. This has nothing to do with a Linux implementation
> problem. It's a sensible workaround for braindamaged hardware at the
> proper abstraction level.
> 
> > Anyway, below is a solution providing a way to disable specific
> > handlers without reworking the way we are modeling shared interrupts.
> > Thomas, I know you were not in favor of the proposed approach, but,
> > AFAICT, it does not add any conditional path to the interrupt handling
> > path (which, IIRC, was one of your requirements), and is simple enough
> > to be used by people really needing it.
> > 
> > There's probably other drawback I haven't noticed, so please don't
> > hesitate to share your thoughts.
> 
> Yes, aside of the bloat, it needs special handling in free_irq() as
> well.

Right. Below is a patch fixing the __setup_irq()/__free_irq() functions.

Anyway, I won't insist on this approach unless you want me to go
further, and I don't plan to spend more time on convincing DT folks that
the virtual demux is what we need (I already spent more time than I
wanted arguing on this feature :-/).

Best Regards,

Boris

--- >8 ---