Re: [PATCH v3 03/15] ACPI: Allow drivers to match using Device Tree compatible property

[Rafael J. Wysocki]

On Friday, October 03, 2014 10:59:10 AM Dmitry Torokhov wrote:
> On Fri, Oct 03, 2014 at 02:43:03PM +0100, Mark Rutland wrote:
> > Hi Rafael,
> > 
> > On Wed, Oct 01, 2014 at 03:10:40AM +0100, Rafael J. Wysocki wrote:
> > > From: Mika Westerberg <mika.westerberg@xxxxxxxxxxxxxxx>
> > > 
> > > We have lots of existing Device Tree enabled drivers and allocating
> > > separate _HID for each is not feasible. Instead we allocate special
> > > _HID "PRP0001" that means that the match should be done using Device
> > > Tree compatible property using driver's .of_match_table instead.
> > 
> > That's hopefully not the precise meaning of "PRP0001" unless we're
> > attempting no semblance of OS independence here?
> > 
> > I'm still of the opinion that marrying ACPI to existing (and often
> > ill-defined) DT bindings is a bad idea. While it's expedient, I believe
> > this is going to be a long-term maintenance nightmare.
> > 
> > I'm very concerned with the prospect of model mismatch between the two
> > (e.g. DT clocks properties where ACPI has traditionally been in charge
> > of clock management). I've not seen any high-level guidelines w.r.t.
> > what should live in _DSD properties and what should not (at least not in
> > the ACPI spec itself). There are almost certainly properties that only
> > make sense if !ACPI, and likely there will be some that only make sense
> > if ACPI.
> > 
> > So I think that in its current level of standardisation, _DSD only makes
> > sense for simple device properties, and not relationships between
> > devices, except where ACPI already has some kind of a model (which
> > currently seems to cover interrupts and GPIOs). I'd also hope that we
> > could expose a 'clean' subset of DT bidnings (i.e. those which aren't
> > known to be kept around only for compatibility with legacy DTBs).
> > 
> > I do not believe it makes sense to share such a low-level interface.
> > Given the aforementioned model differences, and the fact that we don't
> > need to support _every_ device tree binding, I don't see why this can't
> > be handled with separate probe paths in the drivers we care about (as we
> > already do for DT vs platform data).
> 
> Because as a driver writer I do not want to implement N+1 ways of
> getting device configuration. I want one API that works independently of
> the underlying platform. The DT vs platform data is bad enough already.

Well stated, thanks!

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/