Re: [PATCH 1/6] mmc: dw_mmc: remove the deprecated "clock-freq-min-max" property

[Rob Herring <robh@kernel.org>]

On Fri, Feb 23, 2018 at 06:16:39PM +0200, Andy Shevchenko wrote:
> On Fri, Feb 23, 2018 at 4:19 PM, Shawn Lin <shawn.lin@rock-chips.com> wrote:
> > On 2018/2/23 21:27, Andy Shevchenko wrote:
> >> On Fri, Feb 23, 2018 at 8:41 AM, Jaehoon Chung <jh80.chung@samsung.com>
> >> wrote:
> >>>
> >>> 'clock-freq-min-max' property had already deprecated.
> >>> Remove the 'clock-freq-min-max' property that is kept to maintain
> >>> the compatibility.
> >>
> >>
> >> Removing a property without telling the user what to expect is a bad
> >> idea and ABI breakage.
> >>
> >
> > What's the general process to remove a property?
> >
> > I guess we should do:
> > 1) deprecate it in the first place and remove it from all upstream DT

Yes

> > 2) wait some long enough days for expecting the stale of all old DTB
> > containing that property

Yes. How long that is depends on the platform. I think the minimum is 1 
release cycle. Some stable platforms are years. If there are other DT 
changes with new features everyone should want/need, then that can be a 
decision point.

Given this is a shared IP block it's harder to know, so you may need to 
err on the longer side.

> > 3) remove the functionality of the deprecated property from the driver
> > but still leave some warning there

I'd say add a warning in step 1 and combine 3 and 4.

> > 4) remove the left warning finally
> 
> I don't know. Perhaps Rob can shed a light here.
> But I would really OK with removal of some of such properties from
> some drivers where it's more burden to keep them.
> 
> > And for the ABI breakage, we should add something in Documentation/ABI
> > /obsolete  or Documentation/ABI/removed ?

It is only an ABI break if someone notices.

Rob