Re: [PATCH 0/3] mmc: omap_hsmmc: get rid of ti,non-removable

[Tony Lindgren <tony@atomide.com>]

* Sekhar Nori <nsekhar@ti.com> [130916 05:33]:
> On Friday 13 September 2013 09:07 PM, Tony Lindgren wrote:
> > * Sekhar Nori <nsekhar@ti.com> [130913 03:18]:
> >> Get rid of TI specific binding ti,non-removable in favour of the
> >> generic binding present for the same purpose.
> > 
> > Looks like there's a different handling in the MMC driver
> > for no_regulator_off_init that's needed for eMMC. That needs to
> > be sorted out and tested first.
> 
> Okay. I couldn't really get the eMMC on my board to detect, but
> apparently Koen has been able to get it to work. I am not really deep
> into details of HSMMC driver so will check with Balaji on what needs to
> be done here.

I think the issue is that until we have a SCM driver module setting
up the PBIAS register as a regulator, eMMC cards won't work without
ti,non-removable.
 
> >> This patch set does not support the old binding anymore. So, yes,
> >> it does introduce an ABI breakage. IMHO, it is not really worth
> >> supporting both bindings ATM since DT-usage in OMAP is still very
> >> nascent and almost always DTB and uImage are upgraded together.
> > 
> > The old bindings must be supported. It's not like we can just drop
> > them. We should just keep the old binding and parse it the same way
> > as the generic binding. That's a minimal amount of code.
> 
> Yes, its minimal amount of code but its code that will very likely never
> get exercised in future because of the reasons I mentioned above. I fear
> in time we will accumulate a lot of unused code this way.

Hey it's an ABI :)
 
> Anyway, since you don't feel its okay to remove the old binding, I will
> go ahead and mark the old binding as deprecated. But perhaps we can
> come-up with some time frame within which users of old .dtb can upgrade
> their .dtb or consider never upgrading to a newer kernel? This sounds
> eerily similar to feature-removal-schedule.txt and yeah I am aware of
> what happened to that. But I still put it here since this problem of DTB
> compatibility is not going to go away.

But before doing that, we first must fix the issue for still
needing ti,non-removable.
 
> One thing that might help is a kernel configuration that does not
> support deprecated bindings so users of old dtbs can actually see what
> deprecated bindings they are using. This way they can probably plan for
> a dtb upgrade better. So in driver you would do something like:
> 
> 	if (of_support_deprecated_bindings() &&
> 	    of_find_property(np, "ti,non-removable", NULL)) {
> 		.. do something ..
> 	}
> 
> of_support_deprecated_bindings() would return based on a config option.

Hmm yeah maybe that should be discussed separately on the
device tree mailing list.

Regards,

Tony