* Dr. H. Nikolaus Schaller [off-list ref] [150311 10:13]:

Am 11.03.2015 um 17:44 schrieb Tony Lindgren [off-list ref]:

quoted

* Dr. H. Nikolaus Schaller [off-list ref] [150311 09:17]:

quoted

Am 11.03.2015 um 16:24 schrieb Tony Lindgren [off-list ref]:

quoted

Rather than just making platform_data into device tree properties..

Can't you hide the these custom properties behind the compatible flag?

You can initialize that data in the driver based on the compatible
flag and the match data.

This makes sense if you can group things to similar configurations.

Maybe I have not completely understood your proposal.

Do you mean to go back to have big parameter tables for each device/battery
combination in the driver code and the compatible flag (e.g. compatible = ?board17?)
chooses the right data set for the charging map and channels?

If you can somehow group them, then yes.

I don?t see how to group them. Could you make a proposal?

Sorry no idea about that :) I though you may have some ideas based on
dealing with the driver.
 

quoted

quoted

I thought this is what the DT was introduced for - to have the same driver 
code but adapt to different boards depending on hardware variations.

Yeah but you also need to consider the issues related to introducing
new device tree properties. The device tree properties introduced
should be generic where possible.

Yes, that was discussed for a while for this driver?s bindings leading to v4.

Which ones do you think are not generic enough?

It seems maps is the only one then, assuming the cpacity-uah can be made
Linux generic. 

quoted

quoted

And batteries have very different characteristics and vary between devices?

Right. Maybe that has been already agreed on to use capacity-uah for
batteries in general?

Ah, do you mean with generic/not generic the distinction between a ?ti,? prefix
and no prefix?

Well, I don?t know if there is such an agreement and I would have no argument
against calling it ?ti,capacity-uah?.

No no, "capacity-uah" is what we should use, but you need an ack from
the battery and device tree people that this is OK. Let's not add
"ti,capacity-uah? as that can obviously be a generic property.
 

quoted

In that case I have not problem with that as
it?s a generic property :)

Well, many batteries and systems have a fuel gauge chip (e.g. bq27000). This
chip ?knows? the capacity. But therefore it is not needed to specify
it anywhere because it can be read out (usually in uAh).

This driver is to solve the issue that there is no such factory-programmed
battery or fuel gauge chip connected to a twl4030 driver. Nobody can program
that capacity value - except someone matching the device tree with real hardware.

And, by doing and averaging some charge-discharge cycles the fuel gauge
mapping is calibrated.

OK 
 

quoted

quoted

The charging maps are depending on the battery type connected to the twl4030
and which madc channel is which value is also a little hardware dependent
(although the twl4030 doesn?t give much choice).

Just to consider alternatives before introducing driver specific
property for the maps.. Maybe here you could have few different type
of maps and select something safe by default? Of course it could be this
is higly board specific, I think some devices may be able to run below
3.3V for example..

Every battery setup has a different map (which also might depend on the
series resistance of the battery and the wiring).

quoted

quoted

And moving this information into the driver for each board that uses it
would blow up the code.

Right, I'm not saying we should build databases into the kernel drivers.
Just trying to avoid introducing driver specific properties unless
really needed :)

They are not really driver specific, they are battery specific. They
describe properties of the battery:

* capacity - depends on battery
* voltage depending on current charging level - depends on battery (and differs between charging and discharging)
* wiring of MADC inputs to the twl4030 is board dependent.

So all these properties are not driver specific, but describe hardware.
And therefore they had been introduced exactly this way into the old
platform_data driver.

So if you want to see a ?ti," prefix to the capacity property I would be
fine.

Oh if they are battery spicific, then ideally we'd have generic batery
voltage to capacity maps property rather than a custom ti specific
property.

To avoid extra hassles later on, maybe you could submit a generic
binding patch only documenting it to the battery people and the device
tree people? That will make it easier to maintain this driver in the
long run.

Regards,

Tony