[PATCH 0/2] IO voltage domain support for rk3188 and rk3288

[Heiko Stübner]

Am Freitag, 29. August 2014, 21:51:46 schrieb Doug Anderson:
> Santosh,
> 
> On Thu, Aug 28, 2014 at 2:26 PM, Santosh Shilimkar
> 
> <santosh.shilimkar at ti.com> wrote:
> > On Thursday 28 August 2014 03:36 PM, Doug Anderson wrote:
> >> These two patches add support for automatically configuring the IO
> >> voltage domains on rk3188 and rk3288 SoCs.  The first patch adds some
> >> new notification types to the regulator code.  It's used by the second
> >> patch which actually implements the IO voltage domain driver.
> >> 
> >> These two patches were co-developed by Heiko Stübner and Doug Anderson
> >> (proof of concept patches were written by Heiko).  They were tested in
> >> a private branch on an rk3288 board using rk808 instead of mainline
> >> since rk808 support isn't finalized in mainline yet.
> >> 
> >> (sorry if you got this series twice; my mailer seems unhappy with me)
> >> 
> >> Heiko Stübner (2):
> >>   regulator: core: Add REGULATOR_EVENT_PRE_VOLTAGE_CHANGE (and ABORT)
> >>   soc/rockchip: io-domain: add driver handling io domains
> > 
> > Sorry to shot down but your IO domains are nothing but voltage domains
> > and you should really build something in the drivers/power/*
> 
> If everyone agrees that this belongs in drivers/power that's totally
> OK.  Neither Heiko nor I was confident that it should be in
> drivers/soc.  I had even though that the code wouldn't be totally out
> of place in the Rockchip pinctrl driver (adding Linus W since I think
> some SoCs did add code to handle 3.3V vs. 1.8V in pinctrl).

a bit of context for Linus ...

This is essentially the continuation of  the thread "io-domain voltages as 
regulators?" from the beginning of august. After more discussions we found out 
that the io voltage selection I asked about is not an independent supply, but 
instead has to reflect the voltage of the real supplying regulator.

And setting the io-voltage setting to 1.8V while the regulator is supplying 
3.3V for example may actually damage the chip.


So in our current approach here, we added a driver that tracks voltage changes  
of the supplying regulator via a notifier and sets the register bits 
accordingly.


> 
> > Please have a look at the RFC [1]. You should really go on those
> > lines and collaborate to make a generic voltage domain layer instead of
> > throwing the driver under drivers/soc.
> 
> Trying to base things on a 7-month old RFC that hasn't been touched is
> not something I'm going to do.  Maybe that makes me a bad person...
> 
> I would also say that I'm not convinced that we really need a
> complicated framework here.  Maybe when we're talking about things
> like ABB and DevFreq and the like then having a nice framework is a
> good idea.  Really here we're just setting properties associated with
> IO pins.  There's no decisions about latency, power tradeoffs, etc.
> If the pin is connected to 1.8V we need to set the 1.8V bit.  If it's
> connected to 3.3V we need to set the 3.3V bit.  The end.
> 
> The only remotely complicated thing (and why this isn't just a pinctrl
> property) is what happens with dynamic voltages.  SD Card IO lines can
> change voltage depending on UHS negotiation.  In that case the SD Card
> Driver will request that its regulator change from 3.3V to 1.8V.  The
> bit in the IO domain register needs to update in tandem.
> 
> 
> The driver is really quite tiny (333 lines).  If we find that lots of
> people copy it and they have code that's nearly the same then we
> should try to abstract things out then.
> 
> 
> I'd be interested in hearing other opinions, though.
> 
> -Doug