Re: [RFC 7/8] cpufreq: imx6q: Initialize LDO bypass

[Lucas Stach]

Am Mittwoch, den 22.03.2017, 19:48 +0200 schrieb Leonard Crestez:
> On Wed, 2017-03-22 at 18:09 +0100, Lucas Stach wrote:
> > Am Mittwoch, den 22.03.2017, 18:53 +0200 schrieb Leonard Crestez:
> > > 
> > > Several imx6* socs have three built-in regulators LDO_ARM LDO_SOC and
> > > LDO_PU used to control internal chip voltages. "ldo-bypass" mode refers
> > > to placing these regulators in bypass mode and controlling voltages from
> > > an external power management chip instead. This is intended to save
> > > power at the expense of an extra PMIC on the board.
> > > 
> > > The voltages for these regulators are controlled from the imxq6 cpufreq
> > > driver so it makes sense to also control LDO bypass from here. The gpc
> > > driver also fetches a reference to LDO_PU and uses it to gate power to
> > > graphics blocks.
> > > 
> > > The LDO regulator has a minimum dropout voltage of 125mv so enabling
> > > bypass mode will raise the effective voltage by that amount. We need set
> > > the minimum frequency first to avoid overvolting when enabling LDO
> > > bypass.
> > > 
> > > The binding is an u32 fsl,ldo-bypass in the gpc node because that's how
> > > it was defined in the freescale vendor tree for a long time and
> > > compatibility is desirable. Otherwise it would be a bool.
> > > 
> > > Some versions of u-boot shipped by freescale check this same property
> > > and set regulators in bypass mode before linux actually starts so we
> > > check for that scenario as well and finish early.
> > I've not looked at the patch at all, but this feels like the wrong
> > location to implement this. Using bypass mode or not should really be a
> > internal decision of the regulator driver, influenced by a DT property
> > to allow bypass mode.
> > 
> > The regulator driver can also implement the correct sequencing of first
> > lowering external voltage to min + dropout, then going into bypass mode,
> > then lower the external voltage by the amount of the dropout. I don't
> > see why we need a frequency switch for this at all.
> 
> Because minimum voltages are dictated by core frequency. At high frequency
> the (minimum voltage for frequency + dropout) is too high and would go beyond
> the maximum of 1300 mv when bypass is enabled. It doesn't actually instantly
> break, this is based on the "operating ranges" from this document:
> 
> http://www.nxp.com/assets/documents/data/en/data-sheets/IMX6DQCEC.pdf

Urgh, yes, this is unfortunate.

> > Implementing this in the consumers seems like the wrong spot.
> 
> It doesn't belong in drivers for individual regulators either, it's a piece
> of board-level global configuration.

The configuration weather to use the bypass or not is board level, I
agree. But the decision to bypass the regulator should be done either at
the regulator driver level or even in the regulator framework. That is
the point where you are able to match up the constraints of the
consumers with the ability to get the correct voltage from the supplying
regulator.

Regards,
Lucas