[RESEND PATCH v1 03/13] dt-bindings: mfd: ti,tps6594: Add TI TPS65224 PMIC

[Conor Dooley]

On Wed, Feb 14, 2024 at 09:26:13AM -0800, Kevin Hilman wrote:
> Conor Dooley <conor at kernel.org> writes:
> > On Wed, Feb 14, 2024 at 03:01:06PM +0530, Bhargav Raviprakash wrote:
> >> On Fri 2/9/2024 10:41 PM, Conor Dooley wrote:
> >> > On Thu, Feb 08, 2024 at 04:23:33PM +0530, Bhargav Raviprakash wrote:
> >> > > TPS65224 is a Power Management IC with 4 Buck regulators and 3 LDO
> >> > > regulators, it includes additional features like GPIOs, watchdog, ESMs
> >> > > (Error Signal Monitor), and PFSM (Pre-configurable Finite State Machine)
> >> > > managing the state of the device.
> >> > 
> >> > > TPS6594 and TPS65224 have significant functional overlap.
> >> > 
> >> > What does "significant functional overlap" mean? Does one implement a
> >> > compatible subset of the other? I assume the answer is no, given there
> >> > seems to be some core looking registers at different addresses.
> >> 
> >> The intention behind “significant functional overlap” was meant to
> >> indicate a lot of the features between TPS6594 and TPS65224 overlap,
> >> while there are some features specific to TPS65224.
> >> There is compatibility between the PMIC register maps, I2C, PFSM,
> >> and other drivers even though there are some core registers at
> >> different addresses.
> >> 
> >> Would it be more appropriate to say the 2 devices are compatible and have
> >> sufficient feature overlap rather than significant functional overlap?
> >
> > If core registers are at different addresses, then it is unlikely that
> > these devices are compatible.
> 
> That's not necessarily true.  Hardware designers can sometimes be
> creative. :)

Hence "unlikely" in my mail :)

> > In this context, compatible means that existing software intended for
> > the 6594 would run without modification on the 65224, although maybe
> > only supporting a subset of features.  If that's not the case, then
> > the devices are not compatible.
> 
> Compatible is a fuzzy term... so we need to get into the gray area.
> 
> What's going on here is that this new part is derivative in many
> signifcant (but not all) ways from an existing similar part.  When
> writing drivers for new, derivative parts, there's always a choice
> between 1) extending the existing driver (using a new compatible string
> & match table for the diffs) or 2) creating a new driver which will have
> a bunch of duplicated code.
> 
> The first verion of this series[1] took the 2nd approach, but due to the
> significant functional (and feature) overlap, the recommendation was
> instead to take the "reuse" path to avoid signficant amounts of
> duplicated code.
> 
> Of course, it's possible that while going down the "reuse" path, there
> may be a point where creating a separate driver for some aspects might
> make sense, but that needs to be justified.  Based on a quick glance of
> what I see in this series so far (I have not done a detailed review),
> the differences with the new device look to me like they can be handled
> with chip-specific data in a match table.

This is all nice information, but not really relevant here - this is a
binding patch, not a driver one & the conversation stemmed from me
making sure that a fallback compatible was not suitable. Whether or not
there are multiple drivers or not is someone else's problem!

Thanks,
Conor.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20240214/65b10c84/attachment.sig>