Re: [PATCH 01/10] GPIO: gpio-generic: Move initialization up to postcore

[Tony Lindgren <tony@atomide.com>]

* Janusz Krzysztofik <jkrzyszt@tis.icnet.pl> [111214 11:36]:
> (adding Mark Brown and Liam Girdwood - regulator experts - to Cc:)
> 
> On Wednesday 14 of December 2011 at 19:21:49, Tony Lindgren wrote:
> > * Janusz Krzysztofik <jkrzyszt@tis.icnet.pl> [111214 04:40]:
> > > On Tuesday 13 of December 2011 at 00:55:44, Tony Lindgren wrote:
> > > > * Janusz Krzysztofik <jkrzyszt@tis.icnet.pl> [111212 15:13]:
> > > > > On Tuesday 13 of December 2011 at 00:15:20, Tony Lindgren wrote:
> > > > > > 
> > > > > > Might be worth checking if some board specific __initcall helps here
> > > > > > too?
> > > > > 
> > > > > If I only knew how I could insert a board specific __initcall between 
> > > > > two points from where the generic-gpio first, then the 8250 driver, are 
> > > > > called.
> > > > > 
> > > > > Any hints?
> > > > 
> > > > Hmm, can't you do all that in the order you want in
> > > > ams_delta_modem_init()?  Or make that into a late_initcall so
> > > > you have generic-gpio available?
> > > > 
> > > > It seems that the pieces of code you're talking about don't need
> > > > to be initialized early, just needs to be done in the right
> > > > order to get things working.
> > > 
> > > Hi,
> > > I'm almost done with moving registration of all latch dependent devices 
> > > down to a late_initcall hook, however while working on this, I've found 
> > > still another arrangement, yet better in my opinion:
> > > 1) generic-gpio driver registration moved from device_initcall up to 
> > >    subsys_initcall,
> > > 2) latch dependent device registration left at arch_initcall, as it is 
> > >    now,
> > > 3) a temporary hack, removed with the last patch in the series, that 
> > >    requests GPIO pins on behalf of device drivers before those are 
> > >    updated, placed between subsys_initcall and device_initcall, i.e., at 
> > >    fs_initcall or rootfs_initcall; both look ugly, but this is only for 
> > >    a while, in order to keep things working while in the transition,
> > > 4) the modem init hook, once updated with extra GPIO setup that must be 
> > >    done on behalf of the 8250 driver, which is not prepared for 
> > >    accepting any extra init hooks passed with the device platform data, 
> > >    moved down to late_initcall, as suggested,
> > > 5) once all drivers are updated, the hack is removed, and an 
> > >    initialization of unused pins added to that late_initcall modem hook, 
> > >    perhaps renamed in order to not suggest it is still modem only 
> > >    related.
> > > 
> > > What do you think?
> > 
> > Sounds better for sure than what we currently have :)
> 
> Hmm, better doesn't necessarily mean good enough...
> 
> I forgot about the sound device, which shares its latch based GPIO pins 
> with the modem, so should be registered after the modem.
> 
> To make things still more complicated, one of those GPIO pins provides 
> power to both devices, so a still better solution I'd like to introduce 
> would be a GPIO controlled regulator device which feeds both the modem 
> and the sound card with power.
> 
> I've had a look at both generic regulator drivers, fixed.c and gpio-
> regulator.c, both controlling devices over GPIO pins, and found those 
> drivers registered at subsys_initcall level, calling gpio_request*() at 
> probe time. Then again, we would need that base_mmio_gpio driver already 
> registered before subsys_initcall.
> 
> Any suggestions?

Sounds like yet another case for the deferred probe patches posted
few weeks ago?

Regards,

Tony