[PATCH v5 1/3] ARM: mxs: add GPMI-NFC support for imx23/imx28

[u.kleine-koenig@pengutronix.de (Uwe Kleine-König)]

On Fri, Jul 08, 2011 at 12:24:22PM +0200, Lothar Wa?mann wrote:
> Uwe Kleine-K?nig writes:
> > Hello Huang,
> > 
> > On Fri, Jul 08, 2011 at 05:27:11PM +0800, Huang Shijie wrote:
> > > >>>>The init function is used only to set up iomux, so the logical replacement is
> > > >>>>a pointer to the iomux data, and calling mxs_iomux_setup_multiple_pads
> > > >>>>directly from the driver.
> > > >>>Why not put the iomux stuff into the per-machine table and get rid of
> > > >>>the init callback, too?
> > > >>The mmc (ssp) has pin conflict with gpmi on both mx23evk and mx28evk.
> > > >>So, it's better to initialize the pin when the driver(GPMI or MMC)
> > > >>is enabled.
> > > >What do you do to prevent userspace from trying to use both devices?
> > > The board can not support the two devices at the same time.
> > > So the user can only use one device with the board.
> > > 
> > > >I guess you need to configure the hardware somehow to switch between the
> > > >two using a jumper? Isn't it possible to detect the hardware setting and
> > > >setup the muxer accordingly?
> > > >
> > > >IMHO an per-device init-callback is the wrong approach to solve a pin
> > > >conflict.
> > > Do you have any good solution about this?
> > Put the pinmux corresponding to the one device that currently works in
> > the pinmux list!?
> > 
> #define 'that currently works'
> 
> For a dedicated system that may not be a problem. But for development
> kits and modular systems that allow peripheral modules to be plugged
> in there is no 'one device that currently works'.
Yeah, I know that problem. Back when I worked for a company selling
development boards I solved it with clks. Not pretty but more convenient
than kernel parameters or #ifdefs.
The upside of doing it with clks is that if $customer tries to use both
conflicting devices you get an error message instead of breaking device1
when device2 is opened/probed.
IMHO this last scenario must not happen, so I strongly object to setup
the pinmuxing in an .init callback.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-K?nig            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |