Re: 2.6.33-rc3 regression: leds_alix2: can't allocate I/O for GPIO

[Daniel Mack <daniel@caiaq.de>]

Hi,

On Thu, Jan 07, 2010 at 03:00:37PM +0100, Arnd Hannemann wrote:
> Daniel Mack schrieb:
> > On Thu, Jan 07, 2010 at 01:11:31PM +0100, Arnd Hannemann wrote:
> >> Arnd Hannemann schrieb:
> >>> Richard Purdie schrieb:
> >>>> On Thu, 2010-01-07 at 11:26 +0100, Arnd Hannemann wrote:
> >>>>> with 2.6.33-rc3 the LEDs are not working anymore on my Geode embedded board:
> >>>>> If I remember correctly this was working fine with 2.6.32, but certainly with 2.6.31
> >>>>>
> >>>>> relevant dmesg output of 2.6.33-rc3:
> >>>>>
> >>>>> [    0.709643] cs5535_gpio: base=0x6100 mask=0xb003c66 major=253
> >>>>>
> >>>>> [    2.207947] leds_alix2: system is recognized as "PC Engines ALIX.2 v0.99h"
> >>>>> [    2.228678] leds_alix2: can't allocate I/O for GPIO
> >>>>>
> >>>>>
> >>>>> For reference dmesg of 2.6.31.3:
> >>>>>
> >>>>> [    0.611953] cs5535_gpio: base=0x6100 mask=0xb003c66 major=253
> >>>>>
> >>>>> [    2.379114] leds_alix2: system is recognized as "PC Engines ALIX.2"
> >>>>> [    2.399097] Registered led device: alix:1
> >>>>> [    2.411404] Registered led device: alix:2
> >>>>> [    2.423720] Registered led device: alix:3
> >>>> Its likely to be commit 7f131cf3ed96c969d7b092bf629e25c3df50901e. It
> >>>> looks like the request_region() call is failing. Daniel, any idea why?
> >>> Some additional debug information:
> >>>
> >>> elara:~# rdmsr 0x5140000C
> >>> f00100006100
> >>>
> >>> /proc/ioports says:
> >>> 6100-61ff : 0000:00:0f.0
> >>>   6100-61ff : cs5535_gpio
> >>>
> >>> BTW: why is request_region() needed there after all?
> >> Oh, maybe using cs5535_gpio AND leds_alix2 together was never supposed to work?
> > 
> > That is the problem indeed. The problem is that three of these GPIOs are
> > connected to LEDs, so if both drivers are enabled, they will both alter
> > the same GPIOs. Hmm, don't know how to handle this. Maybe the LED driver
> > should use the GPIO interface and that's it?
> 
> I'm not sure that this configuration needs to be supported at all,
> so disallowing that both drivers are configured via Kconfig could be a simple solution.

True.

> If I understand correctly, for off-the-shelf ALIX2 boards, beside the LEDs,
> there is only one other GPIO pin being usable (the "mode switch"), anyway.

Yes. I also wrote a simple polled input device driver to support this
switch, but it was too much of a hack as it ended up being part of
the LED driver. The idea discussed back then was to clean up the code
so that the entire MSR lookup thing is only done once, and a platform
support for ALIX board then adds support for the LEDs via the leds-gpio
LED driver and for the button via gpio-buttons input driver. That would
be much nicer, but I haven't found the time yet to implement it.

Daniel