More GPIO madness on iMX6 - and the crappy ARM port of Linux

[Eric Nelson]

On 01/17/2014 01:33 PM, Arnaud Patard (Rtp) wrote:
> Eric Nelson <eric.nelson at boundarydevices.com> writes:
>
>> On 01/17/2014 12:57 PM, Arnaud Patard (Rtp) wrote:
>>> Russell King - ARM Linux <linux at arm.linux.org.uk> writes:
>>>
>>>> So, we have this wonderful GPIO layer which abstracts GPIO stuff and
>>>> hides stuff.  It's really wonderful, because you don't have to care
>>>> about how the GPIOs are actually accessed in drivers anymore.
>>>>
>>>> However, what about the behaviour of GPIOs?
>>>>
>>>> What about... for example... this sequence:
>>>>
>>>> 	gpio_direction_output(gpio, 1);
>>>> 	val = gpio_get_value(gpio);
>>>>
>>>> What value is "val"?  More importantly, what value is reflected in
>>>> /sys/kernel/debug/gpio ?  Would it indicate that it's high or low?
>>>>
>>>> Now, while you can make reasonable assumptions, such as "it'll return
>>>> that the output is being driven to the requested state" or "it'll
>>>> return the actual state of the pin", what about this instead, which
>>>> happens on iMX hardware - "it'll _always_ return zero".
>>>>
>>>
>>> this is "expected". gpio layer docs are saying that in output case, the
>>> value may be wrong. Not intuitive but documented.
>>>
>>>> Yes, iMX6 at least has this behaviour.  For any output, val as above
>>>> will always be zero, and /proc/sys/kernel/debug/gpio will always
>>>> report that an output is zero... unless the SION bit has been set for
>>>> that GPIO signal.
>>>
>>> afaik at least imx51/53 have some behaviour.
>>>
>>>>
>>>> The reason is that on hardware such as iMX6, reading the GPIO is done
>>>> by reading the pad state register, and this register is _only_ supplied
>>>> the state of the pad when the input path is enabled.  The input path
>>>> is only enabled when the output is disabled, or the SION bit is set
>>>> to force the GPIO input path.
>>>
>>> I sent mails about this same issue for imx51 in Dec 2010 and answer were
>>> that the SION bit should not be set for all gpios:
>>> http://permalink.gmane.org/gmane.linux.ports.arm.kernel/100875
>>>
>>
>> Thanks Arnaud,
>>
>> This bit from the 2010 chain really needs some explanation:
>>
>>>> Arnaud Patard (Rtp) writes:
>>>>
>>>> <snip>
>>>>
>>>> I had done the same, but had some trouble with this.
>>>> E.g. on our board GPIO1_7 is used as a generic GPIO to enable an
>>>> external clock oscillator for the USBH1 ULPI PHY. When the SION bit
>>>> for this pad was set, I got strange errors on the USBH1 port
>>>> (disconnecting low speed devices behind a hub would stall the
>>>> bus). When I removed the SION bit for that pin everything worked
>>>> well.
>>>>
>>
>> Did you ever chase down the symptom here? Was the GPIO output not
>> holding a constant value such that the oscillator wasn't functioning?
>>
>
> It was not me who got this issue. The issue I had was not being able to
> read GPIO value if set as output. This link may be clearer:
> http://archive.arm.linux.org.uk/lurker/message/20101215.151016.ea731aa7.en.html
>
Thanks Arnaud. That link is quite a bit easier to follow.

I'm looping in Lothar and Dinh.

Dinh's comment in this thread makes no sense:

	>>
	>> The SION bit is a "Software Input On" bit. Basically, if
	>> you set the GPIO as an output, you cannot set the SION bit.
	>>

If that were the case, there's no point to either the bit or the PSR
registers. Dinh, can you or another Freescaler give a definitive
explanation here?

We also had some discussion about this on the U-Boot list regarding

Patch:      http://lists.denx.de/pipermail/u-boot/2013-September/163805.html
Discussion: 
http://lists.denx.de/pipermail/u-boot/2013-October/thread.html#163916

In the U-Boot case, we decided to flag those pins used as both
input and output specfically with the SION flag, which has the
benefit of pointing out to a reader that there's something unusual
going on.

Regards,


Eric