>> The main thing I'm missing with this is a coherent explanation of the
>> problem and how the changes proposed fix it.
> 
> Just to emphasize: the main concern here is that the issue is understood
> and that it's not just going to pop up again as soon as something
> changes.

Here are more details Mark.

<1. finish machine driver probe >
[  115.253970] bdw-rt5677 bdw-rt5677: rt5677-dspbuffer <-> 
spi-RT5677AA:00 mapping ok

<2. execute BE dailink .init() and request a gpiod from the codec 
component device>

[  115.254387] rt5677 i2c-RT5677CE:00: plb devm_gpiod_get
[  115.254390] rt5677 i2c-RT5677CE:00: GPIO lookup for consumer 
headphone-enable
[  115.254391] rt5677 i2c-RT5677CE:00: using ACPI for GPIO lookup
[  115.254395] acpi RT5677CE:00: GPIO: looking up headphone-enable-gpios
[  115.254399] acpi RT5677CE:00: GPIO: _DSD returned RT5677CE:00 4 0 0
[  115.254724] rt5677 i2c-RT5677CE:00: GPIO lookup for consumer plug-det
[  115.254725] rt5677 i2c-RT5677CE:00: using ACPI for GPIO lookup
[  115.254727] acpi RT5677CE:00: GPIO: looking up plug-det-gpios
[  115.254729] acpi RT5677CE:00: GPIO: _DSD returned RT5677CE:00 0 0 0
[  115.255451] rt5677 i2c-RT5677CE:00: GPIO lookup for consumer mic-present
[  115.255453] rt5677 i2c-RT5677CE:00: using ACPI for GPIO lookup
[  115.255455] acpi RT5677CE:00: GPIO: looking up mic-present-gpios
[  115.255458] acpi RT5677CE:00: GPIO: _DSD returned RT5677CE:00 1 0 0

<3. gpiod handling complete>

[  115.256293] bdw-rt5677 bdw-rt5677: rt5677-aif1 <-> ssp0-port mapping ok

<4. jack handling complete>
[  115.262040] input: sof-bdw-rt5677 Headphone Jack as 
/devices/pci0000:00/INT3438:00/bdw-rt5677/sound/card1/input11
[  115.262240] input: sof-bdw-rt5677 Mic Jack as 
/devices/pci0000:00/INT3438:00/bdw-rt5677/sound/card1/input12

<5. card fully functional>

<6. rmmod snd_sof-acpi>

<7. rmmod machine driver>

<8. rmmod codec driver>
rmmod: ERROR: Module snd_soc_rt5677 is in use

<9. rmmod -f codec driver>
[  194.118221] gpio gpiochip0: REMOVING GPIOCHIP WITH GPIOS STILL REQUESTED
[  194.118440] rt5677 i2c-RT5677CE:00: plb devm_gpiod_release

So this is a self-inflicted deadlock - broken by design.

When the machine driver is removed, the gpiod is not freed.
Only removing the codec driver can free the gpiod, but the gpio/module 
refcount prevents the codec driver from being removed.

I don't know how to move all the gpio handling in the codec driver, 
since there are platform-dependent ACPI mappings.

The only proposal I can make is to avoid using devm_ but we need a hook 
to call gpiod_put().

using the add_dai_link()/remove_dai_link as I suggested earlier is not 
possible since the runtime is created after this callback is involved.

The proposal suggested by Andy was to have a dual callback to the 
init(), or as in my initial version to call gpiod_put() in the machine 
driver .remove() function, which isn't very elegant but does work.

I also tested a different solution (attached) based on your input where 
the gpiod handing is performed in the machine driver probe, after the 
card registration, and the gpiod_put() called from remove. This is 
simple enough but there might be some issues left with the jack/input 
handling - not sure why the logs for jacks are missing.

Does this clarify the issue and options?

Thanks
-Pierre