[Gta04-owner] Whose GTA04 think that battery is removed? Please report.

[H. Nikolaus Schaller]

> Am 03.05.2017 um 17:15 schrieb Andreas Kemnade <andreas at kemnade.info>:
> 
> Hi,
> 
> On Wed, 3 May 2017 13:05:59 +0200
> "H. Nikolaus Schaller" <hns at goldelico.com> wrote:
> 
>> 
>>> Am 29.04.2017 um 21:03 schrieb Andreas Kemnade <andreas at kemnade.info>:
>>> 
>>> Hi,
>>> 
>>> my GTA04A5 thinks its battery is disconnected when it is suspended
>>> while it is charging.
>>> The Letux 3704 does not do that. Same kernels!
>>> 
>>> This phenomenom does only happen on the gta04a5 when the bit
>>> WATCHDOGEN in BCIMFEN4 is enabled.
>>> 
>>> The bit is enabled by default. I am wondering whether this is a problem
>>> related to the GTA04A5 hardware or my device.
>>> 
>>> Battery "disconnection" leads to immediate wakeup and charging stop.
>>> 
>>> Or the battery of the Letux3704 has stronger ties to the device so that
>>> it cannot be disconnected by the kernel ;-)
>>> 
>>> Hmm, newer chip revision of the tps65950?
>> 
>> I am not aware of. We used the same A3 variant. And the only significant change
>> in the charging circuits is to use a different transistor. But that should not
>> end in a battery-disconnect detection (maybe we should try to find out what
>> situation does trigger that).
>> 
> Well, as far as I understand even the bci does not detect a battery
> disconnect directly. I disabled the corresponding bit in BCIMFEN4.
> (BTW: There are some restrictions writing to that register, so read
> back to check!) Still a battery "disconnection"
> clearing WATCHDOGEN in that register inhibits that disconnection.
> For some unknown reason that watchdog triggers (somehow trouble with
> MADC). It resets the state machine of the charger. And as a side effect
> there seems to be that disconnect.
> 
> So the key point is understanding
> a) that watchdog.
> b) access arbitation to madc.
> - maybe some measurement is started and not finished
>   when it is suspended, but why this happens only on a5
>   not a4?
> c) madc power management
> - madc is enabled once but seems not to be disabled on suspend?
> - some other resource is disabled but needed by madc,

We have the generic-adc-battery driver as well and it uses madc.

I have put Marek on CC.

> but again why
>   only on a5?

Maybe some race condition (RAM faster/slower)?

> 
> I even have a workaround patch in my tree which clears that WATCHDOGEN,
> but I have not found a good place in code to do that because it is
> enabled again at some unknown time. So it does not work properly.
> 
> Regards,
> Andreas

BR,
Nikolaus