Re: [Upstream] [PATCH 0/2] i.MX93 ADC calibration settings

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Jonathan,

On 25. 03. 24 15:45, Jonathan Cameron wrote:
> On Mon, 25 Mar 2024 09:55:23 +0100
> Primoz Fiser <primoz.fiser@xxxxxxxxx> wrote:
> 
>> Hi Jonathan,
>>
>> On 25. 03. 24 09:32, Andrej Picej wrote:
>>> Hi Jonathan,
>>>
>>> On 24. 03. 24 14:55, Jonathan Cameron wrote:  
>>>> On Wed, 20 Mar 2024 11:04:04 +0100
>>>> Andrej Picej <andrej.picej@xxxxxxxxx> wrote:
>>>>  
>>>>> Hi all,
>>>>>
>>>>> we had some problems with failing ADC calibration on the i.MX93 boards.
>>>>> Changing default calibration settings fixed this. The board where this
>>>>> patches are useful is not yet upstream but will be soon (hopefully).  
>>>>
>>>> Tell us more.  My initial instinct is that this shouldn't be board
>>>> specific.
>>>> What's the trade off we are making here?  Time vs precision of
>>>> calibration or
>>>> something else?  If these are set to a level by default that doesn't work
>>>> for our board, maybe we should just change them for all devices?
>>>>  
>>
>> The imx93_adc driver is quite new.
>>
>> If you look at line #162, you will find a comment by the original author:
>>
>>> 	/*
>>> 	 * TODO: we use the default TSAMP/NRSMPL/AVGEN in MCR,
>>> 	 * can add the setting of these bit if need in future.
>>> 	 */  
>>
>> URL:
>> https://github.com/torvalds/linux/blob/master/drivers/iio/adc/imx93_adc.c#L162
>>
>> So, for most use-cases the default setting should work, but why not make
>> them configurable?
>>
>> So this patch-series just implement what was missing from the beginning
>> / was planned for later.
> Hi Primoz,
> 
> I doubt anyone reviewed the comment closely enough to say if what it was
> suggesting was sensible or not, so the fact it was listed as a todo
> doesn't directly impact this discussion.

I agree.

However on the other hand, since we stumbled upon a use-case that
requires adjusting the driver provided default settings of the i.MX93
ADC, this TODO to us is and was a clear indication from the original
author that the driver needs little TLC.

Anyhow, a stance from the author/NXP would be highly welcoming in this
situation.

BR,
Primoz


> 
>>
>> BR,
>> Primoz
>>
>>
>>>
>>> So we have two different boards with the same SoC. On one, the
>>> calibration works with the default values, on the second one the
>>> calibration fails, which makes the ADC unusable. What the ADC lines
>>> measure differ between the boards though. But the implementation is
>>> nothing out of the ordinary.
>>>
>>> We tried different things but the only thing that helped is to use
>>> different calibration properties. We tried deferring the probe and
>>> calibration until later boot and after boot, but it did not help.
>>>
>>> In the Reference Manual [1] (chapter 72.5.1) it is written:
>>>   
>>>> 4. Configure desired calibration settings (default values kept for
>>>> highest accuracy maximum time).  
>>>
>>> So your assumption is correct, longer calibration time (more averaging
>>> samples) -> higher precision. The default values go for a high accuracy.
>>> And since we use a NRSMPL (Number of Averaging Samples) of 32 instead of
>>> default 512, we reduce the accuracy so the calibration values pass the
>>> internal defined limits.
> 
> Ouch.  Let me try to dig into this. Is this effectively relaxing the
> constraints? I guess because a value that is perhaps always biased one way
> is considered within bounds if those acceptable bounds are wider because
> of lower precision?
> 
> I was assuming it was the other way around and the device had fixed constraint
> limits and you needed to take more samples due to higher noise. Seems the
> opposite is true here and that worries me.
> 
> I'll definitely need input from NXP on this as a workaround and their
> strong support to consider it.
> 
>>>
>>> I'm not sure that changing default values is the right solution here. We
>>> saw default values work with one of the boards. And since the NXP kept
>>> these values adjustable I think there is a reason behind it.
> 
> I'd assume trade off between time and calibration precision, not the
> sort of use I think you are describing.
> 
>>>
>>> Note: When I say one of the boards I mean one board form. So same board
>>> version, but different HW.
> 
> Superficially I'm struggling to not see this as broken hardware that it
> is out of expected tolerances in some fashion.  Maybe I misunderstood
> the issue.
> 
> Jonathan
> 
>>>
>>> Best regards,
>>> Andrej
>>>
>>> [1] i.MX 93 Applications Processor Reference Manual, Rev. 4, 12/2023
>>> _______________________________________________
>>> upstream mailing list
>>> upstream@xxxxxxxxxxxxxxx
>>> http://lists.phytec.de/cgi-bin/mailman/listinfo/upstream  
>>
>>
>> _______________________________________________
>> linux-arm-kernel mailing list
>> linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
>> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
> 

-- 
Primoz Fiser                    | phone: +386-41-390-545
<tel:+386-41-390-545> |
---------------------------------------------------------|
Norik systems d.o.o.            | https://www.norik.com
<https://www.norik.com>  |
Your embedded software partner  | email: info@xxxxxxxxx
<mailto:info@xxxxxxxxx> |
Slovenia, EU                    | phone: +386-41-540-545
<tel:+386-41-540-545> |





[Index of Archives]     [Device Tree Compilter]     [Device Tree Spec]     [Linux Driver Backports]     [Video for Linux]     [Linux USB Devel]     [Linux PCI Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Yosemite Backpacking]


  Powered by Linux