Re: [RFC] i2c-stub: make it usable with DT when booting

[Wolfram Sang]

Hi Jean,

> I get the idea and I understand the need to emulate the devices early
> in the boot process. However, how do you deal with the I2C devices
> pre-initialization (setting initial register values)? This is typically
> done in user-space with the i2c-stub-from-dump script, or manual calls
> to i2cset, but that would be too late in your case, wouldn't it?

Yes. In my case, all 0 memory is fine. I needed to check writes which
were enabling "channels" allowing me to do further initialization.

I see the point, though, that we likely want to address your question
before going upstream.

> Code looks sane, but I can't comment on the bindings side of things as
> this isn't my area.

Sure, thanks for checking the code!

Regards,

   Wolfram



signature.asc
Description: PGP signature

Re: [RFC] i2c-stub: make it usable with DT when booting

[Jean Delvare]

Hi Wolfram,

On Tue, 16 May 2017 13:27:58 +0200, Wolfram Sang wrote:
> This patch makes the stub driver parse the device tree when booting and
> create a virtual bus with the desired devices attached. Here is an
> example DTS snipplet making use of it. It is for simulating a more
> complex camera device which has dependencies which needs to be sorted
> out at boot time:
> 
> i2c@42 {
>   compatible = "i2c-stub";
> 
>   #address-cells = <1>;
>   #size-cells = <0>;
> 
>   des@4c {
>   compatible = "maxim,max9268";
>   reg = <0x4c>;
>   };
> 
>   eeprom@57 {
>   compatible = "atmel,24c02";
>   reg = <0x57>;
>   };
> };

I get the idea and I understand the need to emulate the devices early
in the boot process. However, how do you deal with the I2C devices
pre-initialization (setting initial register values)? This is typically
done in user-space with the i2c-stub-from-dump script, or manual calls
to i2cset, but that would be too late in your case, wouldn't it?

> 
> FIXME: can fw_* calls be used instead of of_*?
> 
> Signed-off-by: Wolfram Sang 
> ---
> 
> It was really useful when developing. When using fw_* calls, we probably could
> make use of it with ACPI as well. Still, calling for opinions if we want this
> upstream or if we want to stay module-only? Also, the binding should really
> be marked as "development only". Or is it better to keep it as a separate
> patch to keep the binding un-official?

Code looks sane, but I can't comment on the bindings side of things as
this isn't my area.

-- 
Jean Delvare
SUSE L3 Support