U-Boot sunxi DM_I2C migration?

[Samuel Holland]

Hi all,

On 8/17/21 2:23 PM, Tom Rini wrote:
> On Thu, Aug 12, 2021 at 05:40:34PM -0400, Tom Rini wrote:
>> On Thu, Aug 12, 2021 at 10:22:07PM +0100, Andre Przywara wrote:
>>> On Thu, 12 Aug 2021 13:49:30 -0400
>>> Tom Rini <trini at konsulko.com> wrote:
>>>
>>> Hi Tom,
>>>
>>>> As I try and migrate more CONFIG symbols to be Kconfig only, I see that
>>>> sunxi is still using the legacy I2C drivers.  Is there something
>>>> specific holding up migration?  Thanks!
>>>
>>> Do you mean a DM_I2C conversion, or just moving those base address
>>> symbols to Kconfig? The latter is rather easy, let me know if I should
>>> give it a shot.
>>>
>>> For DM_I2C, it's the usual sunxi issue: we are using both: DM_I2C and
>>> DT based probing for U-Boot proper, and hardcoded CONFIG_ symbols for
>>> the SPL. *Some* boards need the set up the PMIC early (for increasing
>>> the CPU frequency or setting the proper DRAM voltage), so this needs to
>>> be done in the SPL. I don't know how to fix this "properly",
>>> but introducing DM (and DT) into the SPL is surely not worth the
>>> trouble. Alternatively we could have a separate, cut-down SPL-only
>>> driver, like we have for SPI (arch/arm/mach-sunxi/spl_spi_sunxi.c),
>>> but not sure that's really better.
>>>
>>> If there is something in particular that annoys you about the
>>> situation, we could try to improve this particular issue instead?
>>> Moving the non-DM code into a separate file, maybe?
>>
>> I'm working on the symbols now (since it gets tricky).  The first
>> example I pulled out was Hummingbird_A31 which does have legacy I2C and
>> does not set DM_I2C, which is what set me down the more worrying path.
>> If you can confirm that no, really, it's just SPL that's not using
>> DM_I2C that's OK enough for now, we can sort out what the best but
>> probably still unideal solution is long term for SPL.
> 
> As I'm converting the symbols to Kconfig now (and adding
> SPL_SYS_I2C_LEGACY), what I'm seeing on sunxi is a whole lot of platform
> enabling the legacy driver in full U-Boot and not DM_I2C.  Now it's
> entirely possible the follow up solution really is just to switch to
> DM_I2C (and continue SPL_SYS_I2C_LEGACY) and it would be a fairly
> trivial switch.  But I'm not in a good position to make that change and
> test it, so I'm asking you to please, once I've posted this series to
> add SPL_SYS_I2C_LEGACY

I am aware of three places where sunxi boards use legacy I2C in U-Boot proper:
 - The i2c_soft usage enabled by CONFIG_VIDEO_LCD_PANEL_I2C
 - The axp_gpio driver (through pmic_bus_*)
 - do_poweroff() in drivers/power/axp*.c (through pmic_bus_write)

I have a patch series for converting all of these that I will be sending within
the next few days. The two tricky parts are:
 - I have no way to test either board that uses VIDEO_LCD_PANEL_I2C,
   so I do not want to rely on bus number ordering.
 - I am trying to arrange changes in an order that never breaks any
   functionality, so some driver must be temporarily compatible with
   both legacy and DM I2C.

I will rebase this on top of the Kconfig series before sending it.

Regards,
Samuel