VIM3 - A311D - I2C AO Clock Questions

[Lucas Tanure]

On Wed, Mar 9, 2022 at 9:39 AM Lucas Tanure <tanure at linux.com> wrote:
>
> On Tue, Mar 8, 2022 at 6:49 AM Lucas Tanure <tanure at linux.com> wrote:
> >
> > On Mon, Mar 7, 2022 at 8:29 AM Neil Armstrong <narmstrong at baylibre.com> wrote:
> > >
> > > Hi,
> > >
> > > On 06/03/2022 11:23, Lucas Tanure wrote:
> > > > Hi,
> > > >
> > > > I am trying to integrate an I2C device into VIM3 using the I2C_AO bus.
> > > > My device uses a 100000Hz clock, but after changing
> > > > meson-g12b-a311d-khadas-vim3.dts to add my I2C node the driver could
> > > > not properly read its registers.
> > > >
> > > > It reads 0x9100, but the correct data should be 0xcb58.
> > > > The first thing I saw on the bus was that the frequency I requested by
> > > > the device tree was being set after the probe of my driver already
> > > > failed.
> > > > For that, I sent "i2c: meson: Fix wrong speed use from probe" which
> > > > fixes that issue.
> > > >
> > > > But I still get the wrong data, so my next guess is that the clock
> > > > duty cycle is 30%, and I think should be 50%.
> > > > Here is an image of the signals of the bus:
> > > > https://pasteboard.co/TEoxQucsXUUD.png
> > > >
> > > > I have tested other frequencies and all of them uses a 30% duty cycle,
> > > > where SCL is up 30% and low for 70% of the period.
> > >
> > > Indeed it seems by default the duty cycle is for Fast Mode, but standard mode should
> > > be around 50%. Did you try 400khz ?
> >
> > Yes, 400Khz has the same issue of 30% duty cycle.
> >
> > >
> > > Amlogic did some changes on their off-tree linux fork, please look around:
> > > https://github.com/khadas/linux/blob/khadas-vims-5.4.y-fc/drivers/i2c/busses/i2c-meson.c#L143
> >
> > Thanks! I am investigating that code, but so far doesn't work.
> >
> > >
> > > I'm unsure when REG_SLAVE_ADDR BIT(28) appeared, but if you try to port this upstream,
> > > we'll be happy to review it for inclusion.
> >
> > I will try to get this working and push a patch.
> > Thanks
> > Lucas
> >
> > >
> > > Neil
> > >
> > > >
> > > > Is possible to configure the clock to be 100000Hz, but with a 50% duty
> > > > cycle for this I2C bus on A3111D?
> > > >
> > > > Thanks
> > > > Lucas Tanure
> > > >
> > > > _______________________________________________
> > > > linux-amlogic mailing list
> > > > linux-amlogic at lists.infradead.org
> > > > http://lists.infradead.org/mailman/listinfo/linux-amlogic
>
> After some debugging, I got the 50% duty cycle working, but that was
> not the issue.
> The issue was that my device address is 0x22, the same as FUSB302B.
>
> But FUSB302B is not on the device tree, so I didn't there was an
> address collision.
> I will push some patches for the duty cycle later this week.
>
> Thanks
> Lucas Tanure

Hi Jian Hu,

You sent the patch "i2c: meson: add configurable divider factors",
where it sets the div factor for Meson-AXG soc to be 3.
The current state of the clock is a 30% duty cycle, and I would like
to port a 50% duty cycle from Khadas 5.4 driver.

https://github.com/khadas/linux/blob/khadas-vims-5.4.y-fc/drivers/i2c/busses/i2c-meson.c#L143

The problem now is that div_factor must be 4 for VIM3, as the value of
3 (current one) the clock is set wrongly.
The value of 3, plus the 50% patch sets the clock to be 85k instead of
100k, and the duty of 42%.

So, why div_factor is 3 for Meson-AXG?
Can we remove that patch "i2c: meson: add configurable divider
factors" and set div_factor to be 4 for all?

And it seems that for devices at I2C3, i2c at 1c000, the Meson I2C driver
must be built as a module.
If I build the Meson I2C driver as built-in my I2C devices fail to
probe, but works if Meson I2C is a module that probes after 10 seconds
or so.
These 2 devices boot fine at i2c_AO, i2c at 5000, even if Meson I2C is built-in.
But because of an address clash with FUSB302B a 3rd device, I need to
move all of them to I2C3.

Is there any requirement for I2C3 to work during boot? Any power or
clock that needs to be configured before I2C can be used?

Thanks
Lucas