Re: [PATCH] spi: spi-geni-qcom: Use the new method of gpio CS control

[Doug Anderson <dianders@xxxxxxxxxxxx>]

Hi,

On Wed, Dec 2, 2020 at 4:47 PM Stephen Boyd <swboyd@xxxxxxxxxxxx> wrote:
>
> Quoting Stephen Boyd (2020-12-02 15:28:45)
> > Quoting Alexandru M Stan (2020-12-02 14:18:20)
> > > Unfortunately this patch makes my cros-ec (the main EC that used to
> > > work even before my debugging) also fail to probe:
> > > [    0.839533] cros-ec-spi spi6.0: EC failed to respond in time
> > > [    1.040453] cros-ec-spi spi6.0: EC failed to respond in time
> > > [    1.040852] cros-ec-spi spi6.0: Cannot identify the EC: error -110
> > > [    1.040855] cros-ec-spi spi6.0: cannot register EC, fallback to spidev
> > > [    1.040942] cros-ec-spi: probe of spi6.0 failed with error -110
> > >
> > > I wasn't closely looking at this part closely when I was using my
> > > other spi port with spidev, so this is why I haven't noticed it
> > > before.
> > > Doug suggests this might be a polarity issue. More scoping to be had.
> > >
> >
> > Ah I see. It looks like the cs-gpios polarity is wrong for the DTS on
> > sc7180. That's a patch that Doug has sent in for the qcom tree, commit
> > 37dd4b777942 ("arm64: dts: qcom: sc7180: Provide pinconf for SPI to use
> > GPIO for CS") and it is pending for the next release (v5.11). Doug says
> > he will send in a fix for the DTS side, but this patch is still "good"
> > as far as I can tell. It moves us to use gpio descriptors and also finds
> > bugs like this in the DTS file that we would have missed otherwise
> > because the legacy mode doesn't look at the polarity flags in DT.
>
> And that is wrong. With even more investigation and Doug's eagle eyes it
> seems that the cros-ec driver is overriding the spi::mode to clear out
> the SPI_CS_HIGH bit that the spi core sets in there when using the gpio
> descriptors. I'll send a patch for cros-ec-spi shortly.

So do we need any coordinating here, are we OK w/ trogdor devices
being broken for a short period of time?

I think the device tree changes switching to use GPIO for chip select
is already queued in linux-next.  That means if we land this patch
before the fix to cros_ec [1] then we'll end up in a broken state.
Would we be able to do some quick landing to get the cros-ec fix into
v5.10 and then target the SPI patch for 5.11?

-Doug

[1] https://lore.kernel.org/r/20201203011649.1405292-2-swboyd@xxxxxxxxxxxx/