On 2016-09-27 19:00, Viresh Kumar wrote:

On 27-09-16, 12:34, Stefan Agner wrote:

quoted

Added Viresh Kumar to the discussion, he implemented the I2C recovery
functions.

Yes, reordering the pinctrl/gpio_free calls would fix the problem too.

However, I guess there is no explicit rule to that ("request/free GPIOs
only when they are muxed as GPIO"), so I think of it that the issue is
actually in the pinctrl driver.

On top of that it is not entirely trivial to reorder the calls the way
i2c_generic_gpio_recovery and i2c_generic_recovery are set up right now.

AFAICT, these routines don't touch the muxing part at all. Perhaps it is done
internally by the GPIO calls. Can you please elaborate the exact change you are
hinting towards here ?

The i.MX I2C driver touches the pinctrl in its prepare/unprepare
callbacks.

So, on a i.MX or Vybrid, the call chain looks like this:

i2c_generic_gpio_recovery
    -> i2c_get_gpios_for_recovery
       -> gpio_request_one
    -> i2c_generic_recovery
       -> prepare_recovery (i2c_imx_prepare_recovery)
          -> pinctrl_select_state [gpio]
       -> unprepare_recovery (i2c_imx_unprepare_recovery)
          -> pinctrl_select_state [default]
    -> i2c_put_gpios_for_recovery
       -> gpio_free


And for the pinctrl/GPIO driver of Vybrid this is actually a problem
because gpio_free disables the output driver of the pad, and when that
happens after the (I2C) default pinctrl state gets selected the pad is
no longer active.

--
Stefan