[U-Boot] [PATCH V3] imx: mx25: Remove SION bit in all pin-mux that are safe

[Benoît Thébaudeau]

Hi Michael,

On Thu, Jan 25, 2018 at 2:06 PM, Michael Trimarchi
<michael at amarulasolutions.com> wrote:
> SION bit should be used in the situation that we need
> to read back the value of a pin and should not be set by
> default macro.
>
> We get some malfunction as raised by following thread
>
> https://www.spinics.net/lists/linux-usb/msg162574.html
>
> As reported by this application note:
> https://www.nxp.com/docs/en/application-note/AN5078.pdf
>
> The software input on (SION) bit is an option to force an input
> path to be active regardless of the value driven by the
> corresponding module. It is used when the nature direction
> of a pin depending on selected alternative function is an output,
> but it is needed to read the real logic value on a pin.
>
> The SION bit can be used in:
> • Loopback: the module of a selected alternative function drives
> the pad and also receives the pad value as an input
> • GPIO capture: the module of a selected alternative function
> drives the pin and the value is captured by the GPIO
>
> SION bit is not necessary when the pin is configured as a peripheral
> apart specific silicon bug. If an application needs to have this
> set, this should be done in board file or in dts file
>
> Signed-off-by: Michael Trimarchi <michael at amarulasolutions.com>

[...]

I have now tested the following peripherals on i.MX25 without any SION
bit set: AUDMUX/SSI, eSDHC, FEC (except PHY access for which MDIO
might have an issue without SION; Ethernet works but I have not
checked whether this is possible at least partially without access to
the PHY registers), GPIO, I2C, SLCDC, UART, and USB. Everything worked
fine except the lack of SION for eSDHC CMD, which is true for all
eSDHC instances (and not only for eSDHC1, so Linux's
arch/arm/boot/dts/imx25-pinfunc.h should be fixed). Therefore, SION is
mandatory for all the eSDHC CMD functions in
arch/arm/include/asm/arch-mx25/iomux-mx25.h, but it can be removed for
everything else (I still have to carry out one more test to confirm
this for FEC MDIO), and moved to the board files if necessary (but I
don't think that any mainline board code is doing something fancy
enough to need it).

If the bus-status detection example in the documentation of SION in
the reference manual is useful, it means that SION is probably
required for all the signals requiring simultaneous output and input
(such as I²C for device clock stretching or multi-master bus
arbitration, except if the IP toggles between input and output low at
each clock edge rather than between open-drain output high and output
low), because there are no automatic SION signals between the
peripherals and the pads but only direction signals that can request a
single direction at a time. For bidirectional signals that do not
require simultaneous output and input because they work in turns (such
as FEC MDIO), SION can be required or not depending on whether the IP
toggles the direction signal for each turn or always expects an input
feedback while driving an open-drain output high. Even if SION is
required for the I²C example mentioned above (which is unlikely as
basic I²C transfers work fine and clock stretching detection is
automatic and would always need the input state), the need for these
advanced I²C features can be considered board-specific, so SION would
still not be required in iomux-mx25.h.

In the end, for this patch, apart from the pending test for FEC MDIO:
Reviewed-by: Benoît Thébaudeau <benoit.thebaudeau.dev at gmail.com>

Best regards,
Benoît