[U-Boot] Question regarding early pinctrl and driver model in u-boot

[Simon Glass]

Hi Lukasz,

On Thu, 6 Dec 2018 at 01:50, Lukasz Majewski <lukma at denx.de> wrote:
>
> Hi Simon,
>
> > Hi Lukasz,
> >
> > On Mon, 3 Dec 2018 at 15:13, Lukasz Majewski <lukma at denx.de> wrote:
> > >
> > > Dear All,
> > >
> > > I've stumbled upon a following issue:
> > >
> > > - I do have a vybrid SoC which is not using SPL. It only uses U-boot
> > >   proper (u-boot.vyb).
> > >
> > > - In the board_early_init_f() (when we are still in SRAM) I do
> > > perform pinctrl (pinmux) setup for UART1 (this is the console
> > > device).
> > >
> > > - I also do use ucalss-serial.c for this device and have proper
> > > pinmux definition for it in the *.dts file - but this is done
> > > latter.
> > >
> > > The problem is that, when I try to remove pinmux setup code
> > > (imx_iomux_v3_setup_multiple_pads()) from board_early_init_f(),
> > > then I do see uart output late (no U-boot early output - U-Boot
> > > 2018.11-00052-g6552299a40-dirty (Nov 23 2018 - 16:13:51 +0100)).
> > >
> > > The reason for this is the lack of pinctrl UART pins setup.
> > >
> > >
> > > Do you see any idea how to move to full dts support for uart? The
> > > uart driver and pinmux have defined DM_FLAG_PRE_RELOC.
> >
> > Is this due to pinconfig_post_bind() only operating after relocation?
>
> Please correct me if I'm wrong, but post_bind() callback is called with
> device_bind_* family of functions?

Yes

>
> Those functions - e.g. device_bind() are used in an interesting way
> - e.g.: drivers/gpio/s5p_gpio.c -> gpio_exynos_bind().

Yes, they create child devices.

>
> The problem seems to be with the time pinctrl device itself is probed
> (way after board_early_init_f()).
>
> IMHO it would be enough to "probe" pinctrl itself very early (before
> relocation - DDR setup) to just configure pins from iomux's pictrl_hog
> group (and put there DDR and uart pins) - not waiting for other devices
> to be "discovered" (like uart).

Well, even probing a single device will cause pinctrlt to be probed,
so shouldn't this happen automatically?

Regards,
Simon