Re: [PATCH v2 1/2] pinctrl: imx: answer OUTPUT_ENABLE/INPUT_ENABLE queries from the pad register

[Mehmet Fide <mehmet.fide@gmail.com>]

From: Mehmet Fide <mehmet.fide@screeningeagle.com>

> New issues:
> - [High] Dangerous asymmetry between `pin_config_get` (generic) and `pin_config_set` (raw) causes hardware register corruption.
> - [Medium] Incomplete transition to generic pinconf API breaks SCU-based platforms.
>
> Pre-existing issues:
> - [Critical] Out-of-bounds write during Device Tree parsing in imx_pinctrl_parse_pin_mmio

[...]
> [Severity: Medium]
> Does this bypass the generic decoding logic for SCU platforms?
>
> If a generic caller like pinctrl_gpio_get_config() queries an SCU-based
> platform, it appears the SCU path (info->imx_pinconf_get) will overwrite
> the generic parameter in *config with a raw hardware register value instead
> of unpacking it.

Yes, that slipped through: the SCU callback returns the raw pad value
like the mmio one did. v3 makes imx_pinconf_get() return -ENOTSUPP for
IMX_USE_SCU SoCs, so a generic query gets "not supported" there instead
of register bits; the debugfs dump keeps reading the raw value through
the SCU path as before.

[...]
> [Severity: High]
> Does this introduce an asymmetry between the get and set callbacks causing
> hardware corruption?
>
> While the get callback now unpacks generic parameters, imx_pinconf_set_mmio()
> remains completely raw. If a generic caller like pinctrl_gpio_set_config()
> passes a packed config (e.g., PIN_CONFIG_LEVEL), will it be blindly written
> into the hardware pad configuration register here?

It would, and it would have before this patch too; the set side is not
changed here. The fsl,pins binding hands pin_config_set the raw pad
register value and that is the only thing it accepts. The one in-tree
route for a packed config into it is gpiochip_generic_config(), and the
GPIO drivers on these pin controllers do not install it as set_config
(gpio-mmio, gpio-vf610, gpio-mxc); gpio-mxc calls it directly only on
the SCU based SoCs, with the SCU wakeup code the SCU set callback
expects, not a generic parameter. Without a set_config the gpiolib bias
and persist helpers stop at -ENOTSUPP before reaching pinctrl. A raw
value and a packed one cannot be told apart in the callback either, so
decoding on the set side means a real conversion of the driver, not a
fix that belongs in this series. v3 states that boundary in the commit
message.

[...]
> [Severity: Critical]
> This is a pre-existing issue, but can the Device Tree parsing cause an
> out-of-bounds write here?
>
> Looking at imx_pinctrl_parse_pin_mmio() in drivers/pinctrl/freescale/pinctrl-imx.c:
>
>     *pin_id = (mux_reg != -1) ? mux_reg / 4 : conf_reg / 4;
>     pin_reg = &ipctl->pin_regs[*pin_id];
>     pin->pin = *pin_id;
>     pin_reg->mux_reg = mux_reg;
>     pin_reg->conf_reg = conf_reg;
>
> If a maliciously crafted Device Tree overlay provides an excessively large
> mux_reg or conf_reg value, the unvalidated pin_id is used to index into
> ipctl->pin_regs, potentially overwriting arbitrary memory.

Agreed that the index is not checked against info->npins; it has been
like that since the parser was written and this patch does not touch it.
A bounds check with -EINVAL is a one-liner, I will send it as a separate
patch after this series.

Mehmet