gpio_request_by_name for GPIOD_IS_OUT drives GPIO before specifically instructed to

[Tim Harvey]

Greetings,

I'm seeing an issue in U-Boot caused by gpio_request_by_name driving a
GPIO output before it has been given an output level with
{dm_}gpio_set_value.

In my particular instance I have a network PHY that can encounter
errata if it gets reset more than once (fun time with this one!)
declared liks this:

&ethphy0 {
       reset-gpios = <&gpio3 0 GPIO_ACTIVE_LOW>;
       reset-assert-us = <1000>;
       reset-deassert-us = <300000>;
};

Through testing it has been found that this PHY must be held in reset
on power-up and released from reset only once. This is probably the
third time I've run into such an issue in the 20 years I've worked
with hardware - it's rare and certainly a result of poor reset
handling in IC's but not a unique situation (especially at a time
where we can't be too choosy about what parts go on boards).

When eth-phy-uclass.c calls gpio_request_by_name(dev, "reset-gpios",
0, &uc_priv->reset_gpio, GPIOD_IS_OUT) direction_output is called
which sets the GPIO to a high effectively taking the PHY out of reset
before eth_phy_reset is called to do so. Setting the gpio as
GPIO_ACTIVE_LOW will drive it low keeping it in reset until ready to
de-assert it but then the polarity is inverted which results in the
PHY being left in reset permanently.

This affects a wide variety of gpio based things including gpio-hogs.

I'm looking for some input on how to best address this issue. I
haven't finished walking through all the kernel code but I believe
when a gpio is requested it does so without configuring that gpio. Any
ideas?

Best regards,

Tim