Re: [RFC PATCH 3/3] gpio: ej1x8: Add GPIO driver for Etron Tech Inc. EJ168/EJ188/EJ198

[Linus Walleij <linus.walleij@xxxxxxxxxx>]

Hi Martin,

thanks for your patch!

As noted on the earlier patches I think this should be folded into the
existing XHCI USB driver in drivers/usb/host/xhci-pci.c or, if that
gets messy, as a separate bolt-on, something like
xhci-pci-gpio.[c|h] in the drivers/usb/host/* directory.
You can use a Kconfig symbol for the GPIO portions or not.

On Sun, Oct 4, 2020 at 8:00 PM Martin Blumenstingl
<martin.blumenstingl@xxxxxxxxxxxxxx> wrote:

> EJ168/EJ188/EJ198 are USB xHCI controllers. They also contain four GPIO
> lines which are used on some systems to toggle an LED based on whether a
> USB device is connected.
>
> There is no public datasheet available for this hardware. All
> information in this driver is taken from the
> "F9K1115v2.03.97-GPL-10.2.85-20140313" GPL code dump of the Belkin
> F9K1115v2. This board comes with an EJ168 USB xHCI controller and the
> USB 3.0 LED is connected to one of the GPIOs. Inside the GPL source
> archive the related code can be found in:
>   linux/kernels/mips-linux-2.6.31/drivers/usb/host/etxhci-pci.c
>
> Signed-off-by: Martin Blumenstingl <martin.blumenstingl@xxxxxxxxxxxxxx>
(...)

> +config GPIO_EJ1X8
> +       tristate "Etron Tech Inc. EJ168/EJ188/EJ198 GPIO driver"
> +       depends on OF_GPIO && PCI

It is fine to just select GPIOLIB if you want this to always be
compiled-in (if the USB maintainers agree).

> +       help
> +         Selecting this option will enable the GPIO pins present on
> +         the Etron Tech Inc. EJ168/EJ188/EJ198 USB xHCI controllers.
> +
> +         If unsure, say N.

(...)
> +#define EJ1X8_GPIO_INIT                                        0x44
> +#define EJ1X8_GPIO_WRITE                               0x68
> +#define EJ1X8_GPIO_READ                                        0x6c
> +
> +#define EJ1X8_GPIO_CTRL                                        0x18005020
> +#define EJ1X8_GPIO_CTRL_READ_ALL_MASK                  GENMASK(7, 0)
> +#define EJ1X8_GPIO_CTRL_WRITE_ALL_MASK                 GENMASK(23, 16)
> +#define EJ1X8_GPIO_CTRL_OUT_LOW                                0x0
> +#define EJ1X8_GPIO_CTRL_OUT_HIGH                       0x1
> +#define EJ1X8_GPIO_CTRL_IN                             0x2
> +#define EJ1X8_GPIO_CTRL_MASK                           0x3
> +
> +#define EJ1X8_GPIO_MODE                                        0x18005022
> +#define EJ1X8_GPIO_MODE_READ_WRITE_ALL_MASK            GENMASK(23, 16)
> +#define EJ1X8_GPIO_MODE_DISABLE                                0x0
> +#define EJ1X8_GPIO_MODE_ENABLE                         0x1
> +#define EJ1X8_GPIO_MODE_MASK                           0x3

Nice that you got all of this out of reverse-engineering!

> +static LIST_HEAD(ej1x8_gpios);

This should not be necessary. Tie the GPIO state into the PCI device
driver state, possibly using some #ifdefs.

> +static u8 ej1x8_gpio_shift(unsigned int gpio, u8 mask)
> +{
> +       return (gpio * fls(mask));
> +}
> +
> +static u8 ej1x8_gpio_mask(unsigned int gpio, u8 mask)
> +{
> +       return mask << ej1x8_gpio_shift(gpio, mask);
> +}

This looks a bit like regmap but trying to use regmap for this
would probably be overengineering.

Looking at the code I get annoyed that it uses the config space to
manipulate the GPIOs, else you could have used GPIO_GENERIC
but now you can't, how typical.

Other than that the code looks nice, but fold it into the USB
host driver somehow unless there is a compelling argument
as to why not.

Yours,
Linus Walleij