Re: [PATCH v2] gpio: convince line to become input in irq helper

[Linus Walleij <linus.walleij@xxxxxxxxxx>]

I sent a patch for the direction setter to be more careful, but
it's no silver bullet for strange semantics.

On Tue, Jul 5, 2016 at 12:07 PM, Geert Uytterhoeven
<geert@xxxxxxxxxxxxxx> wrote:

> [1] gpio_rcar e6052000.gpio: sense irq = 11, type = 8
>     ravb e6800000.ethernet eth0: Base address at 0xe6800000,
> 2e:09:0a:00:83:1e, IRQ 131.
>     ...
> [2] gpiochip_irq_reqres: gpiochip e6052000.gpio
> [3] gpio_rcar e6052000.gpio: gpio_rcar_direction_input: 11
> [4] gpiochip_irq_reqres: desc->flags = 0x0
(...)
> This configures the GPIO for plain input mode, cfr. [3] above, basically
> undoing the configuration from [1]. Hence interrupts no longer come through,
> and Ethernet fails.

The driver is a bit fragile in that it relies on a certain call semantic,
I guess it is not a widespread problem so we should be able to make
a local fix if necessary.

The .set_direction() call should
set the direction. Why is it turning off interrupts as a side effect?

What happens if you apply this, making the .request() function handle
the pin setup and .set_direction() really just setting the
direction?

diff --git a/drivers/gpio/gpio-rcar.c b/drivers/gpio/gpio-rcar.c
index 681c93fb9e70..68fb0147caf4 100644
--- a/drivers/gpio/gpio-rcar.c
+++ b/drivers/gpio/gpio-rcar.c
@@ -221,7 +221,20 @@ static void
gpio_rcar_config_general_input_output_mode(struct gpio_chip *chip,
     struct gpio_rcar_priv *p = gpiochip_get_data(chip);
     unsigned long flags;

-    /* follow steps in the GPIO documentation for
+    spin_lock_irqsave(&p->lock, flags);
+
+    /* Select Input Mode or Output Mode in INOUTSEL */
+    gpio_rcar_modify_bit(p, INOUTSEL, gpio, output);
+
+    spin_unlock_irqrestore(&p->lock, flags);
+}
+
+static int gpio_rcar_request(struct gpio_chip *chip, unsigned offset)
+{
+    struct gpio_rcar_priv *p = gpiochip_get_data(chip);
+
+    /*
+     * follow steps in the GPIO documentation for
      * "Setting General Output Mode" and
      * "Setting General Input Mode"
      */
@@ -234,14 +247,8 @@ static void
gpio_rcar_config_general_input_output_mode(struct gpio_chip *chip,
     /* Select "General Input/Output Mode" in IOINTSEL */
     gpio_rcar_modify_bit(p, IOINTSEL, gpio, false);

-    /* Select Input Mode or Output Mode in INOUTSEL */
-    gpio_rcar_modify_bit(p, INOUTSEL, gpio, output);
-
     spin_unlock_irqrestore(&p->lock, flags);
-}

-static int gpio_rcar_request(struct gpio_chip *chip, unsigned offset)
-{
     return pinctrl_request_gpio(chip->base + offset);
 }

.request() is always called before .set_direction() when issuing
gpiod_get() anyways, so the order required according to the comment
will be satisfied when the GPIO is first requested, but if we're just
using the interrupt, we still will be able to set the direction right.

Yours,
Linus Walleij
--
To unsubscribe from this list: send the line "unsubscribe linux-gpio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html