gpio: pca953x: fix an incorrect lockdep warning
authorBartosz Golaszewski <bgolaszewski@baylibre.com>
Fri, 16 Sep 2016 16:02:45 +0000 (18:02 +0200)
committerWolfram Sang <wsa@the-dreams.de>
Sat, 24 Sep 2016 08:53:14 +0000 (10:53 +0200)
If an I2C GPIO multiplexer is driven by a GPIO provided by an expander
when there's a second expander using the same device driver on one of
the I2C bus segments, lockdep prints a deadlock warning when trying to
set the direction or the value of the GPIOs provided by the second
expander.

The below diagram presents the setup:

                                               - - - - -
 -------             ---------  Bus segment 1 |         |
|       |           |         |---------------  Devices
|       | SCL/SDA   |         |               |         |
| Linux |-----------| I2C MUX |                - - - - -
|       |    |      |         | Bus segment 2
|       |    |      |         |-------------------
 -------     |       ---------                    |
             |           |                    - - - - -
        ------------     | MUX GPIO          |         |
       |            |    |                     Devices
       |    GPIO    |    |                   |         |
       | Expander 1 |----                     - - - - -
       |            |                             |
        ------------                              | SCL/SDA
                                                  |
                                             ------------
                                            |            |
                                            |    GPIO    |
                                            | Expander 2 |
                                            |            |
                                             ------------

The reason for lockdep warning is that we take the chip->i2c_lock in
pca953x_gpio_set_value() or pca953x_gpio_direction_output() and then
come right back to pca953x_gpio_set_value() when the GPIO mux kicks
in. The locks actually protect different expanders, but for lockdep
both are of the same class, so it says:

  Possible unsafe locking scenario:

        CPU0
        ----
   lock(&chip->i2c_lock);
   lock(&chip->i2c_lock);

  *** DEADLOCK ***

  May be due to missing lock nesting notation

In order to get rid of the warning, retrieve the adapter nesting depth
and use it as lockdep subclass for chip->i2c_lock.

Signed-off-by: Bartosz Golaszewski <bgolaszewski@baylibre.com>
Acked-by: Peter Rosin <peda@axentia.se>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
drivers/gpio/gpio-pca953x.c

index 02f2a56..892dc04 100644 (file)
@@ -786,6 +786,8 @@ static int pca953x_probe(struct i2c_client *client,
        chip->chip_type = PCA_CHIP_TYPE(chip->driver_data);
 
        mutex_init(&chip->i2c_lock);
+       lockdep_set_subclass(&chip->i2c_lock,
+                            i2c_adapter_depth(client->adapter));
 
        /* initialize cached registers from their original values.
         * we can't share this chip with another i2c master.