[2/3] i2c: let I2C masters inherit suspend child ignorance

When using a certain I2C device with runtime PM enabled on
a certain I2C bus adaper the following happens:

struct amba_device *foo
   \
   struct i2c_adapter *bar
      \
      struct i2c_client *baz

The AMBA device foo has its device PM struct set to ignore
children with pm_suspend_ignore_children(&foo->dev, true).
This makes runtime PM work just fine locally in the driver:
the fact that devices on the bus are suspended or resumed
individually does not affect its operation, and the hardware
is not power up unless transferring messages.

However this child ignorance property is not inherited into
the struct i2c_adapter *bar.

On system suspend things will work fine.

On system resume the following annoying phenomenon occurs:

- In the pm_runtime_force_resume() path of
  struct i2c_client *baz, pm_runtime_set_active(&baz->dev); is
  eventually called.

- This becomes __pm_runtime_set_status(&baz->dev, RPM_ACTIVE);

- __pm_runtime_set_status() detects that RPM state is changed,
  and checks whether the parent is:
  not active (RPM_ACTIVE) and not ignoring its children
  If this happens it concludes something is wrong, because
  a parent that is not ignoring its children must be active
  before any children activate.

- Since the struct i2c_adapter *bar has not inherited the
  child ignorance and thus flags that it must indeed go
  online before its children, the check bails out with
  -EBUSY, i.e. the i2c_client *baz thinks it can't work
  because it's parent is not online, and it respects its
  parent.

- In the driver the .resume() callback returns -EBUSY from
  the runtime_force_resume() call as per above. This leaves
  the device in a suspended state, leading to bad behaviour
  later when the device is used. The following debug
  print is made with an extra printg patch but illustrates
  the problem:

[   17.040832] bh1780 2-0029: parent (i2c-2) is not active
               parent->power.ignore_children = 0
[   17.040832] bh1780 2-0029: pm_runtime_force_resume:
               pm_runtime_set_active() failed (-16)
[   17.040863] dpm_run_callback():
               pm_runtime_force_resume+0x0/0x88 returns -16
[   17.040863] PM: Device 2-0029 failed to resume: error -16

Fix this by letting all struct i2c_adapter:s inherit the
child ignorance setting from their parent when registering
an adapter. This way the child will realize it is ignored
and it can proceed to be resumed.

Cc: Wolfram Sang <wsa@the-dreams.de>
Cc: Mark Brown <broonie@kernel.org>
Cc: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>

---
Wolfram: if I can convince you and Rafael that this is the
right thing to do, I guess Rafael could eventually merge
this oneliner through his PM tree with your ACK.
---
 drivers/i2c/i2c-core.c | 1 +
 1 file changed, 1 insertion(+)

-- 
2.4.3

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

On Thu, Apr 7, 2016 at 3:20 PM, Linus Walleij <linus.walleij@linaro.org> wrote:

> @@ -1566,6 +1566,7 @@ static int i2c_register_adapter(struct i2c_adapter *adap)

>

>         pm_runtime_no_callbacks(&adap->dev);

>         pm_runtime_enable(&adap->dev);

> +       pm_suspend_inherit_ignore_children(&adap->dev);


If everyone agrees that i2c devices should never require the bus
to be resumed, we can of course use the big hammer and
just:

pm_suspend_ignore_children(&adap->dev, true);

Same for SPI. (And w1, ...)

Thoughts?

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

On Mon, Apr 11, 2016 at 08:53:40AM +0200, Linus Walleij wrote:

> If everyone agrees that i2c devices should never require the bus

> to be resumed, we can of course use the big hammer and

> just:


> pm_suspend_ignore_children(&adap->dev, true);


> Same for SPI. (And w1, ...)


> Thoughts?


That seems better.