quoted

Right, and I'm not saying it should be, just move the existing logic
into the release callback, and the code flow should be the same and we
don't end up with an "empty" release callback.

But as Russell says, even if we don't have the empty callback, we still
create the problem shown by DEBUG_KOBJECT_RELEASE which wasn't there
before?

IMHO there are two possibilities here:

1. leave it as-is, where we ensure that the remainder of i2c_del_adapter
   does not complete until the release callback has been called.

2. fix it properly by taking (eg) the netdev approach to i2c_adapter,
   or an alternative solution which results in decoupling the lifetime
   of the struct device from the i2c_adapter.

Either of these would be much better than removing the completion and
then moving a chunk of code to make it "look" safer than it actually is
and thereby introducing potential use-after-free bugs.

I agree. As much as I'd love option 2) I don't see that on the horizon.
So, let's keep things as they are. What probably makes sense is to
update the comment with something like this? I took the liberty and used
some wording from Russell:

diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
index e227dff62a85..1c89a08fae2a 100644
--- a/drivers/i2c/i2c-core.c
+++ b/drivers/i2c/i2c-core.c

@@ -1778,11 +1778,14 @@ void i2c_del_adapter(struct i2c_adapter *adap)
 	/* device name is gone after device_unregister */
 	dev_dbg(&adap->dev, "adapter [%s] unregistered\n", adap->name);
 
-	/* clean up the sysfs representation */
+	/* wait until all references to the device are gone
+	 *
+	 * FIXME: This is old code and should ideally be replaced by an
+	 * alternative which results in decoupling the lifetime of the struct
+	 * device from the i2c_adapter, like spi or netdev do.
+	 */
 	init_completion(&adap->dev_released);
 	device_unregister(&adap->dev);
-
-	/* wait for sysfs to drop all references */
 	wait_for_completion(&adap->dev_released);
 
 	/* free bus id */

Thanks for all the input, it is very much appreciated!
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20150120/69ff1196/attachment.sig>