Re: device model update 2/2

[Linus Torvalds]

In article <[e-mail]>,
Patrick Mochel  <[e-mail]> wrote:
>-
> 		/* detach from driver */
> 		if (dev->driver->remove)
> 			dev->driver->remove(dev);
> 		put_driver(dev->driver);
>+
>+		lock_device(dev);
>+		dev->driver = NULL;
>+		unlock_device(dev);

Code like the above just basically can _never_ be correct.

The locking just doesn't make any sense like that. 

Real locking looks something like this:

	lock_device(dev);
	driver = dev->driver;
	dev->driver = NULL;
	unlock_device(dev);

	if (driver->remove)
		driver->remove(dev);
	put_driver(driver);

together with some promise that "dev" cannot go away from under us (ie a
refcount on "dev" itself).

		Linus