Incorrect atomic usage in cx88-alsa driver

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

Hi !

So I see this construct:

	if (test_and_set_bit(0, &chip->opened))
		return -EBUSY;

	.../...

	return 0;
_error:
	dprintk(1,"Error opening PCM!\n");
	clear_bit(0, &chip->opened);
	smp_mb__after_clear_bit();
	return err;

So that's basically an attempt at doing a spinlock. The problem is your
barrier is wrong at the end. Better would be:

done:
	smp_mb__before_clear_bit();
	clear_bit(0, &chip->opened);

Though it's still less optimal that doing:

	if (!spin_trylock(...))
		goto bail;

	.../...

done:
	spin_unlock(...)

If you really want to stick to bitops, then you may want to look at
Nick's upcoming patches adding some bitops with appropriate lock
semantics.


Cheers,
Ben.