During my regression testing I noticed the cadence GPIO driver
fails on the latest gpio for-next tree.

I think the reason is this patch:
commit 96cd559817f2 ("Merge branch 'devel' into for-next")

Here is a part of the test log:

Loopback 8 -> 24
TESTING: gpio: 488: output direction PASSED
TESTING: gpio: 504: input direction PASSED
TESTING: gpio: 488: 0 PASSED
TESTING: gpio: 488 -> 504: 0 PASSED
TESTING: gpio: 488: 1 FAILED
TESTING: gpio: 488 -> 504: 1 FAILED
TESTING: gpio: 488: 0 PASSED
TESTING: gpio: 488 -> 504: 0 PASSED

It looks like the issue is that gc->bgpio_dir has changed its meaning.
It used to be set to the register value (so it was being inverted).

Now it's always set to 1 for output and 0 for input.
However the bgpio_get_set functions were not updated.
So they invert the bit again, which means a wrong register
is being accessed.

This patch fixes that by removing the unnecessary inversion.
Signed-off-by: Jan Kotas <jank@cadence.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>