drm/vc4: crtc: Fix runtime_pm reference counting
authorMaxime Ripard <maxime@cerno.tech>
Thu, 3 Feb 2022 10:20:03 +0000 (11:20 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 2 Mar 2022 10:47:57 +0000 (11:47 +0100)
commiteae86ab320696e71b8ce667398499c1f556bb9b4
tree4b1c085dec082307613b0ce248b2d54946b39ddf
parent163e50b00530d3a64a0dda7be80c0bf25987626b
drm/vc4: crtc: Fix runtime_pm reference counting

commit 6764eb690e77ecded48587d6d4e346ba2e196546 upstream.

At boot on the BCM2711, if the HDMI controllers are running, the CRTC
driver will disable itself and its associated HDMI controller to work
around a hardware bug that would leave some pixels stuck in a FIFO.

In order to avoid that issue, we need to run some operations in lockstep
between the CRTC and HDMI controller, and we need to make sure the HDMI
controller will be powered properly.

However, since we haven't enabled it through KMS, the runtime_pm state
is off at this point so we need to make sure the device is powered
through pm_runtime_resume_and_get, and once the operations are complete,
we call pm_runtime_put.

However, the HDMI controller will do that itself in its
post_crtc_powerdown, which means we'll end up calling pm_runtime_put for
a single pm_runtime_get, throwing the reference counting off. Let's
remove the pm_runtime_put call in the CRTC code in order to have the
proper counting.

Fixes: bca10db67bda ("drm/vc4: crtc: Make sure the HDMI controller is powered when disabling")
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220203102003.1114673-1-maxime@cerno.tech
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/gpu/drm/vc4/vc4_crtc.c