[PATCH] drm/vc4: crtc: Fix runtime_pm reference counting

* [PATCH] drm/vc4: crtc: Fix runtime_pm reference counting
@ 2022-02-03 10:20 Maxime Ripard
  2022-02-17 16:18 ` Javier Martinez Canillas
  2022-02-17 16:34 ` (subset) " Maxime Ripard
  0 siblings, 2 replies; 4+ messages in thread
From: Maxime Ripard @ 2022-02-03 10:20 UTC (permalink / raw)
  To: dri-devel
  Cc: Dom Cobley, Tim Gover, Dave Stevenson, David Airlie,
	Maxime Ripard, Thomas Zimmermann, Daniel Vetter, Phil Elwell

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>
---
 drivers/gpu/drm/vc4/vc4_crtc.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/vc4/vc4_crtc.c b/drivers/gpu/drm/vc4/vc4_crtc.c
index 287dbc89ad64..799aaf8c1abf 100644
--- a/drivers/gpu/drm/vc4/vc4_crtc.c
+++ b/drivers/gpu/drm/vc4/vc4_crtc.c
@@ -525,9 +525,11 @@ int vc4_crtc_disable_at_boot(struct drm_crtc *crtc)
 	if (ret)
 		return ret;
 
-	ret = pm_runtime_put(&vc4_hdmi->pdev->dev);
-	if (ret)
-		return ret;
+	/*
+	 * post_crtc_powerdown will have called pm_runtime_put, so we
+	 * don't need it here otherwise we'll get the reference counting
+	 * wrong.
+	 */
 
 	return 0;
 }
-- 
2.34.1


^ permalink raw reply related	[flat|nested] 4+ messages in thread