drm/stm: ltdc: fix pinctrl recovery after sleep

[dougg3]

I encountered a problem where an RGB TFT LCD panel attached to an STM32MP1 would stop working after the framebuffer was put to sleep and woken back up. The reason was that the sleep pinctrl was applied when the framebuffer went to sleep, but the default pinctrl wasn't being restored after the LTDC woke up. This is because an older commit moved the LTDC default pinctrl configuration to ltdc_encoder_mode_set, which doesn't get called again after it wakes up.

This fix was tested in the v5.4-stm32mp branch on a custom STM32MP1 board. I tested by blanking and then unblanking the framebuffer. Before this fix, the LCD display got garbled after unblanking. With this fix, it recovers correctly.

echo 1 > /sys/class/graphics/fb0/blank
echo 0 > /sys/class/graphics/fb0/blank

[dougg3]

Is there a realistic world in which this PR gets merged at this point? Or should I try submitting it to the upstream kernel instead?

[BernardPuel]

Hello Doug, sorry for the delay and thanks for your contribution. I will push internally for rapid analysis and recovery of normal process.

[BernardPuel]

The fix is already part of last delivery (V4.1). It will be also backported (as your request) on V3.1.3 release (this summer).

[dougg3]

Hi Bernard,

Thanks for looking into this. Looking at the latest code in the v5.15-stm32mp branch of this repository it looks like the problem is still there (ltdc_encoder_enable doesn't restore the default pinctrl) -- is there another place I should be looking at for the fix?

[BernardPuel]

after review, the first patch was not in V4.1 but is planned to be also delivered for V4.1.1. So a review will be done to check all the use cases are taken into account with the 2 patches for V4.1.1.

[no111u3]

@BernardPuel so there is the same behaviour for DSI too but on different kernel version (5.15.67) with run on board STM32MP157F-DK2, but patch fixed only non-DSI part of code.

[BernardPuel]

Here is the patch planned for V4.1.1:

diff --git a/drivers/gpu/drm/stm/ltdc.c b/drivers/gpu/drm/stm/ltdc.c
index 3dcd63f..f7b9258 100644
--- a/drivers/gpu/drm/stm/ltdc.c
+++ b/drivers/gpu/drm/stm/ltdc.c
@@ -1060,6 +1060,20 @@
 	}
 }
 
+static int ltdc_crtc_atomic_check(struct drm_crtc *crtc,
+				  struct drm_atomic_state *state)
+{
+	struct drm_crtc_state *crtc_state = drm_atomic_get_new_crtc_state(state, crtc);
+
+	DRM_DEBUG_ATOMIC("\n");
+
+	/* force a full mode set if active state changed */
+	if (crtc_state->active_changed)
+		crtc_state->mode_changed = true;
+
+	return 0;
+}
+
 static bool ltdc_crtc_get_scanout_position(struct drm_crtc *crtc,
 					   bool in_vblank_irq,
 					   int *vpos, int *hpos,
@@ -1120,6 +1134,7 @@
 	.atomic_flush = ltdc_crtc_atomic_flush,
 	.atomic_enable = ltdc_crtc_atomic_enable,
 	.atomic_disable = ltdc_crtc_atomic_disable,
+	.atomic_check = ltdc_crtc_atomic_check,
 	.get_scanout_position = ltdc_crtc_get_scanout_position,
 };

[no111u3]

@BernardPuel thank you for patch, it works on manual test, will you have plans to release to this repo?
Thank you.

[BernardPuel]

Thanks for your feedback. Yes in this repo but not in 5.10 kernel branch. Only 5.15.