[PATCH 0/1] arm64: dts: rockchip: fix HDMI RX HPD on CM3588 NAS

[Tim Surber <me@timsurber.de>]

The upstream CM3588 NAS device tree uses GPIO3_D4 both as the HDMI
receiver's HPD output and as its 5V detection input. The GPIO setting
wins, so the receiver can never drive HPD. Sources then read the EDID
while the driver is still writing it and keep the damaged copy. This
patch moves the 5V detection to GPIO1_D5, the pin that FriendlyElec's
vendor kernel uses for the CM3588 (rk3588-nanopi6-rev09.dts is the
CM3588 board file there):

  hdmirx-det-gpios = <&gpio1 RK_PD5 GPIO_ACTIVE_LOW>;
  https://github.com/friendlyarm/kernel-rockchip/blob/30d3e1c2470a7282f37b96fdd4354314dda69aa3/arch/arm64/boot/dts/rockchip/rk3588-nanopi6-rev09.dts#L299-L309

  hdmirx_det: rockchip,pins = <1 RK_PD5 RK_FUNC_GPIO &pcfg_pull_up>;
  https://github.com/friendlyarm/kernel-rockchip/blob/30d3e1c2470a7282f37b96fdd4354314dda69aa3/arch/arm64/boot/dts/rockchip/rk3588-nanopi6-rev09.dts#L584-L586

Testing was done on a CM3588 NAS with HDMI0 looped back into HDMI RX.
Both runs used the same kernel, built from linux-rockchip for-next
(c3baae5f1a15) with nothing else applied, and only the DTB was
swapped:

                                    for-next DTB    with this patch
  boots with a good EDID on HDMI0   7/10            10/10
  pin 124 (GPIO3_D4) function       hdmirx GPIO     hdmi (HPD)
  EDID change via VIDIOC_S_EDID     never seen      seen in ~0.48 s,
  reaching the source                               2 HPD IRQs each
  pinctrl "not freeing pin 124"     6/6 unbinds     0/6 unbinds
  warning on unbind
  5V detected (power_present),      yes             yes
  receiver locks to 3840x2160

dtbs_check gives no new warnings.

Tim Surber (1):
  arm64: dts: rockchip: fix HDMI RX 5V detection pin on CM3588 NAS

 .../boot/dts/rockchip/rk3588-friendlyelec-cm3588-nas.dts      | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)


base-commit: c3baae5f1a15645cddadcb2b9abddef5df291f59
-- 
2.55.0