On 11/14/19 2:50 PM, Stefan Agner wrote:
On 2019-11-14 14:17, Marek Vasut wrote:
The bus_flags and bus_format handling logic does not seem to cover all potential usecases. Specifically, this seems to fail with an "edt,etm0700g0edh6" display attached to an 24bit display interface, with interface-pix-fmt = "rgb24" set in DT.
In this specific setup, the panel-simple.c driver entry for the display sets .bus_flags to non-zero value. However, as imxpd->bus_format is set from the DT property "interface-pix-fmt", imx_pd_encoder_atomic_check() will set imx_crtc_state->bus_flags = imxpd->bus_flags even though the imxpd->bus_flags is zero, while the di->bus_flags is correctly set by the panel-simple.c and non-zero. The result is incorrect flags being used for the display configuration and thus an image corruption. (Specifically, DRM_BUS_FLAG_PIXDATA_POSEDGE is not propagated and thus the ipuv3 clocks pixels on the wrong edge).
This patch fixes the problem by overriding the imx_crtc_state->bus_format from the imxpd->bus_format only if the DT property "interface-pix-fmt" is present or if the DI provides no formats. Similarly for bus_flags, which are set from imxpd->bus_flags only if the DI provides no formats.
So this basically prioritizes imxpd->bus_format over what the display provides? Is this correct in all situations?
I was thinking that interface-pix-fmt is the legacy way to define the bus format and it should be provided by the display nowadays.
However, I guess there is the case where you connect a 18-bit display to a 24-bit bus (leaving some bits unconnected). Depending on how the colors/bits are distributed one cannot use 18-bit mode on SoC side but has to use 24-bit. So the bus format becomes a connection specific property... I guess the interface-pix-fmt can serve that role.
So, can this patch go in ? It's fixing an actual issue for me -- very much what Stefan described above.