10:02 <xdarklight> narmstrong: I can finally get CVBS video output on the 32-bit SoCs. kmscube screenshot for reference: https://abload.de/img/vlcsnap-2019-11-30-103qjbc.png

10:02 <narmstrong> xdarklight: wow nice !

10:02 <xdarklight> narmstrong: however, it requires a very weird patch to the meson DRM driver: https://pastebin.com/NR2h632w - without that neither kmscube, nor X nor anything else will start. any idea why that is?

10:04 <narmstrong> xdarklight: interesting, not really it’s not used for hw setup at all, but I did use X11 successfully at the time with this, but maybe the mode strict has evolved and something else is missing to define the aspect ratio

10:05 <narmstrong> xdarklight: can you compare with the same CEA mode on drm_modes ?

10:05 <narmstrong> *in

10:08 <xdarklight> narmstrong: drm_edid.c has edid_cea_modes and that contains a "720x576i" mode with HDMI_PICTURE_ASPECT_4_3 (and another one with HDMI_PICTURE_ASPECT_16_9)

10:09 <narmstrong> Maybe this HDMI_PICTURE_* should only be used with hdmi connectors

10:10 <xdarklight> maybe. I need to find one of my GXL/GXM/G12A boxes with CVBS output and test there and see if it has the same problem

10:10 <xdarklight> oh, unlike Khadas VIM the Le Potato has AV OUT. got one :-)

10:12 <narmstrong> Sun4i doesn’t set it https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/sun4i/sun4i_tv.c#L321

10:12 <narmstrong> Odroid-n2 has one but is not enabled upstream

10:13 <xdarklight> that makes 2 out of 3 which don't set HDMI_PICTURE_ASPECT_*. not setting it makes it default to 0 which is HDMI_PICTURE_ASPECT_NONE

10:13 <xdarklight> I don't have an Odroid-N2, but the Le Potato was right next to my desk :)

10:31 <narmstrong> xdarklight: what error gives kmscube ?

10:33 <xdarklight> narmstrong: you mean when I try it without that patch?

10:38 <xdarklight> https://pastebin.com/WQS8fWKL

10:39 <narmstrong> Can you enable drm.debug=0x3f on the kernel cmdline ? To know what rejects the modeset

10:40 <xdarklight> it actually fails because of meson_venc_cvbs_encoder_atomic_check

10:40 <xdarklight> that drm_mode_equal never returns true when run from kmscube

10:41 <xdarklight> that's how I found the different aspect ratio

10:41 <xdarklight> but I can still enable drm.debug if you want

10:52 <narmstrong> Yeah I suspect that but it wasn’t checking aspect ratio before...

10:54 <narmstrong> Could you replace with drm_mode_match but without aspect ratio flag ?

10:54 <xdarklight> v4.18 is the first kernel which checks for the aspect ratio

10:55 <xdarklight> yep, will try drm_mode_match

11:03 <xdarklight> drm_mode_match without DRM_MODE_MATCH_ASPECT_RATIO works

20:05 <xdarklight> narmstrong: finally powered up my Le Potato: CVBS is also broken there. I'll send a patch later fixing it (by removing the aspect ratio to make behavior of the meson driver identical to the sun4i and zte drivers)

20:06 <narmstrong> xdarklight: Nice, thanks for checking this !

20:21 <xdarklight> narmstrong: last question for today: do you have any hint where to check if I'm not receiving VPU interrupts? the u-boot on my EC-100 board doesn't have video output enabled and it's the only board where CVBS is currently not working. I *think* I have the power domain right as I can read/write VPU registers without hanging the board. I'm "just" missing the interrupt

21:06 <narmstrong> xdarklight: the vpu irq is tied to a clock gate, check by enabling all the gates

21:14 <narmstrong> xdarklight: I’ll try to think of something else

22:52 <xdarklight> narmstrong: I enabled all clk81 gates and all others which seems related to video output, no luck. if you have additional ideas: let me know. I guess I'll call it a day for now