No HDMI monitor after updating to 28.2

For the record reflashing with the modified kernel, following these instructions didn’t work either.

[url]https://docs.nvidia.com/jetson/l4t/index.html#page/Tegra%2520Linux%2520Driver%2520Package%2520Development%2520Guide%2Fdisplay_configuration.html%23wwpID0E06H0HA[/url]

FYI, the get-edid mechanism can fail because the i2c bus doing the query is not always on. So this is to be expected.

Someone knowing the internals of the driver would have to answer the question of why there is an EDID, but ModeDebug does not see it (the EDID definitely has a lot of modes, and some of them very standard and generic). I would expect to see the driver commenting on every single mode with ModeDebug, but not a single EDID mode is ever visible to the driver. Even without ModeDebug I would expect to see more in the logs.

Did you edit the config in the script and specify a mode ? If not, it may just be that your monitor doesn’t support the default 720x480p mode.

You may try to plug your monitor into a Linux PC (preferably connected to a NVIDIA card using NVIDIA driver), and look for a working mode there, and then try to specify this mode into the config. I think this would not modify kernel but rather patch device tree. Be sure you update the DT.

I didn’t but my monitors EDID lists a 480p mode in the DTD section.

Detailed Timing Descriptor
Pixel Clock: 27MHz 
Horizontal Active: 720 
Horizontal Blanking: 138 
Vertical Active: 480 
Vertical Blanking: 45 
Horizontal Sync Offset: 16 
Horizontal Sync Pulse: 62 
Vertical Sync Offset: 9 
Vertical Sync Pulse: 6 
Horizontal Display Size: 509 
Vertical Display Size: 286 
Horizontal Border: 0 
Vertical Border: 0 
Interlaced: false 
Stereo Mode: 0 
Sync Type: 3 


2-Way Line-Interleaved Stereo: false

Unfortunately I have to reinstall the Jetson TX2 on the robot today so I won’t be able to experiment anymore for a few days.

BTW this is the project on which I’m working: http://www.dai-art.ist , it’s an AI art project.

I’ve learned a lot these past few days, thanks to linuxdev and Honey_Patouceul, so even though my problem isn’t resolved I’m very grateful for all the help so far.

Maybe in the meantime someone from nvidia could chime in about how “ModeDebug” doesn’t see the perfectly valid modes that are in the EDID?

There are some limitations on the GPU driver which you won’t see on a desktop PC. Specifically, video modes can only be chosen by EDID, and if the EDID mode of the monitor is not one of the pre-programmed modes in the GPU’s mode pool list, then the mode won’t be accepted. An example would be that no interlaced modes are allowed, and so even if EDID listed one, then that mode wouldn’t be possible.

I am hoping for a comment from NVIDIA though. The fact is that with ModeDebug logging enabled all modes should be listed from the EDID…and a brief comment generated on every single mode as to that mode being accepted or rejected (and if rejected the reason why). Not a single mode was shown with ModeDebug despite having an EDID with valid checksum and many more or less standard modes. Either the driver doesn’t see EDID, or else the driver logging choked.