[Mesa-users] Closing/Disowning DRM FD and then reopening it

[Hugh Cole-Baker]

Hello! I'm trying to write an application which uses EGL and libdrm to render directly to the display without X running. I've been following what examples I could find on the Web but they're from 2011 and 2013 [1] [2] so probably what I'm doing is fairly out of date and I suspect that's related to a problem I've been having.

What I want to do is open() /dev/dri/card0, locate a KMS connector, encoder and CRTC to use, create an EGL surface and render to it, and then display that surface using drmModePageFlip() or drmModeSetCrtc(). This part is working fine. Then I want to de-init all the DRM related resources, close the DRM file descriptor, and launch a subprocess that will also use DRM to render to the display. So there will only be one process using DRM at any time. When that subprocess exits, I want to re-init all the DRM and EGL stuff, and resume drawing and displaying as before.

The problem I'm having is that I can't re-init and resume using DRM in my app after the subprocess has exited. Even with no subprocess launching, I can't use drmModePageFlip() after close()ing the file descriptor for /dev/dri/card0 and reopening it. It works fine the first time, but after closing, reopening and re-initialising EGL in the same process, I always get EACCES when calling drmModePageFlip().

I've modified this kmscube example app to show the problem [3]. The second call to draw_some_frames() always fails. The same thing happens on both a Raspberry Pi (VC4 GPU) with Mesa 11.1, and on Ubuntu 16.04 with Mesa 11.2 (Nouveau driver).

Can anyone give me a pointer on how to completely "disown" the /dev/dri/card0 so that another app can use it, and then reopen it later?

thanks in advance,
--
Hugh

[1] https://github.com/dvdhrm/docs/blob/master/drm-howto/modeset-vsync.c
[2] http://virtuousgeek.org/blog/index.php/jbarnes/2011/10/31/writing_stanalone_programs_with_egl_and_
[3] https://github.com/sigmaris/kmscube/blob/reinit/kmscube.c