Possibly related to fd leaks when sharing surfaces. I'll message you on freenode if I have any trouble repro'ing the issue.
@ChristophHaag got the bug reproduced locally
Any progress? I think I've seen a similar crash in Doom 3 with SteamVR, so I'm not using the patches anymore for now...
@ChristophHaag This is due to libdrm running out of va space to import the new handle. I'll be taking a look at why that happens soon (once I get a new version of the GL semaphore patches out).
Adding correct protection against the segfault when import fails is going to be a slightly bigger task than expected. That problem has a little bit of deep roots.
The latest SteamVR beta correctly detects cases where a shared texture is un-importable and returns an error (as opposed to crashing).
Haven't looked into the exact details as to why the shared-texture is un-importable, but I suspect the VA resources are being exhausted due constant eye texture allocations. Probably worth another look to determine whether the app or steamvr is failing to fully cleanup the old resources.
Driver side fix for this issue is now pushed to mesa master:
https://lists.freedesktop.org/archives/mesa-dev/2018-April/193590.html
The fix above + the latest steam beta should have threaded hifi working correctly:
https://steamcommunity.com/games/250820/announcements/detail/1658891645926260963
Nothing extracted yet.
RX 480, latest radv.
If you want to test it, use this branch https://github.com/ChristophHaag/hifi/commits/threadedsubmit
See the latest commit that enables their threaded submit implementation. It looks like they're basically doing their own reprojection there.
Anyway, at https://github.com/highfidelity/hifi/blob/89425ca210b47f3b66bf461ab62423da4781fa1c/plugins/openvr/src/OpenVrDisplayPlugin.cpp#L214 they have their thread that submits, which works fine for a couple of seconds, but then SteamVR's vrcompositor segfaults.
That probably shouldn't happen, no matter what the application submits.