Thread of 2 posts
So, fun story. I found out from developers related to the Intel Linux DRM ecosystem, specifically dealing with the Xe KMD, have found that the UAPI has padding issues that break 32-bit userspace on 64-bit kernels. What I didn't know was that this was discovered a whole month ago, and that a patch changing things is in review limbo right now.
I have since changed things slightly, by patching the missing compat_ioctl function for 32-bit userspace to reach the basic DRM functions, and first by applying my own xe_drm.h patch, then applying the already proposed one. Result: Now I have working 64-bit and 32-bit userspace OpenGL and Vulkan on the Xe KMD.
Not so hot: Some things are still broken, and some of it isn't related to the Xe KMD. The KMD doesn't have functioning VA-API yet, with an unmerged PR against intel-media-driver, marked "do not merge", mainly because it hard codes a lot, and doesn't detect which driver it should be using.
Also, some OpenGL consuming apps crash on Iris, but not on Zink now, and this is not related to the Xe KMD. Firefox and most web browsers outright crash except on Zink, but on Iris under i915, Firefox has a lovely bug where some web pages or cursor changing sites can cause the whole Gnome Shell compositor to crash back to login if I'm screencasting, which is also something I do most days. I'll attempt to bisect that one possibly in the evening hours of today, since I don't have any major issues going on right now.
Gnome Control Center also crashes on Iris Xe KMD, but not i915. This may or may not be the same bug affecting Firefox. It does launch properly if started with Zink enabled. Too bad Zink isn't capable of running a full desktop yet, if things are going to continue bugging up like this.
HEY! I was right! The Gnome Control Center (and Telegram Desktop, and Firefox) issues all appear to be the same, and they're related to Mesa Git, and not related to Xe KMD. Now Bed.
Bo(miptree/22 exported) with prime_fd unset in xe_batch_submit()