Re: libdrm+kernel patches/fixes from wayland porting work
Taylor R Campbell <[email protected]> Fri, 31 Jan 2025 07:32:25 +0000
| Newsgroups | gmane.os.netbsd.devel.x11 |
|---|---|
| Message-ID | <[email protected]> |
> Date: Thu, 30 Jan 2025 23:13:45 -0800 > From: Jeff Frasca <[email protected]> > > I am back on my project of trying to get my wlroots style Wayland > compositor running on NetBSD. Cool! > My experience with NetBSD is that X11 ends up running anything with > OpenGL via the LLVMpipe software renderer, and I think I figured > out why. > > The libdrm routines that Mesa uses to initialize itself for both the > compositor and the client apps leans heavily on the GET_UNIQUE ioctl. > However, only the master node has a unique string, and the render > nodes do not. For both the compositor and the client, Mesa calls > GET_UNIQUE on the render node's device fd, gets back an error and > then gives up on setting up full acceleration. Last year I made some changes to HEAD to address this: PR kern/58180: render nodes are busted https://gnats.NetBSD.org/58180 https://mail-index.NetBSD.org/source-changes/2024/04/21/msg150921.html https://mail-index.NetBSD.org/source-changes/2024/04/21/msg150922.html The approach was to pass the same information through the GET_UNIQUE ioctl and expose it to render nodes. However, that exposed some other bug which broke Firefox or something, so it's not getting pulled up to 10. Will take a look at the rest later when I have a moment!