kern/55827: variety of Ironlake graphics boards: 3D and 2D acceleration performance problems/crashes on *UXA*

[jbryn%students.wcpss.net@localhost]

>Number:         55827
>Category:       kern
>Synopsis:       Performance problems and non-working 3D on Ironlake
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Fri Nov 27 06:40:00 +0000 2020
>Originator:     Jackson Bryn
>Release:        NetBSD 9.99.75
>Organization:
N/A
>Environment:
System: NetBSD E5510-NetBSD.dell.foo 9.99.75 NetBSD 9.99.75 (GENERIC) #0: Mon Nov 16 14:12:46 UTC 2020 mkrepro%mkrepro.NetBSD.org@localhost:/usr/src/sys/arch/amd64/compile/GENERIC amd64
Architecture: x86_64
Machine: amd64
>Description:
Best demonstrated with X11.

With the intel driver on native Xorg, problems with acceleration rely beyond SNA not working.
Even with UXA enabled instead, while dragging and other basic parts of 2D acceleration does work unlike SNA, complex canvases fail at acceleration and instead appear to fallback to software. 3D support through native MesaLib essentially locks up, and sometimes crashes and reports this message.
`i965: failed to submit batchbuffer`

An actual fix to the KMS code needs to be implemented at some point, instead of haughty workarounds. Note that none of these bugs happen with biosboot plus VESA drivers (disabling modesetting).

Tested with the Dell Latitude E5510 and the HP ProBook 8440p.
>How-To-Repeat:
Build firefox on pkgsrc with native X/Mesa linkage, and watch as the canvas creating tearing and failing to get past 30 to 40fps. On biosboot/VESA, it can achieve a constant 60fps. WebGL context using hardware will also fail to be created, instead resorting to llvmpipe fallback.

Launch glxgears and see that it fails to render anything past the first two frames or so. Drag it around, and it should crash.
>Fix:
Unable to suggest a fix.