XLibre
Restoration of the "Golden Age" XAA: Undoing the 2007–2009 Architectural Sabotage #487
HaplessIdiot
started this conversation in
1 New Ideas
Replies: 1 comment 1 reply
|
This is a fascinating deep dive into X11 history! I admire the dedication to preserving hardware support and Alan Coppersmith's legacy. It's cool that T2 SDE provides a blueprint—looking forward to seeing patches materialize. Restoring XAA could breathe new life into older silicon. Good luck with the effort! |
1 reply
|
Heres the really cool part this patch from T2 can restore the entire XAA folder already https://github.com/rxrbln/t2sde/blob/main/package/xorg/xorg-server/restore-xaa.patch |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Describe the idea
The goal is to perform a systematic revert and restoration of the X Acceleration Architecture (XAA) and its associated legacy hooks that were aggressively purged from the Xorg tree between 2007 and 2009.
We will leverage the battle-tested T2 Linux SDE patches as a blueprint to re-integrate the hw/xfree86/xaa directory and bypass the sabotage commits that attempted to make XAA unbuildable in modern environments.
Pull the hw/xfree86/xaa directory and the ScreenRec/ScrnInfoRec acceleration pointers from the pre-Dave Airlie "Golden Age" (circa 2006/early 2007) that Alan Coopersmith worked so hard to keep alive with Solaris.
Re-insert the AccelInit pointers into core structures to allow drivers like fbdev to talk directly to 2D silicon again.
Strip the mandatory "Modern PC" requirements (PCI probing, x86 Video BIOS, VGA registers) that were baked into the "universal" modesetting path.
It should be done because
The 2007–2009 "Red Wedding" era of commits was a political consolidation disguised as a technical cleanup. This perfectly mirrors today with XLibre vs Wayland, the old war was XAA vs EXA.
For years, Alan Coopersmith was the primary guardian of X11’s "Universal" promise. The mass-deletion of XAA wasn't a consensus-driven evolution; it was a surrender of the very legacy-compatibility and hardware-hardening that Alan had spent his career protecting.
T2 SDE has maintained XAA patches, proving the "KMS makes XAA impossible" narrative was a convenient myth used to justify deleting code for non-x86 architectures.
The Dreamcast (PowerVR CLX2) is an MMIO-based device. Forcing it through a PCI/VESA lens is an architectural mismatch. By restoring the XAA paths that Alan once defended, we give this hardware its 2D performance back without the bloat of a broken 3D/Glamor stack.
We are moving past the era where diverse hardware was treated as "technical debt" to be discarded. We are reclaiming the diversity of the ecosystem that was lost during the 2007 recession’s push for consolidation.
What are the alternatives?
Glamor: Requires a full OpenGL stack. On a 200MHz SH-4, this is a non-starter.
ShadowFB: A software-only fallback that leaves the actual 2D hardware blitters cold and unused.
EXA: A later architecture that, while functional, lacks the close-to-the-metal simplicity for fixed-function legacy chips that XAA excelled at.
Additional context
The T2 SDE project has long been a sanctuary for "unpopular" architectures. By pulling their XAA patches into xlibre, we are joining forces with the developers who refused to let the 2007 "Recession Purge" kill hardware diversity.
Restoring Alan Coopersmith’s "Golden Age" work honors the original Unix philosophy: if it has a framebuffer, it's a display. We are moving away from the PCI-only motherboard tracking and returning to a truly universal display server.
This could also convince Alan to ditch Wayland if he saw our project actually stuck to the golden age of X11 that he tried for 3 years to save back then. It might also undo the damage that he was forced to do with main on Xorg. This mirrors very much like how he was forced to remove his XAA work in the previous recession. I fail to believe he was happy about doing that 2 year revert that's the same thing he fought so hard to avoid in the past happening again.
Extra fields