Desktop and Graphics: compatibility, accel drivers, AI policy

Developers examined long-term binary compatibility under GNU/Linux upgrades and reviewed kernel work on accelerators, memory conversion, and MSR interfaces. A separate KDE thread debated policies for AI and LLM use in contributions.

GNU/Linux upgrades breaking older programs

Richard Stallman opened a thread on emacs-devel about concrete GCC and Glibc changes that break old Emacs builds. Participants explored adjustments to distribution packaging and Emacs practices that could improve long-term compatibility. The discussion matters to anyone maintaining or shipping software expected to survive multiple system upgrades.

Flymake-harper package proposed for GNU ELPA

A proposal to add the flymake-harper package to GNU ELPA led maintainers into an extended review of the Flymake quick-fix API. The exchange clarified expectations for how new checkers should expose fixes. Emacs developers tracking linting and correction tooling have a direct stake in the outcome.

Qualcomm DSP accelerator driver v2

Qualcomm posted a second version of a DRM accel driver for its DSP FastRPC interface. Reviewers rejected the approach as a duplication of the existing fastrpc driver and insisted on a single-driver design. The outcome will shape how Qualcomm DSP acceleration is exposed in the kernel graphics stack.

Clearing pages after private-to-shared conversion

A four-patch v2 series fixes zeroing, sleep, and rollback problems that appear after set_memory_decrypted() in GIC ITS and dma-buf heaps. The work targets Arm CCA with MEC. Correct handling of these conversions is required for confidential-computing and secure dma-buf paths.

Dropping 32-bit x86 MSR interfaces

A thirteen-patch series removes the legacy 32-bit rdmsr and wrmsr macros once every remaining user has been converted to the 64-bit variants. The cleanup reduces dual-path maintenance in the x86 architecture code. Kernel developers working near model-specific registers will see a simpler interface.

KDE AI and LLM contribution policy

A Discourse thread on AI and LLM use in KDE contributions mixed arguments over ethics, accessibility, and maintainer trust. The exchange grew heated and included personal accusations. The debate surfaces unresolved community norms around generative tools in free-software projects.

vmwgfx stable fix flagged by AI bot

The stable maintainer added a vmwgfx NULL-dereference fix to the 6.18-stable tree. An AI bot promptly labeled the change ineffective dead code. The episode shows friction between automated review and graphics-driver stable backports.

Dedicated ROCm compute backend for QEMU

An AMD engineer proposed a ROCm GPU compute backend kept separate from the virglrenderer display path. The design aims to cut queue contention and enable asynchronous events. Improved isolation would benefit users of AMD GPU compute inside QEMU.