Desktop and graphics: ELPA LLM policy, DRM atomic and locking work

Emacs maintainers debated NonGNU ELPA acceptance of LLM-generated code while DRM developers advanced large series on connector state, GPUVM locking, and dma-buf accounting. Several kernel RFCs also touched Arm accelerators, mapping types, and vc4 lock contention.

NonGNU ELPA and LLM-generated code

Richard Stallman stated that NonGNU ELPA will not accept LLM-generated code in reply to a submission inquiry for clutch, an interactive database client. Stefan Monnier and Philip Kaludercic pushed back on applying GNU copyright assignment rules to the NonGNU archive. The heated exchange leaves package authors without a settled policy as generated submissions reach the list.

ekp package submission and the same policy question

A NonGNU ELPA proposal for ekp, a Knuth-Plass line-breaking package, drew a long technical thread that again raised LLM-generated code acceptance. Maintainers weighed the submission against the still-unresolved criteria discussed in the clutch inquiry. Emacs packaging followers should treat the two threads as one governance problem.

DRM connectors convert to atomic_create_state

A 59-patch series converts connector reset hooks to atomic_create_state across DRM drivers and removes the reset hook. The work on dri-devel aims to standardize atomic state creation for every connector path. Driver maintainers will need to clear remaining reset users as the conversion proceeds.

Arm Core Local Accelerator driver RFC

RFC v2 for an Arm Core Local Accelerator driver proposes changes across misc, KVM, and VFIO. Jason Gunthorpe objected to mdev usage in the design. The series matters for how specialized Arm accelerators attach to graphics and virtualization stacks.

Two-pass locking for drm_gpuvm exec

Version 3 adds two-pass locking to drm_gpuvm_exec so evicted private buffer objects can be validated without holding dma-resv locks of unrelated shared resident objects. The change targets lock contention during GPU virtual memory execution. Developers working shared BO paths should review the new locking order.

memcg accounting for dma-buf

RFC v2 adds memcg dma-buf accounting, a charge_fd ioctl, and LSM hooks. Automated review flagged multiple use-after-free and permission bugs in the patches. The goal is to bring GPU buffer memory under normal cgroup and security control.

DMA-buf mapping types

An RFC introduces DMA-buf mapping types so importers and exporters can negotiate alternatives to scatterlist for specialized interconnects. Discussion continues on how that negotiation should work across drivers. The change would affect performance paths that cannot use classic scatterlist.

vc4 job_lock split

A series splits the vc4 job_lock to fix IRQ inversions, seqno races, and preemption issues while cutting contention by roughly ninety percent. The Broadcom VideoCore driver gains finer-grained locking. Platform maintainers should expect shorter lock hold times once the patches land.