Emacs 31 blockers and DRM driver patches

Emacs maintainers triaged release blockers and LLM guidance while DRM developers advanced Rust abstractions, SVM support, and fault handling series. Graphics work centered on GPU memory paths, io_uring integration, and cgroup protection under contention.

Emacs 31 release-blocking bugs

Emacs maintainers listed and triaged bugs blocking the 31.1 release on the emacs-devel list. A security mitigation for a vc-hooks vulnerability is also under discussion. Developers tracking the next major Emacs version have a clear view of remaining blockers and related hardening work.

AGENTS.md for LLM tool steering

Emacs developers discussed adding an AGENTS.md file to steer LLM tools toward policy compliance instead of unrestricted code generation. The proposal would give automated agents explicit project norms to follow. Projects that accept AI-assisted patches gain a lightweight way to encode contribution rules.

Rust dma_fence abstractions at v8

A v8 series adds Rust dma_fence abstractions for dma_buf use by GPU drivers. AI review flagged multiple soundness bugs and deadlock risks in the patches. Rust-for-Linux graphics work must address these issues before the abstractions can land safely.

AMDGPU SVM via shared DRM GPUSVM

AMD posted a v9 series that implements SVM through the shared DRM GPUSVM core rather than KFD-private code, limited to the single-GPU XNACK-on case. The change moves AMDGPU shared virtual memory onto common infrastructure. Users of AMD GPUs gain a clearer path to unified memory features that align with other DRM drivers.

Dmabuf read/write through io_uring

A v5 patchset adds dmabuf-backed registered buffers that work with io_uring fixed read and write operations on NVMe. Key maintainers have reviewed the series. Storage and graphics stacks that already use dmabuf can now exercise direct I/O paths without extra copies.

Writable PTE insertion for DRM faults

A request-for-testing series exports vmf_insert_pfn_mkwrite and updates DRM shmem and vmwgfx to insert writable PTEs directly. The change ensures a .fault handler need not be followed by .pfn_mkwrite for write faults, fixing KVM and VFIO write faults on GEM mappings. Virtualization and DRM combinations become more reliable under write fault load.

pfn_mkwrite interaction with fixup_user_fault

Paolo Bonzini posted a v2 six-patch series that corrects the interaction between .pfn_mkwrite and fixup_user_fault for VM_IO|VM_PFNMAP mappings used by KVM and DRM. The fixes close edge cases that arise when user faults are repaired. KVM and graphics maintainers obtain a cleaner contract for PFNMAP fault handling.

TTM cgroup protection under VRAM pressure

A v8 patchset adds cgroup min and low eviction logic to TTM so that dmem protection is honored when VRAM is contended. The series improves isolation for memory-limited GPU workloads. Containers or cgroups that set dmem limits receive more predictable behavior under memory pressure.