Kbuild speedups, NVIDIA vGPU, and Apple USB4 patches

Kernel traffic concentrated on build-system performance and new device support, with Linus Torvalds reviewing multiple series. Locking RFCs, memory-management changes, and a BPF CVE rounded out the day.

Kbuild series claims large build-time gains

A 21-patch v2 kbuild proposal parallelizes known bottlenecks and reports 36 to 90 percent faster kernel builds. Linus Torvalds and other maintainers are actively reviewing the patches. A separate 20-patch v3 series claims 19 to 36 percent allmodconfig improvements and up to 90 percent on noop builds, with Linus and Kees Cook engaged in the discussion.

NVIDIA posts vGPU manager and VFIO variant driver

NVIDIA submitted a 13-patch series introducing a vGPU manager together with a VFIO variant driver and Rust nova-core components. The posting sparked debate over drvdata refactoring and the state of VFIO Rust bindings. The work matters for virtualized GPU pass-through on systems that already use VFIO.

RFC adds stealable FUTEX_PING locks

An RFC series proposes FUTEX_PING, a stealable futex built on proxy execution for priority-inheritance locks. Peter Zijlstra and others raised deadlock and locking-order concerns. The change would alter how contended user-space locks behave under the kernel futex subsystem.

USB4 and Thunderbolt land for Apple M1/M2/M3

A 22-patch v2 series adds initial USB4/Thunderbolt support, including ACIO, NHI and DART pieces plus device-tree nodes, for Apple M1, M2 and M3 SoCs. The patches constitute the first upstream enablement of these interconnects on the platform. Developers targeting Apple silicon will gain native high-speed I/O paths once the series lands.

Linus NAKs sysctl_field refactoring

Linus Torvalds rejected an RFC that converts SCTP and other net sysctls to a new sysctl_field helper, citing illegible macros and broken indentation. The author acknowledged the problems and began revisions. The exchange highlights ongoing tension over readability of generated sysctl tables.

mTHP support extended beyond PMD pages

A 14-patch v8 series always enables multi-size transparent hugepage support by refactoring has_transparent_hugepage() and introducing pgtable_has_pmd_leaves(). The changes allow mTHP on architectures that lack PMD-sized pages. Memory-management developers gain wider hugepage coverage without waiting for full PMD support.

BPF verifier CVE fixed in stable trees

Stable kernels announced CVE-2026-93127 for a BPF verifier bug that dropped scalar IDs on sign-extending narrowing stack fills. The flaw permitted incorrect scalar tracking and potential out-of-bounds access after certain loads. The fix closes a verifier hole that could be reached by unprivileged BPF programs.