Kernel: hazard pointers debated, kbuild speedups, BPF and arm64 work

Kernel traffic on 2026-09-20 focused on concurrency primitives, build-system throughput, and cross-architecture alignment. Several multi-patch series drew active maintainer review across core subsystems.

Hazard-pointer RFC draws Linus critique

Paul McKenney and Mathieu Desnoyers posted an RFC v3 series of 28 patches adding a simple hazard-pointer implementation plus torture tests. Linus Torvalds questioned the real-world value of the work on linux-kernel. Developers following lockless synchronization now have an open debate on whether the mechanism merits mainline inclusion.

kbuild series claims large build speedups

A 20-patch v3 kbuild series asserts 36-90 percent faster kernel builds through changes to dependency checking, ELF parsing, and linking. Maintainers are reviewing the patches on linux-kernel. Regular kernel builders stand to gain shorter clean and incremental compile times if the series lands.

BPF LSM can set xattrs at inode creation

An 8-patch bpf-next series introduces the bpf_inode_init_xattr kfunc, letting BPF LSM programs set extended attributes atomically inside inode_init_security. Six participants discussed the proposal. The change closes a race window for security modules that must initialize xattrs at creation time.

arm64 converts to generic entry

Version 18 of a 14-patch series moves arm64 onto the generic entry framework already used by x86, RISC-V and others. The conversion shrinks architecture-specific syscall paths. Shared entry code lowers long-term maintenance cost across architectures.

perf finishes move to standalone scripts

A 49-patch v1 series completes perf's transition away from embedded Python and Perl interpreters, routing scripts through a new perf C extension module. The large patchset was posted by a single author. Tooling users obtain a cleaner boundary between the perf binary and its scripting support.

AMD hardware vIOMMU support reaches v5

AMD posted a 24-patch v5 series implementing hardware-accelerated virtualized IOMMU support for guest command, event and PPR logs plus nested translation. Discussion involved two participants. Virtualization stacks on AMD platforms can reduce hypervisor overhead for IOMMU operations once the work merges.

BPF throw paths gain cleanup landing pads

A v2 20-patch bpf-next series adds .bpf_cleanup tables and JIT support so bpf_throw() executes landing-pad destructors on unwind. The work unblocks Rust panic handling inside BPF programs. Four participants reviewed the changes.

NOMMU GUP honours FOLL_PIN

A v3 mm/gup patch makes the NOMMU __get_user_pages_locked path honour FOLL_PIN, correcting a refcount mismatch. The fix prevents use-after-free with io_uring fixed buffers. Three participants handled the security-relevant correction.