Kernel: VMA locks, Apple USB4, CPA and TCP fixes

Kernel development on 2026-09-08 centered on memory-management cleanups, early Apple Silicon USB4 enablement, and several security-related fixes landing or under active review. Arm CCA work for KVM and ftrace improvements also advanced.

Unconditional per-VMA locks advance

A v6 patchset makes per-VMA locks unconditional across all kernel configs, removing configuration-dependent paths. The change enables simplifications in binder and networking code and has been picked up by Andrew Morton. Developers tracking mm and binder subsystems will want to watch the resulting cleanups.

Initial USB4 support for Apple M-series

A 22-patch v2 series introduces the first USB4 and Thunderbolt driver plus device-tree nodes for the ACIO and NHI blocks on Apple M1, M2, and M3 SoCs. The work opens host connectivity paths previously missing on these platforms. Readers building or porting kernels for Apple Silicon gain a concrete base for further peripheral support.

x86 CPA race and UAF fixes merged urgently

Urgent fixes for races and use-after-free issues in the x86 change-page-attributes path reached tip/x86/urgent after earlier misapplied commits and tree problems were corrected. The series drew heated discussion across ten participants. x86 maintainers and anyone running recent tip trees should treat the merge as high priority.

TCP IPv6 race fix queued for stable 6.12

A stable backport addressing a potential race in tcp_v6_syn_recv_sock() (CVE-2026-43198) has been queued for the 6.12 series following a syzbot report. The four-message thread confirms the fix path for long-term kernels. Operators of 6.12-stable and IPv6-heavy workloads have a concrete CVE remediation to track.

Arm CCA Realm support reaches v16 in KVM

The large 45-patch v16 series continues adding Arm Confidential Compute Architecture and Realm support to KVM, with active review focused on RMM versioning and IPA handling. Twelve participants kept the 111-message thread moving. Virtualization developers targeting Arm CCA need the current RMM and IPA direction.

drm/sched fence timeline UAF fix

A v4 series fixes a use-after-free of the fence timeline name inside drm/sched, a path shared by multiple drivers and linked to earlier CVEs. Discussion among seven participants refined the approach. GPU and display driver maintainers should verify the fix against their scheduler users.

arm64 ftrace CALL_OPS under kCFI

An RFC series enables CALL_OPS for arm64 kernels built with kCFI by using Clang 21 prefix sections, cutting BPF trampoline overhead by roughly 10 ns. The work targets both ftrace and BPF performance on hardened arm64 builds. BPF and arm64 tracing users gain a measurable overhead reduction path.

Costly GFP_NORETRY compaction skip

A v4 mm patch clears __GFP_DIRECT_RECLAIM for costly __GFP_NORETRY allocations so they skip direct compaction, addressing an iomap regression. Seven participants reviewed the change. Filesystem and memory-allocation paths that rely on iomap benefit from the corrected reclaim behavior.