Arm CCA in KVM, GPU scheduler regression, and kernel locking fixes

Arm Confidential Compute Architecture support for KVM advanced through a large revision while a late GPU scheduler regression and several locking and race fixes drew maintainer attention. Real-time IPI work also landed in tip trees.

Arm CCA support for KVM reaches v16

A 45-patch series at version 16 adds Arm Confidential Compute Architecture support to KVM, covering confidential VMs and the Realm Management Monitor. Review included hardware testing feedback across 79 messages from 10 participants. Developers following confidential computing on arm64 will want to track this series as it continues toward mainline.

drm/sched FAIR policy regression on 9070XT

A regression report states that the drm/sched FAIR policy causes serious performance degradation at maximum GPU load on the 9070XT. Maintainers discussed a revert in the days before the 7.2 release. GPU driver users and release managers should note the timing relative to the freeze window.

VLAN offload race fixes skb_under_panic

A v2 networking patch fixes a data race on hard_header_len that can trigger skb_under_panic when toggling hardware VLAN offload, and also addresses a related macvlan layering report. Three participants exchanged a dozen messages on netdev. The change closes a panic path that appears when offload settings change under load.

Livepatch BTI fix for Clang 21 on arm64

A patch corrects livepatch BTI exceptions that break when Clang 21 and later omit BTI landing pads for static functions under CONFIG_RANDOMIZE_MODULE_REGION_FULL. Six participants discussed the issue across 22 messages. Arm64 systems that combine livepatch with newer Clang toolchains are directly affected.

Preemption during IPI waits merged to tip

Version 10 of a 14-patch series that allows preemption while waiting for IPI completion to improve real-time performance was merged into the tip smp/core and x86/mm branches. The work reduces blocking during inter-processor interrupt completion. Real-time and low-latency kernel users gain from the reduced wait-side latency.

AppArmor deadlock in complain-mode change_hat

A bug report identifies a deadlock in AppArmor change_hat under complain mode, traced to recursive acquisition of ns->lock introduced in commit 32e92764d6f8, with mainline affected. The report was filed on the stable list. LSM and security maintainers need to correct the lock ordering before it spreads further into stable trees.

Hot page tracking series draws maintainer debate

Version 8 of an eight-patch series proposing hot page tracking and promotion infrastructure for CXL tiering prompted heated debate among maintainers over pghot versus DAMON and existing NUMA balancing. Seven participants exchanged 39 messages. Memory tiering and CXL users should watch how the competing approaches are reconciled.

pidfd namespace ioctl gains missing exec_update_lock

A patch holds exec_update_lock around pidfd namespace ioctls to match the earlier procfs fix for CVE-2026-64371. Four participants reviewed the change in five messages. The addition closes a locking gap that could otherwise leave namespace operations via pidfd exposed.