Kernel verifier, VLAN, and IPsec fixes with subsystem cleanups

Security-oriented backports and race fixes led Kernel and low-level traffic, alongside proposals to retire complex paths in ext4, KVM clocks, and DMA completions. Several series also drew maintainer pushback on scope and necessity.

BPF verifier backport for ld_abs and ld_ind in subprogs

A pair of patches backports a verifier fix for the 6.6, 6.12, and 6.18 stable trees so that failure paths of ld_abs and ld_ind instructions are properly simulated inside subprograms. The change addresses CVE-2026-53090. Readers relying on BPF in production kernels gain a corrected analysis path that prevents incomplete verification of those legacy load forms.

Ext4 RFC to retire inline data writes for regular files

An RFC series proposes phasing out the inline_data write paths used by regular ext4 files. The motivation cites more than fifteen recent bug fixes and eight CVEs that stem from the locking complexity of those paths. Developers maintaining ext4 or filesystems that interact with it should watch the discussion, as the change would simplify a long-standing source of defects.

VLAN offload race fix for skb_under_panic

A netdev patch corrects a data race on hard_header_len that triggers skb_under_panic when hardware VLAN offload is toggled, and it notes a related macvlan layering issue. The fix closes a crash window visible under concurrent configuration changes. Network stack and driver authors who enable or disable VLAN offload at runtime have a concrete race to absorb.

KVM clock cleanup series reaches v7

David Woodhouse posted the seventh version of a 36-patch series that removes pvclock_gtod_data, tightens TSC and guest clock APIs, and addresses imprecision, discontinuities, and migration problems in the KVM clock. The work aims to replace an accumulated set of ad-hoc mechanisms with clearer interfaces. Virtualization maintainers and guests that depend on accurate timekeeping stand to benefit from the reduced mess.

DMAengine migration from tasklets to WQ_BH

A 64-patch v2 series moves DMAengine client completion handling from per-driver tasklets onto a centralized per-channel WQ_BH workqueue model, covering virt-dma and more than ten drivers. The conversion standardizes bottom-half processing that had diverged across implementations. Driver authors using the DMA engine subsystem will need to adapt to the new completion path.

Dispute over removal of the qce crypto driver

A patch that simply deletes the already-BROKEN qce crypto driver prompted a heated exchange among maintainers about in-kernel usage, removal policy, and possible future security features. Opponents argue the driver still has value despite its broken state. Crypto subsystem observers should note the policy tension around keeping versus purging non-functional hardware support.

IPsec and XFRM fix pull for the day

Steffen Klassert sent an eight-patch pull covering OOB write, use-after-free, double-free, deadlock, and NULL dereference issues in the XFRM and IPsec stack. The set consolidates several independent defect reports into a single net tree update. Operators and developers of IPsec configurations gain a batch of memory-safety and locking corrections.

Hot-page tracking infrastructure meets resistance at v8

The eighth version of a patchset introduces a pghot subsystem for tracking and promoting hot pages. Matthew Wilcox objected that customers do not require the feature, keeping the temperature elevated. Memory-management readers will want to follow whether the infrastructure lands or is scaled back in light of the stated lack of demand.