Kernel THP swap, bnxt regression, GSO fixes

Kernel development on 2026-10-08 centered on memory management for transparent huge pages, a networking regression in bnxt_en, and a merged security fix for untrusted GSO handling. Additional work covered IOMMU live update, KVM TLB optimization, and a heated debate on cgroup swap semantics.

PMD-level swap for anonymous THPs

A v8 series of 30 patches proposes PMD-level swap entries for anonymous transparent huge pages so that swapped THPs retain huge mappings across swap-out and swap-in rather than splitting to PTEs. The work targets the mm subsystem and aims to preserve performance characteristics that would otherwise be lost when large pages leave and re-enter memory. Readers tracking THP efficiency should note the series remains under review on linux-kernel.

bnxt_en DMA faults after LL_RESERVED_SPACE change

A regression bisected to commit 447cbe95ebb9 causes IOMMU DMA faults on macvlan and vlan setups with the bnxt_en driver. Discussion on netdev produced and tested a patch that switches to skb_put_padto() to restore correct padding behavior. Operators of Broadcom NetXtreme hardware with virtual networking should watch for the fix landing.

GSO dissection always applied in virtio_net_hdr path

A three-patch series ensures __virtio_net_hdr_to_skb() always dissects GSO packets, closing a bypass of untrusted GSO flow dissection, and corrects a u16 truncation bug in __skb_flow_dissect. The series was merged into net.git. The change matters for anyone relying on correct handling of guest-supplied GSO headers in virtio networking.

IOMMU state preservation for live update

Version 5 of an 18-patch series adds live-update state preservation to the IOMMU subsystem for Intel VT-d and for devices attached through iommufd. The goal is to keep IOMMU configuration intact across kexec-based live updates. Systems that depend on uninterrupted DMA remapping during kernel live update are the primary audience.

nSVM TLB flush optimizations in KVM

A v2 series of 29 patches optimizes nested SVM TLB flushes by introducing per-vCPU L2 ASIDs and removing dynamic allocation. Benchmarks cited in the discussion show gains in the 8-15 percent range. KVM hosts running nested AMD guests stand to benefit once the work is accepted.

Debate over virtualized swap cgroup semantics

Kernel mm developers are debating a path forward for virtualized swap, including possible redefinition of memory.swap cgroup accounting for virtualized and zswap entries. Google participants warned that changes risk breaking existing cgroup v1 and v2 deployments. The long thread (141 messages) signals unresolved tension around how swap is charged in containerized environments.

Seamless Mobility Domain AP support

A 16-patch series for wireless-next implements 802.11bn Seamless Mobility Domain AP roaming support across nl80211, cfg80211, and mac80211. The work enables AP-side SMD behavior for client roaming. WiFi stack maintainers and vendors targeting next-generation roaming features are the relevant audience.

Always-on mTHP support

Version 8 of a 14-patch set decouples has_transparent_hugepage() from PMD-sized page availability so that multi-size THP remains usable when PMD pages cannot be allocated. The change keeps mTHP functional on configurations that lack full PMD THP. Memory-management users who rely on smaller huge-page sizes should track the series.