Arm CCA in KVM, VLAN races, reclaim and hugetlb fixes

Arm Confidential Compute Architecture support for KVM advanced through large successive patch series under active review. Several fixes addressed data races, null dereferences, page table reclaim errors, and buffer write handling.

Arm CCA Realm support in KVM

Successive revisions of a large arm64 patch series added Arm CCA protected VM and Realm support to KVM, including the RMI firmware layer. The v15 37-patch set and v16 45-patch set received sustained maintainer review across dozens of messages. This work enables confidential computing isolation for virtual machines on Arm hardware.

VLAN offload race and skb_under_panic

A net patch fixes a data race on hard_header_len that triggers skb_under_panic when hardware VLAN offload is toggled, along with related macvlan layering issues. Three participants discussed the change on netdev. The fix stabilizes the network stack against a panic reachable during offload reconfiguration.

Wrong address in direct page table reclaim

A patch corrects an incorrect address passed to pte_free_tlb during direct page table reclaim. Andy Lutomirski, Andrew Morton, and Linus Torvalds joined the stable-list discussion. The change prevents mismanaged TLB flushes in the reclaim path.

Proxy execution meets sched_ext

The v10 patchset makes proxy execution compatible with sched_ext by removing their build-time exclusion, adding hooks, and fixing context tracking. Two participants iterated on the series targeting the 7.3 merge window. Extensible scheduler users gain the ability to combine both features.

Null nodemask panic in hugetlb CMA

A v4 patch eliminates a null nodemask dereference in hugetlb_cma_alloc_frozen_folio. Unprivileged users could trigger a kernel panic via gigantic hugepage allocation on NUMA systems with CMA. The fix closes a local denial-of-service vector in the hugepage allocator.

KVM clock cleanup

David Woodhouse posted the v7 36-patch series that removes pvclock_gtod_data, fixes clock imprecision and discontinuities, and improves migration behavior while adding TSC and guest clock APIs. Two participants reviewed the changes. Virtualization hosts obtain more reliable guest timekeeping.

Buffer write errors retain uptodate state

A 19-patch series stops clearing BH_Uptodate when a write fails and introduces BH_Write_EIO instead, touching many filesystems. Four participants discussed the buffer-layer change. Filesystem error handling becomes more precise without discarding valid page state.