Kernel: BPF arenas, forced bind taint, Rust SPDM

BPF development advanced with a large arena-stack series and a verifier bounds fix, while driver-core tainting and media Kconfig choices drew sharp pushback. Rust SPDM work, memory-management races, and proxy-execution refinements rounded out the day.

BPF arena stack and verifier state joining

A BPF maintainer posted an 81-patch series that enables loading Rust programs through an arena stack together with a new verifier algorithm that joins states. The work expands stack handling and state tracking for more complex in-kernel programs. eBPF and Rust-in-kernel developers gain a substantial foundation to review.

TAINT_FORCED_BIND faces VFIO and SRIOV resistance

Greg KH posted a v4 series adding TAINT_FORCED_BIND when userspace manually binds or unbinds devices and drivers through sysfs. VFIO and SRIOV users called the change a regression and demanded a revert in a heated exchange. The flag would mark kernels where forced binding occurred, altering how stability and support are judged.

BPF verifier retains valid packet pointer bytes

A two-patch series fixes a verifier bug that accepted invalid packet accesses after scalar evolution adjusted umax inside counted loops. The change keeps the valid-byte information of a packet pointer during that evolution. Networking and packet-filter BPF programs receive tighter, correct bounds tracking.

Rust SPDM requester reaches v4

A 22-patch v4 series adds a Rust implementation of SPDM as a kernel requester. The stated goal is safer parsing of complex untrusted authentication data. The work continues Rust adoption in areas where parsing defects have historically been costly.

HVO race surfaces while dropping fake head checks

MM developers discussing a patch that drops fake head checks identified a real race in hugeTLB vmemmap optimization. The race allows a speculative page refcount write against read-only vmemmap. What began as cleanup became a concurrency issue with security relevance.

Proxy execution v33 adds sleeping-owner handling

Version 33 of the proxy-execution series introduces handling for sleeping lock owners. Reviewers examined cycle detection, PSI integration, and activation races. Scheduler and locking specialists continue refining the design ahead of any merge.

Media Kconfig exposure draws Linus objection

Linus Torvalds objected to exposing chip-specific media drivers in Kconfig. Sakari posted a revert of the earlier change that had stopped hiding those options. The dispute concerns how much hardware-specific detail belongs in user-visible kernel configuration.

Large mm fixes series lands on stable

A 46-patch series of memory-management fixes, some labeled urgent, was posted to the stable list. The patches cover many subtle concurrency bugs and include AI-generated fixes under active review by numerous participants. Stable-tree consumers and mm watchers face a sizable batch of corrections to evaluate.