On November 30, Linus Torvalds released Linux kernel 6.18, the final stable kernel of 2025 and a strong candidate to become the next Long Term Support (LTS) series. The update brings a quieter set of performance refinements, new hardware enablement, and a notable removal: the bcachefs filesystem has been yanked from the mainline tree and moved to external DKMS maintenance.

The Big Changes Under the Hood

Linux 6.18 is not a headline-grabbing release, but it packs several practical engineering wins that will directly benefit servers, laptops, handhelds, and embedded devices.

Memory Allocator Overhaul with Sheaves

The kernel’s slab allocator gained a significant performance upgrade with the introduction of sheaves—per-CPU cached arrays of freed objects—and a per-NUMA pool called barns. Together, they reduce contention and allocation latency by turning many hot-path allocations into simple local pointer operations instead of lock-intensive atomic operations. Benchmarks published with the patch series show measurable gains in allocation-heavy workloads, particularly on large multi-CPU systems. The feature is opt-in per cache, allowing distributions and operators to test carefully before enabling it broadly.

Persistent Memory Caching with dm-pcache

A new device-mapper target, dm-pcache, uses DAX-capable persistent memory (PMEM) as a high-throughput, crash-consistent write-back cache in front of slower block devices. It includes CRC verification and multi-tree indexing for parallel access. For systems equipped with Optane-like PMEM modules, this extends the storage stack with a flexible acceleration tier. Storage administrators should validate failure scenarios thoroughly before trusting dm-pcache in production, as PMEM remains a relatively niche hardware class.

exFAT Speed Boost and Filesystem Improvements

The exFAT driver received an allocation bitmap optimization that makes loading large partitions dramatically faster. In some benchmarks with large size and small cluster size, mount-time operations ran up to 16 times quicker, as reported by Phoronix. This matters for cameras, mobile devices, and embedded systems that rely on SD cards and microSD storage.

XFS continues its march toward enterprise-grade live metadata repair. The online check and repair functionality now covers more scenarios, letting administrators scrub and fix filesystem inconsistencies without taking large storage volumes offline—a critical availability win for busy servers.

Other filesystem updates: Btrfs saw parallelism gains in certain internal paths, ext4 received functional tweaks, FUSE gained 64-bit copy_file_range support and better virtio-fs handling, and NFSv4 caching can now be completely disabled when desired.

Signed eBPF Programs and AccECN Networking

Preliminary support for cryptographically signing eBPF programs has landed. The kernel now accepts signatures attached to BPF object files via libbpf and bpftool, and validates them against kernel keyrings. This lets security-conscious operators enforce that only trusted, vetted code can be loaded into the kernel, reducing exposure from unprivileged BPF execution. Organizations will need to integrate signing into CI pipelines and manage keys carefully.

The networking stack also gained AccECN (Accurate Explicit Congestion Notification), a TCP feedback mechanism standardized by the IETF in 2025. It enables richer congestion signals for advanced data-center congestion control schemes but requires end-to-end support and conservative deployment.

Rust Milestones: Binder and Mali GPU

The long-awaited Rust version of Android’s Binder IPC driver has been merged. Binder is central to Android’s security model, and rewriting it in Rust aims to eliminate entire classes of memory-safety bugs that have historically plagued the subsystem. The new driver is functionally compatible with its C counterpart.

An early-stage Rust DRM driver for Arm Mali GPUs, named Tyr, also arrived. It supports basic power management and GPU metadata queries and marks a cautious but real step toward Rust-based graphics drivers.

Hardware Enablement: From Handhelds to Apple Silicon

A wide range of hardware now works better out of the box with a stock 6.18 kernel:

  • Laptops and handhelds: ASUS, Lenovo, and GamePad Digital gaming handhelds get improved support; Sony DualSense controllers gain audio jack hot-plug detection and haptics updates; haptic touchpad groundwork appears; Dell/Alienware, HP Omen, and ASUS ROG motherboards receive WMI hooks for better monitoring.
  • Graphics: Nouveau defaults to using NVIDIA’s GSP firmware on Turing/Ampere for improved reclocking and power management; the Rockchip NPU (RKNN) kernel driver and Mesa userspace “Rocket” Gallium driver are upstreamed for AI acceleration on RK3588 boards.
  • Arm ecosystem: Apple M2 device tree and SoC support continues to improve thanks to the Asahi Linux project; the kernel now detects when running under the FreeBSD bhyve hypervisor, including VMs with over 255 vCPUs.

The bcachefs Exit

The most visible change in 6.18 is the removal of the bcachefs core code from the mainline kernel. Introduced experimentally in 6.7 and flagged as “externally maintained” earlier this year, bcachefs is now distributed only as an out-of-tree DKMS module. Its maintainers provide packages for Debian, Ubuntu, Fedora, and openSUSE, and it remains in Arch and NixOS. For users running bcachefs as a root filesystem, this means extra packaging steps and a shift in upgrade expectations—distributions that forbid out-of-tree modules will not ship it by default.

Practical Impacts: Who Gets What

Home and Enthusiast Users
If you use Linux on a laptop, handheld gaming device, or camera with SD cards, you’ll notice faster media mounting, better touchpad feedback, and improved DualSense controller audio and haptics. All of this arrives without manual driver installation once your distribution picks up 6.18.

System Administrators and Data Center Operators
The sheaves allocator improvement can reduce tail latencies on allocation-heavy services like databases or caching layers. XFS online repair shortens maintenance windows for large storage arrays. The arrival of signed eBPF demands a strategic decision: if you run Cilium, Falco, or custom BPF probes, plan a key-management and signing policy now. If you rely on bcachefs, begin testing DKMS-based deployments and verify your recovery procedures under the new packaging model.

Embedded and IoT Developers
Upstreamed Rockchip NPU support and the exFAT speedup are direct wins. Rust Binder offers a safer IPC path for Android-based embedded systems, though you’ll need to ensure your toolchain includes Rust support.

The Backstory: Why bcachefs Left and Rust Arrived

Linux 6.18 fits a longer pattern: the project increasingly demands that in-tree code is actively maintained in cooperation with kernel maintainers. bcachefs fell out of step with that expectation, leading to its removal. Meanwhile, Rust driver work—a multi-year effort—reached a milestone with Binder, proving that memory-safe driver development can happen inside the kernel. These are not dramatic shifts but deliberate, incremental governance choices.

The kernel’s LTS appointment process is separate from the release itself. While 6.18 is widely expected to become the next LTS, the official designation will come later from the LTS working group. Rolling-release distributions (Arch, openSUSE Tumbleweed, Fedora Rawhide) will adopt 6.18 within weeks; enterprise distributions like RHEL and Debian are likely to skip it in favor of their designated LTS kernels, while Ubuntu’s upcoming 26.04 LTS may include 6.18 or a newer point release.

Your Move: What to Test and When

  • For everyone: Boot a live image or test kernel on a spare machine. Exercise your usual workloads—especially storage I/O, virtualization, and any out-of-tree kernel modules you depend on.
  • bcachefs users: Install the DKMS packages, rebuild your initramfs, and confirm that mount, scrub, and recovery still work. If you rely on bcachefs as a root filesystem, this is non-negotiable before upgrading the kernel.
  • eBPF-heavy environments: Decide whether to require signed BPF programs. If so, integrate signing into your artifact pipelines now, and test whether third-party tools still load correctly.
  • PMEM and dm-pcache: If you have persistent memory hardware, allocate time to stress-test failure modes—power loss, metadata corruption, and cache eviction under load.
  • Graphics and firmware: Nouveau users should ensure the GSP firmware is installed; Rockchip NPU users need the matching Mesa 25.x userspace.1

Looking Ahead

Keep an eye on the LTS announcement and your distribution’s mailing lists. The sheaves allocator will likely see fine-tuning over the 6.18.y stable series, and the Rust Mali driver will mature with future merges. Meanwhile, the bcachefs community must demonstrate that an out-of-tree model can sustain filesystem development. For most users, 6.18 is a safe, beneficial upgrade that rewards methodical testing with real-world performance gains.

  1. As first reported by The Register and confirmed by Kernelnewbies, the full list of changes is extensive; this analysis focuses on changes with immediate practical impact. ↩