Linus Torvalds released Linux 7.0 on April 12, 2026. The version number bump is cosmetic (Linus has said so himself), but the release is one of the more feature-dense kernels in recent memory, touching the scheduler, memory subsystem, filesystem layer, networking stack, and the language the kernel is written in.

Self-Healing XFS

The headline feature is autonomous online repair for the XFS filesystem. A new background daemon, xfs_healer, listens to kernel error events via fanotify and automatically triggers repairs on detected corruption while the filesystem stays mounted and online.

The mechanism works by walking parent pointer metadata and reverse mapping information to locate damaged structures and reconstruct them from surrounding consistent data. This requires xfsprogs to have built the filesystem with reverse mapping and parent pointer support enabled, features that have been available for several kernel generations but weren’t the default until now.

For production servers running XFS on database volumes or NAS arrays, this is a significant reliability upgrade. Corruption that previously required unmounting, running xfs_repair, and accepting downtime can now be resolved transparently in the background.

xfs_healer is managed by systemd and auto-started on XFS-backed mounts. Check your distro’s xfsprogs version supports the required metadata features before expecting it to activate automatically.

Scheduler: Time Slice Extension

Linux 7.0 fixes a decade-old scheduler behaviour where threads could be preempted mid-critical-section even when using Restartable Sequences (RSEQ). The new Time Slice Extension feature gives threads a brief grace period to complete RSEQ sections before the scheduler intervenes, reducing micro-stutter on CPU-intensive workloads that rely on userspace atomic operations.

The effect is most visible in game engines, database transaction processing, and any workload using lock-free data structures. Benchmarks from Phoronix show measurable latency reduction in PostgreSQL and Redis under contention.

Rust Is Now Official

Linux 7.0 marks the point where Rust infrastructure in the kernel is no longer considered experimental. In-tree Rust bindings for the GPIO, block device, and network driver subsystems are now stable. Several real drivers, including a new Apple Silicon I2C driver, shipped written entirely in Rust.

The toolchain requirement is Rust 1.82 or newer. Distros shipping kernel 7.0 with Rust support compiled in will need to ensure their build infrastructure meets this.

Hardware Highlights

• Intel Nova Lake: Continued enablement for Intel’s next-generation desktop/laptop platform
• Intel Crescent Island: More accelerator bring-up for Intel’s AI inference silicon
• AMD GPU: New IP blocks enabled: covers RDNA 4 features not yet exposed in 6.x kernels
• ARM64 SBCs: Hardware video decode for several Rockchip RK3588-based boards
• Rock Band 4: Bluetooth controller support (yes, really: it works on Linux now)

Intel TSX: Auto Mode Default

Intel TSX (Transactional Synchronization Extensions) now defaults to auto mode rather than disabled. This restores TSX availability on hardware where the speculative execution mitigations don’t require it to be off, recovering performance for workloads that use transactional memory.

Upgrading

Linux 7.0 is shipping as the default kernel in Ubuntu 26.04 LTS and will follow in Debian Testing shortly. Arch and Gentoo users already have access. Out-of-tree kernel modules (DKMS packages, proprietary GPU drivers) will need rebuilds, check your dkms status before rebooting into 7.0 on production machines.