Linux Kernel 7.0 Is About to Ship — Here Is What It Changes for the Infrastructure That Never Sleeps

Linus Torvalds tagged rc7 on April 5, 2026 and called the code « innocuous » — the signal that Linux 7.0 is done. The final release is expected within days. What makes 7.0 stand out is not a single headline feature: it is the convergence of several long-running engineering projects all reaching maturity in the same release window. Rust graduates from experimental to core, the EEVDF scheduler becomes the sole CPU scheduler in the tree, XFS learns to heal itself online, and a record 2,362 developers shipped code in this cycle.

The kernel now clocks in at 39.2 million lines of source (29.6M of actual code, 4.7M comments, 4.9M blank lines). 225 companies funded the work, led by Intel (10.8% of changesets), Google (7.5%), and AMD (6.6%). A striking 489 contributors submitted their first-ever kernel patch during this cycle — a number Jonathan Corbet at LWN attributes in part to the growing use of AI tools to identify corner cases that would previously have gone unnoticed.

For server administrators, the bottom line is simple: Linux 7.0 is a mechanical gain. More performance without configuration changes, a shrinking attack surface at the driver level, and filesystems that survive metadata corruption without unmounting. Here is what actually changes.

EEVDF Replaces CFS Permanently

The Earliest Eligible Virtual Deadline First (EEVDF) scheduler, introduced incrementally starting with Linux 6.6 (October 2023), becomes the sole scheduler in 7.0. The Completely Fair Scheduler (CFS), which has been running Linux since 2007, is removed from the tree.

The difference is measurable. EEVDF assigns each task a virtual deadline — the point by which it must receive its CPU share — and always serves the task with the nearest deadline first. Under desktop workloads, this translates to 15–25% lower latency according to early benchmarks. For servers, the gain is subtler but real: interactive tasks (HTTP requests, SSH connections) no longer queue behind batch jobs.

The scheduler also ships with a complete hybrid CPU routing overhaul, co-designed with Intel for the Nova Lake platform. Foreground tasks (audio, UI rendering, game engines) are now dynamically pinned to P-cores, while background work (indexers, backups, package managers) migrates to E-cores. Early Phoronix benchmarks show 8–12% better battery life on hybrid-architecture laptops under typical desktop workloads.

For operators managing container fleets, the message is clear: the new scheduler balances load better without requiring you to reconfigure a single cgroup.

Rust Is No Longer an Experiment

Rust entered the kernel with Linux 5.19 in 2022, explicitly marked experimental. Three and a half years later, that label is formally removed. Rust-based drivers and kernel modules are now first-class citizens in the kernel tree.

The operational implications are immediate:

• The Rust NVMe driver is stable and mainline. It delivers the same throughput as its C equivalent, with memory-safety guarantees that eliminate use-after-free and buffer overflow bugs at the storage layer — no runtime overhead.
• The Nova GPU driver (the NVIDIA open-source effort in Rust) continues to mature with improved command submission handling.
• PCI, DMA, and interrupt abstractions are now fully covered by the kernel’s Rust API. A developer can write a complete hardware driver without touching a line of C.

Why this matters: 70% of kernel CVEs are memory-safety bugs (Google Project Zero, 2023). Every driver rewritten in Rust mechanically shrinks the kernel’s attack surface. Linux 7.0 does not rewrite the entire kernel in Rust — nobody is claiming that — but it validates the infrastructure that makes such a transition possible, driver by driver, release by release.

XFS Heals Without Unmounting

This is the most concrete change for server operators. XFS, the default filesystem on RHEL and the majority of Linux server deployments, now includes built-in online self-healing.

Before 7.0, XFS metadata corruption — from a power failure, hardware fault, or bad sector — would invariably result in:

XFS (sda1): Metadata corruption detected at xfs_inode_buf_ops
XFS (sda1): Unmount and run xfs_repair

That Unmount is the entire problem. Unmounting the root volume on a production server means booting from rescue media and running xfs_repair manually — minutes of downtime for a corruption that might only affect a single inode.

With 7.0, XFS cross-references its redundant metadata structures, reconstructs the correct data, and writes the repaired block back without unmounting:

XFS (sda1): Repairing corrupt inode btree in AG 2.
XFS (sda1): Self-healing complete. Filesystem operational.

The administrator gets a log entry; the filesystem keeps running. Across a fleet where every unmount is a human intervention, this directly reduces operational risk.

Swap Doubles in Speed, Networking Follows

Swap table phase II lands in 7.0. Phase I (Linux 6.18, December 2025) had already delivered 5–20% gains on memory-heavy workloads. Phase II cleans up and accelerates the swapping code further: benchmarks on Redis show an additional 20% speedup for memory-saturating tasks. zram also gains compressed data writeback — pages are no longer decompressed before being written to physical storage, saving CPU cycles on mobile devices.

On the networking side, io_uring v3 introduces zero-copy networking: web servers and databases can push data directly to the network card without an intermediate kernel-space copy. Nginx benchmarks show 40% more throughput. AccECN (Accurate Explicit Congestion Notification) is now enabled by default, improving TCP congestion control. The UDP layer has been rearchitected to reduce lock contention on multi-core systems — a direct gain for DNS resolvers, VPN gateways, and real-time telemetry pipelines.

RISC-V Reaches the Desktop, Apple Silicon Solidifies

Linux 7.0 adds mainline support for the SpacemiT K3, the first RISC-V RVA23 desktop-class processor. RISC-V, long confined to embedded systems, now has a genuine desktop target with upstream kernel support — mainstream distributions can ship RISC-V builds without out-of-tree patches.

On the Apple Silicon side, support continues to mature. The NVMe, GPIO, I2C, and SPI controllers on M4 and M5 chips receive driver updates this cycle, bringing full mainline support closer — roughly comparable to what the x86 embedded controller ecosystem gets. Asahi Linux integrates these contributions as they land; 7.0 consolidates the low-level infrastructure.

Hybrid chips are not left behind: the Snapdragon X2 Elite (Qualcomm) receives initial support, and AMD ROCm benefits from kernel changes that will enable native packaging in Ubuntu 26.04.

Security Advances on Three Fronts Simultaneously

Hardware shadow stacks are now enabled by default on x86_64, blocking ROP (Return-Oriented Programming) attacks at the processor level. KASLR (Kernel Address Space Layout Randomization) entropy increases from 9 bits to 13 bits, making brute-force address attacks 16× harder.

Landlock, the per-process sandboxing LSM, receives finer-grained filesystem permission controls. And ML-DSA post-quantum signatures (FIPS 204) are now supported for kernel module authentication — a security primitive designed to withstand attacks from quantum computers capable of breaking current RSA and ECDSA signatures.

The development tooling keeps pace: Clang can now perform static locking analysis during kernel compilation, catching potential deadlocks before the code ever runs.

Verdict: Upgrade Now or Wait for 7.1

Linux 7.0 is not a Long Term Support release — no LTS has been announced for the 7.x branch yet. The question for an administrator is not « should I upgrade? » but « when? »

Upgrade now if you run database servers or memory-intensive workloads on XFS. The swap table phase II gains and XFS self-healing are zero-config improvements — you get them without changing anything in your application stack. Users of hybrid-architecture workstations (Intel P-core/E-core) will also see better battery life and responsiveness.

Wait for 7.1 if you are on production-critical deployments with exotic hardware. The 7.0 cycle was unusually busy — rc3 and rc4 were « some of the biggest in recent history, » per Torvalds — and early community feedback reports minor boot issues on custom builds. Stable distributions (Ubuntu, Debian, RHEL) will not package 7.0 until the 7.1 cycle begins, expected around June 2026.

For homelabbers and enthusiasts: go for it. The scheduler alone justifies the update on a desktop, and Arch Linux already has packages available via the AUR.

What makes 7.0 remarkable is not a version number — Linus himself notes that he bumped the major « purely because I get easily confused with big numbers. » What is remarkable is that in a single two-month cycle, the kernel simultaneously improved scheduling, filesystem reliability, networking throughput, swap performance, security hardening, and driver coverage — without breaking anything. That is the definition of a mature infrastructure project.

References

• Linus Torvalds — Linux 7.0-rc7 announcement — LKML, April 5, 2026
• Jonathan Corbet — Development statistics for the 7.0 kernel — LWN.net, April 13, 2026
• LWN — 7.0 Merge window, part 1 — February 2026
• LWN — 7.0 Merge window, part 2 — March 2026
• OSTechNix — Linux Kernel 7.0 is Officially Released — April 14, 2026
• FOSS Linux — Linux Kernel 7.0: The Complete Feature Breakdown — April 15, 2026
• SudoFlare — Linux Kernel 7.0: Changes for Developers — April 2026
• CNX Software — Linux 7.0 Release: Main Changes, Arm, RISC-V, and MIPS — April 13, 2026
• Phoronix — Modern AMD Graphics Driver Surpasses Six Million Lines — March 2026
• Rust for Linux — NVMe Driver — rust-for-linux.com
• LKML — Linux 7.0 official release — April 12, 2026