Linux 7.2: The Biggest Changes in the New Kernel Release

This is not a universal performance switch: the benefit depends on the processor topology and workload. It is best understood as a scheduler improvement aimed at systems where moving related work between cache domains creates avoidable overhead.

Storage and filesystem work targets real bottlenecks

Several Linux 7.2 changes focus on I/O paths rather than user-facing features.

• Pipe-write changes are associated with reported throughput improvements of roughly 6% to 28%, depending on the workload and test configuration.
• ext4 receives a reworked fast-commit path intended to reduce lock contention and deadlock risk, along with faster directory-hash processing.
• iomap-related work improves parts of the ext4 and XFS paths, with testing showing gains in some high-IOPS NVMe and io_uring workloads. These results should not be treated as a general promise for every disk or application.
• Btrfs large-folio support is no longer experimental and is enabled by default. Linux 7.2 also introduces experimental huge-folio support up to 2 MB.
• exFAT is converted further to the iomap infrastructure, improving buffered I/O, direct I/O, and related operations. Reports describe substantial gains under heavy-load tests, but the result is workload-specific rather than a guaranteed percentage for all exFAT users.

The practical takeaway is that Linux 7.2's performance story is distributed across several subsystems. Users are more likely to see meaningful improvements when their workloads stress CPU-cache locality, filesystem metadata, high-throughput storage, pipes, or Btrfs memory management.

Hardware support expands across Apple, AMD, Intel, and NVIDIA

The release includes several important enablement milestones:

• Apple M3 MacBooks can boot a mainline kernel to a serial console through upstream Asahi Linux work. This is a significant bring-up step, but it is not the same as complete, polished desktop support for every MacBook component.
• Intel gains support for USB4STREAM, a protocol for sending data streams through USB4 cables.
• AMDGPU receives initial HDMI 2.1 Fixed Rate Link support, while AMD platform code adds early preparation for Zen 6 systems.
• The Intel Xe driver gains initial CRI-platform support, Nouveau adds NVIDIA GA100 compute-accelerator support through GSP, and Apple Silicon receives battery-reporting improvements.
• Rust support is extended to IBM's s390 architecture, broadening the platforms on which kernel components written in Rust can be developed.

Other additions span networking, virtualization, security, embedded systems, and power telemetry, making Linux 7.2 a platform-enablement release as much as a performance release.

Legacy support is removed—and one old x86 assumption becomes explicit

Linux 7.2 removes more than 13,000 lines of i486 x87 floating-point-emulation code. It also drops AppleTalk and support for legacy ISA and PCMCIA ARCnet adapters.

The kernel now requires a usable x86 time-stamp counter, or TSC, unconditionally. Modern x86 systems already generally depend on one, so the change mainly affects exceptionally old or unusual hardware rather than current desktops and servers.

The cycle also included the completion of a long-running cleanup to remove uses of the deprecated strncpy() API from the kernel tree, replacing them with more explicit string-handling functions.

GPU scheduling was pulled back after regressions

Linux 7.2 briefly included a new “fair” GPU scheduler intended to improve the way GPU jobs are ordered. AMD graphics regressions led developers to revert that scheduler to FIFO before release.

That rollback is a useful reminder that a feature appearing in a development cycle is not necessarily ready for stable users. In this case, maintainers prioritized predictable graphics behavior over shipping a more ambitious scheduling model.

A nine-week cycle shows the cost of a busier kernel tree

The Linux 7.2 development cycle lasted about nine weeks and involved more than 2,100 contributors, setting a reported merge-window participation record. Analysis cited by reporting estimated that roughly 5% of commits carried an AI-assistance attribution.

The central concern was not that AI independently authored Linux's headline features. Instead, AI tools are contributing to a larger flow of vulnerability reports, suggested patches, review findings, and follow-up fixes. Some accepted changes address problems identified through AI-assisted analysis, but every proposed change still creates work for human maintainers and reviewers.

Linus Torvalds's description of oversized late updates as the “new normal” reads more as acceptance of a growing operational reality than as an endorsement of lower review standards. The release made it out on schedule, but its late-stage churn demonstrates why kernel maintainers remain cautious about volume, regressions, and patch quality.

What comes next for Linux 7.3 and Ubuntu 26.10?

The Linux 7.3 merge window opened immediately after 7.2, with reporting indicating that about 40 pull requests were already pending. The next couple of weeks will determine which major changes enter that cycle.

For Ubuntu 26.10, Canonical's kernel team has identified Linux 7.2 as the current target version. Ubuntu 26.10 is scheduled for October 15, 2026, with a kernel freeze planned for October 1. That timing makes 7.2 the more established target, but distribution plans can still change if testing exposes serious regressions or integration problems.

For kernel developers, the next step is continued regression testing and review. For users, the most relevant question is not whether Linux 7.2 contains the largest feature list, but whether their hardware and workload align with its strongest improvements: cache locality, filesystem throughput, newer platform support, or better device enablement.