Linux 7.2 has been released and includes new features in USB4STREAM, KVM and Ext4

Key points:
  • USB4STREAM integration for direct data transfer between computers via USB4 ports without networking.
  • Critical performance optimizations in Btrfs (large folios by default) and Ext4 (fast commit redesign).
  • KVM hypervisor support for Intel MBEC and AMD GMET extensions, optimizing virtual machines.
  • Reduced swap memory consumption and support for nested schedulers via BPF.
  • Removal of support for legacy i486 processors and purging of the obsolete AppleTalk protocol.

Tux, the mascot of the Linux Kernel

After two intense months of development and more than seventeen thousand bug fixes, Linus Torvalds has officially released the Linux kernel 7.2. The new version not only cleans up thousands of lines of legacy code, but also introduces performance optimizations, improves memory management, and debuts new direct transmission technologies between devices.

At the heart of the iomap mechanism, the elimination of redundant calls has managed to increase input/output operations per second (IOPS) by five percent when using systems such as ext4 and xfs under intensive loads on NVMe drives.

Btrfs now supports large folios by default to maximize sequential input and output, while Ext4 has redesigned its quick commit mechanism to eradicate bottlenecks and double the hash calculation speed in complex directories.

In addition, USB4STREAM is introduced , a mechanism that allows direct data transfer between computers connected via USB4 ports . By exposing a virtual device on the system, users can read and write information between computers using standard commands, completely eliminating the need to configure complex network bridges or sockets.

Furthermore, the swap subsystem has drastically reduced its memory footprint by unifying anonymous and shared memory , saving up to half a gigabyte of RAM for every terabyte of swap mounted. At the processing level, the SCHED_EXT mechanism driven by eBPF now allows the creation of nested schedulers , giving each control group the ability to manage its own tasks in isolation.

The kernel scheduler now supports cache- aware load balancing , grouping intelligent threads to share the same high-level cache and significantly improving data access efficiency. This is further enhanced by the SCHED_EXT mechanism, which now allows the creation of nested subschedulers , giving each control group (cgroup) the ability to manage its own execution rules.

In addition, it is worth noting that an innovative Fair GPU Scheduler was initially introduced, but had to be disabled at the last minute due to regressions that collapsed performance when running games through the Proton compatibility layer, forcing a temporary return to the traditional FIFO model.

Meanwhile, the kernel continues its necessary cleanup, definitively removing support for the old i486 processors that lack a floating-point unit, as well as discontinuing the venerable AppleTalk network protocol after decades of service.

Security and virtualization are advancing by closing historical attack vectors. The KVM hypervisor now leverages Intel MBEC and AMD GMET hardware extensions , enabling independent management of execution rights between the kernel and user space in virtual machines, thus mitigating costly control transfers to the hypervisor.

In the codebase, the unsafe `strncpy` function has been eradicated, an effort that took years and hundreds of patches to replace it with overflow-safe alternatives. Additionally, AF_ALG has been purged of its hardware accelerators, prioritizing software implementations in user space, and the Landlock module has expanded its protection to include strict control over UDP network sockets.

This release will also be remembered for its cleanup of legacy code and adoption of new standards. Support for i486 processors has been dropped , a fate also suffered by the iconic AppleTalk protocol, ARCnet interfaces, and aging ATM network connections.

In contrast, modernity shines with the PPPoE controller, which by embracing the GRO and GSO offloading mechanisms has managed to quintuple the performance in packet reassembly.

Finally, hardware support is crowned with the arrival of HDMI 2.1 FRL technology to the open AMDGPU driver for 8K transmissions, while the Rust-for-Linux ecosystem continues to mature steadily, adding support for the s390 architecture and advancing the integration of the Nova driver to govern modern NVIDIA GPUs based on the Turing architecture and above.

Finally, if you are interested in learning more, you can find the details at the following link.


Add as preferred source in Google