Linux 7.2 is now available as a stable release, bringing changes of particular interest to system administrators, infrastructure teams and Linux users working with many-core servers, Btrfs storage or virtualization. The new kernel improves how the scheduler takes advantage of CPU cache topology, introduces memory reclaim enhancements, expands KVM capabilities and updates several storage and networking subsystems. On desktop systems, AMDGPU also takes a long-awaited step towards HDMI 2.1 FRL support.
Linux 7.2 highlights in 20 seconds
- The new Cache Aware Scheduling mechanism tries to keep related workloads within the same LLC cache domain.
- Btrfs enables large folios by default and continues advancing support for larger folio sizes.
- MGLRU receives changes designed to improve memory reclaim and writeback in some scenarios.
- KVM adds new capabilities for AMD and Intel processors.
- AMDGPU introduces the foundations for HDMI 2.1 FRL, although some features are still maturing.
For system administrators, the importance of Linux 7.2 lies less in a single headline feature and more in changes to components that directly influence the everyday behaviour of servers and virtual machines: scheduling, memory, filesystems, virtualization and networking.
| New feature | Area | Why does it matter? |
|---|---|---|
| Cache Aware Scheduling | CPU | Reduces movement between LLC domains for suitable workloads |
| Large folios in Btrfs | Storage | Allows memory and I/O to be processed using larger units |
| Huge folios up to 2 MiB in Btrfs | Storage | Opens the door to lower overhead for certain workloads |
| MGLRU improvements | Memory | Refines page reclaim under memory pressure |
| Larger NFS blocks | Network/storage | Can improve NFS transfers on systems with sufficient memory |
| KVM improvements | Virtualization | Expands VM execution and isolation capabilities |
| HDMI 2.1 FRL in AMDGPU | Graphics | Brings Linux closer to modern HDMI 2.1 display capabilities |
| MPTCP, PPPoE and TCP-AO improvements | Networking | Expands the capabilities of the Linux network stack |
| Rust for s390 | Architectures | Extends Rust support to IBM Z environments |
Cache Aware Scheduling: the scheduler starts looking beyond the CPU core
One of the most technically interesting additions in Linux 7.2 is Cache Aware Scheduling, work developed primarily by Intel engineers that allows the scheduler to take the layout of the Last Level Cache (LLC) into account.
The idea addresses a familiar problem in modern many-core systems. Finding an idle CPU does not necessarily mean that moving a thread to it is the most efficient decision.
Some processor architectures contain multiple shared cache domains. When a multithreaded process works with common data, moving its threads between these domains can cause information to be fetched again from slower levels of the memory hierarchy.
Linux 7.2 introduces infrastructure designed to keep certain related threads within the same LLC domain whenever doing so makes sense.
The kernel does not apply this policy indiscriminately. It considers workload characteristics and effective cache capacity before deciding whether grouping threads is worthwhile. Single-threaded processes, for example, generally do not justify the additional tracking overhead.
Linux also avoids concentrating workloads when their memory footprint could substantially exceed the available LLC capacity.
For server administrators, this behaviour could be particularly interesting on machines with large core counts and complex cache topologies, although the actual benefit will depend entirely on the workload. It should not be interpreted as a universal performance improvement for every server running Linux 7.2.
Btrfs gives larger folios a more prominent role
Btrfs accounts for another significant part of the work in this release.
Linux has spent several kernel versions expanding its use of folios, an abstraction that can manage groups of memory pages more efficiently. Linux 7.2 continues that transition and enables large folios by default in Btrfs.
The filesystem also adds support for huge folios of up to 2 MiB, although this should be distinguished from the more mature functionality and remains an area where some caution is appropriate.
The basic objective is straightforward: certain operations can work with larger chunks of data instead of constantly handling small pages, potentially reducing internal operations and memory-management overhead.
There are also specific I/O path optimizations.
Among the changes included in Linux 7.2 are development tests showing around a 15% improvement in sequential writes in a particular scenario after changing the size of certain operations during writeback.
Another change removes unnecessary serialization of direct I/O operations left behind after an API migration. Tests associated with that patch recovered as much as 60% of throughput in that specific scenario.
These figures come from specific development tests. They are not a promise that every Btrfs server will automatically gain 15% or 60% performance after upgrading.
For production environments, that distinction matters: Linux 7.2 contains promising optimizations, but their behaviour should be validated against each infrastructure’s workloads, storage devices and configuration.
MGLRU improves behaviour when memory starts running low
There is also important work in Multi-Gen LRU (MGLRU), the mechanism that helps the kernel determine which pages should remain in memory and which can be reclaimed as memory pressure increases.
One of the changes affects dirty pages waiting to be written.
With MGLRU, there were situations where the mechanism responsible for triggering writeback could react later than with the traditional LRU implementation. Linux 7.2 changes part of this behaviour so memory reclaim can respond better when write activity is high.
On servers with plenty of RAM, these situations may go unnoticed much of the time. The behaviour of the subsystem becomes far more relevant when multiple processes compete for memory, the page cache grows and the kernel needs to reclaim RAM while I/O operations continue.
Linux 7.2 also includes swap-related changes and continues developing support for multi-size Transparent Huge Pages through khugepaged.
NFS, SMB and networking also receive updates
The release is not limited to local storage.
NFS increases the default size of certain blocks to 4 MB on systems with at least 16 GB of RAM, a potentially relevant change for servers using Network File System storage.
SMB, meanwhile, receives improvements related to compressed file handling.
On the networking side, Linux 7.2 adds IPv6 address signalling support for Multipath TCP (MPTCP), as well as GRO and GSO support for PPPoE.
GRO (Generic Receive Offload) and GSO (Generic Segmentation Offload) allow network traffic to be aggregated or segmented more efficiently within the network stack, reducing some of the packet-processing overhead in suitable scenarios.
Linux 7.2 also incorporates an implementation of TCP Authentication Option (TCP-AO), a mechanism for authenticating TCP segments using shared keys that can be particularly useful for certain network protocols and infrastructure environments.
KVM expands its capabilities for AMD and Intel
Virtualization once again accounts for a substantial portion of the kernel changes.
KVM adds support related to AMD’s Guest Mode Execute Trap (GMET) and Intel’s Mode-Based Execute Control, alongside further work around Intel Trust Domain Extensions (TDX).
These capabilities are intended to expand the execution, control and isolation options available to virtual machines on compatible hardware.
For system administrators, it is worth remembering that the presence of a feature in the kernel does not automatically mean it can be used on every platform. Effective availability may depend on the processor generation, firmware, QEMU, libvirt, host configuration and the virtualization tools running above KVM.
This is particularly relevant for technologies such as TDX, where several layers of the platform are involved.
Linux 7.2 provides kernel-level pieces, but the complete software and hardware stack ultimately determines which capabilities are actually ready for production.
AMDGPU gets the foundations for HDMI 2.1 FRL
Although less relevant to a headless server, one of the Linux 7.2 changes likely to attract the most attention is found in AMDGPU.
AMD has introduced initial support for HDMI 2.1 FRL (Fixed Rate Link) alongside work related to DSC (Display Stream Compression) over FRL.
FRL makes it possible to overcome the bandwidth limitations of the older TMDS mechanism used by HDMI and is necessary for certain resolution and refresh-rate combinations associated with HDMI 2.1.
This is particularly relevant for Linux workstations and PCs using Radeon GPUs connected to modern monitors or televisions.
However, it should not yet be interpreted as meaning that every HDMI 2.1 feature now works completely on every Radeon GPU under Linux. Development of VRR and FreeSync over FRL connections has continued, meaning some parts of the stack are still maturing.
The new DRM scheduler will have to wait
Linux 7.2 also provides an interesting example of a feature that did not ultimately arrive as originally planned.
The DRM subsystem had prepared a new GPU job scheduler using a fairer scheduling policy that was intended to replace FIFO as the default behaviour.
Regressions appeared during the final stages of development.
The decision was therefore made to revert the change and temporarily return to FIFO, while the new scheduler remains experimental. Work is expected to continue during the Linux 7.3 development cycle.
It was not the only change reverted before release. Modifications related to I2C DesignWare, mt76 and Qualcomm components were also rolled back. One of the I2C issues could even prevent the touchpad from working on some ThinkPad laptops.
This is a less visible part of a new kernel release, but an important one from a systems perspective: the final weeks of a stable development cycle are not only about adding functionality. They are also about removing changes that are still causing regressions.
Linux 7.2 therefore arrives with considerably more infrastructure-focused work than its long-awaited Radeon graphics improvements might initially suggest.
Administrators running Btrfs, NFS, KVM, many-core servers or systems frequently exposed to memory pressure have good reasons to follow its adoption by the major distributions. As always in production, however, the fact that an improvement is available upstream does not mean an immediate upgrade is necessarily the right decision. The kernel supported by each distribution, available backports, hardware compatibility and the applications running on top all need to be considered.
Frequently asked questions
Which Linux 7.2 features are most relevant to servers?
Cache Aware Scheduling, Btrfs and MGLRU improvements, NFS and networking changes, and new KVM capabilities are among the most relevant additions for system administrators.
Does Linux 7.2 make Btrfs faster?
It includes several optimizations that improved performance in specific tests, alongside enabling large folios by default. Performance figures published for individual patches relate to specific scenarios and should not be extrapolated to every Btrfs server.
Should production servers be upgraded to Linux 7.2 immediately?
That depends on the distribution and infrastructure. In production environments, it is generally preferable to follow the kernel supported by the distribution and evaluate compatibility, regressions and specific requirements before moving to a new kernel branch.
Does Linux 7.2 provide complete HDMI 2.1 support for AMD Radeon?
Linux 7.2 incorporates an important part of the infrastructure required for HDMI 2.1 FRL and DSC over FRL in AMDGPU, but some functionality related to VRR and FreeSync over FRL continues to evolve.