Linux 5.7 was released by Linus Torvalds on May 31st, and as with many kernel releases, the changelog reads like infrastructure plumbing rather than flashy features. But this one has several changes that developers and sysadmins should pay attention to, particularly if you’re running virtualized workloads or dealing with storage performance.
The ExFAT Driver: A Microsoft Collaboration#
Perhaps the most symbolically interesting addition is the new ExFAT file system driver, contributed by Samsung and developed with Microsoft’s blessing. Microsoft published the ExFAT specification last year and encouraged a proper in-kernel implementation. The previous ExFAT support in Linux was a staging driver with known issues and questionable legal standing.
The new driver is clean, performant, and properly licensed. For those of us who regularly shuttle data between Linux and Windows systems — or deal with SD cards and USB drives formatted as ExFAT — this is a genuine quality-of-life improvement. It’s also another data point in the ongoing Microsoft-loves-Linux narrative that would have been unthinkable a decade ago.
I’ve tested the new driver with a few large file transfers and it’s noticeably more stable than the old staging driver. No more occasional corruption on unmount, which was a persistent annoyance.
Split-Lock Detection: Protecting Multi-Tenant Performance#
The split-lock detection feature is one of those kernel changes that most developers will never interact with directly but that matters enormously for anyone running shared infrastructure. A split-lock occurs when an atomic operation spans a cache line boundary, forcing the CPU to lock the entire memory bus to maintain coherence. This is extremely expensive — a single split-lock can stall all cores for hundreds of cycles.
In a bare-metal single-tenant environment, this is mostly a self-inflicted performance wound. But in virtualized or multi-tenant environments — read: the cloud — one misbehaving VM can degrade performance for every other VM sharing the same physical host.
Kernel 5.7 adds the ability to detect split-lock operations and either warn about them or kill the offending process. For cloud providers, this is a meaningful tool for ensuring fair resource allocation. I expect we’ll see this enabled by default in cloud kernels fairly quickly.
The configuration options are sensible: split_lock_detect=fatal kills processes that trigger split-locks, while split_lock_detect=warn just logs them. For development environments, the warning mode is useful for catching performance issues before they hit production.
Userfaultfd Write Protection: Better Live Migration#
The userfaultfd subsystem — which allows user-space programs to handle page faults — got write-protection tracking support. This sounds obscure, but it’s a key enabler for two important use cases: database snapshotting and VM live migration.
For live migration specifically, write-protection tracking allows the hypervisor to efficiently identify which memory pages have been modified during migration. Instead of re-scanning all of memory, the kernel notifies user-space when a page is dirtied. This translates to faster migration times and shorter blackout periods when moving VMs between hosts.
If you’re running KVM-based virtualization, this is relevant. QEMU already has patches to take advantage of this feature, and I’d expect production hypervisors to adopt it over the next few kernel release cycles.
Thermal Management and ARM Progress#
The kernel’s thermal management framework got a significant overhaul in 5.7, with the new power_allocator governor providing more intelligent thermal throttling. Instead of the crude step-down approach used previously, the new governor uses a PID controller to maintain target temperatures while maximizing performance.
This matters particularly for ARM-based systems, which are increasingly showing up in edge computing and server scenarios. The Raspberry Pi 4, various ARM development boards, and AWS Graviton2 instances all benefit from better thermal management.
Speaking of ARM, 5.7 continues the trend of improving ARM64 support with new SoC additions and driver improvements. The kernel’s ARM ecosystem has matured significantly — three years ago, running a mainline kernel on ARM hardware was an adventure. Today, it’s largely unremarkable, which is exactly where it should be.
Other Noteworthy Changes#
A few other additions worth mentioning:
- BPF improvements: The BPF subsystem continues to expand, with new helper functions and better BTF (BPF Type Format) support. BPF is quietly becoming one of the most important subsystems in the kernel for observability and networking.
- Habana Labs Gaudi AI accelerator support: Another AI training accelerator gets kernel support, reflecting the growing diversity of hardware in the ML training space.
- Zonefs file system: A new file system for zoned block devices (primarily SMR hard drives), co-developed by Western Digital. As storage media becomes more diverse, the kernel needs to support different access patterns.
My Take#
Linux 5.7 is a “boring” release in the best possible sense. No dramatic restructuring, no contentious API changes — just steady improvement of the infrastructure that runs most of the internet. The split-lock detection and userfaultfd improvements are exactly the kind of changes that make Linux better for cloud workloads without breaking anything for desktop or embedded users.
I’ve been running Linux kernels in production since the 2.4 days, and what strikes me about the modern development process is how mature it’s become. The merge window is predictable, the release candidates are genuinely stable, and the range of hardware and use cases covered by each release is extraordinary.
The ExFAT driver is a nice cherry on top — not because the technology is particularly interesting, but because it represents the kind of cross-company collaboration that was rare in open source just a few years ago. Microsoft contributing patent rights so Linux can properly support their file system is a concrete act, not just a press release.
If you’re running production systems, I’d recommend updating to 5.7 after it lands in your distribution’s update channel. The split-lock detection alone is worth it for anyone running multi-tenant workloads. For the rest of us, it’s another solid release from a project that’s been consistently delivering for nearly 30 years.