Linux 6.18: all the new features of the new kernel

Linux 6.18 comes with much more substance than it seems at first glanceIt's not just another kernel version: we're talking about a meticulously crafted cycle focused on performance, memory, networking, security, and hardware support, with all the makings of becoming the next LTS branch that will set the pace for distributions for years to come.

The development of Linux 6.18 has been particularly intense and strategicDeep new features have been introduced (such as sheaves in the memory manager, the new swap model, and the dm-pcache target), and at the same time, stability has been meticulously maintained in the release candidate versions, polishing regressions in virtual memoryfile systems, networks, and GPUs. All of this fits into a curious context: More and more users are tired of Windows 11, rise of the IA and more eyes than ever on Linux as an alternative.

Linux Kernel 6.18: Why This Version Matters So Much

Linus Torvalds has blessed Linux 6.18 as a major, change-laden releaseFrom the first release candidates, a very active cycle was expected, and according to Linus himself, the priority in recent weeks has been to reduce the noise: fewer experiments and more fine-tuning to leave the tree stable and ready for a long-term release.

Everything points to Linux 6.18 being the reference LTS kernel by 2025with extended long-term support. This explains why so much attention has been paid to key subsystems such as memory, networking, GPU drivers, security (SELinux, BPF, XFRM), and emerging architectures like longarch and RISC-V. The idea is that this kernel can serve as a solid foundation for enterprise distributions, general-purpose desktops, and cloud and virtualization environments.

The -rc phases have been particularly revealing of the cycle strategyIn version 6.18-rc6, a virtual memory regression occurred, forcing an on-the-fly fix for a late-stage bug. The fix was simple, but it served as a reminder that any oversight at this stage could result in a week's delay. Version 6.18-rc7 arrived with less noise and many minor adjustments to drivers, networking, and security, reinforcing the impression that a stable release was practically complete.

Intermediate -rcs, such as 6.18-rc3, have also been used to support delicate subsystemsIn this week's update, SMB Direct (RDMA) took center stage, receiving a substantial set of client and server patches for improved management of credits, queues, and disconnection states, along with fixes for XFS, io_uring, DRM, Rust Binder, networking, and the block layer. These changes aren't immediately noticeable to the end user, but they make a significant difference in data centers and high-performance environments.

In parallel, the kernel community has been keeping a watchful eye on the ecosystem's climate.More and more users are tired of Windows 11There's a lot of interest in Linux for gaming and development, and a boom in AI workloads. That's why this release combines a strong push for compatibility with modern hardware (gaming laptops, PC consoles, Apple Silicon, NPUs, GPUs) with improvements in virtualization, security, and performance.

Memory and performance improvements: sheaves, swap, and more

One of the major features of Linux 6.18 is the new "sheave" mechanism in the SLUB allocatorThis feature introduces small object caches per CPU, drastically reducing contention between cores when memory is allocated and deallocated. In practice, this greatly speeds up hot paths associated with RCUs and other structures that are constantly being created and destroyed.

The result of sheaves is more agile multitasking and fewer internal kernel locks.On servers with many CPUs, workloads that heavily utilize small structures (descriptors, inodes in memory, network structures, etc.) benefit significantly. For the end user, this translates into lower latency and improved responsiveness under pressure.

Linux 6.18 also takes an important step in the swap subsystem With the first phase of the new "swap table," an abstraction designed to improve how the swap memory cache is managed, developers have seen performance improvements of between 5% and 20% in certain scenarios, as well as much more stable behavior when the system experiences significant memory pressure.

Another key building block is the introduction of memdesc_flags_t, closely tied to struct pageThis change is part of a medium-term plan to reduce the size of struct pages and make memory management more flexible. For now, it's an intermediate step, but it lays the groundwork for future optimizations without abruptly breaking compatibility.

Minor memory management optimizations have also been applied to the x86 architecture.These improvements are complemented by enhancements to the accounting of reserved pages (memblock), the handling of HIGHMEM pages within the mempool, and details of timers and ticks that prevent race conditions with null pointers during timer shutdown. All of this contributes to a more predictable kernel behavior on systems with large amounts of memory and many CPUs.

File systems, storage, and dm-pcache

The area of storage and file systems get a good boost in Linux 6.18One of the key changes is the new device-mapper target. dm-pcachewhich allows the use of persistent memory (e.g., CXL or DAX devices) as an ultra-fast cache in front of disks or SSD slower. This opens the door to highly flexible hybrid storage architectures without having to rewrite applications.

In file systems there are several key pieces that are game-changers in practice.OverlayFS gains support for case foldingThis is important if you need compatibility with case-insensitive systems. FUSE now handles large range backups more efficiently, and XFS cleans up obsolete mount options while enabling the following by default: fsck online, better integrating hot testing and repair.

EXT4 continues its modernization process in 6.18: Adds support for 32-bit user and group IDs (essential in environments with many users or high UID/GIDs) and incorporates a new ioctl interface to manage the superblock more directly, improving administration and debugging tools.

There are more settings in other file systems and in the VFS: it is forced that sb_min_blocksize() be treated as __must_check In vfat, exfat, isofs, and XFS, to avoid silently ignoring errors, out-of-bounds reads in XFS are corrected by repairing symbolic links, and efivarfs is ensured to "freeze" before hibernating or suspending, preventing data corruption. firmware.

One point of contention is the definitive withdrawal of Bcachefs from this branchAlthough there was quite a buzz when it was initially integrated as a "promising" file system, the decision for version 6.18 has been to remove it, at least for now. Other file systems like Btrfs and EROFS are receiving robustness patches: fixes for reference checking, send/receive, and hardening against malicious images that could cause loops or corruption.

Networks: TCP, UDP, MPTCP and SMB Direct finer

The Linux 6.18 network stack also comes in strong, both in new features and fine-tuning.TCP stands out with its initial support for Accurate ECN (Explicit Congestion Notification), an improvement that allows for much more precise congestion measurement. This helps to optimize bandwidth usage without blindly monitoring and without generating the typical latency and signal loss spikes.

At UDP, the reception path has been rewritten with NUMA systems in mind.The goal is to reduce contention between CPUs and improve performance when there are many queues and many cores receiving packets. In addition, the default receive buffer is increased to 4 MB, and previous buffer sharing improvements are integrated, which in internal tests has resulted in performance increases of 40-50% under heavy network loads.

Multipath TCP (MPTCP) receives another batch of important fixesRace conditions are adjusted in timers and deferred jobs, ACKs are fine-tuned in fallback scenarios to classic TCP, the return to traditional mode is delayed in the presence of out-of-order packets, and the quick closure logic is separated from the normal TCP path. All of this is designed to make multipath sessions more stable and predictable.

In the -rc3 and later candidates, extensive work has been done on high-performance networks and RDMA.SMB Direct restructures read/write credit management, refines send and flush queues, and ensures that disconnect states wake up all involved threads. mlx5/mlx5e, phylink, DSA, veth, and other drivers fix leaks, null dereferences, and stat alignment issues, all with the aim of not disrupting high-traffic environments.

Other loose pieces of netting complete the pictureL2TP corrects the control buffer reset, virtio-net correctly initializes hash fields, bonding adjusts its list of slave interfaces in broadcast mode, and HSR is prevented from mounting cross interfaces between different netns. These are minor details, yes, but a poorly implemented network issue can lead to outages and latencies that are difficult to diagnose later.

Virtualization, KVM, and support for more vCPUs and hypervisors

Linux 6.18 places a lot of emphasis on virtualization, where the kernel has a strong foothold.The combination of KVM, AMD-specific enhancements, and IntelThe ability to detect new hypervisors makes this release a very attractive piece for cloud and on-premise environments.

KVM x86 brings support for CET (Control-flow Enforcement Technology) This update applies to both AMD and Intel processors, strengthening protections against attacks that attempt to hijack the execution flow. AMD Secure AVIC is also activated, improving virtual interrupt management on EPYC platforms, and the handling of more than 255 vCPUs on large servers is reinforced.

Integration with Microsoft Hyper-V is also improved.Kexec and Kdump now work better under this hypervisor, confidential virtual machines in Azure have been fine-tuned, and several compatibility issues have been addressed. Formal detection of the FreeBSD Bhyve hypervisor has also been added to adjust guest behavior when Linux runs on top of it.

In arm64, KVM receives corrections in the calculation and activation of FGT trapsThis prevents activating bits before configuring the rest of the system, and ensures that certain identification registers are initialized only once per virtual machine. In SVM (AMD), unnecessary LBR MSR intercept updates are avoided, reducing noise and potential bugs.

Finally, the VFIO ecosystem expands with support for new platforms, including support for NVIDIA GB300 within NVGRACE VFIO, which facilitates the virtualization of this hardware in compute-intensive environments and data centers that rely on GPUs and NPUs at scale.

CPUs, architectures, and emerging hardware

In terms of architecture, Linux 6.18 is one of those releases that makes a difference in the long run.RISC-V gains maturity with support for MIPS vendor extensions and the correction of the MIPS vendor ID in the ecosystem, which now aligns to 0x127 in both the kernel and QEMU. Better-aligned CSR definitions are also added, and maximum physical memory bits (MAX_POSSIBLE_PHYSMEM_BITS) are marked to improve zsmalloc.

LoongArch (Loongson) receives deep adjustments in NUMA node parsing, in CPU ordered presentation in /proc/cpuinfoIn addition, its security engine provides support for RNG, TPM2, and cryptographic acceleration, including interrupt masking during kexec/kdump and disabling the BPF trampoline for traces in module functions where it might be problematic.

MIPS corrects problems that could lead to random segmentation faults.The Malta platform (MMIO SOC-it with !EVA) is adjusted, TLB deactivation is avoided at critical times during initialization, and errors in device trees of some specific SoCs are fixed.

ARM64 sees a veritable deluge of improvements in DTS and SoCRockchip gains OPP board adjustments, PCIe and USB power on BigTreeTech boards; Orange Pi 5 corrects voltages; Raspberry Pi 5 (BCM2712) aligns ethernet0 alias; MSI issues are fixed and OTG switching is polished on NXP/imx; and NVIDIA Tegra marks certain PHYs as wake sources on Jetson Xavier NX, improving sleep behavior.

Apple Silicon remains a top priorityThanks to the work of Asahi Linux, more comprehensive device trees are now integrated for the Apple M2 Pro, Max, and Ultra, resulting in improved support for CPUs, GPUs, power, and peripherals. New EDAC drivers have also been added for Intel, AMD, and ARM platforms, along with preparations for Zen 6, patches for Bulldozer, changes to Intel FRED, and sliding power support for Panther Lake.

Graphics, GPUs, NPUs and displays

The graphics stack and DRM is another area where Linux 6.18 doesn't fall short.To begin with, Nouveau now uses the GSP firmware by default on NVIDIA Turing and Ampere GPUs, bringing the open-source driver closer to finer control over power management and functionality in these generations, useful in environments with CUDA.

AMDGPU and DRM/AMD fix multiple display and performance issuesClock gating, bandwidth unit conversion (pbn→kbps), DPCD read times and retries, sleep settings, and additional checks for VCN resets in SR-IOV environments. Everything is designed to prevent video queues from getting stuck and unexpected crashes from occurring when using a shared GPU.

On the Intel side, the i915 controller and the new Xe driver continue to evolveDMC firmware 30.02 for Xe3_LPD is loaded, phy detections are refined, PCI identifiers are separated by subplatform, and proper handling of MSI-X interrupts is added. Support for Intel “Wildcat Lake” displays appears in the display area, a relevant point for laptops and next-generation computers.

ARM Mali GPUs get double the love: Panthor and TyrPanthor expands its compatibility with G710, G510, G310 GPUs and Gx15/Gx20/Gx25 series, while Tyr enters as a new driver written in Rust for Mali with the aim of eventually replacing Panthor, offering a more secure and maintainable foundation.

Beyond GPUs, a new "Rocket" controller for NPUs in recent Rockchip SoCs has been added.Strengthening support for AI accelerators on Linux. And, as a bonus, the DRM "panic mode" is improved with a logo and QR code for debugging, preventing overlaps, divisions by zero, and memory leaks when painting in 24-bit modes.

Hardware support: PC consoles, laptops and peripherals

If you use Linux for gaming Or on modern laptops, 6.18 is of great interest to you.New HWMON drivers for GPD Win 4 and Win Max 2 have been added, providing fine-tuning control of fans and sensors. ASUS ROG Ally and Lenovo Legion Go 2 receives fixes for NVMe, interruptions, and other details that, in practice, make the difference between a stable system and one that crashes when exiting a game.

Controllers and input devices also benefitThe DualSense audio jack finally works correctly on Linux, HID drivers for mice and keyboards (including SONiX and ELECOM models) have been fine-tuned, and memory leaks in drivers such as playstationuclogic, ntrig or pidff, and special keys are set up on devices such as Lenovo Yoga Slim 7x or HONOR laptops.

Initial support for haptic touchpads is added, and compatibility with Snapdragon X Elite laptops is expanded.This opens up a range of options for high-performance ARM systems focused on energy efficiency. Alienware and Dell G feature a new controller that manages fans and RGB lighting, along with improved thermal profiles; HP Omen, Victus, and recent Asus motherboards also receive tweaks to WMI modules and event management.

There's a trickle of improvements in network cards, storage, panels, and audio devices.From new IDs in USB and PCI drivers to improved PTP profiles in network controllers like ICE, MLX5, and ENE, the kernel is increasingly recognizing and managing today's hardware more effectively, without the need to wait years for your device to reach the stable branch.

Security, BPF, SELinux and cryptography

Security is another pillar of Linux 6.18, with changes ranging from BPF to the XFRM subsystemOne of the most visible new features is the ability to load signed BPF programs, which allows for verification of their integrity and better integration into strict security policies in both server environments and sensitive deployments.

SELinux is undergoing a minor internal refactoring that, although invisible to the user, is very relevant.. The structure task_security_struct happens to be called cred_security_struct and the alias is standardized crsecThe access decision cache is linked per task. The goal is to facilitate code reasoning and reduce confusion that could lead to subtle errors.

In XFRM, the subsystem that manages IPsec and other network transformations, a series of fixes are being implemented.Improved handling of Security Association references in updates, removal of tunnels that were never added, marking states as DEAD when migrations fail, cleaning up device states when moves fail, improved EXTACK messages, and hunting for residual states when removing network spaces. Fewer dark corners, fewer surprises.

There are also difficult decisions, such as deactivating TPM Bus Encryption is considered defective and has a negative impact on performanceIt is preferable to be conservative and not offer a "false sense" of security that, in practice, penalizes without contributing much.

In parallel, the kernel refines mitigations for new attack surfaces. Like VMSCAPE, it updates attack vector controls on x86 and strengthens dependencies on certain cryptographic primitives, such as Poly1305, when combined with tools like KMSAN. All of this aligns with the ongoing effort to keep the kernel robust against modern threats.

Tools, io_uring, Rust Binder, and code quality

Linux 6.18 not only introduces new features: it also greatly refines the internal "plumbing"..io_uring receives several fixes in wait routes (io_waitid_prep()), lock annotations, auto-commit buffers for commands multishot and the CPU accounting of the sqpoll thread, so that getrusage() It should be better managed and not misleadingly inflated.

Rust Binder, the effort to bring Binder to Rust, is progressing with short but steady steps.Warnings about orphaned mappings have been removed, the handling of process freeze notifications has been adjusted, the deletion of listeners still in use has been avoided, and reporting details when a process is actually frozen have been corrected. Objtool now recognizes new Rust features like 'noreturn', and Clippy warnings have been cleaned up, helping to keep the tree clean and compileable in more configurations.

There's also fine work being done in memory, slab, and DAMON.Races with structures are corrected obj_extsThe consumption of poisoned pages is blocked when splitting THP, and mappings are correctly recounted in mremap() and leaks are sealed in DAMON filters. Hugetlbfs moves assertions to avoid premature firing on early return routes.

The overall result is a more internally consistent kernel, with fewer rough edges in advanced tools and subsystems.For most users, this will translate into fewer strange bugs, fewer inexplicable kernel panics, and a more robust experience on desktops, servers, and containers.

Looking at the set of changes—more refined memory, smarter networking, hardened file systems, better virtualization, and much broader hardware support—Linux 6.18 is shaping up to be a very complete and mature kernel.Prepared to serve as a foundation for both desktop and gaming distributions as well as enterprise and cloud offerings, and with everything to become the next LTS to migrate to with relative ease when distros begin to adopt it.