The seventh kernel release candidate, Linux 6.18-rc7, It's already in the Git tree And it arrives as the weekly pulse that marks the final stretch before the regular publication. Barring any last-minute setbacks, the official launch is expected on Sunday, November 30th. In this installment, the focus is clearly on the Regression fixes, driver polishing, and minor tweaks that fine-tune a cycle that is coming in very busy.
There's a strategic detail that doesn't go unnoticed: everything points to Linux 6.18 being the annual LTS (extended support) versionThis explains the care taken in addressing architectural fixes, network subsystems, GPUs, and security. Furthermore, this RC7 release comes after a scare in rc6A regression in virtual memory (VM) that was fixed with a simple patch, but served as a reminder that, so close to closure, Any loose end could cost an extra week.
What Linux 6.18-rc7 offers and why it matters
The goal of rc7 is to consolidate stability: variations have been reduced compared to rc6 and most changes fall into usual categories: Driver updates (with a focus on GPU and network), architectural fixes (longarch, mips, arm64)There's also some kernel-level networking, and improvements to tools and documentation. There's also a notable patch to SELinux prompted by a variable-naming mix-up that led to a renaming to clear the way.
Highlights of this week's news
Among the visible changes, the most notable is the correction of the vendor identifier for MIPS in the RISC-V ecosystem. The value was incorrect in both the kernel and the QEMU test code, and it is now set to 0x127, bringing both fronts into line. These kinds of subtle adjustments prevent incompatibilities that are difficult to diagnose during development and virtualization.
In terms of consumer hardware, the RC7 enables "two-handed" operation of the AYANEO Flip DS's touch panels. This means both digitizers are activated, a key feature for anyone using this device as a... portable PC console with touch controlsImprovements are also coming to the ASUS ROG Ally family, the Lenovo Legion Go, and several Alienware laptops—a wave of fine-tuning that expands compatibility and polishes details in popular gaming machines.
What Linus Torvalds has said
In the announcement message, Linus acknowledged that RC6 wasn't the best week due to that VM slip-up, although the fix turned out to be trivial. Even so, he remains confident: barring a major surprise, the The final launch should be ready next weekend.This RC7 is more restrained than the previous one, and that's a good sign: little "turbulence" and mostly normal changes.
Free translation of the spirit of the messageThe regression of the virtual memory subsystem in RC6 was a late blow, but the patch was simple and doesn't address a deep-seated problem. In RC7, the volume decreases; we see driver tweaks—mainly GPU and network—architectural fixes (with longarch, MIPS, and arm64 at the forefront), network kernel improvements, and some tooling and documentation. There's a larger-than-usual SELinux patch, prompted by a renaming variables to avoid confusionLet's use this last week to catch any stragglers.
LWN has reinforced the interpretation that this is probably the last -rc before a stable 6.18. If something unpleasant is discovered, an RC8 is a possibility, but, as of today, the expectation is that May the 6.18 arrive without further delays.
Architectures: longarch, mips, arm64 and others
In loongarch, critical aspects have been corrected: parsing NUMA nodes with numa_memblks, consolidating CPU names in /proc/cpuinfo, masking interrupts during kexec/kdump to avoid surprises, and even a tweak that Disable the BPF springboard for traces in module functionsUAPI types have also been aligned in the ptrace header, cleaning up interfaces.
MIPS includes a diverse set of fixes: improvements for Malta platforms (MMIO SOC-it with !EVA), prevention of TLB shutdown during initial unification, corrections in DTS (econet/EN751221), and a solution to random segmentation faults which affected stability. In parallel, in RISC-V the MIPS vendor ID is updated to 0x127, synchronizing the kernel and test tooling.
ARM64 receives a good number of changes from DTS: Rockchip sees adjustments to OPP boards (RK3576 and RK3588), PCIe and USB power pins on BigTreeTech boards, voltages on Orange Pi 5, the alias ethernet0 on Raspberry Pi 5 (BCM2712), and the disabling of HS400 in certain profiles for avoid unstable high-speed modesOn NXP/Imx, there are MSI fixes in imx95, OTG switching in imx8mp, and other tweaks that polish the platform. On NVIDIA Tegra, PHYs are marked as wake sources in Jetson Xavier NX.
Drivers: GPU, HID inputs, and x86 platform
GPU and display bring a lot to the table. AMD corrects premature DTBCLK gating, improves pbn-to-kbps conversion accuracy, corrects retries and timing in DPCD reads, bypasses the "power ungate" in sleep mode for VPE, and adds checks for VCN queue reset in SR-IOVOn Intel, the i915/xe3 loads DMC 30.02 for Xe3_LPD, refines c10phy detection to PHY A only, and separates wcl subplatform PCI identifiers. Radeon removes a fence path that could cause blocking. Xe includes overflow defenses with BIT(), duplicate option cleanup, and MSI-X vector0 interrupt handling.
On the HID/input front, in addition to support for the SONiX AK870 PRO and ELECOM M-XT3URBK, there are fixes for memory leaks in controllers such as PlayStation, uclogic, ntrig, and pidff; improvements to keyboards (Lenovo Yoga Slim 7x), AMD SFH sensors (stop before restart), and a change in input ownership, which is now called INPUT_PROP_PRESSUREPADSpecial keys are also being fine-tuned on HONOR equipment, and Goodix adds the ACPI ID GDIX1003 with pin handling cleanups.
The x86 platform comes loaded: the amd/pmc module recognizes the Van Gogh SoC, support for the spurious 8042 is added to ROG Ally, and features for Lenovo Legion Go 2 are included. In the HP/Alienware world, there are new thermal and cooling profiles for Omen and Victus, and AWCC support in Alienware 16 Aurora; coverage is also expanded to include the entire “M”, “X”, and “G” families in the alienware-wmi-wmax driver. MSI and Acer receive load settings and events in its WMI modules.
Networking and storage: MPTCP, L2TP, NVMe and more
MPTCP tops the list of fixes: race conditions in job scheduling and timers are corrected, ACK generation is improved in fallback scenarios, premature crashes are prevented, the fallback is delayed if there are out-of-order packets, and fastclose is decoupled from traditional TCP logic. In L2TP, the reset is now performed correctly. SKB buffer control in transmissionwhile veth gains robustness to avoid getting stuck due to races when managing TX queues.
NVMe and its Fibre Channel transport optimize the management request queue lifetime and ensure the cancellation of pending jobs by removing drivers. A lockdep warning during partition scans is addressed in multipath. The error path in idpf is also smoothed to prevent null dereferences, and drivers such as ice (PTP cleanup), mlx5 (IRQ handling), mlxsw (statistics and checks), and Open vSwitch (removing NSH support that never worked) are refined. phylink (supported modes in fixed links) and DSA (microchip lan937x and hellcreek).
File systems and VFS
efivarfs is always frozen before hibernation/suspend to prevent data corruption, and XFS fixes out-of-bounds reads by repairing symbolic links. Several systems (vfat, exfat, isofs, and again xfs) now check the return value of sb_min_blocksize(), and the block subsystem marks that API as __must_check so that no one ignores it. tmpfs/shmem fixes reconfiguration with noswap, bfs rebuilds the file type on load, and improvements are introduced to the VFS path (iput_not_last) and Landlock to prevent splats from calls that may be sleeping.
Security: SELinux and XFRM
SELinux applies a renaming of key variables (task_security_struct becomes cred_security_struct and is standardized to crsec where applicable) and moves the access decision cache to one context per task, facilitating reasoning and avoiding confusionIn XFRM there is a battery of fixes: SA references in updates, deletion of non-added tunnels, states marked as DEAD when migrations fail, device state cleanup on failed move, improved error/EXTACK messages, and hunting for residual states when deleting a netns.
Memory, times and tools
Fine-tuning is done in memory and timers: memblock correctly calculates free pages with soft-reserve, mempool doesn't unduly poison HIGHMEM pages, ticks fixes an idle condition, and timers prevent a race with null function pointers when shutdown_sync is executed. Timekeeping closes a resource leak, and mm/huge_memory correctly initializes the huge zero folioio_uring adjusts argument types, perf fixes counters to zero in cpu-clock, and RISC-V tools correct CSR definition misalignments.
Virtualization and ACPI
KVM/arm64 computes FGT traps before activating them and finalizes ID registers only once per VM; in SVM (AMD), redundant updates of LBR MSR intercepts are avoided. ACPI/APEI/EINJ fixes initialization and injection in version 2, and several kernel-doc-type documentation elements in Intel platform modules, tee uapi, and others are aligned, helping the tree compile cleanly and Improve the quality of the reference.
Judging by the activity and tone of the announcement, RC7 leaves 6.18 with a pretty smooth path: specific fixes in VMs after the RC6 scare, numerous driver corrections—with special emphasis on GPU and network drivers—improvements to DTS for ARM64, cleanup and clarifications in SELinux and XFRM, and improvements to widely used devices (AYANEO Flip DS, ROG Ally, Legion Go, Alienware). The plan is to finalize the stable release next Sunday, unless a major bug appears that warrants an RC8 update; otherwise, everything indicates that 6.18, a candidate for LTS status, will be released with good form factor and extensive hardware support.
