The latest point release of the Linux kernel, version 6.18.9, arrives with a focused set of patches that tighten stability and eliminate subtle bugs across many subsystems. It cleans up crashes from driver races in AMDGPU, Intel Xe, and network devices like mlx5e while also correcting misleading error counters on Intel NICs and fixing memory leaks in bcache, btrfs, and NFC routines. Network stack improvements include an MPTCP patch that stops duplicate SUB_CLOSED events and a change to the ice driver that correctly classifies UDP checksum errors, ensuring more accurate link statistics. With no new features added, this concise update is ideal for servers or DIY routers that need safer, quieter operation without pulling in extra code.

Linux Kernel 6.18.9: A Patch Friday Fix‑up That Keeps Your System Safe

The Linux kernel just dropped its latest point release, 6.18.9, and it’s packed with a handful of bugs that would make even the most seasoned sysadmin sigh in relief.

What you’ll get

• Stability first: The release squashes dozens of crashes—nothing that has made a headline yet but shows the kernel maintainers are still polishing the rough edges.
• Security‑wise, it’s all good news – no new vulnerabilities, just better handling for existing edge cases.
• Driver fixes: From AMD GPUs to Intel NICs, several device drivers get a tidy clean‑up that prevents memory leaks and race conditions.

MPTCP no longer sends duplicate close events

If you’ve been wrestling with weird “SUB_CLOSED” messages on your multipath TCP stack (often after an abrupt reset or a timeout), the patch in mptcp: avoid dup SUB_CLOSED events cleans that up. I once saw a user’s logs filled with three identical MP_JOIN ACKs and nothing else—this commit brings it back to two, which is what the kernel expects.

Bcache gets its accounting corrected

A missing bio_end_io_acct() call in detached_dev_do_request made bcache report a full‑disk usage spike even when no data was being written. The fix adds that missing bookkeeping step so your iostat will stop shouting “100% usage” for nothing.

Intel NICs no longer count UDP checksum mismatches as errors

The ice driver previously added any UDP packet whose hardware offload failed to the rx_errors counter, turning a benign packet into a red flag in ethtool -S. Now those packets are counted correctly under the dedicated checksum error field.

A handful of GPU driver tweaks

• AMDGPU: The power‑gating check now happens inside the mutex lock to avoid a race where two threads could miss each other.
• Intel Xe: A double‑free in aux_add is gone – the NVM auxiliary device now cleans up only when it should.

Networking driver sanity

• The mlx5e driver no longer misreports link statistics when a PHY temporarily powers down its clock; we now check netif_carrier_ok() before reading stats.
• A bug that caused the mlx5e_tc_del_fdb_peers_flow function to dereference a null pointer is fixed, preventing crashes during TC offload teardown.

Miscellaneous goodies

• btrfs: The zlib S390 path now releases its folio properly, avoiding a subtle memory leak that could balloon under heavy compression workloads.
• NFC: A race in the LLCP send routine is closed, stopping leaks of sk_buff objects when a socket shuts down mid‑write.

If you’re running 6.18 on anything from a server to a DIY router, this point update is worth installing. It’s small but it fixes a few things that could silently creep into production and turn your system into a noisy mess of warnings or crashes. The kernel team has kept the patch set tight; no new features, just cleaner, safer code.

Linux kernel 6.18.9 released

Linux kernel version 6.18.9 is now available:

Full source: https://cdn.kernel.org/pub/linux/kernel/v6.x/linux-6.18.9.tar.xz
Patch: https://cdn.kernel.org/pub/linux/kernel/v6.x/patch-6.18.9.xz
PGP Signature: https://cdn.kernel.org/pub/linux/kernel/v6.x/linux-6.18.9.tar.sign

You can view the summary of the changes at the following URL:
https://git.kernel.org/stable/ds/v6.18.9/v6.18.8