Greg Kroah-Hartman tagged Linux 7.1.1 today at 13:50 UTC+2 (June 19, 2026), marking the first stable point release of the 7.1 series just five days after Linus Torvalds published the mainline 7.1 kernel. The release carries ten cherry-picked fixes dominated by a CVE-level arm64 hardware errata mitigation now extended to Microsoft's Azure Cobalt 100, NVIDIA's Olympus silicon, and additional Arm CPU families — alongside driver core hardening, an AMD GPU memory-management fix, and a SOFTIRQ deadlock closure in the kernel's async-signaling path.
CVE-2025-10263: The TLBI Completion Errata
The dominant theme in 7.1.1 is five commits mitigating CVE-2025-10263, a hardware-level ordering defect in the arm64 Translation Lookaside Buffer Invalidation (TLBI) instruction. On affected processors, a broadcastTLBI;DSB sequence may complete before memory writes are globally observed across all cores. The consequence: a TLB entry that was supposed to be invalidated might be re-validated against stale page table data before all cores have seen the updated memory mapping.The mitigation has existed in the kernel as
ARM64_WORKAROUND_REPEAT_TLBI — it issues a second TLBI;DSB to enforce ordering — but was only applied to a subset of earlier CPU designs. Two preparatory commits in 7.1.1 first add MIDR_EL1 register definitions for the Arm C1-Premium and Arm C1-Ultra (TRM references 109416/0100 and 108014/0100 respectively), which are required groundwork before the errata table entries can be registered.Affected Hardware
Three separate enable-patches land in 7.1.1, each contributed by the affected silicon vendor:Microsoft Azure Cobalt 100 — Will Deacon's patch registers the Neoverse N2 r0p0 stepping in the repeat-TLBI workaround table. Azure Cobalt 100 is Microsoft's custom Arm chip deployed across the Azure Cobalt VM series; Linux VMs running on those instances gain the memory-ordering protection with this kernel update.
NVIDIA Olympus — Shanker Donthineni at NVIDIA adds
MIDR_NVIDIA_OLYMPUS to the ARM64_ERRATUM_4118414 workaround list — the same erratum number used for the Azure Cobalt 100 patch, confirming this is the same silicon-level defect across both vendors' designs.Generic Arm CPUs — Mark Rutland's base patch extends the repeat-TLBI workaround to a broader set of Arm CPU models not covered by earlier entries in the errata table.
Other Fixes
driver core — reject devices with unregistered buses: A regression introduced with the const-bus rework in kernel 6.3 caused registering a device against an unregistered bus to silently succeed. The device would never actually bind and its sysfs attributes would be absent — hard to diagnose. Johan Hovold's fix returns an explicit error, restoring correct behavior for all kernels from 6.3 onward.driver core faux bus — use-after-free and missing sanity check: A companion Hovold patch (7.0+ only) fixes the faux bus root device's missing
release function (which freed the device structure without cleaning up the embedded kobject reference) and adds a null-check on bus registration failure, closing a use-after-free window on error paths.drm/amdgpu — broken HMM retry loop removal: Honglei Huang at AMD removes a
while-loop in amdgpu_hmm_range_get_pages() that was busy-spinning for approximately one second on -EBUSY returns. The loop was already broken by an earlier refactor that moved mmu_interval_read_begin() outside it, making -EBUSY unrecoverable inside the loop. The fix propagates -EAGAIN to callers immediately instead of spinning uselessly.HID — battery devm list cleanup: Rafael Passos fixes dangling
list_head nodes in hdev->batteries that occurred when devm_kzalloc freed battery entries on driver unbind or reprobe, leaving the list with stale pointers that could cause use-after-free on the next probe cycle.fs/fcntl — SOFTIRQ deadlock in fasync signaling: Mingyu Wang fixes a lock-ordering inversion in
send_sigio() and send_sigurg() where tasklist_lock was acquired with read_lock() (SOFTIRQ-unsafe) while holding the file descriptor table lock (SOFTIRQ-safe). TCP URG packets deliver send_sigurg() from softirq context, making this a potential remote deadlock vector on systems using FASYNC process-group signaling. The fix replaces read_lock(&tasklist_lock) with rcu_read_lock(), safe because PID hash traversal is already RCU-protected.Key Takeaways
- Linux 7.1.1 ships 10 fixes as the first stable point release of the 7.1 series, arriving five days after the June 14 mainline drop
- CVE-2025-10263 mitigation now covers Microsoft Azure Cobalt 100 (Neoverse N2 r0p0), NVIDIA Olympus, and additional Arm CPU families — the
TLBI;DSBcompletion ordering defect can allow stale TLB entries to persist in affected hardware - Azure Cobalt 100 is one of the newly protected targets — Linux workloads on Azure Cobalt VMs benefit directly from this update
- drm/amdgpu removes a broken 1-second busy-wait that was spinning on an unrecoverable error code after an earlier refactor
- fs/fcntl closes a remote-triggerable SOFTIRQ deadlock reachable via TCP URG packets on FASYNC-enabled process groups
- Same-day coordinated stable releases: 7.0.13, 6.18.36, 6.12.94, 6.6.143, 6.1.176, 5.15.210, 5.10.259
uname -r.References
- Linux 7.1.1 changelog — cdn.kernel.org
- kernel.org release listing
The Linux Kernel Archives
www.kernel.org
- Stable release announcement (lore.kernel.org)