Introduction

Live kernel updates—where a running system transitions to a new kernel version without a full reboot—are a crucial capability for minimizing downtime in production environments. The Linux kernel has been steadily advancing in this area, with features like kexec handover and the live update orchestrator. However, one remaining hurdle is preserving HugeTLB (hugetlbfs) memory across such updates. At the 2026 Linux Storage, Filesystem, Memory Management, and BPF Summit, memory-management track lead Pratyush Yadav presented a session focused on solving this very problem.

What Is HugeTLB and Why Does Preservation Matter?

HugeTLB is a Linux kernel feature that provides large memory pages (typically 2 MB or 1 GB) to reduce TLB (Translation Lookaside Buffer) misses and improve performance for memory-intensive workloads such as databases, virtual machine monitors, and scientific computing. The pages are allocated from a hugetlbfs filesystem and are often pinned in physical memory for the entire lifetime of the process.

During a live update, the kernel must transfer all active memory mappings to the new kernel image. While ordinary pages can be remapped relatively easily, HugeTLB pages pose unique challenges:

• Their large size makes them expensive to copy or refault.
• They may be referenced by multiple processes or by kernel subsystems (e.g., KVM, DPDK).
• The page-table metadata in the old kernel must be seamlessly transferred to the new kernel without breaking access.

Current State of Live Update Infrastructure

The live update path relies heavily on the kexec mechanism and the live update orchestrator (which coordinates the transition). In a typical flow:

1. The new kernel is loaded into memory via kexec.
2. A reboot (actually a kernel handover) is performed, but without a full hardware reset.
3. The old kernel's memory is preserved and handed over to the new kernel.

However, the current implementation has limited support for preserving HugeTLB pages. As a result, workloads using hugetlbfs may experience data loss or require a full reboot—defeating the purpose of a live update.

The Session: Identifying the Gaps

In his summit session, Yadav outlined the key issues that must be addressed:

• Tracking pinned HugeTLB pages: The live update code needs to know which huge pages are in use and where they reside.
• Remapping page-table entries: The new kernel must reconstruct the same virtual-to-physical mappings for huge pages, while ensuring consistency with the old kernel’s state.
• Handling page cache and file-backed huge pages: hugetlbfs supports both anonymous and file-backed huge pages, each requiring different treatment.

Yadav suggested that the solution might involve extending the existing kexec handover protocol to include a region descriptor for HugeTLB pages, similar to how standard memory regions are passed. Another approach could be to leverage userfaultfd or memory events to allow userspace to re‑register huge pages after the update.

Related Work and Technical Considerations

Preserving large pages is not just a matter of copying data; the new kernel must respect the page-table formatting of the old kernel. For instance, if the old kernel used 2 MB pages, the new kernel must also use huge pages for the same physical range—or else it might attempt to break them into 4 KB pages, leading to performance degradation.

Additionally, the live update orchestrator must coordinate with the memory management subsystem to ensure that:

• No TLB shootdowns happen unnecessarily.
• The page migration logic does not interfere.
• Any huge page pool reservations are preserved.

Yadav also noted that the work is still in early stages and that contributions from the community are needed to refine the design. He pointed to existing patches on the mailing list as a starting point.

Conclusion

The ability to maintain HugeTLB memory across live updates is a critical enabler for systems that cannot afford downtime—such as high‑frequency trading platforms, real‑time analytics, and cloud infrastructure. The session led by Pratyush Yadav at the 2026 Linux Summit highlighted both the progress made and the remaining challenges. As the Linux kernel continues to mature its live update capabilities, addressing HugeTLB preservation will bring the vision of true zero‑downtime updates one step closer to reality.

For more details, see related articles on kexec handover and the live update orchestrator.