Re: [LSF/MM/BPF TOPIC] HugeTLB file preservation across live update

[Pasha Tatashin <pasha.tatashin@soleen.com>]

On Fri, Feb 6, 2026 at 12:48 PM Pratyush Yadav <pratyush@kernel.org> wrote:
>
> Hi all,
>
> I would like to propose the topic of HugeTLB file preservation across

+1

> live update for discussion at the upcoming LSF/MM/BPF summit.
>
> Background
> ==========
>
> Support for performing a live update was recently added to the kernel
> via the Live Update Orchestrator (LUO) subsystem [0]. Live update allows
> a full kernel update over kexec without terminating running workloads.
> The main target is live updating the hypervisor without terminating VMs.
>
> For more information on live update and LUO, see its kernel
> documentation [1], the LUO patchset [0], or this LPC 2025 presentation
> [2].
>
> MEMFD preservation
> ==================
>
> MEMFD is the first file type that added support for live update. This
> support was merged with the LUO series. The memfd preservation code can
> be found in mm/memfd_luo.c [3].
>
> MEMFD can use pages backed by shmem or HugeTLB. The live update support
> in mainline currently only works with shmem-backed MEMFDs.
>
> HugeTLB preservation
> ====================
>
> While shmem support for live update is useful for many things, using it
> for VM memory (and other memory-intensive workloads) is inefficient. The
> next step for MEMFD support for live update is to allow preserving
> HugeTLB-backed files.
>
> Note that we do not attempt to preserve the whole HugeTLB file system.
> We only preserve the memory pages for the files specified by userspace.
> The HugeTLB file is re-created fresh on the next kernel and the pages
> are inserted back in. This lets us keep the maintenance footprint
> smaller.
>
> Current status
> --------------
>
> This support was sent out on the list as an RFC [4] in December 2025,
> and was also presented at LPC 2025 [5] in the Live Update
> microconference.
>
> While the patchset has got some initial reviews from live update
> maintainers, it has not yet had any review from HugeTLB developers or
> maintainers.
>
> Goals of the discussion
> -----------------------
>
> This session will aim to present the topic, explain why it is needed,
> show how the support works, and most importantly, collect feedback on
> how to proceed with this work.
>
> PS: if any HugeTLB developers or maintainers are reading this, it would
> be great to have your initial thoughts on the RFC [4] so I can present a
> more polished patchset at the conference and make the most of the
> discussion.

Thank you Pratyush for proposing this topic. A few more thoughts:
While the discussion should be about memfd+hugetlb support, the long
term goal is to make this work as an enabler for guestmemfd+hugetlb
preservation support.

Pasha

>
> Key attendees
> =============
>
> Live update:
>   - Alexander Graf
>   - David Matlack
>   - Jason Gunthorpe
>   - Mike Rapoport
>   - Pasha Tatashin
>
> HugeTLB:
>   - David Hildenbrand
>   - Muchun Song
>   - Oscar Salvador
>
> References
> ==========
>
> [0] https://lore.kernel.org/linux-mm/20251125165850.3389713-1-pasha.tatashin@soleen.com/T/#u
> [1] https://docs.kernel.org/core-api/liveupdate.html
> [2] https://lpc.events/event/19/contributions/2052/
> [3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/mm/memfd_luo.c
> [4] https://lore.kernel.org/linux-mm/20251206230222.853493-1-pratyush@kernel.org/T/#u
> [5] https://lpc.events/event/19/contributions/2044/
>
> --
> Regards,
> Pratyush Yadav