Re: Survey: Is anyone using 32-bit builds of the CephFS kernel client?

Hi everyone,

Thanks for the feedback.

So far, I have received several replies regarding the 32-bit CephFS
kernel client usage. Based on the responses, there do not appear to be
any active users of 32-bit builds, and I have not received any
objections to moving forward with a 64-bit-only implementation.

Therefore, I will continue working on the patch to remove 32-bit
support and proceed with the 64-bit-only approach.

I will make sure the change is properly documented, including the
deprecation/removal of 32-bit support in the relevant release notes.

Thanks again for the feedback and discussion.

Best regards,
Xiubo

On Thu, 16 Jul 2026 at 16:53, Xiubo Li <[email protected]> wrote:
>
> Hi everyone,
>
> We are considering a performance improvement for the CephFS kernel
> client (kclient) that would be significantly simpler and cleaner to
> implement as 64-bit-only. Supporting 32-bit in this case adds
> non-trivial complexity and ongoing maintenance burden, so we'd like to
> understand whether continuing 32-bit support is still necessary.
>
> Before making any decision, we want to hear from the community:
>
> - Are you currently using, or planning to use, 32-bit (arm/armhf,
> i386/i686) builds of the CephFS kernel client?
> - If yes, is this for production workloads? Or legacy/testing purposes?
> - What is your platform (distribution, kernel version, architecture)?
>
> If there are no active 32-bit users, we would proceed with a
> 64-bit-only implementation and clearly document the deprecation in the
> release notes with appropriate warnings.
>
> If 32-bit support is still needed, we will explore alternative
> approaches — for example, keeping the current code path for 32-bit
> while the new fast path is 64-bit-only, or finding a way to cleanly
> abstract the difference.
>
> Please reply by August 6, 2026 so we can factor your feedback into the plan.
>
> Thanks!
> Xiubo Li