Re: [PATCH v2 00/14] userfaultfd: working set tracking for VM guest memory
From: Kiryl Shutsemau
Date: Fri May 08 2026 - 18:49:09 EST
On Fri, May 08, 2026 at 10:32:20AM -0700, Andrew Morton wrote:
> On Fri, 8 May 2026 16:55:12 +0100 "Kiryl Shutsemau (Meta)" <kas@xxxxxxxxxx> wrote:
>
> > This series adds userfaultfd support for tracking the working set of
> > VM guest memory, so a VMM can identify cold pages and evict them to
> > tiered or remote storage.
> >
> > v1: https://lore.kernel.org/all/20260427114607.4068647-1-kas@xxxxxxxxxx/
>
> Thanks. I'll duck v2 for now, await more review.
Sure.
> > Assisted-by: Claude:claude-opus-4-6
>
> For my education, and perhaps for others: can you please explain how
> you used Claude in the preparation of this series?
I'm no expert by any means, but here's how I used it here.
For this particular project there was quite a bit of path-finding.
I had a phase where I bounced ideas off Claude. It helped me
understand the problem space better and formulate possible solutions.
Rubber ducking on steroids.
Once it's clear _what_ to do, we formulate a plan on _how_. It also
involves back and forth.
Once the plan was done, I gave the go-ahead on executing it.
Userfaultfd already had a test suite, and it was extended to cover the new
functionality. I have some scripts to build the kernel and run it in a VM.
Claude knows how to use them, so at the end of plan execution I had a
functional feature.
Then the review phase. The most time-consuming and draining part.
I carefully reviewed all patches.
At this stage I use Claude as an editor.
Some of the changes I asked for required substantial rework of the whole
patchset, and I had to start the review from scratch. A good test suite and
build-test harness help to keep the whole thing from falling apart.
It took me quite a few review rounds before I was happy with the result.
Maybe between 8 and 10. I think better instructions can cut this number down.
And I need to rethink how I do the review. Reading the git log in
parallel with examining the code in the editor and giving instructions to
Claude is not very ergonomic. There's room for improvement.
Once I was happy with the patchset to give it Signed-off-by, I ran it
through Chris' review prompts several times, addressing the issues.
I hope it is helpful. I would also be glad if other folks shared their
workflow. There is probably a better way to achieve the same result.
I am new to the game.
--
Kiryl Shutsemau / Kirill A. Shutemov