On 4/24/26 7:58 AM, Andrew Morton wrote:
> On Sun, 19 Apr 2026 12:57:37 -0600 Nico Pache <npache@redhat.com> wrote:
>
>> The following series provides khugepaged with the capability to collapse
>> anonymous memory regions to mTHPs.
>
> Lots of stuff here:
>       https://sashiko.dev/#/patchset/20260419185750.260784-1-npache@redhat.com
>
> It's going to take some time.  Hopefully worthwhile.
>
> As always, it's useful to hear about the usefulness of the AI review.

Ok I went through those! Can you please pull the changes from the
staging branch so I can resend it soon (probably after LSFMM, so no
rush)?

Heres my report:

patch 2 - rather useless as memcg does not have per order/mthp stats

patch 4 - good point, although kinda minor, same as what Usama brought up

patch 5 - either the concern is nonsense or im too dumb to understand
it. Maybe someone else can confirm

Patch 5.2 - Not a real concern (I don't think), given we've already
fully locked down the PTE table, nothing should be able to reach it.
It notes a specific config might be an issue, I will test with that
on.

Patch 9 - was a real concern, and my fault for semi-lazily stripping
out a variable without fully considering the effects. David noted this
too. Good news is that it got the reasoning for why it is bad correct.
Oddly, I did not see a bug during testing which I would expected to
show up in the madvise tests. I just reverted my changes, I will try
to clean this up in a future cleanup series... Although there may be
no good way around this madvise behavior.

Patch 10.1 - Good point, I had considered this during my design, but
then convinced myself I was incorrect. This actually saves us a lot of
heap space :) Gotta retest a lot though. First few tests show no issue

Patch 10.2 - Not a concern and if we made it here, it's already been
checked. Furthermore, the result would be the same. Although not a bad
question from the AI

Patch 10.3 - I dont think this is a valid concern at all

Patch 10.4 - I don't think this is a valid concern at all

Patch 10.5 - The first half is valid (although it's what the next
patch does), so it's not really that valid, it's just missing the
series context. second half hmm

Patch 10.6 - Not a real concern

Patch 11 - Not a bad consideration

Patch 12 - real bug from my last refactor

Patch 12.2 - Decent consideration, but not a real concern, just a design choice.

So yes overall very smart to check sashiko :) But as someone currently
actively working on sashiko I was already a fan.

Cheers,
-- Nico



>