Did you read what I wrote in the first comment?

Yes. What I'm trying to say is

1. I would prefer doing this for all arch at once instead of just doing it for amd64. Or at least split this into two commits where the second one handles MD part specifically (e.g. amd64: ....). So far I don't see any commits under commits tab.
2. You said armv7 is broken in principle due to this patch. I didn't know the recent tradition says we allow commits that breaks build or run. Even recently, I heard from other committers that we still want all commits to be buildable and runnable. Otherwise it makes bisect harder. Maybe I'm misunderstanding, so please correct me if I'm wrong.

This is a diff against the branch, which contains around 20 commits.

I do not see a reason to spend time on machines I cannot even test, until the idea of this change is circulated in public and get accepted enough for the MD work to make sense.

For armv7, the arch maintainers or people who want the platform to stay alive, need to finally implement ifuncs for the kernel. It is not much work, much less than for userspace (where it is not too hard as well).

Make sense. We can ask other arch developers for testing and once they report success (hopefully with additional investigation in the MD code) it can be merged. There are many voices in the mailing list who want to keep support for armv7, but I'm not sure how many of them will actually test this and do the ifunc implementation.

Since the idea of scheduler as kernel module takes time for more investigation, I'm fine with this approach. Once merged, I'll ask 4BSD users on the mailing list to benchmark 4BSD vs ULE.

Since the idea of scheduler as kernel module takes time for more investigation, I'm fine with this approach. Once merged, I'll ask 4BSD users on the mailing list to benchmark 4BSD vs ULE.

Please coordinate at least with me before sending anything. We have to be careful in the choice of words, since the really important thing is that users do these tests after my ULE fixes. If people want to test 4BSD vs. ULE in the meantime, that's of course fine, but we don't want to press them to do that.

Got that.