Yes, that's right. In fact, the smaller mbufs are actually still okay with multipage slabs due to their power-of-2 size, but I thought it was better to make the constraint explicit.

markj feedback: style(9)

missed one, and messed one up

markj feedback, don't print threads

In D22963#506510, @cem wrote:

I don't think expecting misaligned atomics to work is reasonable. We could KASSERT alignment? Asking for 2-byte alignment isn't asking much.

I only looked at _atomic_subword.h.

Looks good. Did this fix the bus-sc accounting?

jeff feedback, s/ipers_nohdr/ipers_offpage/ too

forgot jeff feedback s/eff_nohdr/eff_offpage/

jeff & markj feedback

Code looks good.

@jeff Will do for the rest of the comments. A couple clarifying questions.

Jeff asked that the old code be left in place for posting the diff, for readability of the diff. The #ifdef 0 code will be deleted before commit.

Latest revision LGTM.

This one blames to 1998, so I guess it was legit :)

Here's what changing it to no-error for the clang build does (amd64):

Looks good, after fixing up the keg_alloc_slab comment and the keg_dtor printf format.

Rebase LGTM.

Looks good but I do have a question about how uma_zone_reserve() is supposed to work now.

Looks good. Recommend building at least INVARIANTS and non-INVARIANTS kernels before commit.

Fixups look good. I agree with leaving an XXX and coming back later to the problem of transitioning between no-limit and limit.

Generally looking good, but I do have a few more comments.

Fixups look good.

Fixups look good.

I can prepare a follow up. What I did here was exactly what we already do for clang. I will change both. And then I assume you'd have the same comment for the user build?

In D22876#501221, @jhb wrote:

Maybe include share/mk/bsd.sys.mk at the same time as userland needs it as well (at least it did for me)?

In D22888#501195, @imp wrote:

So it worked! Yippy! Of course I'd love you to commit this :)

imp's suggestion

Ah. Yes. I'll check if that builds. Thanks.

A non-intrusive Makefile solution could be to lower camcontrol's WARNS to 3, either unconditionally or conditional on gcc.

In D22888#501055, @imp wrote:

It is actually properly aligned. Let's not do silly crap like this. Let's say it really is properly aligned.

Replace alloca

Looks good.

Looks good, just some nitpicks.

I went ahead and committed this in r355746 along with re-application of D22759.

The problem I was originally trying to solve was allowing embedded slabs on the bucket zones for D22768. This means the slab memory for the bucket zones must come from uma_small_alloc. It can't come from the startup pages because PHYS_TO_VM_PAGE doesn't work on the startup pages since they aren't in the page array (although, we could possibly change that, with some potential other cost, see the vm_page.c bit in D11784). So I added a check at keg creation time that only enables embedded slabs if the keg was created after BOOT_PAGEALLOC. That disabled embedding for bucket zones, which were being created before then.