[uma-multipage 2/3] uma: convert mbuf_jumbo_alloc & pmap_pdpt_alloc to UMA_ZONE_CONTIG

[rlibby]

Test Plan

kyua test -k /usr/tests/sys/Kyuafile

[glebius]

Why did you convert all mbuf zones to contig? Only jumbo 9k and jumbo 16k really require that.

[jeff]

else if ((keg->uk_flags & UMA_ZONE_CONTIG) != 0 && keg->uk_ppera > 1)

		keg->uk_allocf = contig_alloc;

Note that CONTIG just specifies a constraint. It does not force the allocator change unless we're allocating more than one page. This was specified for mbufs because drivers do expect them to be contig and if UMA makes a layout decision that forces them into multiple pages we need to be sure to honor the constraint.

[glebius]

I see. That's because in next review we may produce multipage slabs with an allocation crossing a boundary.

[rlibby]

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]

sys/i386/i386/pmap.c

907 ↗

(On Diff #66923)

Aren't you relaxing a constraint here? vm_paddr_t is always 64 bits wide but the PDPT physical address has to fit in CR3, which is a 32-bit register.

[rlibby]

sys/i386/i386/pmap.c

907 ↗

(On Diff #66923)

You're right, I'll revert the pmap changes and just add UMA_ZONE_CONTIG to the flags here.

[rlibby]

review feedback: leave pmap_pdpt_allocf alone, just tag the zone.

Also tag the zones using slb_uma_real_alloc (it does contig allocs).

[rlibby]

Fix CTR/KASSERT after dropping some local variables

[rlibby]

In D23238#515865, @rlibby wrote:

Fix CTR/KASSERT after dropping some local variables

Oops, ignore, that was intended for the next patch.