SoC: fixes for 5.18, part 1

The introduction of vmap-stack on 32-bit arm caused a regression
on a few omap3/omap4 machines that pass a stack variable into
a firmware interface.

The early pre-ACPI AMD Seattle machines have been broken for
a while, Ard Biesheuvel has a series to bring them back for now.

A few machines with multiple DMA channels used on a device have
the channels in the wrong order according to the binding, which
causes a harmless warning. Reversing the order is easier than
fixing the tools to suppress the warning.