Main part is that kern_copyin on amd64 after LA57 should query the top
of UVA for correct operations. In fact it should started doing that
after the workaround for AMD bug with IRET in the last user page was
fixed by reducing UVA by a page.
Also since we started calculating top of UVA, fix MIPS according to
the comment.
Reported by: lwhsu
PR: 248933
Reviewed by: alc, markj
Sponsored by: The FreeBSD Foundation
Differential revision: https://reviews.freebsd.org/D26312