This looked wrong at the first glance for arm32 because it is not multi-copy atomic, but uses support_IRIW_for_not_multiple_copy_atomic_cpu = false.
However, it seems to work, because OrderAcess::loadload() is implemented in a much stronger way than on other platforms. Relying on this in shared code is probably very unexpected to most developers. There should be a more comprehensive implementation.
In addition, if support_IRIW_for_not_multiple_copy_atomic_cpu gets changed on PPC64 in the future (e.g. for experiments or for evaluating options for JEP 188: Java Memory Model Update) this will become incorrect for that platform, too.
(There are no plans to support this officially. Just for experiments.)