They were introduced before UMA gained support for cross-domain frees.
Before that point, uma_zfree() would place the item in a per-CPU
bucket in the common case, so over time the per-CPU caches could become
polluted with items from remote domains if cross-domain frees were
common. free_domain() was intended to avoid polluting the caches in
cases where an expensive free() operation was permissible.
Now UMA handles cross-domain frees automatically for zones that follow
the (default) first-touch policy, which includes malloc(9) zones. The
the above-mentioned pollution is handled with an extra caching layer.
This makes free_domain() redundant. To simplify our allocator
interfaces, I propose removing it and uma_zfree_domain(). The latter is
only used to implement free_domain().