Change Details

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 which 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().

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().

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 whichthat 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().