P4
HeapBaseMinAddress is only honored when using +CompressedOops.
So the description of HeapBaseMinAddress should be fixed to add that it has only effect with +UseCompressedOops.
In addition to that, there is a behavioral bug in that the flag will be ignored if we cannot place the heap at any preferred address. In that case, we use whatever address the OS gives us, and HeapBaseMinAddress is ignored.
Can very easily be reproduced by handing a very high address: e.g. "-XX:UseHeapBaseMinAddress=256T" will consistently map on my machine at areas < 256T.
We should either document that behavior or make the JVM fail in that case.
Note however, that we may run on specially hardened kernels that don't allow mmap wish addresses at all - in those cases all we *can* do is to accept whatever address the OS hands to us. Failing always on such kernels may not be an option.
So the description of HeapBaseMinAddress should be fixed to add that it has only effect with +UseCompressedOops.
In addition to that, there is a behavioral bug in that the flag will be ignored if we cannot place the heap at any preferred address. In that case, we use whatever address the OS gives us, and HeapBaseMinAddress is ignored.
Can very easily be reproduced by handing a very high address: e.g. "-XX:UseHeapBaseMinAddress=256T" will consistently map on my machine at areas < 256T.
We should either document that behavior or make the JVM fail in that case.
Note however, that we may run on specially hardened kernels that don't allow mmap wish addresses at all - in those cases all we *can* do is to accept whatever address the OS hands to us. Failing always on such kernels may not be an option.