RFR (S): 8077255: TracePageSizes output reports wrong page size on Windows with G1
jon.masamitsu at oracle.com
Wed Apr 8 21:15:30 UTC 2015
On 4/8/2015 6:57 AM, Thomas Schatzl wrote:
> Hi all,
> can I have reviews for the following change that fixes the output of
> -XX:+TracePageSizes to report the expected values for page size.
> Instead of returning the value of the page size to be used for
> allocations, this returned the OS'es allocation granularity which is the
> same on all but Windows, where this test subsequently fails.
> I am not completely sure why this did not fail with the corresponding
> push. The test most of the time still passes.
> The change simply calculates the preferred page size and passes it to
> the ReservedSpace constructor where it is used if set. Otherwise, the
> ReservedSpace constructor calculates that value by itself as before.
So the behavior of create_aux_memory_mapper() is unchanged with the fix
except that preferred_page_size is calculated before the call to
and passed in to create_aux_memory_mapper() and printed for TracePageSizes.
If I'm understanding that correctly, change looks good.
> JPRT (supposedly running that test), manual verification of the -XX:
> +TracePageSizes output before and after the change with and without -XX:
More information about the hotspot-gc-dev