G1, Paralle and Serial all log incorrect page sizes with os::trace_pages().
For G1 the parameter order is wrong and in Parallel and Serial the large_page_size will always be used even if the mappings fail.
I've created a test that catch those issues but there are also problems when tracing the code heaps (seeJDK-8261029), so the test can't be added until a fix for that is integrated as well.
The test is available here for now:
https://github.com/kstefanj/jdk/tree/test-for-trace-page-sizes
For G1 the parameter order is wrong and in Parallel and Serial the large_page_size will always be used even if the mappings fail.
I've created a test that catch those issues but there are also problems when tracing the code heaps (see
The test is available here for now:
https://github.com/kstefanj/jdk/tree/test-for-trace-page-sizes
- relates to
-
JDK-8259496 Parallel GC insists on using large pages although it failed to use them
-
- Closed
-
-
JDK-8261527 Record page size used for underlying mapping in ReservedSpace
-
- Resolved
-
-
JDK-8262188 Add test to verify trace page sizes logging on Linux
-
- Resolved
-
-
JDK-8261029 Code heap page sizes not traced correctly using os::trace_page_sizes
-
- Resolved
-
(1 links to)