When committing pages g1 has an optimization to re-use OS-level zero filling of newly committed pages.
This current does not work properly for the G1RegionsSmallerThanCommitSizeMapper::commit() call, e.g. for the card counts table.
A 4k page on x86 spans two (1M) regions; for the first one of that set the code correctly passes zero_filled == true to the on_commit handler, but the second does not - which means that useless work is done.
This impacts startup time slightly
This current does not work properly for the G1RegionsSmallerThanCommitSizeMapper::commit() call, e.g. for the card counts table.
A 4k page on x86 spans two (1M) regions; for the first one of that set the code correctly passes zero_filled == true to the on_commit handler, but the second does not - which means that useless work is done.
This impacts startup time slightly
- duplicates
-
JDK-8241848 G1RegionsSmallerThanCommitSizeMapper::commit_regions() calculate all_zero_filled wrong for callback
-
- Closed
-