RFR (S/M): 8169703: G1 crashes with guarantee(pretouch_gang != NULL) failed: No pretouch gang specified
thomas.schatzl at oracle.com
Fri Nov 18 08:47:56 UTC 2016
can I have reviews for this change that fixes use of AlwaysPreTouch
Basically, AlwaysPreTouch is apparently not used very often, so some
use cases were missed, causing a guarantee to fail.
The patch adds support for:
- parallel pretouch during (potentially concurrent) humongous object
- allocating CDS/archive regions during archive creation. In this case
the code uses only one thread because archive regions are always
allocated region by region anyway.
- mapping in a memory range covered by a CDS archive
All these cases are covered by new regression tests.
runtime tier0-5 tests with -XX:+AlwaysPreTouch, new tests
More information about the hotspot-gc-dev