RFR (XXS): 8134857: Inconsistency in maximum TLAB/PLAB size and humonguous object size

Thomas Schatzl thomas.schatzl at oracle.com
Tue Sep 1 16:07:13 UTC 2015


Hi all,

  can I have another review about PLAB argument processing? G1 now only
allows a maximum TLAB/PLAB size of heap region size / 2 - 1 words
(because with JDK-8078555 the bounds are actually checked).

This is wrong. Only heap region size / 2 + 1 objects are considered
humongous (see G1CollectedHeap::is_humongous()). This bug disallows
specification of Young- and OldPLABSizes of exactly heap region size /
2, which is inconvenient.

Finding out the exact value for 2^(heap region size words)/2-1 for some
testing is much more work than just specifying the "round" value.

There is also no reason to disallow 2^(heap region size words) word
sized T/PLABs.

CR:
https://bugs.openjdk.java.net/browse/JDK-8134857
Webrev:
http://cr.openjdk.java.net/~tschatzl/8134857/webrev/
Testing:
jprt

Thanks,
  Thomas





More information about the hotspot-gc-dev mailing list