EVALUATION

http://hg.openjdk.java.net/jdk7/hotspot-gc/hotspot/rev/ed52bcc32739
                                     

SUGGESTED FIX

Change G1CollectedHeap::max_capacity() to

size_t G1CollectedHeap::max_capacity() const {
  return _g1_reserved.byte_size();
}
                                     

EVALUATION

The reason for this is that we return the committed size from method G1CollectedHeap::max_capacity(), instead of the reserved size:

size_t G1CollectedHeap::max_capacity() const {
  return _g1_committed.byte_size();
}
                                     

SUGGESTED FIX

This is the behavior before (on my 2G Linux x86 laptop):

tonys-laptop:~> java -XX:+UnlockExperimentalVMOptions -XX:+UseG1GC -jar ~/Projects/MMTest/dist/MMTest.jar
max memory = 33554432

tonys-laptop:~> java -Xms32m -Xmx128m -XX:+UnlockExperimentalVMOptions -XX:+UseG1GC -jar ~/Projects/MMTest/dist/MMTest.jar
max memory = 33554432

This is the behavior after, which seems correct:

tonys-laptop:~> java -XX:+UnlockExperimentalVMOptions -XX:+UseG1GC -jar ~/Projects/MMTest/dist/MMTest.jar
max memory = 524288000

tonys-laptop:~> java -Xms32m -Xmx128m -XX:+UnlockExperimentalVMOptions -XX:+UseG1GC -jar ~/Projects/MMTest/dist/MMTest.jarVM
max memory = 134217728

This is the same app with ParallelScavenge:

tonys-laptop:~> java -jar ~/Projects/MMTest/dist/MMTest.jar
max memory = 504889344

tonys-laptop:~> java -Xms32m -Xmx128m -jar ~/Projects/MMTest/dist/MMTest.jar
max memory = 129302528

This is slightly different to what G1 returns, because ParallelScavenge subtracts the space of the unused survivor from the reserved size. We won't do that in G1.
                                     

SUGGESTED FIX

I'll piggyback a very minor cosmetic change in g1MemoryPool.{hpp,cpp} on this CR.