The fix for 6787254 intended to increase the capacity of TaskQueues in 64-bit builds by increasing the size of the TaskQueue::Age struct to 64 bits.  A typo prevented the change from taking effect.
Posted Date : 2009-03-25 22:48:28.0

N/A

The following code defines the type used for the two member fields of the Age struct:

#ifdef LP64
typedef juint TAG_TYPE;
// for a taskqueue size of 4M
#define LOG_TASKQ_SIZE 22
#else
typedef jushort TAG_TYPE;
// for a taskqueue size of 16K
#define LOG_TASKQ_SIZE 14
#endif

The LP64 should be _LP64.  Because of this typo, the larger data type is never used, so Age remains a 32-bit struct even in the 64-bit VM and the intended increase in task queue size for the 64-bit VM never took effect.
Posted Date : 2009-03-25 22:48:28.0

In addition to the _LP64 typo, there are several places which still access the Age struct as if were always 32-bits wide by dereferencing through an unsigned int* or volatile unsigned int*. The main examples are the get_age() and set_age() methods in TaskQueueSuper, but there are also calls to Atomic::cmpxchg() and asserts that must be updated to use size_t or another appropriate type.
Posted Date : 2009-03-25 22:48:28.0

The Age class should be better encapsulated so that TaskQueue* methods do not access fields directly, but use accessors which have the necessary attributes (e.g., volatile).  Adding a cmpxchg method to Age would simplify TaskQueue* methods and eliminate many of the casts.  Because the Age struct is now even larger in 64-bit builds, the fix for alignment of the Age struct from 6821507 must be done simultaneously.  That turns out to further simplify the code.
Posted Date : 2009-04-07 19:50:26.0

http://hg.openjdk.java.net/jdk7/hotspot-gc/hotspot/rev/3ee342e25e57
Posted Date : 2009-08-06 13:25:04.0

http://hg.openjdk.java.net/jdk7/hotspot/hotspot/rev/3ee342e25e57
Posted Date : 2009-08-10 05:12:01.0