In a full GC, move the verification after the GC to after RSet rebuilding. Verify RSet entries during a full GC under control of a flag.
Reviewed-by: tschatzl, brutisso
Create G1ParGCAllocBufferContainer to contain two buffers instead of previously using one buffer, in order to hold the first priority buffer longer. Thus, when some large objects hits the value of free space left in the first priority buffer it has an alternative to fit in the second priority buffer while the first priority buffer is given more chances to try allocating smaller objects. Overall, it will improve heap space efficiency.
Reviewed-by: johnc, jmasa, brutisso
Fixed the output of G1SummarizeRSetStats: too small datatype for the number of concurrently processed cards, added concurrent remembered set thread time retrieval for Linux and Windows (BSD uses os::elapsedTime() now), and other cleanup. The information presented during VM operation is now relative to the previous output, not always cumulative if G1SummarizeRSetStatsPeriod > 0. At VM exit, the code prints a cumulative summary.
Reviewed-by: johnc, jwilhelm
The fix of 8010992, disable use of global operator new and new[] which caused failure on some tests. This takes two of the bugs also add ALLOW_OPERATOR_NEW_USAGE to prevent crash for third party code calling operator new of jvm on certain platforms.
Reviewed-by: coleenp, dholmes, zgu
Refactor G1's hot card cache and card counts table into their own files. Simplify the card counts table, including removing the encoding of the card index in each entry. The card counts table now has a 1:1 correspondence with the cards spanned by heap. Space for the card counts table is reserved from virtual memory (rather than C heap) during JVM startup and is committed/expanded when the heap is expanded. Changes were also reviewed-by Vitaly Davidovich.
Reviewed-by: tschatzl, jmasa
Move the calculation and recording of eden capacity to the start of a GC and print a detailed heap transition for full GCs.
Reviewed-by: tschatzl, jmasa
Some parts of the G1 heap can only be walked during a safepoint. Skip verifying these parts of the heap when verifying during JVM startup.
Reviewed-by: brutisso, tschatzl
If the marking stack overflows while the marking tasks are draining the SATB buffers, remark will exit with some SATB buffers left unprocessed. Relax the guarantee to allow for overflow.
Reviewed-by: jmasa, brutisso
Skip reference processing if the global marking stack overflows during remark. Refactor and rename set_phase(); move code that sets the concurrency level into its own routine. Do not call set_phase() from within parallel reference processing; use the concurrency level routine instead. The marking state should only set reset by CMTask[0] during the concurrent phase of the marking cycle; if an overflow occurs at any stage during the remark, the marking state will be reset after reference processing.
Reviewed-by: brutisso, jmasa
In CMTask::do_marking_step(), Skip offering termination and entering the first and second synchronization barriers if called from a serial context, i.e. the VM thread.
Reviewed-by: brutisso, tschatzl
Update the task local finger to the start of the next object when marking aborts, in order to avoid the redundant scanning of all 0's when the marking task restarts, if otherwise updating to the next word. In addition, reuse the routine nextObject() in routine iterate().
Reviewed-by: johnc, ysr
Stop adding old regions to collection set when the remaining reclaimable bytes reaches, or goes below, G1HeapWastePercent. Changes were also reviewed by Vitaly Davidovich <vitalyd@gmail.com>.
Reviewed-by: brutisso
