6871109: G1: remove the concept of the scan only prefix
Removed scan only regions and associated code. The young portion of the collection set is now constructed incrementally - when a young region is retired as the current allocation region it is added to the collection set.
Reviewed-by: apetrusenko, iveresov, tonyp
The outgoing region map is not maintained properly and it's causing an assert failure. Given that we don't actually use it, I'm removing it. I'm piggy-backing a small change on this which removes a message that it's printed before a Full GC when DisableExplicitGC is set.
Reviewed-by: apetrusenko, ysr
The expression that calculates the live bytes for a heap region can overflow for a suitably large humongous region/object. Cache the object size in a suitably sized local variable so that the expression is converted to a wider type.
Reviewed-by: tonyp, jmasa, iveresov, apetrusenko
Humongous regions may contain multiple objects as a result of being retained as to-space from a previous GC and then re-used as to-space after being tagged as humongous. These changes include a check that causes retained to-space regions that are now tagged as humongous to be disregarded and a new to-space region allocated.
Reviewed-by: tonyp, iveresov
Make sure that two marking cycles do not overlap, i.e., a new one can only start after the concurrent marking thread finishes all its work. In the fix I piggy-back a couple of minor extra fixes: some general code reformatting for consistency (only around the code I modified), the removal of a field (G1CollectorPolicy::_should_initiate_conc_mark) which doesn't seem to be used at all (it's only set but never read), as well as moving the "is GC locker active" test earlier into the G1 pause / Full GC and using a more appropriate method for it.
Reviewed-by: johnc, jmasa, jcoomes, ysr
Calling the methods region_stack_push() and region_stack_pop() concurrent is not MT-safe. The assumption is that we will only call region_stack_push() during a GC pause and region_stack_pop() during marking. Unfortunately, we also call region_stack_push() during marking which seems to be introducing subtle marking failures. This change introduces lock-based methods for pushing / popping to be called during marking.
Reviewed-by: iveresov, johnc
Remove the G1GCPercent parameter, that specifies the desired GC overhead percentage in G1, and observe the GCTimeRatio parameter instead.
Reviewed-by: jmasa, johnc
Newly-created threads always had the active field of their SATB queue initialized to false, even if they were created during marking. As a result, updates from threads created during a marking cycle were never enqueued and never processed. The fix includes remaining a method from active() to is_active() for readability and naming consistency.
Reviewed-by: ysr, johnc
If CM task was aborted while scanning the last object of the specified region and the size of that object is equal to bitmap's granularity then the next offset would be equal or over the region limit which is exactly what the assertion states.
Reviewed-by: ysr, tonyp, jmasa
A TLAB became large enough to be considered a humongous object allowing multiple objects to be allocated in a humongous region, which violates a basic assumption about humongous regions. The changes ensure that TLABs cannot be regarded as humongous.
Reviewed-by: iveresov, tonyp
Replace the combination of the G1SteadyStateUsed / G1SteadyStateUsedDelta parameteres to decide the marking initiation threshold and instead use InitiatingHeapOccupancyPercent.
Reviewed-by: ysr, johnc
Implemented block-based work stealing. Moved copying during the rset scanning phase to the main copying phase. Made the size of rset table depend on the region size.
Reviewed-by: apetrusenko, tonyp
The operations of re-using a RSet component and expanding the same RSet component were not mutually exlusive, and this could lead to RSets getting corrupted and entries being dropped.
Reviewed-by: iveresov, johnc
G1 now answers "no" to the query can_elide_initializing_store_barrier() in the product build. A debug flag allows alternate behaviour in debug builds.
Reviewed-by: iveresov, tonyp
When enqueuing a completed PtrQueue buffer, cache a local pointer to the buffer and clear the field in the PtrQueue prior to unlocking the mutex referenced by the _lock field and pass the cached local value to the enqueuing routine. This will prevent the same completed buffer being enqueued multiple times, which causes the hang.
Reviewed-by: ysr
Deferred card marks are now flushed during the gc prologue. Parallel[Scavege,OldGC] and SerialGC no longer defer card marks generated by COMPILER2 as a result of ReduceInitialCardMarks. For these cases, introduced a diagnostic option to defer the card marks, only for the purposes of testing and diagnostics. CMS and G1 continue to defer card marks. Potential performance concern related to single-threaded flushing of deferred card marks in the gc prologue will be addressed in the future.
Reviewed-by: never, johnc
Fixes a race on the dirty card queue completed buffer list between worker thread(s) performing a flush of a deferred store barrier (enqueueing a newly completed buffer) and worker thread(s) in the RSet updating code claiming completed buffers. Removed the routine that removes elements from the completed update buffer queue using a CAS.
Reviewed-by: ysr, tonyp
ScavengeALot now causes an incremental (but possibly partially young, in the G1 sense) collection. Some such collections may be abandoned on account of MMU specs. Band-aided a native leak associated with abandoned pauses, as well as an MMU tracker overflow related to frequent scavenge events in the face of a large MMU denominator interval; the latter is protected by a product flag that defaults to false.
Reviewed-by: tonyp
Protected stats dump with a new develop flag; other than for the dump, reconciled product and non-product behaviour in face of the error.
Reviewed-by: tonyp
Added instrumentation and (temporary) assert in non-product mode; clipped the value when found out-of-bounds in product mode. Fix of original issue will follow collection of data from this instrumentation.
Reviewed-by: jcoomes, tonyp
The type in the NEW_C_HEAP_ARRRY and FREE_C_HEAP_ARRAY calls in the buffer allocation code was changed from void* to char as the size argument had already been mulitipled by the byte size of an object pointer.
Reviewed-by: ysr, tonyp
The fix addresses two memory leaks in G1 code: (1) _evac_failure_scan_stack - a resource object allocated on the C heap was not freed; (2) RSHashTable were linked into deleted list which was only cleared at full GC.
Reviewed-by: tonyp, iveresov
6889757: G1: enable card mark elision for initializing writes from compiled code (ReduceInitialCardMarks)
Defer the (compiler-elided) card-mark upon a slow-path allocation until after the store and before the next subsequent safepoint; G1 now answers yes to can_elide_tlab_write_barriers().
Reviewed-by: jcoomes, kvn, never
When we copy objects to survivors during marking, we incorrectly set NTAMS to bottom, which causes marking to miss visiting some of those objects.
Reviewed-by: apetrusenko, iveresov
It tidies up the G1 heap verification a bit. In particular, when the verification is done in parallel and there is a failure, this is propagated to the top level and the heap is dumped at the end, not by every thread that encounters a failure.
Reviewed-by: johnc, jmasa
