diff src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp @ 1145:e018e6884bd8

6631166: CMS: better heuristics when combatting fragmentation Summary: Autonomic per-worker free block cache sizing, tunable coalition policies, fixes to per-size block statistics, retuned gain and bandwidth of some feedback loop filters to allow quicker reactivity to abrupt changes in ambient demand, and other heuristics to reduce fragmentation of the CMS old gen. Also tightened some assertions, including those related to locking. Reviewed-by: jmasa
author ysr
date Wed, 23 Dec 2009 09:23:54 -0800
parents 0e2d7ae2bc67
children 4788266644c1
line wrap: on
line diff
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp	Wed Dec 16 15:12:51 2009 -0800
+++ b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp	Wed Dec 23 09:23:54 2009 -0800
@@ -253,7 +253,6 @@
   }
 }
 
-
 void ConcurrentMarkSweepGeneration::ref_processor_init() {
   assert(collector() != NULL, "no collector");
   collector()->ref_processor_init();
@@ -341,6 +340,14 @@
   _icms_duty_cycle = CMSIncrementalDutyCycle;
 }
 
+double CMSStats::cms_free_adjustment_factor(size_t free) const {
+  // TBD: CR 6909490
+  return 1.0;
+}
+
+void CMSStats::adjust_cms_free_adjustment_factor(bool fail, size_t free) {
+}
+
 // If promotion failure handling is on use
 // the padded average size of the promotion for each
 // young generation collection.
@@ -361,7 +368,11 @@
 
     // Adjust by the safety factor.
     double cms_free_dbl = (double)cms_free;
-    cms_free_dbl = cms_free_dbl * (100.0 - CMSIncrementalSafetyFactor) / 100.0;
+    double cms_adjustment = (100.0 - CMSIncrementalSafetyFactor)/100.0;
+    // Apply a further correction factor which tries to adjust
+    // for recent occurance of concurrent mode failures.
+    cms_adjustment = cms_adjustment * cms_free_adjustment_factor(cms_free);
+    cms_free_dbl = cms_free_dbl * cms_adjustment;
 
     if (PrintGCDetails && Verbose) {
       gclog_or_tty->print_cr("CMSStats::time_until_cms_gen_full: cms_free "
@@ -395,6 +406,8 @@
   // late.
   double work = cms_duration() + gc0_period();
   double deadline = time_until_cms_gen_full();
+  // If a concurrent mode failure occurred recently, we want to be
+  // more conservative and halve our expected time_until_cms_gen_full()
   if (work > deadline) {
     if (Verbose && PrintGCDetails) {
       gclog_or_tty->print(
@@ -556,7 +569,8 @@
   _should_unload_classes(false),
   _concurrent_cycles_since_last_unload(0),
   _roots_scanning_options(0),
-  _sweep_estimate(CMS_SweepWeight, CMS_SweepPadding)
+  _inter_sweep_estimate(CMS_SweepWeight, CMS_SweepPadding),
+  _intra_sweep_estimate(CMS_SweepWeight, CMS_SweepPadding)
 {
   if (ExplicitGCInvokesConcurrentAndUnloadsClasses) {
     ExplicitGCInvokesConcurrent = true;
@@ -773,7 +787,7 @@
   NOT_PRODUCT(_overflow_counter = CMSMarkStackOverflowInterval;)
   _gc_counters = new CollectorCounters("CMS", 1);
   _completed_initialization = true;
-  _sweep_timer.start();  // start of time
+  _inter_sweep_timer.start();  // start of time
 }
 
 const char* ConcurrentMarkSweepGeneration::name() const {
@@ -900,6 +914,14 @@
   return result;
 }
 
+// At a promotion failure dump information on block layout in heap
+// (cms old generation).
+void ConcurrentMarkSweepGeneration::promotion_failure_occurred() {
+  if (CMSDumpAtPromotionFailure) {
+    cmsSpace()->dump_at_safepoint_with_locks(collector(), gclog_or_tty);
+  }
+}
+
 CompactibleSpace*
 ConcurrentMarkSweepGeneration::first_compaction_space() const {
   return _cmsSpace;
@@ -1368,12 +1390,7 @@
 ConcurrentMarkSweepGeneration::
 par_promote_alloc_done(int thread_num) {
   CMSParGCThreadState* ps = _par_gc_thread_states[thread_num];
-  ps->lab.retire();
-#if CFLS_LAB_REFILL_STATS
-  if (thread_num == 0) {
-    _cmsSpace->print_par_alloc_stats();
-  }
-#endif
+  ps->lab.retire(thread_num);
 }
 
 void
@@ -1974,11 +1991,14 @@
   // We must adjust the allocation statistics being maintained
   // in the free list space. We do so by reading and clearing
   // the sweep timer and updating the block flux rate estimates below.
-  assert(_sweep_timer.is_active(), "We should never see the timer inactive");
-  _sweep_timer.stop();
-  // Note that we do not use this sample to update the _sweep_estimate.
-  _cmsGen->cmsSpace()->beginSweepFLCensus((float)(_sweep_timer.seconds()),
-                                          _sweep_estimate.padded_average());
+  assert(!_intra_sweep_timer.is_active(), "_intra_sweep_timer should be inactive");
+  if (_inter_sweep_timer.is_active()) {
+    _inter_sweep_timer.stop();
+    // Note that we do not use this sample to update the _inter_sweep_estimate.
+    _cmsGen->cmsSpace()->beginSweepFLCensus((float)(_inter_sweep_timer.seconds()),
+                                            _inter_sweep_estimate.padded_average(),
+                                            _intra_sweep_estimate.padded_average());
+  }
 
   GenMarkSweep::invoke_at_safepoint(_cmsGen->level(),
     ref_processor(), clear_all_soft_refs);
@@ -2015,10 +2035,10 @@
   }
 
   // Adjust the per-size allocation stats for the next epoch.
-  _cmsGen->cmsSpace()->endSweepFLCensus(sweepCount() /* fake */);
-  // Restart the "sweep timer" for next epoch.
-  _sweep_timer.reset();
-  _sweep_timer.start();
+  _cmsGen->cmsSpace()->endSweepFLCensus(sweep_count() /* fake */);
+  // Restart the "inter sweep timer" for the next epoch.
+  _inter_sweep_timer.reset();
+  _inter_sweep_timer.start();
 
   // Sample collection pause time and reset for collection interval.
   if (UseAdaptiveSizePolicy) {
@@ -2676,7 +2696,7 @@
   // Also reset promotion tracking in par gc thread states.
   if (ParallelGCThreads > 0) {
     for (uint i = 0; i < ParallelGCThreads; i++) {
-      _par_gc_thread_states[i]->promo.stopTrackingPromotions();
+      _par_gc_thread_states[i]->promo.stopTrackingPromotions(i);
     }
   }
 }
@@ -2771,7 +2791,7 @@
   bool do_bit(size_t offset) {
     HeapWord* addr = _marks->offsetToHeapWord(offset);
     if (!_marks->isMarked(addr)) {
-      oop(addr)->print();
+      oop(addr)->print_on(gclog_or_tty);
       gclog_or_tty->print_cr(" ("INTPTR_FORMAT" should have been marked)", addr);
       _failed = true;
     }
@@ -2820,7 +2840,7 @@
   // Clear any marks from a previous round
   verification_mark_bm()->clear_all();
   assert(verification_mark_stack()->isEmpty(), "markStack should be empty");
-  assert(overflow_list_is_empty(), "overflow list should be empty");
+  verify_work_stacks_empty();
 
   GenCollectedHeap* gch = GenCollectedHeap::heap();
   gch->ensure_parsability(false);  // fill TLABs, but no need to retire them
@@ -2893,8 +2913,8 @@
   verification_mark_bm()->iterate(&vcl);
   if (vcl.failed()) {
     gclog_or_tty->print("Verification failed");
-    Universe::heap()->print();
-    fatal(" ... aborting");
+    Universe::heap()->print_on(gclog_or_tty);
+    fatal("CMS: failed marking verification after remark");
   }
 }
 
@@ -3314,7 +3334,7 @@
     Universe::heap()->barrier_set()->resize_covered_region(mr);
     // Hmmmm... why doesn't CFLS::set_end verify locking?
     // This is quite ugly; FIX ME XXX
-    _cmsSpace->assert_locked();
+    _cmsSpace->assert_locked(freelistLock());
     _cmsSpace->set_end((HeapWord*)_virtual_space.high());
 
     // update the space and generation capacity counters
@@ -5868,9 +5888,9 @@
   check_correct_thread_executing();
   verify_work_stacks_empty();
   verify_overflow_empty();
-  incrementSweepCount();
-  _sweep_timer.stop();
-  _sweep_estimate.sample(_sweep_timer.seconds());
+  increment_sweep_count();
+  _inter_sweep_timer.stop();
+  _inter_sweep_estimate.sample(_inter_sweep_timer.seconds());
   size_policy()->avg_cms_free_at_sweep()->sample(_cmsGen->free());
 
   // PermGen verification support: If perm gen sweeping is disabled in
@@ -5893,6 +5913,9 @@
     }
   }
 
+  assert(!_intra_sweep_timer.is_active(), "Should not be active");
+  _intra_sweep_timer.reset();
+  _intra_sweep_timer.start();
   if (asynch) {
     TraceCPUTime tcpu(PrintGCDetails, true, gclog_or_tty);
     CMSPhaseAccounting pa(this, "sweep", !PrintGCDetails);
@@ -5937,8 +5960,11 @@
   verify_work_stacks_empty();
   verify_overflow_empty();
 
-  _sweep_timer.reset();
-  _sweep_timer.start();
+  _intra_sweep_timer.stop();
+  _intra_sweep_estimate.sample(_intra_sweep_timer.seconds());
+
+  _inter_sweep_timer.reset();
+  _inter_sweep_timer.start();
 
   update_time_of_last_gc(os::javaTimeMillis());
 
@@ -5981,11 +6007,11 @@
 // FIX ME!!! Looks like this belongs in CFLSpace, with
 // CMSGen merely delegating to it.
 void ConcurrentMarkSweepGeneration::setNearLargestChunk() {
-  double nearLargestPercent = 0.999;
+  double nearLargestPercent = FLSLargestBlockCoalesceProximity;
   HeapWord*  minAddr        = _cmsSpace->bottom();
   HeapWord*  largestAddr    =
     (HeapWord*) _cmsSpace->dictionary()->findLargestDict();
-  if (largestAddr == 0) {
+  if (largestAddr == NULL) {
     // The dictionary appears to be empty.  In this case
     // try to coalesce at the end of the heap.
     largestAddr = _cmsSpace->end();
@@ -5993,6 +6019,13 @@
   size_t largestOffset     = pointer_delta(largestAddr, minAddr);
   size_t nearLargestOffset =
     (size_t)((double)largestOffset * nearLargestPercent) - MinChunkSize;
+  if (PrintFLSStatistics != 0) {
+    gclog_or_tty->print_cr(
+      "CMS: Large Block: " PTR_FORMAT ";"
+      " Proximity: " PTR_FORMAT " -> " PTR_FORMAT,
+      largestAddr,
+      _cmsSpace->nearLargestChunk(), minAddr + nearLargestOffset);
+  }
   _cmsSpace->set_nearLargestChunk(minAddr + nearLargestOffset);
 }
 
@@ -6072,9 +6105,11 @@
   assert_lock_strong(gen->freelistLock());
   assert_lock_strong(bitMapLock());
 
-  assert(!_sweep_timer.is_active(), "Was switched off in an outer context");
-  gen->cmsSpace()->beginSweepFLCensus((float)(_sweep_timer.seconds()),
-                                      _sweep_estimate.padded_average());
+  assert(!_inter_sweep_timer.is_active(), "Was switched off in an outer context");
+  assert(_intra_sweep_timer.is_active(),  "Was switched on  in an outer context");
+  gen->cmsSpace()->beginSweepFLCensus((float)(_inter_sweep_timer.seconds()),
+                                      _inter_sweep_estimate.padded_average(),
+                                      _intra_sweep_estimate.padded_average());
   gen->setNearLargestChunk();
 
   {
@@ -6087,7 +6122,7 @@
     // end-of-sweep-census below will be off by a little bit.
   }
   gen->cmsSpace()->sweep_completed();
-  gen->cmsSpace()->endSweepFLCensus(sweepCount());
+  gen->cmsSpace()->endSweepFLCensus(sweep_count());
   if (should_unload_classes()) {                // unloaded classes this cycle,
     _concurrent_cycles_since_last_unload = 0;   // ... reset count
   } else {                                      // did not unload classes,