Merge

src/os/linux/vm/os_linux.cpp

@@ -2767,7 +2767,19 @@ void os::numa_make_global(char *addr, size_t bytes) {
   Linux::numa_interleave_memory(addr, bytes);
 }
 
+// Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
+// bind policy to MPOL_PREFERRED for the current thread.
+#define USE_MPOL_PREFERRED 0
+
 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
+  // To make NUMA and large pages more robust when both enabled, we need to ease
+  // the requirements on where the memory should be allocated. MPOL_BIND is the
+  // default policy and it will force memory to be allocated on the specified
+  // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
+  // the specified node, but will not force it. Using this policy will prevent
+  // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
+  // free large pages.
+  Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
 }
 
@@ -2869,6 +2881,8 @@ bool os::Linux::libnuma_init() {
                                             libnuma_dlsym(handle, "numa_tonode_memory")));
       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
                                             libnuma_dlsym(handle, "numa_interleave_memory")));
+      set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
+                                            libnuma_dlsym(handle, "numa_set_bind_policy")));
 
 
       if (numa_available() != -1) {
@@ -2935,6 +2949,7 @@ os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
 os::Linux::numa_available_func_t os::Linux::_numa_available;
 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
+os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
 unsigned long* os::Linux::_numa_all_nodes;
 
 bool os::pd_uncommit_memory(char* addr, size_t size) {

src/os/linux/vm/os_linux.hpp

@@ -235,6 +235,7 @@ private:
   typedef int (*numa_available_func_t)(void);
   typedef int (*numa_tonode_memory_func_t)(void *start, size_t size, int node);
   typedef void (*numa_interleave_memory_func_t)(void *start, size_t size, unsigned long *nodemask);
+  typedef void (*numa_set_bind_policy_func_t)(int policy);
 
   static sched_getcpu_func_t _sched_getcpu;
   static numa_node_to_cpus_func_t _numa_node_to_cpus;
@@ -242,6 +243,7 @@ private:
   static numa_available_func_t _numa_available;
   static numa_tonode_memory_func_t _numa_tonode_memory;
   static numa_interleave_memory_func_t _numa_interleave_memory;
+  static numa_set_bind_policy_func_t _numa_set_bind_policy;
   static unsigned long* _numa_all_nodes;
 
   static void set_sched_getcpu(sched_getcpu_func_t func) { _sched_getcpu = func; }
@@ -250,6 +252,7 @@ private:
   static void set_numa_available(numa_available_func_t func) { _numa_available = func; }
   static void set_numa_tonode_memory(numa_tonode_memory_func_t func) { _numa_tonode_memory = func; }
   static void set_numa_interleave_memory(numa_interleave_memory_func_t func) { _numa_interleave_memory = func; }
+  static void set_numa_set_bind_policy(numa_set_bind_policy_func_t func) { _numa_set_bind_policy = func; }
   static void set_numa_all_nodes(unsigned long* ptr) { _numa_all_nodes = ptr; }
   static int sched_getcpu_syscall(void);
 public:
@@ -267,6 +270,11 @@ public:
       _numa_interleave_memory(start, size, _numa_all_nodes);
     }
   }
+  static void numa_set_bind_policy(int policy) {
+    if (_numa_set_bind_policy != NULL) {
+      _numa_set_bind_policy(policy);
+    }
+  }
   static int get_node_by_cpu(int cpu_id);
 };
 

src/os_cpu/solaris_x86/vm/globals_solaris_x86.hpp

@@ -44,6 +44,6 @@ define_pd_global(uintx,JVMInvokeMethodSlack,     10*K);
 define_pd_global(intx, CompilerThreadStackSize,  0);
 
 // Used on 64 bit platforms for UseCompressedOops base address
-define_pd_global(uintx,HeapBaseMinAddress,       256*M);
+define_pd_global(uintx,HeapBaseMinAddress,       2*G);
 
 #endif // OS_CPU_SOLARIS_X86_VM_GLOBALS_SOLARIS_X86_HPP

src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp

@@ -2493,11 +2493,11 @@ void G1CollectedHeap::register_concurrent_cycle_start(jlong start_time) {
 
 void G1CollectedHeap::register_concurrent_cycle_end() {
   if (_concurrent_cycle_started) {
-    _gc_timer_cm->register_gc_end(os::elapsed_counter());
-
     if (_cm->has_aborted()) {
       _gc_tracer_cm->report_concurrent_mode_failure();
     }
+
+    _gc_timer_cm->register_gc_end(os::elapsed_counter());
     _gc_tracer_cm->report_gc_end(_gc_timer_cm->gc_end(), _gc_timer_cm->time_partitions());
 
     _concurrent_cycle_started = false;

src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp

@@ -168,7 +168,15 @@ G1CollectorPolicy::G1CollectorPolicy() :
   // Set up the region size and associated fields. Given that the
   // policy is created before the heap, we have to set this up here,
   // so it's done as soon as possible.
-  HeapRegion::setup_heap_region_size(Arguments::min_heap_size());
+
+  // It would have been natural to pass initial_heap_byte_size() and
+  // max_heap_byte_size() to setup_heap_region_size() but those have
+  // not been set up at this point since they should be aligned with
+  // the region size. So, there is a circular dependency here. We base
+  // the region size on the heap size, but the heap size should be
+  // aligned with the region size. To get around this we use the
+  // unaligned values for the heap.
+  HeapRegion::setup_heap_region_size(InitialHeapSize, MaxHeapSize);
   HeapRegionRemSet::setup_remset_size();
 
   G1ErgoVerbose::initialize();

src/share/vm/gc_implementation/g1/heapRegion.cpp

@@ -149,18 +149,11 @@ void HeapRegionDCTOC::walk_mem_region_with_cl(MemRegion mr,
 // many regions in the heap (based on the min heap size).
 #define TARGET_REGION_NUMBER          2048
 
-void HeapRegion::setup_heap_region_size(uintx min_heap_size) {
-  // region_size in bytes
+void HeapRegion::setup_heap_region_size(size_t initial_heap_size, size_t max_heap_size) {
   uintx region_size = G1HeapRegionSize;
   if (FLAG_IS_DEFAULT(G1HeapRegionSize)) {
-    // We base the automatic calculation on the min heap size. This
-    // can be problematic if the spread between min and max is quite
-    // wide, imagine -Xms128m -Xmx32g. But, if we decided it based on
-    // the max size, the region size might be way too large for the
-    // min size. Either way, some users might have to set the region
-    // size manually for some -Xms / -Xmx combos.
-
-    region_size = MAX2(min_heap_size / TARGET_REGION_NUMBER,
+    size_t average_heap_size = (initial_heap_size + max_heap_size) / 2;
+    region_size = MAX2(average_heap_size / TARGET_REGION_NUMBER,
                        (uintx) MIN_REGION_SIZE);
   }
 

src/share/vm/gc_implementation/g1/heapRegion.hpp

@@ -361,7 +361,7 @@ class HeapRegion: public G1OffsetTableContigSpace {
   // CardsPerRegion). All those fields are considered constant
   // throughout the JVM's execution, therefore they should only be set
   // up once during initialization time.
-  static void setup_heap_region_size(uintx min_heap_size);
+  static void setup_heap_region_size(size_t initial_heap_size, size_t max_heap_size);
 
   enum ClaimValues {
     InitialClaimValue          = 0,

src/share/vm/gc_implementation/shared/gcTraceSend.cpp

@@ -28,6 +28,7 @@
 #include "gc_implementation/shared/gcTrace.hpp"
 #include "gc_implementation/shared/gcWhen.hpp"
 #include "gc_implementation/shared/copyFailedInfo.hpp"
+#include "runtime/os.hpp"
 #include "trace/tracing.hpp"
 #include "trace/traceBackend.hpp"
 #if INCLUDE_ALL_GCS
@@ -54,11 +55,12 @@ void GCTracer::send_garbage_collection_event() const {
 }
 
 void GCTracer::send_reference_stats_event(ReferenceType type, size_t count) const {
-  EventGCReferenceStatistics e;
+  EventGCReferenceStatistics e(UNTIMED);
   if (e.should_commit()) {
       e.set_gcId(_shared_gc_info.id());
       e.set_type((u1)type);
       e.set_count(count);
+      e.set_endtime(os::elapsed_counter());
       e.commit();
   }
 }
@@ -105,20 +107,22 @@ static TraceStructCopyFailed to_trace_struct(const CopyFailedInfo& cf_info) {
 }
 
 void YoungGCTracer::send_promotion_failed_event(const PromotionFailedInfo& pf_info) const {
-  EventPromotionFailed e;
+  EventPromotionFailed e(UNTIMED);
   if (e.should_commit()) {
     e.set_gcId(_shared_gc_info.id());
     e.set_data(to_trace_struct(pf_info));
     e.set_thread(pf_info.thread()->thread_id());
+    e.set_endtime(os::elapsed_counter());
     e.commit();
   }
 }
 
 // Common to CMS and G1
 void OldGCTracer::send_concurrent_mode_failure_event() {
-  EventConcurrentModeFailure e;
+  EventConcurrentModeFailure e(UNTIMED);
   if (e.should_commit()) {
     e.set_gcId(_shared_gc_info.id());
+    e.set_endtime(os::elapsed_counter());
     e.commit();
   }
 }
@@ -136,7 +140,7 @@ void G1NewTracer::send_g1_young_gc_event() {
 }
 
 void G1NewTracer::send_evacuation_info_event(EvacuationInfo* info) {
-  EventEvacuationInfo e;
+  EventEvacuationInfo e(UNTIMED);
   if (e.should_commit()) {
     e.set_gcId(_shared_gc_info.id());
     e.set_cSetRegions(info->collectionset_regions());
@@ -147,15 +151,17 @@ void G1NewTracer::send_evacuation_info_event(EvacuationInfo* info) {
     e.set_allocRegionsUsedAfter(info->alloc_regions_used_before() + info->bytes_copied());
     e.set_bytesCopied(info->bytes_copied());
     e.set_regionsFreed(info->regions_freed());
+    e.set_endtime(os::elapsed_counter());
     e.commit();
   }
 }
 
 void G1NewTracer::send_evacuation_failed_event(const EvacuationFailedInfo& ef_info) const {
-  EventEvacuationFailed e;
+  EventEvacuationFailed e(UNTIMED);
   if (e.should_commit()) {
     e.set_gcId(_shared_gc_info.id());
     e.set_data(to_trace_struct(ef_info));
+    e.set_endtime(os::elapsed_counter());
     e.commit();
   }
 }
@@ -189,12 +195,13 @@ class GCHeapSummaryEventSender : public GCHeapSummaryVisitor {
   void visit(const GCHeapSummary* heap_summary) const {
     const VirtualSpaceSummary& heap_space = heap_summary->heap();
 
-    EventGCHeapSummary e;
+    EventGCHeapSummary e(UNTIMED);
     if (e.should_commit()) {
       e.set_gcId(_id);
       e.set_when((u1)_when);
       e.set_heapSpace(to_trace_struct(heap_space));
       e.set_heapUsed(heap_summary->used());
+      e.set_endtime(os::elapsed_counter());
       e.commit();
     }
   }
@@ -209,7 +216,7 @@ class GCHeapSummaryEventSender : public GCHeapSummaryVisitor {
     const SpaceSummary& from_space = ps_heap_summary->from();
     const SpaceSummary& to_space = ps_heap_summary->to();
 
-    EventPSHeapSummary e;
+    EventPSHeapSummary e(UNTIMED);
     if (e.should_commit()) {
       e.set_gcId(_id);
       e.set_when((u1)_when);
@@ -220,6 +227,7 @@ class GCHeapSummaryEventSender : public GCHeapSummaryVisitor {
       e.set_edenSpace(to_trace_struct(ps_heap_summary->eden()));
       e.set_fromSpace(to_trace_struct(ps_heap_summary->from()));
       e.set_toSpace(to_trace_struct(ps_heap_summary->to()));
+      e.set_endtime(os::elapsed_counter());
       e.commit();
     }
   }
@@ -241,13 +249,14 @@ static TraceStructMetaspaceSizes to_trace_struct(const MetaspaceSizes& sizes) {
 }
 
 void GCTracer::send_meta_space_summary_event(GCWhen::Type when, const MetaspaceSummary& meta_space_summary) const {
-  EventMetaspaceSummary e;
+  EventMetaspaceSummary e(UNTIMED);
   if (e.should_commit()) {
     e.set_gcId(_shared_gc_info.id());
     e.set_when((u1) when);
     e.set_metaspace(to_trace_struct(meta_space_summary.meta_space()));
     e.set_dataSpace(to_trace_struct(meta_space_summary.data_space()));
     e.set_classSpace(to_trace_struct(meta_space_summary.class_space()));
+    e.set_endtime(os::elapsed_counter());
     e.commit();
   }
 }
@@ -282,8 +291,6 @@ class PhaseSender : public PhaseVisitor {
       default: /* Ignore sending this phase */ break;
     }
   }
-
-#undef send_phase
 };
 
 void GCTracer::send_phase_events(TimePartitions* time_partitions) const {

src/share/vm/runtime/arguments.cpp

@@ -1605,17 +1605,6 @@ julong Arguments::limit_by_allocatable_memory(julong limit) {
   return result;
 }
 
-void Arguments::set_heap_base_min_address() {
-  if (FLAG_IS_DEFAULT(HeapBaseMinAddress) && UseG1GC && HeapBaseMinAddress < 1*G) {
-    // By default HeapBaseMinAddress is 2G on all platforms except Solaris x86.
-    // G1 currently needs a lot of C-heap, so on Solaris we have to give G1
-    // some extra space for the C-heap compared to other collectors.
-    // Use FLAG_SET_DEFAULT here rather than FLAG_SET_ERGO to make sure that
-    // code that checks for default values work correctly.
-    FLAG_SET_DEFAULT(HeapBaseMinAddress, 1*G);
-  }
-}
-
 void Arguments::set_heap_size() {
   if (!FLAG_IS_DEFAULT(DefaultMaxRAMFraction)) {
     // Deprecated flag
@@ -3537,8 +3526,6 @@ jint Arguments::parse(const JavaVMInitArgs* args) {
     }
   }
 
-  set_heap_base_min_address();
-
   // Set heap size based on available physical memory
   set_heap_size();
 

src/share/vm/runtime/arguments.hpp

@@ -334,8 +334,6 @@ class Arguments : AllStatic {
   // limits the given memory size by the maximum amount of memory this process is
   // currently allowed to allocate or reserve.
   static julong limit_by_allocatable_memory(julong size);
-  // Setup HeapBaseMinAddress
-  static void set_heap_base_min_address();
   // Setup heap size
   static void set_heap_size();
   // Based on automatic selection criteria, should the