6973963: SEGV in ciBlock::start_bci() with EA

Summary: Added more checks into ResourceObj and growableArray to verify correctness of allocation type.
Reviewed-by: never, coleenp, dholmes

8 files changed, 45 insertions, 26 deletions

diff --git a/src/share/vm/opto/block.cpp b/src/share/vm/opto/block.cpp
index d7cf639c9..1d7a84048 100644
--- a/src/share/vm/opto/block.cpp
+++ b/src/share/vm/opto/block.cpp
@@ -353,7 +353,8 @@ void Block::dump( const Block_Array *bbs ) const {
 PhaseCFG::PhaseCFG( Arena *a, RootNode *r, Matcher &m ) :
   Phase(CFG),
   _bbs(a),
-  _root(r)
+  _root(r),
+  _node_latency(NULL)
 #ifndef PRODUCT
   , _trace_opto_pipelining(TraceOptoPipelining || C->method_has_option("TraceOptoPipelining"))
 #endif
diff --git a/src/share/vm/opto/block.hpp b/src/share/vm/opto/block.hpp
index 9bd989539..cc06263f1 100644
--- a/src/share/vm/opto/block.hpp
+++ b/src/share/vm/opto/block.hpp
@@ -374,7 +374,7 @@ class PhaseCFG : public Phase {
   float _outer_loop_freq;       // Outmost loop frequency
 
   // Per node latency estimation, valid only during GCM
-  GrowableArray<uint> _node_latency;
+  GrowableArray<uint> *_node_latency;
 
 #ifndef PRODUCT
   bool _trace_opto_pipelining;  // tracing flag
diff --git a/src/share/vm/opto/c2_globals.hpp b/src/share/vm/opto/c2_globals.hpp
index ac0a991d4..3fdaf8427 100644
--- a/src/share/vm/opto/c2_globals.hpp
+++ b/src/share/vm/opto/c2_globals.hpp
@@ -281,6 +281,12 @@
   product(bool, InsertMemBarAfterArraycopy, true,                           \
           "Insert memory barrier after arraycopy call")                     \
                                                                             \
+  develop(bool, SubsumeLoads, true,                                         \
+          "Attempt to compile while subsuming loads into machine instructions.") \
+                                                                            \
+  develop(bool, StressRecompilation, false,                                 \
+          "Recompile each compiled method without subsuming loads or escape analysis.") \
+                                                                            \
   /* controls for tier 1 compilations */                                    \
                                                                             \
   develop(bool, Tier1CountInvocations, true,                                \
diff --git a/src/share/vm/opto/c2compiler.cpp b/src/share/vm/opto/c2compiler.cpp
index ff4fa6bad..142273b26 100644
--- a/src/share/vm/opto/c2compiler.cpp
+++ b/src/share/vm/opto/c2compiler.cpp
@@ -103,13 +103,14 @@ void C2Compiler::compile_method(ciEnv* env,
   if (!is_initialized()) {
     initialize();
   }
-  bool subsume_loads = true;
+  bool subsume_loads = SubsumeLoads;
   bool do_escape_analysis = DoEscapeAnalysis &&
     !env->jvmti_can_access_local_variables();
   while (!env->failing()) {
     // Attempt to compile while subsuming loads into machine instructions.
     Compile C(env, this, target, entry_bci, subsume_loads, do_escape_analysis);
 
+
     // Check result and retry if appropriate.
     if (C.failure_reason() != NULL) {
       if (C.failure_reason_is(retry_no_subsuming_loads())) {
@@ -127,6 +128,16 @@ void C2Compiler::compile_method(ciEnv* env,
       // on the ciEnv via env->record_method_not_compilable().
       env->record_failure(C.failure_reason());
     }
+    if (StressRecompilation) {
+      if (subsume_loads) {
+        subsume_loads = false;
+        continue;  // retry
+      }
+      if (do_escape_analysis) {
+        do_escape_analysis = false;
+        continue;  // retry
+      }
+    }
 
     // No retry; just break the loop.
     break;
diff --git a/src/share/vm/opto/chaitin.cpp b/src/share/vm/opto/chaitin.cpp
index ed9124b24..4a8e0e3f5 100644
--- a/src/share/vm/opto/chaitin.cpp
+++ b/src/share/vm/opto/chaitin.cpp
@@ -569,7 +569,7 @@ void PhaseChaitin::gather_lrg_masks( bool after_aggressive ) {
         if (trace_spilling() && lrg._def != NULL) {
           // collect defs for MultiDef printing
           if (lrg._defs == NULL) {
-            lrg._defs = new (_ifg->_arena) GrowableArray<Node*>();
+            lrg._defs = new (_ifg->_arena) GrowableArray<Node*>(_ifg->_arena, 2, 0, NULL);
             lrg._defs->append(lrg._def);
           }
           lrg._defs->append(n);
diff --git a/src/share/vm/opto/compile.cpp b/src/share/vm/opto/compile.cpp
index 632066521..e5d3c4fdd 100644
--- a/src/share/vm/opto/compile.cpp
+++ b/src/share/vm/opto/compile.cpp
@@ -904,8 +904,8 @@ void Compile::Init(int aliaslevel) {
   probe_alias_cache(NULL)->_index = AliasIdxTop;
 
   _intrinsics = NULL;
-  _macro_nodes = new GrowableArray<Node*>(comp_arena(), 8,  0, NULL);
-  _predicate_opaqs = new GrowableArray<Node*>(comp_arena(), 8,  0, NULL);
+  _macro_nodes = new(comp_arena()) GrowableArray<Node*>(comp_arena(), 8,  0, NULL);
+  _predicate_opaqs = new(comp_arena()) GrowableArray<Node*>(comp_arena(), 8,  0, NULL);
   register_library_intrinsics();
 }
 
diff --git a/src/share/vm/opto/gcm.cpp b/src/share/vm/opto/gcm.cpp
index 5fc0b1c1b..f222001d3 100644
--- a/src/share/vm/opto/gcm.cpp
+++ b/src/share/vm/opto/gcm.cpp
@@ -841,7 +841,7 @@ void PhaseCFG::partial_latency_of_defs(Node *n) {
 #ifndef PRODUCT
   if (trace_opto_pipelining()) {
     tty->print("# latency_to_inputs: node_latency[%d] = %d for node",
-               n->_idx, _node_latency.at_grow(n->_idx));
+               n->_idx, _node_latency->at_grow(n->_idx));
     dump();
   }
 #endif
@@ -853,7 +853,7 @@ void PhaseCFG::partial_latency_of_defs(Node *n) {
     return;
 
   uint nlen = n->len();
-  uint use_latency = _node_latency.at_grow(n->_idx);
+  uint use_latency = _node_latency->at_grow(n->_idx);
   uint use_pre_order = _bbs[n->_idx]->_pre_order;
 
   for ( uint j=0; j<nlen; j++ ) {
@@ -884,15 +884,15 @@ void PhaseCFG::partial_latency_of_defs(Node *n) {
     uint delta_latency = n->latency(j);
     uint current_latency = delta_latency + use_latency;
 
-    if (_node_latency.at_grow(def->_idx) < current_latency) {
-      _node_latency.at_put_grow(def->_idx, current_latency);
+    if (_node_latency->at_grow(def->_idx) < current_latency) {
+      _node_latency->at_put_grow(def->_idx, current_latency);
     }
 
 #ifndef PRODUCT
     if (trace_opto_pipelining()) {
       tty->print_cr("#      %d + edge_latency(%d) == %d -> %d, node_latency[%d] = %d",
                     use_latency, j, delta_latency, current_latency, def->_idx,
-                    _node_latency.at_grow(def->_idx));
+                    _node_latency->at_grow(def->_idx));
     }
 #endif
   }
@@ -926,7 +926,7 @@ int PhaseCFG::latency_from_use(Node *n, const Node *def, Node *use) {
       return 0;
 
     uint nlen = use->len();
-    uint nl = _node_latency.at_grow(use->_idx);
+    uint nl = _node_latency->at_grow(use->_idx);
 
     for ( uint j=0; j<nlen; j++ ) {
       if (use->in(j) == n) {
@@ -962,7 +962,7 @@ void PhaseCFG::latency_from_uses(Node *n) {
 #ifndef PRODUCT
   if (trace_opto_pipelining()) {
     tty->print("# latency_from_outputs: node_latency[%d] = %d for node",
-               n->_idx, _node_latency.at_grow(n->_idx));
+               n->_idx, _node_latency->at_grow(n->_idx));
     dump();
   }
 #endif
@@ -975,7 +975,7 @@ void PhaseCFG::latency_from_uses(Node *n) {
     if (latency < l) latency = l;
   }
 
-  _node_latency.at_put_grow(n->_idx, latency);
+  _node_latency->at_put_grow(n->_idx, latency);
 }
 
 //------------------------------hoist_to_cheaper_block-------------------------
@@ -985,9 +985,9 @@ Block* PhaseCFG::hoist_to_cheaper_block(Block* LCA, Block* early, Node* self) {
   const double delta = 1+PROB_UNLIKELY_MAG(4);
   Block* least       = LCA;
   double least_freq  = least->_freq;
-  uint target        = _node_latency.at_grow(self->_idx);
-  uint start_latency = _node_latency.at_grow(LCA->_nodes[0]->_idx);
-  uint end_latency   = _node_latency.at_grow(LCA->_nodes[LCA->end_idx()]->_idx);
+  uint target        = _node_latency->at_grow(self->_idx);
+  uint start_latency = _node_latency->at_grow(LCA->_nodes[0]->_idx);
+  uint end_latency   = _node_latency->at_grow(LCA->_nodes[LCA->end_idx()]->_idx);
   bool in_latency    = (target <= start_latency);
   const Block* root_block = _bbs[_root->_idx];
 
@@ -1005,7 +1005,7 @@ Block* PhaseCFG::hoist_to_cheaper_block(Block* LCA, Block* early, Node* self) {
 #ifndef PRODUCT
   if (trace_opto_pipelining()) {
     tty->print("# Find cheaper block for latency %d: ",
-      _node_latency.at_grow(self->_idx));
+      _node_latency->at_grow(self->_idx));
     self->dump();
     tty->print_cr("#   B%d: start latency for [%4d]=%d, end latency for [%4d]=%d, freq=%g",
       LCA->_pre_order,
@@ -1032,9 +1032,9 @@ Block* PhaseCFG::hoist_to_cheaper_block(Block* LCA, Block* early, Node* self) {
     if (mach && LCA == root_block)
       break;
 
-    uint start_lat = _node_latency.at_grow(LCA->_nodes[0]->_idx);
+    uint start_lat = _node_latency->at_grow(LCA->_nodes[0]->_idx);
     uint end_idx   = LCA->end_idx();
-    uint end_lat   = _node_latency.at_grow(LCA->_nodes[end_idx]->_idx);
+    uint end_lat   = _node_latency->at_grow(LCA->_nodes[end_idx]->_idx);
     double LCA_freq = LCA->_freq;
 #ifndef PRODUCT
     if (trace_opto_pipelining()) {
@@ -1073,7 +1073,7 @@ Block* PhaseCFG::hoist_to_cheaper_block(Block* LCA, Block* early, Node* self) {
       tty->print_cr("#  Change latency for [%4d] from %d to %d", self->_idx, target, end_latency);
     }
 #endif
-    _node_latency.at_put_grow(self->_idx, end_latency);
+    _node_latency->at_put_grow(self->_idx, end_latency);
     partial_latency_of_defs(self);
   }
 
@@ -1255,8 +1255,7 @@ void PhaseCFG::GlobalCodeMotion( Matcher &matcher, uint unique, Node_List &proj_
 
   // Compute the latency information (via backwards walk) for all the
   // instructions in the graph
-  GrowableArray<uint> node_latency;
-  _node_latency = node_latency;
+  _node_latency = new GrowableArray<uint>(); // resource_area allocation
 
   if( C->do_scheduling() )
     ComputeLatenciesBackwards(visited, stack);
@@ -1341,6 +1340,8 @@ void PhaseCFG::GlobalCodeMotion( Matcher &matcher, uint unique, Node_List &proj_
     }
   }
 #endif
+  // Dead.
+  _node_latency = (GrowableArray<uint> *)0xdeadbeef;
 }
 
 
diff --git a/src/share/vm/opto/lcm.cpp b/src/share/vm/opto/lcm.cpp
index 0afde9016..8692746cf 100644
--- a/src/share/vm/opto/lcm.cpp
+++ b/src/share/vm/opto/lcm.cpp
@@ -461,7 +461,7 @@ Node *Block::select(PhaseCFG *cfg, Node_List &worklist, int *ready_cnt, VectorSe
       n_choice = 1;
     }
 
-    uint n_latency = cfg->_node_latency.at_grow(n->_idx);
+    uint n_latency = cfg->_node_latency->at_grow(n->_idx);
     uint n_score   = n->req();   // Many inputs get high score to break ties
 
     // Keep best latency found
@@ -738,7 +738,7 @@ bool Block::schedule_local(PhaseCFG *cfg, Matcher &matcher, int *ready_cnt, Vect
         Node     *n = _nodes[j];
         int     idx = n->_idx;
         tty->print("#   ready cnt:%3d  ", ready_cnt[idx]);
-        tty->print("latency:%3d  ", cfg->_node_latency.at_grow(idx));
+        tty->print("latency:%3d  ", cfg->_node_latency->at_grow(idx));
         tty->print("%4d: %s\n", idx, n->Name());
       }
     }
@@ -765,7 +765,7 @@ bool Block::schedule_local(PhaseCFG *cfg, Matcher &matcher, int *ready_cnt, Vect
 #ifndef PRODUCT
     if (cfg->trace_opto_pipelining()) {
       tty->print("#    select %d: %s", n->_idx, n->Name());
-      tty->print(", latency:%d", cfg->_node_latency.at_grow(n->_idx));
+      tty->print(", latency:%d", cfg->_node_latency->at_grow(n->_idx));
       n->dump();
       if (Verbose) {
         tty->print("#   ready list:");