8230434: [C1, C2] Release barrier for volatile field stores in constructors implemented inconsistently
Reviewed-by: shade, lucy
--- a/src/hotspot/share/c1/c1_GraphBuilder.cpp Wed Sep 04 09:47:55 2019 +0200
+++ b/src/hotspot/share/c1/c1_GraphBuilder.cpp Wed Sep 04 14:21:54 2019 +0200
@@ -1467,11 +1467,12 @@
call_register_finalizer();
}
+ // The conditions for a memory barrier are described in Parse::do_exits().
bool need_mem_bar = false;
if (method()->name() == ciSymbol::object_initializer_name() &&
- (scope()->wrote_final() || (AlwaysSafeConstructors && scope()->wrote_fields())
- || (support_IRIW_for_not_multiple_copy_atomic_cpu && scope()->wrote_volatile())
- )){
+ (scope()->wrote_final() ||
+ (AlwaysSafeConstructors && scope()->wrote_fields()) ||
+ (support_IRIW_for_not_multiple_copy_atomic_cpu && scope()->wrote_volatile()))) {
need_mem_bar = true;
}
--- a/src/hotspot/share/opto/parse1.cpp Wed Sep 04 09:47:55 2019 +0200
+++ b/src/hotspot/share/opto/parse1.cpp Wed Sep 04 14:21:54 2019 +0200
@@ -979,15 +979,18 @@
// Rather than put a barrier on only those writes which are required
// to complete, we force all writes to complete.
//
- // 2. On PPC64, also add MemBarRelease for constructors which write
- // volatile fields. As support_IRIW_for_not_multiple_copy_atomic_cpu
- // is set on PPC64, no sync instruction is issued after volatile
- // stores. We want to guarantee the same behavior as on platforms
- // with total store order, although this is not required by the Java
- // memory model. So as with finals, we add a barrier here.
+ // 2. Experimental VM option is used to force the barrier if any field
+ // was written out in the constructor.
//
- // 3. Experimental VM option is used to force the barrier if any field
- // was written out in the constructor.
+ // 3. On processors which are not CPU_MULTI_COPY_ATOMIC (e.g. PPC64),
+ // support_IRIW_for_not_multiple_copy_atomic_cpu selects that
+ // MemBarVolatile is used before volatile load instead of after volatile
+ // store, so there's no barrier after the store.
+ // We want to guarantee the same behavior as on platforms with total store
+ // order, although this is not required by the Java memory model.
+ // In this case, we want to enforce visibility of volatile field
+ // initializations which are performed in constructors.
+ // So as with finals, we add a barrier here.
//
// "All bets are off" unless the first publication occurs after a
// normal return from the constructor. We do not attempt to detect
@@ -995,9 +998,9 @@
// exceptional returns, since they cannot publish normally.
//
if (method()->is_initializer() &&
- (wrote_final() ||
- PPC64_ONLY(wrote_volatile() ||)
- (AlwaysSafeConstructors && wrote_fields()))) {
+ (wrote_final() ||
+ (AlwaysSafeConstructors && wrote_fields()) ||
+ (support_IRIW_for_not_multiple_copy_atomic_cpu && wrote_volatile()))) {
_exits.insert_mem_bar(Op_MemBarRelease, alloc_with_final());
// If Memory barrier is created for final fields write