8201491: G1 support for java.lang.ref.Reference precleaning
Summary: Implement single-threaded concurrent reference precleaning for G1.
Reviewed-by: sangheki, kbarrett
/*
* Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation.
*
* This code is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
* version 2 for more details (a copy is included in the LICENSE file that
* accompanied this code).
*
* You should have received a copy of the GNU General Public License version
* 2 along with this work; if not, write to the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
*
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
* or visit www.oracle.com if you need additional information or have any
* questions.
*
*/
#ifndef SHARE_VM_MEMORY_ITERATOR_HPP
#define SHARE_VM_MEMORY_ITERATOR_HPP
#include "memory/allocation.hpp"
#include "memory/memRegion.hpp"
#include "oops/oopsHierarchy.hpp"
class CodeBlob;
class nmethod;
class ReferenceDiscoverer;
class DataLayout;
class KlassClosure;
class ClassLoaderData;
class Symbol;
// The following classes are C++ `closures` for iterating over objects, roots and spaces
class Closure : public StackObj { };
// OopClosure is used for iterating through references to Java objects.
class OopClosure : public Closure {
public:
virtual void do_oop(oop* o) = 0;
virtual void do_oop(narrowOop* o) = 0;
};
class DoNothingClosure : public OopClosure {
public:
virtual void do_oop(oop* p) {}
virtual void do_oop(narrowOop* p) {}
};
extern DoNothingClosure do_nothing_cl;
// ExtendedOopClosure adds extra code to be run during oop iterations.
// This is needed by the GC and is extracted to a separate type to not
// pollute the OopClosure interface.
class ExtendedOopClosure : public OopClosure {
private:
ReferenceDiscoverer* _ref_discoverer;
protected:
ExtendedOopClosure(ReferenceDiscoverer* rd) : _ref_discoverer(rd) { }
ExtendedOopClosure() : _ref_discoverer(NULL) { }
~ExtendedOopClosure() { }
void set_ref_discoverer_internal(ReferenceDiscoverer* rd) { _ref_discoverer = rd; }
public:
ReferenceDiscoverer* ref_discoverer() const { return _ref_discoverer; }
// Iteration of InstanceRefKlasses differ depending on the closure,
// the below enum describes the different alternatives.
enum ReferenceIterationMode {
DO_DISCOVERY, // Apply closure and discover references
DO_DISCOVERED_AND_DISCOVERY, // Apply closure to discovered field and do discovery
DO_FIELDS // Apply closure to all fields
};
// The default iteration mode is to do discovery.
virtual ReferenceIterationMode reference_iteration_mode() { return DO_DISCOVERY; }
// If the do_metadata functions return "true",
// we invoke the following when running oop_iterate():
//
// 1) do_klass on the header klass pointer.
// 2) do_klass on the klass pointer in the mirrors.
// 3) do_cld on the class loader data in class loaders.
//
// The virtual (without suffix) and the non-virtual (with _nv suffix) need
// to be updated together, or else the devirtualization will break.
//
// Providing default implementations of the _nv functions unfortunately
// removes the compile-time safeness, but reduces the clutter for the
// ExtendedOopClosures that don't need to walk the metadata.
// Currently, only CMS and G1 need these.
bool do_metadata_nv() { return false; }
virtual bool do_metadata() { return do_metadata_nv(); }
void do_klass_nv(Klass* k) { ShouldNotReachHere(); }
virtual void do_klass(Klass* k) { do_klass_nv(k); }
void do_cld_nv(ClassLoaderData* cld) { ShouldNotReachHere(); }
virtual void do_cld(ClassLoaderData* cld) { do_cld_nv(cld); }
// True iff this closure may be safely applied more than once to an oop
// location without an intervening "major reset" (like the end of a GC).
virtual bool idempotent() { return false; }
#ifdef ASSERT
// Default verification of each visited oop field.
template <typename T> void verify(T* p);
// Can be used by subclasses to turn off the default verification of oop fields.
virtual bool should_verify_oops() { return true; }
#endif
};
// Wrapper closure only used to implement oop_iterate_no_header().
class NoHeaderExtendedOopClosure : public ExtendedOopClosure {
OopClosure* _wrapped_closure;
public:
NoHeaderExtendedOopClosure(OopClosure* cl) : _wrapped_closure(cl) {}
// Warning: this calls the virtual version do_oop in the the wrapped closure.
void do_oop_nv(oop* p) { _wrapped_closure->do_oop(p); }
void do_oop_nv(narrowOop* p) { _wrapped_closure->do_oop(p); }
void do_oop(oop* p) { assert(false, "Only the _nv versions should be used");
_wrapped_closure->do_oop(p); }
void do_oop(narrowOop* p) { assert(false, "Only the _nv versions should be used");
_wrapped_closure->do_oop(p);}
};
class KlassClosure : public Closure {
public:
virtual void do_klass(Klass* k) = 0;
};
class CLDClosure : public Closure {
public:
virtual void do_cld(ClassLoaderData* cld) = 0;
};
class CLDToOopClosure : public CLDClosure {
OopClosure* _oop_closure;
bool _must_claim_cld;
public:
CLDToOopClosure(OopClosure* oop_closure, bool must_claim_cld = true) :
_oop_closure(oop_closure),
_must_claim_cld(must_claim_cld) {}
void do_cld(ClassLoaderData* cld);
};
// The base class for all concurrent marking closures,
// that participates in class unloading.
// It's used to proxy through the metadata to the oops defined in them.
class MetadataAwareOopClosure: public ExtendedOopClosure {
public:
MetadataAwareOopClosure() : ExtendedOopClosure() { }
MetadataAwareOopClosure(ReferenceDiscoverer* rd) : ExtendedOopClosure(rd) { }
bool do_metadata_nv() { return true; }
virtual bool do_metadata() { return do_metadata_nv(); }
void do_klass_nv(Klass* k);
virtual void do_klass(Klass* k) { do_klass_nv(k); }
void do_cld_nv(ClassLoaderData* cld);
virtual void do_cld(ClassLoaderData* cld) { do_cld_nv(cld); }
};
// ObjectClosure is used for iterating through an object space
class ObjectClosure : public Closure {
public:
// Called for each object.
virtual void do_object(oop obj) = 0;
};
class BoolObjectClosure : public Closure {
public:
virtual bool do_object_b(oop obj) = 0;
};
class AlwaysTrueClosure: public BoolObjectClosure {
public:
bool do_object_b(oop p) { return true; }
};
class AlwaysFalseClosure : public BoolObjectClosure {
public:
bool do_object_b(oop p) { return false; }
};
// Applies an oop closure to all ref fields in objects iterated over in an
// object iteration.
class ObjectToOopClosure: public ObjectClosure {
ExtendedOopClosure* _cl;
public:
void do_object(oop obj);
ObjectToOopClosure(ExtendedOopClosure* cl) : _cl(cl) {}
};
// A version of ObjectClosure that is expected to be robust
// in the face of possibly uninitialized objects.
class ObjectClosureCareful : public ObjectClosure {
public:
virtual size_t do_object_careful_m(oop p, MemRegion mr) = 0;
virtual size_t do_object_careful(oop p) = 0;
};
// The following are used in CompactibleFreeListSpace and
// ConcurrentMarkSweepGeneration.
// Blk closure (abstract class)
class BlkClosure : public StackObj {
public:
virtual size_t do_blk(HeapWord* addr) = 0;
};
// A version of BlkClosure that is expected to be robust
// in the face of possibly uninitialized objects.
class BlkClosureCareful : public BlkClosure {
public:
size_t do_blk(HeapWord* addr) {
guarantee(false, "call do_blk_careful instead");
return 0;
}
virtual size_t do_blk_careful(HeapWord* addr) = 0;
};
// SpaceClosure is used for iterating over spaces
class Space;
class CompactibleSpace;
class SpaceClosure : public StackObj {
public:
// Called for each space
virtual void do_space(Space* s) = 0;
};
class CompactibleSpaceClosure : public StackObj {
public:
// Called for each compactible space
virtual void do_space(CompactibleSpace* s) = 0;
};
// CodeBlobClosure is used for iterating through code blobs
// in the code cache or on thread stacks
class CodeBlobClosure : public Closure {
public:
// Called for each code blob.
virtual void do_code_blob(CodeBlob* cb) = 0;
};
// Applies an oop closure to all ref fields in code blobs
// iterated over in an object iteration.
class CodeBlobToOopClosure : public CodeBlobClosure {
OopClosure* _cl;
bool _fix_relocations;
protected:
void do_nmethod(nmethod* nm);
public:
// If fix_relocations(), then cl must copy objects to their new location immediately to avoid
// patching nmethods with the old locations.
CodeBlobToOopClosure(OopClosure* cl, bool fix_relocations) : _cl(cl), _fix_relocations(fix_relocations) {}
virtual void do_code_blob(CodeBlob* cb);
bool fix_relocations() const { return _fix_relocations; }
const static bool FixRelocations = true;
};
class MarkingCodeBlobClosure : public CodeBlobToOopClosure {
public:
MarkingCodeBlobClosure(OopClosure* cl, bool fix_relocations) : CodeBlobToOopClosure(cl, fix_relocations) {}
// Called for each code blob, but at most once per unique blob.
virtual void do_code_blob(CodeBlob* cb);
};
// MonitorClosure is used for iterating over monitors in the monitors cache
class ObjectMonitor;
class MonitorClosure : public StackObj {
public:
// called for each monitor in cache
virtual void do_monitor(ObjectMonitor* m) = 0;
};
// A closure that is applied without any arguments.
class VoidClosure : public StackObj {
public:
// I would have liked to declare this a pure virtual, but that breaks
// in mysterious ways, for unknown reasons.
virtual void do_void();
};
// YieldClosure is intended for use by iteration loops
// to incrementalize their work, allowing interleaving
// of an interruptable task so as to allow other
// threads to run (which may not otherwise be able to access
// exclusive resources, for instance). Additionally, the
// closure also allows for aborting an ongoing iteration
// by means of checking the return value from the polling
// call.
class YieldClosure : public StackObj {
public:
virtual bool should_return() = 0;
// Yield on a fine-grain level. The check in case of not yielding should be very fast.
virtual bool should_return_fine_grain() { return false; }
};
// Abstract closure for serializing data (read or write).
class SerializeClosure : public Closure {
public:
// Return bool indicating whether closure implements read or write.
virtual bool reading() const = 0;
// Read/write the void pointer pointed to by p.
virtual void do_ptr(void** p) = 0;
// Read/write the 32-bit unsigned integer pointed to by p.
virtual void do_u4(u4* p) = 0;
// Read/write the region specified.
virtual void do_region(u_char* start, size_t size) = 0;
// Check/write the tag. If reading, then compare the tag against
// the passed in value and fail is they don't match. This allows
// for verification that sections of the serialized data are of the
// correct length.
virtual void do_tag(int tag) = 0;
// Read/write the oop
virtual void do_oop(oop* o) = 0;
bool writing() {
return !reading();
}
};
class SymbolClosure : public StackObj {
public:
virtual void do_symbol(Symbol**) = 0;
// Clear LSB in symbol address; it can be set by CPSlot.
static Symbol* load_symbol(Symbol** p) {
return (Symbol*)(intptr_t(*p) & ~1);
}
// Store symbol, adjusting new pointer if the original pointer was adjusted
// (symbol references in constant pool slots have their LSB set to 1).
static void store_symbol(Symbol** p, Symbol* sym) {
*p = (Symbol*)(intptr_t(sym) | (intptr_t(*p) & 1));
}
};
// The two class template specializations are used to dispatch calls
// to the ExtendedOopClosure functions. If use_non_virtual_call is true,
// the non-virtual versions are called (E.g. do_oop_nv), otherwise the
// virtual versions are called (E.g. do_oop).
template <bool use_non_virtual_call>
class Devirtualizer {};
// Dispatches to the non-virtual functions.
template <> class Devirtualizer<true> {
public:
template <class OopClosureType, typename T> static void do_oop(OopClosureType* closure, T* p);
template <class OopClosureType> static void do_klass(OopClosureType* closure, Klass* k);
template <class OopClosureType> static void do_cld(OopClosureType* closure, ClassLoaderData* cld);
template <class OopClosureType> static bool do_metadata(OopClosureType* closure);
};
// Dispatches to the virtual functions.
template <> class Devirtualizer<false> {
public:
template <class OopClosureType, typename T> static void do_oop(OopClosureType* closure, T* p);
template <class OopClosureType> static void do_klass(OopClosureType* closure, Klass* k);
template <class OopClosureType> static void do_cld(OopClosureType* closure, ClassLoaderData* cld);
template <class OopClosureType> static bool do_metadata(OopClosureType* closure);
};
#endif // SHARE_VM_MEMORY_ITERATOR_HPP