Mercurial > hg > graal-compiler
view src/share/vm/oops/cpCacheOop.hpp @ 2149:7e37af9d69ef
7011379: G1: overly long concurrent marking cycles
Summary: This changeset introduces filtering of SATB buffers at the point when they are about to be enqueued. If this filtering clears enough entries on each buffer, the buffer can then be re-used and not enqueued. This cuts down the number of SATB buffers that need to be processed by the concurrent marking threads.
Reviewed-by: johnc, ysr
author | tonyp |
---|---|
date | Wed, 19 Jan 2011 09:35:17 -0500 |
parents | f95d63e2154a |
children | c5a923563727 |
line wrap: on
line source
/* * Copyright (c) 1998, 2010, 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_OOPS_CPCACHEOOP_HPP #define SHARE_VM_OOPS_CPCACHEOOP_HPP #include "interpreter/bytecodes.hpp" #include "memory/allocation.hpp" #include "oops/arrayOop.hpp" #include "utilities/array.hpp" // A ConstantPoolCacheEntry describes an individual entry of the constant // pool cache. There's 2 principal kinds of entries: field entries for in- // stance & static field access, and method entries for invokes. Some of // the entry layout is shared and looks as follows: // // bit number |31 0| // bit length |-8--|-8--|---16----| // -------------------------------- // _indices [ b2 | b1 | index ] // _f1 [ entry specific ] // _f2 [ entry specific ] // _flags [t|f|vf|v|m|h|unused|field_index] (for field entries) // bit length |4|1|1 |1|1|0|---7--|----16-----] // _flags [t|f|vf|v|m|h|unused|eidx|psze] (for method entries) // bit length |4|1|1 |1|1|1|---7--|-8--|-8--] // -------------------------------- // // with: // index = original constant pool index // b1 = bytecode 1 // b2 = bytecode 2 // psze = parameters size (method entries only) // eidx = interpreter entry index (method entries only) // field_index = index into field information in holder instanceKlass // The index max is 0xffff (max number of fields in constant pool) // and is multiplied by (instanceKlass::next_offset) when accessing. // t = TosState (see below) // f = field is marked final (see below) // vf = virtual, final (method entries only : is_vfinal()) // v = field is volatile (see below) // m = invokeinterface used for method in class Object (see below) // h = RedefineClasses/Hotswap bit (see below) // // The flags after TosState have the following interpretation: // bit 27: f flag true if field is marked final // bit 26: vf flag true if virtual final method // bit 25: v flag true if field is volatile (only for fields) // bit 24: m flag true if invokeinterface used for method in class Object // bit 23: 0 for fields, 1 for methods // // The flags 31, 30, 29, 28 together build a 4 bit number 0 to 8 with the // following mapping to the TosState states: // // btos: 0 // ctos: 1 // stos: 2 // itos: 3 // ltos: 4 // ftos: 5 // dtos: 6 // atos: 7 // vtos: 8 // // Entry specific: field entries: // _indices = get (b1 section) and put (b2 section) bytecodes, original constant pool index // _f1 = field holder // _f2 = field offset in words // _flags = field type information, original field index in field holder // (field_index section) // // Entry specific: method entries: // _indices = invoke code for f1 (b1 section), invoke code for f2 (b2 section), // original constant pool index // _f1 = method for all but virtual calls, unused by virtual calls // (note: for interface calls, which are essentially virtual, // contains klassOop for the corresponding interface. // for invokedynamic, f1 contains the CallSite object for the invocation // _f2 = method/vtable index for virtual calls only, unused by all other // calls. The vf flag indicates this is a method pointer not an // index. // _flags = field type info (f section), // virtual final entry (vf), // interpreter entry index (eidx section), // parameter size (psze section) // // Note: invokevirtual & invokespecial bytecodes can share the same constant // pool entry and thus the same constant pool cache entry. All invoke // bytecodes but invokevirtual use only _f1 and the corresponding b1 // bytecode, while invokevirtual uses only _f2 and the corresponding // b2 bytecode. The value of _flags is shared for both types of entries. // // The fields are volatile so that they are stored in the order written in the // source code. The _indices field with the bytecode must be written last. class ConstantPoolCacheEntry VALUE_OBJ_CLASS_SPEC { friend class VMStructs; friend class constantPoolCacheKlass; friend class constantPoolOopDesc; //resolve_constant_at_impl => set_f1 private: volatile intx _indices; // constant pool index & rewrite bytecodes volatile oop _f1; // entry specific oop field volatile intx _f2; // entry specific int/oop field volatile intx _flags; // flags #ifdef ASSERT bool same_methodOop(oop cur_f1, oop f1); #endif void set_bytecode_1(Bytecodes::Code code); void set_bytecode_2(Bytecodes::Code code); void set_f1(oop f1) { oop existing_f1 = _f1; // read once assert(existing_f1 == NULL || existing_f1 == f1, "illegal field change"); oop_store(&_f1, f1); } void set_f1_if_null_atomic(oop f1); void set_f2(intx f2) { assert(_f2 == 0 || _f2 == f2, "illegal field change"); _f2 = f2; } int as_flags(TosState state, bool is_final, bool is_vfinal, bool is_volatile, bool is_method_interface, bool is_method); void set_flags(intx flags) { _flags = flags; } public: // specific bit values in flag field // Note: the interpreter knows this layout! enum FlagBitValues { hotSwapBit = 23, methodInterface = 24, volatileField = 25, vfinalMethod = 26, finalField = 27 }; enum { field_index_mask = 0xFFFF }; // start of type bits in flags // Note: the interpreter knows this layout! enum FlagValues { tosBits = 28 }; // Initialization void initialize_entry(int original_index); // initialize primary entry void initialize_secondary_entry(int main_index); // initialize secondary entry void set_field( // sets entry to resolved field state Bytecodes::Code get_code, // the bytecode used for reading the field Bytecodes::Code put_code, // the bytecode used for writing the field KlassHandle field_holder, // the object/klass holding the field int orig_field_index, // the original field index in the field holder int field_offset, // the field offset in words in the field holder TosState field_type, // the (machine) field type bool is_final, // the field is final bool is_volatile // the field is volatile ); void set_method( // sets entry to resolved method entry Bytecodes::Code invoke_code, // the bytecode used for invoking the method methodHandle method, // the method/prototype if any (NULL, otherwise) int vtable_index // the vtable index if any, else negative ); void set_interface_call( methodHandle method, // Resolved method int index // Method index into interface ); void set_dynamic_call( Handle call_site, // Resolved java.dyn.CallSite (f1) methodHandle signature_invoker // determines signature information ); // For JVM_CONSTANT_InvokeDynamic cache entries: void initialize_bootstrap_method_index_in_cache(int bsm_cache_index); int bootstrap_method_index_in_cache(); void set_parameter_size(int value) { assert(parameter_size() == 0 || parameter_size() == value, "size must not change"); // Setting the parameter size by itself is only safe if the // current value of _flags is 0, otherwise another thread may have // updated it and we don't want to overwrite that value. Don't // bother trying to update it once it's nonzero but always make // sure that the final parameter size agrees with what was passed. if (_flags == 0) { Atomic::cmpxchg_ptr((value & 0xFF), &_flags, 0); } guarantee(parameter_size() == value, "size must not change"); } // Which bytecode number (1 or 2) in the index field is valid for this bytecode? // Returns -1 if neither is valid. static int bytecode_number(Bytecodes::Code code) { switch (code) { case Bytecodes::_getstatic : // fall through case Bytecodes::_getfield : // fall through case Bytecodes::_invokespecial : // fall through case Bytecodes::_invokestatic : // fall through case Bytecodes::_invokedynamic : // fall through case Bytecodes::_invokeinterface : return 1; case Bytecodes::_putstatic : // fall through case Bytecodes::_putfield : // fall through case Bytecodes::_invokevirtual : return 2; default : break; } return -1; } // Has this bytecode been resolved? Only valid for invokes and get/put field/static. bool is_resolved(Bytecodes::Code code) const { switch (bytecode_number(code)) { case 1: return (bytecode_1() == code); case 2: return (bytecode_2() == code); } return false; // default: not resolved } // Accessors bool is_secondary_entry() const { return (_indices & 0xFFFF) == 0; } int constant_pool_index() const { assert((_indices & 0xFFFF) != 0, "must be main entry"); return (_indices & 0xFFFF); } int main_entry_index() const { assert((_indices & 0xFFFF) == 0, "must be secondary entry"); return ((uintx)_indices >> 16); } Bytecodes::Code bytecode_1() const { return Bytecodes::cast((_indices >> 16) & 0xFF); } Bytecodes::Code bytecode_2() const { return Bytecodes::cast((_indices >> 24) & 0xFF); } volatile oop f1() const { return _f1; } bool is_f1_null() const { return (oop)_f1 == NULL; } // classifies a CPC entry as unbound intx f2() const { return _f2; } int field_index() const; int parameter_size() const { return _flags & 0xFF; } bool is_vfinal() const { return ((_flags & (1 << vfinalMethod)) == (1 << vfinalMethod)); } bool is_volatile() const { return ((_flags & (1 << volatileField)) == (1 << volatileField)); } bool is_methodInterface() const { return ((_flags & (1 << methodInterface)) == (1 << methodInterface)); } bool is_byte() const { return (((uintx) _flags >> tosBits) == btos); } bool is_char() const { return (((uintx) _flags >> tosBits) == ctos); } bool is_short() const { return (((uintx) _flags >> tosBits) == stos); } bool is_int() const { return (((uintx) _flags >> tosBits) == itos); } bool is_long() const { return (((uintx) _flags >> tosBits) == ltos); } bool is_float() const { return (((uintx) _flags >> tosBits) == ftos); } bool is_double() const { return (((uintx) _flags >> tosBits) == dtos); } bool is_object() const { return (((uintx) _flags >> tosBits) == atos); } TosState flag_state() const { assert( ( (_flags >> tosBits) & 0x0F ) < number_of_states, "Invalid state in as_flags"); return (TosState)((_flags >> tosBits) & 0x0F); } // Code generation support static WordSize size() { return in_WordSize(sizeof(ConstantPoolCacheEntry) / HeapWordSize); } static ByteSize size_in_bytes() { return in_ByteSize(sizeof(ConstantPoolCacheEntry)); } static ByteSize indices_offset() { return byte_offset_of(ConstantPoolCacheEntry, _indices); } static ByteSize f1_offset() { return byte_offset_of(ConstantPoolCacheEntry, _f1); } static ByteSize f2_offset() { return byte_offset_of(ConstantPoolCacheEntry, _f2); } static ByteSize flags_offset() { return byte_offset_of(ConstantPoolCacheEntry, _flags); } // GC Support void oops_do(void f(oop*)); void oop_iterate(OopClosure* blk); void oop_iterate_m(OopClosure* blk, MemRegion mr); void follow_contents(); void adjust_pointers(); #ifndef SERIALGC // Parallel Old void follow_contents(ParCompactionManager* cm); #endif // SERIALGC void update_pointers(); void update_pointers(HeapWord* beg_addr, HeapWord* end_addr); // RedefineClasses() API support: // If this constantPoolCacheEntry refers to old_method then update it // to refer to new_method. // trace_name_printed is set to true if the current call has // printed the klass name so that other routines in the adjust_* // group don't print the klass name. bool adjust_method_entry(methodOop old_method, methodOop new_method, bool * trace_name_printed); bool is_interesting_method_entry(klassOop k); bool is_field_entry() const { return (_flags & (1 << hotSwapBit)) == 0; } bool is_method_entry() const { return (_flags & (1 << hotSwapBit)) != 0; } // Debugging & Printing void print (outputStream* st, int index) const; void verify(outputStream* st) const; static void verify_tosBits() { assert(tosBits == 28, "interpreter now assumes tosBits is 28"); } }; // A constant pool cache is a runtime data structure set aside to a constant pool. The cache // holds interpreter runtime information for all field access and invoke bytecodes. The cache // is created and initialized before a class is actively used (i.e., initialized), the indivi- // dual cache entries are filled at resolution (i.e., "link") time (see also: rewriter.*). class constantPoolCacheOopDesc: public oopDesc { friend class VMStructs; private: int _length; constantPoolOop _constant_pool; // the corresponding constant pool // If true, safe for concurrent GC processing, // Set unconditionally in constantPoolCacheKlass::allocate() volatile bool _is_conc_safe; // Sizing debug_only(friend class ClassVerifier;) public: int length() const { return _length; } private: void set_length(int length) { _length = length; } static int header_size() { return sizeof(constantPoolCacheOopDesc) / HeapWordSize; } static int object_size(int length) { return align_object_size(header_size() + length * in_words(ConstantPoolCacheEntry::size())); } int object_size() { return object_size(length()); } // Helpers constantPoolOop* constant_pool_addr() { return &_constant_pool; } ConstantPoolCacheEntry* base() const { return (ConstantPoolCacheEntry*)((address)this + in_bytes(base_offset())); } friend class constantPoolCacheKlass; friend class ConstantPoolCacheEntry; public: // Initialization void initialize(intArray& inverse_index_map); // Secondary indexes. // They must look completely different from normal indexes. // The main reason is that byte swapping is sometimes done on normal indexes. // Also, some of the CP accessors do different things for secondary indexes. // Finally, it is helpful for debugging to tell the two apart. static bool is_secondary_index(int i) { return (i < 0); } static int decode_secondary_index(int i) { assert(is_secondary_index(i), ""); return ~i; } static int encode_secondary_index(int i) { assert(!is_secondary_index(i), ""); return ~i; } // Accessors void set_constant_pool(constantPoolOop pool) { oop_store_without_check((oop*)&_constant_pool, (oop)pool); } constantPoolOop constant_pool() const { return _constant_pool; } // Fetches the entry at the given index. // The entry may be either primary or secondary. // In either case the index must not be encoded or byte-swapped in any way. ConstantPoolCacheEntry* entry_at(int i) const { assert(0 <= i && i < length(), "index out of bounds"); return base() + i; } // Fetches the secondary entry referred to by index. // The index may be a secondary index, and must not be byte-swapped. ConstantPoolCacheEntry* secondary_entry_at(int i) const { int raw_index = i; if (is_secondary_index(i)) { // correct these on the fly raw_index = decode_secondary_index(i); } assert(entry_at(raw_index)->is_secondary_entry(), "not a secondary entry"); return entry_at(raw_index); } // Given a primary or secondary index, fetch the corresponding primary entry. // Indirect through the secondary entry, if the index is encoded as a secondary index. // The index must not be byte-swapped. ConstantPoolCacheEntry* main_entry_at(int i) const { int primary_index = i; if (is_secondary_index(i)) { // run through an extra level of indirection: int raw_index = decode_secondary_index(i); primary_index = entry_at(raw_index)->main_entry_index(); } assert(!entry_at(primary_index)->is_secondary_entry(), "only one level of indirection"); return entry_at(primary_index); } // GC support // If the _length field has not been set, the size of the // constantPoolCache cannot be correctly calculated. bool is_conc_safe() { return _is_conc_safe; } void set_is_conc_safe(bool v) { _is_conc_safe = v; } // Code generation static ByteSize base_offset() { return in_ByteSize(sizeof(constantPoolCacheOopDesc)); } static ByteSize entry_offset(int raw_index) { int index = raw_index; if (is_secondary_index(raw_index)) index = decode_secondary_index(raw_index); return (base_offset() + ConstantPoolCacheEntry::size_in_bytes() * index); } // RedefineClasses() API support: // If any entry of this constantPoolCache points to any of // old_methods, replace it with the corresponding new_method. // trace_name_printed is set to true if the current call has // printed the klass name so that other routines in the adjust_* // group don't print the klass name. void adjust_method_entries(methodOop* old_methods, methodOop* new_methods, int methods_length, bool * trace_name_printed); }; #endif // SHARE_VM_OOPS_CPCACHEOOP_HPP