OpenJDK / jdk / jdk10
changeset 25957:100a882dcffa
Merge
| author | dcubed |
|---|---|
| date | Wed, 13 Aug 2014 13:29:45 -0700 |
| parents | c8aa7a0bf7d0 99be217ac88d |
| children | 729cd80956ae |
| files | hotspot/src/share/vm/memory/collectorPolicy.cpp hotspot/src/share/vm/services/memPtr.cpp hotspot/src/share/vm/services/memPtr.hpp hotspot/src/share/vm/services/memPtrArray.hpp hotspot/src/share/vm/services/memRecorder.cpp hotspot/src/share/vm/services/memRecorder.hpp hotspot/src/share/vm/services/memSnapshot.cpp hotspot/src/share/vm/services/memSnapshot.hpp hotspot/src/share/vm/services/memTrackWorker.cpp hotspot/src/share/vm/services/memTrackWorker.hpp |
| diffstat | 152 files changed, 6748 insertions(+), 7959 deletions(-) [+] |
line wrap: on
line diff
--- a/hotspot/agent/src/os/solaris/proc/saproc.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/agent/src/os/solaris/proc/saproc.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -314,7 +314,7 @@ handle = dlopen(name, mode); } if (_libsaproc_debug) { - printf("libsaproc DEBUG: pathmap_dlopen %s return 0x%x\n", name, handle); + printf("libsaproc DEBUG: pathmap_dlopen %s return 0x%lx\n", name, (unsigned long) handle); } return handle; }
--- a/hotspot/make/excludeSrc.make Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/make/excludeSrc.make Wed Aug 13 13:29:45 2014 -0700 @@ -119,8 +119,8 @@ CFLAGS += -DINCLUDE_NMT=0 Src_Files_EXCLUDE += \ - memBaseline.cpp memPtr.cpp memRecorder.cpp memReporter.cpp memSnapshot.cpp memTrackWorker.cpp \ - memTracker.cpp nmtDCmd.cpp + memBaseline.cpp memReporter.cpp mallocTracker.cpp virtualMemoryTracker.cpp nmtCommon.cpp \ + memTracker.cpp nmtDCmd.cpp mallocSiteTable.cpp endif -include $(HS_ALT_MAKE)/excludeSrc.make
--- a/hotspot/src/cpu/ppc/vm/cppInterpreterGenerator_ppc.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/ppc/vm/cppInterpreterGenerator_ppc.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2002, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2002, 2014, Oracle and/or its affiliates. All rights reserved. * Copyright 2012, 2013 SAP AG. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * @@ -26,8 +26,9 @@ #ifndef CPU_PPC_VM_CPPINTERPRETERGENERATOR_PPC_HPP #define CPU_PPC_VM_CPPINTERPRETERGENERATOR_PPC_HPP - address generate_normal_entry(void); - address generate_native_entry(void); + address generate_normal_entry(bool synchronized); + address generate_native_entry(bool synchronized); + address generate_math_entry(AbstractInterpreter::MethodKind kind) { return NULL; } void lock_method(void); void unlock_method(void);
--- a/hotspot/src/cpu/ppc/vm/cppInterpreter_ppc.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/ppc/vm/cppInterpreter_ppc.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -938,8 +938,9 @@ // Interpreter stub for calling a native method. (C++ interpreter) // This sets up a somewhat different looking stack for calling the native method // than the typical interpreter frame setup. +// The synchronized parameter is ignored. // -address CppInterpreterGenerator::generate_native_entry(void) { +address CppInterpreterGenerator::generate_native_entry(bool synchronized) { if (native_entry != NULL) return native_entry; address entry = __ pc(); @@ -1729,7 +1730,8 @@ __ std(R0, BasicObjectLock::obj_offset_in_bytes(), stack_base); // Mark lock as unused } -address CppInterpreterGenerator::generate_normal_entry(void) { +// The synchronized parameter is ignored +address CppInterpreterGenerator::generate_normal_entry(bool synchronized) { if (interpreter_frame_manager != NULL) return interpreter_frame_manager; address entry = __ pc(); @@ -2789,38 +2791,6 @@ return interpreter_frame_manager; } -// Generate code for various sorts of method entries -// -address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) { - address entry_point = NULL; - - switch (kind) { - case Interpreter::zerolocals : break; - case Interpreter::zerolocals_synchronized : break; - case Interpreter::native : // Fall thru - case Interpreter::native_synchronized : entry_point = ((CppInterpreterGenerator*)this)->generate_native_entry(); break; - case Interpreter::empty : break; - case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break; - case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break; - // These are special interpreter intrinsics which we don't support so far. - case Interpreter::java_lang_math_sin : break; - case Interpreter::java_lang_math_cos : break; - case Interpreter::java_lang_math_tan : break; - case Interpreter::java_lang_math_abs : break; - case Interpreter::java_lang_math_log : break; - case Interpreter::java_lang_math_log10 : break; - case Interpreter::java_lang_math_sqrt : break; - case Interpreter::java_lang_math_pow : break; - case Interpreter::java_lang_math_exp : break; - case Interpreter::java_lang_ref_reference_get: entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break; - default : ShouldNotReachHere(); break; - } - - if (entry_point) { - return entry_point; - } - return ((InterpreterGenerator*)this)->generate_normal_entry(); -} InterpreterGenerator::InterpreterGenerator(StubQueue* code) : CppInterpreterGenerator(code) {
--- a/hotspot/src/cpu/ppc/vm/interpreterGenerator_ppc.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/ppc/vm/interpreterGenerator_ppc.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2002, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2002, 2014, Oracle and/or its affiliates. All rights reserved. * Copyright 2012, 2013 SAP AG. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * @@ -31,7 +31,12 @@ private: address generate_abstract_entry(void); - address generate_accessor_entry(void); + address generate_jump_to_normal_entry(void); + address generate_accessor_entry(void) { return generate_jump_to_normal_entry(); } + address generate_empty_entry(void) { return generate_jump_to_normal_entry(); } address generate_Reference_get_entry(void); + // Not supported + address generate_CRC32_update_entry() { return NULL; } + address generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { return NULL; } #endif // CPU_PPC_VM_INTERPRETERGENERATOR_PPC_HPP
--- a/hotspot/src/cpu/ppc/vm/interpreter_ppc.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/ppc/vm/interpreter_ppc.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -428,6 +428,19 @@ return entry; } + +// Call an accessor method (assuming it is resolved, otherwise drop into +// vanilla (slow path) entry. +address InterpreterGenerator::generate_jump_to_normal_entry(void) { + address entry = __ pc(); + address normal_entry = Interpreter::entry_for_kind(Interpreter::zerolocals); + assert(normal_entry != NULL, "should already be generated."); + __ branch_to_entry(normal_entry, R11_scratch1); + __ flush(); + + return entry; +} + // Abstract method entry. // address InterpreterGenerator::generate_abstract_entry(void) { @@ -485,203 +498,6 @@ return entry; } -// Call an accessor method (assuming it is resolved, otherwise drop into -// vanilla (slow path) entry. -address InterpreterGenerator::generate_accessor_entry(void) { - if (!UseFastAccessorMethods && (!FLAG_IS_ERGO(UseFastAccessorMethods))) { - return NULL; - } - - Label Lslow_path, Lacquire; - - const Register - Rclass_or_obj = R3_ARG1, - Rconst_method = R4_ARG2, - Rcodes = Rconst_method, - Rcpool_cache = R5_ARG3, - Rscratch = R11_scratch1, - Rjvmti_mode = Rscratch, - Roffset = R12_scratch2, - Rflags = R6_ARG4, - Rbtable = R7_ARG5; - - static address branch_table[number_of_states]; - - address entry = __ pc(); - - // Check for safepoint: - // Ditch this, real man don't need safepoint checks. - - // Also check for JVMTI mode - // Check for null obj, take slow path if so. - __ ld(Rclass_or_obj, Interpreter::stackElementSize, CC_INTERP_ONLY(R17_tos) NOT_CC_INTERP(R15_esp)); - __ lwz(Rjvmti_mode, thread_(interp_only_mode)); - __ cmpdi(CCR1, Rclass_or_obj, 0); - __ cmpwi(CCR0, Rjvmti_mode, 0); - __ crorc(/*CCR0 eq*/2, /*CCR1 eq*/4+2, /*CCR0 eq*/2); - __ beq(CCR0, Lslow_path); // this==null or jvmti_mode!=0 - - // Do 2 things in parallel: - // 1. Load the index out of the first instruction word, which looks like this: - // <0x2a><0xb4><index (2 byte, native endianess)>. - // 2. Load constant pool cache base. - __ ld(Rconst_method, in_bytes(Method::const_offset()), R19_method); - __ ld(Rcpool_cache, in_bytes(ConstMethod::constants_offset()), Rconst_method); - - __ lhz(Rcodes, in_bytes(ConstMethod::codes_offset()) + 2, Rconst_method); // Lower half of 32 bit field. - __ ld(Rcpool_cache, ConstantPool::cache_offset_in_bytes(), Rcpool_cache); - - // Get the const pool entry by means of <index>. - const int codes_shift = exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord); - __ slwi(Rscratch, Rcodes, codes_shift); // (codes&0xFFFF)<<codes_shift - __ add(Rcpool_cache, Rscratch, Rcpool_cache); - - // Check if cpool cache entry is resolved. - // We are resolved if the indices offset contains the current bytecode. - ByteSize cp_base_offset = ConstantPoolCache::base_offset(); - // Big Endian: - __ lbz(Rscratch, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::indices_offset()) + 7 - 2, Rcpool_cache); - __ cmpwi(CCR0, Rscratch, Bytecodes::_getfield); - __ bne(CCR0, Lslow_path); - __ isync(); // Order succeeding loads wrt. load of _indices field from cpool_cache. - - // Finally, start loading the value: Get cp cache entry into regs. - __ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcpool_cache); - __ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcpool_cache); - - // Following code is from templateTable::getfield_or_static - // Load pointer to branch table - __ load_const_optimized(Rbtable, (address)branch_table, Rscratch); - - // Get volatile flag - __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // extract volatile bit - // note: sync is needed before volatile load on PPC64 - - // Check field type - __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits); - -#ifdef ASSERT - Label LFlagInvalid; - __ cmpldi(CCR0, Rflags, number_of_states); - __ bge(CCR0, LFlagInvalid); - - __ ld(R9_ARG7, 0, R1_SP); - __ ld(R10_ARG8, 0, R21_sender_SP); - __ cmpd(CCR0, R9_ARG7, R10_ARG8); - __ asm_assert_eq("backlink", 0x543); -#endif // ASSERT - __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. - - // Load from branch table and dispatch (volatile case: one instruction ahead) - __ sldi(Rflags, Rflags, LogBytesPerWord); - __ cmpwi(CCR6, Rscratch, 1); // volatile? - if (support_IRIW_for_not_multiple_copy_atomic_cpu) { - __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // volatile ? size of 1 instruction : 0 - } - __ ldx(Rbtable, Rbtable, Rflags); - - if (support_IRIW_for_not_multiple_copy_atomic_cpu) { - __ subf(Rbtable, Rscratch, Rbtable); // point to volatile/non-volatile entry point - } - __ mtctr(Rbtable); - __ bctr(); - -#ifdef ASSERT - __ bind(LFlagInvalid); - __ stop("got invalid flag", 0x6541); - - bool all_uninitialized = true, - all_initialized = true; - for (int i = 0; i<number_of_states; ++i) { - all_uninitialized = all_uninitialized && (branch_table[i] == NULL); - all_initialized = all_initialized && (branch_table[i] != NULL); - } - assert(all_uninitialized != all_initialized, "consistency"); // either or - - __ fence(); // volatile entry point (one instruction before non-volatile_entry point) - if (branch_table[vtos] == 0) branch_table[vtos] = __ pc(); // non-volatile_entry point - if (branch_table[dtos] == 0) branch_table[dtos] = __ pc(); // non-volatile_entry point - if (branch_table[ftos] == 0) branch_table[ftos] = __ pc(); // non-volatile_entry point - __ stop("unexpected type", 0x6551); -#endif - - if (branch_table[itos] == 0) { // generate only once - __ align(32, 28, 28); // align load - __ fence(); // volatile entry point (one instruction before non-volatile_entry point) - branch_table[itos] = __ pc(); // non-volatile_entry point - __ lwax(R3_RET, Rclass_or_obj, Roffset); - __ beq(CCR6, Lacquire); - __ blr(); - } - - if (branch_table[ltos] == 0) { // generate only once - __ align(32, 28, 28); // align load - __ fence(); // volatile entry point (one instruction before non-volatile_entry point) - branch_table[ltos] = __ pc(); // non-volatile_entry point - __ ldx(R3_RET, Rclass_or_obj, Roffset); - __ beq(CCR6, Lacquire); - __ blr(); - } - - if (branch_table[btos] == 0) { // generate only once - __ align(32, 28, 28); // align load - __ fence(); // volatile entry point (one instruction before non-volatile_entry point) - branch_table[btos] = __ pc(); // non-volatile_entry point - __ lbzx(R3_RET, Rclass_or_obj, Roffset); - __ extsb(R3_RET, R3_RET); - __ beq(CCR6, Lacquire); - __ blr(); - } - - if (branch_table[ctos] == 0) { // generate only once - __ align(32, 28, 28); // align load - __ fence(); // volatile entry point (one instruction before non-volatile_entry point) - branch_table[ctos] = __ pc(); // non-volatile_entry point - __ lhzx(R3_RET, Rclass_or_obj, Roffset); - __ beq(CCR6, Lacquire); - __ blr(); - } - - if (branch_table[stos] == 0) { // generate only once - __ align(32, 28, 28); // align load - __ fence(); // volatile entry point (one instruction before non-volatile_entry point) - branch_table[stos] = __ pc(); // non-volatile_entry point - __ lhax(R3_RET, Rclass_or_obj, Roffset); - __ beq(CCR6, Lacquire); - __ blr(); - } - - if (branch_table[atos] == 0) { // generate only once - __ align(32, 28, 28); // align load - __ fence(); // volatile entry point (one instruction before non-volatile_entry point) - branch_table[atos] = __ pc(); // non-volatile_entry point - __ load_heap_oop(R3_RET, (RegisterOrConstant)Roffset, Rclass_or_obj); - __ verify_oop(R3_RET); - //__ dcbt(R3_RET); // prefetch - __ beq(CCR6, Lacquire); - __ blr(); - } - - __ align(32, 12); - __ bind(Lacquire); - __ twi_0(R3_RET); - __ isync(); // acquire - __ blr(); - -#ifdef ASSERT - for (int i = 0; i<number_of_states; ++i) { - assert(branch_table[i], "accessor_entry initialization"); - //tty->print_cr("accessor_entry: branch_table[%d] = 0x%llx (opcode 0x%llx)", i, branch_table[i], *((unsigned int*)branch_table[i])); - } -#endif - - __ bind(Lslow_path); - __ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), Rscratch); - __ flush(); - - return entry; -} - // Interpreter intrinsic for WeakReference.get(). // 1. Don't push a full blown frame and go on dispatching, but fetch the value // into R8 and return quickly @@ -713,7 +529,6 @@ // and so we don't need to call the G1 pre-barrier. Thus we can use the // regular method entry code to generate the NPE. // - // This code is based on generate_accessor_enty. address entry = __ pc(); @@ -768,7 +583,7 @@ return entry; } else { - return generate_accessor_entry(); + return generate_jump_to_normal_entry(); } }
--- a/hotspot/src/cpu/ppc/vm/templateInterpreterGenerator_ppc.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/ppc/vm/templateInterpreterGenerator_ppc.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -30,7 +30,6 @@ address generate_normal_entry(bool synchronized); address generate_native_entry(bool synchronized); address generate_math_entry(AbstractInterpreter::MethodKind kind); - address generate_empty_entry(void); void lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded=false); void unlock_method(bool check_exceptions = true);
--- a/hotspot/src/cpu/ppc/vm/templateInterpreter_ppc.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/ppc/vm/templateInterpreter_ppc.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -602,48 +602,6 @@ // End of helpers -// ============================================================================ -// Various method entries -// - -// Empty method, generate a very fast return. We must skip this entry if -// someone's debugging, indicated by the flag -// "interp_mode" in the Thread obj. -// Note: empty methods are generated mostly methods that do assertions, which are -// disabled in the "java opt build". -address TemplateInterpreterGenerator::generate_empty_entry(void) { - if (!UseFastEmptyMethods) { - NOT_PRODUCT(__ should_not_reach_here();) - return Interpreter::entry_for_kind(Interpreter::zerolocals); - } - - Label Lslow_path; - const Register Rjvmti_mode = R11_scratch1; - address entry = __ pc(); - - __ lwz(Rjvmti_mode, thread_(interp_only_mode)); - __ cmpwi(CCR0, Rjvmti_mode, 0); - __ bne(CCR0, Lslow_path); // jvmti_mode!=0 - - // Noone's debuggin: Simply return. - // Pop c2i arguments (if any) off when we return. -#ifdef ASSERT - __ ld(R9_ARG7, 0, R1_SP); - __ ld(R10_ARG8, 0, R21_sender_SP); - __ cmpd(CCR0, R9_ARG7, R10_ARG8); - __ asm_assert_eq("backlink", 0x545); -#endif // ASSERT - __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. - - // And we're done. - __ blr(); - - __ bind(Lslow_path); - __ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1); - __ flush(); - - return entry; -} // Support abs and sqrt like in compiler. // For others we can use a normal (native) entry. @@ -1289,45 +1247,6 @@ return entry; } -// ============================================================================= -// Entry points - -address AbstractInterpreterGenerator::generate_method_entry( - AbstractInterpreter::MethodKind kind) { - // Determine code generation flags. - bool synchronized = false; - address entry_point = NULL; - - switch (kind) { - case Interpreter::zerolocals : break; - case Interpreter::zerolocals_synchronized: synchronized = true; break; - case Interpreter::native : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break; - case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true); break; - case Interpreter::empty : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry(); break; - case Interpreter::accessor : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry(); break; - case Interpreter::abstract : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry(); break; - - case Interpreter::java_lang_math_sin : // fall thru - case Interpreter::java_lang_math_cos : // fall thru - case Interpreter::java_lang_math_tan : // fall thru - case Interpreter::java_lang_math_abs : // fall thru - case Interpreter::java_lang_math_log : // fall thru - case Interpreter::java_lang_math_log10 : // fall thru - case Interpreter::java_lang_math_sqrt : // fall thru - case Interpreter::java_lang_math_pow : // fall thru - case Interpreter::java_lang_math_exp : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind); break; - case Interpreter::java_lang_ref_reference_get - : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break; - default : ShouldNotReachHere(); break; - } - - if (entry_point) { - return entry_point; - } - - return ((InterpreterGenerator*) this)->generate_normal_entry(synchronized); -} - // These should never be compiled since the interpreter will prefer // the compiled version to the intrinsic version. bool AbstractInterpreter::can_be_compiled(methodHandle m) { @@ -1355,7 +1274,7 @@ int callee_locals, bool is_top_frame) { // Note: This calculation must exactly parallel the frame setup - // in AbstractInterpreterGenerator::generate_method_entry. + // in InterpreterGenerator::generate_fixed_frame. assert(Interpreter::stackElementWords == 1, "sanity"); const int max_alignment_space = StackAlignmentInBytes / Interpreter::stackElementSize; const int abi_scratch = is_top_frame ? (frame::abi_reg_args_size / Interpreter::stackElementSize) :
--- a/hotspot/src/cpu/ppc/vm/vm_version_ppc.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/ppc/vm/vm_version_ppc.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -29,6 +29,7 @@ #include "compiler/disassembler.hpp" #include "memory/resourceArea.hpp" #include "runtime/java.hpp" +#include "runtime/os.hpp" #include "runtime/stubCodeGenerator.hpp" #include "utilities/defaultStream.hpp" #include "vm_version_ppc.hpp" @@ -108,7 +109,7 @@ (has_vand() ? " vand" : "") // Make sure number of %s matches num_features! ); - _features_str = strdup(buf); + _features_str = os::strdup(buf); NOT_PRODUCT(if (Verbose) print_features();); // PPC64 supports 8-byte compare-exchange operations (see
--- a/hotspot/src/cpu/sparc/vm/cppInterpreter_sparc.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/sparc/vm/cppInterpreter_sparc.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2007, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2007, 2014, 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 @@ -29,6 +29,7 @@ #include "interpreter/interpreter.hpp" #include "interpreter/interpreterGenerator.hpp" #include "interpreter/interpreterRuntime.hpp" +#include "interpreter/interp_masm.hpp" #include "oops/arrayOop.hpp" #include "oops/methodData.hpp" #include "oops/method.hpp" @@ -68,9 +69,7 @@ #define STATE(field_name) Lstate, in_bytes(byte_offset_of(BytecodeInterpreter, field_name)) #define __ _masm-> -Label frame_manager_entry; -Label fast_accessor_slow_entry_path; // fast accessor methods need to be able to jmp to unsynchronized - // c++ interpreter entry point this holds that entry point label. +Label frame_manager_entry; // c++ interpreter entry point this holds that entry point label. static address unctrap_frame_manager_entry = NULL; @@ -452,110 +451,6 @@ return NULL; } -// Call an accessor method (assuming it is resolved, otherwise drop into -// vanilla (slow path) entry - -// Generates code to elide accessor methods -// Uses G3_scratch and G1_scratch as scratch -address InterpreterGenerator::generate_accessor_entry(void) { - - // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; - // parameter size = 1 - // Note: We can only use this code if the getfield has been resolved - // and if we don't have a null-pointer exception => check for - // these conditions first and use slow path if necessary. - address entry = __ pc(); - Label slow_path; - - if ( UseFastAccessorMethods) { - // Check if we need to reach a safepoint and generate full interpreter - // frame if so. - AddressLiteral sync_state(SafepointSynchronize::address_of_state()); - __ load_contents(sync_state, G3_scratch); - __ cmp(G3_scratch, SafepointSynchronize::_not_synchronized); - __ br(Assembler::notEqual, false, Assembler::pn, slow_path); - __ delayed()->nop(); - - // Check if local 0 != NULL - __ ld_ptr(Gargs, G0, Otos_i ); // get local 0 - __ tst(Otos_i); // check if local 0 == NULL and go the slow path - __ brx(Assembler::zero, false, Assembler::pn, slow_path); - __ delayed()->nop(); - - - // read first instruction word and extract bytecode @ 1 and index @ 2 - // get first 4 bytes of the bytecodes (big endian!) - __ ld_ptr(Address(G5_method, in_bytes(Method::const_offset())), G1_scratch); - __ ld(Address(G1_scratch, in_bytes(ConstMethod::codes_offset())), G1_scratch); - - // move index @ 2 far left then to the right most two bytes. - __ sll(G1_scratch, 2*BitsPerByte, G1_scratch); - __ srl(G1_scratch, 2*BitsPerByte - exact_log2(in_words( - ConstantPoolCacheEntry::size()) * BytesPerWord), G1_scratch); - - // get constant pool cache - __ ld_ptr(G5_method, in_bytes(Method::const_offset()), G3_scratch); - __ ld_ptr(G3_scratch, in_bytes(ConstMethod::constants_offset()), G3_scratch); - __ ld_ptr(G3_scratch, ConstantPool::cache_offset_in_bytes(), G3_scratch); - - // get specific constant pool cache entry - __ add(G3_scratch, G1_scratch, G3_scratch); - - // Check the constant Pool cache entry to see if it has been resolved. - // If not, need the slow path. - ByteSize cp_base_offset = ConstantPoolCache::base_offset(); - __ ld_ptr(G3_scratch, in_bytes(cp_base_offset + ConstantPoolCacheEntry::indices_offset()), G1_scratch); - __ srl(G1_scratch, 2*BitsPerByte, G1_scratch); - __ and3(G1_scratch, 0xFF, G1_scratch); - __ cmp(G1_scratch, Bytecodes::_getfield); - __ br(Assembler::notEqual, false, Assembler::pn, slow_path); - __ delayed()->nop(); - - // Get the type and return field offset from the constant pool cache - __ ld_ptr(G3_scratch, in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset()), G1_scratch); - __ ld_ptr(G3_scratch, in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset()), G3_scratch); - - Label xreturn_path; - // Need to differentiate between igetfield, agetfield, bgetfield etc. - // because they are different sizes. - // Get the type from the constant pool cache - __ srl(G1_scratch, ConstantPoolCacheEntry::tos_state_shift, G1_scratch); - // Make sure we don't need to mask G1_scratch after the above shift - ConstantPoolCacheEntry::verify_tos_state_shift(); - __ cmp(G1_scratch, atos ); - __ br(Assembler::equal, true, Assembler::pt, xreturn_path); - __ delayed()->ld_ptr(Otos_i, G3_scratch, Otos_i); - __ cmp(G1_scratch, itos); - __ br(Assembler::equal, true, Assembler::pt, xreturn_path); - __ delayed()->ld(Otos_i, G3_scratch, Otos_i); - __ cmp(G1_scratch, stos); - __ br(Assembler::equal, true, Assembler::pt, xreturn_path); - __ delayed()->ldsh(Otos_i, G3_scratch, Otos_i); - __ cmp(G1_scratch, ctos); - __ br(Assembler::equal, true, Assembler::pt, xreturn_path); - __ delayed()->lduh(Otos_i, G3_scratch, Otos_i); -#ifdef ASSERT - __ cmp(G1_scratch, btos); - __ br(Assembler::equal, true, Assembler::pt, xreturn_path); - __ delayed()->ldsb(Otos_i, G3_scratch, Otos_i); - __ should_not_reach_here(); -#endif - __ ldsb(Otos_i, G3_scratch, Otos_i); - __ bind(xreturn_path); - - // _ireturn/_areturn - __ retl(); // return from leaf routine - __ delayed()->mov(O5_savedSP, SP); - - // Generate regular method entry - __ bind(slow_path); - __ ba(fast_accessor_slow_entry_path); - __ delayed()->nop(); - return entry; - } - return NULL; -} - address InterpreterGenerator::generate_Reference_get_entry(void) { #if INCLUDE_ALL_GCS if (UseG1GC) { @@ -573,7 +468,7 @@ // If G1 is not enabled then attempt to go through the accessor entry point // Reference.get is an accessor - return generate_accessor_entry(); + return generate_jump_to_normal_entry(); } // @@ -1870,23 +1765,6 @@ __ ba(call_interpreter_2); __ delayed()->st_ptr(O1, STATE(_stack)); - - // Fast accessor methods share this entry point. - // This works because frame manager is in the same codelet - // This can either be an entry via call_stub/c1/c2 or a recursive interpreter call - // we need to do a little register fixup here once we distinguish the two of them - if (UseFastAccessorMethods && !synchronized) { - // Call stub_return address still in O7 - __ bind(fast_accessor_slow_entry_path); - __ set((intptr_t)return_from_native_method - 8, Gtmp1); - __ cmp(Gtmp1, O7); // returning to interpreter? - __ brx(Assembler::equal, true, Assembler::pt, re_dispatch); // yep - __ delayed()->nop(); - __ ba(re_dispatch); - __ delayed()->mov(G0, prevState); // initial entry - - } - // interpreter returning to native code (call_stub/c1/c2) // convert result and unwind initial activation // L2_scratch - scaled result type index
--- a/hotspot/src/cpu/sparc/vm/interpreterGenerator_sparc.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/sparc/vm/interpreterGenerator_sparc.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, 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 @@ -32,9 +32,11 @@ address generate_normal_entry(bool synchronized); address generate_native_entry(bool synchronized); address generate_abstract_entry(void); - address generate_math_entry(AbstractInterpreter::MethodKind kind); - address generate_empty_entry(void); - address generate_accessor_entry(void); + // there are no math intrinsics on sparc + address generate_math_entry(AbstractInterpreter::MethodKind kind) { return NULL; } + address generate_jump_to_normal_entry(void); + address generate_accessor_entry(void) { return generate_jump_to_normal_entry(); } + address generate_empty_entry(void) { return generate_jump_to_normal_entry(); } address generate_Reference_get_entry(void); void lock_method(void); void save_native_result(void); @@ -43,4 +45,7 @@ void generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue); void generate_counter_overflow(Label& Lcontinue); + // Not supported + address generate_CRC32_update_entry() { return NULL; } + address generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { return NULL; } #endif // CPU_SPARC_VM_INTERPRETERGENERATOR_SPARC_HPP
--- a/hotspot/src/cpu/sparc/vm/interpreter_sparc.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/sparc/vm/interpreter_sparc.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -241,6 +241,15 @@ // Various method entries +address InterpreterGenerator::generate_jump_to_normal_entry(void) { + address entry = __ pc(); + assert(Interpreter::entry_for_kind(Interpreter::zerolocals) != NULL, "should already be generated"); + AddressLiteral al(Interpreter::entry_for_kind(Interpreter::zerolocals)); + __ jump_to(al, G3_scratch); + __ delayed()->nop(); + return entry; +} + // Abstract method entry // Attempt to execute abstract method. Throw exception // @@ -255,159 +264,6 @@ } - -//---------------------------------------------------------------------------------------------------- -// Entry points & stack frame layout -// -// Here we generate the various kind of entries into the interpreter. -// The two main entry type are generic bytecode methods and native call method. -// These both come in synchronized and non-synchronized versions but the -// frame layout they create is very similar. The other method entry -// types are really just special purpose entries that are really entry -// and interpretation all in one. These are for trivial methods like -// accessor, empty, or special math methods. -// -// When control flow reaches any of the entry types for the interpreter -// the following holds -> -// -// C2 Calling Conventions: -// -// The entry code below assumes that the following registers are set -// when coming in: -// G5_method: holds the Method* of the method to call -// Lesp: points to the TOS of the callers expression stack -// after having pushed all the parameters -// -// The entry code does the following to setup an interpreter frame -// pop parameters from the callers stack by adjusting Lesp -// set O0 to Lesp -// compute X = (max_locals - num_parameters) -// bump SP up by X to accomadate the extra locals -// compute X = max_expression_stack -// + vm_local_words -// + 16 words of register save area -// save frame doing a save sp, -X, sp growing towards lower addresses -// set Lbcp, Lmethod, LcpoolCache -// set Llocals to i0 -// set Lmonitors to FP - rounded_vm_local_words -// set Lesp to Lmonitors - 4 -// -// The frame has now been setup to do the rest of the entry code - -// Try this optimization: Most method entries could live in a -// "one size fits all" stack frame without all the dynamic size -// calculations. It might be profitable to do all this calculation -// statically and approximately for "small enough" methods. - -//----------------------------------------------------------------------------------------------- - -// C1 Calling conventions -// -// Upon method entry, the following registers are setup: -// -// g2 G2_thread: current thread -// g5 G5_method: method to activate -// g4 Gargs : pointer to last argument -// -// -// Stack: -// -// +---------------+ <--- sp -// | | -// : reg save area : -// | | -// +---------------+ <--- sp + 0x40 -// | | -// : extra 7 slots : note: these slots are not really needed for the interpreter (fix later) -// | | -// +---------------+ <--- sp + 0x5c -// | | -// : free : -// | | -// +---------------+ <--- Gargs -// | | -// : arguments : -// | | -// +---------------+ -// | | -// -// -// -// AFTER FRAME HAS BEEN SETUP for method interpretation the stack looks like: -// -// +---------------+ <--- sp -// | | -// : reg save area : -// | | -// +---------------+ <--- sp + 0x40 -// | | -// : extra 7 slots : note: these slots are not really needed for the interpreter (fix later) -// | | -// +---------------+ <--- sp + 0x5c -// | | -// : : -// | | <--- Lesp -// +---------------+ <--- Lmonitors (fp - 0x18) -// | VM locals | -// +---------------+ <--- fp -// | | -// : reg save area : -// | | -// +---------------+ <--- fp + 0x40 -// | | -// : extra 7 slots : note: these slots are not really needed for the interpreter (fix later) -// | | -// +---------------+ <--- fp + 0x5c -// | | -// : free : -// | | -// +---------------+ -// | | -// : nonarg locals : -// | | -// +---------------+ -// | | -// : arguments : -// | | <--- Llocals -// +---------------+ <--- Gargs -// | | - -address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) { - // determine code generation flags - bool synchronized = false; - address entry_point = NULL; - - switch (kind) { - case Interpreter::zerolocals : break; - case Interpreter::zerolocals_synchronized: synchronized = true; break; - case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break; - case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break; - case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break; - case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break; - case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break; - - case Interpreter::java_lang_math_sin : break; - case Interpreter::java_lang_math_cos : break; - case Interpreter::java_lang_math_tan : break; - case Interpreter::java_lang_math_sqrt : break; - case Interpreter::java_lang_math_abs : break; - case Interpreter::java_lang_math_log : break; - case Interpreter::java_lang_math_log10 : break; - case Interpreter::java_lang_math_pow : break; - case Interpreter::java_lang_math_exp : break; - case Interpreter::java_lang_ref_reference_get - : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break; - default: - fatal(err_msg("unexpected method kind: %d", kind)); - break; - } - - if (entry_point) return entry_point; - - return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized); -} - - bool AbstractInterpreter::can_be_compiled(methodHandle m) { // No special entry points that preclude compilation return true;
--- a/hotspot/src/cpu/sparc/vm/templateInterpreter_sparc.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/sparc/vm/templateInterpreter_sparc.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -456,6 +456,115 @@ // Generate a fixed interpreter frame. This is identical setup for interpreted // methods and for native methods hence the shared code. + +//---------------------------------------------------------------------------------------------------- +// Stack frame layout +// +// When control flow reaches any of the entry types for the interpreter +// the following holds -> +// +// C2 Calling Conventions: +// +// The entry code below assumes that the following registers are set +// when coming in: +// G5_method: holds the Method* of the method to call +// Lesp: points to the TOS of the callers expression stack +// after having pushed all the parameters +// +// The entry code does the following to setup an interpreter frame +// pop parameters from the callers stack by adjusting Lesp +// set O0 to Lesp +// compute X = (max_locals - num_parameters) +// bump SP up by X to accomadate the extra locals +// compute X = max_expression_stack +// + vm_local_words +// + 16 words of register save area +// save frame doing a save sp, -X, sp growing towards lower addresses +// set Lbcp, Lmethod, LcpoolCache +// set Llocals to i0 +// set Lmonitors to FP - rounded_vm_local_words +// set Lesp to Lmonitors - 4 +// +// The frame has now been setup to do the rest of the entry code + +// Try this optimization: Most method entries could live in a +// "one size fits all" stack frame without all the dynamic size +// calculations. It might be profitable to do all this calculation +// statically and approximately for "small enough" methods. + +//----------------------------------------------------------------------------------------------- + +// C1 Calling conventions +// +// Upon method entry, the following registers are setup: +// +// g2 G2_thread: current thread +// g5 G5_method: method to activate +// g4 Gargs : pointer to last argument +// +// +// Stack: +// +// +---------------+ <--- sp +// | | +// : reg save area : +// | | +// +---------------+ <--- sp + 0x40 +// | | +// : extra 7 slots : note: these slots are not really needed for the interpreter (fix later) +// | | +// +---------------+ <--- sp + 0x5c +// | | +// : free : +// | | +// +---------------+ <--- Gargs +// | | +// : arguments : +// | | +// +---------------+ +// | | +// +// +// +// AFTER FRAME HAS BEEN SETUP for method interpretation the stack looks like: +// +// +---------------+ <--- sp +// | | +// : reg save area : +// | | +// +---------------+ <--- sp + 0x40 +// | | +// : extra 7 slots : note: these slots are not really needed for the interpreter (fix later) +// | | +// +---------------+ <--- sp + 0x5c +// | | +// : : +// | | <--- Lesp +// +---------------+ <--- Lmonitors (fp - 0x18) +// | VM locals | +// +---------------+ <--- fp +// | | +// : reg save area : +// | | +// +---------------+ <--- fp + 0x40 +// | | +// : extra 7 slots : note: these slots are not really needed for the interpreter (fix later) +// | | +// +---------------+ <--- fp + 0x5c +// | | +// : free : +// | | +// +---------------+ +// | | +// : nonarg locals : +// | | +// +---------------+ +// | | +// : arguments : +// | | <--- Llocals +// +---------------+ <--- Gargs +// | | + void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { // // @@ -599,136 +708,6 @@ } -// Empty method, generate a very fast return. - -address InterpreterGenerator::generate_empty_entry(void) { - - // A method that does nother but return... - - address entry = __ pc(); - Label slow_path; - - // do nothing for empty methods (do not even increment invocation counter) - if ( UseFastEmptyMethods) { - // If we need a safepoint check, generate full interpreter entry. - AddressLiteral sync_state(SafepointSynchronize::address_of_state()); - __ set(sync_state, G3_scratch); - __ cmp_and_br_short(G3_scratch, SafepointSynchronize::_not_synchronized, Assembler::notEqual, Assembler::pn, slow_path); - - // Code: _return - __ retl(); - __ delayed()->mov(O5_savedSP, SP); - - __ bind(slow_path); - (void) generate_normal_entry(false); - - return entry; - } - return NULL; -} - -// Call an accessor method (assuming it is resolved, otherwise drop into -// vanilla (slow path) entry - -// Generates code to elide accessor methods -// Uses G3_scratch and G1_scratch as scratch -address InterpreterGenerator::generate_accessor_entry(void) { - - // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; - // parameter size = 1 - // Note: We can only use this code if the getfield has been resolved - // and if we don't have a null-pointer exception => check for - // these conditions first and use slow path if necessary. - address entry = __ pc(); - Label slow_path; - - - // XXX: for compressed oops pointer loading and decoding doesn't fit in - // delay slot and damages G1 - if ( UseFastAccessorMethods && !UseCompressedOops ) { - // Check if we need to reach a safepoint and generate full interpreter - // frame if so. - AddressLiteral sync_state(SafepointSynchronize::address_of_state()); - __ load_contents(sync_state, G3_scratch); - __ cmp(G3_scratch, SafepointSynchronize::_not_synchronized); - __ cmp_and_br_short(G3_scratch, SafepointSynchronize::_not_synchronized, Assembler::notEqual, Assembler::pn, slow_path); - - // Check if local 0 != NULL - __ ld_ptr(Gargs, G0, Otos_i ); // get local 0 - // check if local 0 == NULL and go the slow path - __ br_null_short(Otos_i, Assembler::pn, slow_path); - - - // read first instruction word and extract bytecode @ 1 and index @ 2 - // get first 4 bytes of the bytecodes (big endian!) - __ ld_ptr(G5_method, Method::const_offset(), G1_scratch); - __ ld(G1_scratch, ConstMethod::codes_offset(), G1_scratch); - - // move index @ 2 far left then to the right most two bytes. - __ sll(G1_scratch, 2*BitsPerByte, G1_scratch); - __ srl(G1_scratch, 2*BitsPerByte - exact_log2(in_words( - ConstantPoolCacheEntry::size()) * BytesPerWord), G1_scratch); - - // get constant pool cache - __ ld_ptr(G5_method, Method::const_offset(), G3_scratch); - __ ld_ptr(G3_scratch, ConstMethod::constants_offset(), G3_scratch); - __ ld_ptr(G3_scratch, ConstantPool::cache_offset_in_bytes(), G3_scratch); - - // get specific constant pool cache entry - __ add(G3_scratch, G1_scratch, G3_scratch); - - // Check the constant Pool cache entry to see if it has been resolved. - // If not, need the slow path. - ByteSize cp_base_offset = ConstantPoolCache::base_offset(); - __ ld_ptr(G3_scratch, cp_base_offset + ConstantPoolCacheEntry::indices_offset(), G1_scratch); - __ srl(G1_scratch, 2*BitsPerByte, G1_scratch); - __ and3(G1_scratch, 0xFF, G1_scratch); - __ cmp_and_br_short(G1_scratch, Bytecodes::_getfield, Assembler::notEqual, Assembler::pn, slow_path); - - // Get the type and return field offset from the constant pool cache - __ ld_ptr(G3_scratch, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), G1_scratch); - __ ld_ptr(G3_scratch, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), G3_scratch); - - Label xreturn_path; - // Need to differentiate between igetfield, agetfield, bgetfield etc. - // because they are different sizes. - // Get the type from the constant pool cache - __ srl(G1_scratch, ConstantPoolCacheEntry::tos_state_shift, G1_scratch); - // Make sure we don't need to mask G1_scratch after the above shift - ConstantPoolCacheEntry::verify_tos_state_shift(); - __ cmp(G1_scratch, atos ); - __ br(Assembler::equal, true, Assembler::pt, xreturn_path); - __ delayed()->ld_ptr(Otos_i, G3_scratch, Otos_i); - __ cmp(G1_scratch, itos); - __ br(Assembler::equal, true, Assembler::pt, xreturn_path); - __ delayed()->ld(Otos_i, G3_scratch, Otos_i); - __ cmp(G1_scratch, stos); - __ br(Assembler::equal, true, Assembler::pt, xreturn_path); - __ delayed()->ldsh(Otos_i, G3_scratch, Otos_i); - __ cmp(G1_scratch, ctos); - __ br(Assembler::equal, true, Assembler::pt, xreturn_path); - __ delayed()->lduh(Otos_i, G3_scratch, Otos_i); -#ifdef ASSERT - __ cmp(G1_scratch, btos); - __ br(Assembler::equal, true, Assembler::pt, xreturn_path); - __ delayed()->ldsb(Otos_i, G3_scratch, Otos_i); - __ should_not_reach_here(); -#endif - __ ldsb(Otos_i, G3_scratch, Otos_i); - __ bind(xreturn_path); - - // _ireturn/_areturn - __ retl(); // return from leaf routine - __ delayed()->mov(O5_savedSP, SP); - - // Generate regular method entry - __ bind(slow_path); - (void) generate_normal_entry(false); - return entry; - } - return NULL; -} - // Method entry for java.lang.ref.Reference.get. address InterpreterGenerator::generate_Reference_get_entry(void) { #if INCLUDE_ALL_GCS @@ -806,7 +785,7 @@ // If G1 is not enabled then attempt to go through the accessor entry point // Reference.get is an accessor - return generate_accessor_entry(); + return generate_jump_to_normal_entry(); } // @@ -1242,8 +1221,6 @@ // Generic method entry to (asm) interpreter -//------------------------------------------------------------------------------------------------------------------------ -// address InterpreterGenerator::generate_normal_entry(bool synchronized) { address entry = __ pc(); @@ -1410,123 +1387,6 @@ return entry; } - -//---------------------------------------------------------------------------------------------------- -// Entry points & stack frame layout -// -// Here we generate the various kind of entries into the interpreter. -// The two main entry type are generic bytecode methods and native call method. -// These both come in synchronized and non-synchronized versions but the -// frame layout they create is very similar. The other method entry -// types are really just special purpose entries that are really entry -// and interpretation all in one. These are for trivial methods like -// accessor, empty, or special math methods. -// -// When control flow reaches any of the entry types for the interpreter -// the following holds -> -// -// C2 Calling Conventions: -// -// The entry code below assumes that the following registers are set -// when coming in: -// G5_method: holds the Method* of the method to call -// Lesp: points to the TOS of the callers expression stack -// after having pushed all the parameters -// -// The entry code does the following to setup an interpreter frame -// pop parameters from the callers stack by adjusting Lesp -// set O0 to Lesp -// compute X = (max_locals - num_parameters) -// bump SP up by X to accomadate the extra locals -// compute X = max_expression_stack -// + vm_local_words -// + 16 words of register save area -// save frame doing a save sp, -X, sp growing towards lower addresses -// set Lbcp, Lmethod, LcpoolCache -// set Llocals to i0 -// set Lmonitors to FP - rounded_vm_local_words -// set Lesp to Lmonitors - 4 -// -// The frame has now been setup to do the rest of the entry code - -// Try this optimization: Most method entries could live in a -// "one size fits all" stack frame without all the dynamic size -// calculations. It might be profitable to do all this calculation -// statically and approximately for "small enough" methods. - -//----------------------------------------------------------------------------------------------- - -// C1 Calling conventions -// -// Upon method entry, the following registers are setup: -// -// g2 G2_thread: current thread -// g5 G5_method: method to activate -// g4 Gargs : pointer to last argument -// -// -// Stack: -// -// +---------------+ <--- sp -// | | -// : reg save area : -// | | -// +---------------+ <--- sp + 0x40 -// | | -// : extra 7 slots : note: these slots are not really needed for the interpreter (fix later) -// | | -// +---------------+ <--- sp + 0x5c -// | | -// : free : -// | | -// +---------------+ <--- Gargs -// | | -// : arguments : -// | | -// +---------------+ -// | | -// -// -// -// AFTER FRAME HAS BEEN SETUP for method interpretation the stack looks like: -// -// +---------------+ <--- sp -// | | -// : reg save area : -// | | -// +---------------+ <--- sp + 0x40 -// | | -// : extra 7 slots : note: these slots are not really needed for the interpreter (fix later) -// | | -// +---------------+ <--- sp + 0x5c -// | | -// : : -// | | <--- Lesp -// +---------------+ <--- Lmonitors (fp - 0x18) -// | VM locals | -// +---------------+ <--- fp -// | | -// : reg save area : -// | | -// +---------------+ <--- fp + 0x40 -// | | -// : extra 7 slots : note: these slots are not really needed for the interpreter (fix later) -// | | -// +---------------+ <--- fp + 0x5c -// | | -// : free : -// | | -// +---------------+ -// | | -// : nonarg locals : -// | | -// +---------------+ -// | | -// : arguments : -// | | <--- Llocals -// +---------------+ <--- Gargs -// | | - static int size_activation_helper(int callee_extra_locals, int max_stack, int monitor_size) { // Figure out the size of an interpreter frame (in words) given that we have a fully allocated
--- a/hotspot/src/cpu/sparc/vm/vm_version_sparc.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/sparc/vm/vm_version_sparc.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -26,6 +26,7 @@ #include "asm/macroAssembler.inline.hpp" #include "memory/resourceArea.hpp" #include "runtime/java.hpp" +#include "runtime/os.hpp" #include "runtime/stubCodeGenerator.hpp" #include "vm_version_sparc.hpp" @@ -249,7 +250,7 @@ (!has_hardware_fsmuld() ? ", no-fsmuld" : "")); // buf is started with ", " or is empty - _features_str = strdup(strlen(buf) > 2 ? buf + 2 : buf); + _features_str = os::strdup(strlen(buf) > 2 ? buf + 2 : buf); // There are three 64-bit SPARC families that do not overlap, e.g., // both is_ultra3() and is_sparc64() cannot be true at the same time.
--- a/hotspot/src/cpu/x86/vm/cppInterpreterGenerator_x86.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/x86/vm/cppInterpreterGenerator_x86.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, 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 @@ -27,21 +27,6 @@ protected: -#if 0 - address generate_asm_interpreter_entry(bool synchronized); - address generate_native_entry(bool synchronized); - address generate_abstract_entry(void); - address generate_math_entry(AbstractInterpreter::MethodKind kind); - address generate_empty_entry(void); - address generate_accessor_entry(void); - address generate_Reference_get_entry(void); - void lock_method(void); - void generate_stack_overflow_check(void); - - void generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue); - void generate_counter_overflow(Label* do_continue); -#endif - void generate_more_monitors(); void generate_deopt_handling(); address generate_interpreter_frame_manager(bool synchronized); // C++ interpreter only
--- a/hotspot/src/cpu/x86/vm/cppInterpreter_x86.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/x86/vm/cppInterpreter_x86.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2007, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2007, 2014, 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 @@ -66,9 +66,6 @@ #define __ _masm-> #define STATE(field_name) (Address(state, byte_offset_of(BytecodeInterpreter, field_name))) -Label fast_accessor_slow_entry_path; // fast accessor methods need to be able to jmp to unsynchronized - // c++ interpreter entry point this holds that entry point label. - // default registers for state and sender_sp // state and sender_sp are the same on 32bit because we have no choice. // state could be rsi on 64bit but it is an arg reg and not callee save @@ -660,7 +657,6 @@ // generate_method_entry) so the guard should work for them too. // - // monitor entry size: see picture of stack set (generate_method_entry) and frame_i486.hpp const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; // total overhead size: entry_size + (saved rbp, thru expr stack bottom). @@ -794,156 +790,6 @@ __ lock_object(monitor); } -// Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry - -address InterpreterGenerator::generate_accessor_entry(void) { - - // rbx: Method* - - // rsi/r13: senderSP must preserved for slow path, set SP to it on fast path - - Label xreturn_path; - - // do fastpath for resolved accessor methods - if (UseFastAccessorMethods) { - - address entry_point = __ pc(); - - Label slow_path; - // If we need a safepoint check, generate full interpreter entry. - ExternalAddress state(SafepointSynchronize::address_of_state()); - __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), - SafepointSynchronize::_not_synchronized); - - __ jcc(Assembler::notEqual, slow_path); - // ASM/C++ Interpreter - // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1 - // Note: We can only use this code if the getfield has been resolved - // and if we don't have a null-pointer exception => check for - // these conditions first and use slow path if necessary. - // rbx,: method - // rcx: receiver - __ movptr(rax, Address(rsp, wordSize)); - - // check if local 0 != NULL and read field - __ testptr(rax, rax); - __ jcc(Assembler::zero, slow_path); - - // read first instruction word and extract bytecode @ 1 and index @ 2 - __ movptr(rdx, Address(rbx, Method::const_offset())); - __ movptr(rdi, Address(rdx, ConstMethod::constants_offset())); - __ movl(rdx, Address(rdx, ConstMethod::codes_offset())); - // Shift codes right to get the index on the right. - // The bytecode fetched looks like <index><0xb4><0x2a> - __ shrl(rdx, 2*BitsPerByte); - __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); - __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes())); - - // rax,: local 0 - // rbx,: method - // rcx: receiver - do not destroy since it is needed for slow path! - // rcx: scratch - // rdx: constant pool cache index - // rdi: constant pool cache - // rsi/r13: sender sp - - // check if getfield has been resolved and read constant pool cache entry - // check the validity of the cache entry by testing whether _indices field - // contains Bytecode::_getfield in b1 byte. - assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below"); - __ movl(rcx, - Address(rdi, - rdx, - Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset())); - __ shrl(rcx, 2*BitsPerByte); - __ andl(rcx, 0xFF); - __ cmpl(rcx, Bytecodes::_getfield); - __ jcc(Assembler::notEqual, slow_path); - - // Note: constant pool entry is not valid before bytecode is resolved - __ movptr(rcx, - Address(rdi, - rdx, - Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset())); - __ movl(rdx, - Address(rdi, - rdx, - Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset())); - - Label notByte, notShort, notChar; - const Address field_address (rax, rcx, Address::times_1); - - // Need to differentiate between igetfield, agetfield, bgetfield etc. - // because they are different sizes. - // Use the type from the constant pool cache - __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift); - // Make sure we don't need to mask rdx after the above shift - ConstantPoolCacheEntry::verify_tos_state_shift(); -#ifdef _LP64 - Label notObj; - __ cmpl(rdx, atos); - __ jcc(Assembler::notEqual, notObj); - // atos - __ movptr(rax, field_address); - __ jmp(xreturn_path); - - __ bind(notObj); -#endif // _LP64 - __ cmpl(rdx, btos); - __ jcc(Assembler::notEqual, notByte); - __ load_signed_byte(rax, field_address); - __ jmp(xreturn_path); - - __ bind(notByte); - __ cmpl(rdx, stos); - __ jcc(Assembler::notEqual, notShort); - __ load_signed_short(rax, field_address); - __ jmp(xreturn_path); - - __ bind(notShort); - __ cmpl(rdx, ctos); - __ jcc(Assembler::notEqual, notChar); - __ load_unsigned_short(rax, field_address); - __ jmp(xreturn_path); - - __ bind(notChar); -#ifdef ASSERT - Label okay; -#ifndef _LP64 - __ cmpl(rdx, atos); - __ jcc(Assembler::equal, okay); -#endif // _LP64 - __ cmpl(rdx, itos); - __ jcc(Assembler::equal, okay); - __ stop("what type is this?"); - __ bind(okay); -#endif // ASSERT - // All the rest are a 32 bit wordsize - __ movl(rax, field_address); - - __ bind(xreturn_path); - - // _ireturn/_areturn - __ pop(rdi); // get return address - __ mov(rsp, sender_sp_on_entry); // set sp to sender sp - __ jmp(rdi); - - // generate a vanilla interpreter entry as the slow path - __ bind(slow_path); - // We will enter c++ interpreter looking like it was - // called by the call_stub this will cause it to return - // a tosca result to the invoker which might have been - // the c++ interpreter itself. - - __ jmp(fast_accessor_slow_entry_path); - return entry_point; - - } else { - return NULL; - } - -} - address InterpreterGenerator::generate_Reference_get_entry(void) { #if INCLUDE_ALL_GCS if (UseG1GC) { @@ -961,7 +807,7 @@ // If G1 is not enabled then attempt to go through the accessor entry point // Reference.get is an accessor - return generate_accessor_entry(); + return generate_jump_to_normal_entry(); } // @@ -1670,10 +1516,6 @@ address entry_point = __ pc(); - // Fast accessor methods share this entry point. - // This works because frame manager is in the same codelet - if (UseFastAccessorMethods && !synchronized) __ bind(fast_accessor_slow_entry_path); - Label dispatch_entry_2; __ movptr(rcx, sender_sp_on_entry); __ movptr(state, (int32_t)NULL_WORD); // no current activation @@ -2212,40 +2054,6 @@ return entry_point; } -address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) { - // determine code generation flags - bool synchronized = false; - address entry_point = NULL; - - switch (kind) { - case Interpreter::zerolocals : break; - case Interpreter::zerolocals_synchronized: synchronized = true; break; - case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break; - case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break; - case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break; - case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break; - case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break; - - case Interpreter::java_lang_math_sin : // fall thru - case Interpreter::java_lang_math_cos : // fall thru - case Interpreter::java_lang_math_tan : // fall thru - case Interpreter::java_lang_math_abs : // fall thru - case Interpreter::java_lang_math_log : // fall thru - case Interpreter::java_lang_math_log10 : // fall thru - case Interpreter::java_lang_math_sqrt : // fall thru - case Interpreter::java_lang_math_pow : // fall thru - case Interpreter::java_lang_math_exp : // fall thru - entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break; - case Interpreter::java_lang_ref_reference_get - : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break; - default : ShouldNotReachHere(); break; - } - - if (entry_point) return entry_point; - - return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized); - -} InterpreterGenerator::InterpreterGenerator(StubQueue* code) : CppInterpreterGenerator(code) {
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/hotspot/src/cpu/x86/vm/interpreterGenerator_x86.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -0,0 +1,66 @@ +/* + * Copyright (c) 2014, 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. + * + */ + +#include "precompiled.hpp" +#include "asm/macroAssembler.hpp" +#include "interpreter/interpreter.hpp" +#include "interpreter/interpreterGenerator.hpp" +#include "interpreter/interpreterRuntime.hpp" +#include "interpreter/interp_masm.hpp" + +#define __ _masm-> + +// Jump into normal path for accessor and empty entry to jump to normal entry +// The "fast" optimization don't update compilation count therefore can disable inlining +// for these functions that should be inlined. +address InterpreterGenerator::generate_jump_to_normal_entry(void) { + address entry_point = __ pc(); + + assert(Interpreter::entry_for_kind(Interpreter::zerolocals) != NULL, "should already be generated"); + __ jump(RuntimeAddress(Interpreter::entry_for_kind(Interpreter::zerolocals))); + return entry_point; +} + +// Abstract method entry +// Attempt to execute abstract method. Throw exception +address InterpreterGenerator::generate_abstract_entry(void) { + + address entry_point = __ pc(); + + // abstract method entry + +#ifndef CC_INTERP + // pop return address, reset last_sp to NULL + __ empty_expression_stack(); + __ restore_bcp(); // rsi must be correct for exception handler (was destroyed) + __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) +#endif + + // throw exception + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError)); + // the call_VM checks for exception, so we should never return here. + __ should_not_reach_here(); + + return entry_point; +}
--- a/hotspot/src/cpu/x86/vm/interpreterGenerator_x86.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/x86/vm/interpreterGenerator_x86.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, 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 @@ -36,8 +36,9 @@ address generate_native_entry(bool synchronized); address generate_abstract_entry(void); address generate_math_entry(AbstractInterpreter::MethodKind kind); - address generate_empty_entry(void); - address generate_accessor_entry(void); + address generate_jump_to_normal_entry(void); + address generate_accessor_entry(void) { return generate_jump_to_normal_entry(); } + address generate_empty_entry(void) { return generate_jump_to_normal_entry(); } address generate_Reference_get_entry(); address generate_CRC32_update_entry(); address generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind);
--- a/hotspot/src/cpu/x86/vm/interpreter_x86_32.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/x86/vm/interpreter_x86_32.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -67,45 +67,6 @@ } -// -// Various method entries (that c++ and asm interpreter agree upon) -//------------------------------------------------------------------------------------------------------------------------ -// -// - -// Empty method, generate a very fast return. - -address InterpreterGenerator::generate_empty_entry(void) { - - // rbx,: Method* - // rcx: receiver (unused) - // rsi: previous interpreter state (C++ interpreter) must preserve - // rsi: sender sp must set sp to this value on return - - if (!UseFastEmptyMethods) return NULL; - - address entry_point = __ pc(); - - // If we need a safepoint check, generate full interpreter entry. - Label slow_path; - ExternalAddress state(SafepointSynchronize::address_of_state()); - __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), - SafepointSynchronize::_not_synchronized); - __ jcc(Assembler::notEqual, slow_path); - - // do nothing for empty methods (do not even increment invocation counter) - // Code: _return - // _return - // return w/o popping parameters - __ pop(rax); - __ mov(rsp, rsi); - __ jmp(rax); - - __ bind(slow_path); - (void) generate_normal_entry(false); - return entry_point; -} - address InterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) { // rbx,: Method* @@ -216,36 +177,6 @@ } -// Abstract method entry -// Attempt to execute abstract method. Throw exception -address InterpreterGenerator::generate_abstract_entry(void) { - - // rbx,: Method* - // rcx: receiver (unused) - // rsi: previous interpreter state (C++ interpreter) must preserve - - // rsi: sender SP - - address entry_point = __ pc(); - - // abstract method entry - -#ifndef CC_INTERP - // pop return address, reset last_sp to NULL - __ empty_expression_stack(); - __ restore_bcp(); // rsi must be correct for exception handler (was destroyed) - __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) -#endif - - // throw exception - __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError)); - // the call_VM checks for exception, so we should never return here. - __ should_not_reach_here(); - - return entry_point; -} - - void Deoptimization::unwind_callee_save_values(frame* f, vframeArray* vframe_array) { // This code is sort of the equivalent of C2IAdapter::setup_stack_frame back in
--- a/hotspot/src/cpu/x86/vm/interpreter_x86_64.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/x86/vm/interpreter_x86_64.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -301,66 +301,6 @@ return entry_point; } - -// Abstract method entry -// Attempt to execute abstract method. Throw exception -address InterpreterGenerator::generate_abstract_entry(void) { - // rbx: Method* - // r13: sender SP - - address entry_point = __ pc(); - - // abstract method entry - -#ifndef CC_INTERP - // pop return address, reset last_sp to NULL - __ empty_expression_stack(); - __ restore_bcp(); // rsi must be correct for exception handler (was destroyed) - __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) -#endif - - // throw exception - __ call_VM(noreg, CAST_FROM_FN_PTR(address, - InterpreterRuntime::throw_AbstractMethodError)); - // the call_VM checks for exception, so we should never return here. - __ should_not_reach_here(); - - return entry_point; -} - - -// Empty method, generate a very fast return. - -address InterpreterGenerator::generate_empty_entry(void) { - // rbx: Method* - // r13: sender sp must set sp to this value on return - - if (!UseFastEmptyMethods) { - return NULL; - } - - address entry_point = __ pc(); - - // If we need a safepoint check, generate full interpreter entry. - Label slow_path; - __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), - SafepointSynchronize::_not_synchronized); - __ jcc(Assembler::notEqual, slow_path); - - // do nothing for empty methods (do not even increment invocation counter) - // Code: _return - // _return - // return w/o popping parameters - __ pop(rax); - __ mov(rsp, r13); - __ jmp(rax); - - __ bind(slow_path); - (void) generate_normal_entry(false); - return entry_point; - -} - void Deoptimization::unwind_callee_save_values(frame* f, vframeArray* vframe_array) { // This code is sort of the equivalent of C2IAdapter::setup_stack_frame back in
--- a/hotspot/src/cpu/x86/vm/templateInterpreter_x86.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/x86/vm/templateInterpreter_x86.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -38,7 +38,7 @@ int callee_locals, bool is_top_frame) { // Note: This calculation must exactly parallel the frame setup - // in AbstractInterpreterGenerator::generate_method_entry. + // in InterpreterGenerator::generate_fixed_frame. // fixed size of an interpreter frame: int overhead = frame::sender_sp_offset -
--- a/hotspot/src/cpu/x86/vm/templateInterpreter_x86_32.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/x86/vm/templateInterpreter_x86_32.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -468,10 +468,10 @@ // rax, // NOTE: since the additional locals are also always pushed (wasn't obvious in - // generate_method_entry) so the guard should work for them too. + // generate_fixed_frame) so the guard should work for them too. // - // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp + // monitor entry size: see picture of stack in frame_x86.hpp const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; // total overhead size: entry_size + (saved rbp, thru expr stack bottom). @@ -633,145 +633,6 @@ __ movptr(Address(rsp, 0), rsp); // set expression stack bottom } -// End of helpers - -// -// Various method entries -//------------------------------------------------------------------------------------------------------------------------ -// -// - -// Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry - -address InterpreterGenerator::generate_accessor_entry(void) { - - // rbx,: Method* - // rcx: receiver (preserve for slow entry into asm interpreter) - - // rsi: senderSP must preserved for slow path, set SP to it on fast path - - address entry_point = __ pc(); - Label xreturn_path; - - // do fastpath for resolved accessor methods - if (UseFastAccessorMethods) { - Label slow_path; - // If we need a safepoint check, generate full interpreter entry. - ExternalAddress state(SafepointSynchronize::address_of_state()); - __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), - SafepointSynchronize::_not_synchronized); - - __ jcc(Assembler::notEqual, slow_path); - // ASM/C++ Interpreter - // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1 - // Note: We can only use this code if the getfield has been resolved - // and if we don't have a null-pointer exception => check for - // these conditions first and use slow path if necessary. - // rbx,: method - // rcx: receiver - __ movptr(rax, Address(rsp, wordSize)); - - // check if local 0 != NULL and read field - __ testptr(rax, rax); - __ jcc(Assembler::zero, slow_path); - - // read first instruction word and extract bytecode @ 1 and index @ 2 - __ movptr(rdx, Address(rbx, Method::const_offset())); - __ movptr(rdi, Address(rdx, ConstMethod::constants_offset())); - __ movl(rdx, Address(rdx, ConstMethod::codes_offset())); - // Shift codes right to get the index on the right. - // The bytecode fetched looks like <index><0xb4><0x2a> - __ shrl(rdx, 2*BitsPerByte); - __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); - __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes())); - - // rax,: local 0 - // rbx,: method - // rcx: receiver - do not destroy since it is needed for slow path! - // rcx: scratch - // rdx: constant pool cache index - // rdi: constant pool cache - // rsi: sender sp - - // check if getfield has been resolved and read constant pool cache entry - // check the validity of the cache entry by testing whether _indices field - // contains Bytecode::_getfield in b1 byte. - assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below"); - __ movl(rcx, - Address(rdi, - rdx, - Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset())); - __ shrl(rcx, 2*BitsPerByte); - __ andl(rcx, 0xFF); - __ cmpl(rcx, Bytecodes::_getfield); - __ jcc(Assembler::notEqual, slow_path); - - // Note: constant pool entry is not valid before bytecode is resolved - __ movptr(rcx, - Address(rdi, - rdx, - Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset())); - __ movl(rdx, - Address(rdi, - rdx, - Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset())); - - Label notByte, notShort, notChar; - const Address field_address (rax, rcx, Address::times_1); - - // Need to differentiate between igetfield, agetfield, bgetfield etc. - // because they are different sizes. - // Use the type from the constant pool cache - __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift); - // Make sure we don't need to mask rdx after the above shift - ConstantPoolCacheEntry::verify_tos_state_shift(); - __ cmpl(rdx, btos); - __ jcc(Assembler::notEqual, notByte); - __ load_signed_byte(rax, field_address); - __ jmp(xreturn_path); - - __ bind(notByte); - __ cmpl(rdx, stos); - __ jcc(Assembler::notEqual, notShort); - __ load_signed_short(rax, field_address); - __ jmp(xreturn_path); - - __ bind(notShort); - __ cmpl(rdx, ctos); - __ jcc(Assembler::notEqual, notChar); - __ load_unsigned_short(rax, field_address); - __ jmp(xreturn_path); - - __ bind(notChar); -#ifdef ASSERT - Label okay; - __ cmpl(rdx, atos); - __ jcc(Assembler::equal, okay); - __ cmpl(rdx, itos); - __ jcc(Assembler::equal, okay); - __ stop("what type is this?"); - __ bind(okay); -#endif // ASSERT - // All the rest are a 32 bit wordsize - // This is ok for now. Since fast accessors should be going away - __ movptr(rax, field_address); - - __ bind(xreturn_path); - - // _ireturn/_areturn - __ pop(rdi); // get return address - __ mov(rsp, rsi); // set sp to sender sp - __ jmp(rdi); - - // generate a vanilla interpreter entry as the slow path - __ bind(slow_path); - - (void) generate_normal_entry(false); - return entry_point; - } - return NULL; - -} // Method entry for java.lang.ref.Reference.get. address InterpreterGenerator::generate_Reference_get_entry(void) { @@ -862,7 +723,7 @@ // If G1 is not enabled then attempt to go through the accessor entry point // Reference.get is an accessor - return generate_accessor_entry(); + return generate_jump_to_normal_entry(); } /** @@ -1557,100 +1418,6 @@ return entry_point; } -//------------------------------------------------------------------------------------------------------------------------ -// Entry points -// -// Here we generate the various kind of entries into the interpreter. -// The two main entry type are generic bytecode methods and native call method. -// These both come in synchronized and non-synchronized versions but the -// frame layout they create is very similar. The other method entry -// types are really just special purpose entries that are really entry -// and interpretation all in one. These are for trivial methods like -// accessor, empty, or special math methods. -// -// When control flow reaches any of the entry types for the interpreter -// the following holds -> -// -// Arguments: -// -// rbx,: Method* -// rcx: receiver -// -// -// Stack layout immediately at entry -// -// [ return address ] <--- rsp -// [ parameter n ] -// ... -// [ parameter 1 ] -// [ expression stack ] (caller's java expression stack) - -// Assuming that we don't go to one of the trivial specialized -// entries the stack will look like below when we are ready to execute -// the first bytecode (or call the native routine). The register usage -// will be as the template based interpreter expects (see interpreter_x86.hpp). -// -// local variables follow incoming parameters immediately; i.e. -// the return address is moved to the end of the locals). -// -// [ monitor entry ] <--- rsp -// ... -// [ monitor entry ] -// [ expr. stack bottom ] -// [ saved rsi ] -// [ current rdi ] -// [ Method* ] -// [ saved rbp, ] <--- rbp, -// [ return address ] -// [ local variable m ] -// ... -// [ local variable 1 ] -// [ parameter n ] -// ... -// [ parameter 1 ] <--- rdi - -address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) { - // determine code generation flags - bool synchronized = false; - address entry_point = NULL; - InterpreterGenerator* ig_this = (InterpreterGenerator*)this; - - switch (kind) { - case Interpreter::zerolocals : break; - case Interpreter::zerolocals_synchronized: synchronized = true; break; - case Interpreter::native : entry_point = ig_this->generate_native_entry(false); break; - case Interpreter::native_synchronized : entry_point = ig_this->generate_native_entry(true); break; - case Interpreter::empty : entry_point = ig_this->generate_empty_entry(); break; - case Interpreter::accessor : entry_point = ig_this->generate_accessor_entry(); break; - case Interpreter::abstract : entry_point = ig_this->generate_abstract_entry(); break; - - case Interpreter::java_lang_math_sin : // fall thru - case Interpreter::java_lang_math_cos : // fall thru - case Interpreter::java_lang_math_tan : // fall thru - case Interpreter::java_lang_math_abs : // fall thru - case Interpreter::java_lang_math_log : // fall thru - case Interpreter::java_lang_math_log10 : // fall thru - case Interpreter::java_lang_math_sqrt : // fall thru - case Interpreter::java_lang_math_pow : // fall thru - case Interpreter::java_lang_math_exp : entry_point = ig_this->generate_math_entry(kind); break; - case Interpreter::java_lang_ref_reference_get - : entry_point = ig_this->generate_Reference_get_entry(); break; - case Interpreter::java_util_zip_CRC32_update - : entry_point = ig_this->generate_CRC32_update_entry(); break; - case Interpreter::java_util_zip_CRC32_updateBytes - : // fall thru - case Interpreter::java_util_zip_CRC32_updateByteBuffer - : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break; - default: - fatal(err_msg("unexpected method kind: %d", kind)); - break; - } - - if (entry_point) return entry_point; - - return ig_this->generate_normal_entry(synchronized); - -} // These should never be compiled since the interpreter will prefer // the compiled version to the intrinsic version.
--- a/hotspot/src/cpu/x86/vm/templateInterpreter_x86_64.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/x86/vm/templateInterpreter_x86_64.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -400,7 +400,7 @@ // page mechanism will work for that. // // NOTE: Since the additional locals are also always pushed (wasn't -// obvious in generate_method_entry) so the guard should work for them +// obvious in generate_fixed_frame) so the guard should work for them // too. // // Args: @@ -411,8 +411,7 @@ // rax void InterpreterGenerator::generate_stack_overflow_check(void) { - // monitor entry size: see picture of stack set - // (generate_method_entry) and frame_amd64.hpp + // monitor entry size: see picture of stack in frame_x86.hpp const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; // total overhead size: entry_size + (saved rbp through expr stack @@ -600,153 +599,6 @@ // End of helpers -// Various method entries -//------------------------------------------------------------------------------------------------------------------------ -// -// - -// Call an accessor method (assuming it is resolved, otherwise drop -// into vanilla (slow path) entry -address InterpreterGenerator::generate_accessor_entry(void) { - // rbx: Method* - - // r13: senderSP must preserver for slow path, set SP to it on fast path - - address entry_point = __ pc(); - Label xreturn_path; - - // do fastpath for resolved accessor methods - if (UseFastAccessorMethods) { - // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites - // thereof; parameter size = 1 - // Note: We can only use this code if the getfield has been resolved - // and if we don't have a null-pointer exception => check for - // these conditions first and use slow path if necessary. - Label slow_path; - // If we need a safepoint check, generate full interpreter entry. - __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), - SafepointSynchronize::_not_synchronized); - - __ jcc(Assembler::notEqual, slow_path); - // rbx: method - __ movptr(rax, Address(rsp, wordSize)); - - // check if local 0 != NULL and read field - __ testptr(rax, rax); - __ jcc(Assembler::zero, slow_path); - - // read first instruction word and extract bytecode @ 1 and index @ 2 - __ movptr(rdx, Address(rbx, Method::const_offset())); - __ movptr(rdi, Address(rdx, ConstMethod::constants_offset())); - __ movl(rdx, Address(rdx, ConstMethod::codes_offset())); - // Shift codes right to get the index on the right. - // The bytecode fetched looks like <index><0xb4><0x2a> - __ shrl(rdx, 2 * BitsPerByte); - __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); - __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes())); - - // rax: local 0 - // rbx: method - // rdx: constant pool cache index - // rdi: constant pool cache - - // check if getfield has been resolved and read constant pool cache entry - // check the validity of the cache entry by testing whether _indices field - // contains Bytecode::_getfield in b1 byte. - assert(in_words(ConstantPoolCacheEntry::size()) == 4, - "adjust shift below"); - __ movl(rcx, - Address(rdi, - rdx, - Address::times_8, - ConstantPoolCache::base_offset() + - ConstantPoolCacheEntry::indices_offset())); - __ shrl(rcx, 2 * BitsPerByte); - __ andl(rcx, 0xFF); - __ cmpl(rcx, Bytecodes::_getfield); - __ jcc(Assembler::notEqual, slow_path); - - // Note: constant pool entry is not valid before bytecode is resolved - __ movptr(rcx, - Address(rdi, - rdx, - Address::times_8, - ConstantPoolCache::base_offset() + - ConstantPoolCacheEntry::f2_offset())); - // edx: flags - __ movl(rdx, - Address(rdi, - rdx, - Address::times_8, - ConstantPoolCache::base_offset() + - ConstantPoolCacheEntry::flags_offset())); - - Label notObj, notInt, notByte, notShort; - const Address field_address(rax, rcx, Address::times_1); - - // Need to differentiate between igetfield, agetfield, bgetfield etc. - // because they are different sizes. - // Use the type from the constant pool cache - __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift); - // Make sure we don't need to mask edx after the above shift - ConstantPoolCacheEntry::verify_tos_state_shift(); - - __ cmpl(rdx, atos); - __ jcc(Assembler::notEqual, notObj); - // atos - __ load_heap_oop(rax, field_address); - __ jmp(xreturn_path); - - __ bind(notObj); - __ cmpl(rdx, itos); - __ jcc(Assembler::notEqual, notInt); - // itos - __ movl(rax, field_address); - __ jmp(xreturn_path); - - __ bind(notInt); - __ cmpl(rdx, btos); - __ jcc(Assembler::notEqual, notByte); - // btos - __ load_signed_byte(rax, field_address); - __ jmp(xreturn_path); - - __ bind(notByte); - __ cmpl(rdx, stos); - __ jcc(Assembler::notEqual, notShort); - // stos - __ load_signed_short(rax, field_address); - __ jmp(xreturn_path); - - __ bind(notShort); -#ifdef ASSERT - Label okay; - __ cmpl(rdx, ctos); - __ jcc(Assembler::equal, okay); - __ stop("what type is this?"); - __ bind(okay); -#endif - // ctos - __ load_unsigned_short(rax, field_address); - - __ bind(xreturn_path); - - // _ireturn/_areturn - __ pop(rdi); - __ mov(rsp, r13); - __ jmp(rdi); - __ ret(0); - - // generate a vanilla interpreter entry as the slow path - __ bind(slow_path); - (void) generate_normal_entry(false); - } else { - (void) generate_normal_entry(false); - } - - return entry_point; -} - // Method entry for java.lang.ref.Reference.get. address InterpreterGenerator::generate_Reference_get_entry(void) { #if INCLUDE_ALL_GCS @@ -773,8 +625,6 @@ // and so we don't need to call the G1 pre-barrier. Thus we can use the // regular method entry code to generate the NPE. // - // This code is based on generate_accessor_enty. - // // rbx: Method* // r13: senderSP must preserve for slow path, set SP to it on fast path @@ -832,7 +682,7 @@ // If G1 is not enabled then attempt to go through the accessor entry point // Reference.get is an accessor - return generate_accessor_entry(); + return generate_jump_to_normal_entry(); } /** @@ -1566,100 +1416,6 @@ return entry_point; } -// Entry points -// -// Here we generate the various kind of entries into the interpreter. -// The two main entry type are generic bytecode methods and native -// call method. These both come in synchronized and non-synchronized -// versions but the frame layout they create is very similar. The -// other method entry types are really just special purpose entries -// that are really entry and interpretation all in one. These are for -// trivial methods like accessor, empty, or special math methods. -// -// When control flow reaches any of the entry types for the interpreter -// the following holds -> -// -// Arguments: -// -// rbx: Method* -// -// Stack layout immediately at entry -// -// [ return address ] <--- rsp -// [ parameter n ] -// ... -// [ parameter 1 ] -// [ expression stack ] (caller's java expression stack) - -// Assuming that we don't go to one of the trivial specialized entries -// the stack will look like below when we are ready to execute the -// first bytecode (or call the native routine). The register usage -// will be as the template based interpreter expects (see -// interpreter_amd64.hpp). -// -// local variables follow incoming parameters immediately; i.e. -// the return address is moved to the end of the locals). -// -// [ monitor entry ] <--- rsp -// ... -// [ monitor entry ] -// [ expr. stack bottom ] -// [ saved r13 ] -// [ current r14 ] -// [ Method* ] -// [ saved ebp ] <--- rbp -// [ return address ] -// [ local variable m ] -// ... -// [ local variable 1 ] -// [ parameter n ] -// ... -// [ parameter 1 ] <--- r14 - -address AbstractInterpreterGenerator::generate_method_entry( - AbstractInterpreter::MethodKind kind) { - // determine code generation flags - bool synchronized = false; - address entry_point = NULL; - InterpreterGenerator* ig_this = (InterpreterGenerator*)this; - - switch (kind) { - case Interpreter::zerolocals : break; - case Interpreter::zerolocals_synchronized: synchronized = true; break; - case Interpreter::native : entry_point = ig_this->generate_native_entry(false); break; - case Interpreter::native_synchronized : entry_point = ig_this->generate_native_entry(true); break; - case Interpreter::empty : entry_point = ig_this->generate_empty_entry(); break; - case Interpreter::accessor : entry_point = ig_this->generate_accessor_entry(); break; - case Interpreter::abstract : entry_point = ig_this->generate_abstract_entry(); break; - - case Interpreter::java_lang_math_sin : // fall thru - case Interpreter::java_lang_math_cos : // fall thru - case Interpreter::java_lang_math_tan : // fall thru - case Interpreter::java_lang_math_abs : // fall thru - case Interpreter::java_lang_math_log : // fall thru - case Interpreter::java_lang_math_log10 : // fall thru - case Interpreter::java_lang_math_sqrt : // fall thru - case Interpreter::java_lang_math_pow : // fall thru - case Interpreter::java_lang_math_exp : entry_point = ig_this->generate_math_entry(kind); break; - case Interpreter::java_lang_ref_reference_get - : entry_point = ig_this->generate_Reference_get_entry(); break; - case Interpreter::java_util_zip_CRC32_update - : entry_point = ig_this->generate_CRC32_update_entry(); break; - case Interpreter::java_util_zip_CRC32_updateBytes - : // fall thru - case Interpreter::java_util_zip_CRC32_updateByteBuffer - : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break; - default: - fatal(err_msg("unexpected method kind: %d", kind)); - break; - } - - if (entry_point) { - return entry_point; - } - - return ig_this->generate_normal_entry(synchronized); -} // These should never be compiled since the interpreter will prefer // the compiled version to the intrinsic version.
--- a/hotspot/src/cpu/x86/vm/vm_version_x86.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/x86/vm/vm_version_x86.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -27,6 +27,7 @@ #include "asm/macroAssembler.inline.hpp" #include "memory/resourceArea.hpp" #include "runtime/java.hpp" +#include "runtime/os.hpp" #include "runtime/stubCodeGenerator.hpp" #include "vm_version_x86.hpp" @@ -514,7 +515,7 @@ (supports_tscinv() ? ", tscinv": ""), (supports_bmi1() ? ", bmi1" : ""), (supports_bmi2() ? ", bmi2" : "")); - _features_str = strdup(buf); + _features_str = os::strdup(buf); // UseSSE is set to the smaller of what hardware supports and what // the command line requires. I.e., you cannot set UseSSE to 2 on
--- a/hotspot/src/cpu/zero/vm/cppInterpreter_zero.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/zero/vm/cppInterpreter_zero.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2003, 2014, Oracle and/or its affiliates. All rights reserved. * Copyright 2007, 2008, 2009, 2010, 2011 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * @@ -831,60 +831,6 @@ return generate_entry((address) CppInterpreter::normal_entry); } -address AbstractInterpreterGenerator::generate_method_entry( - AbstractInterpreter::MethodKind kind) { - address entry_point = NULL; - - switch (kind) { - case Interpreter::zerolocals: - case Interpreter::zerolocals_synchronized: - break; - - case Interpreter::native: - entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); - break; - - case Interpreter::native_synchronized: - entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); - break; - - case Interpreter::empty: - entry_point = ((InterpreterGenerator*) this)->generate_empty_entry(); - break; - - case Interpreter::accessor: - entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry(); - break; - - case Interpreter::abstract: - entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry(); - break; - - case Interpreter::java_lang_math_sin: - case Interpreter::java_lang_math_cos: - case Interpreter::java_lang_math_tan: - case Interpreter::java_lang_math_abs: - case Interpreter::java_lang_math_log: - case Interpreter::java_lang_math_log10: - case Interpreter::java_lang_math_sqrt: - case Interpreter::java_lang_math_pow: - case Interpreter::java_lang_math_exp: - entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind); - break; - - case Interpreter::java_lang_ref_reference_get: - entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); - break; - - default: - ShouldNotReachHere(); - } - - if (entry_point == NULL) - entry_point = ((InterpreterGenerator*) this)->generate_normal_entry(false); - - return entry_point; -} InterpreterGenerator::InterpreterGenerator(StubQueue* code) : CppInterpreterGenerator(code) {
--- a/hotspot/src/cpu/zero/vm/globals_zero.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/zero/vm/globals_zero.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2000, 2014, Oracle and/or its affiliates. All rights reserved. * Copyright 2007, 2008, 2009, 2010, 2011 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * @@ -61,6 +61,12 @@ define_pd_global(uintx, TypeProfileLevel, 0); -#define ARCH_FLAGS(develop, product, diagnostic, experimental, notproduct) +#define ARCH_FLAGS(develop, product, diagnostic, experimental, notproduct) \ + product(bool, UseFastEmptyMethods, true, \ + "Use fast method entry code for empty methods") \ + \ + product(bool, UseFastAccessorMethods, true, \ + "Use fast method entry code for accessor methods") \ + \ #endif // CPU_ZERO_VM_GLOBALS_ZERO_HPP
--- a/hotspot/src/cpu/zero/vm/interpreterGenerator_zero.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/cpu/zero/vm/interpreterGenerator_zero.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. * Copyright 2007 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * @@ -39,4 +39,7 @@ address generate_accessor_entry(); address generate_Reference_get_entry(); + // Not supported + address generate_CRC32_update_entry() { return NULL; } + address generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { return NULL; } #endif // CPU_ZERO_VM_INTERPRETERGENERATOR_ZERO_HPP
--- a/hotspot/src/os/aix/vm/os_aix.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/os/aix/vm/os_aix.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -58,6 +58,7 @@ #include "runtime/mutexLocker.hpp" #include "runtime/objectMonitor.hpp" #include "runtime/orderAccess.inline.hpp" +#include "runtime/os.hpp" #include "runtime/osThread.hpp" #include "runtime/perfMemory.hpp" #include "runtime/sharedRuntime.hpp" @@ -378,10 +379,10 @@ // default should be 4K. size_t data_page_size = SIZE_4K; { - void* p = ::malloc(SIZE_16M); + void* p = os::malloc(SIZE_16M, mtInternal); guarantee(p != NULL, "malloc failed"); data_page_size = os::Aix::query_pagesize(p); - ::free(p); + os::free(p); } // query default shm page size (LDR_CNTRL SHMPSIZE)
--- a/hotspot/src/os/aix/vm/porting_aix.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/os/aix/vm/porting_aix.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -24,6 +24,8 @@ #include "asm/assembler.hpp" #include "memory/allocation.hpp" +#include "memory/allocation.inline.hpp" +#include "runtime/os.hpp" #include "loadlib_aix.hpp" #include "porting_aix.hpp" #include "utilities/debug.hpp" @@ -83,7 +85,7 @@ while (n) { node* p = n; n = n->next; - free(p->v); + os::free(p->v); delete p; } } @@ -95,7 +97,7 @@ } } node* p = new node; - p->v = strdup(s); + p->v = os::strdup_check_oom(s); p->next = first; first = p; return p->v;
--- a/hotspot/src/os/bsd/vm/os_bsd.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/os/bsd/vm/os_bsd.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -2439,23 +2439,25 @@ } // The memory is committed - MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC); + MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); return addr; } bool os::release_memory_special(char* base, size_t bytes) { - MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); - // detaching the SHM segment will also delete it, see reserve_memory_special() - int rslt = shmdt(base); - if (rslt == 0) { - tkr.record((address)base, bytes); - return true; + if (MemTracker::tracking_level() > NMT_minimal) { + Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); + // detaching the SHM segment will also delete it, see reserve_memory_special() + int rslt = shmdt(base); + if (rslt == 0) { + tkr.record((address)base, bytes); + return true; + } else { + return false; + } } else { - tkr.discard(); - return false; + return shmdt(base) == 0; } - } size_t os::large_page_size() {
--- a/hotspot/src/os/bsd/vm/perfMemory_bsd.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/os/bsd/vm/perfMemory_bsd.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -753,7 +753,7 @@ (void)::memset((void*) mapAddress, 0, size); // it does not go through os api, the operation has to record from here - MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC); + MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, CURRENT_PC, mtInternal); return mapAddress; } @@ -918,7 +918,7 @@ } // it does not go through os api, the operation has to record from here - MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC); + MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, CURRENT_PC, mtInternal); *addr = mapAddress; *sizep = size;
--- a/hotspot/src/os/linux/vm/os_linux.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/os/linux/vm/os_linux.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -3504,9 +3504,12 @@ assert(is_ptr_aligned(start, alignment), "Must be"); - // os::reserve_memory_special will record this memory area. - // Need to release it here to prevent overlapping reservations. - MemTracker::record_virtual_memory_release((address)start, bytes); + if (MemTracker::tracking_level() > NMT_minimal) { + // os::reserve_memory_special will record this memory area. + // Need to release it here to prevent overlapping reservations. + Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); + tkr.record((address)start, bytes); + } char* end = start + bytes; @@ -3601,7 +3604,7 @@ } // The memory is committed - MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC); + MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); } return addr; @@ -3617,24 +3620,30 @@ } bool os::release_memory_special(char* base, size_t bytes) { + bool res; + if (MemTracker::tracking_level() > NMT_minimal) { + Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); + res = os::Linux::release_memory_special_impl(base, bytes); + if (res) { + tkr.record((address)base, bytes); + } + + } else { + res = os::Linux::release_memory_special_impl(base, bytes); + } + return res; +} + +bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { assert(UseLargePages, "only for large pages"); - - MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); - bool res; + if (UseSHM) { res = os::Linux::release_memory_special_shm(base, bytes); } else { assert(UseHugeTLBFS, "must be"); res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); } - - if (res) { - tkr.record((address)base, bytes); - } else { - tkr.discard(); - } - return res; }
--- a/hotspot/src/os/linux/vm/os_linux.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/os/linux/vm/os_linux.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -108,6 +108,7 @@ static char* reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec); static char* reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec); + static bool release_memory_special_impl(char* base, size_t bytes); static bool release_memory_special_shm(char* base, size_t bytes); static bool release_memory_special_huge_tlbfs(char* base, size_t bytes);
--- a/hotspot/src/os/linux/vm/perfMemory_linux.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/os/linux/vm/perfMemory_linux.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -753,7 +753,7 @@ (void)::memset((void*) mapAddress, 0, size); // it does not go through os api, the operation has to record from here - MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC); + MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, CURRENT_PC, mtInternal); return mapAddress; } @@ -924,7 +924,7 @@ } // it does not go through os api, the operation has to record from here - MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC); + MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, CURRENT_PC, mtInternal); *addr = mapAddress; *sizep = size;
--- a/hotspot/src/os/posix/vm/os_posix.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/os/posix/vm/os_posix.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -75,21 +75,41 @@ VMError::report_coredump_status(buffer, success); } -address os::get_caller_pc(int n) { +int os::get_native_stack(address* stack, int frames, int toSkip) { #ifdef _NMT_NOINLINE_ - n ++; + toSkip++; #endif + + int frame_idx = 0; + int num_of_frames; // number of frames captured frame fr = os::current_frame(); - while (n > 0 && fr.pc() && - !os::is_first_C_frame(&fr) && fr.sender_pc()) { - fr = os::get_sender_for_C_frame(&fr); - n --; + while (fr.pc() && frame_idx < frames) { + if (toSkip > 0) { + toSkip --; + } else { + stack[frame_idx ++] = fr.pc(); + } + if (fr.fp() == NULL || os::is_first_C_frame(&fr) + ||fr.sender_pc() == NULL || fr.cb() != NULL) break; + + if (fr.sender_pc() && !os::is_first_C_frame(&fr)) { + fr = os::get_sender_for_C_frame(&fr); + } else { + break; + } } - if (n == 0) { - return fr.pc(); - } else { - return NULL; + num_of_frames = frame_idx; + for (; frame_idx < frames; frame_idx ++) { + stack[frame_idx] = NULL; } + + return num_of_frames; +} + + +bool os::unsetenv(const char* name) { + assert(name != NULL, "Null pointer"); + return (::unsetenv(name) == 0); } int os::get_last_error() {
--- a/hotspot/src/os/solaris/vm/perfMemory_solaris.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/os/solaris/vm/perfMemory_solaris.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -770,7 +770,8 @@ (void)::memset((void*) mapAddress, 0, size); // it does not go through os api, the operation has to record from here - MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC); + MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, + size, CURRENT_PC, mtInternal); return mapAddress; } @@ -941,7 +942,8 @@ } // it does not go through os api, the operation has to record from here - MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC); + MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, + size, CURRENT_PC, mtInternal); *addr = mapAddress; *sizep = size;
--- a/hotspot/src/os/windows/vm/os_windows.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/os/windows/vm/os_windows.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -138,9 +138,8 @@ // Workaround for issue when a custom launcher doesn't call // DestroyJavaVM and NMT is trying to track memory when free is // called from a static destructor - if (MemTracker::is_on()) { - MemTracker::shutdown(MemTracker::NMT_normal); - } + MemTracker::shutdown(); + break; default: break; @@ -163,6 +162,10 @@ return result > 0 && result < len; } +bool os::unsetenv(const char* name) { + assert(name != NULL, "Null pointer"); + return (SetEnvironmentVariable(name, NULL) == TRUE); +} // No setuid programs under Windows. bool os::have_special_privileges() { @@ -319,15 +322,16 @@ * So far, this method is only used by Native Memory Tracking, which is * only supported on Windows XP or later. */ -address os::get_caller_pc(int n) { +int os::get_native_stack(address* stack, int frames, int toSkip) { #ifdef _NMT_NOINLINE_ - n++; + toSkip ++; #endif - address pc; - if (os::Kernel32Dll::RtlCaptureStackBackTrace(n + 1, 1, (PVOID*)&pc, NULL) == 1) { - return pc; - } - return NULL; + int captured = Kernel32Dll::RtlCaptureStackBackTrace(toSkip + 1, frames, + (PVOID*)stack, NULL); + for (int index = captured; index < frames; index ++) { + stack[index] = NULL; + } + return captured; } @@ -2901,7 +2905,7 @@ PAGE_READWRITE); // If reservation failed, return NULL if (p_buf == NULL) return NULL; - MemTracker::record_virtual_memory_reserve((address)p_buf, size_of_reserve, mtNone, CALLER_PC); + MemTracker::record_virtual_memory_reserve((address)p_buf, size_of_reserve, CALLER_PC); os::release_memory(p_buf, bytes + chunk_size); // we still need to round up to a page boundary (in case we are using large pages) @@ -2967,7 +2971,7 @@ // need to create a dummy 'reserve' record to match // the release. MemTracker::record_virtual_memory_reserve((address)p_buf, - bytes_to_release, mtNone, CALLER_PC); + bytes_to_release, CALLER_PC); os::release_memory(p_buf, bytes_to_release); } #ifdef ASSERT @@ -2986,11 +2990,10 @@ } // Although the memory is allocated individually, it is returned as one. // NMT records it as one block. - address pc = CALLER_PC; if ((flags & MEM_COMMIT) != 0) { - MemTracker::record_virtual_memory_reserve_and_commit((address)p_buf, bytes, mtNone, pc); + MemTracker::record_virtual_memory_reserve_and_commit((address)p_buf, bytes, CALLER_PC); } else { - MemTracker::record_virtual_memory_reserve((address)p_buf, bytes, mtNone, pc); + MemTracker::record_virtual_memory_reserve((address)p_buf, bytes, CALLER_PC); } // made it this far, success @@ -3188,8 +3191,7 @@ DWORD flag = MEM_RESERVE | MEM_COMMIT | MEM_LARGE_PAGES; char * res = (char *)VirtualAlloc(addr, bytes, flag, prot); if (res != NULL) { - address pc = CALLER_PC; - MemTracker::record_virtual_memory_reserve_and_commit((address)res, bytes, mtNone, pc); + MemTracker::record_virtual_memory_reserve_and_commit((address)res, bytes, CALLER_PC); } return res;
--- a/hotspot/src/os/windows/vm/perfMemory_windows.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/os/windows/vm/perfMemory_windows.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2001, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2001, 2014, 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 @@ -29,6 +29,7 @@ #include "oops/oop.inline.hpp" #include "os_windows.inline.hpp" #include "runtime/handles.inline.hpp" +#include "runtime/os.hpp" #include "runtime/perfMemory.hpp" #include "services/memTracker.hpp" #include "utilities/exceptions.hpp" @@ -1388,7 +1389,7 @@ // the file has been successfully created and the file mapping // object has been created. sharedmem_fileHandle = fh; - sharedmem_fileName = strdup(filename); + sharedmem_fileName = os::strdup(filename); return fmh; } @@ -1498,7 +1499,8 @@ (void)memset(mapAddress, '\0', size); // it does not go through os api, the operation has to record from here - MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC); + MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, + size, CURRENT_PC, mtInternal); return (char*) mapAddress; } @@ -1680,7 +1682,8 @@ } // it does not go through os api, the operation has to record from here - MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC); + MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, + CURRENT_PC, mtInternal); *addrp = (char*)mapAddress; @@ -1834,10 +1837,14 @@ return; } - MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); - remove_file_mapping(addr); - // it does not go through os api, the operation has to record from here - tkr.record((address)addr, bytes); + if (MemTracker::tracking_level() > NMT_minimal) { + // it does not go through os api, the operation has to record from here + Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); + remove_file_mapping(addr); + tkr.record((address)addr, bytes); + } else { + remove_file_mapping(addr); + } } char* PerfMemory::backing_store_filename() {
--- a/hotspot/src/os_cpu/solaris_sparc/vm/vm_version_solaris_sparc.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/os_cpu/solaris_sparc/vm/vm_version_solaris_sparc.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -23,6 +23,8 @@ */ #include "precompiled.hpp" +#include "memory/allocation.hpp" +#include "memory/allocation.inline.hpp" #include "runtime/os.hpp" #include "vm_version_sparc.hpp" @@ -48,7 +50,7 @@ // All SI defines used below must be supported. guarantee(bufsize != -1, "must be supported"); - char* buf = (char*) malloc(bufsize); + char* buf = (char*) os::malloc(bufsize, mtInternal); if (buf == NULL) return; @@ -60,7 +62,7 @@ } } - free(buf); + os::free(buf); } int VM_Version::platform_features(int features) { @@ -161,7 +163,7 @@ char tmp; size_t bufsize = sysinfo(SI_ISALIST, &tmp, 1); - char* buf = (char*) malloc(bufsize); + char* buf = (char*) os::malloc(bufsize, mtInternal); if (buf != NULL) { if (sysinfo(SI_ISALIST, buf, bufsize) == bufsize) { @@ -184,7 +186,7 @@ if (vis[3] == '2') features |= vis2_instructions_m; } } - free(buf); + os::free(buf); } } @@ -228,7 +230,7 @@ } #endif // Convert to UPPER case before compare. - char* impl = strdup(implementation); + char* impl = os::strdup_check_oom(implementation); for (int i = 0; impl[i] != 0; i++) impl[i] = (char)toupper((uint)impl[i]); @@ -252,7 +254,7 @@ implementation = "SPARC"; } } - free((void*)impl); + os::free((void*)impl); break; } } // for(
--- a/hotspot/src/share/vm/asm/codeBuffer.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/asm/codeBuffer.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -269,7 +269,7 @@ GrowableArray<int>* CodeBuffer::create_patch_overflow() { if (_overflow_arena == NULL) { - _overflow_arena = new (mtCode) Arena(); + _overflow_arena = new (mtCode) Arena(mtCode); } return new (_overflow_arena) GrowableArray<int>(_overflow_arena, 8, 0, 0); }
--- a/hotspot/src/share/vm/c1/c1_Compiler.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/c1/c1_Compiler.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -48,7 +48,7 @@ void Compiler::init_c1_runtime() { BufferBlob* buffer_blob = CompilerThread::current()->get_buffer_blob(); - Arena* arena = new (mtCompiler) Arena(); + Arena* arena = new (mtCompiler) Arena(mtCompiler); Runtime1::initialize(buffer_blob); FrameMap::initialize(); // initialize data structures
--- a/hotspot/src/share/vm/c1/c1_LIRAssembler.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/c1/c1_LIRAssembler.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -30,6 +30,7 @@ #include "c1/c1_MacroAssembler.hpp" #include "c1/c1_ValueStack.hpp" #include "ci/ciInstance.hpp" +#include "runtime/os.hpp" void LIR_Assembler::patching_epilog(PatchingStub* patch, LIR_PatchCode patch_code, Register obj, CodeEmitInfo* info) { // we must have enough patching space so that call can be inserted @@ -848,7 +849,7 @@ stringStream st; st.print("bad oop %s at %d", r->as_Register()->name(), _masm->offset()); #ifdef SPARC - _masm->_verify_oop(r->as_Register(), strdup(st.as_string()), __FILE__, __LINE__); + _masm->_verify_oop(r->as_Register(), os::strdup(st.as_string(), mtCompiler), __FILE__, __LINE__); #else _masm->verify_oop(r->as_Register()); #endif
--- a/hotspot/src/share/vm/ci/ciEnv.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/ci/ciEnv.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -86,7 +86,8 @@ // ------------------------------------------------------------------ // ciEnv::ciEnv -ciEnv::ciEnv(CompileTask* task, int system_dictionary_modification_counter) { +ciEnv::ciEnv(CompileTask* task, int system_dictionary_modification_counter) + : _ciEnv_arena(mtCompiler) { VM_ENTRY_MARK; // Set up ciEnv::current immediately, for the sake of ciObjectFactory, etc. @@ -144,7 +145,7 @@ _jvmti_can_pop_frame = false; } -ciEnv::ciEnv(Arena* arena) { +ciEnv::ciEnv(Arena* arena) : _ciEnv_arena(mtCompiler) { ASSERT_IN_VM; // Set up ciEnv::current immediately, for the sake of ciObjectFactory, etc.
--- a/hotspot/src/share/vm/ci/ciObjectFactory.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/ci/ciObjectFactory.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1999, 2014, 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 @@ -112,7 +112,7 @@ // This Arena is long lived and exists in the resource mark of the // compiler thread that initializes the initial ciObjectFactory which // creates the shared ciObjects that all later ciObjectFactories use. - Arena* arena = new (mtCompiler) Arena(); + Arena* arena = new (mtCompiler) Arena(mtCompiler); ciEnv initial(arena); ciEnv* env = ciEnv::current(); env->_factory->init_shared_objects();
--- a/hotspot/src/share/vm/classfile/classLoader.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/classfile/classLoader.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -273,13 +273,17 @@ } LazyClassPathEntry::LazyClassPathEntry(char* path, const struct stat* st) : ClassPathEntry() { - _path = strdup(path); + _path = os::strdup_check_oom(path); _st = *st; _meta_index = NULL; _resolved_entry = NULL; _has_error = false; } +LazyClassPathEntry::~LazyClassPathEntry() { + os::free(_path); +} + bool LazyClassPathEntry::is_jar_file() { return ((_st.st_mode & S_IFREG) == S_IFREG); } @@ -416,7 +420,7 @@ default: { if (!skipCurrentJar && cur_entry != NULL) { - char* new_name = strdup(package_name); + char* new_name = os::strdup_check_oom(package_name); boot_class_path_packages.append(new_name); } } @@ -438,7 +442,7 @@ void ClassLoader::setup_bootstrap_search_path() { assert(_first_entry == NULL, "should not setup bootstrap class search path twice"); - char* sys_class_path = os::strdup(Arguments::get_sysclasspath()); + char* sys_class_path = os::strdup_check_oom(Arguments::get_sysclasspath()); if (TraceClassLoading && Verbose) { tty->print_cr("[Bootstrap loader class path=%s]", sys_class_path); } @@ -460,6 +464,7 @@ end++; } } + os::free(sys_class_path); } ClassPathEntry* ClassLoader::create_class_path_entry(char *path, const struct stat* st, bool lazy, TRAPS) {
--- a/hotspot/src/share/vm/classfile/classLoader.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/classfile/classLoader.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -128,6 +128,8 @@ bool is_jar_file(); const char* name() { return _path; } LazyClassPathEntry(char* path, const struct stat* st); + virtual ~LazyClassPathEntry(); + ClassFileStream* open_stream(const char* name, TRAPS); void set_meta_index(MetaIndex* meta_index) { _meta_index = meta_index; } virtual bool is_lazy();
--- a/hotspot/src/share/vm/classfile/symbolTable.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/classfile/symbolTable.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -70,9 +70,9 @@ void SymbolTable::initialize_symbols(int arena_alloc_size) { // Initialize the arena for global symbols, size passed in depends on CDS. if (arena_alloc_size == 0) { - _arena = new (mtSymbol) Arena(); + _arena = new (mtSymbol) Arena(mtSymbol); } else { - _arena = new (mtSymbol) Arena(arena_alloc_size); + _arena = new (mtSymbol) Arena(mtSymbol, arena_alloc_size); } }
--- a/hotspot/src/share/vm/classfile/verifier.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/classfile/verifier.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -2217,6 +2217,181 @@ } } +// Look at the method's handlers. If the bci is in the handler's try block +// then check if the handler_pc is already on the stack. If not, push it. +void ClassVerifier::push_handlers(ExceptionTable* exhandlers, + GrowableArray<u4>* handler_stack, + u4 bci) { + int exlength = exhandlers->length(); + for(int x = 0; x < exlength; x++) { + if (bci >= exhandlers->start_pc(x) && bci < exhandlers->end_pc(x)) { + handler_stack->append_if_missing(exhandlers->handler_pc(x)); + } + } +} + +// Return TRUE if all code paths starting with start_bc_offset end in +// bytecode athrow or loop. +bool ClassVerifier::ends_in_athrow(u4 start_bc_offset) { + ResourceMark rm; + // Create bytecode stream. + RawBytecodeStream bcs(method()); + u4 code_length = method()->code_size(); + bcs.set_start(start_bc_offset); + u4 target; + // Create stack for storing bytecode start offsets for if* and *switch. + GrowableArray<u4>* bci_stack = new GrowableArray<u4>(30); + // Create stack for handlers for try blocks containing this handler. + GrowableArray<u4>* handler_stack = new GrowableArray<u4>(30); + // Create list of visited branch opcodes (goto* and if*). + GrowableArray<u4>* visited_branches = new GrowableArray<u4>(30); + ExceptionTable exhandlers(_method()); + + while (true) { + if (bcs.is_last_bytecode()) { + // if no more starting offsets to parse or if at the end of the + // method then return false. + if ((bci_stack->is_empty()) || ((u4)bcs.end_bci() == code_length)) + return false; + // Pop a bytecode starting offset and scan from there. + bcs.set_start(bci_stack->pop()); + } + Bytecodes::Code opcode = bcs.raw_next(); + u4 bci = bcs.bci(); + + // If the bytecode is in a TRY block, push its handlers so they + // will get parsed. + push_handlers(&exhandlers, handler_stack, bci); + + switch (opcode) { + case Bytecodes::_if_icmpeq: + case Bytecodes::_if_icmpne: + case Bytecodes::_if_icmplt: + case Bytecodes::_if_icmpge: + case Bytecodes::_if_icmpgt: + case Bytecodes::_if_icmple: + case Bytecodes::_ifeq: + case Bytecodes::_ifne: + case Bytecodes::_iflt: + case Bytecodes::_ifge: + case Bytecodes::_ifgt: + case Bytecodes::_ifle: + case Bytecodes::_if_acmpeq: + case Bytecodes::_if_acmpne: + case Bytecodes::_ifnull: + case Bytecodes::_ifnonnull: + target = bcs.dest(); + if (visited_branches->contains(bci)) { + if (bci_stack->is_empty()) return true; + // Pop a bytecode starting offset and scan from there. + bcs.set_start(bci_stack->pop()); + } else { + if (target > bci) { // forward branch + if (target >= code_length) return false; + // Push the branch target onto the stack. + bci_stack->push(target); + // then, scan bytecodes starting with next. + bcs.set_start(bcs.next_bci()); + } else { // backward branch + // Push bytecode offset following backward branch onto the stack. + bci_stack->push(bcs.next_bci()); + // Check bytecodes starting with branch target. + bcs.set_start(target); + } + // Record target so we don't branch here again. + visited_branches->append(bci); + } + break; + + case Bytecodes::_goto: + case Bytecodes::_goto_w: + target = (opcode == Bytecodes::_goto ? bcs.dest() : bcs.dest_w()); + if (visited_branches->contains(bci)) { + if (bci_stack->is_empty()) return true; + // Been here before, pop new starting offset from stack. + bcs.set_start(bci_stack->pop()); + } else { + if (target >= code_length) return false; + // Continue scanning from the target onward. + bcs.set_start(target); + // Record target so we don't branch here again. + visited_branches->append(bci); + } + break; + + // Check that all switch alternatives end in 'athrow' bytecodes. Since it + // is difficult to determine where each switch alternative ends, parse + // each switch alternative until either hit a 'return', 'athrow', or reach + // the end of the method's bytecodes. This is gross but should be okay + // because: + // 1. tableswitch and lookupswitch byte codes in handlers for ctor explicit + // constructor invocations should be rare. + // 2. if each switch alternative ends in an athrow then the parsing should be + // short. If there is no athrow then it is bogus code, anyway. + case Bytecodes::_lookupswitch: + case Bytecodes::_tableswitch: + { + address aligned_bcp = (address) round_to((intptr_t)(bcs.bcp() + 1), jintSize); + u4 default_offset = Bytes::get_Java_u4(aligned_bcp) + bci; + int keys, delta; + if (opcode == Bytecodes::_tableswitch) { + jint low = (jint)Bytes::get_Java_u4(aligned_bcp + jintSize); + jint high = (jint)Bytes::get_Java_u4(aligned_bcp + 2*jintSize); + // This is invalid, but let the regular bytecode verifier + // report this because the user will get a better error message. + if (low > high) return true; + keys = high - low + 1; + delta = 1; + } else { + keys = (int)Bytes::get_Java_u4(aligned_bcp + jintSize); + delta = 2; + } + // Invalid, let the regular bytecode verifier deal with it. + if (keys < 0) return true; + + // Push the offset of the next bytecode onto the stack. + bci_stack->push(bcs.next_bci()); + + // Push the switch alternatives onto the stack. + for (int i = 0; i < keys; i++) { + u4 target = bci + (jint)Bytes::get_Java_u4(aligned_bcp+(3+i*delta)*jintSize); + if (target > code_length) return false; + bci_stack->push(target); + } + + // Start bytecode parsing for the switch at the default alternative. + if (default_offset > code_length) return false; + bcs.set_start(default_offset); + break; + } + + case Bytecodes::_return: + return false; + + case Bytecodes::_athrow: + { + if (bci_stack->is_empty()) { + if (handler_stack->is_empty()) { + return true; + } else { + // Parse the catch handlers for try blocks containing athrow. + bcs.set_start(handler_stack->pop()); + } + } else { + // Pop a bytecode offset and starting scanning from there. + bcs.set_start(bci_stack->pop()); + } + } + break; + + default: + ; + } // end switch + } // end while loop + + return false; +} + void ClassVerifier::verify_invoke_init( RawBytecodeStream* bcs, u2 ref_class_index, VerificationType ref_class_type, StackMapFrame* current_frame, u4 code_length, bool *this_uninit, @@ -2236,18 +2411,26 @@ return; } - // Make sure that this call is not done from within a TRY block because - // that can result in returning an incomplete object. Simply checking - // (bci >= start_pc) also ensures that this call is not done after a TRY - // block. That is also illegal because this call must be the first Java - // statement in the constructor. + // Check if this call is done from inside of a TRY block. If so, make + // sure that all catch clause paths end in a throw. Otherwise, this + // can result in returning an incomplete object. ExceptionTable exhandlers(_method()); int exlength = exhandlers.length(); for(int i = 0; i < exlength; i++) { - if (bci >= exhandlers.start_pc(i)) { - verify_error(ErrorContext::bad_code(bci), - "Bad <init> method call from after the start of a try block"); - return; + u2 start_pc = exhandlers.start_pc(i); + u2 end_pc = exhandlers.end_pc(i); + + if (bci >= start_pc && bci < end_pc) { + if (!ends_in_athrow(exhandlers.handler_pc(i))) { + verify_error(ErrorContext::bad_code(bci), + "Bad <init> method call from after the start of a try block"); + return; + } else if (VerboseVerification) { + ResourceMark rm; + tty->print_cr( + "Survived call to ends_in_athrow(): %s", + current_class()->name()->as_C_string()); + } } }
--- a/hotspot/src/share/vm/classfile/verifier.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/classfile/verifier.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -30,6 +30,7 @@ #include "oops/klass.hpp" #include "oops/method.hpp" #include "runtime/handles.hpp" +#include "utilities/growableArray.hpp" #include "utilities/exceptions.hpp" // The verifier class @@ -303,6 +304,16 @@ StackMapFrame* current_frame, u4 code_length, bool* this_uninit, constantPoolHandle cp, TRAPS); + // Used by ends_in_athrow() to push all handlers that contain bci onto + // the handler_stack, if the handler is not already on the stack. + void push_handlers(ExceptionTable* exhandlers, + GrowableArray<u4>* handler_stack, + u4 bci); + + // Returns true if all paths starting with start_bc_offset end in athrow + // bytecode or loop. + bool ends_in_athrow(u4 start_bc_offset); + void verify_invoke_instructions( RawBytecodeStream* bcs, u4 code_length, StackMapFrame* current_frame, bool* this_uninit, VerificationType return_type,
--- a/hotspot/src/share/vm/compiler/compilerOracle.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/compiler/compilerOracle.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -33,6 +33,7 @@ #include "oops/symbol.hpp" #include "runtime/handles.inline.hpp" #include "runtime/jniHandles.hpp" +#include "runtime/os.hpp" class MethodMatcher : public CHeapObj<mtCompiler> { public: @@ -175,7 +176,11 @@ Symbol* method_name, Mode method_mode, Symbol* signature, const char * opt, MethodMatcher* next): MethodMatcher(class_name, class_mode, method_name, method_mode, signature, next) { - option = opt; + option = os::strdup_check_oom(opt); + } + + virtual ~MethodOptionMatcher() { + os::free((void*)option); } bool match(methodHandle method, const char* opt) { @@ -498,7 +503,7 @@ tty->print("CompilerOracle: %s ", command_names[command]); match->print(); } - match = add_option_string(c_name, c_match, m_name, m_match, signature, strdup(option)); + match = add_option_string(c_name, c_match, m_name, m_match, signature, option); line += bytes_read; } } else {
--- a/hotspot/src/share/vm/gc_implementation/concurrentMarkSweep/cmsCollectorPolicy.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/gc_implementation/concurrentMarkSweep/cmsCollectorPolicy.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2007, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2007, 2014, 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 @@ -52,7 +52,8 @@ } void ConcurrentMarkSweepPolicy::initialize_generations() { - _generations = NEW_C_HEAP_ARRAY3(GenerationSpecPtr, number_of_generations(), mtGC, 0, AllocFailStrategy::RETURN_NULL); + _generations = NEW_C_HEAP_ARRAY3(GenerationSpecPtr, number_of_generations(), mtGC, + CURRENT_PC, AllocFailStrategy::RETURN_NULL); if (_generations == NULL) vm_exit_during_initialization("Unable to allocate gen spec");
--- a/hotspot/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -289,7 +289,7 @@ } _fine_grain_regions = NEW_C_HEAP_ARRAY3(PerRegionTablePtr, _max_fine_entries, - mtGC, 0, AllocFailStrategy::RETURN_NULL); + mtGC, CURRENT_PC, AllocFailStrategy::RETURN_NULL); if (_fine_grain_regions == NULL) { vm_exit_out_of_memory(sizeof(void*)*_max_fine_entries, OOM_MALLOC_ERROR,
--- a/hotspot/src/share/vm/interpreter/abstractInterpreter.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/interpreter/abstractInterpreter.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -280,9 +280,6 @@ address generate_result_handler_for(BasicType type); address generate_slow_signature_handler(); - // entry point generator - address generate_method_entry(AbstractInterpreter::MethodKind kind); - void bang_stack_shadow_pages(bool native_call); void generate_all();
--- a/hotspot/src/share/vm/interpreter/cppInterpreter.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/interpreter/cppInterpreter.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, 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 @@ -108,7 +108,7 @@ } -#define method_entry(kind) Interpreter::_entry_table[Interpreter::kind] = generate_method_entry(Interpreter::kind) +#define method_entry(kind) Interpreter::_entry_table[Interpreter::kind] = ((InterpreterGenerator*)this)->generate_method_entry(Interpreter::kind) { CodeletMark cm(_masm, "(kind = frame_manager)"); // all non-native method kinds
--- a/hotspot/src/share/vm/interpreter/interpreter.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/interpreter/interpreter.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -29,6 +29,7 @@ #include "interpreter/bytecodeHistogram.hpp" #include "interpreter/bytecodeInterpreter.hpp" #include "interpreter/interpreter.hpp" +#include "interpreter/interpreterGenerator.hpp" #include "interpreter/interpreterRuntime.hpp" #include "interpreter/interp_masm.hpp" #include "interpreter/templateTable.hpp" @@ -261,7 +262,7 @@ // Special intrinsic method? // Note: This test must come _after_ the test for native methods, // otherwise we will run into problems with JDK 1.2, see also - // AbstractInterpreterGenerator::generate_method_entry() for + // InterpreterGenerator::generate_method_entry() for // for details. switch (m->intrinsic_id()) { case vmIntrinsics::_dsin : return java_lang_math_sin ; @@ -521,3 +522,50 @@ Interpreter::_entry_table[kind] = Interpreter::_entry_table[Interpreter::abstract]; } } + +// Generate method entries +address InterpreterGenerator::generate_method_entry( + AbstractInterpreter::MethodKind kind) { + // determine code generation flags + bool synchronized = false; + address entry_point = NULL; + + switch (kind) { + case Interpreter::zerolocals : break; + case Interpreter::zerolocals_synchronized: synchronized = true; break; + case Interpreter::native : entry_point = generate_native_entry(false); break; + case Interpreter::native_synchronized : entry_point = generate_native_entry(true); break; + case Interpreter::empty : entry_point = generate_empty_entry(); break; + case Interpreter::accessor : entry_point = generate_accessor_entry(); break; + case Interpreter::abstract : entry_point = generate_abstract_entry(); break; + + case Interpreter::java_lang_math_sin : // fall thru + case Interpreter::java_lang_math_cos : // fall thru + case Interpreter::java_lang_math_tan : // fall thru + case Interpreter::java_lang_math_abs : // fall thru + case Interpreter::java_lang_math_log : // fall thru + case Interpreter::java_lang_math_log10 : // fall thru + case Interpreter::java_lang_math_sqrt : // fall thru + case Interpreter::java_lang_math_pow : // fall thru + case Interpreter::java_lang_math_exp : entry_point = generate_math_entry(kind); break; + case Interpreter::java_lang_ref_reference_get + : entry_point = generate_Reference_get_entry(); break; +#ifndef CC_INTERP + case Interpreter::java_util_zip_CRC32_update + : entry_point = generate_CRC32_update_entry(); break; + case Interpreter::java_util_zip_CRC32_updateBytes + : // fall thru + case Interpreter::java_util_zip_CRC32_updateByteBuffer + : entry_point = generate_CRC32_updateBytes_entry(kind); break; +#endif // CC_INTERP + default: + fatal(err_msg("unexpected method kind: %d", kind)); + break; + } + + if (entry_point) { + return entry_point; + } + + return generate_normal_entry(synchronized); +}
--- a/hotspot/src/share/vm/interpreter/interpreterGenerator.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/interpreter/interpreterGenerator.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, 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 @@ -37,9 +37,11 @@ class InterpreterGenerator: public CC_INTERP_ONLY(CppInterpreterGenerator) NOT_CC_INTERP(TemplateInterpreterGenerator) { -public: + public: -InterpreterGenerator(StubQueue* _code); + InterpreterGenerator(StubQueue* _code); + // entry point generator + address generate_method_entry(AbstractInterpreter::MethodKind kind); #ifdef TARGET_ARCH_x86 # include "interpreterGenerator_x86.hpp"
--- a/hotspot/src/share/vm/interpreter/templateInterpreter.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/interpreter/templateInterpreter.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -364,7 +364,7 @@ #define method_entry(kind) \ { CodeletMark cm(_masm, "method entry point (kind = " #kind ")"); \ - Interpreter::_entry_table[Interpreter::kind] = generate_method_entry(Interpreter::kind); \ + Interpreter::_entry_table[Interpreter::kind] = ((InterpreterGenerator*)this)->generate_method_entry(Interpreter::kind); \ } // all non-native method kinds
--- a/hotspot/src/share/vm/interpreter/templateInterpreterGenerator.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/interpreter/templateInterpreterGenerator.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, 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 @@ -59,9 +59,6 @@ address generate_safept_entry_for(TosState state, address runtime_entry); void generate_throw_exception(); - // entry point generator -// address generate_method_entry(AbstractInterpreter::MethodKind kind); - // Instruction generation void generate_and_dispatch (Template* t, TosState tos_out = ilgl); void set_vtos_entry_points (Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep);
--- a/hotspot/src/share/vm/memory/allocation.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/memory/allocation.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -422,26 +422,23 @@ } //------------------------------Arena------------------------------------------ -NOT_PRODUCT(volatile jint Arena::_instance_count = 0;) -Arena::Arena(size_t init_size) { +Arena::Arena(MEMFLAGS flag, size_t init_size) : _flags(flag), _size_in_bytes(0) { size_t round_size = (sizeof (char *)) - 1; init_size = (init_size+round_size) & ~round_size; _first = _chunk = new (AllocFailStrategy::EXIT_OOM, init_size) Chunk(init_size); _hwm = _chunk->bottom(); // Save the cached hwm, max _max = _chunk->top(); - _size_in_bytes = 0; + MemTracker::record_new_arena(flag); set_size_in_bytes(init_size); - NOT_PRODUCT(Atomic::inc(&_instance_count);) } -Arena::Arena() { +Arena::Arena(MEMFLAGS flag) : _flags(flag), _size_in_bytes(0) { _first = _chunk = new (AllocFailStrategy::EXIT_OOM, Chunk::init_size) Chunk(Chunk::init_size); _hwm = _chunk->bottom(); // Save the cached hwm, max _max = _chunk->top(); - _size_in_bytes = 0; + MemTracker::record_new_arena(flag); set_size_in_bytes(Chunk::init_size); - NOT_PRODUCT(Atomic::inc(&_instance_count);) } Arena *Arena::move_contents(Arena *copy) { @@ -463,7 +460,7 @@ Arena::~Arena() { destruct_contents(); - NOT_PRODUCT(Atomic::dec(&_instance_count);) + MemTracker::record_arena_free(_flags); } void* Arena::operator new(size_t size) throw() { @@ -479,21 +476,21 @@ // dynamic memory type binding void* Arena::operator new(size_t size, MEMFLAGS flags) throw() { #ifdef ASSERT - void* p = (void*)AllocateHeap(size, flags|otArena, CALLER_PC); + void* p = (void*)AllocateHeap(size, flags, CALLER_PC); if (PrintMallocFree) trace_heap_malloc(size, "Arena-new", p); return p; #else - return (void *) AllocateHeap(size, flags|otArena, CALLER_PC); + return (void *) AllocateHeap(size, flags, CALLER_PC); #endif } void* Arena::operator new(size_t size, const std::nothrow_t& nothrow_constant, MEMFLAGS flags) throw() { #ifdef ASSERT - void* p = os::malloc(size, flags|otArena, CALLER_PC); + void* p = os::malloc(size, flags, CALLER_PC); if (PrintMallocFree) trace_heap_malloc(size, "Arena-new", p); return p; #else - return os::malloc(size, flags|otArena, CALLER_PC); + return os::malloc(size, flags, CALLER_PC); #endif } @@ -518,8 +515,9 @@ // change the size void Arena::set_size_in_bytes(size_t size) { if (_size_in_bytes != size) { + long delta = (long)(size - size_in_bytes()); _size_in_bytes = size; - MemTracker::record_arena_size((address)this, size); + MemTracker::record_arena_size_change(delta, _flags); } }
--- a/hotspot/src/share/vm/memory/allocation.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/memory/allocation.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, 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 @@ -133,51 +133,34 @@ /* - * MemoryType bitmap layout: - * | 16 15 14 13 12 11 10 09 | 08 07 06 05 | 04 03 02 01 | - * | memory type | object | reserved | - * | | type | | + * Memory types */ enum MemoryType { // Memory type by sub systems. It occupies lower byte. - mtNone = 0x0000, // undefined - mtClass = 0x0100, // memory class for Java classes - mtThread = 0x0200, // memory for thread objects - mtThreadStack = 0x0300, - mtCode = 0x0400, // memory for generated code - mtGC = 0x0500, // memory for GC - mtCompiler = 0x0600, // memory for compiler - mtInternal = 0x0700, // memory used by VM, but does not belong to + mtJavaHeap = 0x00, // Java heap + mtClass = 0x01, // memory class for Java classes + mtThread = 0x02, // memory for thread objects + mtThreadStack = 0x03, + mtCode = 0x04, // memory for generated code + mtGC = 0x05, // memory for GC + mtCompiler = 0x06, // memory for compiler + mtInternal = 0x07, // memory used by VM, but does not belong to // any of above categories, and not used for // native memory tracking - mtOther = 0x0800, // memory not used by VM - mtSymbol = 0x0900, // symbol - mtNMT = 0x0A00, // memory used by native memory tracking - mtChunk = 0x0B00, // chunk that holds content of arenas - mtJavaHeap = 0x0C00, // Java heap - mtClassShared = 0x0D00, // class data sharing - mtTest = 0x0E00, // Test type for verifying NMT - mtTracing = 0x0F00, // memory used for Tracing - mt_number_of_types = 0x000F, // number of memory types (mtDontTrack + mtOther = 0x08, // memory not used by VM + mtSymbol = 0x09, // symbol + mtNMT = 0x0A, // memory used by native memory tracking + mtClassShared = 0x0B, // class data sharing + mtChunk = 0x0C, // chunk that holds content of arenas + mtTest = 0x0D, // Test type for verifying NMT + mtTracing = 0x0E, // memory used for Tracing + mtNone = 0x0F, // undefined + mt_number_of_types = 0x10 // number of memory types (mtDontTrack // is not included as validate type) - mtDontTrack = 0x0F00, // memory we do not or cannot track - mt_masks = 0x7F00, - - // object type mask - otArena = 0x0010, // an arena object - otNMTRecorder = 0x0020, // memory recorder object - ot_masks = 0x00F0 }; -#define IS_MEMORY_TYPE(flags, type) ((flags & mt_masks) == type) -#define HAS_VALID_MEMORY_TYPE(flags)((flags & mt_masks) != mtNone) -#define FLAGS_TO_MEMORY_TYPE(flags) (flags & mt_masks) +typedef MemoryType MEMFLAGS; -#define IS_ARENA_OBJ(flags) ((flags & ot_masks) == otArena) -#define IS_NMT_RECORDER(flags) ((flags & ot_masks) == otNMTRecorder) -#define NMT_CAN_TRACK(flags) (!IS_NMT_RECORDER(flags) && !(IS_MEMORY_TYPE(flags, mtDontTrack))) - -typedef unsigned short MEMFLAGS; #if INCLUDE_NMT @@ -189,27 +172,23 @@ #endif // INCLUDE_NMT -// debug build does not inline -#if defined(_NMT_NOINLINE_) - #define CURRENT_PC (NMT_track_callsite ? os::get_caller_pc(1) : 0) - #define CALLER_PC (NMT_track_callsite ? os::get_caller_pc(2) : 0) - #define CALLER_CALLER_PC (NMT_track_callsite ? os::get_caller_pc(3) : 0) -#else - #define CURRENT_PC (NMT_track_callsite? os::get_caller_pc(0) : 0) - #define CALLER_PC (NMT_track_callsite ? os::get_caller_pc(1) : 0) - #define CALLER_CALLER_PC (NMT_track_callsite ? os::get_caller_pc(2) : 0) -#endif - +class NativeCallStack; template <MEMFLAGS F> class CHeapObj ALLOCATION_SUPER_CLASS_SPEC { public: - _NOINLINE_ void* operator new(size_t size, address caller_pc = 0) throw(); + _NOINLINE_ void* operator new(size_t size, const NativeCallStack& stack) throw(); + _NOINLINE_ void* operator new(size_t size) throw(); _NOINLINE_ void* operator new (size_t size, const std::nothrow_t& nothrow_constant, - address caller_pc = 0) throw(); - _NOINLINE_ void* operator new [](size_t size, address caller_pc = 0) throw(); + const NativeCallStack& stack) throw(); + _NOINLINE_ void* operator new (size_t size, const std::nothrow_t& nothrow_constant) + throw(); + _NOINLINE_ void* operator new [](size_t size, const NativeCallStack& stack) throw(); + _NOINLINE_ void* operator new [](size_t size) throw(); _NOINLINE_ void* operator new [](size_t size, const std::nothrow_t& nothrow_constant, - address caller_pc = 0) throw(); + const NativeCallStack& stack) throw(); + _NOINLINE_ void* operator new [](size_t size, const std::nothrow_t& nothrow_constant) + throw(); void operator delete(void* p); void operator delete [] (void* p); }; @@ -384,13 +363,15 @@ //------------------------------Arena------------------------------------------ // Fast allocation of memory -class Arena : public CHeapObj<mtNone|otArena> { +class Arena : public CHeapObj<mtNone> { protected: friend class ResourceMark; friend class HandleMark; friend class NoHandleMark; friend class VMStructs; + MEMFLAGS _flags; // Memory tracking flags + Chunk *_first; // First chunk Chunk *_chunk; // current chunk char *_hwm, *_max; // High water mark and max in current chunk @@ -418,8 +399,8 @@ } public: - Arena(); - Arena(size_t init_size); + Arena(MEMFLAGS memflag); + Arena(MEMFLAGS memflag, size_t init_size); ~Arena(); void destruct_contents(); char* hwm() const { return _hwm; } @@ -518,8 +499,6 @@ static void free_malloced_objects(Chunk* chunk, char* hwm, char* max, char* hwm2) PRODUCT_RETURN; static void free_all(char** start, char** end) PRODUCT_RETURN; - // how many arena instances - NOT_PRODUCT(static volatile jint _instance_count;) private: // Reset this Arena to empty, access will trigger grow if necessary void reset(void) { @@ -681,7 +660,7 @@ NEW_C_HEAP_ARRAY3(type, (size), memflags, pc, AllocFailStrategy::RETURN_NULL) #define NEW_C_HEAP_ARRAY_RETURN_NULL(type, size, memflags)\ - NEW_C_HEAP_ARRAY3(type, (size), memflags, (address)0, AllocFailStrategy::RETURN_NULL) + NEW_C_HEAP_ARRAY3(type, (size), memflags, CURRENT_PC, AllocFailStrategy::RETURN_NULL) #define REALLOC_C_HEAP_ARRAY(type, old, size, memflags)\ (type*) (ReallocateHeap((char*)(old), (size) * sizeof(type), memflags))
--- a/hotspot/src/share/vm/memory/allocation.inline.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/memory/allocation.inline.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, 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 @@ -27,6 +27,7 @@ #include "runtime/atomic.inline.hpp" #include "runtime/os.hpp" +#include "services/memTracker.hpp" // Explicit C-heap memory management @@ -49,12 +50,10 @@ #endif // allocate using malloc; will fail if no memory available -inline char* AllocateHeap(size_t size, MEMFLAGS flags, address pc = 0, +inline char* AllocateHeap(size_t size, MEMFLAGS flags, + const NativeCallStack& stack, AllocFailType alloc_failmode = AllocFailStrategy::EXIT_OOM) { - if (pc == 0) { - pc = CURRENT_PC; - } - char* p = (char*) os::malloc(size, flags, pc); + char* p = (char*) os::malloc(size, flags, stack); #ifdef ASSERT if (PrintMallocFree) trace_heap_malloc(size, "AllocateHeap", p); #endif @@ -63,10 +62,14 @@ } return p; } +inline char* AllocateHeap(size_t size, MEMFLAGS flags, + AllocFailType alloc_failmode = AllocFailStrategy::EXIT_OOM) { + return AllocateHeap(size, flags, CURRENT_PC, alloc_failmode); +} -inline char* ReallocateHeap(char *old, size_t size, MEMFLAGS flags, +inline char* ReallocateHeap(char *old, size_t size, MEMFLAGS flag, AllocFailType alloc_failmode = AllocFailStrategy::EXIT_OOM) { - char* p = (char*) os::realloc(old, size, flags, CURRENT_PC); + char* p = (char*) os::realloc(old, size, flag, CURRENT_PC); #ifdef ASSERT if (PrintMallocFree) trace_heap_malloc(size, "ReallocateHeap", p); #endif @@ -85,8 +88,22 @@ template <MEMFLAGS F> void* CHeapObj<F>::operator new(size_t size, - address caller_pc) throw() { - void* p = (void*)AllocateHeap(size, F, (caller_pc != 0 ? caller_pc : CALLER_PC)); + const NativeCallStack& stack) throw() { + void* p = (void*)AllocateHeap(size, F, stack); +#ifdef ASSERT + if (PrintMallocFree) trace_heap_malloc(size, "CHeapObj-new", p); +#endif + return p; +} + +template <MEMFLAGS F> void* CHeapObj<F>::operator new(size_t size) throw() { + return CHeapObj<F>::operator new(size, CALLER_PC); +} + +template <MEMFLAGS F> void* CHeapObj<F>::operator new (size_t size, + const std::nothrow_t& nothrow_constant, const NativeCallStack& stack) throw() { + void* p = (void*)AllocateHeap(size, F, stack, + AllocFailStrategy::RETURN_NULL); #ifdef ASSERT if (PrintMallocFree) trace_heap_malloc(size, "CHeapObj-new", p); #endif @@ -94,23 +111,28 @@ } template <MEMFLAGS F> void* CHeapObj<F>::operator new (size_t size, - const std::nothrow_t& nothrow_constant, address caller_pc) throw() { - void* p = (void*)AllocateHeap(size, F, (caller_pc != 0 ? caller_pc : CALLER_PC), - AllocFailStrategy::RETURN_NULL); -#ifdef ASSERT - if (PrintMallocFree) trace_heap_malloc(size, "CHeapObj-new", p); -#endif - return p; + const std::nothrow_t& nothrow_constant) throw() { + return CHeapObj<F>::operator new(size, nothrow_constant, CALLER_PC); } template <MEMFLAGS F> void* CHeapObj<F>::operator new [](size_t size, - address caller_pc) throw() { - return CHeapObj<F>::operator new(size, caller_pc); + const NativeCallStack& stack) throw() { + return CHeapObj<F>::operator new(size, stack); +} + +template <MEMFLAGS F> void* CHeapObj<F>::operator new [](size_t size) + throw() { + return CHeapObj<F>::operator new(size, CALLER_PC); } template <MEMFLAGS F> void* CHeapObj<F>::operator new [](size_t size, - const std::nothrow_t& nothrow_constant, address caller_pc) throw() { - return CHeapObj<F>::operator new(size, nothrow_constant, caller_pc); + const std::nothrow_t& nothrow_constant, const NativeCallStack& stack) throw() { + return CHeapObj<F>::operator new(size, nothrow_constant, stack); +} + +template <MEMFLAGS F> void* CHeapObj<F>::operator new [](size_t size, + const std::nothrow_t& nothrow_constant) throw() { + return CHeapObj<F>::operator new(size, nothrow_constant, CALLER_PC); } template <MEMFLAGS F> void CHeapObj<F>::operator delete(void* p){
--- a/hotspot/src/share/vm/memory/cardTableRS.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/memory/cardTableRS.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -56,7 +56,7 @@ #endif set_bs(_ct_bs); _last_cur_val_in_gen = NEW_C_HEAP_ARRAY3(jbyte, GenCollectedHeap::max_gens + 1, - mtGC, 0, AllocFailStrategy::RETURN_NULL); + mtGC, CURRENT_PC, AllocFailStrategy::RETURN_NULL); if (_last_cur_val_in_gen == NULL) { vm_exit_during_initialization("Could not create last_cur_val_in_gen array."); }
--- a/hotspot/src/share/vm/memory/collectorPolicy.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/memory/collectorPolicy.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -905,7 +905,8 @@ } void MarkSweepPolicy::initialize_generations() { - _generations = NEW_C_HEAP_ARRAY3(GenerationSpecPtr, number_of_generations(), mtGC, 0, AllocFailStrategy::RETURN_NULL); + _generations = NEW_C_HEAP_ARRAY3(GenerationSpecPtr, number_of_generations(), mtGC, CURRENT_PC, + AllocFailStrategy::RETURN_NULL); if (_generations == NULL) { vm_exit_during_initialization("Unable to allocate gen spec"); }
--- a/hotspot/src/share/vm/memory/heapInspection.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/memory/heapInspection.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -135,7 +135,7 @@ _ref = (HeapWord*) Universe::boolArrayKlassObj(); _buckets = (KlassInfoBucket*) AllocateHeap(sizeof(KlassInfoBucket) * _num_buckets, - mtInternal, 0, AllocFailStrategy::RETURN_NULL); + mtInternal, CURRENT_PC, AllocFailStrategy::RETURN_NULL); if (_buckets != NULL) { _size = _num_buckets; for (int index = 0; index < _size; index++) {
--- a/hotspot/src/share/vm/memory/memRegion.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/memory/memRegion.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2000, 2014, 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 @@ -103,11 +103,13 @@ } void* MemRegion::operator new(size_t size) throw() { - return (address)AllocateHeap(size, mtGC, 0, AllocFailStrategy::RETURN_NULL); + return (address)AllocateHeap(size, mtGC, CURRENT_PC, + AllocFailStrategy::RETURN_NULL); } void* MemRegion::operator new [](size_t size) throw() { - return (address)AllocateHeap(size, mtGC, 0, AllocFailStrategy::RETURN_NULL); + return (address)AllocateHeap(size, mtGC, CURRENT_PC, + AllocFailStrategy::RETURN_NULL); } void MemRegion::operator delete(void* p) { FreeHeap(p, mtGC);
--- a/hotspot/src/share/vm/memory/resourceArea.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/memory/resourceArea.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, 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 @@ -49,11 +49,11 @@ debug_only(static int _warned;) // to suppress multiple warnings public: - ResourceArea() { + ResourceArea() : Arena(mtThread) { debug_only(_nesting = 0;) } - ResourceArea(size_t init_size) : Arena(init_size) { + ResourceArea(size_t init_size) : Arena(mtThread, init_size) { debug_only(_nesting = 0;); } @@ -64,7 +64,7 @@ if (UseMallocOnly) { // use malloc, but save pointer in res. area for later freeing char** save = (char**)internal_malloc_4(sizeof(char*)); - return (*save = (char*)os::malloc(size, mtThread)); + return (*save = (char*)os::malloc(size, mtThread, CURRENT_PC)); } #endif return (char*)Amalloc(size, alloc_failmode);
--- a/hotspot/src/share/vm/opto/compile.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/opto/compile.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -647,6 +647,10 @@ _printer(IdealGraphPrinter::printer()), #endif _congraph(NULL), + _comp_arena(mtCompiler), + _node_arena(mtCompiler), + _old_arena(mtCompiler), + _Compile_types(mtCompiler), _replay_inline_data(NULL), _late_inlines(comp_arena(), 2, 0, NULL), _string_late_inlines(comp_arena(), 2, 0, NULL), @@ -954,6 +958,10 @@ _in_dump_cnt(0), _printer(NULL), #endif + _comp_arena(mtCompiler), + _node_arena(mtCompiler), + _old_arena(mtCompiler), + _Compile_types(mtCompiler), _dead_node_list(comp_arena()), _dead_node_count(0), _congraph(NULL),
--- a/hotspot/src/share/vm/opto/runtime.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/opto/runtime.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1381,11 +1381,11 @@ } NamedCounter* c; if (tag == NamedCounter::BiasedLockingCounter) { - c = new BiasedLockingNamedCounter(strdup(st.as_string())); + c = new BiasedLockingNamedCounter(st.as_string()); } else if (tag == NamedCounter::RTMLockingCounter) { - c = new RTMLockingNamedCounter(strdup(st.as_string())); + c = new RTMLockingNamedCounter(st.as_string()); } else { - c = new NamedCounter(strdup(st.as_string()), tag); + c = new NamedCounter(st.as_string(), tag); } // atomically add the new counter to the head of the list. We only
--- a/hotspot/src/share/vm/opto/runtime.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/opto/runtime.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -75,11 +75,17 @@ public: NamedCounter(const char *n, CounterTag tag = NoTag): - _name(n), + _name(n == NULL ? NULL : os::strdup(n)), _count(0), _next(NULL), _tag(tag) {} + ~NamedCounter() { + if (_name != NULL) { + os::free((void*)_name); + } + } + const char * name() const { return _name; } int count() const { return _count; } address addr() { return (address)&_count; }
--- a/hotspot/src/share/vm/opto/type.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/opto/type.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -265,7 +265,7 @@ // locking. Arena* save = current->type_arena(); - Arena* shared_type_arena = new (mtCompiler)Arena(); + Arena* shared_type_arena = new (mtCompiler)Arena(mtCompiler); current->set_type_arena(shared_type_arena); _shared_type_dict =
--- a/hotspot/src/share/vm/precompiled/precompiled.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/precompiled/precompiled.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2010, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2010, 2014, 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 @@ -222,10 +222,17 @@ # include "runtime/vmThread.hpp" # include "runtime/vm_operations.hpp" # include "runtime/vm_version.hpp" +# include "services/allocationSite.hpp" # include "services/lowMemoryDetector.hpp" +# include "services/mallocTracker.hpp" +# include "services/memBaseline.hpp" # include "services/memoryPool.hpp" # include "services/memoryService.hpp" # include "services/memoryUsage.hpp" +# include "services/memReporter.hpp" +# include "services/memTracker.hpp" +# include "services/nmtCommon.hpp" +# include "services/virtualMemoryTracker.hpp" # include "utilities/accessFlags.hpp" # include "utilities/array.hpp" # include "utilities/bitMap.hpp" @@ -240,6 +247,7 @@ # include "utilities/hashtable.hpp" # include "utilities/histogram.hpp" # include "utilities/macros.hpp" +# include "utilities/nativeCallStack.hpp" # include "utilities/numberSeq.hpp" # include "utilities/ostream.hpp" # include "utilities/preserveException.hpp"
--- a/hotspot/src/share/vm/prims/jni.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/prims/jni.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -74,6 +74,7 @@ #include "runtime/signature.hpp" #include "runtime/thread.inline.hpp" #include "runtime/vm_operations.hpp" +#include "services/memTracker.hpp" #include "services/runtimeService.hpp" #include "trace/tracing.hpp" #include "utilities/defaultStream.hpp" @@ -2697,6 +2698,7 @@ if (bad_address != NULL) { os::protect_memory(bad_address, size, os::MEM_PROT_READ, /*is_committed*/false); + MemTracker::record_virtual_memory_type((void*)bad_address, mtInternal); } } return bad_address; @@ -3857,6 +3859,7 @@ void TestKlass_test(); void TestBitMap_test(); void TestAsUtf8(); +void Test_linked_list(); #if INCLUDE_ALL_GCS void TestOldFreeSpaceCalculation_test(); void TestG1BiasedArray_test(); @@ -3887,6 +3890,7 @@ run_unit_test(TestBitMap_test()); run_unit_test(TestAsUtf8()); run_unit_test(ObjectMonitor::sanity_checks()); + run_unit_test(Test_linked_list()); #if INCLUDE_VM_STRUCTS run_unit_test(VMStructs::test()); #endif
--- a/hotspot/src/share/vm/prims/jvmtiManageCapabilities.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/prims/jvmtiManageCapabilities.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2003, 2014, 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 @@ -316,6 +316,7 @@ avail.can_generate_frame_pop_events || avail.can_generate_method_entry_events || avail.can_generate_method_exit_events; +#ifdef ZERO bool enter_all_methods = interp_events || avail.can_generate_breakpoint_events; @@ -324,6 +325,7 @@ UseFastEmptyMethods = false; UseFastAccessorMethods = false; } +#endif // ZERO if (avail.can_generate_breakpoint_events) { RewriteFrequentPairs = false;
--- a/hotspot/src/share/vm/prims/whitebox.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/prims/whitebox.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -52,8 +52,10 @@ #include "gc_implementation/g1/heapRegionRemSet.hpp" #endif // INCLUDE_ALL_GCS -#ifdef INCLUDE_NMT +#if INCLUDE_NMT +#include "services/mallocSiteTable.hpp" #include "services/memTracker.hpp" +#include "utilities/nativeCallStack.hpp" #endif // INCLUDE_NMT #include "compiler/compileBroker.hpp" @@ -255,12 +257,16 @@ // NMT picks it up correctly WB_ENTRY(jlong, WB_NMTMalloc(JNIEnv* env, jobject o, jlong size)) jlong addr = 0; + addr = (jlong)(uintptr_t)os::malloc(size, mtTest); + return addr; +WB_END - if (MemTracker::is_on() && !MemTracker::shutdown_in_progress()) { - addr = (jlong)(uintptr_t)os::malloc(size, mtTest); - } - - return addr; +// Alloc memory with pseudo call stack. The test can create psudo malloc +// allocation site to stress the malloc tracking. +WB_ENTRY(jlong, WB_NMTMallocWithPseudoStack(JNIEnv* env, jobject o, jlong size, jint pseudo_stack)) + address pc = (address)(size_t)pseudo_stack; + NativeCallStack stack(&pc, 1); + return (jlong)os::malloc(size, mtTest, stack); WB_END // Free the memory allocated by NMTAllocTest @@ -271,10 +277,8 @@ WB_ENTRY(jlong, WB_NMTReserveMemory(JNIEnv* env, jobject o, jlong size)) jlong addr = 0; - if (MemTracker::is_on() && !MemTracker::shutdown_in_progress()) { addr = (jlong)(uintptr_t)os::reserve_memory(size); MemTracker::record_virtual_memory_type((address)addr, mtTest); - } return addr; WB_END @@ -293,20 +297,20 @@ os::release_memory((char *)(uintptr_t)addr, size); WB_END -// Block until the current generation of NMT data to be merged, used to reliably test the NMT feature -WB_ENTRY(jboolean, WB_NMTWaitForDataMerge(JNIEnv* env)) - - if (!MemTracker::is_on() || MemTracker::shutdown_in_progress()) { - return false; - } - - return MemTracker::wbtest_wait_for_data_merge(); +WB_ENTRY(jboolean, WB_NMTIsDetailSupported(JNIEnv* env)) + return MemTracker::tracking_level() == NMT_detail; WB_END -WB_ENTRY(jboolean, WB_NMTIsDetailSupported(JNIEnv* env)) - return MemTracker::tracking_level() == MemTracker::NMT_detail; +WB_ENTRY(void, WB_NMTOverflowHashBucket(JNIEnv* env, jobject o, jlong num)) + address pc = (address)1; + for (jlong index = 0; index < num; index ++) { + NativeCallStack stack(&pc, 1); + os::malloc(0, mtTest, stack); + pc += MallocSiteTable::hash_buckets(); + } WB_END + #endif // INCLUDE_NMT static jmethodID reflected_method_to_jmid(JavaThread* thread, JNIEnv* env, jobject method) { @@ -843,12 +847,13 @@ #endif // INCLUDE_ALL_GCS #if INCLUDE_NMT {CC"NMTMalloc", CC"(J)J", (void*)&WB_NMTMalloc }, + {CC"NMTMallocWithPseudoStack", CC"(JI)J", (void*)&WB_NMTMallocWithPseudoStack}, {CC"NMTFree", CC"(J)V", (void*)&WB_NMTFree }, {CC"NMTReserveMemory", CC"(J)J", (void*)&WB_NMTReserveMemory }, {CC"NMTCommitMemory", CC"(JJ)V", (void*)&WB_NMTCommitMemory }, {CC"NMTUncommitMemory", CC"(JJ)V", (void*)&WB_NMTUncommitMemory }, {CC"NMTReleaseMemory", CC"(JJ)V", (void*)&WB_NMTReleaseMemory }, - {CC"NMTWaitForDataMerge", CC"()Z", (void*)&WB_NMTWaitForDataMerge}, + {CC"NMTOverflowHashBucket", CC"(J)V", (void*)&WB_NMTOverflowHashBucket}, {CC"NMTIsDetailSupported",CC"()Z", (void*)&WB_NMTIsDetailSupported}, #endif // INCLUDE_NMT {CC"deoptimizeAll", CC"()V", (void*)&WB_DeoptimizeAll },
--- a/hotspot/src/share/vm/runtime/arguments.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/runtime/arguments.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -300,6 +300,11 @@ { "UseNewReflection", JDK_Version::jdk(9), JDK_Version::jdk(10) }, { "ReflectionWrapResolutionErrors",JDK_Version::jdk(9), JDK_Version::jdk(10) }, { "VerifyReflectionBytecodes", JDK_Version::jdk(9), JDK_Version::jdk(10) }, + { "AutoShutdownNMT", JDK_Version::jdk(9), JDK_Version::jdk(10) }, +#ifndef ZERO + { "UseFastAccessorMethods", JDK_Version::jdk(9), JDK_Version::jdk(10) }, + { "UseFastEmptyMethods", JDK_Version::jdk(9), JDK_Version::jdk(10) }, +#endif // ZERO { NULL, JDK_Version(0), JDK_Version(0) } }; @@ -799,7 +804,7 @@ } else { *bldarray = REALLOC_C_HEAP_ARRAY(char*, *bldarray, new_count, mtInternal); } - (*bldarray)[*count] = strdup(arg); + (*bldarray)[*count] = os::strdup_check_oom(arg); *count = new_count; } @@ -1070,16 +1075,6 @@ UseCompiler = true; UseLoopCounter = true; -#ifndef ZERO - // Turn these off for mixed and comp. Leave them on for Zero. - if (FLAG_IS_DEFAULT(UseFastAccessorMethods)) { - UseFastAccessorMethods = (mode == _int); - } - if (FLAG_IS_DEFAULT(UseFastEmptyMethods)) { - UseFastEmptyMethods = (mode == _int); - } -#endif - // Default values may be platform/compiler dependent - // use the saved values ClipInlining = Arguments::_ClipInlining; @@ -1885,7 +1880,7 @@ } void Arguments::process_java_launcher_argument(const char* launcher, void* extra_info) { - _sun_java_launcher = strdup(launcher); + _sun_java_launcher = os::strdup_check_oom(launcher); } bool Arguments::created_by_java_launcher() { @@ -2388,7 +2383,7 @@ if (PrintNMTStatistics) { #if INCLUDE_NMT - if (MemTracker::tracking_level() == MemTracker::NMT_off) { + if (MemTracker::tracking_level() == NMT_off) { #endif // INCLUDE_NMT warning("PrintNMTStatistics is disabled, because native memory tracking is not enabled"); PrintNMTStatistics = false; @@ -2995,7 +2990,7 @@ // Redirect GC output to the file. -Xloggc:<filename> // ostream_init_log(), when called will use this filename // to initialize a fileStream. - _gc_log_filename = strdup(tail); + _gc_log_filename = os::strdup_check_oom(tail); if (!is_filename_valid(_gc_log_filename)) { jio_fprintf(defaultStream::output_stream(), "Invalid file name for use with -Xloggc: Filename can only contain the " @@ -3598,15 +3593,24 @@ CommandLineFlags::printFlags(tty, false); vm_exit(0); } +#if INCLUDE_NMT if (match_option(option, "-XX:NativeMemoryTracking", &tail)) { -#if INCLUDE_NMT - MemTracker::init_tracking_options(tail); -#else - jio_fprintf(defaultStream::error_stream(), - "Native Memory Tracking is not supported in this VM\n"); - return JNI_ERR; + // The launcher did not setup nmt environment variable properly. +// if (!MemTracker::check_launcher_nmt_support(tail)) { +// warning("Native Memory Tracking did not setup properly, using wrong launcher?"); +// } + + // Verify if nmt option is valid. + if (MemTracker::verify_nmt_option()) { + // Late initialization, still in single-threaded mode. + if (MemTracker::tracking_level() >= NMT_summary) { + MemTracker::init(); + } + } else { + vm_exit_during_initialization("Syntax error, expecting -XX:NativeMemoryTracking=[off|summary|detail]", NULL); + } + } #endif - } #ifndef PRODUCT
--- a/hotspot/src/share/vm/runtime/fprofiler.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/runtime/fprofiler.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -629,10 +629,16 @@ } vmNode(const char* name, const TickPosition where) : ProfilerNode() { - _name = name; + _name = os::strdup(name); update(where); } + ~vmNode() { + if (_name != NULL) { + os::free((void*)_name); + } + } + const char *name() const { return _name; } bool is_compiled() const { return true; } @@ -784,7 +790,7 @@ assert(index >= 0, "Must be positive"); // Note that we call strdup below since the symbol may be resource allocated if (!table[index]) { - table[index] = new (this) vmNode(os::strdup(name), where); + table[index] = new (this) vmNode(name, where); } else { ProfilerNode* prev = table[index]; for(ProfilerNode* node = prev; node; node = node->next()) { @@ -794,7 +800,7 @@ } prev = node; } - prev->set_next(new (this) vmNode(os::strdup(name), where)); + prev->set_next(new (this) vmNode(name, where)); } }
--- a/hotspot/src/share/vm/runtime/globals.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/runtime/globals.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -945,11 +945,6 @@ diagnostic(bool, PrintNMTStatistics, false, \ "Print native memory tracking summary data if it is on") \ \ - diagnostic(bool, AutoShutdownNMT, true, \ - "Automatically shutdown native memory tracking under stress " \ - "situations. When set to false, native memory tracking tries to " \ - "stay alive at the expense of JVM performance") \ - \ diagnostic(bool, LogCompilation, false, \ "Log compilation activity in detail to LogFile") \ \ @@ -2789,12 +2784,6 @@ product(bool, UseLoopCounter, true, \ "Increment invocation counter on backward branch") \ \ - product(bool, UseFastEmptyMethods, true, \ - "Use fast method entry code for empty methods") \ - \ - product(bool, UseFastAccessorMethods, true, \ - "Use fast method entry code for accessor methods") \ - \ product_pd(bool, UseOnStackReplacement, \ "Use on stack replacement, calls runtime if invoc. counter " \ "overflows in loop") \
--- a/hotspot/src/share/vm/runtime/handles.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/runtime/handles.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, 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 @@ -227,7 +227,7 @@ HandleArea* _prev; // link to outer (older) area public: // Constructor - HandleArea(HandleArea* prev) : Arena(Chunk::tiny_size) { + HandleArea(HandleArea* prev) : Arena(mtThread, Chunk::tiny_size) { debug_only(_handle_mark_nesting = 0); debug_only(_no_handle_mark_nesting = 0); _prev = prev;
--- a/hotspot/src/share/vm/runtime/init.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/runtime/init.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, 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 @@ -34,8 +34,10 @@ #include "runtime/init.hpp" #include "runtime/safepoint.hpp" #include "runtime/sharedRuntime.hpp" +#include "services/memTracker.hpp" #include "utilities/macros.hpp" + // Initialization done by VM thread in vm_init_globals() void check_ThreadShadow(); void eventlog_init(); @@ -131,6 +133,12 @@ javaClasses_init(); // must happen after vtable initialization stubRoutines_init2(); // note: StubRoutines need 2-phase init +#if INCLUDE_NMT + // Solaris stack is walkable only after stubRoutines are set up. + // On Other platforms, the stack is always walkable. + NMT_stack_walkable = true; +#endif // INCLUDE_NMT + // All the flags that get adjusted by VM_Version_init and os::init_2 // have been set so dump the flags now. if (PrintFlagsFinal) {
--- a/hotspot/src/share/vm/runtime/java.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/runtime/java.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -57,7 +57,6 @@ #include "runtime/thread.inline.hpp" #include "runtime/timer.hpp" #include "runtime/vm_operations.hpp" -#include "services/memReporter.hpp" #include "services/memTracker.hpp" #include "trace/tracing.hpp" #include "utilities/dtrace.hpp" @@ -349,12 +348,7 @@ #endif // ENABLE_ZAP_DEAD_LOCALS // Native memory tracking data if (PrintNMTStatistics) { - if (MemTracker::is_on()) { - BaselineTTYOutputer outputer(tty); - MemTracker::print_memory_usage(outputer, K, false); - } else { - tty->print_cr("%s", MemTracker::reason()); - } + MemTracker::final_report(tty); } } @@ -390,12 +384,7 @@ // Native memory tracking data if (PrintNMTStatistics) { - if (MemTracker::is_on()) { - BaselineTTYOutputer outputer(tty); - MemTracker::print_memory_usage(outputer, K, false); - } else { - tty->print_cr("%s", MemTracker::reason()); - } + MemTracker::final_report(tty); } } @@ -544,10 +533,6 @@ BeforeExit_lock->notify_all(); } - // Shutdown NMT before exit. Otherwise, - // it will run into trouble when system destroys static variables. - MemTracker::shutdown(MemTracker::NMT_normal); - if (VerifyStringTableAtExit) { int fail_cnt = 0; {
--- a/hotspot/src/share/vm/runtime/os.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/runtime/os.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -52,6 +52,7 @@ #include "runtime/thread.inline.hpp" #include "runtime/vm_version.hpp" #include "services/attachListener.hpp" +#include "services/nmtCommon.hpp" #include "services/memTracker.hpp" #include "services/threadService.hpp" #include "utilities/defaultStream.hpp" @@ -516,6 +517,14 @@ return dup_str; } +char* os::strdup_check_oom(const char* str, MEMFLAGS flags) { + char* p = os::strdup(str, flags); + if (p == NULL) { + vm_exit_out_of_memory(strlen(str) + 1, OOM_MALLOC_ERROR, "os::strdup_check_oom"); + } + return p; +} + #define paranoid 0 /* only set to 1 if you suspect checking code has bug */ @@ -553,7 +562,11 @@ return ptr; } -void* os::malloc(size_t size, MEMFLAGS memflags, address caller) { +void* os::malloc(size_t size, MEMFLAGS flags) { + return os::malloc(size, flags, CALLER_PC); +} + +void* os::malloc(size_t size, MEMFLAGS memflags, const NativeCallStack& stack) { NOT_PRODUCT(inc_stat_counter(&num_mallocs, 1)); NOT_PRODUCT(inc_stat_counter(&alloc_bytes, size)); @@ -579,11 +592,15 @@ size = 1; } + // NMT support + NMT_TrackingLevel level = MemTracker::tracking_level(); + size_t nmt_header_size = MemTracker::malloc_header_size(level); + #ifndef ASSERT - const size_t alloc_size = size; + const size_t alloc_size = size + nmt_header_size; #else - const size_t alloc_size = GuardedMemory::get_total_size(size); - if (size > alloc_size) { // Check for rollover. + const size_t alloc_size = GuardedMemory::get_total_size(size + nmt_header_size); + if (size + nmt_header_size > alloc_size) { // Check for rollover. return NULL; } #endif @@ -602,7 +619,7 @@ return NULL; } // Wrap memory with guard - GuardedMemory guarded(ptr, size); + GuardedMemory guarded(ptr, size + nmt_header_size); ptr = guarded.get_user_ptr(); #endif if ((intptr_t)ptr == (intptr_t)MallocCatchPtr) { @@ -615,48 +632,50 @@ } // we do not track guard memory - MemTracker::record_malloc((address)ptr, size, memflags, caller == 0 ? CALLER_PC : caller); - - return ptr; + return MemTracker::record_malloc((address)ptr, size, memflags, stack, level); } +void* os::realloc(void *memblock, size_t size, MEMFLAGS flags) { + return os::realloc(memblock, size, flags, CALLER_PC); +} -void* os::realloc(void *memblock, size_t size, MEMFLAGS memflags, address caller) { +void* os::realloc(void *memblock, size_t size, MEMFLAGS memflags, const NativeCallStack& stack) { #ifndef ASSERT NOT_PRODUCT(inc_stat_counter(&num_mallocs, 1)); NOT_PRODUCT(inc_stat_counter(&alloc_bytes, size)); - MemTracker::Tracker tkr = MemTracker::get_realloc_tracker(); - void* ptr = ::realloc(memblock, size); - if (ptr != NULL) { - tkr.record((address)memblock, (address)ptr, size, memflags, - caller == 0 ? CALLER_PC : caller); - } else { - tkr.discard(); - } - return ptr; + // NMT support + void* membase = MemTracker::record_free(memblock); + NMT_TrackingLevel level = MemTracker::tracking_level(); + size_t nmt_header_size = MemTracker::malloc_header_size(level); + void* ptr = ::realloc(membase, size + nmt_header_size); + return MemTracker::record_malloc(ptr, size, memflags, stack, level); #else if (memblock == NULL) { - return os::malloc(size, memflags, (caller == 0 ? CALLER_PC : caller)); + return os::malloc(size, memflags, stack); } if ((intptr_t)memblock == (intptr_t)MallocCatchPtr) { tty->print_cr("os::realloc caught " PTR_FORMAT, memblock); breakpoint(); } - verify_memory(memblock); + // NMT support + void* membase = MemTracker::malloc_base(memblock); + verify_memory(membase); NOT_PRODUCT(if (MallocVerifyInterval > 0) check_heap()); if (size == 0) { return NULL; } // always move the block - void* ptr = os::malloc(size, memflags, caller == 0 ? CALLER_PC : caller); + void* ptr = os::malloc(size, memflags, stack); if (PrintMalloc) { tty->print_cr("os::remalloc " SIZE_FORMAT " bytes, " PTR_FORMAT " --> " PTR_FORMAT, size, memblock, ptr); } // Copy to new memory if malloc didn't fail if ( ptr != NULL ) { - GuardedMemory guarded(memblock); - memcpy(ptr, memblock, MIN2(size, guarded.get_user_size())); - if (paranoid) verify_memory(ptr); + GuardedMemory guarded(MemTracker::malloc_base(memblock)); + // Guard's user data contains NMT header + size_t memblock_size = guarded.get_user_size() - MemTracker::malloc_header_size(memblock); + memcpy(ptr, memblock, MIN2(size, memblock_size)); + if (paranoid) verify_memory(MemTracker::malloc_base(ptr)); if ((intptr_t)ptr == (intptr_t)MallocCatchPtr) { tty->print_cr("os::realloc caught, " SIZE_FORMAT " bytes --> " PTR_FORMAT, size, ptr); breakpoint(); @@ -669,7 +688,6 @@ void os::free(void *memblock, MEMFLAGS memflags) { - address trackp = (address) memblock; NOT_PRODUCT(inc_stat_counter(&num_frees, 1)); #ifdef ASSERT if (memblock == NULL) return; @@ -677,20 +695,22 @@ if (tty != NULL) tty->print_cr("os::free caught " PTR_FORMAT, memblock); breakpoint(); } - verify_memory(memblock); + void* membase = MemTracker::record_free(memblock); + verify_memory(membase); NOT_PRODUCT(if (MallocVerifyInterval > 0) check_heap()); - GuardedMemory guarded(memblock); + GuardedMemory guarded(membase); size_t size = guarded.get_user_size(); inc_stat_counter(&free_bytes, size); - memblock = guarded.release_for_freeing(); + membase = guarded.release_for_freeing(); if (PrintMalloc && tty != NULL) { - fprintf(stderr, "os::free " SIZE_FORMAT " bytes --> " PTR_FORMAT "\n", size, (uintptr_t)memblock); + fprintf(stderr, "os::free " SIZE_FORMAT " bytes --> " PTR_FORMAT "\n", size, (uintptr_t)membase); } + ::free(membase); +#else + void* membase = MemTracker::record_free(memblock); + ::free(membase); #endif - MemTracker::record_free(trackp, memflags); - - ::free(memblock); } void os::init_random(long initval) { @@ -1404,7 +1424,7 @@ char* os::reserve_memory(size_t bytes, char* addr, size_t alignment_hint) { char* result = pd_reserve_memory(bytes, addr, alignment_hint); if (result != NULL) { - MemTracker::record_virtual_memory_reserve((address)result, bytes, mtNone, CALLER_PC); + MemTracker::record_virtual_memory_reserve((address)result, bytes, CALLER_PC); } return result; @@ -1414,7 +1434,7 @@ MEMFLAGS flags) { char* result = pd_reserve_memory(bytes, addr, alignment_hint); if (result != NULL) { - MemTracker::record_virtual_memory_reserve((address)result, bytes, mtNone, CALLER_PC); + MemTracker::record_virtual_memory_reserve((address)result, bytes, CALLER_PC); MemTracker::record_virtual_memory_type((address)result, flags); } @@ -1424,7 +1444,7 @@ char* os::attempt_reserve_memory_at(size_t bytes, char* addr) { char* result = pd_attempt_reserve_memory_at(bytes, addr); if (result != NULL) { - MemTracker::record_virtual_memory_reserve((address)result, bytes, mtNone, CALLER_PC); + MemTracker::record_virtual_memory_reserve((address)result, bytes, CALLER_PC); } return result; } @@ -1464,23 +1484,29 @@ } bool os::uncommit_memory(char* addr, size_t bytes) { - MemTracker::Tracker tkr = MemTracker::get_virtual_memory_uncommit_tracker(); - bool res = pd_uncommit_memory(addr, bytes); - if (res) { - tkr.record((address)addr, bytes); + bool res; + if (MemTracker::tracking_level() > NMT_minimal) { + Tracker tkr = MemTracker::get_virtual_memory_uncommit_tracker(); + res = pd_uncommit_memory(addr, bytes); + if (res) { + tkr.record((address)addr, bytes); + } } else { - tkr.discard(); + res = pd_uncommit_memory(addr, bytes); } return res; } bool os::release_memory(char* addr, size_t bytes) { - MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); - bool res = pd_release_memory(addr, bytes); - if (res) { - tkr.record((address)addr, bytes); + bool res; + if (MemTracker::tracking_level() > NMT_minimal) { + Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); + res = pd_release_memory(addr, bytes); + if (res) { + tkr.record((address)addr, bytes); + } } else { - tkr.discard(); + res = pd_release_memory(addr, bytes); } return res; } @@ -1491,7 +1517,7 @@ bool allow_exec) { char* result = pd_map_memory(fd, file_name, file_offset, addr, bytes, read_only, allow_exec); if (result != NULL) { - MemTracker::record_virtual_memory_reserve_and_commit((address)result, bytes, mtNone, CALLER_PC); + MemTracker::record_virtual_memory_reserve_and_commit((address)result, bytes, CALLER_PC); } return result; } @@ -1504,12 +1530,15 @@ } bool os::unmap_memory(char *addr, size_t bytes) { - MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); - bool result = pd_unmap_memory(addr, bytes); - if (result) { - tkr.record((address)addr, bytes); + bool result; + if (MemTracker::tracking_level() > NMT_minimal) { + Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); + result = pd_unmap_memory(addr, bytes); + if (result) { + tkr.record((address)addr, bytes); + } } else { - tkr.discard(); + result = pd_unmap_memory(addr, bytes); } return result; }
--- a/hotspot/src/share/vm/runtime/os.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/runtime/os.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -65,6 +65,8 @@ class Event; class DLL; class FileHandle; +class NativeCallStack; + template<class E> class GrowableArray; // %%%%% Moved ThreadState, START_FN, OSThread to new osThread.hpp. -- Rose @@ -96,9 +98,11 @@ // Typedef for structured exception handling support typedef void (*java_call_t)(JavaValue* value, methodHandle* method, JavaCallArguments* args, Thread* thread); +class MallocTracker; + class os: AllStatic { friend class VMStructs; - + friend class MallocTracker; public: enum { page_sizes_max = 9 }; // Size of _page_sizes array (8 plus a sentinel) @@ -160,7 +164,10 @@ // Override me as needed static int file_name_strcmp(const char* s1, const char* s2); + // get/unset environment variable static bool getenv(const char* name, char* buffer, int len); + static bool unsetenv(const char* name); + static bool have_special_privileges(); static jlong javaTimeMillis(); @@ -207,8 +214,13 @@ // Interface for detecting multiprocessor system static inline bool is_MP() { +#if !INCLUDE_NMT assert(_processor_count > 0, "invalid processor count"); return _processor_count > 1 || AssumeMP; +#else + // NMT needs atomic operations before this initialization. + return true; +#endif } static julong available_memory(); static julong physical_memory(); @@ -635,15 +647,25 @@ static void* thread_local_storage_at(int index); static void free_thread_local_storage(int index); - // Stack walk - static address get_caller_pc(int n = 0); + // Retrieve native stack frames. + // Parameter: + // stack: an array to storage stack pointers. + // frames: size of above array. + // toSkip: number of stack frames to skip at the beginning. + // Return: number of stack frames captured. + static int get_native_stack(address* stack, int size, int toSkip = 0); // General allocation (must be MT-safe) - static void* malloc (size_t size, MEMFLAGS flags, address caller_pc = 0); - static void* realloc (void *memblock, size_t size, MEMFLAGS flags, address caller_pc = 0); + static void* malloc (size_t size, MEMFLAGS flags, const NativeCallStack& stack); + static void* malloc (size_t size, MEMFLAGS flags); + static void* realloc (void *memblock, size_t size, MEMFLAGS flag, const NativeCallStack& stack); + static void* realloc (void *memblock, size_t size, MEMFLAGS flag); + static void free (void *memblock, MEMFLAGS flags = mtNone); static bool check_heap(bool force = false); // verify C heap integrity static char* strdup(const char *, MEMFLAGS flags = mtInternal); // Like strdup + // Like strdup, but exit VM when strdup() returns NULL + static char* strdup_check_oom(const char*, MEMFLAGS flags = mtInternal); #ifndef PRODUCT static julong num_mallocs; // # of calls to malloc/realloc
--- a/hotspot/src/share/vm/runtime/safepoint.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/runtime/safepoint.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -52,7 +52,6 @@ #include "runtime/sweeper.hpp" #include "runtime/synchronizer.hpp" #include "runtime/thread.inline.hpp" -#include "services/memTracker.hpp" #include "services/runtimeService.hpp" #include "utilities/events.hpp" #include "utilities/macros.hpp" @@ -527,10 +526,6 @@ TraceTime t7("purging class loader data graph", TraceSafepointCleanupTime); ClassLoaderDataGraph::purge_if_needed(); } - - if (MemTracker::is_on()) { - MemTracker::sync(); - } }
--- a/hotspot/src/share/vm/runtime/thread.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/runtime/thread.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -297,8 +297,7 @@ #if INCLUDE_NMT // record thread's native stack, stack grows downward address stack_low_addr = stack_base() - stack_size(); - MemTracker::record_thread_stack(stack_low_addr, stack_size(), this, - CURRENT_PC); + MemTracker::record_thread_stack(stack_low_addr, stack_size()); #endif // INCLUDE_NMT } @@ -316,7 +315,7 @@ #if INCLUDE_NMT if (_stack_base != NULL) { address low_stack_addr = stack_base() - stack_size(); - MemTracker::release_thread_stack(low_stack_addr, stack_size(), this); + MemTracker::release_thread_stack(low_stack_addr, stack_size()); #ifdef ASSERT set_stack_base(NULL); #endif @@ -1425,9 +1424,6 @@ set_monitor_chunks(NULL); set_next(NULL); set_thread_state(_thread_new); -#if INCLUDE_NMT - set_recorder(NULL); -#endif _terminated = _not_terminated; _privileged_stack_top = NULL; _array_for_gc = NULL; @@ -1503,7 +1499,6 @@ _jni_attach_state = _not_attaching_via_jni; } assert(deferred_card_mark().is_empty(), "Default MemRegion ctor"); - _safepoint_visible = false; } bool JavaThread::reguard_stack(address cur_sp) { @@ -1566,7 +1561,6 @@ thr_type = entry_point == &compiler_thread_entry ? os::compiler_thread : os::java_thread; os::create_thread(this, thr_type, stack_sz); - _safepoint_visible = false; // The _osthread may be NULL here because we ran out of memory (too many threads active). // We need to throw and OutOfMemoryError - however we cannot do this here because the caller // may hold a lock and all locks must be unlocked before throwing the exception (throwing @@ -1584,13 +1578,6 @@ tty->print_cr("terminate thread %p", this); } - // By now, this thread should already be invisible to safepoint, - // and its per-thread recorder also collected. - assert(!is_safepoint_visible(), "wrong state"); -#if INCLUDE_NMT - assert(get_recorder() == NULL, "Already collected"); -#endif // INCLUDE_NMT - // JSR166 -- return the parker to the free list Parker::Release(_parker); _parker = NULL; @@ -3359,11 +3346,6 @@ // initialize TLS ThreadLocalStorage::init(); - // Bootstrap native memory tracking, so it can start recording memory - // activities before worker thread is started. This is the first phase - // of bootstrapping, VM is currently running in single-thread mode. - MemTracker::bootstrap_single_thread(); - // Initialize output stream logging ostream_init_log(); @@ -3414,9 +3396,6 @@ // Initialize Java-Level synchronization subsystem ObjectMonitor::Initialize(); - // Second phase of bootstrapping, VM is about entering multi-thread mode - MemTracker::bootstrap_multi_thread(); - // Initialize global modules jint status = init_globals(); if (status != JNI_OK) { @@ -3438,9 +3417,6 @@ // real raw monitor. VM is setup enough here for raw monitor enter. JvmtiExport::transition_pending_onload_raw_monitors(); - // Fully start NMT - MemTracker::start(); - // Create the VMThread { TraceTime timer("Start VMThread", TraceStartupTime); VMThread::create(); @@ -3995,8 +3971,6 @@ daemon = false; } - p->set_safepoint_visible(true); - ThreadService::add_thread(p, daemon); // Possible GC point. @@ -4042,13 +4016,6 @@ // to do callbacks into the safepoint code. However, the safepoint code is not aware // of this thread since it is removed from the queue. p->set_terminated_value(); - - // Now, this thread is not visible to safepoint - p->set_safepoint_visible(false); - // once the thread becomes safepoint invisible, we can not use its per-thread - // recorder. And Threads::do_threads() no longer walks this thread, so we have - // to release its per-thread recorder here. - MemTracker::thread_exiting(p); } // unlock Threads_lock // Since Events::log uses a lock, we grab it outside the Threads_lock
--- a/hotspot/src/share/vm/runtime/thread.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/runtime/thread.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -43,10 +43,6 @@ #include "runtime/unhandledOops.hpp" #include "utilities/macros.hpp" -#if INCLUDE_NMT -#include "services/memRecorder.hpp" -#endif // INCLUDE_NMT - #include "trace/traceBackend.hpp" #include "trace/traceMacros.hpp" #include "utilities/exceptions.hpp" @@ -1036,16 +1032,6 @@ bool do_not_unlock_if_synchronized() { return _do_not_unlock_if_synchronized; } void set_do_not_unlock_if_synchronized(bool val) { _do_not_unlock_if_synchronized = val; } -#if INCLUDE_NMT - // native memory tracking - inline MemRecorder* get_recorder() const { return (MemRecorder*)_recorder; } - inline void set_recorder(MemRecorder* rc) { _recorder = rc; } - - private: - // per-thread memory recorder - MemRecorder* volatile _recorder; -#endif // INCLUDE_NMT - // Suspend/resume support for JavaThread private: inline void set_ext_suspended(); @@ -1485,19 +1471,6 @@ return result; } - // NMT (Native memory tracking) support. - // This flag helps NMT to determine if this JavaThread will be blocked - // at safepoint. If not, ThreadCritical is needed for writing memory records. - // JavaThread is only safepoint visible when it is in Threads' thread list, - // it is not visible until it is added to the list and becomes invisible - // once it is removed from the list. - public: - bool is_safepoint_visible() const { return _safepoint_visible; } - void set_safepoint_visible(bool visible) { _safepoint_visible = visible; } - private: - bool _safepoint_visible; - - // Static operations public: // Returns the running thread as a JavaThread static inline JavaThread* current();
--- a/hotspot/src/share/vm/runtime/vmStructs.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/runtime/vmStructs.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -52,6 +52,7 @@ #include "interpreter/bytecodes.hpp" #include "interpreter/interpreter.hpp" #include "memory/allocation.hpp" +#include "memory/allocation.inline.hpp" #include "memory/cardTableRS.hpp" #include "memory/defNewGeneration.hpp" #include "memory/freeBlockDictionary.hpp" @@ -93,6 +94,7 @@ #include "runtime/globals.hpp" #include "runtime/java.hpp" #include "runtime/javaCalls.hpp" +#include "runtime/os.hpp" #include "runtime/perfMemory.hpp" #include "runtime/serviceThread.hpp" #include "runtime/sharedRuntime.hpp" @@ -3296,14 +3298,14 @@ } } if (strstr(typeName, " const") == typeName + len - 6) { - char * s = strdup(typeName); + char * s = os::strdup_check_oom(typeName); s[len - 6] = '\0'; // tty->print_cr("checking \"%s\" for \"%s\"", s, typeName); if (recursiveFindType(origtypes, s, true) == 1) { - free(s); + os::free(s); return 1; } - free(s); + os::free(s); } if (!isRecurse) { tty->print_cr("type \"%s\" not found", typeName);
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/hotspot/src/share/vm/services/allocationSite.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -0,0 +1,57 @@ +/* + * Copyright (c) 2014, 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_SERVICES_ALLOCATION_SITE_HPP +#define SHARE_VM_SERVICES_ALLOCATION_SITE_HPP + +#include "memory/allocation.hpp" +#include "utilities/nativeCallStack.hpp" + +// Allocation site represents a code path that makes a memory +// allocation +template <class E> class AllocationSite VALUE_OBJ_CLASS_SPEC { + private: + NativeCallStack _call_stack; + E e; + public: + AllocationSite(const NativeCallStack& stack) : _call_stack(stack) { } + int hash() const { return _call_stack.hash(); } + bool equals(const NativeCallStack& stack) const { + return _call_stack.equals(stack); + } + + bool equals(const AllocationSite<E>& other) const { + return other.equals(_call_stack); + } + + const NativeCallStack* call_stack() const { + return &_call_stack; + } + + // Information regarding this allocation + E* data() { return &e; } + const E* peek() const { return &e; } +}; + +#endif // SHARE_VM_SERVICES_ALLOCATION_SITE_HPP
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/hotspot/src/share/vm/services/mallocSiteTable.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -0,0 +1,261 @@ +/* + * Copyright (c) 2014, 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. + * + */ +#include "precompiled.hpp" + + +#include "memory/allocation.inline.hpp" +#include "runtime/atomic.hpp" +#include "services/mallocSiteTable.hpp" + +/* + * Early os::malloc() calls come from initializations of static variables, long before entering any + * VM code. Upon the arrival of the first os::malloc() call, malloc site hashtable has to be + * initialized, along with the allocation site for the hashtable entries. + * To ensure that malloc site hashtable can be initialized without triggering any additional os::malloc() + * call, the hashtable bucket array and hashtable entry allocation site have to be static. + * It is not a problem for hashtable bucket, since it is an array of pointer type, C runtime just + * allocates a block memory and zero the memory for it. + * But for hashtable entry allocation site object, things get tricky. C runtime not only allocates + * memory for it, but also calls its constructor at some later time. If we initialize the allocation site + * at the first os::malloc() call, the object will be reinitialized when its constructor is called + * by C runtime. + * To workaround above issue, we declare a static size_t array with the size of the CallsiteHashtableEntry, + * the memory is used to instantiate CallsiteHashtableEntry for the hashtable entry allocation site. + * Given it is a primitive type array, C runtime will do nothing other than assign the memory block for the variable, + * which is exactly what we want. + * The same trick is also applied to create NativeCallStack object for CallsiteHashtableEntry memory allocation. + * + * Note: C++ object usually aligns to particular alignment, depends on compiler implementation, we declare + * the memory as size_t arrays, to ensure the memory is aligned to native machine word alignment. + */ + +// Reserve enough memory for NativeCallStack and MallocSiteHashtableEntry objects +size_t MallocSiteTable::_hash_entry_allocation_stack[CALC_OBJ_SIZE_IN_TYPE(NativeCallStack, size_t)]; +size_t MallocSiteTable::_hash_entry_allocation_site[CALC_OBJ_SIZE_IN_TYPE(MallocSiteHashtableEntry, size_t)]; + +// Malloc site hashtable buckets +MallocSiteHashtableEntry* MallocSiteTable::_table[MallocSiteTable::table_size]; + +// concurrent access counter +volatile int MallocSiteTable::_access_count = 0; + +// Tracking hashtable contention +NOT_PRODUCT(int MallocSiteTable::_peak_count = 0;) + + +/* + * Initialize malloc site table. + * Hashtable entry is malloc'd, so it can cause infinite recursion. + * To avoid above problem, we pre-initialize a hash entry for + * this allocation site. + * The method is called during C runtime static variable initialization + * time, it is in single-threaded mode from JVM perspective. + */ +bool MallocSiteTable::initialize() { + assert(sizeof(_hash_entry_allocation_stack) >= sizeof(NativeCallStack), "Sanity Check"); + assert(sizeof(_hash_entry_allocation_site) >= sizeof(MallocSiteHashtableEntry), + "Sanity Check"); + assert((size_t)table_size <= MAX_MALLOCSITE_TABLE_SIZE, "Hashtable overflow"); + + // Fake the call stack for hashtable entry allocation + assert(NMT_TrackingStackDepth > 1, "At least one tracking stack"); + + // Create pseudo call stack for hashtable entry allocation + address pc[3]; + if (NMT_TrackingStackDepth >= 3) { + pc[2] = (address)MallocSiteTable::allocation_at; + } + if (NMT_TrackingStackDepth >= 2) { + pc[1] = (address)MallocSiteTable::lookup_or_add; + } + pc[0] = (address)MallocSiteTable::new_entry; + + // Instantiate NativeCallStack object, have to use placement new operator. (see comments above) + NativeCallStack* stack = ::new ((void*)_hash_entry_allocation_stack) + NativeCallStack(pc, MIN2(((int)(sizeof(pc) / sizeof(address))), ((int)NMT_TrackingStackDepth))); + + // Instantiate hash entry for hashtable entry allocation callsite + MallocSiteHashtableEntry* entry = ::new ((void*)_hash_entry_allocation_site) + MallocSiteHashtableEntry(*stack); + + // Add the allocation site to hashtable. + int index = hash_to_index(stack->hash()); + _table[index] = entry; + + return true; +} + +// Walks entries in the hashtable. +// It stops walk if the walker returns false. +bool MallocSiteTable::walk(MallocSiteWalker* walker) { + MallocSiteHashtableEntry* head; + for (int index = 0; index < table_size; index ++) { + head = _table[index]; + while (head != NULL) { + if (!walker->do_malloc_site(head->peek())) { + return false; + } + head = (MallocSiteHashtableEntry*)head->next(); + } + } + return true; +} + +/* + * The hashtable does not have deletion policy on individual entry, + * and each linked list node is inserted via compare-and-swap, + * so each linked list is stable, the contention only happens + * at the end of linked list. + * This method should not return NULL under normal circumstance. + * If NULL is returned, it indicates: + * 1. Out of memory, it cannot allocate new hash entry. + * 2. Overflow hash bucket. + * Under any of above circumstances, caller should handle the situation. + */ +MallocSite* MallocSiteTable::lookup_or_add(const NativeCallStack& key, size_t* bucket_idx, + size_t* pos_idx) { + int index = hash_to_index(key.hash()); + assert(index >= 0, "Negative index"); + *bucket_idx = (size_t)index; + *pos_idx = 0; + + // First entry for this hash bucket + if (_table[index] == NULL) { + MallocSiteHashtableEntry* entry = new_entry(key); + // OOM check + if (entry == NULL) return NULL; + + // swap in the head + if (Atomic::cmpxchg_ptr((void*)entry, (volatile void *)&_table[index], NULL) == NULL) { + return entry->data(); + } + + delete entry; + } + + MallocSiteHashtableEntry* head = _table[index]; + while (head != NULL && (*pos_idx) <= MAX_BUCKET_LENGTH) { + MallocSite* site = head->data(); + if (site->equals(key)) { + // found matched entry + return head->data(); + } + + if (head->next() == NULL && (*pos_idx) < MAX_BUCKET_LENGTH) { + MallocSiteHashtableEntry* entry = new_entry(key); + // OOM check + if (entry == NULL) return NULL; + if (head->atomic_insert(entry)) { + (*pos_idx) ++; + return entry->data(); + } + // contended, other thread won + delete entry; + } + head = (MallocSiteHashtableEntry*)head->next(); + (*pos_idx) ++; + } + return NULL; +} + +// Access malloc site +MallocSite* MallocSiteTable::malloc_site(size_t bucket_idx, size_t pos_idx) { + assert(bucket_idx < table_size, "Invalid bucket index"); + MallocSiteHashtableEntry* head = _table[bucket_idx]; + for (size_t index = 0; index < pos_idx && head != NULL; + index ++, head = (MallocSiteHashtableEntry*)head->next()); + assert(head != NULL, "Invalid position index"); + return head->data(); +} + +// Allocates MallocSiteHashtableEntry object. Special call stack +// (pre-installed allocation site) has to be used to avoid infinite +// recursion. +MallocSiteHashtableEntry* MallocSiteTable::new_entry(const NativeCallStack& key) { + void* p = AllocateHeap(sizeof(MallocSiteHashtableEntry), mtNMT, + *hash_entry_allocation_stack(), AllocFailStrategy::RETURN_NULL); + return ::new (p) MallocSiteHashtableEntry(key); +} + +void MallocSiteTable::reset() { + for (int index = 0; index < table_size; index ++) { + MallocSiteHashtableEntry* head = _table[index]; + _table[index] = NULL; + delete_linked_list(head); + } +} + +void MallocSiteTable::delete_linked_list(MallocSiteHashtableEntry* head) { + MallocSiteHashtableEntry* p; + while (head != NULL) { + p = head; + head = (MallocSiteHashtableEntry*)head->next(); + if (p != (MallocSiteHashtableEntry*)_hash_entry_allocation_site) { + delete p; + } + } +} + +void MallocSiteTable::shutdown() { + AccessLock locker(&_access_count); + locker.exclusiveLock(); + reset(); +} + +bool MallocSiteTable::walk_malloc_site(MallocSiteWalker* walker) { + assert(walker != NULL, "NuLL walker"); + AccessLock locker(&_access_count); + if (locker.sharedLock()) { + NOT_PRODUCT(_peak_count = MAX2(_peak_count, _access_count);) + return walk(walker); + } + return false; +} + + +void MallocSiteTable::AccessLock::exclusiveLock() { + jint target; + jint val; + + assert(_lock_state != ExclusiveLock, "Can only call once"); + assert(*_lock >= 0, "Can not content exclusive lock"); + + // make counter negative to block out shared locks + do { + val = *_lock; + target = _MAGIC_ + *_lock; + } while (Atomic::cmpxchg(target, _lock, val) != val); + + // wait for all readers to exit + while (*_lock != _MAGIC_) { +#ifdef _WINDOWS + os::naked_short_sleep(1); +#else + os::naked_yield(); +#endif + } + _lock_state = ExclusiveLock; +} + +
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/hotspot/src/share/vm/services/mallocSiteTable.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -0,0 +1,268 @@ +/* + * Copyright (c) 2014, 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_SERVICES_MALLOC_SITE_TABLE_HPP +#define SHARE_VM_SERVICES_MALLOC_SITE_TABLE_HPP + +#if INCLUDE_NMT + +#include "memory/allocation.hpp" +#include "runtime/atomic.hpp" +#include "services/allocationSite.hpp" +#include "services/mallocTracker.hpp" +#include "services/nmtCommon.hpp" + +// MallocSite represents a code path that eventually calls +// os::malloc() to allocate memory +class MallocSite : public AllocationSite<MemoryCounter> { + public: + MallocSite() : + AllocationSite<MemoryCounter>(emptyStack) { } + + MallocSite(const NativeCallStack& stack) : + AllocationSite<MemoryCounter>(stack) { } + + void allocate(size_t size) { data()->allocate(size); } + void deallocate(size_t size) { data()->deallocate(size); } + + // Memory allocated from this code path + size_t size() const { return peek()->size(); } + // The number of calls were made + size_t count() const { return peek()->count(); } +}; + +// Malloc site hashtable entry +class MallocSiteHashtableEntry : public CHeapObj<mtNMT> { + private: + MallocSite _malloc_site; + MallocSiteHashtableEntry* _next; + + public: + MallocSiteHashtableEntry() : _next(NULL) { } + + MallocSiteHashtableEntry(NativeCallStack stack): + _malloc_site(stack), _next(NULL) { } + + inline const MallocSiteHashtableEntry* next() const { + return _next; + } + + // Insert an entry atomically. + // Return true if the entry is inserted successfully. + // The operation can be failed due to contention from other thread. + bool atomic_insert(const MallocSiteHashtableEntry* entry) { + return (Atomic::cmpxchg_ptr((void*)entry, (volatile void*)&_next, + NULL) == NULL); + } + + void set_callsite(const MallocSite& site) { + _malloc_site = site; + } + + inline const MallocSite* peek() const { return &_malloc_site; } + inline MallocSite* data() { return &_malloc_site; } + + inline long hash() const { return _malloc_site.hash(); } + inline bool equals(const NativeCallStack& stack) const { + return _malloc_site.equals(stack); + } + // Allocation/deallocation on this allocation site + inline void allocate(size_t size) { _malloc_site.allocate(size); } + inline void deallocate(size_t size) { _malloc_site.deallocate(size); } + // Memory counters + inline size_t size() const { return _malloc_site.size(); } + inline size_t count() const { return _malloc_site.count(); } +}; + +// The walker walks every entry on MallocSiteTable +class MallocSiteWalker : public StackObj { + public: + virtual bool do_malloc_site(const MallocSite* e) { return false; } +}; + +/* + * Native memory tracking call site table. + * The table is only needed when detail tracking is enabled. + */ +class MallocSiteTable : AllStatic { + private: + // The number of hash bucket in this hashtable. The number should + // be tuned if malloc activities changed significantly. + // The statistics data can be obtained via Jcmd + // jcmd <pid> VM.native_memory statistics. + + // Currently, (number of buckets / number of entires) ratio is + // about 1 / 6 + enum { + table_base_size = 128, // The base size is calculated from statistics to give + // table ratio around 1:6 + table_size = (table_base_size * NMT_TrackingStackDepth - 1) + }; + + + // This is a very special lock, that allows multiple shared accesses (sharedLock), but + // once exclusive access (exclusiveLock) is requested, all shared accesses are + // rejected forever. + class AccessLock : public StackObj { + enum LockState { + NoLock, + SharedLock, + ExclusiveLock + }; + + private: + // A very large negative number. The only possibility to "overflow" + // this number is when there are more than -min_jint threads in + // this process, which is not going to happen in foreseeable future. + const static int _MAGIC_ = min_jint; + + LockState _lock_state; + volatile int* _lock; + public: + AccessLock(volatile int* lock) : + _lock(lock), _lock_state(NoLock) { + } + + ~AccessLock() { + if (_lock_state == SharedLock) { + Atomic::dec((volatile jint*)_lock); + } + } + // Acquire shared lock. + // Return true if shared access is granted. + inline bool sharedLock() { + jint res = Atomic::add(1, _lock); + if (res < 0) { + Atomic::add(-1, _lock); + return false; + } + _lock_state = SharedLock; + return true; + } + // Acquire exclusive lock + void exclusiveLock(); + }; + + public: + static bool initialize(); + static void shutdown(); + + NOT_PRODUCT(static int access_peak_count() { return _peak_count; }) + + // Number of hash buckets + static inline int hash_buckets() { return (int)table_size; } + + // Access and copy a call stack from this table. Shared lock should be + // acquired before access the entry. + static inline bool access_stack(NativeCallStack& stack, size_t bucket_idx, + size_t pos_idx) { + AccessLock locker(&_access_count); + if (locker.sharedLock()) { + NOT_PRODUCT(_peak_count = MAX2(_peak_count, _access_count);) + MallocSite* site = malloc_site(bucket_idx, pos_idx); + if (site != NULL) { + stack = *site->call_stack(); + return true; + } + } + return false; + } + + // Record a new allocation from specified call path. + // Return true if the allocation is recorded successfully, bucket_idx + // and pos_idx are also updated to indicate the entry where the allocation + // information was recorded. + // Return false only occurs under rare scenarios: + // 1. out of memory + // 2. overflow hash bucket + static inline bool allocation_at(const NativeCallStack& stack, size_t size, + size_t* bucket_idx, size_t* pos_idx) { + AccessLock locker(&_access_count); + if (locker.sharedLock()) { + NOT_PRODUCT(_peak_count = MAX2(_peak_count, _access_count);) + MallocSite* site = lookup_or_add(stack, bucket_idx, pos_idx); + if (site != NULL) site->allocate(size); + return site != NULL; + } + return false; + } + + // Record memory deallocation. bucket_idx and pos_idx indicate where the allocation + // information was recorded. + static inline bool deallocation_at(size_t size, size_t bucket_idx, size_t pos_idx) { + AccessLock locker(&_access_count); + if (locker.sharedLock()) { + NOT_PRODUCT(_peak_count = MAX2(_peak_count, _access_count);) + MallocSite* site = malloc_site(bucket_idx, pos_idx); + if (site != NULL) { + site->deallocate(size); + return true; + } + } + return false; + } + + // Walk this table. + static bool walk_malloc_site(MallocSiteWalker* walker); + + private: + static MallocSiteHashtableEntry* new_entry(const NativeCallStack& key); + static void reset(); + + // Delete a bucket linked list + static void delete_linked_list(MallocSiteHashtableEntry* head); + + static MallocSite* lookup_or_add(const NativeCallStack& key, size_t* bucket_idx, size_t* pos_idx); + static MallocSite* malloc_site(size_t bucket_idx, size_t pos_idx); + static bool walk(MallocSiteWalker* walker); + + static inline int hash_to_index(int hash) { + hash = (hash > 0) ? hash : (-hash); + return (hash % table_size); + } + + static inline const NativeCallStack* hash_entry_allocation_stack() { + return (NativeCallStack*)_hash_entry_allocation_stack; + } + + private: + // Counter for counting concurrent access + static volatile int _access_count; + + // The callsite hashtable. It has to be a static table, + // since malloc call can come from C runtime linker. + static MallocSiteHashtableEntry* _table[table_size]; + + + // Reserve enough memory for placing the objects + + // The memory for hashtable entry allocation stack object + static size_t _hash_entry_allocation_stack[CALC_OBJ_SIZE_IN_TYPE(NativeCallStack, size_t)]; + // The memory for hashtable entry allocation callsite object + static size_t _hash_entry_allocation_site[CALC_OBJ_SIZE_IN_TYPE(MallocSiteHashtableEntry, size_t)]; + NOT_PRODUCT(static int _peak_count;) +}; + +#endif // INCLUDE_NMT +#endif // SHARE_VM_SERVICES_MALLOC_SITE_TABLE_HPP
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/hotspot/src/share/vm/services/mallocTracker.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -0,0 +1,200 @@ +/* + * Copyright (c) 2014, 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. + * + */ +#include "precompiled.hpp" + +#include "runtime/atomic.hpp" +#include "runtime/atomic.inline.hpp" +#include "services/mallocSiteTable.hpp" +#include "services/mallocTracker.hpp" +#include "services/mallocTracker.inline.hpp" +#include "services/memTracker.hpp" + +size_t MallocMemorySummary::_snapshot[CALC_OBJ_SIZE_IN_TYPE(MallocMemorySnapshot, size_t)]; + +// Total malloc'd memory amount +size_t MallocMemorySnapshot::total() const { + size_t amount = 0; + for (int index = 0; index < mt_number_of_types; index ++) { + amount += _malloc[index].malloc_size(); + } + amount += _tracking_header.size() + total_arena(); + return amount; +} + +// Total malloc'd memory used by arenas +size_t MallocMemorySnapshot::total_arena() const { + size_t amount = 0; + for (int index = 0; index < mt_number_of_types; index ++) { + amount += _malloc[index].arena_size(); + } + return amount; +} + + +void MallocMemorySnapshot::reset() { + _tracking_header.reset(); + for (int index = 0; index < mt_number_of_types; index ++) { + _malloc[index].reset(); + } +} + +// Make adjustment by subtracting chunks used by arenas +// from total chunks to get total free chunck size +void MallocMemorySnapshot::make_adjustment() { + size_t arena_size = total_arena(); + int chunk_idx = NMTUtil::flag_to_index(mtChunk); + _malloc[chunk_idx].record_free(arena_size); +} + + +void MallocMemorySummary::initialize() { + assert(sizeof(_snapshot) >= sizeof(MallocMemorySnapshot), "Sanity Check"); + // Uses placement new operator to initialize static area. + ::new ((void*)_snapshot)MallocMemorySnapshot(); +} + +void MallocHeader::release() const { + // Tracking already shutdown, no housekeeping is needed anymore + if (MemTracker::tracking_level() <= NMT_minimal) return; + + MallocMemorySummary::record_free(size(), flags()); + MallocMemorySummary::record_free_malloc_header(sizeof(MallocHeader)); + if (tracking_level() == NMT_detail) { + MallocSiteTable::deallocation_at(size(), _bucket_idx, _pos_idx); + } +} + +bool MallocHeader::record_malloc_site(const NativeCallStack& stack, size_t size, + size_t* bucket_idx, size_t* pos_idx) const { + bool ret = MallocSiteTable::allocation_at(stack, size, bucket_idx, pos_idx); + + // Something went wrong, could be OOM or overflow malloc site table. + // We want to keep tracking data under OOM circumstance, so transition to + // summary tracking. + if (!ret) { + MemTracker::transition_to(NMT_summary); + } + return ret; +} + +bool MallocHeader::get_stack(NativeCallStack& stack) const { + return MallocSiteTable::access_stack(stack, _bucket_idx, _pos_idx); +} + +bool MallocTracker::initialize(NMT_TrackingLevel level) { + if (level >= NMT_summary) { + MallocMemorySummary::initialize(); + } + + if (level == NMT_detail) { + return MallocSiteTable::initialize(); + } + return true; +} + +bool MallocTracker::transition(NMT_TrackingLevel from, NMT_TrackingLevel to) { + assert(from != NMT_off, "Can not transition from off state"); + assert(to != NMT_off, "Can not transition to off state"); + if (from == NMT_minimal) { + MallocMemorySummary::reset(); + } + + if (to == NMT_detail) { + assert(from == NMT_minimal || from == NMT_summary, "Just check"); + return MallocSiteTable::initialize(); + } else if (from == NMT_detail) { + assert(to == NMT_minimal || to == NMT_summary, "Just check"); + MallocSiteTable::shutdown(); + } + return true; +} + +// Record a malloc memory allocation +void* MallocTracker::record_malloc(void* malloc_base, size_t size, MEMFLAGS flags, + const NativeCallStack& stack, NMT_TrackingLevel level) { + void* memblock; // the address for user data + MallocHeader* header = NULL; + + if (malloc_base == NULL) { + return NULL; + } + + // Check malloc size, size has to <= MAX_MALLOC_SIZE. This is only possible on 32-bit + // systems, when malloc size >= 1GB, but is is safe to assume it won't happen. + if (size > MAX_MALLOC_SIZE) { + fatal("Should not use malloc for big memory block, use virtual memory instead"); + } + // Uses placement global new operator to initialize malloc header + switch(level) { + case NMT_off: + return malloc_base; + case NMT_minimal: { + MallocHeader* hdr = ::new (malloc_base) MallocHeader(); + break; + } + case NMT_summary: { + header = ::new (malloc_base) MallocHeader(size, flags); + break; + } + case NMT_detail: { + header = ::new (malloc_base) MallocHeader(size, flags, stack); + break; + } + default: + ShouldNotReachHere(); + } + memblock = (void*)((char*)malloc_base + sizeof(MallocHeader)); + + // The alignment check: 8 bytes alignment for 32 bit systems. + // 16 bytes alignment for 64-bit systems. + assert(((size_t)memblock & (sizeof(size_t) * 2 - 1)) == 0, "Alignment check"); + + // Sanity check + assert(get_memory_tracking_level(memblock) == level, + "Wrong tracking level"); + +#ifdef ASSERT + if (level > NMT_minimal) { + // Read back + assert(get_size(memblock) == size, "Wrong size"); + assert(get_flags(memblock) == flags, "Wrong flags"); + } +#endif + + return memblock; +} + +void* MallocTracker::record_free(void* memblock) { + // Never turned on + if (MemTracker::tracking_level() == NMT_off || + memblock == NULL) { + return memblock; + } + MallocHeader* header = malloc_header(memblock); + header->release(); + + return (void*)header; +} + +
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/hotspot/src/share/vm/services/mallocTracker.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -0,0 +1,424 @@ +/* + * Copyright (c) 2014, 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_SERVICES_MALLOC_TRACKER_HPP +#define SHARE_VM_SERVICES_MALLOC_TRACKER_HPP + +#if INCLUDE_NMT + +#include "memory/allocation.hpp" +#include "runtime/atomic.hpp" +#include "services/nmtCommon.hpp" +#include "utilities/nativeCallStack.hpp" + +/* + * This counter class counts memory allocation and deallocation, + * records total memory allocation size and number of allocations. + * The counters are updated atomically. + */ +class MemoryCounter VALUE_OBJ_CLASS_SPEC { + private: + size_t _count; + size_t _size; + + DEBUG_ONLY(size_t _peak_count;) + DEBUG_ONLY(size_t _peak_size; ) + + public: + MemoryCounter() : _count(0), _size(0) { + DEBUG_ONLY(_peak_count = 0;) + DEBUG_ONLY(_peak_size = 0;) + } + + // Reset counters + void reset() { + _size = 0; + _count = 0; + DEBUG_ONLY(_peak_size = 0;) + DEBUG_ONLY(_peak_count = 0;) + } + + inline void allocate(size_t sz) { + Atomic::add(1, (volatile MemoryCounterType*)&_count); + if (sz > 0) { + Atomic::add((MemoryCounterType)sz, (volatile MemoryCounterType*)&_size); + DEBUG_ONLY(_peak_size = MAX2(_peak_size, _size)); + } + DEBUG_ONLY(_peak_count = MAX2(_peak_count, _count);) + } + + inline void deallocate(size_t sz) { + assert(_count > 0, "Negative counter"); + assert(_size >= sz, "Negative size"); + Atomic::add(-1, (volatile MemoryCounterType*)&_count); + if (sz > 0) { + Atomic::add(-(MemoryCounterType)sz, (volatile MemoryCounterType*)&_size); + } + } + + inline void resize(long sz) { + if (sz != 0) { + Atomic::add((MemoryCounterType)sz, (volatile MemoryCounterType*)&_size); + DEBUG_ONLY(_peak_size = MAX2(_size, _peak_size);) + } + } + + inline size_t count() const { return _count; } + inline size_t size() const { return _size; } + DEBUG_ONLY(inline size_t peak_count() const { return _peak_count; }) + DEBUG_ONLY(inline size_t peak_size() const { return _peak_size; }) + +}; + +/* + * Malloc memory used by a particular subsystem. + * It includes the memory acquired through os::malloc() + * call and arena's backing memory. + */ +class MallocMemory VALUE_OBJ_CLASS_SPEC { + private: + MemoryCounter _malloc; + MemoryCounter _arena; + + public: + MallocMemory() { } + + inline void record_malloc(size_t sz) { + _malloc.allocate(sz); + } + + inline void record_free(size_t sz) { + _malloc.deallocate(sz); + } + + inline void record_new_arena() { + _arena.allocate(0); + } + + inline void record_arena_free() { + _arena.deallocate(0); + } + + inline void record_arena_size_change(long sz) { + _arena.resize(sz); + } + + void reset() { + _malloc.reset(); + _arena.reset(); + } + + inline size_t malloc_size() const { return _malloc.size(); } + inline size_t malloc_count() const { return _malloc.count();} + inline size_t arena_size() const { return _arena.size(); } + inline size_t arena_count() const { return _arena.count(); } + + DEBUG_ONLY(inline const MemoryCounter& malloc_counter() const { return _malloc; }) + DEBUG_ONLY(inline const MemoryCounter& arena_counter() const { return _arena; }) +}; + +class MallocMemorySummary; + +// A snapshot of malloc'd memory, includes malloc memory +// usage by types and memory used by tracking itself. +class MallocMemorySnapshot : public ResourceObj { + friend class MallocMemorySummary; + + private: + MallocMemory _malloc[mt_number_of_types]; + MemoryCounter _tracking_header; + + + public: + inline MallocMemory* by_type(MEMFLAGS flags) { + int index = NMTUtil::flag_to_index(flags); + return &_malloc[index]; + } + + inline MallocMemory* by_index(int index) { + assert(index >= 0, "Index out of bound"); + assert(index < mt_number_of_types, "Index out of bound"); + return &_malloc[index]; + } + + inline MemoryCounter* malloc_overhead() { + return &_tracking_header; + } + + // Total malloc'd memory amount + size_t total() const; + // Total malloc'd memory used by arenas + size_t total_arena() const; + + inline size_t thread_count() { + return by_type(mtThreadStack)->malloc_count(); + } + + void reset(); + + void copy_to(MallocMemorySnapshot* s) { + s->_tracking_header = _tracking_header; + for (int index = 0; index < mt_number_of_types; index ++) { + s->_malloc[index] = _malloc[index]; + } + } + + // Make adjustment by subtracting chunks used by arenas + // from total chunks to get total free chunk size + void make_adjustment(); +}; + +/* + * This class is for collecting malloc statistics at summary level + */ +class MallocMemorySummary : AllStatic { + private: + // Reserve memory for placement of MallocMemorySnapshot object + static size_t _snapshot[CALC_OBJ_SIZE_IN_TYPE(MallocMemorySnapshot, size_t)]; + + public: + static void initialize(); + + static inline void record_malloc(size_t size, MEMFLAGS flag) { + as_snapshot()->by_type(flag)->record_malloc(size); + } + + static inline void record_free(size_t size, MEMFLAGS flag) { + as_snapshot()->by_type(flag)->record_free(size); + } + + static inline void record_new_arena(MEMFLAGS flag) { + as_snapshot()->by_type(flag)->record_new_arena(); + } + + static inline void record_arena_free(MEMFLAGS flag) { + as_snapshot()->by_type(flag)->record_arena_free(); + } + + static inline void record_arena_size_change(long size, MEMFLAGS flag) { + as_snapshot()->by_type(flag)->record_arena_size_change(size); + } + + static void snapshot(MallocMemorySnapshot* s) { + as_snapshot()->copy_to(s); + s->make_adjustment(); + } + + // Record memory used by malloc tracking header + static inline void record_new_malloc_header(size_t sz) { + as_snapshot()->malloc_overhead()->allocate(sz); + } + + static inline void record_free_malloc_header(size_t sz) { + as_snapshot()->malloc_overhead()->deallocate(sz); + } + + // The memory used by malloc tracking headers + static inline size_t tracking_overhead() { + return as_snapshot()->malloc_overhead()->size(); + } + + // Reset all counters to zero + static void reset() { + as_snapshot()->reset(); + } + + static MallocMemorySnapshot* as_snapshot() { + return (MallocMemorySnapshot*)_snapshot; + } +}; + + +/* + * Malloc tracking header. + * To satisfy malloc alignment requirement, NMT uses 2 machine words for tracking purpose, + * which ensures 8-bytes alignment on 32-bit systems and 16-bytes on 64-bit systems (Product build). + */ + +class MallocHeader VALUE_OBJ_CLASS_SPEC { +#ifdef _LP64 + size_t _size : 62; + size_t _level : 2; + size_t _flags : 8; + size_t _pos_idx : 16; + size_t _bucket_idx: 40; +#define MAX_MALLOCSITE_TABLE_SIZE ((size_t)1 << 40) +#define MAX_BUCKET_LENGTH ((size_t)(1 << 16)) +#define MAX_MALLOC_SIZE (((size_t)1 << 62) - 1) +#else + size_t _size : 30; + size_t _level : 2; + size_t _flags : 8; + size_t _pos_idx : 8; + size_t _bucket_idx: 16; +#define MAX_MALLOCSITE_TABLE_SIZE ((size_t)(1 << 16)) +#define MAX_BUCKET_LENGTH ((size_t)(1 << 8)) +// Max malloc size = 1GB - 1 on 32 bit system, such has total 4GB memory +#define MAX_MALLOC_SIZE ((size_t)(1 << 30) - 1) +#endif // _LP64 + + public: + // Summary tracking header + MallocHeader(size_t size, MEMFLAGS flags) { + assert(sizeof(MallocHeader) == sizeof(void*) * 2, + "Wrong header size"); + + _level = NMT_summary; + _flags = flags; + set_size(size); + MallocMemorySummary::record_malloc(size, flags); + MallocMemorySummary::record_new_malloc_header(sizeof(MallocHeader)); + } + // Detail tracking header + MallocHeader(size_t size, MEMFLAGS flags, const NativeCallStack& stack) { + assert(sizeof(MallocHeader) == sizeof(void*) * 2, + "Wrong header size"); + + _level = NMT_detail; + _flags = flags; + set_size(size); + size_t bucket_idx; + size_t pos_idx; + if (record_malloc_site(stack, size, &bucket_idx, &pos_idx)) { + assert(bucket_idx <= MAX_MALLOCSITE_TABLE_SIZE, "Overflow bucket index"); + assert(pos_idx <= MAX_BUCKET_LENGTH, "Overflow bucket position index"); + _bucket_idx = bucket_idx; + _pos_idx = pos_idx; + } + MallocMemorySummary::record_malloc(size, flags); + MallocMemorySummary::record_new_malloc_header(sizeof(MallocHeader)); + } + // Minimal tracking header + MallocHeader() { + assert(sizeof(MallocHeader) == sizeof(void*) * 2, + "Wrong header size"); + + _level = (unsigned short)NMT_minimal; + } + + inline NMT_TrackingLevel tracking_level() const { + return (NMT_TrackingLevel)_level; + } + + inline size_t size() const { return _size; } + inline MEMFLAGS flags() const { return (MEMFLAGS)_flags; } + bool get_stack(NativeCallStack& stack) const; + + // Cleanup tracking information before the memory is released. + void release() const; + + private: + inline void set_size(size_t size) { + assert(size <= MAX_MALLOC_SIZE, "Malloc size too large, should use virtual memory?"); + _size = size; + } + bool record_malloc_site(const NativeCallStack& stack, size_t size, + size_t* bucket_idx, size_t* pos_idx) const; +}; + + +// Main class called from MemTracker to track malloc activities +class MallocTracker : AllStatic { + public: + // Initialize malloc tracker for specific tracking level + static bool initialize(NMT_TrackingLevel level); + + static bool transition(NMT_TrackingLevel from, NMT_TrackingLevel to); + + // malloc tracking header size for specific tracking level + static inline size_t malloc_header_size(NMT_TrackingLevel level) { + return (level == NMT_off) ? 0 : sizeof(MallocHeader); + } + + // Parameter name convention: + // memblock : the beginning address for user data + // malloc_base: the beginning address that includes malloc tracking header + // + // The relationship: + // memblock = (char*)malloc_base + sizeof(nmt header) + // + + // Record malloc on specified memory block + static void* record_malloc(void* malloc_base, size_t size, MEMFLAGS flags, + const NativeCallStack& stack, NMT_TrackingLevel level); + + // Record free on specified memory block + static void* record_free(void* memblock); + + // Get tracking level of specified memory block + static inline NMT_TrackingLevel get_memory_tracking_level(void* memblock); + + + // Offset memory address to header address + static inline void* get_base(void* memblock); + static inline void* get_base(void* memblock, NMT_TrackingLevel level) { + if (memblock == NULL || level == NMT_off) return memblock; + return (char*)memblock - malloc_header_size(level); + } + + // Get memory size + static inline size_t get_size(void* memblock) { + MallocHeader* header = malloc_header(memblock); + assert(header->tracking_level() >= NMT_summary, + "Wrong tracking level"); + return header->size(); + } + + // Get memory type + static inline MEMFLAGS get_flags(void* memblock) { + MallocHeader* header = malloc_header(memblock); + assert(header->tracking_level() >= NMT_summary, + "Wrong tracking level"); + return header->flags(); + } + + // Get header size + static inline size_t get_header_size(void* memblock) { + return (memblock == NULL) ? 0 : sizeof(MallocHeader); + } + + static inline void record_new_arena(MEMFLAGS flags) { + MallocMemorySummary::record_new_arena(flags); + } + + static inline void record_arena_free(MEMFLAGS flags) { + MallocMemorySummary::record_arena_free(flags); + } + + static inline void record_arena_size_change(int size, MEMFLAGS flags) { + MallocMemorySummary::record_arena_size_change(size, flags); + } + private: + static inline MallocHeader* malloc_header(void *memblock) { + assert(memblock != NULL, "NULL pointer"); + MallocHeader* header = (MallocHeader*)((char*)memblock - sizeof(MallocHeader)); + assert(header->tracking_level() >= NMT_minimal, "Bad header"); + return header; + } +}; + +#endif // INCLUDE_NMT + + +#endif //SHARE_VM_SERVICES_MALLOC_TRACKER_HPP
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/hotspot/src/share/vm/services/mallocTracker.inline.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -0,0 +1,43 @@ +/* + * Copyright (c) 2014, 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_SERVICES_MALLOC_TRACKER_INLINE_HPP +#define SHARE_VM_SERVICES_MALLOC_TRACKER_INLINE_HPP + +#include "services/mallocTracker.hpp" +#include "services/memTracker.hpp" + +inline NMT_TrackingLevel MallocTracker::get_memory_tracking_level(void* memblock) { + assert(memblock != NULL, "Sanity check"); + if (MemTracker::tracking_level() == NMT_off) return NMT_off; + MallocHeader* header = malloc_header(memblock); + return header->tracking_level(); +} + +inline void* MallocTracker::get_base(void* memblock){ + return get_base(memblock, MemTracker::tracking_level()); +} + +#endif // SHARE_VM_SERVICES_MALLOC_TRACKER_INLINE_HPP +
--- a/hotspot/src/share/vm/services/management.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/services/management.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1914,7 +1914,7 @@ ResourceMark rm(THREAD); // thread->name() uses ResourceArea assert(thread->name() != NULL, "All threads should have a name"); - _names_chars[_count] = strdup(thread->name()); + _names_chars[_count] = os::strdup(thread->name()); _times->long_at_put(_count, os::is_thread_cpu_time_supported() ? os::thread_cpu_time(thread) : -1); _count++; @@ -1932,7 +1932,7 @@ ThreadTimesClosure::~ThreadTimesClosure() { for (int i = 0; i < _count; i++) { - free(_names_chars[i]); + os::free(_names_chars[i]); } FREE_C_HEAP_ARRAY(char *, _names_chars, mtInternal); }
--- a/hotspot/src/share/vm/services/memBaseline.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/services/memBaseline.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2012, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012, 2014, 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 @@ -22,471 +22,301 @@ * */ #include "precompiled.hpp" + #include "memory/allocation.hpp" #include "runtime/safepoint.hpp" #include "runtime/thread.inline.hpp" #include "services/memBaseline.hpp" #include "services/memTracker.hpp" - -MemType2Name MemBaseline::MemType2NameMap[NUMBER_OF_MEMORY_TYPE] = { - {mtJavaHeap, "Java Heap"}, - {mtClass, "Class"}, - {mtThreadStack,"Thread Stack"}, - {mtThread, "Thread"}, - {mtCode, "Code"}, - {mtGC, "GC"}, - {mtCompiler, "Compiler"}, - {mtInternal, "Internal"}, - {mtOther, "Other"}, - {mtSymbol, "Symbol"}, - {mtNMT, "Memory Tracking"}, - {mtTracing, "Tracing"}, - {mtChunk, "Pooled Free Chunks"}, - {mtClassShared,"Shared spaces for classes"}, - {mtTest, "Test"}, - {mtNone, "Unknown"} // It can happen when type tagging records are lagging - // behind -}; - -MemBaseline::MemBaseline() { - _baselined = false; - - for (int index = 0; index < NUMBER_OF_MEMORY_TYPE; index ++) { - _malloc_data[index].set_type(MemType2NameMap[index]._flag); - _vm_data[index].set_type(MemType2NameMap[index]._flag); - _arena_data[index].set_type(MemType2NameMap[index]._flag); +/* + * Sizes are sorted in descenting order for reporting + */ +int compare_malloc_size(const MallocSite& s1, const MallocSite& s2) { + if (s1.size() == s2.size()) { + return 0; + } else if (s1.size() > s2.size()) { + return -1; + } else { + return 1; } - - _malloc_cs = NULL; - _vm_cs = NULL; - _vm_map = NULL; - - _number_of_classes = 0; - _number_of_threads = 0; } -void MemBaseline::clear() { - if (_malloc_cs != NULL) { - delete _malloc_cs; - _malloc_cs = NULL; +int compare_virtual_memory_size(const VirtualMemoryAllocationSite& s1, + const VirtualMemoryAllocationSite& s2) { + if (s1.reserved() == s2.reserved()) { + return 0; + } else if (s1.reserved() > s2.reserved()) { + return -1; + } else { + return 1; } +} - if (_vm_cs != NULL) { - delete _vm_cs; - _vm_cs = NULL; - } - - if (_vm_map != NULL) { - delete _vm_map; - _vm_map = NULL; - } - - reset(); +// Sort into allocation site addresses order for baseline comparison +int compare_malloc_site(const MallocSite& s1, const MallocSite& s2) { + return s1.call_stack()->compare(*s2.call_stack()); } -void MemBaseline::reset() { - _baselined = false; - _total_vm_reserved = 0; - _total_vm_committed = 0; - _total_malloced = 0; - _number_of_classes = 0; - - if (_malloc_cs != NULL) _malloc_cs->clear(); - if (_vm_cs != NULL) _vm_cs->clear(); - if (_vm_map != NULL) _vm_map->clear(); - - for (int index = 0; index < NUMBER_OF_MEMORY_TYPE; index ++) { - _malloc_data[index].clear(); - _vm_data[index].clear(); - _arena_data[index].clear(); - } +int compare_virtual_memory_site(const VirtualMemoryAllocationSite& s1, + const VirtualMemoryAllocationSite& s2) { + return s1.call_stack()->compare(*s2.call_stack()); } -MemBaseline::~MemBaseline() { - clear(); +/* + * Walker to walk malloc allocation site table + */ +class MallocAllocationSiteWalker : public MallocSiteWalker { + private: + SortedLinkedList<MallocSite, compare_malloc_size, ResourceObj::ARENA> + _malloc_sites; + size_t _count; + + // Entries in MallocSiteTable with size = 0 and count = 0, + // when the malloc site is not longer there. + public: + MallocAllocationSiteWalker(Arena* arena) : _count(0), _malloc_sites(arena) { + } + + inline size_t count() const { return _count; } + + LinkedList<MallocSite>* malloc_sites() { + return &_malloc_sites; + } + + bool do_malloc_site(const MallocSite* site) { + if (site->size() >= MemBaseline::SIZE_THRESHOLD) { + if (_malloc_sites.add(*site) != NULL) { + _count++; + return true; + } else { + return false; // OOM + } + } else { + // malloc site does not meet threshold, ignore and continue + return true; + } + } +}; + +// Compare virtual memory region's base address +int compare_virtual_memory_base(const ReservedMemoryRegion& r1, const ReservedMemoryRegion& r2) { + return r1.compare(r2); } -// baseline malloc'd memory records, generate overall summary and summaries by -// memory types -bool MemBaseline::baseline_malloc_summary(const MemPointerArray* malloc_records) { - MemPointerArrayIteratorImpl malloc_itr((MemPointerArray*)malloc_records); - MemPointerRecord* malloc_ptr = (MemPointerRecord*)malloc_itr.current(); - size_t used_arena_size = 0; - int index; - while (malloc_ptr != NULL) { - index = flag2index(FLAGS_TO_MEMORY_TYPE(malloc_ptr->flags())); - size_t size = malloc_ptr->size(); - if (malloc_ptr->is_arena_memory_record()) { - // We do have anonymous arenas, they are either used as value objects, - // which are embedded inside other objects, or used as stack objects. - _arena_data[index].inc(size); - used_arena_size += size; - } else { - _total_malloced += size; - _malloc_data[index].inc(size); - if (malloc_ptr->is_arena_record()) { - // see if arena memory record present - MemPointerRecord* next_malloc_ptr = (MemPointerRecordEx*)malloc_itr.peek_next(); - if (next_malloc_ptr != NULL && next_malloc_ptr->is_arena_memory_record()) { - assert(next_malloc_ptr->is_memory_record_of_arena(malloc_ptr), - "Arena records do not match"); - size = next_malloc_ptr->size(); - _arena_data[index].inc(size); - used_arena_size += size; - malloc_itr.next(); - } +// Walk all virtual memory regions for baselining +class VirtualMemoryAllocationWalker : public VirtualMemoryWalker { + private: + SortedLinkedList<ReservedMemoryRegion, compare_virtual_memory_base, ResourceObj::ARENA> + _virtual_memory_regions; + size_t _count; + + public: + VirtualMemoryAllocationWalker(Arena* a) : _count(0), _virtual_memory_regions(a) { + } + + bool do_allocation_site(const ReservedMemoryRegion* rgn) { + if (rgn->size() >= MemBaseline::SIZE_THRESHOLD) { + if (_virtual_memory_regions.add(*rgn) != NULL) { + _count ++; + return true; + } else { + return false; } } - malloc_ptr = (MemPointerRecordEx*)malloc_itr.next(); + return true; } - // substract used arena size to get size of arena chunk in free list - index = flag2index(mtChunk); - _malloc_data[index].reduce(used_arena_size); - // we really don't know how many chunks in free list, so just set to - // 0 - _malloc_data[index].overwrite_counter(0); + LinkedList<ReservedMemoryRegion>* virtual_memory_allocations() { + return &_virtual_memory_regions; + } +}; + + +bool MemBaseline::baseline_summary() { + assert(_malloc_memory_snapshot == NULL, "Malloc baseline not yet reset"); + assert(_virtual_memory_snapshot == NULL, "Virtual baseline not yet reset"); + + _malloc_memory_snapshot = new (arena()) MallocMemorySnapshot(); + _virtual_memory_snapshot = new (arena()) VirtualMemorySnapshot(); + if (_malloc_memory_snapshot == NULL || _virtual_memory_snapshot == NULL) { + return false; + } + MallocMemorySummary::snapshot(_malloc_memory_snapshot); + VirtualMemorySummary::snapshot(_virtual_memory_snapshot); + return true; +} + +bool MemBaseline::baseline_allocation_sites() { + assert(arena() != NULL, "Just check"); + // Malloc allocation sites + MallocAllocationSiteWalker malloc_walker(arena()); + if (!MallocSiteTable::walk_malloc_site(&malloc_walker)) { + return false; + } + + _malloc_sites.set_head(malloc_walker.malloc_sites()->head()); + // The malloc sites are collected in size order + _malloc_sites_order = by_size; + + // Virtual memory allocation sites + VirtualMemoryAllocationWalker virtual_memory_walker(arena()); + if (!VirtualMemoryTracker::walk_virtual_memory(&virtual_memory_walker)) { + return false; + } + + // Virtual memory allocations are collected in call stack order + _virtual_memory_allocations.set_head(virtual_memory_walker.virtual_memory_allocations()->head()); + + if (!aggregate_virtual_memory_allocation_sites()) { + return false; + } + // Virtual memory allocation sites are aggregrated in call stack order + _virtual_memory_sites_order = by_address; return true; } -// check if there is a safepoint in progress, if so, block the thread -// for the safepoint -void MemBaseline::check_safepoint(JavaThread* thr) { - if (SafepointSynchronize::is_synchronizing()) { - // grab and drop the SR_lock to honor the safepoint protocol - MutexLocker ml(thr->SR_lock()); - } -} - -// baseline mmap'd memory records, generate overall summary and summaries by -// memory types -bool MemBaseline::baseline_vm_summary(const MemPointerArray* vm_records) { - MemPointerArrayIteratorImpl vm_itr((MemPointerArray*)vm_records); - VMMemRegion* vm_ptr = (VMMemRegion*)vm_itr.current(); - int index; - while (vm_ptr != NULL) { - if (vm_ptr->is_reserved_region()) { - index = flag2index(FLAGS_TO_MEMORY_TYPE(vm_ptr->flags())); - // we use the number of thread stack to count threads - if (IS_MEMORY_TYPE(vm_ptr->flags(), mtThreadStack)) { - _number_of_threads ++; - } - _total_vm_reserved += vm_ptr->size(); - _vm_data[index].inc(vm_ptr->size(), 0); - } else { - _total_vm_committed += vm_ptr->size(); - _vm_data[index].inc(0, vm_ptr->size()); - } - vm_ptr = (VMMemRegion*)vm_itr.next(); - } - return true; -} - -// baseline malloc'd memory by callsites, but only the callsites with memory allocation -// over 1KB are stored. -bool MemBaseline::baseline_malloc_details(const MemPointerArray* malloc_records) { - assert(MemTracker::track_callsite(), "detail tracking is off"); - - MemPointerArrayIteratorImpl malloc_itr(const_cast<MemPointerArray*>(malloc_records)); - MemPointerRecordEx* malloc_ptr = (MemPointerRecordEx*)malloc_itr.current(); - MallocCallsitePointer malloc_callsite; - - // initailize malloc callsite array - if (_malloc_cs == NULL) { - _malloc_cs = new (std::nothrow) MemPointerArrayImpl<MallocCallsitePointer>(64); - // out of native memory - if (_malloc_cs == NULL || _malloc_cs->out_of_memory()) { - return false; - } - } else { - _malloc_cs->clear(); - } - - MemPointerArray* malloc_data = const_cast<MemPointerArray*>(malloc_records); - - // sort into callsite pc order. Details are aggregated by callsites - malloc_data->sort((FN_SORT)malloc_sort_by_pc); - bool ret = true; - - // baseline memory that is totaled over 1 KB - while (malloc_ptr != NULL) { - if (!MemPointerRecord::is_arena_memory_record(malloc_ptr->flags())) { - // skip thread stacks - if (!IS_MEMORY_TYPE(malloc_ptr->flags(), mtThreadStack)) { - if (malloc_callsite.addr() != malloc_ptr->pc()) { - if ((malloc_callsite.amount()/K) > 0) { - if (!_malloc_cs->append(&malloc_callsite)) { - ret = false; - break; - } - } - malloc_callsite = MallocCallsitePointer(malloc_ptr->pc()); - } - malloc_callsite.inc(malloc_ptr->size()); - } - } - malloc_ptr = (MemPointerRecordEx*)malloc_itr.next(); +bool MemBaseline::baseline(bool summaryOnly) { + if (arena() == NULL) { + _arena = new (std::nothrow, mtNMT) Arena(mtNMT); + if (arena() == NULL) return false; } - // restore to address order. Snapshot malloc data is maintained in memory - // address order. - malloc_data->sort((FN_SORT)malloc_sort_by_addr); - - if (!ret) { - return false; - } - // deal with last record - if (malloc_callsite.addr() != 0 && (malloc_callsite.amount()/K) > 0) { - if (!_malloc_cs->append(&malloc_callsite)) { - return false; - } - } - return true; -} - -// baseline mmap'd memory by callsites -bool MemBaseline::baseline_vm_details(const MemPointerArray* vm_records) { - assert(MemTracker::track_callsite(), "detail tracking is off"); - - VMCallsitePointer vm_callsite; - VMCallsitePointer* cur_callsite = NULL; - MemPointerArrayIteratorImpl vm_itr((MemPointerArray*)vm_records); - VMMemRegionEx* vm_ptr = (VMMemRegionEx*)vm_itr.current(); - - // initialize virtual memory map array - if (_vm_map == NULL) { - _vm_map = new (std::nothrow) MemPointerArrayImpl<VMMemRegionEx>(vm_records->length()); - if (_vm_map == NULL || _vm_map->out_of_memory()) { - return false; - } - } else { - _vm_map->clear(); - } - - // initialize virtual memory callsite array - if (_vm_cs == NULL) { - _vm_cs = new (std::nothrow) MemPointerArrayImpl<VMCallsitePointer>(64); - if (_vm_cs == NULL || _vm_cs->out_of_memory()) { - return false; - } - } else { - _vm_cs->clear(); - } - - // consolidate virtual memory data - VMMemRegionEx* reserved_rec = NULL; - VMMemRegionEx* committed_rec = NULL; + reset(); - // vm_ptr is coming in increasing base address order - while (vm_ptr != NULL) { - if (vm_ptr->is_reserved_region()) { - // consolidate reserved memory regions for virtual memory map. - // The criteria for consolidation is: - // 1. two adjacent reserved memory regions - // 2. belong to the same memory type - // 3. reserved from the same callsite - if (reserved_rec == NULL || - reserved_rec->base() + reserved_rec->size() != vm_ptr->addr() || - FLAGS_TO_MEMORY_TYPE(reserved_rec->flags()) != FLAGS_TO_MEMORY_TYPE(vm_ptr->flags()) || - reserved_rec->pc() != vm_ptr->pc()) { - if (!_vm_map->append(vm_ptr)) { - return false; - } - // inserted reserved region, we need the pointer to the element in virtual - // memory map array. - reserved_rec = (VMMemRegionEx*)_vm_map->at(_vm_map->length() - 1); - } else { - reserved_rec->expand_region(vm_ptr->addr(), vm_ptr->size()); - } + _class_count = InstanceKlass::number_of_instance_classes(); - if (cur_callsite != NULL && !_vm_cs->append(cur_callsite)) { - return false; - } - vm_callsite = VMCallsitePointer(vm_ptr->pc()); - cur_callsite = &vm_callsite; - vm_callsite.inc(vm_ptr->size(), 0); - } else { - // consolidate committed memory regions for virtual memory map - // The criterial is: - // 1. two adjacent committed memory regions - // 2. committed from the same callsite - if (committed_rec == NULL || - committed_rec->base() + committed_rec->size() != vm_ptr->addr() || - committed_rec->pc() != vm_ptr->pc()) { - if (!_vm_map->append(vm_ptr)) { - return false; - } - committed_rec = (VMMemRegionEx*)_vm_map->at(_vm_map->length() - 1); - } else { - committed_rec->expand_region(vm_ptr->addr(), vm_ptr->size()); - } - vm_callsite.inc(0, vm_ptr->size()); - } - vm_ptr = (VMMemRegionEx*)vm_itr.next(); - } - // deal with last record - if (cur_callsite != NULL && !_vm_cs->append(cur_callsite)) { + if (!baseline_summary()) { return false; } - // sort it into callsite pc order. Details are aggregated by callsites - _vm_cs->sort((FN_SORT)bl_vm_sort_by_pc); + _baseline_type = Summary_baselined; - // walk the array to consolidate record by pc - MemPointerArrayIteratorImpl itr(_vm_cs); - VMCallsitePointer* callsite_rec = (VMCallsitePointer*)itr.current(); - VMCallsitePointer* next_rec = (VMCallsitePointer*)itr.next(); - while (next_rec != NULL) { - assert(callsite_rec != NULL, "Sanity check"); - if (next_rec->addr() == callsite_rec->addr()) { - callsite_rec->inc(next_rec->reserved_amount(), next_rec->committed_amount()); - itr.remove(); - next_rec = (VMCallsitePointer*)itr.current(); - } else { - callsite_rec = next_rec; - next_rec = (VMCallsitePointer*)itr.next(); - } + // baseline details + if (!summaryOnly && + MemTracker::tracking_level() == NMT_detail) { + baseline_allocation_sites(); + _baseline_type = Detail_baselined; } return true; } -// baseline a snapshot. If summary_only = false, memory usages aggregated by -// callsites are also baselined. -// The method call can be lengthy, especially when detail tracking info is -// requested. So the method checks for safepoint explicitly. -bool MemBaseline::baseline(MemSnapshot& snapshot, bool summary_only) { - Thread* THREAD = Thread::current(); - assert(THREAD->is_Java_thread(), "must be a JavaThread"); - MutexLocker snapshot_locker(snapshot._lock); - reset(); - _baselined = baseline_malloc_summary(snapshot._alloc_ptrs); - if (_baselined) { - check_safepoint((JavaThread*)THREAD); - _baselined = baseline_vm_summary(snapshot._vm_ptrs); +int compare_allocation_site(const VirtualMemoryAllocationSite& s1, + const VirtualMemoryAllocationSite& s2) { + return s1.call_stack()->compare(*s2.call_stack()); +} + +bool MemBaseline::aggregate_virtual_memory_allocation_sites() { + SortedLinkedList<VirtualMemoryAllocationSite, compare_allocation_site, ResourceObj::ARENA> + allocation_sites(arena()); + + VirtualMemoryAllocationIterator itr = virtual_memory_allocations(); + const ReservedMemoryRegion* rgn; + VirtualMemoryAllocationSite* site; + while ((rgn = itr.next()) != NULL) { + VirtualMemoryAllocationSite tmp(*rgn->call_stack()); + site = allocation_sites.find(tmp); + if (site == NULL) { + LinkedListNode<VirtualMemoryAllocationSite>* node = + allocation_sites.add(tmp); + if (node == NULL) return false; + site = node->data(); + } + site->reserve_memory(rgn->size()); + site->commit_memory(rgn->committed_size()); } - _number_of_classes = snapshot.number_of_classes(); - if (!summary_only && MemTracker::track_callsite() && _baselined) { - check_safepoint((JavaThread*)THREAD); - _baselined = baseline_malloc_details(snapshot._alloc_ptrs); - if (_baselined) { - check_safepoint((JavaThread*)THREAD); - _baselined = baseline_vm_details(snapshot._vm_ptrs); - } - } - return _baselined; + _virtual_memory_sites.set_head(allocation_sites.head()); + return true; } - -int MemBaseline::flag2index(MEMFLAGS flag) const { - for (int index = 0; index < NUMBER_OF_MEMORY_TYPE; index ++) { - if (MemType2NameMap[index]._flag == flag) { - return index; - } +MallocSiteIterator MemBaseline::malloc_sites(SortingOrder order) { + assert(!_malloc_sites.is_empty(), "Detail baseline?"); + switch(order) { + case by_size: + malloc_sites_to_size_order(); + break; + case by_site: + malloc_sites_to_allocation_site_order(); + break; + case by_address: + default: + ShouldNotReachHere(); } - assert(false, "no type"); - return -1; + return MallocSiteIterator(_malloc_sites.head()); } -const char* MemBaseline::type2name(MEMFLAGS type) { - for (int index = 0; index < NUMBER_OF_MEMORY_TYPE; index ++) { - if (MemType2NameMap[index]._flag == type) { - return MemType2NameMap[index]._name; - } +VirtualMemorySiteIterator MemBaseline::virtual_memory_sites(SortingOrder order) { + assert(!_virtual_memory_sites.is_empty(), "Detail baseline?"); + switch(order) { + case by_size: + virtual_memory_sites_to_size_order(); + break; + case by_site: + virtual_memory_sites_to_reservation_site_order(); + break; + case by_address: + default: + ShouldNotReachHere(); } - assert(false, err_msg("bad type %x", type)); - return NULL; + return VirtualMemorySiteIterator(_virtual_memory_sites.head()); } -MemBaseline& MemBaseline::operator=(const MemBaseline& other) { - _total_malloced = other._total_malloced; - _total_vm_reserved = other._total_vm_reserved; - _total_vm_committed = other._total_vm_committed; - - _baselined = other._baselined; - _number_of_classes = other._number_of_classes; - - for (int index = 0; index < NUMBER_OF_MEMORY_TYPE; index ++) { - _malloc_data[index] = other._malloc_data[index]; - _vm_data[index] = other._vm_data[index]; - _arena_data[index] = other._arena_data[index]; - } +// Sorting allocations sites in different orders +void MemBaseline::malloc_sites_to_size_order() { + if (_malloc_sites_order != by_size) { + SortedLinkedList<MallocSite, compare_malloc_size, ResourceObj::ARENA> + tmp(arena()); - if (MemTracker::track_callsite()) { - assert(_malloc_cs != NULL && _vm_cs != NULL, "out of memory"); - assert(other._malloc_cs != NULL && other._vm_cs != NULL, - "not properly baselined"); - _malloc_cs->clear(); - _vm_cs->clear(); - int index; - for (index = 0; index < other._malloc_cs->length(); index ++) { - _malloc_cs->append(other._malloc_cs->at(index)); - } - - for (index = 0; index < other._vm_cs->length(); index ++) { - _vm_cs->append(other._vm_cs->at(index)); - } + // Add malloc sites to sorted linked list to sort into size order + tmp.move(&_malloc_sites); + _malloc_sites.set_head(tmp.head()); + tmp.set_head(NULL); + _malloc_sites_order = by_size; } - return *this; } -/* compare functions for sorting */ - -// sort snapshot malloc'd records in callsite pc order -int MemBaseline::malloc_sort_by_pc(const void* p1, const void* p2) { - assert(MemTracker::track_callsite(),"Just check"); - const MemPointerRecordEx* mp1 = (const MemPointerRecordEx*)p1; - const MemPointerRecordEx* mp2 = (const MemPointerRecordEx*)p2; - return UNSIGNED_COMPARE(mp1->pc(), mp2->pc()); +void MemBaseline::malloc_sites_to_allocation_site_order() { + if (_malloc_sites_order != by_site) { + SortedLinkedList<MallocSite, compare_malloc_site, ResourceObj::ARENA> + tmp(arena()); + // Add malloc sites to sorted linked list to sort into site (address) order + tmp.move(&_malloc_sites); + _malloc_sites.set_head(tmp.head()); + tmp.set_head(NULL); + _malloc_sites_order = by_site; + } } -// sort baselined malloc'd records in size order -int MemBaseline::bl_malloc_sort_by_size(const void* p1, const void* p2) { - assert(MemTracker::is_on(), "Just check"); - const MallocCallsitePointer* mp1 = (const MallocCallsitePointer*)p1; - const MallocCallsitePointer* mp2 = (const MallocCallsitePointer*)p2; - return UNSIGNED_COMPARE(mp2->amount(), mp1->amount()); -} +void MemBaseline::virtual_memory_sites_to_size_order() { + if (_virtual_memory_sites_order != by_size) { + SortedLinkedList<VirtualMemoryAllocationSite, compare_virtual_memory_size, ResourceObj::ARENA> + tmp(arena()); -// sort baselined malloc'd records in callsite pc order -int MemBaseline::bl_malloc_sort_by_pc(const void* p1, const void* p2) { - assert(MemTracker::is_on(), "Just check"); - const MallocCallsitePointer* mp1 = (const MallocCallsitePointer*)p1; - const MallocCallsitePointer* mp2 = (const MallocCallsitePointer*)p2; - return UNSIGNED_COMPARE(mp1->addr(), mp2->addr()); + tmp.move(&_virtual_memory_sites); + + _virtual_memory_sites.set_head(tmp.head()); + tmp.set_head(NULL); + _virtual_memory_sites_order = by_size; + } } +void MemBaseline::virtual_memory_sites_to_reservation_site_order() { + if (_virtual_memory_sites_order != by_size) { + SortedLinkedList<VirtualMemoryAllocationSite, compare_virtual_memory_site, ResourceObj::ARENA> + tmp(arena()); -// sort baselined mmap'd records in size (reserved size) order -int MemBaseline::bl_vm_sort_by_size(const void* p1, const void* p2) { - assert(MemTracker::is_on(), "Just check"); - const VMCallsitePointer* mp1 = (const VMCallsitePointer*)p1; - const VMCallsitePointer* mp2 = (const VMCallsitePointer*)p2; - return UNSIGNED_COMPARE(mp2->reserved_amount(), mp1->reserved_amount()); + tmp.add(&_virtual_memory_sites); + + _virtual_memory_sites.set_head(tmp.head()); + tmp.set_head(NULL); + + _virtual_memory_sites_order = by_size; + } } -// sort baselined mmap'd records in callsite pc order -int MemBaseline::bl_vm_sort_by_pc(const void* p1, const void* p2) { - assert(MemTracker::is_on(), "Just check"); - const VMCallsitePointer* mp1 = (const VMCallsitePointer*)p1; - const VMCallsitePointer* mp2 = (const VMCallsitePointer*)p2; - return UNSIGNED_COMPARE(mp1->addr(), mp2->addr()); -} - - -// sort snapshot malloc'd records in memory block address order -int MemBaseline::malloc_sort_by_addr(const void* p1, const void* p2) { - assert(MemTracker::is_on(), "Just check"); - const MemPointerRecord* mp1 = (const MemPointerRecord*)p1; - const MemPointerRecord* mp2 = (const MemPointerRecord*)p2; - int delta = UNSIGNED_COMPARE(mp1->addr(), mp2->addr()); - assert(p1 == p2 || delta != 0, "dup pointer"); - return delta; -} -
--- a/hotspot/src/share/vm/services/memBaseline.hpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/services/memBaseline.hpp Wed Aug 13 13:29:45 2014 -0700 @@ -1,5 +1,5 @@ /* - * Copyright (c) 2012, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012, 2014, 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 @@ -25,425 +25,205 @@ #ifndef SHARE_VM_SERVICES_MEM_BASELINE_HPP #define SHARE_VM_SERVICES_MEM_BASELINE_HPP +#if INCLUDE_NMT + #include "memory/allocation.hpp" #include "runtime/mutex.hpp" -#include "services/memPtr.hpp" -#include "services/memSnapshot.hpp" +#include "services/mallocSiteTable.hpp" +#include "services/mallocTracker.hpp" +#include "services/nmtCommon.hpp" +#include "services/virtualMemoryTracker.hpp" +#include "utilities/linkedlist.hpp" -// compare unsigned number -#define UNSIGNED_COMPARE(a, b) ((a > b) ? 1 : ((a == b) ? 0 : -1)) +typedef LinkedListIterator<MallocSite> MallocSiteIterator; +typedef LinkedListIterator<VirtualMemoryAllocationSite> VirtualMemorySiteIterator; +typedef LinkedListIterator<ReservedMemoryRegion> VirtualMemoryAllocationIterator; /* - * MallocCallsitePointer and VMCallsitePointer are used - * to baseline memory blocks with their callsite information. - * They are only available when detail tracking is turned - * on. + * Baseline a memory snapshot */ - -/* baselined malloc record aggregated by callsite */ -class MallocCallsitePointer : public MemPointer { - private: - size_t _count; // number of malloc invocation from this callsite - size_t _amount; // total amount of memory malloc-ed from this callsite - +class MemBaseline VALUE_OBJ_CLASS_SPEC { public: - MallocCallsitePointer() { - _count = 0; - _amount = 0; - } - - MallocCallsitePointer(address pc) : MemPointer(pc) { - _count = 0; - _amount = 0; - } + enum BaselineThreshold { + SIZE_THRESHOLD = K // Only allocation size over this threshold will be baselined. + }; - MallocCallsitePointer& operator=(const MallocCallsitePointer& p) { - MemPointer::operator=(p); - _count = p.count(); - _amount = p.amount(); - return *this; - } + enum BaselineType { + Not_baselined, + Summary_baselined, + Detail_baselined + }; - inline void inc(size_t size) { - _count ++; - _amount += size; + enum SortingOrder { + by_address, // by memory address + by_size, // by memory size + by_site // by call site where the memory is allocated from }; - inline size_t count() const { - return _count; - } + private: + // All baseline data is stored in this arena + Arena* _arena; + + // Summary information + MallocMemorySnapshot* _malloc_memory_snapshot; + VirtualMemorySnapshot* _virtual_memory_snapshot; + + size_t _class_count; - inline size_t amount() const { - return _amount; - } -}; + // Allocation sites information + // Malloc allocation sites + LinkedListImpl<MallocSite, ResourceObj::ARENA> + _malloc_sites; + + // All virtual memory allocations + LinkedListImpl<ReservedMemoryRegion, ResourceObj::ARENA> + _virtual_memory_allocations; -// baselined virtual memory record aggregated by callsite -class VMCallsitePointer : public MemPointer { - private: - size_t _count; // number of invocation from this callsite - size_t _reserved_amount; // total reserved amount - size_t _committed_amount; // total committed amount + // Virtual memory allocations by allocation sites, always in by_address + // order + LinkedListImpl<VirtualMemoryAllocationSite, ResourceObj::ARENA> + _virtual_memory_sites; + + SortingOrder _malloc_sites_order; + SortingOrder _virtual_memory_sites_order; + + BaselineType _baseline_type; public: - VMCallsitePointer() { - _count = 0; - _reserved_amount = 0; - _committed_amount = 0; - } - - VMCallsitePointer(address pc) : MemPointer(pc) { - _count = 0; - _reserved_amount = 0; - _committed_amount = 0; - } - - VMCallsitePointer& operator=(const VMCallsitePointer& p) { - MemPointer::operator=(p); - _count = p.count(); - _reserved_amount = p.reserved_amount(); - _committed_amount = p.committed_amount(); - return *this; - } - - inline void inc(size_t reserved, size_t committed) { - _count ++; - _reserved_amount += reserved; - _committed_amount += committed; - } - - inline size_t count() const { - return _count; - } - - inline size_t reserved_amount() const { - return _reserved_amount; + // create a memory baseline + MemBaseline(): + _baseline_type(Not_baselined), + _class_count(0), + _arena(NULL), + _malloc_memory_snapshot(NULL), + _virtual_memory_snapshot(NULL), + _malloc_sites(NULL) { } - inline size_t committed_amount() const { - return _committed_amount; - } -}; - -// maps a memory type flag to readable name -typedef struct _memType2Name { - MEMFLAGS _flag; - const char* _name; -} MemType2Name; - - -// This class aggregates malloc'd records by memory type -class MallocMem VALUE_OBJ_CLASS_SPEC { - private: - MEMFLAGS _type; - - size_t _count; - size_t _amount; - - public: - MallocMem() { - _type = mtNone; - _count = 0; - _amount = 0; + ~MemBaseline() { + reset(); + if (_arena != NULL) { + delete _arena; + } } - MallocMem(MEMFLAGS flags) { - assert(HAS_VALID_MEMORY_TYPE(flags), "no type"); - _type = FLAGS_TO_MEMORY_TYPE(flags); - _count = 0; - _amount = 0; - } + bool baseline(bool summaryOnly = true); - inline void set_type(MEMFLAGS flag) { - _type = flag; - } + BaselineType baseline_type() const { return _baseline_type; } - inline void clear() { - _count = 0; - _amount = 0; - _type = mtNone; + MallocMemorySnapshot* malloc_memory_snapshot() const { + return _malloc_memory_snapshot; } - MallocMem& operator=(const MallocMem& m) { - assert(_type == m.type(), "different type"); - _count = m.count(); - _amount = m.amount(); - return *this; + VirtualMemorySnapshot* virtual_memory_snapshot() const { + return _virtual_memory_snapshot; } - inline void inc(size_t amt) { - _amount += amt; - _count ++; - } + MallocSiteIterator malloc_sites(SortingOrder order); + VirtualMemorySiteIterator virtual_memory_sites(SortingOrder order); - inline void reduce(size_t amt) { - assert(_amount >= amt, "Just check"); - _amount -= amt; - } - - inline void overwrite_counter(size_t count) { - _count = count; + // Virtual memory allocation iterator always returns in virtual memory + // base address order. + VirtualMemoryAllocationIterator virtual_memory_allocations() { + assert(!_virtual_memory_allocations.is_empty(), "Not detail baseline"); + return VirtualMemoryAllocationIterator(_virtual_memory_allocations.head()); } - inline MEMFLAGS type() const { - return _type; - } - - inline bool is_type(MEMFLAGS flags) const { - return FLAGS_TO_MEMORY_TYPE(flags) == _type; - } - - inline size_t count() const { - return _count; + // Total reserved memory = total malloc'd memory + total reserved virtual + // memory + size_t total_reserved_memory() const { + assert(baseline_type() != Not_baselined, "Not yet baselined"); + assert(_virtual_memory_snapshot != NULL, "No virtual memory snapshot"); + assert(_malloc_memory_snapshot != NULL, "No malloc memory snapshot"); + size_t amount = _malloc_memory_snapshot->total() + + _virtual_memory_snapshot->total_reserved(); + return amount; } - inline size_t amount() const { - return _amount; - } -}; - -// This class records live arena's memory usage -class ArenaMem : public MallocMem { - public: - ArenaMem(MEMFLAGS typeflag): MallocMem(typeflag) { - } - ArenaMem() { } -}; - -// This class aggregates virtual memory by its memory type -class VMMem VALUE_OBJ_CLASS_SPEC { - private: - MEMFLAGS _type; - - size_t _count; - size_t _reserved_amount; - size_t _committed_amount; - - public: - VMMem() { - _type = mtNone; - _count = 0; - _reserved_amount = 0; - _committed_amount = 0; + // Total committed memory = total malloc'd memory + total committed + // virtual memory + size_t total_committed_memory() const { + assert(baseline_type() != Not_baselined, "Not yet baselined"); + assert(_virtual_memory_snapshot != NULL, + "Not a snapshot"); + size_t amount = _malloc_memory_snapshot->total() + + _virtual_memory_snapshot->total_committed(); + return amount; } - VMMem(MEMFLAGS flags) { - assert(HAS_VALID_MEMORY_TYPE(flags), "no type"); - _type = FLAGS_TO_MEMORY_TYPE(flags); - _count = 0; - _reserved_amount = 0; - _committed_amount = 0; + size_t total_arena_memory() const { + assert(baseline_type() != Not_baselined, "Not yet baselined"); + assert(_malloc_memory_snapshot != NULL, "Not yet baselined"); + return _malloc_memory_snapshot->total_arena(); } - inline void clear() { - _type = mtNone; - _count = 0; - _reserved_amount = 0; - _committed_amount = 0; + size_t malloc_tracking_overhead() const { + assert(baseline_type() != Not_baselined, "Not yet baselined"); + return _malloc_memory_snapshot->malloc_overhead()->size(); } - inline void set_type(MEMFLAGS flags) { - _type = FLAGS_TO_MEMORY_TYPE(flags); + const MallocMemory* malloc_memory(MEMFLAGS flag) const { + assert(_malloc_memory_snapshot != NULL, "Not a snapshot"); + return _malloc_memory_snapshot->by_type(flag); } - VMMem& operator=(const VMMem& m) { - assert(_type == m.type(), "different type"); - - _count = m.count(); - _reserved_amount = m.reserved_amount(); - _committed_amount = m.committed_amount(); - return *this; + const VirtualMemory* virtual_memory(MEMFLAGS flag) const { + assert(_virtual_memory_snapshot != NULL, "Not a snapshot"); + return _virtual_memory_snapshot->by_type(flag); } - inline MEMFLAGS type() const { - return _type; - } - - inline bool is_type(MEMFLAGS flags) const { - return FLAGS_TO_MEMORY_TYPE(flags) == _type; - } - - inline void inc(size_t reserved_amt, size_t committed_amt) { - _reserved_amount += reserved_amt; - _committed_amount += committed_amt; - _count ++; - } - - inline size_t count() const { - return _count; - } - - inline size_t reserved_amount() const { - return _reserved_amount; + size_t class_count() const { + assert(baseline_type() != Not_baselined, "Not yet baselined"); + return _class_count; } - inline size_t committed_amount() const { - return _committed_amount; + size_t thread_count() const { + assert(baseline_type() != Not_baselined, "Not yet baselined"); + assert(_malloc_memory_snapshot != NULL, "Baselined?"); + return _malloc_memory_snapshot->thread_count(); } -}; - - - -#define NUMBER_OF_MEMORY_TYPE (mt_number_of_types + 1) - -class BaselineReporter; -class BaselineComparisonReporter; - -/* - * This class baselines current memory snapshot. - * A memory baseline summarizes memory usage by memory type, - * aggregates memory usage by callsites when detail tracking - * is on. - */ -class MemBaseline VALUE_OBJ_CLASS_SPEC { - friend class BaselineReporter; - friend class BaselineComparisonReporter; - - private: - // overall summaries - size_t _total_malloced; - size_t _total_vm_reserved; - size_t _total_vm_committed; - size_t _number_of_classes; - size_t _number_of_threads; - - // if it has properly baselined - bool _baselined; - - // we categorize memory into three categories within the memory type - MallocMem _malloc_data[NUMBER_OF_MEMORY_TYPE]; - VMMem _vm_data[NUMBER_OF_MEMORY_TYPE]; - ArenaMem _arena_data[NUMBER_OF_MEMORY_TYPE]; - - // memory records that aggregate memory usage by callsites. - // only available when detail tracking is on. - MemPointerArray* _malloc_cs; - MemPointerArray* _vm_cs; - // virtual memory map - MemPointerArray* _vm_map; - - private: - static MemType2Name MemType2NameMap[NUMBER_OF_MEMORY_TYPE]; - - private: - // should not use copy constructor - MemBaseline(MemBaseline& copy) { ShouldNotReachHere(); } - - // check and block at a safepoint - static inline void check_safepoint(JavaThread* thr); - - public: - // create a memory baseline - MemBaseline(); - - ~MemBaseline(); - - inline bool baselined() const { - return _baselined; - } - - MemBaseline& operator=(const MemBaseline& other); // reset the baseline for reuse - void clear(); - - // baseline the snapshot - bool baseline(MemSnapshot& snapshot, bool summary_only = true); - - bool baseline(const MemPointerArray* malloc_records, - const MemPointerArray* vm_records, - bool summary_only = true); + void reset() { + _baseline_type = Not_baselined; + _malloc_memory_snapshot = NULL; + _virtual_memory_snapshot = NULL; + _class_count = 0; - // total malloc'd memory of specified memory type - inline size_t malloc_amount(MEMFLAGS flag) const { - return _malloc_data[flag2index(flag)].amount(); - } - // number of malloc'd memory blocks of specified memory type - inline size_t malloc_count(MEMFLAGS flag) const { - return _malloc_data[flag2index(flag)].count(); - } - // total memory used by arenas of specified memory type - inline size_t arena_amount(MEMFLAGS flag) const { - return _arena_data[flag2index(flag)].amount(); - } - // number of arenas of specified memory type - inline size_t arena_count(MEMFLAGS flag) const { - return _arena_data[flag2index(flag)].count(); - } - // total reserved memory of specified memory type - inline size_t reserved_amount(MEMFLAGS flag) const { - return _vm_data[flag2index(flag)].reserved_amount(); - } - // total committed memory of specified memory type - inline size_t committed_amount(MEMFLAGS flag) const { - return _vm_data[flag2index(flag)].committed_amount(); - } - // total memory (malloc'd + mmap'd + arena) of specified - // memory type - inline size_t total_amount(MEMFLAGS flag) const { - int index = flag2index(flag); - return _malloc_data[index].amount() + - _vm_data[index].reserved_amount() + - _arena_data[index].amount(); + _malloc_sites = NULL; + _virtual_memory_sites = NULL; + _virtual_memory_allocations = NULL; + + if (_arena != NULL) { + _arena->destruct_contents(); + } } - /* overall summaries */ + private: + // Baseline summary information + bool baseline_summary(); - // total malloc'd memory in snapshot - inline size_t total_malloc_amount() const { - return _total_malloced; - } - // total mmap'd memory in snapshot - inline size_t total_reserved_amount() const { - return _total_vm_reserved; - } - // total committed memory in snapshot - inline size_t total_committed_amount() const { - return _total_vm_committed; - } - // number of loaded classes - inline size_t number_of_classes() const { - return _number_of_classes; - } - // number of running threads - inline size_t number_of_threads() const { - return _number_of_threads; - } - // lookup human readable name of a memory type - static const char* type2name(MEMFLAGS type); + // Baseline allocation sites (detail tracking only) + bool baseline_allocation_sites(); + + // Aggregate virtual memory allocation by allocation sites + bool aggregate_virtual_memory_allocation_sites(); - private: - // convert memory flag to the index to mapping table - int flag2index(MEMFLAGS flag) const; - - // reset baseline values - void reset(); - - // summarize the records in global snapshot - bool baseline_malloc_summary(const MemPointerArray* malloc_records); - bool baseline_vm_summary(const MemPointerArray* vm_records); - bool baseline_malloc_details(const MemPointerArray* malloc_records); - bool baseline_vm_details(const MemPointerArray* vm_records); + Arena* arena() { return _arena; } - // print a line of malloc'd memory aggregated by callsite - void print_malloc_callsite(outputStream* st, address pc, size_t size, - size_t count, int diff_amt, int diff_count) const; - // print a line of mmap'd memory aggregated by callsite - void print_vm_callsite(outputStream* st, address pc, size_t rsz, - size_t csz, int diff_rsz, int diff_csz) const; + // Sorting allocation sites in different orders + // Sort allocation sites in size order + void malloc_sites_to_size_order(); + // Sort allocation sites in call site address order + void malloc_sites_to_allocation_site_order(); - // sorting functions for raw records - static int malloc_sort_by_pc(const void* p1, const void* p2); - static int malloc_sort_by_addr(const void* p1, const void* p2); - - private: - // sorting functions for baselined records - static int bl_malloc_sort_by_size(const void* p1, const void* p2); - static int bl_vm_sort_by_size(const void* p1, const void* p2); - static int bl_malloc_sort_by_pc(const void* p1, const void* p2); - static int bl_vm_sort_by_pc(const void* p1, const void* p2); + // Sort allocation sites in reserved size order + void virtual_memory_sites_to_size_order(); + // Sort allocation sites in call site address order + void virtual_memory_sites_to_reservation_site_order(); }; +#endif // INCLUDE_NMT #endif // SHARE_VM_SERVICES_MEM_BASELINE_HPP
--- a/hotspot/src/share/vm/services/memPtr.cpp Fri Aug 08 16:24:16 2014 -0700 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,43 +0,0 @@ -/* - * Copyright (c) 2012, 2013, 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. - * - */ - -#include "precompiled.hpp" -#include "runtime/atomic.inline.hpp" -#include "services/memPtr.hpp" -#include "services/memTracker.hpp" - -volatile jint SequenceGenerator::_seq_number = 1; -volatile unsigned long SequenceGenerator::_generation = 1; -NOT_PRODUCT(jint SequenceGenerator::_max_seq_number = 1;) - -jint SequenceGenerator::next() { - jint seq = Atomic::add(1, &_seq_number); - if (seq < 0) { - MemTracker::shutdown(MemTracker::NMT_sequence_overflow); - } else { - NOT_PRODUCT(_max_seq_number = (seq > _max_seq_number) ? seq : _max_seq_number;) - } - return seq; -} -
--- a/hotspot/src/share/vm/services/memPtr.hpp Fri Aug 08 16:24:16 2014 -0700 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,509 +0,0 @@ -/* - * Copyright (c) 2012, 2013, 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_SERVICES_MEM_PTR_HPP -#define SHARE_VM_SERVICES_MEM_PTR_HPP - -#include "memory/allocation.hpp" -#include "runtime/os.hpp" -#include "runtime/safepoint.hpp" - -/* - * global sequence generator that generates sequence numbers to serialize - * memory records. - */ -class SequenceGenerator : AllStatic { - public: - static jint next(); - - // peek last sequence number - static jint peek() { - return _seq_number; - } - - // reset sequence number - static void reset() { - assert(SafepointSynchronize::is_at_safepoint(), "Safepoint required"); - _seq_number = 1; - _generation ++; - }; - - static unsigned long current_generation() { return _generation; } - NOT_PRODUCT(static jint max_seq_num() { return _max_seq_number; }) - - private: - static volatile jint _seq_number; - static volatile unsigned long _generation; - NOT_PRODUCT(static jint _max_seq_number; ) -}; - -/* - * followings are the classes that are used to hold memory activity records in different stages. - * MemPointer - * |--------MemPointerRecord - * | - * |----MemPointerRecordEx - * | | - * | |-------SeqMemPointerRecordEx - * | - * |----SeqMemPointerRecord - * | - * |----VMMemRegion - * | - * |-----VMMemRegionEx - * - * - * prefix 'Seq' - sequenced, the record contains a sequence number - * surfix 'Ex' - extension, the record contains a caller's pc - * - * per-thread recorder : SeqMemPointerRecord(Ex) - * snapshot staging : SeqMemPointerRecord(Ex) - * snapshot : MemPointerRecord(Ex) and VMMemRegion(Ex) - * - */ - -/* - * class that wraps an address to a memory block, - * the memory pointer either points to a malloc'd - * memory block, or a mmap'd memory block - */ -class MemPointer VALUE_OBJ_CLASS_SPEC { - public: - MemPointer(): _addr(0) { } - MemPointer(address addr): _addr(addr) { } - - MemPointer(const MemPointer& copy_from) { - _addr = copy_from.addr(); - } - - inline address addr() const { - return _addr; - } - - inline operator address() const { - return addr(); - } - - inline bool operator == (const MemPointer& other) const { - return addr() == other.addr(); - } - - inline MemPointer& operator = (const MemPointer& other) { - _addr = other.addr(); - return *this; - } - - protected: - inline void set_addr(address addr) { _addr = addr; } - - protected: - // memory address - address _addr; -}; - -/* MemPointerRecord records an activityand associated - * attributes on a memory block. - */ -class MemPointerRecord : public MemPointer { - private: - MEMFLAGS _flags; - size_t _size; - -public: - /* extension of MemoryType enum - * see share/vm/memory/allocation.hpp for details. - * - * The tag values are associated to sorting orders, so be - * careful if changes are needed. - * The allocation records should be sorted ahead of tagging - * records, which in turn ahead of deallocation records - */ - enum MemPointerTags { - tag_alloc = 0x0001, // malloc or reserve record - tag_commit = 0x0002, // commit record - tag_type = 0x0003, // tag virtual memory to a memory type - tag_uncommit = 0x0004, // uncommit record - tag_release = 0x0005, // free or release record - tag_size = 0x0006, // arena size - tag_masks = 0x0007, // all tag bits - vmBit = 0x0008 - }; - - /* helper functions to interpret the tagging flags */ - - inline static bool is_allocation_record(MEMFLAGS flags) { - return (flags & tag_masks) == tag_alloc; - } - - inline static bool is_deallocation_record(MEMFLAGS flags) { - return (flags & tag_masks) == tag_release; - } - - inline static bool is_arena_record(MEMFLAGS flags) { - return (flags & (otArena | tag_size)) == otArena; - } - - inline static bool is_arena_memory_record(MEMFLAGS flags) { - return (flags & (otArena | tag_size)) == (otArena | tag_size); - } - - inline static bool is_virtual_memory_record(MEMFLAGS flags) { - return (flags & vmBit) != 0; - } - - inline static bool is_virtual_memory_reserve_record(MEMFLAGS flags) { - return (flags & 0x0F) == (tag_alloc | vmBit); - } - - inline static bool is_virtual_memory_commit_record(MEMFLAGS flags) { - return (flags & 0x0F) == (tag_commit | vmBit); - } - - inline static bool is_virtual_memory_uncommit_record(MEMFLAGS flags) { - return (flags & 0x0F) == (tag_uncommit | vmBit); - } - - inline static bool is_virtual_memory_release_record(MEMFLAGS flags) { - return (flags & 0x0F) == (tag_release | vmBit); - } - - inline static bool is_virtual_memory_type_record(MEMFLAGS flags) { - return (flags & 0x0F) == (tag_type | vmBit); - } - - /* tagging flags */ - inline static MEMFLAGS malloc_tag() { return tag_alloc; } - inline static MEMFLAGS free_tag() { return tag_release; } - inline static MEMFLAGS arena_size_tag() { return tag_size | otArena; } - inline static MEMFLAGS virtual_memory_tag() { return vmBit; } - inline static MEMFLAGS virtual_memory_reserve_tag() { return (tag_alloc | vmBit); } - inline static MEMFLAGS virtual_memory_commit_tag() { return (tag_commit | vmBit); } - inline static MEMFLAGS virtual_memory_uncommit_tag(){ return (tag_uncommit | vmBit); } - inline static MEMFLAGS virtual_memory_release_tag() { return (tag_release | vmBit); } - inline static MEMFLAGS virtual_memory_type_tag() { return (tag_type | vmBit); } - - public: - MemPointerRecord(): _size(0), _flags(mtNone) { } - - MemPointerRecord(address addr, MEMFLAGS memflags, size_t size = 0): - MemPointer(addr), _flags(memflags), _size(size) { } - - MemPointerRecord(const MemPointerRecord& copy_from): - MemPointer(copy_from), _flags(copy_from.flags()), - _size(copy_from.size()) { - } - - /* MemPointerRecord is not sequenced, it always return - * 0 to indicate non-sequenced - */ - virtual jint seq() const { return 0; } - - inline size_t size() const { return _size; } - inline void set_size(size_t size) { _size = size; } - - inline MEMFLAGS flags() const { return _flags; } - inline void set_flags(MEMFLAGS flags) { _flags = flags; } - - MemPointerRecord& operator= (const MemPointerRecord& ptr) { - MemPointer::operator=(ptr); - _flags = ptr.flags(); -#ifdef ASSERT - if (IS_ARENA_OBJ(_flags)) { - assert(!is_vm_pointer(), "wrong flags"); - assert((_flags & ot_masks) == otArena, "wrong flags"); - } -#endif - _size = ptr.size(); - return *this; - } - - // if the pointer represents a malloc-ed memory address - inline bool is_malloced_pointer() const { - return !is_vm_pointer(); - } - - // if the pointer represents a virtual memory address - inline bool is_vm_pointer() const { - return is_virtual_memory_record(_flags); - } - - // if this record records a 'malloc' or virtual memory - // 'reserve' call - inline bool is_allocation_record() const { - return is_allocation_record(_flags); - } - - // if this record records a size information of an arena - inline bool is_arena_memory_record() const { - return is_arena_memory_record(_flags); - } - - // if this pointer represents an address to an arena object - inline bool is_arena_record() const { - return is_arena_record(_flags); - } - - // if this record represents a size information of specific arena - inline bool is_memory_record_of_arena(const MemPointerRecord* arena_rc) { - assert(is_arena_memory_record(), "not size record"); - assert(arena_rc->is_arena_record(), "not arena record"); - return (arena_rc->addr() + sizeof(void*)) == addr(); - } - - // if this record records a 'free' or virtual memory 'free' call - inline bool is_deallocation_record() const { - return is_deallocation_record(_flags); - } - - // if this record records a virtual memory 'commit' call - inline bool is_commit_record() const { - return is_virtual_memory_commit_record(_flags); - } - - // if this record records a virtual memory 'uncommit' call - inline bool is_uncommit_record() const { - return is_virtual_memory_uncommit_record(_flags); - } - - // if this record is a tagging record of a virtual memory block - inline bool is_type_tagging_record() const { - return is_virtual_memory_type_record(_flags); - } - - // if the two memory pointer records actually represent the same - // memory block - inline bool is_same_region(const MemPointerRecord* other) const { - return (addr() == other->addr() && size() == other->size()); - } - - // if this memory region fully contains another one - inline bool contains_region(const MemPointerRecord* other) const { - return contains_region(other->addr(), other->size()); - } - - // if this memory region fully contains specified memory range - inline bool contains_region(address add, size_t sz) const { - return (addr() <= add && addr() + size() >= add + sz); - } - - inline bool contains_address(address add) const { - return (addr() <= add && addr() + size() > add); - } - - // if this memory region overlaps another region - inline bool overlaps_region(const MemPointerRecord* other) const { - assert(other != NULL, "Just check"); - assert(size() > 0 && other->size() > 0, "empty range"); - return contains_address(other->addr()) || - contains_address(other->addr() + other->size() - 1) || // exclude end address - other->contains_address(addr()) || - other->contains_address(addr() + size() - 1); // exclude end address - } - -}; - -// MemPointerRecordEx also records callsite pc, from where -// the memory block is allocated -class MemPointerRecordEx : public MemPointerRecord { - private: - address _pc; // callsite pc - - public: - MemPointerRecordEx(): _pc(0) { } - - MemPointerRecordEx(address addr, MEMFLAGS memflags, size_t size = 0, address pc = 0): - MemPointerRecord(addr, memflags, size), _pc(pc) {} - - MemPointerRecordEx(const MemPointerRecordEx& copy_from): - MemPointerRecord(copy_from), _pc(copy_from.pc()) {} - - inline address pc() const { return _pc; } - - void init(const MemPointerRecordEx* mpe) { - MemPointerRecord::operator=(*mpe); - _pc = mpe->pc(); - } - - void init(const MemPointerRecord* mp) { - MemPointerRecord::operator=(*mp); - _pc = 0; - } -}; - -// a virtual memory region. The region can represent a reserved -// virtual memory region or a committed memory region -class VMMemRegion : public MemPointerRecord { -public: - VMMemRegion() { } - - void init(const MemPointerRecord* mp) { - assert(mp->is_vm_pointer(), "Sanity check"); - _addr = mp->addr(); - set_size(mp->size()); - set_flags(mp->flags()); - } - - VMMemRegion& operator=(const VMMemRegion& other) { - MemPointerRecord::operator=(other); - return *this; - } - - inline bool is_reserved_region() const { - return is_allocation_record(); - } - - inline bool is_committed_region() const { - return is_commit_record(); - } - - /* base address of this virtual memory range */ - inline address base() const { - return addr(); - } - - /* tag this virtual memory range to the specified memory type */ - inline void tag(MEMFLAGS f) { - set_flags(flags() | (f & mt_masks)); - } - - // expand this region to also cover specified range. - // The range has to be on either end of the memory region. - void expand_region(address addr, size_t sz) { - if (addr < base()) { - assert(addr + sz == base(), "Sanity check"); - _addr = addr; - set_size(size() + sz); - } else { - assert(base() + size() == addr, "Sanity check"); - set_size(size() + sz); - } - } - - // exclude the specified address range from this region. - // The excluded memory range has to be on either end of this memory - // region. - inline void exclude_region(address add, size_t sz) { - assert(is_reserved_region() || is_committed_region(), "Sanity check"); - assert(addr() != NULL && size() != 0, "Sanity check"); - assert(add >= addr() && add < addr() + size(), "Sanity check"); - assert(add == addr() || (add + sz) == (addr() + size()), - "exclude in the middle"); - if (add == addr()) { - set_addr(add + sz); - set_size(size() - sz); - } else { - set_size(size() - sz); - } - } -}; - -class VMMemRegionEx : public VMMemRegion { - private: - jint _seq; // sequence number - - public: - VMMemRegionEx(): _pc(0) { } - - void init(const MemPointerRecordEx* mpe) { - VMMemRegion::init(mpe); - _pc = mpe->pc(); - } - - void init(const MemPointerRecord* mpe) { - VMMemRegion::init(mpe); - _pc = 0; - } - - VMMemRegionEx& operator=(const VMMemRegionEx& other) { - VMMemRegion::operator=(other); - _pc = other.pc(); - return *this; - } - - inline address pc() const { return _pc; } - private: - address _pc; -}; - -/* - * Sequenced memory record - */ -class SeqMemPointerRecord : public MemPointerRecord { - private: - jint _seq; // sequence number - - public: - SeqMemPointerRecord(): _seq(0){ } - - SeqMemPointerRecord(address addr, MEMFLAGS flags, size_t size, jint seq) - : MemPointerRecord(addr, flags, size), _seq(seq) { - } - - SeqMemPointerRecord(const SeqMemPointerRecord& copy_from) - : MemPointerRecord(copy_from) { - _seq = copy_from.seq(); - } - - SeqMemPointerRecord& operator= (const SeqMemPointerRecord& ptr) { - MemPointerRecord::operator=(ptr); - _seq = ptr.seq(); - return *this; - } - - inline jint seq() const { - return _seq; - } -}; - - - -class SeqMemPointerRecordEx : public MemPointerRecordEx { - private: - jint _seq; // sequence number - - public: - SeqMemPointerRecordEx(): _seq(0) { } - - SeqMemPointerRecordEx(address addr, MEMFLAGS flags, size_t size, - jint seq, address pc): - MemPointerRecordEx(addr, flags, size, pc), _seq(seq) { - } - - SeqMemPointerRecordEx(const SeqMemPointerRecordEx& copy_from) - : MemPointerRecordEx(copy_from) { - _seq = copy_from.seq(); - } - - SeqMemPointerRecordEx& operator= (const SeqMemPointerRecordEx& ptr) { - MemPointerRecordEx::operator=(ptr); - _seq = ptr.seq(); - return *this; - } - - inline jint seq() const { - return _seq; - } -}; - -#endif // SHARE_VM_SERVICES_MEM_PTR_HPP
--- a/hotspot/src/share/vm/services/memPtrArray.hpp Fri Aug 08 16:24:16 2014 -0700 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,306 +0,0 @@ -/* - * Copyright (c) 2012, 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_UTILITIES_MEM_PTR_ARRAY_HPP -#define SHARE_VM_UTILITIES_MEM_PTR_ARRAY_HPP - -#include "memory/allocation.hpp" -#include "services/memPtr.hpp" - -class MemPtr; -class MemRecorder; -class ArenaInfo; -class MemSnapshot; - -extern "C" { - typedef int (*FN_SORT)(const void *, const void *); -} - - -// Memory pointer array interface. This array is used by NMT to hold -// various memory block information. -// The memory pointer arrays are usually walked with their iterators. - -class MemPointerArray : public CHeapObj<mtNMT> { - public: - virtual ~MemPointerArray() { } - - // return true if it can not allocate storage for the data - virtual bool out_of_memory() const = 0; - virtual bool is_empty() const = 0; - virtual bool is_full() = 0; - virtual int length() const = 0; - virtual void clear() = 0; - virtual bool append(MemPointer* ptr) = 0; - virtual bool insert_at(MemPointer* ptr, int pos) = 0; - virtual bool remove_at(int pos) = 0; - virtual MemPointer* at(int index) const = 0; - virtual void sort(FN_SORT fn) = 0; - virtual size_t instance_size() const = 0; - virtual bool shrink() = 0; - - NOT_PRODUCT(virtual int capacity() const = 0;) -}; - -// Iterator interface -class MemPointerArrayIterator VALUE_OBJ_CLASS_SPEC { - public: - // return the pointer at current position - virtual MemPointer* current() const = 0; - // return the next pointer and advance current position - virtual MemPointer* next() = 0; - // return next pointer without advancing current position - virtual MemPointer* peek_next() const = 0; - // return previous pointer without changing current position - virtual MemPointer* peek_prev() const = 0; - // remove the pointer at current position - virtual void remove() = 0; - // insert the pointer at current position - virtual bool insert(MemPointer* ptr) = 0; - // insert specified element after current position and - // move current position to newly inserted position - virtual bool insert_after(MemPointer* ptr) = 0; -}; - -// implementation class -class MemPointerArrayIteratorImpl : public MemPointerArrayIterator { - protected: - MemPointerArray* _array; - int _pos; - - public: - MemPointerArrayIteratorImpl(MemPointerArray* arr) { - assert(arr != NULL, "Parameter check"); - _array = arr; - _pos = 0; - } - - virtual MemPointer* current() const { - if (_pos < _array->length()) { - return _array->at(_pos); - } - return NULL; - } - - virtual MemPointer* next() { - if (_pos + 1 < _array->length()) { - return _array->at(++_pos); - } - _pos = _array->length(); - return NULL; - } - - virtual MemPointer* peek_next() const { - if (_pos + 1 < _array->length()) { - return _array->at(_pos + 1); - } - return NULL; - } - - virtual MemPointer* peek_prev() const { - if (_pos > 0) { - return _array->at(_pos - 1); - } - return NULL; - } - - virtual void remove() { - if (_pos < _array->length()) { - _array->remove_at(_pos); - } - } - - virtual bool insert(MemPointer* ptr) { - return _array->insert_at(ptr, _pos); - } - - virtual bool insert_after(MemPointer* ptr) { - if (_array->insert_at(ptr, _pos + 1)) { - _pos ++; - return true; - } - return false; - } -}; - - - -// Memory pointer array implementation. -// This implementation implements expandable array -#define DEFAULT_PTR_ARRAY_SIZE 1024 - -template <class E> class MemPointerArrayImpl : public MemPointerArray { - private: - int _max_size; - int _size; - bool _init_elements; - E* _data; - - public: - MemPointerArrayImpl(int initial_size = DEFAULT_PTR_ARRAY_SIZE, bool init_elements = true): - _max_size(initial_size), _size(0), _init_elements(init_elements) { - _data = (E*)raw_allocate(sizeof(E), initial_size); - if (_init_elements) { - for (int index = 0; index < _max_size; index ++) { - ::new ((void*)&_data[index]) E(); - } - } - } - - virtual ~MemPointerArrayImpl() { - if (_data != NULL) { - raw_free(_data); - } - } - - public: - bool out_of_memory() const { - return (_data == NULL); - } - - size_t instance_size() const { - return sizeof(MemPointerArrayImpl<E>) + _max_size * sizeof(E); - } - - bool is_empty() const { - assert(_data != NULL, "Just check"); - return _size == 0; - } - - bool is_full() { - assert(_data != NULL, "Just check"); - if (_size < _max_size) { - return false; - } else { - return !expand_array(); - } - } - - int length() const { - assert(_data != NULL, "Just check"); - return _size; - } - - NOT_PRODUCT(int capacity() const { return _max_size; }) - - void clear() { - assert(_data != NULL, "Just check"); - _size = 0; - } - - bool append(MemPointer* ptr) { - assert(_data != NULL, "Just check"); - if (is_full()) { - return false; - } - _data[_size ++] = *(E*)ptr; - return true; - } - - bool insert_at(MemPointer* ptr, int pos) { - assert(_data != NULL, "Just check"); - if (is_full()) { - return false; - } - for (int index = _size; index > pos; index --) { - _data[index] = _data[index - 1]; - } - _data[pos] = *(E*)ptr; - _size ++; - return true; - } - - bool remove_at(int pos) { - assert(_data != NULL, "Just check"); - if (_size <= pos && pos >= 0) { - return false; - } - -- _size; - - for (int index = pos; index < _size; index ++) { - _data[index] = _data[index + 1]; - } - return true; - } - - MemPointer* at(int index) const { - assert(_data != NULL, "Just check"); - assert(index >= 0 && index < _size, "illegal index"); - return &_data[index]; - } - - bool shrink() { - float used = ((float)_size) / ((float)_max_size); - if (used < 0.40) { - E* old_ptr = _data; - int new_size = ((_max_size) / (2 * DEFAULT_PTR_ARRAY_SIZE) + 1) * DEFAULT_PTR_ARRAY_SIZE; - _data = (E*)raw_reallocate(_data, sizeof(E), new_size); - if (_data == NULL) { - _data = old_ptr; - return false; - } else { - _max_size = new_size; - return true; - } - } - return false; - } - - void sort(FN_SORT fn) { - assert(_data != NULL, "Just check"); - qsort((void*)_data, _size, sizeof(E), fn); - } - - private: - bool expand_array() { - assert(_data != NULL, "Not yet allocated"); - E* old_ptr = _data; - if ((_data = (E*)raw_reallocate((void*)_data, sizeof(E), - _max_size + DEFAULT_PTR_ARRAY_SIZE)) == NULL) { - _data = old_ptr; - return false; - } else { - _max_size += DEFAULT_PTR_ARRAY_SIZE; - if (_init_elements) { - for (int index = _size; index < _max_size; index ++) { - ::new ((void*)&_data[index]) E(); - } - } - return true; - } - } - - void* raw_allocate(size_t elementSize, int items) { - return os::malloc(elementSize * items, mtNMT); - } - - void* raw_reallocate(void* ptr, size_t elementSize, int items) { - return os::realloc(ptr, elementSize * items, mtNMT); - } - - void raw_free(void* ptr) { - os::free(ptr, mtNMT); - } -}; - -#endif // SHARE_VM_UTILITIES_MEM_PTR_ARRAY_HPP
--- a/hotspot/src/share/vm/services/memRecorder.cpp Fri Aug 08 16:24:16 2014 -0700 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,171 +0,0 @@ -/* - * Copyright (c) 2012, 2013, 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. - * - */ - -#include "precompiled.hpp" - -#include "runtime/atomic.inline.hpp" -#include "services/memBaseline.hpp" -#include "services/memRecorder.hpp" -#include "services/memPtr.hpp" -#include "services/memTracker.hpp" - -MemPointer* SequencedRecordIterator::next_record() { - MemPointerRecord* itr_cur = (MemPointerRecord*)_itr.current(); - if (itr_cur == NULL) { - return itr_cur; - } - - MemPointerRecord* itr_next = (MemPointerRecord*)_itr.next(); - - // don't collapse virtual memory records - while (itr_next != NULL && !itr_cur->is_vm_pointer() && - !itr_next->is_vm_pointer() && - same_kind(itr_cur, itr_next)) { - itr_cur = itr_next; - itr_next = (MemPointerRecord*)_itr.next(); - } - - return itr_cur; -} - - -volatile jint MemRecorder::_instance_count = 0; - -MemRecorder::MemRecorder() { - assert(MemTracker::is_on(), "Native memory tracking is off"); - Atomic::inc(&_instance_count); - set_generation(); - - if (MemTracker::track_callsite()) { - _pointer_records = new (std::nothrow)FixedSizeMemPointerArray<SeqMemPointerRecordEx, - DEFAULT_RECORDER_PTR_ARRAY_SIZE>(); - } else { - _pointer_records = new (std::nothrow)FixedSizeMemPointerArray<SeqMemPointerRecord, - DEFAULT_RECORDER_PTR_ARRAY_SIZE>(); - } - _next = NULL; - - - if (_pointer_records != NULL) { - // recode itself - address pc = CURRENT_PC; - record((address)this, (MemPointerRecord::malloc_tag()|mtNMT|otNMTRecorder), - sizeof(MemRecorder), SequenceGenerator::next(), pc); - record((address)_pointer_records, (MemPointerRecord::malloc_tag()|mtNMT|otNMTRecorder), - _pointer_records->instance_size(), SequenceGenerator::next(), pc); - } -} - -MemRecorder::~MemRecorder() { - if (_pointer_records != NULL) { - if (MemTracker::is_on()) { - MemTracker::record_free((address)_pointer_records, mtNMT); - MemTracker::record_free((address)this, mtNMT); - } - delete _pointer_records; - } - // delete all linked recorders - while (_next != NULL) { - MemRecorder* tmp = _next; - _next = _next->next(); - tmp->set_next(NULL); - delete tmp; - } - Atomic::dec(&_instance_count); -} - -// Sorting order: -// 1. memory block address -// 2. mem pointer record tags -// 3. sequence number -int MemRecorder::sort_record_fn(const void* e1, const void* e2) { - const MemPointerRecord* p1 = (const MemPointerRecord*)e1; - const MemPointerRecord* p2 = (const MemPointerRecord*)e2; - int delta = UNSIGNED_COMPARE(p1->addr(), p2->addr()); - if (delta == 0) { - int df = UNSIGNED_COMPARE((p1->flags() & MemPointerRecord::tag_masks), - (p2->flags() & MemPointerRecord::tag_masks)); - if (df == 0) { - assert(p1->seq() != p2->seq(), "dup seq"); - return p1->seq() - p2->seq(); - } else { - return df; - } - } else { - return delta; - } -} - -bool MemRecorder::record(address p, MEMFLAGS flags, size_t size, jint seq, address pc) { - assert(seq > 0, "No sequence number"); -#ifdef ASSERT - if (MemPointerRecord::is_virtual_memory_record(flags)) { - assert((flags & MemPointerRecord::tag_masks) != 0, "bad virtual memory record"); - } else { - assert((flags & MemPointerRecord::tag_masks) == MemPointerRecord::malloc_tag() || - (flags & MemPointerRecord::tag_masks) == MemPointerRecord::free_tag() || - IS_ARENA_OBJ(flags), - "bad malloc record"); - } - // a recorder should only hold records within the same generation - unsigned long cur_generation = SequenceGenerator::current_generation(); - assert(cur_generation == _generation, - "this thread did not enter sync point"); -#endif - - if (MemTracker::track_callsite()) { - SeqMemPointerRecordEx ap(p, flags, size, seq, pc); - debug_only(check_dup_seq(ap.seq());) - return _pointer_records->append(&ap); - } else { - SeqMemPointerRecord ap(p, flags, size, seq); - debug_only(check_dup_seq(ap.seq());) - return _pointer_records->append(&ap); - } -} - - // iterator for alloc pointers -SequencedRecordIterator MemRecorder::pointer_itr() { - assert(_pointer_records != NULL, "just check"); - _pointer_records->sort((FN_SORT)sort_record_fn); - return SequencedRecordIterator(_pointer_records); -} - - -void MemRecorder::set_generation() { - _generation = SequenceGenerator::current_generation(); -} - -#ifdef ASSERT - -void MemRecorder::check_dup_seq(jint seq) const { - MemPointerArrayIteratorImpl itr(_pointer_records); - MemPointerRecord* rc = (MemPointerRecord*)itr.current(); - while (rc != NULL) { - assert(rc->seq() != seq, "dup seq"); - rc = (MemPointerRecord*)itr.next(); - } -} - -#endif
--- a/hotspot/src/share/vm/services/memRecorder.hpp Fri Aug 08 16:24:16 2014 -0700 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,271 +0,0 @@ -/* - * Copyright (c) 2012, 2013, 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_SERVICES_MEM_RECORDER_HPP -#define SHARE_VM_SERVICES_MEM_RECORDER_HPP - -#include "memory/allocation.hpp" -#include "runtime/os.hpp" -#include "services/memPtrArray.hpp" - -class MemSnapshot; -class MemTracker; -class MemTrackWorker; - -// Fixed size memory pointer array implementation -template <class E, int SIZE> class FixedSizeMemPointerArray : - public MemPointerArray { - // This implementation is for memory recorder only - friend class MemRecorder; - - private: - E _data[SIZE]; - int _size; - - protected: - FixedSizeMemPointerArray(bool init_elements = false): - _size(0){ - if (init_elements) { - for (int index = 0; index < SIZE; index ++) { - ::new ((void*)&_data[index]) E(); - } - } - } - - void* operator new(size_t size, const std::nothrow_t& nothrow_constant) throw() { - // the instance is part of memRecorder, needs to be tagged with 'otNMTRecorder' - // to avoid recursion - return os::malloc(size, (mtNMT | otNMTRecorder)); - } - - void* operator new(size_t size) throw() { - assert(false, "use nothrow version"); - return NULL; - } - - void operator delete(void* p) { - os::free(p, (mtNMT | otNMTRecorder)); - } - - // instance size - inline size_t instance_size() const { - return sizeof(FixedSizeMemPointerArray<E, SIZE>); - } - - NOT_PRODUCT(int capacity() const { return SIZE; }) - - public: - // implementation of public interface - bool out_of_memory() const { return false; } - bool is_empty() const { return _size == 0; } - bool is_full() { return length() >= SIZE; } - int length() const { return _size; } - - void clear() { - _size = 0; - } - - bool append(MemPointer* ptr) { - if (is_full()) return false; - _data[_size ++] = *(E*)ptr; - return true; - } - - virtual bool insert_at(MemPointer* p, int pos) { - assert(false, "append only"); - return false; - } - - virtual bool remove_at(int pos) { - assert(false, "not supported"); - return false; - } - - MemPointer* at(int index) const { - assert(index >= 0 && index < length(), - "parameter check"); - return ((E*)&_data[index]); - } - - void sort(FN_SORT fn) { - qsort((void*)_data, _size, sizeof(E), fn); - } - - bool shrink() { - return false; - } -}; - - -// This iterator requires pre-sorted MemPointerArray, which is sorted by: -// 1. address -// 2. allocation type -// 3. sequence number -// During the array walking, iterator collapses pointers with the same -// address and allocation type, and only returns the one with highest -// sequence number. -// -// This is read-only iterator, update methods are asserted. -class SequencedRecordIterator : public MemPointerArrayIterator { - private: - MemPointerArrayIteratorImpl _itr; - MemPointer* _cur; - - public: - SequencedRecordIterator(const MemPointerArray* arr): - _itr(const_cast<MemPointerArray*>(arr)) { - _cur = next_record(); - } - - SequencedRecordIterator(const SequencedRecordIterator& itr): - _itr(itr._itr) { - _cur = next_record(); - } - - // return the pointer at current position - virtual MemPointer* current() const { - return _cur; - }; - - // return the next pointer and advance current position - virtual MemPointer* next() { - _cur = next_record(); - return _cur; - } - - // return the next pointer without advancing current position - virtual MemPointer* peek_next() const { - assert(false, "not implemented"); - return NULL; - - } - // return the previous pointer without changing current position - virtual MemPointer* peek_prev() const { - assert(false, "not implemented"); - return NULL; - } - - // remove the pointer at current position - virtual void remove() { - assert(false, "read-only iterator"); - }; - // insert the pointer at current position - virtual bool insert(MemPointer* ptr) { - assert(false, "read-only iterator"); - return false; - } - - virtual bool insert_after(MemPointer* ptr) { - assert(false, "read-only iterator"); - return false; - } - private: - // collapse the 'same kind' of records, and return this 'kind' of - // record with highest sequence number - MemPointer* next_record(); - - // Test if the two records are the same kind: the same memory block and allocation - // type. - inline bool same_kind(const MemPointerRecord* p1, const MemPointerRecord* p2) const { - assert(!p1->is_vm_pointer() && !p2->is_vm_pointer(), "malloc pointer only"); - return (p1->addr() == p2->addr() && - (p1->flags() &MemPointerRecord::tag_masks) == - (p2->flags() & MemPointerRecord::tag_masks)); - } -}; - - - -#define DEFAULT_RECORDER_PTR_ARRAY_SIZE 512 - -class MemRecorder : public CHeapObj<mtNMT|otNMTRecorder> { - friend class MemSnapshot; - friend class MemTracker; - friend class MemTrackWorker; - friend class GenerationData; - - protected: - // the array that holds memory records - MemPointerArray* _pointer_records; - - private: - // used for linked list - MemRecorder* _next; - // active recorder can only record a certain generation data - unsigned long _generation; - - protected: - _NOINLINE_ MemRecorder(); - ~MemRecorder(); - - // record a memory operation - bool record(address addr, MEMFLAGS flags, size_t size, jint seq, address caller_pc = 0); - - // linked list support - inline void set_next(MemRecorder* rec) { - _next = rec; - } - - inline MemRecorder* next() const { - return _next; - } - - // if the recorder is full - inline bool is_full() const { - assert(_pointer_records != NULL, "just check"); - return _pointer_records->is_full(); - } - - // if running out of memory when initializing recorder's internal - // data - inline bool out_of_memory() const { - return (_pointer_records == NULL || - _pointer_records->out_of_memory()); - } - - inline void clear() { - assert(_pointer_records != NULL, "Just check"); - _pointer_records->clear(); - } - - SequencedRecordIterator pointer_itr(); - - // return the generation of this recorder which it belongs to - unsigned long get_generation() const { return _generation; } - protected: - // number of MemRecorder instance - static volatile jint _instance_count; - - private: - // sorting function, sort records into following order - // 1. memory address - // 2. allocation type - // 3. sequence number - static int sort_record_fn(const void* e1, const void* e2); - - debug_only(void check_dup_seq(jint seq) const;) - void set_generation(); -}; - -#endif // SHARE_VM_SERVICES_MEM_RECORDER_HPP
--- a/hotspot/src/share/vm/services/memReporter.cpp Fri Aug 08 16:24:16 2014 -0700 +++ b/hotspot/src/share/vm/services/memReporter.cpp Wed Aug 13 13:29:45 2014 -0700 @@ -22,618 +22,595 @@ * */ #include "precompiled.hpp" -#include "classfile/systemDictionary.hpp" -#include "runtime/os.hpp" + +#include "memory/allocation.hpp" +#include "services/mallocTracker.hpp" #include "services/memReporter.hpp" -#include "services/memPtrArray.hpp" -#include "services/memTracker.hpp" +#include "services/virtualMemoryTracker.hpp" +#include "utilities/globalDefinitions.hpp" + +size_t MemReporterBase::reserved_total(const MallocMemory* malloc, const VirtualMemory* vm) const { + return malloc->malloc_size() + malloc->arena_size() + vm->reserved(); +} + +size_t MemReporterBase::committed_total(const MallocMemory* malloc, const VirtualMemory* vm) const { + return malloc->malloc_size() + malloc->arena_size() + vm->committed(); +} -PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC +void MemReporterBase::print_total(size_t reserved, size_t committed) const { + const char* scale = current_scale(); + output()->print("reserved=" SIZE_FORMAT "%s, committed=" SIZE_FORMAT "%s", + amount_in_current_scale(reserved), scale, amount_in_current_scale(committed), scale); +} + +void MemReporterBase::print_malloc(size_t amount, size_t count) const { + const char* scale = current_scale(); + outputStream* out = output(); + out->print("(malloc=" SIZE_FORMAT "%s", + amount_in_current_scale(amount), scale); + + if (count > 0) { + out->print(" #" SIZE_FORMAT "", count); + } -const char* BaselineOutputer::memory_unit(size_t scale) { - switch(scale) { - case K: return "KB"; - case M: return "MB"; - case G: return "GB"; - } - ShouldNotReachHere(); - return NULL; + out->print(")"); +} + +void MemReporterBase::print_virtual_memory(size_t reserved, size_t committed) const { + const char* scale = current_scale(); + output()->print("(mmap: reserved=" SIZE_FORMAT "%s, committed=" SIZE_FORMAT "%s)", + amount_in_current_scale(reserved), scale, amount_in_current_scale(committed), scale); +} + +void MemReporterBase::print_malloc_line(size_t amount, size_t count) const { + output()->print("%28s", " "); + print_malloc(amount, count); + output()->print_cr(" "); +} + +void MemReporterBase::print_virtual_memory_line(size_t reserved, size_t committed) const { + output()->print("%28s", " "); + print_virtual_memory(reserved, committed); + output()->print_cr(" "); +} + +void MemReporterBase::print_arena_line(size_t amount, size_t count) const { + const char* scale = current_scale(); + output()->print_cr("%27s (arena=" SIZE_FORMAT "%s #" SIZE_FORMAT ")", " ", + amount_in_current_scale(amount), scale, count); +} + +void MemReporterBase::print_virtual_memory_region(const char* type, address base, size_t size) const { + const char* scale = current_scale(); + output()->print("[" PTR_FORMAT " - " PTR_FORMAT "] %s " SIZE_FORMAT "%s", + p2i(base), p2i(base + size), type, amount_in_current_scale(size), scale); } -void BaselineReporter::report_baseline(const MemBaseline& baseline, bool summary_only) { - assert(MemTracker::is_on(), "Native memory tracking is off"); - _outputer.start(scale()); - _outputer.total_usage( - amount_in_current_scale(baseline.total_malloc_amount() + baseline.total_reserved_amount()), - amount_in_current_scale(baseline.total_malloc_amount() + baseline.total_committed_amount())); - - _outputer.num_of_classes(baseline.number_of_classes()); - _outputer.num_of_threads(baseline.number_of_threads()); - - report_summaries(baseline); - if (!summary_only && MemTracker::track_callsite()) { - report_virtual_memory_map(baseline); - report_callsites(baseline); - } - _outputer.done(); -} - -void BaselineReporter::report_summaries(const MemBaseline& baseline) { - _outputer.start_category_summary(); - MEMFLAGS type; +void MemSummaryReporter::report() { + const char* scale = current_scale(); + outputStream* out = output(); + size_t total_reserved_amount = _malloc_snapshot->total() + + _vm_snapshot->total_reserved(); + size_t total_committed_amount = _malloc_snapshot->total() + + _vm_snapshot->total_committed(); - for (int index = 0; index < NUMBER_OF_MEMORY_TYPE; index ++) { - type = MemBaseline::MemType2NameMap[index]._flag; - _outputer.category_summary(type, - amount_in_current_scale(baseline.reserved_amount(type)), - amount_in_current_scale(baseline.committed_amount(type)), - amount_in_current_scale(baseline.malloc_amount(type)), - baseline.malloc_count(type), - amount_in_current_scale(baseline.arena_amount(type)), - baseline.arena_count(type)); - } - - _outputer.done_category_summary(); -} + // Overall total + out->print_cr("\nNative Memory Tracking:\n"); + out->print("Total: "); + print_total(total_reserved_amount, total_committed_amount); + out->print("\n"); -void BaselineReporter::report_virtual_memory_map(const MemBaseline& baseline) { - _outputer.start_virtual_memory_map(); - MemBaseline* pBL = const_cast<MemBaseline*>(&baseline); - MemPointerArrayIteratorImpl itr = MemPointerArrayIteratorImpl(pBL->_vm_map); - VMMemRegionEx* rgn = (VMMemRegionEx*)itr.current(); - while (rgn != NULL) { - if (rgn->is_reserved_region()) { - _outputer.reserved_memory_region(FLAGS_TO_MEMORY_TYPE(rgn->flags()), - rgn->base(), rgn->base() + rgn->size(), amount_in_current_scale(rgn->size()), rgn->pc()); - } else { - _outputer.committed_memory_region(rgn->base(), rgn->base() + rgn->size(), - amount_in_current_scale(rgn->size()), rgn->pc()); - } - rgn = (VMMemRegionEx*)itr.next(); + // Summary by memory type + for (int index = 0; index < mt_number_of_types; index ++) { + MEMFLAGS flag = NMTUtil::index_to_flag(index); + // thread stack is reported as part of thread category + if (flag == mtThreadStack) continue; + MallocMemory* malloc_memory = _malloc_snapshot->by_type(flag); + VirtualMemory* virtual_memory = _vm_snapshot->by_type(flag); + + report_summary_of_type(flag, malloc_memory, virtual_memory); } - - _outputer.done_virtual_memory_map(); } -void BaselineReporter::report_callsites(const MemBaseline& baseline) { - _outputer.start_callsite(); - MemBaseline* pBL = const_cast<MemBaseline*>(&baseline); +void MemSummaryReporter::report_summary_of_type(MEMFLAGS flag, + MallocMemory* malloc_memory, VirtualMemory* virtual_memory) { - pBL->_malloc_cs->sort((FN_SORT)MemBaseline::bl_malloc_sort_by_size); - pBL->_vm_cs->sort((FN_SORT)MemBaseline::bl_vm_sort_by_size); + size_t reserved_amount = reserved_total (malloc_memory, virtual_memory); + size_t committed_amount = committed_total(malloc_memory, virtual_memory); - // walk malloc callsites - MemPointerArrayIteratorImpl malloc_itr(pBL->_malloc_cs); - MallocCallsitePointer* malloc_callsite = - (MallocCallsitePointer*)malloc_itr.current(); - while (malloc_callsite != NULL) { - _outputer.malloc_callsite(malloc_callsite->addr(), - amount_in_current_scale(malloc_callsite->amount()), malloc_callsite->count()); - malloc_callsite = (MallocCallsitePointer*)malloc_itr.next(); + // Count thread's native stack in "Thread" category + if (flag == mtThread) { + const VirtualMemory* thread_stack_usage = + (const VirtualMemory*)_vm_snapshot->by_type(mtThreadStack); + reserved_amount += thread_stack_usage->reserved(); + committed_amount += thread_stack_usage->committed(); + } else if (flag == mtNMT) { + // Count malloc headers in "NMT" category + reserved_amount += _malloc_snapshot->malloc_overhead()->size(); + committed_amount += _malloc_snapshot->malloc_overhead()->size(); } - // walk virtual memory callsite - MemPointerArrayIteratorImpl vm_itr(pBL->_vm_cs); - VMCallsitePointer* vm_callsite = (VMCallsitePointer*)vm_itr.current(); - while (vm_callsite != NULL) { - _outputer.virtual_memory_callsite(vm_callsite->addr(), - amount_in_current_scale(vm_callsite->reserved_amount()), - amount_in_current_scale(vm_callsite->committed_amount())); - vm_callsite = (VMCallsitePointer*)vm_itr.next(); - } - pBL->_malloc_cs->sort((FN_SORT)MemBaseline::bl_malloc_sort_by_pc); - pBL->_vm_cs->sort((FN_SORT)MemBaseline::bl_vm_sort_by_pc); - _outputer.done_callsite(); -} + if (amount_in_current_scale(reserved_amount) > 0) { + outputStream* out = output(); + const char* scale = current_scale(); + out->print("-%26s (", NMTUtil::flag_to_name(flag)); + print_total(reserved_amount, committed_amount); + out->print_cr(")"); -void BaselineReporter::diff_baselines(const MemBaseline& cur, const MemBaseline& prev, - bool summary_only) { - assert(MemTracker::is_on(), "Native memory tracking is off"); - _outputer.start(scale()); - size_t total_reserved = cur.total_malloc_amount() + cur.total_reserved_amount(); - size_t total_committed = cur.total_malloc_amount() + cur.total_committed_amount(); - - _outputer.diff_total_usage( - amount_in_current_scale(total_reserved), amount_in_current_scale(total_committed), - diff_in_current_scale(total_reserved, (prev.total_malloc_amount() + prev.total_reserved_amount())), - diff_in_current_scale(total_committed, (prev.total_committed_amount() + prev.total_malloc_amount()))); + if (flag == mtClass) { + // report class count + out->print_cr("%27s (classes #" SIZE_FORMAT ")", " ", _class_count); + } else if (flag == mtThread) { + // report thread count + out->print_cr("%27s (thread #" SIZE_FORMAT ")", " ", _malloc_snapshot->thread_count()); + const VirtualMemory* thread_stack_usage = + _vm_snapshot->by_type(mtThreadStack); + out->print("%27s (stack: ", " "); + print_total(thread_stack_usage->reserved(), thread_stack_usage->committed()); + out->print_cr(")"); + } - _outputer.diff_num_of_classes(cur.number_of_classes(), - diff(cur.number_of_classes(), prev.number_of_classes())); - _outputer.diff_num_of_threads(cur.number_of_threads(), - diff(cur.number_of_threads(), prev.number_of_threads())); + // report malloc'd memory + if (amount_in_current_scale(malloc_memory->malloc_size()) > 0) { + // We don't know how many arena chunks are in used, so don't report the count + size_t count = (flag == mtChunk) ? 0 : malloc_memory->malloc_count(); + print_malloc_line(malloc_memory->malloc_size(), count); + } - diff_summaries(cur, prev); - if (!summary_only && MemTracker::track_callsite()) { - diff_callsites(cur, prev); - } - _outputer.done(); -} - -void BaselineReporter::diff_summaries(const MemBaseline& cur, const MemBaseline& prev) { - _outputer.start_category_summary(); - MEMFLAGS type; + if (amount_in_current_scale(virtual_memory->reserved()) > 0) { + print_virtual_memory_line(virtual_memory->reserved(), virtual_memory->committed()); + } - for (int index = 0; index < NUMBER_OF_MEMORY_TYPE; index ++) { - type = MemBaseline::MemType2NameMap[index]._flag; - _outputer.diff_category_summary(type, - amount_in_current_scale(cur.reserved_amount(type)), - amount_in_current_scale(cur.committed_amount(type)), - amount_in_current_scale(cur.malloc_amount(type)), - cur.malloc_count(type), - amount_in_current_scale(cur.arena_amount(type)), - cur.arena_count(type), - diff_in_current_scale(cur.reserved_amount(type), prev.reserved_amount(type)), - diff_in_current_scale(cur.committed_amount(type), prev.committed_amount(type)), - diff_in_current_scale(cur.malloc_amount(type), prev.malloc_amount(type)), - diff(cur.malloc_count(type), prev.malloc_count(type)), - diff_in_current_scale(cur.arena_amount(type), prev.arena_amount(type)), - diff(cur.arena_count(type), prev.arena_count(type))); + if (amount_in_current_scale(malloc_memory->arena_size()) > 0) { + print_arena_line(malloc_memory->arena_size(), malloc_memory->arena_count()); + } + + if (flag == mtNMT && + amount_in_current_scale(_malloc_snapshot->malloc_overhead()->size()) > 0) { + out->print_cr("%27s (tracking overhead=" SIZE_FORMAT "%s)", " ", + amount_in_current_scale(_malloc_snapshot->malloc_overhead()->size()), scale); + } + + out->print_cr(" "); } - - _outputer.done_category_summary(); } -void BaselineReporter::diff_callsites(const MemBaseline& cur, const MemBaseline& prev) { - _outputer.start_callsite(); - MemBaseline* pBL_cur = const_cast<MemBaseline*>(&cur); - MemBaseline* pBL_prev = const_cast<MemBaseline*>(&prev); +void MemDetailReporter::report_detail() { + // Start detail report + outputStream* out = output(); + out->print_cr("Details:\n"); + + report_malloc_sites(); + report_virtual_memory_allocation_sites(); +} + +void MemDetailReporter::report_malloc_sites() { + MallocSiteIterator malloc_itr = _baseline.malloc_sites(MemBaseline::by_size); + if (malloc_itr.is_empty()) return; + + outputStream* out = output(); - // walk malloc callsites - MemPointerArrayIteratorImpl cur_malloc_itr(pBL_cur->_malloc_cs); - MemPointerArrayIteratorImpl prev_malloc_itr(pBL_prev->_malloc_cs); + const MallocSite* malloc_site; + while ((malloc_site = malloc_itr.next()) != NULL) { + // Don't report if size is too small + if (amount_in_current_scale(malloc_site->size()) == 0) + continue; - MallocCallsitePointer* cur_malloc_callsite = - (MallocCallsitePointer*)cur_malloc_itr.current(); - MallocCallsitePointer* prev_malloc_callsite = - (MallocCallsitePointer*)prev_malloc_itr.current(); + const NativeCallStack* stack = malloc_site->call_stack(); + stack->print_on(out); + out->print("%29s", " "); + print_malloc(malloc_site->size(), malloc_site->count()); + out->print_cr("\n"); + } +} + +void MemDetailReporter::report_virtual_memory_allocation_sites() { + VirtualMemorySiteIterator virtual_memory_itr = + _baseline.virtual_memory_sites(MemBaseline::by_size); + + if (virtual_memory_itr.is_empty()) return; + + outputStream* out = output(); + const VirtualMemoryAllocationSite* virtual_memory_site; - while (cur_malloc_callsite != NULL || prev_malloc_callsite != NULL) { - if (prev_malloc_callsite == NULL) { - assert(cur_malloc_callsite != NULL, "sanity check"); - // this is a new callsite - _outputer.diff_malloc_callsite(cur_malloc_callsite->addr(), - amount_in_current_scale(cur_malloc_callsite->amount()), - cur_malloc_callsite->count(), - diff_in_current_scale(cur_malloc_callsite->amount(), 0), - diff(cur_malloc_callsite->count(), 0)); - cur_malloc_callsite = (MallocCallsitePointer*)cur_malloc_itr.next(); - } else if (cur_malloc_callsite == NULL) { - assert(prev_malloc_callsite != NULL, "Sanity check"); - // this callsite is already gone - _outputer.diff_malloc_callsite(prev_malloc_callsite->addr(), - 0, 0, - diff_in_current_scale(0, prev_malloc_callsite->amount()), - diff(0, prev_malloc_callsite->count())); - prev_malloc_callsite = (MallocCallsitePointer*)prev_malloc_itr.next(); - } else { - assert(cur_malloc_callsite != NULL, "Sanity check"); - assert(prev_malloc_callsite != NULL, "Sanity check"); - if (cur_malloc_callsite->addr() < prev_malloc_callsite->addr()) { - // this is a new callsite - _outputer.diff_malloc_callsite(cur_malloc_callsite->addr(), - amount_in_current_scale(cur_malloc_callsite->amount()), - cur_malloc_callsite->count(), - diff_in_current_scale(cur_malloc_callsite->amount(), 0), - diff(cur_malloc_callsite->count(), 0)); - cur_malloc_callsite = (MallocCallsitePointer*)cur_malloc_itr.next(); - } else if (cur_malloc_callsite->addr() > prev_malloc_callsite->addr()) { - // this callsite is already gone - _outputer.diff_malloc_callsite(prev_malloc_callsite->addr(), - 0, 0, - diff_in_current_scale(0, prev_malloc_callsite->amount()), - diff(0, prev_malloc_callsite->count())); - prev_malloc_callsite = (MallocCallsitePointer*)prev_malloc_itr.next(); - } else { - // the same callsite - _outputer.diff_malloc_callsite(cur_malloc_callsite->addr(), - amount_in_current_scale(cur_malloc_callsite->amount()), - cur_malloc_callsite->count(), - diff_in_current_scale(cur_malloc_callsite->amount(), prev_malloc_callsite->amount()), - diff(cur_malloc_callsite->count(), prev_malloc_callsite->count())); - cur_malloc_callsite = (MallocCallsitePointer*)cur_malloc_itr.next(); - prev_malloc_callsite = (MallocCallsitePointer*)prev_malloc_itr.next(); - } - } + while ((virtual_memory_site = virtual_memory_itr.next()) != NULL) { + // Don't report if size is too small + if (amount_in_current_scale(virtual_memory_site->reserved()) == 0) + continue; + + const NativeCallStack* stack = virtual_memory_site->call_stack(); + stack->print_on(out); + out->print("%28s (", " "); + print_total(virtual_memory_site->reserved(), virtual_memory_site->committed()); + out->print_cr(")\n"); + } +} + + +void MemDetailReporter::report_virtual_memory_map() { + // Virtual memory map always in base address order + VirtualMemoryAllocationIterator itr = _baseline.virtual_memory_allocations(); + const ReservedMemoryRegion* rgn; + + output()->print_cr("Virtual memory map:"); + while ((rgn = itr.next()) != NULL) { + report_virtual_memory_region(rgn); + } +} + +void MemDetailReporter::report_virtual_memory_region(const ReservedMemoryRegion* reserved_rgn) { + assert(reserved_rgn != NULL, "NULL pointer"); + + // Don't report if size is too small + if (amount_in_current_scale(reserved_rgn->size()) == 0) return; + + outputStream* out = output(); + const char* scale = current_scale(); + const NativeCallStack* stack = reserved_rgn->call_stack(); + bool all_committed = reserved_rgn->all_committed(); + const char* region_type = (all_committed ? "reserved and committed" : "reserved"); + out->print_cr(" "); + print_virtual_memory_region(region_type, reserved_rgn->base(), reserved_rgn->size()); + out->print(" for %s", NMTUtil::flag_to_name(reserved_rgn->flag())); + if (stack->is_empty()) { + out->print_cr(" "); + } else { + out->print_cr(" from"); + stack->print_on(out, 4); } - // walk virtual memory callsite - MemPointerArrayIteratorImpl cur_vm_itr(pBL_cur->_vm_cs); - MemPointerArrayIteratorImpl prev_vm_itr(pBL_prev->_vm_cs); - VMCallsitePointer* cur_vm_callsite = (VMCallsitePointer*)cur_vm_itr.current(); - VMCallsitePointer* prev_vm_callsite = (VMCallsitePointer*)prev_vm_itr.current(); - while (cur_vm_callsite != NULL || prev_vm_callsite != NULL) { - if (prev_vm_callsite == NULL || cur_vm_callsite->addr() < prev_vm_callsite->addr()) { - // this is a new callsite - _outputer.diff_virtual_memory_callsite(cur_vm_callsite->addr(), - amount_in_current_scale(cur_vm_callsite->reserved_amount()), - amount_in_current_scale(cur_vm_callsite->committed_amount()), - diff_in_current_scale(cur_vm_callsite->reserved_amount(), 0), - diff_in_current_scale(cur_vm_callsite->committed_amount(), 0)); - cur_vm_callsite = (VMCallsitePointer*)cur_vm_itr.next(); - } else if (cur_vm_callsite == NULL || cur_vm_callsite->addr() > prev_vm_callsite->addr()) { - // this callsite is already gone - _outputer.diff_virtual_memory_callsite(prev_vm_callsite->addr(), - amount_in_current_scale(0), - amount_in_current_scale(0), - diff_in_current_scale(0, prev_vm_callsite->reserved_amount()), - diff_in_current_scale(0, prev_vm_callsite->committed_amount())); - prev_vm_callsite = (VMCallsitePointer*)prev_vm_itr.next(); - } else { // the same callsite - _outputer.diff_virtual_memory_callsite(cur_vm_callsite->addr(), - amount_in_current_scale(cur_vm_callsite->reserved_amount()), - amount_in_current_scale(cur_vm_callsite->committed_amount()), - diff_in_current_scale(cur_vm_callsite->reserved_amount(), prev_vm_callsite->reserved_amount()), - diff_in_current_scale(cur_vm_callsite->committed_amount(), prev_vm_callsite->committed_amount())); - cur_vm_callsite = (VMCallsitePointer*)cur_vm_itr.next(); - prev_vm_callsite = (VMCallsitePointer*)prev_vm_itr.next(); - } - } - - _outputer.done_callsite(); -} - -size_t BaselineReporter::amount_in_current_scale(size_t amt) const { - return (size_t)(((float)amt/(float)_scale) + 0.5); -} - -int BaselineReporter::diff_in_current_scale(size_t value1, size_t value2) const { - return (int)(((float)value1 - (float)value2)/((float)_scale) + 0.5); -} - -int BaselineReporter::diff(size_t value1, size_t value2) const { - return ((int)value1 - (int)value2); -} - -void BaselineTTYOutputer::start(size_t scale, bool report_diff) { - _scale = scale; - _output->print_cr(" "); - _output->print_cr("Native Memory Tracking:"); - _output->print_cr(" "); -} - -void BaselineTTYOutputer::done() { - -} + if (all_committed) return; -void BaselineTTYOutputer::total_usage(size_t total_reserved, size_t total_committed) { - const char* unit = memory_unit(_scale); - _output->print_cr("Total: reserved=%d%s, committed=%d%s", - total_reserved, unit, total_committed, unit); -} - -void BaselineTTYOutputer::start_category_summary() { - _output->print_cr(" "); -} - -/** - * report a summary of memory type - */ -void BaselineTTYOutputer::category_summary(MEMFLAGS type, - size_t reserved_amt, size_t committed_amt, size_t malloc_amt, - size_t malloc_count, size_t arena_amt, size_t arena_count) { - - // we report mtThreadStack under mtThread category - if (type == mtThreadStack) { - assert(malloc_amt == 0 && malloc_count == 0 && arena_amt == 0, - "Just check"); - _thread_stack_reserved = reserved_amt; - _thread_stack_committed = committed_amt; - } else { - const char* unit = memory_unit(_scale); - size_t total_reserved = (reserved_amt + malloc_amt + arena_amt); - size_t total_committed = (committed_amt + malloc_amt + arena_amt); - if (type == mtThread) { - total_reserved += _thread_stack_reserved; - total_committed += _thread_stack_committed; - } - - if (total_reserved > 0) { - _output->print_cr("-%26s (reserved=%d%s, committed=%d%s)", - MemBaseline::type2name(type), total_reserved, unit, - total_committed, unit); - - if (type == mtClass) { - _output->print_cr("%27s (classes #%d)", " ", _num_of_classes); - } else if (type == mtThread) { - _output->print_cr("%27s (thread #%d)", " ", _num_of_threads); - _output->print_cr("%27s (stack: reserved=%d%s, committed=%d%s)", " ", - _thread_stack_reserved, unit, _thread_stack_committed, unit); - } - - if (malloc_amt > 0) { - if (type != mtChunk) { - _output->print_cr("%27s (malloc=%d%s, #%d)", " ", malloc_amt, unit, - malloc_count); - } else { - _output->print_cr("%27s (malloc=%d%s)", " ", malloc_amt, unit); - } - } - - if (reserved_amt > 0) { - _output->print_cr("%27s (mmap: reserved=%d%s, committed=%d%s)", - " ", reserved_amt, unit, committed_amt, unit); - } - - if (arena_amt > 0) { - _output->print_cr("%27s (arena=%d%s, #%d)", " ", arena_amt, unit, arena_count); - } - - _output->print_cr(" "); + CommittedRegionIterator itr = reserved_rgn->iterate_committed_regions(); + const CommittedMemoryRegion* committed_rgn; + while ((committed_rgn = itr.next()) != NULL) { + // Don't report if size is too small + if (amount_in_current_scale(committed_rgn->size()) == 0) continue; + stack = committed_rgn->call_stack(); + out->print("\n\t"); + print_virtual_memory_region("committed", committed_rgn->base(), committed_rgn->size()); + if (stack->is_empty()) { + out->print_cr(" "); + } else { + out->print_cr(" from"); + stack->print_on(out, 12); } } } -void BaselineTTYOutputer::done_category_summary() { - _output->print_cr(" "); -} +void MemSummaryDiffReporter::report_diff() { + const char* scale = current_scale(); + outputStream* out = output(); + out->print_cr("\nNative Memory Tracking:\n"); - -void BaselineTTYOutputer::start_virtual_memory_map() { - _output->print_cr("Virtual memory map:"); -} + // Overall diff + out->print("Total: "); + print_virtual_memory_diff(_current_baseline.total_reserved_memory(), + _current_baseline.total_committed_memory(), _early_baseline.total_reserved_memory(), + _early_baseline.total_committed_memory()); -void BaselineTTYOutputer::reserved_memory_region(MEMFLAGS type, address base, address end, - size_t size, address pc) { - const char* unit = memory_unit(_scale); - char buf[128]; - int offset; - _output->print_cr(" "); - _output->print_cr("[" PTR_FORMAT " - " PTR_FORMAT "] reserved %d%s for %s", base, end, size, unit, - MemBaseline::type2name(type)); - if (os::dll_address_to_function_name(pc, buf, sizeof(buf), &offset)) { - _output->print_cr("\t\tfrom [%s+0x%x]", buf, offset); - } -} + out->print_cr("\n"); -void BaselineTTYOutputer::committed_memory_region(address base, address end, size_t size, address pc) { - const char* unit = memory_unit(_scale); - char buf[128]; - int offset; - _output->print("\t[" PTR_FORMAT " - " PTR_FORMAT "] committed %d%s", base, end, size, unit); - if (os::dll_address_to_function_name(pc, buf, sizeof(buf), &offset)) { - _output->print_cr(" from [%s+0x%x]", buf, offset); + // Summary diff by memory type + for (int index = 0; index < mt_number_of_types; index ++) { + MEMFLAGS flag = NMTUtil::index_to_flag(index); + // thread stack is reported as part of thread category + if (flag == mtThreadStack) continue; + diff_summary_of_type(flag, _early_baseline.malloc_memory(flag), + _early_baseline.virtual_memory(flag), _current_baseline.malloc_memory(flag), + _current_baseline.virtual_memory(flag)); } } -void BaselineTTYOutputer::done_virtual_memory_map() { - _output->print_cr(" "); -} - - - -void BaselineTTYOutputer::start_callsite() { - _output->print_cr("Details:"); - _output->print_cr(" "); -} - -void BaselineTTYOutputer::done_callsite() { - _output->print_cr(" "); -} +void MemSummaryDiffReporter::print_malloc_diff(size_t current_amount, size_t current_count, + size_t early_amount, size_t early_count) const { + const char* scale = current_scale(); + outputStream* out = output(); -void BaselineTTYOutputer::malloc_callsite(address pc, size_t malloc_amt, - size_t malloc_count) { - if (malloc_amt > 0) { - const char* unit = memory_unit(_scale); - char buf[128]; - int offset; - if (pc == 0) { - _output->print("[BOOTSTRAP]%18s", " "); - } else if (os::dll_address_to_function_name(pc, buf, sizeof(buf), &offset)) { - _output->print_cr("[" PTR_FORMAT "] %s+0x%x", pc, buf, offset); - _output->print("%28s", " "); - } else { - _output->print("[" PTR_FORMAT "]%18s", pc, " "); + out->print("malloc=" SIZE_FORMAT "%s", amount_in_current_scale(current_amount), scale); + long amount_diff = diff_in_current_scale(current_amount, early_amount); + if (amount_diff != 0) { + out->print(" %+ld%s", amount_diff, scale); + } + if (current_count > 0) { + out->print(" #" SIZE_FORMAT "", current_count); + if (current_count != early_count) { + out->print(" %+d", (int)(current_count - early_count)); } - - _output->print_cr("(malloc=%d%s #%d)", malloc_amt, unit, malloc_count); - _output->print_cr(" "); } } -void BaselineTTYOutputer::virtual_memory_callsite(address pc, size_t reserved_amt, - size_t committed_amt) { - if (reserved_amt > 0) { - const char* unit = memory_unit(_scale); - char buf[128]; - int offset; - if (pc == 0) { - _output->print("[BOOTSTRAP]%18s", " "); - } else if (os::dll_address_to_function_name(pc, buf, sizeof(buf), &offset)) { - _output->print_cr("[" PTR_FORMAT "] %s+0x%x", pc, buf, offset); - _output->print("%28s", " "); - } else { - _output->print("[" PTR_FORMAT "]%18s", pc, " "); - } +void MemSummaryDiffReporter::print_arena_diff(size_t current_amount, size_t current_count, + size_t early_amount, size_t early_count) const { + const char* scale = current_scale(); + outputStream* out = output(); + out->print("arena=" SIZE_FORMAT "%s", amount_in_current_scale(current_amount), scale); + if (diff_in_current_scale(current_amount, early_amount) != 0) { + out->print(" %+ld", diff_in_current_scale(current_amount, early_amount)); + } + + out->print(" #" SIZE_FORMAT "", current_count); + if (current_count != early_count) { + out->print(" %+d", (int)(current_count - early_count)); + } +} - _output->print_cr("(mmap: reserved=%d%s, committed=%d%s)", - reserved_amt, unit, committed_amt, unit); - _output->print_cr(" "); +void MemSummaryDiffReporter::print_virtual_memory_diff(size_t current_reserved, size_t current_committed, + size_t early_reserved, size_t early_committed) const {