diff options
110 files changed, 2347 insertions, 1483 deletions
@@ -115,3 +115,4 @@ b4acf10eb134fe930802c97e36db65e7ccb544b5 jdk7-b104 cc3fdfeb54b049f18edcf3463e6ab051d0b7b609 hs19-b05 688a538aa65412178286ae2a6b0c00b6711e121b hs19-b06 bf496cbe9b74dda5975a1559da7ecfdd313e509e jdk7-b107 +0000000000000000000000000000000000000000 hs19-b06 diff --git a/src/cpu/sparc/vm/vm_version_sparc.cpp b/src/cpu/sparc/vm/vm_version_sparc.cpp index ba44fe6a5..ca8a0ddaf 100644 --- a/src/cpu/sparc/vm/vm_version_sparc.cpp +++ b/src/cpu/sparc/vm/vm_version_sparc.cpp @@ -112,6 +112,11 @@ void VM_Version::initialize() { } } +#ifdef COMPILER2 + // Currently not supported anywhere. + FLAG_SET_DEFAULT(UseFPUForSpilling, false); +#endif + char buf[512]; jio_snprintf(buf, sizeof(buf), "%s%s%s%s%s%s%s%s%s%s%s%s", (has_v8() ? ", has_v8" : ""), diff --git a/src/cpu/x86/vm/vm_version_x86.cpp b/src/cpu/x86/vm/vm_version_x86.cpp index d6968a4f8..9fb37169a 100644 --- a/src/cpu/x86/vm/vm_version_x86.cpp +++ b/src/cpu/x86/vm/vm_version_x86.cpp @@ -482,6 +482,15 @@ void VM_Version::get_processor_features() { } } +#ifdef COMPILER2 + if (UseFPUForSpilling) { + if (UseSSE < 2) { + // Only supported with SSE2+ + FLAG_SET_DEFAULT(UseFPUForSpilling, false); + } + } +#endif + assert(0 <= ReadPrefetchInstr && ReadPrefetchInstr <= 3, "invalid value"); assert(0 <= AllocatePrefetchInstr && AllocatePrefetchInstr <= 3, "invalid value"); @@ -520,6 +529,11 @@ void VM_Version::get_processor_features() { if( supports_sse4_2() && supports_ht() ) { // Nehalem based cpus AllocatePrefetchDistance = 192; AllocatePrefetchLines = 4; +#ifdef COMPILER2 + if (AggressiveOpts && FLAG_IS_DEFAULT(UseFPUForSpilling)) { + FLAG_SET_DEFAULT(UseFPUForSpilling, true); + } +#endif } } assert(AllocatePrefetchDistance % AllocatePrefetchStepSize == 0, "invalid value"); diff --git a/src/cpu/x86/vm/x86_32.ad b/src/cpu/x86/vm/x86_32.ad index b4c57056f..a9cdabf8f 100644 --- a/src/cpu/x86/vm/x86_32.ad +++ b/src/cpu/x86/vm/x86_32.ad @@ -852,6 +852,39 @@ static int impl_movx_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int dst } } +static int impl_movgpr2x_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo, + int src_hi, int dst_hi, int size, outputStream* st ) { + // 32-bit + if (cbuf) { + emit_opcode(*cbuf, 0x66); + emit_opcode(*cbuf, 0x0F); + emit_opcode(*cbuf, 0x6E); + emit_rm(*cbuf, 0x3, Matcher::_regEncode[dst_lo] & 7, Matcher::_regEncode[src_lo] & 7); +#ifndef PRODUCT + } else if (!do_size) { + st->print("movdl %s, %s\t# spill", Matcher::regName[dst_lo], Matcher::regName[src_lo]); +#endif + } + return 4; +} + + +static int impl_movx2gpr_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo, + int src_hi, int dst_hi, int size, outputStream* st ) { + // 32-bit + if (cbuf) { + emit_opcode(*cbuf, 0x66); + emit_opcode(*cbuf, 0x0F); + emit_opcode(*cbuf, 0x7E); + emit_rm(*cbuf, 0x3, Matcher::_regEncode[src_lo] & 7, Matcher::_regEncode[dst_lo] & 7); +#ifndef PRODUCT + } else if (!do_size) { + st->print("movdl %s, %s\t# spill", Matcher::regName[dst_lo], Matcher::regName[src_lo]); +#endif + } + return 4; +} + static int impl_mov_helper( CodeBuffer *cbuf, bool do_size, int src, int dst, int size, outputStream* st ) { if( cbuf ) { emit_opcode(*cbuf, 0x8B ); @@ -947,6 +980,12 @@ uint MachSpillCopyNode::implementation( CodeBuffer *cbuf, PhaseRegAlloc *ra_, bo if( dst_first_rc == rc_int && src_first_rc == rc_stack ) size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_first),dst_first,0x8B,"MOV ",size, st); + // Check for integer reg-xmm reg copy + if( src_first_rc == rc_int && dst_first_rc == rc_xmm ) { + assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad), + "no 64 bit integer-float reg moves" ); + return impl_movgpr2x_helper(cbuf,do_size,src_first,dst_first,src_second, dst_second, size, st); + } // -------------------------------------- // Check for float reg-reg copy if( src_first_rc == rc_float && dst_first_rc == rc_float ) { @@ -1018,6 +1057,13 @@ uint MachSpillCopyNode::implementation( CodeBuffer *cbuf, PhaseRegAlloc *ra_, bo return impl_movx_helper(cbuf,do_size,src_first,dst_first,src_second, dst_second, size, st); } + // Check for xmm reg-integer reg copy + if( src_first_rc == rc_xmm && dst_first_rc == rc_int ) { + assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad), + "no 64 bit float-integer reg moves" ); + return impl_movx2gpr_helper(cbuf,do_size,src_first,dst_first,src_second, dst_second, size, st); + } + // Check for xmm store if( src_first_rc == rc_xmm && dst_first_rc == rc_stack ) { return impl_x_helper(cbuf,do_size,false,ra_->reg2offset(dst_first),src_first, src_second, size, st); diff --git a/src/cpu/x86/vm/x86_64.ad b/src/cpu/x86/vm/x86_64.ad index ba91cb799..ba24322a9 100644 --- a/src/cpu/x86/vm/x86_64.ad +++ b/src/cpu/x86/vm/x86_64.ad @@ -1607,8 +1607,8 @@ uint MachSpillCopyNode::implementation(CodeBuffer* cbuf, emit_opcode(*cbuf, 0x0F); emit_opcode(*cbuf, 0x7E); emit_rm(*cbuf, 0x3, - Matcher::_regEncode[dst_first] & 7, - Matcher::_regEncode[src_first] & 7); + Matcher::_regEncode[src_first] & 7, + Matcher::_regEncode[dst_first] & 7); #ifndef PRODUCT } else if (!do_size) { st->print("movdq %s, %s\t# spill", @@ -1637,8 +1637,8 @@ uint MachSpillCopyNode::implementation(CodeBuffer* cbuf, emit_opcode(*cbuf, 0x0F); emit_opcode(*cbuf, 0x7E); emit_rm(*cbuf, 0x3, - Matcher::_regEncode[dst_first] & 7, - Matcher::_regEncode[src_first] & 7); + Matcher::_regEncode[src_first] & 7, + Matcher::_regEncode[dst_first] & 7); #ifndef PRODUCT } else if (!do_size) { st->print("movdl %s, %s\t# spill", diff --git a/src/cpu/zero/vm/bytecodeInterpreter_zero.inline.hpp b/src/cpu/zero/vm/bytecodeInterpreter_zero.inline.hpp index 702a165ef..a35809f0a 100644 --- a/src/cpu/zero/vm/bytecodeInterpreter_zero.inline.hpp +++ b/src/cpu/zero/vm/bytecodeInterpreter_zero.inline.hpp @@ -1,6 +1,6 @@ /* * Copyright (c) 2002, Oracle and/or its affiliates. All rights reserved. - * Copyright 2007 Red Hat, Inc. + * Copyright 2007, 2010 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -268,7 +268,7 @@ inline jint BytecodeInterpreter::VMintSub(jint op1, jint op2) { return op1 - op2; } -inline jint BytecodeInterpreter::VMintUshr(jint op1, jint op2) { +inline juint BytecodeInterpreter::VMintUshr(jint op1, jint op2) { return ((juint) op1) >> (op2 & 0x1F); } diff --git a/src/cpu/zero/vm/javaFrameAnchor_zero.hpp b/src/cpu/zero/vm/javaFrameAnchor_zero.hpp index 5748301c0..0763790b6 100644 --- a/src/cpu/zero/vm/javaFrameAnchor_zero.hpp +++ b/src/cpu/zero/vm/javaFrameAnchor_zero.hpp @@ -82,6 +82,10 @@ return _last_Java_fp; } + address last_Java_pc() const { + return _last_Java_pc; + } + static ByteSize last_Java_fp_offset() { return byte_offset_of(JavaFrameAnchor, _last_Java_fp); } diff --git a/src/os_cpu/linux_sparc/vm/orderAccess_linux_sparc.inline.hpp b/src/os_cpu/linux_sparc/vm/orderAccess_linux_sparc.inline.hpp index 4ff88cfdb..2770baaab 100644 --- a/src/os_cpu/linux_sparc/vm/orderAccess_linux_sparc.inline.hpp +++ b/src/os_cpu/linux_sparc/vm/orderAccess_linux_sparc.inline.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2003, 2008, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -36,8 +36,8 @@ inline void OrderAccess::acquire() { } inline void OrderAccess::release() { - jint* dummy = (jint*)&dummy; - __asm__ volatile("stw %%g0, [%0]" : : "r" (dummy) : "memory"); + jint* local_dummy = (jint*)&local_dummy; + __asm__ volatile("stw %%g0, [%0]" : : "r" (local_dummy) : "memory"); } inline void OrderAccess::fence() { diff --git a/src/os_cpu/linux_x86/vm/orderAccess_linux_x86.inline.hpp b/src/os_cpu/linux_x86/vm/orderAccess_linux_x86.inline.hpp index a4f6c17d7..d487ece69 100644 --- a/src/os_cpu/linux_x86/vm/orderAccess_linux_x86.inline.hpp +++ b/src/os_cpu/linux_x86/vm/orderAccess_linux_x86.inline.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2003, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -30,16 +30,18 @@ inline void OrderAccess::loadstore() { acquire(); } inline void OrderAccess::storeload() { fence(); } inline void OrderAccess::acquire() { - volatile intptr_t dummy; + volatile intptr_t local_dummy; #ifdef AMD64 - __asm__ volatile ("movq 0(%%rsp), %0" : "=r" (dummy) : : "memory"); + __asm__ volatile ("movq 0(%%rsp), %0" : "=r" (local_dummy) : : "memory"); #else - __asm__ volatile ("movl 0(%%esp),%0" : "=r" (dummy) : : "memory"); + __asm__ volatile ("movl 0(%%esp),%0" : "=r" (local_dummy) : : "memory"); #endif // AMD64 } inline void OrderAccess::release() { - dummy = 0; + // Avoid hitting the same cache-line from + // different threads. + volatile jint local_dummy = 0; } inline void OrderAccess::fence() { diff --git a/src/os_cpu/linux_zero/vm/os_linux_zero.cpp b/src/os_cpu/linux_zero/vm/os_linux_zero.cpp index ce9d86fdd..fd78628b3 100644 --- a/src/os_cpu/linux_zero/vm/os_linux_zero.cpp +++ b/src/os_cpu/linux_zero/vm/os_linux_zero.cpp @@ -435,22 +435,22 @@ extern "C" { void _Copy_arrayof_conjoint_bytes(HeapWord* from, HeapWord* to, size_t count) { - ShouldNotCallThis(); + memmove(to, from, count); } void _Copy_arrayof_conjoint_jshorts(HeapWord* from, HeapWord* to, size_t count) { - ShouldNotCallThis(); + memmove(to, from, count * 2); } void _Copy_arrayof_conjoint_jints(HeapWord* from, HeapWord* to, size_t count) { - ShouldNotCallThis(); + memmove(to, from, count * 4); } void _Copy_arrayof_conjoint_jlongs(HeapWord* from, HeapWord* to, size_t count) { - ShouldNotCallThis(); + memmove(to, from, count * 8); } }; diff --git a/src/os_cpu/linux_zero/vm/thread_linux_zero.cpp b/src/os_cpu/linux_zero/vm/thread_linux_zero.cpp index 15dba5fff..8551eafce 100644 --- a/src/os_cpu/linux_zero/vm/thread_linux_zero.cpp +++ b/src/os_cpu/linux_zero/vm/thread_linux_zero.cpp @@ -1,6 +1,6 @@ /* * Copyright (c) 1997, 2007, Oracle and/or its affiliates. All rights reserved. - * Copyright 2009 Red Hat, Inc. + * Copyright 2009, 2010 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -23,6 +23,9 @@ * */ -// This file is intentionally empty +#include "incls/_precompiled.incl" +#include "incls/_thread_linux_zero.cpp.incl" -void JavaThread::cache_global_variables() { } +void JavaThread::cache_global_variables() { + // nothing to do +} diff --git a/src/os_cpu/solaris_sparc/vm/orderAccess_solaris_sparc.inline.hpp b/src/os_cpu/solaris_sparc/vm/orderAccess_solaris_sparc.inline.hpp index 2646977ce..6075e0106 100644 --- a/src/os_cpu/solaris_sparc/vm/orderAccess_solaris_sparc.inline.hpp +++ b/src/os_cpu/solaris_sparc/vm/orderAccess_solaris_sparc.inline.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2003, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -42,8 +42,8 @@ inline void OrderAccess::acquire() { } inline void OrderAccess::release() { - jint* dummy = (jint*)&dummy; - __asm__ volatile("stw %%g0, [%0]" : : "r" (dummy) : "memory"); + jint* local_dummy = (jint*)&local_dummy; + __asm__ volatile("stw %%g0, [%0]" : : "r" (local_dummy) : "memory"); } inline void OrderAccess::fence() { @@ -57,7 +57,9 @@ inline void OrderAccess::acquire() { } inline void OrderAccess::release() { - dummy = 0; + // Avoid hitting the same cache-line from + // different threads. + volatile jint local_dummy = 0; } inline void OrderAccess::fence() { diff --git a/src/os_cpu/solaris_x86/vm/orderAccess_solaris_x86.inline.hpp b/src/os_cpu/solaris_x86/vm/orderAccess_solaris_x86.inline.hpp index 7165cde66..84c1ce923 100644 --- a/src/os_cpu/solaris_x86/vm/orderAccess_solaris_x86.inline.hpp +++ b/src/os_cpu/solaris_x86/vm/orderAccess_solaris_x86.inline.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2003, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -40,7 +40,9 @@ inline void OrderAccess::acquire() { } inline void OrderAccess::release() { - dummy = 0; + // Avoid hitting the same cache-line from + // different threads. + volatile jint local_dummy = 0; } inline void OrderAccess::fence() { @@ -53,11 +55,11 @@ inline void OrderAccess::fence() { extern "C" { inline void _OrderAccess_acquire() { - volatile intptr_t dummy; + volatile intptr_t local_dummy; #ifdef AMD64 - __asm__ volatile ("movq 0(%%rsp), %0" : "=r" (dummy) : : "memory"); + __asm__ volatile ("movq 0(%%rsp), %0" : "=r" (local_dummy) : : "memory"); #else - __asm__ volatile ("movl 0(%%esp),%0" : "=r" (dummy) : : "memory"); + __asm__ volatile ("movl 0(%%esp),%0" : "=r" (local_dummy) : : "memory"); #endif // AMD64 } inline void _OrderAccess_fence() { diff --git a/src/os_cpu/windows_x86/vm/orderAccess_windows_x86.inline.hpp b/src/os_cpu/windows_x86/vm/orderAccess_windows_x86.inline.hpp index 4a5ac91b1..1ccddc4d3 100644 --- a/src/os_cpu/windows_x86/vm/orderAccess_windows_x86.inline.hpp +++ b/src/os_cpu/windows_x86/vm/orderAccess_windows_x86.inline.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2003, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -41,7 +41,7 @@ inline void OrderAccess::acquire() { inline void OrderAccess::release() { // A volatile store has release semantics. - dummy = 0; + volatile jint local_dummy = 0; } inline void OrderAccess::fence() { diff --git a/src/share/vm/code/nmethod.cpp b/src/share/vm/code/nmethod.cpp index af0f18333..1b6472a3c 100644 --- a/src/share/vm/code/nmethod.cpp +++ b/src/share/vm/code/nmethod.cpp @@ -433,6 +433,10 @@ void nmethod::init_defaults() { _unload_reported = false; // jvmti state NOT_PRODUCT(_has_debug_info = false); +#ifdef ASSERT + _oops_are_stale = false; +#endif + _oops_do_mark_link = NULL; _jmethod_id = NULL; _osr_link = NULL; @@ -1230,11 +1234,10 @@ void nmethod::log_state_change() const { bool nmethod::make_not_entrant_or_zombie(unsigned int state) { assert(state == zombie || state == not_entrant, "must be zombie or not_entrant"); - bool was_alive = false; - // Make sure neither the nmethod nor the method is flushed in case of a safepoint in code below. nmethodLocker nml(this); methodHandle the_method(method()); + No_Safepoint_Verifier nsv; { // If the method is already zombie there is nothing to do @@ -1303,13 +1306,27 @@ bool nmethod::make_not_entrant_or_zombie(unsigned int state) { // state will be flushed later when the transition to zombie // happens or they get unloaded. if (state == zombie) { - // zombie only - if a JVMTI agent has enabled the CompiledMethodUnload event - // and it hasn't already been reported for this nmethod then report it now. - // (the event may have been reported earilier if the GC marked it for unloading). - post_compiled_method_unload(); + { + // Flushing dependecies must be done before any possible + // safepoint can sneak in, otherwise the oops used by the + // dependency logic could have become stale. + MutexLockerEx mu(CodeCache_lock, Mutex::_no_safepoint_check_flag); + flush_dependencies(NULL); + } - MutexLockerEx mu(CodeCache_lock, Mutex::_no_safepoint_check_flag); - flush_dependencies(NULL); + { + // zombie only - if a JVMTI agent has enabled the CompiledMethodUnload event + // and it hasn't already been reported for this nmethod then report it now. + // (the event may have been reported earilier if the GC marked it for unloading). + Pause_No_Safepoint_Verifier pnsv(&nsv); + post_compiled_method_unload(); + } + +#ifdef ASSERT + // It's no longer safe to access the oops section since zombie + // nmethods aren't scanned for GC. + _oops_are_stale = true; +#endif } else { assert(state == not_entrant, "other cases may need to be handled differently"); } diff --git a/src/share/vm/code/nmethod.hpp b/src/share/vm/code/nmethod.hpp index b57cb5e3d..88fe784cf 100644 --- a/src/share/vm/code/nmethod.hpp +++ b/src/share/vm/code/nmethod.hpp @@ -177,6 +177,10 @@ class nmethod : public CodeBlob { // Protected by Patching_lock unsigned char _state; // {alive, not_entrant, zombie, unloaded) +#ifdef ASSERT + bool _oops_are_stale; // indicates that it's no longer safe to access oops section +#endif + enum { alive = 0, not_entrant = 1, // uncommon trap has happened but activations may still exist zombie = 2, @@ -434,6 +438,7 @@ class nmethod : public CodeBlob { oop* oop_addr_at(int index) const { // for GC // relocation indexes are biased by 1 (because 0 is reserved) assert(index > 0 && index <= oops_size(), "must be a valid non-zero index"); + assert(!_oops_are_stale, "oops are stale"); return &oops_begin()[index - 1]; } diff --git a/src/share/vm/compiler/compileBroker.cpp b/src/share/vm/compiler/compileBroker.cpp index 7288cac07..478d3f47b 100644 --- a/src/share/vm/compiler/compileBroker.cpp +++ b/src/share/vm/compiler/compileBroker.cpp @@ -1652,12 +1652,10 @@ void CompileBroker::invoke_compiler_on_method(CompileTask* task) { void CompileBroker::handle_full_code_cache() { UseInterpreter = true; if (UseCompiler || AlwaysCompileLoopMethods ) { - CompilerThread* thread = CompilerThread::current(); - CompileLog* log = thread->log(); - if (log != NULL) { - log->begin_elem("code_cache_full"); - log->stamp(); - log->end_elem(); + if (xtty != NULL) { + xtty->begin_elem("code_cache_full"); + xtty->stamp(); + xtty->end_elem(); } warning("CodeCache is full. Compiler has been disabled."); warning("Try increasing the code cache size using -XX:ReservedCodeCacheSize="); diff --git a/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.cpp b/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.cpp index 28f6ef4b0..30982f3dd 100644 --- a/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.cpp +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2001, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -402,6 +402,29 @@ size_t CompactibleFreeListSpace::max_alloc_in_words() const { return res; } +void LinearAllocBlock::print_on(outputStream* st) const { + st->print_cr(" LinearAllocBlock: ptr = " PTR_FORMAT ", word_size = " SIZE_FORMAT + ", refillsize = " SIZE_FORMAT ", allocation_size_limit = " SIZE_FORMAT, + _ptr, _word_size, _refillSize, _allocation_size_limit); +} + +void CompactibleFreeListSpace::print_on(outputStream* st) const { + st->print_cr("COMPACTIBLE FREELIST SPACE"); + st->print_cr(" Space:"); + Space::print_on(st); + + st->print_cr("promoInfo:"); + _promoInfo.print_on(st); + + st->print_cr("_smallLinearAllocBlock"); + _smallLinearAllocBlock.print_on(st); + + // dump_memory_block(_smallLinearAllocBlock->_ptr, 128); + + st->print_cr(" _fitStrategy = %s, _adaptive_freelists = %s", + _fitStrategy?"true":"false", _adaptive_freelists?"true":"false"); +} + void CompactibleFreeListSpace::print_indexed_free_lists(outputStream* st) const { reportIndexedFreeListStatistics(); @@ -557,13 +580,15 @@ size_t CompactibleFreeListSpace::maxChunkSizeInIndexedFreeLists() const { void CompactibleFreeListSpace::set_end(HeapWord* value) { HeapWord* prevEnd = end(); assert(prevEnd != value, "unnecessary set_end call"); - assert(prevEnd == NULL || value >= unallocated_block(), "New end is below unallocated block"); + assert(prevEnd == NULL || !BlockOffsetArrayUseUnallocatedBlock || value >= unallocated_block(), + "New end is below unallocated block"); _end = value; if (prevEnd != NULL) { // Resize the underlying block offset table. _bt.resize(pointer_delta(value, bottom())); if (value <= prevEnd) { - assert(value >= unallocated_block(), "New end is below unallocated block"); + assert(!BlockOffsetArrayUseUnallocatedBlock || value >= unallocated_block(), + "New end is below unallocated block"); } else { // Now, take this new chunk and add it to the free blocks. // Note that the BOT has not yet been updated for this block. @@ -938,7 +963,6 @@ HeapWord* CompactibleFreeListSpace::block_start_careful(const void* p) const { size_t CompactibleFreeListSpace::block_size(const HeapWord* p) const { NOT_PRODUCT(verify_objects_initialized()); - assert(MemRegion(bottom(), end()).contains(p), "p not in space"); // This must be volatile, or else there is a danger that the compiler // will compile the code below into a sometimes-infinite loop, by keeping // the value read the first time in a register. @@ -957,7 +981,7 @@ size_t CompactibleFreeListSpace::block_size(const HeapWord* p) const { // must read from what 'p' points to in each loop. klassOop k = ((volatile oopDesc*)p)->klass_or_null(); if (k != NULL) { - assert(k->is_oop(true /* ignore mark word */), "Should really be klass oop."); + assert(k->is_oop(true /* ignore mark word */), "Should be klass oop"); oop o = (oop)p; assert(o->is_parsable(), "Should be parsable"); assert(o->is_oop(true /* ignore mark word */), "Should be an oop."); @@ -1231,7 +1255,6 @@ HeapWord* CompactibleFreeListSpace::allocate_adaptive_freelists(size_t size) { // satisfy the request. This is different that // evm. // Don't record chunk off a LinAB? smallSplitBirth(size); - } else { // Raid the exact free lists larger than size, even if they are not // overpopulated. @@ -1449,6 +1472,7 @@ CompactibleFreeListSpace::getChunkFromLinearAllocBlock(LinearAllocBlock *blk, // Update BOT last so that other (parallel) GC threads see a consistent // view of the BOT and free blocks. // Above must occur before BOT is updated below. + OrderAccess::storestore(); _bt.split_block(res, blk_size, size); // adjust block offset table } return res; @@ -1477,6 +1501,7 @@ HeapWord* CompactibleFreeListSpace::getChunkFromLinearAllocBlockRemainder( // Update BOT last so that other (parallel) GC threads see a consistent // view of the BOT and free blocks. // Above must occur before BOT is updated below. + OrderAccess::storestore(); _bt.split_block(res, blk_size, size); // adjust block offset table _bt.allocated(res, size); } @@ -1856,6 +1881,8 @@ CompactibleFreeListSpace::splitChunkAndReturnRemainder(FreeChunk* chunk, ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads. // Above must occur before BOT is updated below. // adjust block offset table + OrderAccess::storestore(); + assert(chunk->isFree() && ffc->isFree(), "Error"); _bt.split_block((HeapWord*)chunk, chunk->size(), new_size); if (rem_size < SmallForDictionary) { bool is_par = (SharedHeap::heap()->n_par_threads() > 0); @@ -1911,8 +1938,7 @@ void CompactibleFreeListSpace::save_marks() { // mark the "end" of the used space at the time of this call; // note, however, that promoted objects from this point // on are tracked in the _promoInfo below. - set_saved_mark_word(BlockOffsetArrayUseUnallocatedBlock ? - unallocated_block() : end()); + set_saved_mark_word(unallocated_block()); // inform allocator that promotions should be tracked. assert(_promoInfo.noPromotions(), "_promoInfo inconsistency"); _promoInfo.startTrackingPromotions(); @@ -2238,8 +2264,7 @@ void CompactibleFreeListSpace::split(size_t from, size_t to1) { } void CompactibleFreeListSpace::print() const { - tty->print(" CompactibleFreeListSpace"); - Space::print(); + Space::print_on(tty); } void CompactibleFreeListSpace::prepare_for_verify() { @@ -2253,18 +2278,28 @@ class VerifyAllBlksClosure: public BlkClosure { private: const CompactibleFreeListSpace* _sp; const MemRegion _span; + HeapWord* _last_addr; + size_t _last_size; + bool _last_was_obj; + bool _last_was_live; public: VerifyAllBlksClosure(const CompactibleFreeListSpace* sp, - MemRegion span) : _sp(sp), _span(span) { } + MemRegion span) : _sp(sp), _span(span), + _last_addr(NULL), _last_size(0), + _last_was_obj(false), _last_was_live(false) { } virtual size_t do_blk(HeapWord* addr) { size_t res; + bool was_obj = false; + bool was_live = false; if (_sp->block_is_obj(addr)) { + was_obj = true; oop p = oop(addr); guarantee(p->is_oop(), "Should be an oop"); res = _sp->adjustObjectSize(p->size()); if (_sp->obj_is_alive(addr)) { + was_live = true; p->verify(); } } else { @@ -2275,7 +2310,20 @@ class VerifyAllBlksClosure: public BlkClosure { "Chunk should be on a free list"); } } - guarantee(res != 0, "Livelock: no rank reduction!"); + if (res == 0) { + gclog_or_tty->print_cr("Livelock: no rank reduction!"); + gclog_or_tty->print_cr( + " Current: addr = " PTR_FORMAT ", size = " SIZE_FORMAT ", obj = %s, live = %s \n" + " Previous: addr = " PTR_FORMAT ", size = " SIZE_FORMAT ", obj = %s, live = %s \n", + addr, res, was_obj ?"true":"false", was_live ?"true":"false", + _last_addr, _last_size, _last_was_obj?"true":"false", _last_was_live?"true":"false"); + _sp->print_on(gclog_or_tty); + guarantee(false, "Seppuku!"); + } + _last_addr = addr; + _last_size = res; + _last_was_obj = was_obj; + _last_was_live = was_live; return res; } }; @@ -2521,7 +2569,7 @@ void CFLS_LAB::modify_initialization(size_t n, unsigned wt) { HeapWord* CFLS_LAB::alloc(size_t word_sz) { FreeChunk* res; - word_sz = _cfls->adjustObjectSize(word_sz); + guarantee(word_sz == _cfls->adjustObjectSize(word_sz), "Error"); if (word_sz >= CompactibleFreeListSpace::IndexSetSize) { // This locking manages sync with other large object allocations. MutexLockerEx x(_cfls->parDictionaryAllocLock(), @@ -2667,12 +2715,12 @@ void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n (cur_sz < CompactibleFreeListSpace::IndexSetSize) && (CMSSplitIndexedFreeListBlocks || k <= 1); k++, cur_sz = k * word_sz) { - FreeList* gfl = &_indexedFreeList[cur_sz]; FreeList fl_for_cur_sz; // Empty. fl_for_cur_sz.set_size(cur_sz); { MutexLockerEx x(_indexedFreeListParLocks[cur_sz], Mutex::_no_safepoint_check_flag); + FreeList* gfl = &_indexedFreeList[cur_sz]; if (gfl->count() != 0) { // nn is the number of chunks of size cur_sz that // we'd need to split k-ways each, in order to create @@ -2685,9 +2733,9 @@ void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n // we increment the split death count by the number of blocks // we just took from the cur_sz-size blocks list and which // we will be splitting below. - ssize_t deaths = _indexedFreeList[cur_sz].splitDeaths() + + ssize_t deaths = gfl->splitDeaths() + fl_for_cur_sz.count(); - _indexedFreeList[cur_sz].set_splitDeaths(deaths); + gfl->set_splitDeaths(deaths); } } } @@ -2703,18 +2751,25 @@ void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n // access the main chunk sees it as a single free block until we // change it. size_t fc_size = fc->size(); + assert(fc->isFree(), "Error"); for (int i = k-1; i >= 0; i--) { FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz); + assert((i != 0) || + ((fc == ffc) && ffc->isFree() && + (ffc->size() == k*word_sz) && (fc_size == word_sz)), + "Counting error"); ffc->setSize(word_sz); - ffc->linkNext(NULL); ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads. + ffc->linkNext(NULL); // Above must occur before BOT is updated below. - // splitting from the right, fc_size == (k - i + 1) * wordsize - _bt.mark_block((HeapWord*)ffc, word_sz); + OrderAccess::storestore(); + // splitting from the right, fc_size == i * word_sz + _bt.mark_block((HeapWord*)ffc, word_sz, true /* reducing */); fc_size -= word_sz; - _bt.verify_not_unallocated((HeapWord*)ffc, ffc->size()); + assert(fc_size == i*word_sz, "Error"); + _bt.verify_not_unallocated((HeapWord*)ffc, word_sz); _bt.verify_single_block((HeapWord*)fc, fc_size); - _bt.verify_single_block((HeapWord*)ffc, ffc->size()); + _bt.verify_single_block((HeapWord*)ffc, word_sz); // Push this on "fl". fl->returnChunkAtHead(ffc); } @@ -2744,7 +2799,7 @@ void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n _dictionary->minSize()), FreeBlockDictionary::atLeast); if (fc != NULL) { - _bt.allocated((HeapWord*)fc, fc->size()); // update _unallocated_blk + _bt.allocated((HeapWord*)fc, fc->size(), true /* reducing */); // update _unallocated_blk dictionary()->dictCensusUpdate(fc->size(), true /*split*/, false /*birth*/); @@ -2754,8 +2809,10 @@ void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n } } if (fc == NULL) return; - assert((ssize_t)n >= 1, "Control point invariant"); // Otherwise, split up that block. + assert((ssize_t)n >= 1, "Control point invariant"); + assert(fc->isFree(), "Error: should be a free block"); + _bt.verify_single_block((HeapWord*)fc, fc->size()); const size_t nn = fc->size() / word_sz; n = MIN2(nn, n); assert((ssize_t)n >= 1, "Control point invariant"); @@ -2773,6 +2830,7 @@ void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n // dictionary and return, leaving "fl" empty. if (n == 0) { returnChunkToDictionary(fc); + assert(fl->count() == 0, "We never allocated any blocks"); return; } @@ -2785,11 +2843,14 @@ void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n size_t prefix_size = n * word_sz; rem_fc = (FreeChunk*)((HeapWord*)fc + prefix_size); rem_fc->setSize(rem); - rem_fc->linkNext(NULL); rem_fc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads. + rem_fc->linkNext(NULL); // Above must occur before BOT is updated below. assert((ssize_t)n > 0 && prefix_size > 0 && rem_fc > fc, "Error"); + OrderAccess::storestore(); _bt.split_block((HeapWord*)fc, fc->size(), prefix_size); + assert(fc->isFree(), "Error"); + fc->setSize(prefix_size); if (rem >= IndexSetSize) { returnChunkToDictionary(rem_fc); dictionary()->dictCensusUpdate(rem, true /*split*/, true /*birth*/); @@ -2815,11 +2876,12 @@ void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n for (ssize_t i = n-1; i > 0; i--) { FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz); ffc->setSize(word_sz); - ffc->linkNext(NULL); ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads. + ffc->linkNext(NULL); // Above must occur before BOT is updated below. + OrderAccess::storestore(); // splitting from the right, fc_size == (n - i + 1) * wordsize - _bt.mark_block((HeapWord*)ffc, word_sz); + _bt.mark_block((HeapWord*)ffc, word_sz, true /* reducing */); fc_size -= word_sz; _bt.verify_not_unallocated((HeapWord*)ffc, ffc->size()); _bt.verify_single_block((HeapWord*)ffc, ffc->size()); @@ -2828,9 +2890,11 @@ void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n fl->returnChunkAtHead(ffc); } // First chunk + assert(fc->isFree() && fc->size() == n*word_sz, "Error: should still be a free block"); + // The blocks above should show their new sizes before the first block below fc->setSize(word_sz); + fc->linkPrev(NULL); // idempotent wrt free-ness, see assert above fc->linkNext(NULL); - fc->linkPrev(NULL); _bt.verify_not_unallocated((HeapWord*)fc, fc->size()); _bt.verify_single_block((HeapWord*)fc, fc->size()); fl->returnChunkAtHead(fc); diff --git a/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp b/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp index 8bca5df52..6d67e9411 100644 --- a/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2001, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -48,6 +48,8 @@ class LinearAllocBlock VALUE_OBJ_CLASS_SPEC { size_t _word_size; size_t _refillSize; size_t _allocation_size_limit; // largest size that will be allocated + + void print_on(outputStream* st) const; }; // Concrete subclass of CompactibleSpace that implements @@ -249,10 +251,14 @@ class CompactibleFreeListSpace: public CompactibleSpace { size_t numFreeBlocksInIndexedFreeLists() const; // Accessor HeapWord* unallocated_block() const { - HeapWord* ub = _bt.unallocated_block(); - assert(ub >= bottom() && - ub <= end(), "space invariant"); - return ub; + if (BlockOffsetArrayUseUnallocatedBlock) { + HeapWord* ub = _bt.unallocated_block(); + assert(ub >= bottom() && + ub <= end(), "space invariant"); + return ub; + } else { + return end(); + } } void freed(HeapWord* start, size_t size) { _bt.freed(start, size); @@ -476,6 +482,7 @@ class CompactibleFreeListSpace: public CompactibleSpace { // Debugging support void print() const; + void print_on(outputStream* st) const; void prepare_for_verify(); void verify(bool allow_dirty) const; void verifyFreeLists() const PRODUCT_RETURN; diff --git a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp index 1fac91a54..f5d19544e 100644 --- a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp @@ -1019,7 +1019,7 @@ HeapWord* ConcurrentMarkSweepGeneration::allocate(size_t size, } HeapWord* ConcurrentMarkSweepGeneration::have_lock_and_allocate(size_t size, - bool tlab) { + bool tlab /* ignored */) { assert_lock_strong(freelistLock()); size_t adjustedSize = CompactibleFreeListSpace::adjustObjectSize(size); HeapWord* res = cmsSpace()->allocate(adjustedSize); @@ -1032,6 +1032,11 @@ HeapWord* ConcurrentMarkSweepGeneration::have_lock_and_allocate(size_t size, // allowing the object to be blackened (and its references scanned) // either during a preclean phase or at the final checkpoint. if (res != NULL) { + // We may block here with an uninitialized object with + // its mark-bit or P-bits not yet set. Such objects need + // to be safely navigable by block_start(). + assert(oop(res)->klass_or_null() == NULL, "Object should be uninitialized here."); + assert(!((FreeChunk*)res)->isFree(), "Error, block will look free but show wrong size"); collector()->direct_allocated(res, adjustedSize); _direct_allocated_words += adjustedSize; // allocation counters @@ -1061,8 +1066,14 @@ void CMSCollector::direct_allocated(HeapWord* start, size_t size) { // [see comments preceding SweepClosure::do_blk() below for details] // 1. need to mark the object as live so it isn't collected // 2. need to mark the 2nd bit to indicate the object may be uninitialized - // 3. need to mark the end of the object so sweeper can skip over it - // if it's uninitialized when the sweeper reaches it. + // 3. need to mark the end of the object so marking, precleaning or sweeping + // can skip over uninitialized or unparsable objects. An allocated + // object is considered uninitialized for our purposes as long as + // its klass word is NULL. (Unparsable objects are those which are + // initialized in the sense just described, but whose sizes can still + // not be correctly determined. Note that the class of unparsable objects + // can only occur in the perm gen. All old gen objects are parsable + // as soon as they are initialized.) _markBitMap.mark(start); // object is live _markBitMap.mark(start + 1); // object is potentially uninitialized? _markBitMap.mark(start + size - 1); @@ -1088,7 +1099,13 @@ void CMSCollector::promoted(bool par, HeapWord* start, // We don't need to mark the object as uninitialized (as // in direct_allocated above) because this is being done with the // world stopped and the object will be initialized by the - // time the sweeper gets to look at it. + // time the marking, precleaning or sweeping get to look at it. + // But see the code for copying objects into the CMS generation, + // where we need to ensure that concurrent readers of the + // block offset table are able to safely navigate a block that + // is in flux from being free to being allocated (and in + // transition while being copied into) and subsequently + // becoming a bona-fide object when the copy/promotion is complete. assert(SafepointSynchronize::is_at_safepoint(), "expect promotion only at safepoints"); @@ -1304,6 +1321,48 @@ ConcurrentMarkSweepGeneration::allocation_limit_reached(Space* space, return collector()->allocation_limit_reached(space, top, word_sz); } +// IMPORTANT: Notes on object size recognition in CMS. +// --------------------------------------------------- +// A block of storage in the CMS generation is always in +// one of three states. A free block (FREE), an allocated +// object (OBJECT) whose size() method reports the correct size, +// and an intermediate state (TRANSIENT) in which its size cannot +// be accurately determined. +// STATE IDENTIFICATION: (32 bit and 64 bit w/o COOPS) +// ----------------------------------------------------- +// FREE: klass_word & 1 == 1; mark_word holds block size +// +// OBJECT: klass_word installed; klass_word != 0 && klass_word & 0 == 0; +// obj->size() computes correct size +// [Perm Gen objects needs to be "parsable" before they can be navigated] +// +// TRANSIENT: klass_word == 0; size is indeterminate until we become an OBJECT +// +// STATE IDENTIFICATION: (64 bit+COOPS) +// ------------------------------------ +// FREE: mark_word & CMS_FREE_BIT == 1; mark_word & ~CMS_FREE_BIT gives block_size +// +// OBJECT: klass_word installed; klass_word != 0; +// obj->size() computes correct size +// [Perm Gen comment above continues to hold] +// +// TRANSIENT: klass_word == 0; size is indeterminate until we become an OBJECT +// +// +// STATE TRANSITION DIAGRAM +// +// mut / parnew mut / parnew +// FREE --------------------> TRANSIENT ---------------------> OBJECT --| +// ^ | +// |------------------------ DEAD <------------------------------------| +// sweep mut +// +// While a block is in TRANSIENT state its size cannot be determined +// so readers will either need to come back later or stall until +// the size can be determined. Note that for the case of direct +// allocation, P-bits, when available, may be used to determine the +// size of an object that may not yet have been initialized. + // Things to support parallel young-gen collection. oop ConcurrentMarkSweepGeneration::par_promote(int thread_num, @@ -1331,33 +1390,39 @@ ConcurrentMarkSweepGeneration::par_promote(int thread_num, } } assert(promoInfo->has_spooling_space(), "Control point invariant"); - HeapWord* obj_ptr = ps->lab.alloc(word_sz); + const size_t alloc_sz = CompactibleFreeListSpace::adjustObjectSize(word_sz); + HeapWord* obj_ptr = ps->lab.alloc(alloc_sz); if (obj_ptr == NULL) { - obj_ptr = expand_and_par_lab_allocate(ps, word_sz); + obj_ptr = expand_and_par_lab_allocate(ps, alloc_sz); if (obj_ptr == NULL) { return NULL; } } oop obj = oop(obj_ptr); + OrderAccess::storestore(); assert(obj->klass_or_null() == NULL, "Object should be uninitialized here."); + assert(!((FreeChunk*)obj_ptr)->isFree(), "Error, block will look free but show wrong size"); + // IMPORTANT: See note on object initialization for CMS above. // Otherwise, copy the object. Here we must be careful to insert the // klass pointer last, since this marks the block as an allocated object. // Except with compressed oops it's the mark word. HeapWord* old_ptr = (HeapWord*)old; + // Restore the mark word copied above. + obj->set_mark(m); + assert(obj->klass_or_null() == NULL, "Object should be uninitialized here."); + assert(!((FreeChunk*)obj_ptr)->isFree(), "Error, block will look free but show wrong size"); + OrderAccess::storestore(); + + if (UseCompressedOops) { + // Copy gap missed by (aligned) header size calculation below + obj->set_klass_gap(old->klass_gap()); + } if (word_sz > (size_t)oopDesc::header_size()) { Copy::aligned_disjoint_words(old_ptr + oopDesc::header_size(), obj_ptr + oopDesc::header_size(), word_sz - oopDesc::header_size()); } - if (UseCompressedOops) { - // Copy gap missed by (aligned) header size calculation above - obj->set_klass_gap(old->klass_gap()); - } - - // Restore the mark word copied above. - obj->set_mark(m); - // Now we can track the promoted object, if necessary. We take care // to delay the transition from uninitialized to full object // (i.e., insertion of klass pointer) until after, so that it @@ -1365,18 +1430,22 @@ ConcurrentMarkSweepGeneration::par_promote(int thread_num, if (promoInfo->tracking()) { promoInfo->track((PromotedObject*)obj, old->klass()); } + assert(obj->klass_or_null() == NULL, "Object should be uninitialized here."); + assert(!((FreeChunk*)obj_ptr)->isFree(), "Error, block will look free but show wrong size"); + assert(old->is_oop(), "Will use and dereference old klass ptr below"); // Finally, install the klass pointer (this should be volatile). + OrderAccess::storestore(); obj->set_klass(old->klass()); + // We should now be able to calculate the right size for this object + assert(obj->is_oop() && obj->size() == (int)word_sz, "Error, incorrect size computed for promoted object"); - assert(old->is_oop(), "Will dereference klass ptr below"); collector()->promoted(true, // parallel obj_ptr, old->is_objArray(), word_sz); NOT_PRODUCT( - Atomic::inc(&_numObjectsPromoted); - Atomic::add((jint)CompactibleFreeListSpace::adjustObjectSize(obj->size()), - &_numWordsPromoted); + Atomic::inc_ptr(&_numObjectsPromoted); + Atomic::add_ptr(alloc_sz, &_numWordsPromoted); ) return obj; @@ -1965,6 +2034,9 @@ void CMSCollector::do_compaction_work(bool clear_all_soft_refs) { _intra_sweep_estimate.padded_average()); } + { + TraceCMSMemoryManagerStats(); + } GenMarkSweep::invoke_at_safepoint(_cmsGen->level(), ref_processor(), clear_all_soft_refs); #ifdef ASSERT @@ -3415,6 +3487,7 @@ CMSPhaseAccounting::~CMSPhaseAccounting() { void CMSCollector::checkpointRootsInitial(bool asynch) { assert(_collectorState == InitialMarking, "Wrong collector state"); check_correct_thread_executing(); + TraceCMSMemoryManagerStats tms(_collectorState); ReferenceProcessor* rp = ref_processor(); SpecializationStats::clear(); assert(_restart_addr == NULL, "Control point invariant"); @@ -4748,6 +4821,7 @@ void CMSCollector::checkpointRootsFinal(bool asynch, // world is stopped at this checkpoint assert(SafepointSynchronize::is_at_safepoint(), "world should be stopped"); + TraceCMSMemoryManagerStats tms(_collectorState); verify_work_stacks_empty(); verify_overflow_empty(); @@ -5849,6 +5923,8 @@ void CMSCollector::sweep(bool asynch) { verify_work_stacks_empty(); verify_overflow_empty(); increment_sweep_count(); + TraceCMSMemoryManagerStats tms(_collectorState); + _inter_sweep_timer.stop(); _inter_sweep_estimate.sample(_inter_sweep_timer.seconds()); size_policy()->avg_cms_free_at_sweep()->sample(_cmsGen->free()); @@ -7861,14 +7937,20 @@ size_t SweepClosure::do_blk_careful(HeapWord* addr) { FreeChunk* fc = (FreeChunk*)addr; size_t res; - // check if we are done sweepinrg - if (addr == _limit) { // we have swept up to the limit, do nothing more + // Check if we are done sweeping. Below we check "addr >= _limit" rather + // than "addr == _limit" because although _limit was a block boundary when + // we started the sweep, it may no longer be one because heap expansion + // may have caused us to coalesce the block ending at the address _limit + // with a newly expanded chunk (this happens when _limit was set to the + // previous _end of the space), so we may have stepped past _limit; see CR 6977970. + if (addr >= _limit) { // we have swept up to or past the limit, do nothing more assert(_limit >= _sp->bottom() && _limit <= _sp->end(), "sweep _limit out of bounds"); + assert(addr < _sp->end(), "addr out of bounds"); // help the closure application finish - return pointer_delta(_sp->end(), _limit); + return pointer_delta(_sp->end(), addr); } - assert(addr <= _limit, "sweep invariant"); + assert(addr < _limit, "sweep invariant"); // check if we should yield do_yield_check(addr); @@ -9121,3 +9203,57 @@ size_t MarkDeadObjectsClosure::do_blk(HeapWord* addr) { } return res; } + +TraceCMSMemoryManagerStats::TraceCMSMemoryManagerStats(CMSCollector::CollectorState phase): TraceMemoryManagerStats() { + + switch (phase) { + case CMSCollector::InitialMarking: + initialize(true /* fullGC */ , + true /* recordGCBeginTime */, + true /* recordPreGCUsage */, + false /* recordPeakUsage */, + false /* recordPostGCusage */, + true /* recordAccumulatedGCTime */, + false /* recordGCEndTime */, + false /* countCollection */ ); + break; + + case CMSCollector::FinalMarking: + initialize(true /* fullGC */ , + false /* recordGCBeginTime */, + false /* recordPreGCUsage */, + false /* recordPeakUsage */, + false /* recordPostGCusage */, + true /* recordAccumulatedGCTime */, + false /* recordGCEndTime */, + false /* countCollection */ ); + break; + + case CMSCollector::Sweeping: + initialize(true /* fullGC */ , + false /* recordGCBeginTime */, + false /* recordPreGCUsage */, + true /* recordPeakUsage */, + true /* recordPostGCusage */, + false /* recordAccumulatedGCTime */, + true /* recordGCEndTime */, + true /* countCollection */ ); + break; + + default: + ShouldNotReachHere(); + } +} + +// when bailing out of cms in concurrent mode failure +TraceCMSMemoryManagerStats::TraceCMSMemoryManagerStats(): TraceMemoryManagerStats() { + initialize(true /* fullGC */ , + true /* recordGCBeginTime */, + true /* recordPreGCUsage */, + true /* recordPeakUsage */, + true /* recordPostGCusage */, + true /* recordAccumulatedGCTime */, + true /* recordGCEndTime */, + true /* countCollection */ ); +} + diff --git a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.hpp b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.hpp index 451538472..8455ded4a 100644 --- a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.hpp +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.hpp @@ -507,6 +507,7 @@ class CMSCollector: public CHeapObj { friend class VM_CMS_Operation; friend class VM_CMS_Initial_Mark; friend class VM_CMS_Final_Remark; + friend class TraceCMSMemoryManagerStats; private: jlong _time_of_last_gc; @@ -1009,10 +1010,10 @@ class ConcurrentMarkSweepGeneration: public CardGeneration { // Non-product stat counters NOT_PRODUCT( - int _numObjectsPromoted; - int _numWordsPromoted; - int _numObjectsAllocated; - int _numWordsAllocated; + size_t _numObjectsPromoted; + size_t _numWordsPromoted; + size_t _numObjectsAllocated; + size_t _numWordsAllocated; ) // Used for sizing decisions @@ -1858,3 +1859,11 @@ public: _dead_bit_map(dead_bit_map) {} size_t do_blk(HeapWord* addr); }; + +class TraceCMSMemoryManagerStats : public TraceMemoryManagerStats { + + public: + TraceCMSMemoryManagerStats(CMSCollector::CollectorState phase); + TraceCMSMemoryManagerStats(); +}; + diff --git a/src/share/vm/gc_implementation/concurrentMarkSweep/freeChunk.hpp b/src/share/vm/gc_implementation/concurrentMarkSweep/freeChunk.hpp index 19e3c1c0d..760fabf43 100644 --- a/src/share/vm/gc_implementation/concurrentMarkSweep/freeChunk.hpp +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/freeChunk.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2001, 2008, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -110,15 +110,21 @@ class FreeChunk VALUE_OBJ_CLASS_SPEC { } void linkNext(FreeChunk* ptr) { _next = ptr; } void linkPrev(FreeChunk* ptr) { - LP64_ONLY(if (UseCompressedOops) _prev = ptr; else) - _prev = (FreeChunk*)((intptr_t)ptr | 0x1); + LP64_ONLY(if (UseCompressedOops) _prev = ptr; else) + _prev = (FreeChunk*)((intptr_t)ptr | 0x1); } void clearPrev() { _prev = NULL; } void clearNext() { _next = NULL; } void markNotFree() { - LP64_ONLY(if (UseCompressedOops) set_mark(markOopDesc::prototype());) - // Also set _prev to null - _prev = NULL; + // Set _prev (klass) to null before (if) clearing the mark word below + _prev = NULL; +#ifdef _LP64 + if (UseCompressedOops) { + OrderAccess::storestore(); + set_mark(markOopDesc::prototype()); + } +#endif + assert(!isFree(), "Error"); } // Return the address past the end of this chunk diff --git a/src/share/vm/gc_implementation/concurrentMarkSweep/promotionInfo.cpp b/src/share/vm/gc_implementation/concurrentMarkSweep/promotionInfo.cpp index 9b1fdf07a..09602b5b7 100644 --- a/src/share/vm/gc_implementation/concurrentMarkSweep/promotionInfo.cpp +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/promotionInfo.cpp @@ -330,7 +330,7 @@ void PromotionInfo::verify() const { void PromotionInfo::print_on(outputStream* st) const { SpoolBlock* curSpool = NULL; size_t i = 0; - st->print_cr("start & end indices: [" SIZE_FORMAT ", " SIZE_FORMAT ")", + st->print_cr(" start & end indices: [" SIZE_FORMAT ", " SIZE_FORMAT ")", _firstIndex, _nextIndex); for (curSpool = _spoolHead; curSpool != _spoolTail && curSpool != NULL; curSpool = curSpool->nextSpoolBlock) { @@ -350,7 +350,7 @@ void PromotionInfo::print_on(outputStream* st) const { st->print_cr(" free "); i++; } - st->print_cr(SIZE_FORMAT " header spooling blocks", i); + st->print_cr(" " SIZE_FORMAT " header spooling blocks", i); } void SpoolBlock::print_on(outputStream* st) const { diff --git a/src/share/vm/gc_implementation/g1/concurrentG1Refine.cpp b/src/share/vm/gc_implementation/g1/concurrentG1Refine.cpp index 78d3d5cb2..e144aa7cc 100644 --- a/src/share/vm/gc_implementation/g1/concurrentG1Refine.cpp +++ b/src/share/vm/gc_implementation/g1/concurrentG1Refine.cpp @@ -339,7 +339,9 @@ jbyte* ConcurrentG1Refine::cache_insert(jbyte* card_ptr, bool* defer) { return res; } -void ConcurrentG1Refine::clean_up_cache(int worker_i, G1RemSet* g1rs) { +void ConcurrentG1Refine::clean_up_cache(int worker_i, + G1RemSet* g1rs, + DirtyCardQueue* into_cset_dcq) { assert(!use_cache(), "cache should be disabled"); int start_idx; @@ -353,7 +355,19 @@ void ConcurrentG1Refine::clean_up_cache(int worker_i, G1RemSet* g1rs) { for (int i = start_idx; i < end_idx; i++) { jbyte* entry = _hot_cache[i]; if (entry != NULL) { - g1rs->concurrentRefineOneCard(entry, worker_i); + if (g1rs->concurrentRefineOneCard(entry, worker_i, true)) { + // 'entry' contains references that point into the current + // collection set. We need to record 'entry' in the DCQS + // that's used for that purpose. + // + // The only time we care about recording cards that contain + // references that point into the collection set is during + // RSet updating while within an evacuation pause. + // In this case worker_i should be the id of a GC worker thread + assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause"); + assert(worker_i < (int) DirtyCardQueueSet::num_par_ids(), "incorrect worker id"); + into_cset_dcq->enqueue(entry); + } } } } diff --git a/src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp b/src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp index 85a50f7c7..f5c3653f8 100644 --- a/src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp +++ b/src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2001, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -184,7 +184,7 @@ class ConcurrentG1Refine: public CHeapObj { jbyte* cache_insert(jbyte* card_ptr, bool* defer); // Process the cached entries. - void clean_up_cache(int worker_i, G1RemSet* g1rs); + void clean_up_cache(int worker_i, G1RemSet* g1rs, DirtyCardQueue* into_cset_dcq); // Set up for parallel processing of the cards in the hot cache void clear_hot_cache_claimed_index() { diff --git a/src/share/vm/gc_implementation/g1/concurrentMark.cpp b/src/share/vm/gc_implementation/g1/concurrentMark.cpp index ddbad890d..4dd197ebd 100644 --- a/src/share/vm/gc_implementation/g1/concurrentMark.cpp +++ b/src/share/vm/gc_implementation/g1/concurrentMark.cpp @@ -2586,9 +2586,6 @@ void ConcurrentMark::complete_marking_in_collection_set() { double end_time = os::elapsedTime(); double elapsed_time_ms = (end_time - start) * 1000.0; g1h->g1_policy()->record_mark_closure_time(elapsed_time_ms); - if (PrintGCDetails) { - gclog_or_tty->print_cr("Mark closure took %5.2f ms.", elapsed_time_ms); - } ClearMarksInHRClosure clr(nextMarkBitMap()); g1h->collection_set_iterate(&clr); diff --git a/src/share/vm/gc_implementation/g1/dirtyCardQueue.cpp b/src/share/vm/gc_implementation/g1/dirtyCardQueue.cpp index 760f36301..c8100c477 100644 --- a/src/share/vm/gc_implementation/g1/dirtyCardQueue.cpp +++ b/src/share/vm/gc_implementation/g1/dirtyCardQueue.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2001, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -178,13 +178,14 @@ DirtyCardQueueSet::get_completed_buffer(int stop_at) { } bool DirtyCardQueueSet:: -apply_closure_to_completed_buffer_helper(int worker_i, +apply_closure_to_completed_buffer_helper(CardTableEntryClosure* cl, + int worker_i, BufferNode* nd) { if (nd != NULL) { void **buf = BufferNode::make_buffer_from_node(nd); size_t index = nd->index(); bool b = - DirtyCardQueue::apply_closure_to_buffer(_closure, buf, + DirtyCardQueue::apply_closure_to_buffer(cl, buf, index, _sz, true, worker_i); if (b) { @@ -199,17 +200,24 @@ apply_closure_to_completed_buffer_helper(int worker_i, } } -bool DirtyCardQueueSet::apply_closure_to_completed_buffer(int worker_i, +bool DirtyCardQueueSet::apply_closure_to_completed_buffer(CardTableEntryClosure* cl, + int worker_i, int stop_at, - bool during_pause) -{ + bool during_pause) { assert(!during_pause || stop_at == 0, "Should not leave any completed buffers during a pause"); BufferNode* nd = get_completed_buffer(stop_at); - bool res = apply_closure_to_completed_buffer_helper(worker_i, nd); + bool res = apply_closure_to_completed_buffer_helper(cl, worker_i, nd); if (res) Atomic::inc(&_processed_buffers_rs_thread); return res; } +bool DirtyCardQueueSet::apply_closure_to_completed_buffer(int worker_i, + int stop_at, + bool during_pause) { + return apply_closure_to_completed_buffer(_closure, worker_i, + stop_at, during_pause); +} + void DirtyCardQueueSet::apply_closure_to_all_completed_buffers() { BufferNode* nd = _completed_buffers_head; while (nd != NULL) { @@ -222,8 +230,8 @@ void DirtyCardQueueSet::apply_closure_to_all_completed_buffers() { } } -void DirtyCardQueueSet::abandon_logs() { - assert(SafepointSynchronize::is_at_safepoint(), "Must be at safepoint."); +// Deallocates any completed log buffers +void DirtyCardQueueSet::clear() { BufferNode* buffers_to_delete = NULL; { MutexLockerEx x(_cbl_mon, Mutex::_no_safepoint_check_flag); @@ -242,6 +250,12 @@ void DirtyCardQueueSet::abandon_logs() { buffers_to_delete = nd->next(); deallocate_buffer(BufferNode::make_buffer_from_node(nd)); } + +} + +void DirtyCardQueueSet::abandon_logs() { + assert(SafepointSynchronize::is_at_safepoint(), "Must be at safepoint."); + clear(); // Since abandon is done only at safepoints, we can safely manipulate // these queues. for (JavaThread* t = Threads::first(); t; t = t->next()) { diff --git a/src/share/vm/gc_implementation/g1/dirtyCardQueue.hpp b/src/share/vm/gc_implementation/g1/dirtyCardQueue.hpp index e1b689816..524c0c256 100644 --- a/src/share/vm/gc_implementation/g1/dirtyCardQueue.hpp +++ b/src/share/vm/gc_implementation/g1/dirtyCardQueue.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2001, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -123,7 +123,21 @@ public: int stop_at = 0, bool during_pause = false); - bool apply_closure_to_completed_buffer_helper(int worker_i, + // If there exists some completed buffer, pop it, then apply the + // specified closure to all its elements, nulling out those elements + // processed. If all elements are processed, returns "true". If no + // completed buffers exist, returns false. If a completed buffer exists, + // but is only partially completed before a "yield" happens, the + // partially completed buffer (with its processed elements set to NULL) + // is returned to the completed buffer set, and this call returns false. + bool apply_closure_to_completed_buffer(CardTableEntryClosure* cl, + int worker_i = 0, + int stop_at = 0, + bool during_pause = false); + + // Helper routine for the above. + bool apply_closure_to_completed_buffer_helper(CardTableEntryClosure* cl, + int worker_i, BufferNode* nd); BufferNode* get_completed_buffer(int stop_at); @@ -136,6 +150,9 @@ public: return &_shared_dirty_card_queue; } + // Deallocate any completed log buffers + void clear(); + // If a full collection is happening, reset partial logs, and ignore // completed ones: the full collection will make them all irrelevant. void abandon_logs(); diff --git a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp index 658ac777f..e71c10978 100644 --- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp @@ -56,7 +56,12 @@ public: _sts(sts), _g1rs(g1rs), _cg1r(cg1r), _concurrent(true) {} bool do_card_ptr(jbyte* card_ptr, int worker_i) { - _g1rs->concurrentRefineOneCard(card_ptr, worker_i); + bool oops_into_cset = _g1rs->concurrentRefineOneCard(card_ptr, worker_i, false); + // This path is executed by the concurrent refine or mutator threads, + // concurrently, and so we do not care if card_ptr contains references + // that point into the collection set. + assert(!oops_into_cset, "should be"); + if (_concurrent && _sts->should_yield()) { // Caller will actually yield. return false; @@ -1039,29 +1044,56 @@ resize_if_necessary_after_full_collection(size_t word_size) { const size_t capacity_after_gc = capacity(); const size_t free_after_gc = capacity_after_gc - used_after_gc; + // This is enforced in arguments.cpp. + assert(MinHeapFreeRatio <= MaxHeapFreeRatio, + "otherwise the code below doesn't make sense"); + // We don't have floating point command-line arguments - const double minimum_free_percentage = (double) MinHeapFreeRatio / 100; + const double minimum_free_percentage = (double) MinHeapFreeRatio / 100.0; const double maximum_used_percentage = 1.0 - minimum_free_percentage; - const double maximum_free_percentage = (double) MaxHeapFreeRatio / 100; + const double maximum_free_percentage = (double) MaxHeapFreeRatio / 100.0; const double minimum_used_percentage = 1.0 - maximum_free_percentage; - size_t minimum_desired_capacity = (size_t) (used_after_gc / maximum_used_percentage); - size_t maximum_desired_capacity = (size_t) (used_after_gc / minimum_used_percentage); - - // Don't shrink less than the initial size. - minimum_desired_capacity = - MAX2(minimum_desired_capacity, - collector_policy()->initial_heap_byte_size()); - maximum_desired_capacity = - MAX2(maximum_desired_capacity, - collector_policy()->initial_heap_byte_size()); - - // We are failing here because minimum_desired_capacity is - assert(used_after_gc <= minimum_desired_capacity, "sanity check"); - assert(minimum_desired_capacity <= maximum_desired_capacity, "sanity check"); + const size_t min_heap_size = collector_policy()->min_heap_byte_size(); + const size_t max_heap_size = collector_policy()->max_heap_byte_size(); + + // We have to be careful here as these two calculations can overflow + // 32-bit size_t's. + double used_after_gc_d = (double) used_after_gc; + double minimum_desired_capacity_d = used_after_gc_d / maximum_used_percentage; + double maximum_desired_capacity_d = used_after_gc_d / minimum_used_percentage; + + // Let's make sure that they are both under the max heap size, which + // by default will make them fit into a size_t. + double desired_capacity_upper_bound = (double) max_heap_size; + minimum_desired_capacity_d = MIN2(minimum_desired_capacity_d, + desired_capacity_upper_bound); + maximum_desired_capacity_d = MIN2(maximum_desired_capacity_d, + desired_capacity_upper_bound); + + // We can now safely turn them into size_t's. + size_t minimum_desired_capacity = (size_t) minimum_desired_capacity_d; + size_t maximum_desired_capacity = (size_t) maximum_desired_capacity_d; + + // This assert only makes sense here, before we adjust them + // with respect to the min and max heap size. + assert(minimum_desired_capacity <= maximum_desired_capacity, + err_msg("minimum_desired_capacity = "SIZE_FORMAT", " + "maximum_desired_capacity = "SIZE_FORMAT, + minimum_desired_capacity, maximum_desired_capacity)); + + // Should not be greater than the heap max size. No need to adjust + // it with respect to the heap min size as it's a lower bound (i.e., + // we'll try to make the capacity larger than it, not smaller). + minimum_desired_capacity = MIN2(minimum_desired_capacity, max_heap_size); + // Should not be less than the heap min size. No need to adjust it + // with respect to the heap max size as it's an upper bound (i.e., + // we'll try to make the capacity smaller than it, not greater). + maximum_desired_capacity = MAX2(maximum_desired_capacity, min_heap_size); if (PrintGC && Verbose) { - const double free_percentage = ((double)free_after_gc) / capacity(); + const double free_percentage = + (double) free_after_gc / (double) capacity_after_gc; gclog_or_tty->print_cr("Computing new size after full GC "); gclog_or_tty->print_cr(" " " minimum_free_percentage: %6.2f", @@ -1073,45 +1105,47 @@ resize_if_necessary_after_full_collection(size_t word_size) { " capacity: %6.1fK" " minimum_desired_capacity: %6.1fK" " maximum_desired_capacity: %6.1fK", - capacity() / (double) K, - minimum_desired_capacity / (double) K, - maximum_desired_capacity / (double) K); + (double) capacity_after_gc / (double) K, + (double) minimum_desired_capacity / (double) K, + (double) maximum_desired_capacity / (double) K); gclog_or_tty->print_cr(" " - " free_after_gc : %6.1fK" - " used_after_gc : %6.1fK", - free_after_gc / (double) K, - used_after_gc / (double) K); + " free_after_gc: %6.1fK" + " used_after_gc: %6.1fK", + (double) free_after_gc / (double) K, + (double) used_after_gc / (double) K); gclog_or_tty->print_cr(" " " free_percentage: %6.2f", free_percentage); } - if (capacity() < minimum_desired_capacity) { + if (capacity_after_gc < minimum_desired_capacity) { // Don't expand unless it's significant size_t expand_bytes = minimum_desired_capacity - capacity_after_gc; expand(expand_bytes); if (PrintGC && Verbose) { - gclog_or_tty->print_cr(" expanding:" + gclog_or_tty->print_cr(" " + " expanding:" + " max_heap_size: %6.1fK" " minimum_desired_capacity: %6.1fK" " expand_bytes: %6.1fK", - minimum_desired_capacity / (double) K, - expand_bytes / (double) K); + (double) max_heap_size / (double) K, + (double) minimum_desired_capacity / (double) K, + (double) expand_bytes / (double) K); } // No expansion, now see if we want to shrink - } else if (capacity() > maximum_desired_capacity) { + } else if (capacity_after_gc > maximum_desired_capacity) { // Capacity too large, compute shrinking size size_t shrink_bytes = capacity_after_gc - maximum_desired_capacity; shrink(shrink_bytes); if (PrintGC && Verbose) { gclog_or_tty->print_cr(" " " shrinking:" - " initSize: %.1fK" - " maximum_desired_capacity: %.1fK", - collector_policy()->initial_heap_byte_size() / (double) K, - maximum_desired_capacity / (double) K); - gclog_or_tty->print_cr(" " - " shrink_bytes: %.1fK", - shrink_bytes / (double) K); + " min_heap_size: %6.1fK" + " maximum_desired_capacity: %6.1fK" + " shrink_bytes: %6.1fK", + (double) min_heap_size / (double) K, + (double) maximum_desired_capacity / (double) K, + (double) shrink_bytes / (double) K); } } } @@ -1322,6 +1356,7 @@ G1CollectedHeap::G1CollectedHeap(G1CollectorPolicy* policy_) : SharedHeap(policy_), _g1_policy(policy_), _dirty_card_queue_set(false), + _into_cset_dirty_card_queue_set(false), _ref_processor(NULL), _process_strong_tasks(new SubTasksDone(G1H_PS_NumElements)), _bot_shared(NULL), @@ -1572,6 +1607,16 @@ jint G1CollectedHeap::initialize() { Shared_DirtyCardQ_lock, &JavaThread::dirty_card_queue_set()); } + + // Initialize the card queue set used to hold cards containing + // references into the collection set. + _into_cset_dirty_card_queue_set.initialize(DirtyCardQ_CBL_mon, + DirtyCardQ_FL_lock, + -1, // never trigger processing + -1, // no limit on length + Shared_DirtyCardQ_lock, + &JavaThread::dirty_card_queue_set()); + // In case we're keeping closure specialization stats, initialize those // counts and that mechanism. SpecializationStats::clear(); @@ -1603,14 +1648,16 @@ size_t G1CollectedHeap::capacity() const { return _g1_committed.byte_size(); } -void G1CollectedHeap::iterate_dirty_card_closure(bool concurrent, +void G1CollectedHeap::iterate_dirty_card_closure(CardTableEntryClosure* cl, + DirtyCardQueue* into_cset_dcq, + bool concurrent, int worker_i) { // Clean cards in the hot card cache - concurrent_g1_refine()->clean_up_cache(worker_i, g1_rem_set()); + concurrent_g1_refine()->clean_up_cache(worker_i, g1_rem_set(), into_cset_dcq); DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); int n_completed_buffers = 0; - while (dcqs.apply_closure_to_completed_buffer(worker_i, 0, true)) { + while (dcqs.apply_closure_to_completed_buffer(cl, worker_i, 0, true)) { n_completed_buffers++; } g1_policy()->record_update_rs_processed_buffers(worker_i, @@ -2147,9 +2194,12 @@ size_t G1CollectedHeap::unsafe_max_tlab_alloc(Thread* ignored) const { } } -HeapWord* G1CollectedHeap::allocate_new_tlab(size_t size) { +HeapWord* G1CollectedHeap::allocate_new_tlab(size_t word_size) { + assert(!isHumongous(word_size), + err_msg("a TLAB should not be of humongous size, " + "word_size = "SIZE_FORMAT, word_size)); bool dummy; - return G1CollectedHeap::mem_allocate(size, false, true, &dummy); + return G1CollectedHeap::mem_allocate(word_size, false, true, &dummy); } bool G1CollectedHeap::allocs_are_zero_filled() { @@ -2692,6 +2742,35 @@ struct PrepareForRSScanningClosure : public HeapRegionClosure { } }; +#if TASKQUEUE_STATS +void G1CollectedHeap::print_taskqueue_stats_hdr(outputStream* const st) { + st->print_raw_cr("GC Task Stats"); + st->print_raw("thr "); TaskQueueStats::print_header(1, st); st->cr(); + st->print_raw("--- "); TaskQueueStats::print_header(2, st); st->cr(); +} + +void G1CollectedHeap::print_taskqueue_stats(outputStream* const st) const { + print_taskqueue_stats_hdr(st); + + TaskQueueStats totals; + const int n = MAX2(workers()->total_workers(), 1); + for (int i = 0; i < n; ++i) { + st->print("%3d ", i); task_queue(i)->stats.print(st); st->cr(); + totals += task_queue(i)->stats; + } + st->print_raw("tot "); totals.print(st); st->cr(); + + DEBUG_ONLY(totals.verify()); +} + +void G1CollectedHeap::reset_taskqueue_stats() { + const int n = MAX2(workers()->total_workers(), 1); + for (int i = 0; i < n; ++i) { + task_queue(i)->stats.reset(); + } +} +#endif // TASKQUEUE_STATS + void G1CollectedHeap::do_collection_pause_at_safepoint(double target_pause_time_ms) { if (GC_locker::check_active_before_gc()) { @@ -2825,93 +2904,57 @@ G1CollectedHeap::do_collection_pause_at_safepoint(double target_pause_time_ms) { g1_policy()->print_collection_set(g1_policy()->inc_cset_head(), gclog_or_tty); #endif // YOUNG_LIST_VERBOSE - // Now choose the CS. We may abandon a pause if we find no - // region that will fit in the MMU pause. - bool abandoned = g1_policy()->choose_collection_set(target_pause_time_ms); + g1_policy()->choose_collection_set(target_pause_time_ms); // Nothing to do if we were unable to choose a collection set. - if (!abandoned) { #if G1_REM_SET_LOGGING - gclog_or_tty->print_cr("\nAfter pause, heap:"); - print(); + gclog_or_tty->print_cr("\nAfter pause, heap:"); + print(); #endif - PrepareForRSScanningClosure prepare_for_rs_scan; - collection_set_iterate(&prepare_for_rs_scan); + PrepareForRSScanningClosure prepare_for_rs_scan; + collection_set_iterate(&prepare_for_rs_scan); - setup_surviving_young_words(); + setup_surviving_young_words(); - // Set up the gc allocation regions. - get_gc_alloc_regions(); + // Set up the gc allocation regions. + get_gc_alloc_regions(); - // Actually do the work... - evacuate_collection_set(); + // Actually do the work... + evacuate_collection_set(); - free_collection_set(g1_policy()->collection_set()); - g1_policy()->clear_collection_set(); + free_collection_set(g1_policy()->collection_set()); + g1_policy()->clear_collection_set(); - cleanup_surviving_young_words(); + cleanup_surviving_young_words(); - // Start a new incremental collection set for the next pause. - g1_policy()->start_incremental_cset_building(); + // Start a new incremental collection set for the next pause. + g1_policy()->start_incremental_cset_building(); - // Clear the _cset_fast_test bitmap in anticipation of adding - // regions to the incremental collection set for the next - // evacuation pause. - clear_cset_fast_test(); + // Clear the _cset_fast_test bitmap in anticipation of adding + // regions to the incremental collection set for the next + // evacuation pause. + clear_cset_fast_test(); - if (g1_policy()->in_young_gc_mode()) { - _young_list->reset_sampled_info(); + if (g1_policy()->in_young_gc_mode()) { + _young_list->reset_sampled_info(); - // Don't check the whole heap at this point as the - // GC alloc regions from this pause have been tagged - // as survivors and moved on to the survivor list. - // Survivor regions will fail the !is_young() check. - assert(check_young_list_empty(false /* check_heap */), - "young list should be empty"); + // Don't check the whole heap at this point as the + // GC alloc regions from this pause have been tagged + // as survivors and moved on to the survivor list. + // Survivor regions will fail the !is_young() check. + assert(check_young_list_empty(false /* check_heap */), + "young list should be empty"); #if YOUNG_LIST_VERBOSE - gclog_or_tty->print_cr("Before recording survivors.\nYoung List:"); - _young_list->print(); + gclog_or_tty->print_cr("Before recording survivors.\nYoung List:"); + _young_list->print(); #endif // YOUNG_LIST_VERBOSE - g1_policy()->record_survivor_regions(_young_list->survivor_length(), + g1_policy()->record_survivor_regions(_young_list->survivor_length(), _young_list->first_survivor_region(), _young_list->last_survivor_region()); - _young_list->reset_auxilary_lists(); - } - } else { - // We have abandoned the current collection. This can only happen - // if we're not doing young or partially young collections, and - // we didn't find an old region that we're able to collect within - // the allowed time. - - assert(g1_policy()->collection_set() == NULL, "should be"); - assert(_young_list->length() == 0, "because it should be"); - - // This should be a no-op. - abandon_collection_set(g1_policy()->inc_cset_head()); - - g1_policy()->clear_incremental_cset(); - g1_policy()->stop_incremental_cset_building(); - - // Start a new incremental collection set for the next pause. - g1_policy()->start_incremental_cset_building(); - - // Clear the _cset_fast_test bitmap in anticipation of adding - // regions to the incremental collection set for the next - // evacuation pause. - clear_cset_fast_test(); - - // This looks confusing, because the DPT should really be empty - // at this point -- since we have not done any collection work, - // there should not be any derived pointers in the table to update; - // however, there is some additional state in the DPT which is - // reset at the end of the (null) "gc" here via the following call. - // A better approach might be to split off that state resetting work - // into a separate method that asserts that the DPT is empty and call - // that here. That is deferred for now. - COMPILER2_PRESENT(DerivedPointerTable::update_pointers()); + _young_list->reset_auxilary_lists(); } if (evacuation_failed()) { @@ -2945,7 +2988,7 @@ G1CollectedHeap::do_collection_pause_at_safepoint(double target_pause_time_ms) { double end_time_sec = os::elapsedTime(); double pause_time_ms = (end_time_sec - start_time_sec) * MILLIUNITS; g1_policy()->record_pause_time_ms(pause_time_ms); - g1_policy()->record_collection_pause_end(abandoned); + g1_policy()->record_collection_pause_end(); assert(regions_accounted_for(), "Region leakage."); @@ -2988,6 +3031,9 @@ G1CollectedHeap::do_collection_pause_at_safepoint(double target_pause_time_ms) { } } + TASKQUEUE_STATS_ONLY(if (ParallelGCVerbose) print_taskqueue_stats()); + TASKQUEUE_STATS_ONLY(reset_taskqueue_stats()); + if (PrintHeapAtGC) { Universe::print_heap_after_gc(); } @@ -3346,25 +3392,6 @@ public: } }; -class UpdateRSetImmediate : public OopsInHeapRegionClosure { -private: - G1CollectedHeap* _g1; - G1RemSet* _g1_rem_set; -public: - UpdateRSetImmediate(G1CollectedHeap* g1) : - _g1(g1), _g1_rem_set(g1->g1_rem_set()) {} - - virtual void do_oop(narrowOop* p) { do_oop_work(p); } - virtual void do_oop( oop* p) { do_oop_work(p); } - template <class T> void do_oop_work(T* p) { - assert(_from->is_in_reserved(p), "paranoia"); - T heap_oop = oopDesc::load_heap_oop(p); - if (!oopDesc::is_null(heap_oop) && !_from->is_survivor()) { - _g1_rem_set->par_write_ref(_from, p, 0); - } - } -}; - class UpdateRSetDeferred : public OopsInHeapRegionClosure { private: G1CollectedHeap* _g1; @@ -3389,8 +3416,6 @@ public: } }; - - class RemoveSelfPointerClosure: public ObjectClosure { private: G1CollectedHeap* _g1; @@ -3453,7 +3478,7 @@ public: }; void G1CollectedHeap::remove_self_forwarding_pointers() { - UpdateRSetImmediate immediate_update(_g1h); + UpdateRSetImmediate immediate_update(_g1h->g1_rem_set()); DirtyCardQueue dcq(&_g1h->dirty_card_queue_set()); UpdateRSetDeferred deferred_update(_g1h, &dcq); OopsInHeapRegionClosure *cl; @@ -3583,7 +3608,7 @@ void G1CollectedHeap::handle_evacuation_failure_common(oop old, markOop m) { if (!r->evacuation_failed()) { r->set_evacuation_failed(true); if (G1PrintHeapRegions) { - gclog_or_tty->print("evacuation failed in heap region "PTR_FORMAT" " + gclog_or_tty->print("overflow in heap region "PTR_FORMAT" " "["PTR_FORMAT","PTR_FORMAT")\n", r, r->bottom(), r->end()); } @@ -3617,6 +3642,10 @@ void G1CollectedHeap::preserve_mark_if_necessary(oop obj, markOop m) { HeapWord* G1CollectedHeap::par_allocate_during_gc(GCAllocPurpose purpose, size_t word_size) { + assert(!isHumongous(word_size), + err_msg("we should not be seeing humongous allocation requests " + "during GC, word_size = "SIZE_FORMAT, word_size)); + HeapRegion* alloc_region = _gc_alloc_regions[purpose]; // let the caller handle alloc failure if (alloc_region == NULL) return NULL; @@ -3649,6 +3678,10 @@ G1CollectedHeap::allocate_during_gc_slow(GCAllocPurpose purpose, HeapRegion* alloc_region, bool par, size_t word_size) { + assert(!isHumongous(word_size), + err_msg("we should not be seeing humongous allocation requests " + "during GC, word_size = "SIZE_FORMAT, word_size)); + HeapWord* block = NULL; // In the parallel case, a previous thread to obtain the lock may have // already assigned a new gc_alloc_region. @@ -3754,10 +3787,6 @@ G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, int queue_num) _surviving_alloc_buffer(g1h->desired_plab_sz(GCAllocForSurvived)), _tenured_alloc_buffer(g1h->desired_plab_sz(GCAllocForTenured)), _age_table(false), -#if G1_DETAILED_STATS - _pushes(0), _pops(0), _steals(0), - _steal_attempts(0), _overflow_pushes(0), -#endif _strong_roots_time(0), _term_time(0), _alloc_buffer_waste(0), _undo_waste(0) { @@ -3777,14 +3806,41 @@ G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, int queue_num) _surviving_young_words = _surviving_young_words_base + PADDING_ELEM_NUM; memset(_surviving_young_words, 0, real_length * sizeof(size_t)); - _overflowed_refs = new OverflowQueue(10); - _alloc_buffers[GCAllocForSurvived] = &_surviving_alloc_buffer; _alloc_buffers[GCAllocForTenured] = &_tenured_alloc_buffer; _start = os::elapsedTime(); } +void +G1ParScanThreadState::print_termination_stats_hdr(outputStream* const st) +{ + st->print_raw_cr("GC Termination Stats"); + st->print_raw_cr(" elapsed --strong roots-- -------termination-------" + " ------waste (KiB)------"); + st->print_raw_cr("thr ms ms % ms % attempts" + " total alloc undo"); + st->print_raw_cr("--- --------- --------- ------ --------- ------ --------" + " ------- ------- -------"); +} + +void +G1ParScanThreadState::print_termination_stats(int i, + outputStream* const st) const +{ + const double elapsed_ms = elapsed_time() * 1000.0; + const double s_roots_ms = strong_roots_time() * 1000.0; + const double term_ms = term_time() * 1000.0; + st->print_cr("%3d %9.2f %9.2f %6.2f " + "%9.2f %6.2f " SIZE_FORMAT_W(8) " " + SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7), + i, elapsed_ms, s_roots_ms, s_roots_ms * 100 / elapsed_ms, + term_ms, term_ms * 100 / elapsed_ms, term_attempts(), + (alloc_buffer_waste() + undo_waste()) * HeapWordSize / K, + alloc_buffer_waste() * HeapWordSize / K, + undo_waste() * HeapWordSize / K); +} + G1ParClosureSuper::G1ParClosureSuper(G1CollectedHeap* g1, G1ParScanThreadState* par_scan_state) : _g1(g1), _g1_rem(_g1->g1_rem_set()), _cm(_g1->concurrent_mark()), _par_scan_state(par_scan_state) { } @@ -3991,12 +4047,9 @@ public: G1ParScanThreadState* pss = par_scan_state(); while (true) { pss->trim_queue(); - IF_G1_DETAILED_STATS(pss->note_steal_attempt()); StarTask stolen_task; if (queues()->steal(pss->queue_num(), pss->hash_seed(), stolen_task)) { - IF_G1_DETAILED_STATS(pss->note_steal()); - // slightly paranoid tests; I'm trying to catch potential // problems before we go into push_on_queue to know where the // problem is coming from @@ -4115,35 +4168,9 @@ public: // Clean up any par-expanded rem sets. HeapRegionRemSet::par_cleanup(); - MutexLocker x(stats_lock()); if (ParallelGCVerbose) { - gclog_or_tty->print("Thread %d complete:\n", i); -#if G1_DETAILED_STATS - gclog_or_tty->print(" Pushes: %7d Pops: %7d Overflows: %7d Steals %7d (in %d attempts)\n", - pss.pushes(), - pss.pops(), - pss.overflow_pushes(), - pss.steals(), - pss.steal_attempts()); -#endif - double elapsed = pss.elapsed(); - double strong_roots = pss.strong_roots_time(); - double term = pss.term_time(); - gclog_or_tty->print(" Elapsed: %7.2f ms.\n" - " Strong roots: %7.2f ms (%6.2f%%)\n" - " Termination: %7.2f ms (%6.2f%%) " - "(in "SIZE_FORMAT" entries)\n", - elapsed * 1000.0, - strong_roots * 1000.0, (strong_roots*100.0/elapsed), - term * 1000.0, (term*100.0/elapsed), - pss.term_attempts()); - size_t total_waste = pss.alloc_buffer_waste() + pss.undo_waste(); - gclog_or_tty->print(" Waste: %8dK\n" - " Alloc Buffer: %8dK\n" - " Undo: %8dK\n", - (total_waste * HeapWordSize) / K, - (pss.alloc_buffer_waste() * HeapWordSize) / K, - (pss.undo_waste() * HeapWordSize) / K); + MutexLocker x(stats_lock()); + pss.print_termination_stats(i); } assert(pss.refs_to_scan() == 0, "Task queue should be empty"); @@ -4260,6 +4287,7 @@ void G1CollectedHeap::evacuate_collection_set() { if (ParallelGCThreads > 0) { // The individual threads will set their evac-failure closures. StrongRootsScope srs(this); + if (ParallelGCVerbose) G1ParScanThreadState::print_termination_stats_hdr(); workers()->run_task(&g1_par_task); } else { StrongRootsScope srs(this); @@ -4293,7 +4321,7 @@ void G1CollectedHeap::evacuate_collection_set() { if (evacuation_failed()) { remove_self_forwarding_pointers(); if (PrintGCDetails) { - gclog_or_tty->print(" (evacuation failed)"); + gclog_or_tty->print(" (to-space overflow)"); } else if (PrintGC) { gclog_or_tty->print("--"); } diff --git a/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp b/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp index 74606c18b..a342d698d 100644 --- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp @@ -46,17 +46,7 @@ class ConcurrentMarkThread; class ConcurrentG1Refine; class ConcurrentZFThread; -// If want to accumulate detailed statistics on work queues -// turn this on. -#define G1_DETAILED_STATS 0 - -#if G1_DETAILED_STATS -# define IF_G1_DETAILED_STATS(code) code -#else -# define IF_G1_DETAILED_STATS(code) -#endif - -typedef GenericTaskQueue<StarTask> RefToScanQueue; +typedef OverflowTaskQueue<StarTask> RefToScanQueue; typedef GenericTaskQueueSet<RefToScanQueue> RefToScanQueueSet; typedef int RegionIdx_t; // needs to hold [ 0..max_regions() ) @@ -471,6 +461,12 @@ protected: virtual void shrink(size_t expand_bytes); void shrink_helper(size_t expand_bytes); + #if TASKQUEUE_STATS + static void print_taskqueue_stats_hdr(outputStream* const st = gclog_or_tty); + void print_taskqueue_stats(outputStream* const st = gclog_or_tty) const; + void reset_taskqueue_stats(); + #endif // TASKQUEUE_STATS + // Do an incremental collection: identify a collection set, and evacuate // its live objects elsewhere. virtual void do_collection_pause(); @@ -505,6 +501,12 @@ protected: // A function to check the consistency of dirty card logs. void check_ct_logs_at_safepoint(); + // A DirtyCardQueueSet that is used to hold cards that contain + // references into the current collection set. This is used to + // update the remembered sets of the regions in the collection + // set in the event of an evacuation failure. + DirtyCardQueueSet _into_cset_dirty_card_queue_set; + // After a collection pause, make the regions in the CS into free // regions. void free_collection_set(HeapRegion* cs_head); @@ -656,11 +658,18 @@ protected: public: void set_refine_cte_cl_concurrency(bool concurrent); - RefToScanQueue *task_queue(int i); + RefToScanQueue *task_queue(int i) const; // A set of cards where updates happened during the GC DirtyCardQueueSet& dirty_card_queue_set() { return _dirty_card_queue_set; } + // A DirtyCardQueueSet that is used to hold cards that contain + // references into the current collection set. This is used to + // update the remembered sets of the regions in the collection + // set in the event of an evacuation failure. + DirtyCardQueueSet& into_cset_dirty_card_queue_set() + { return _into_cset_dirty_card_queue_set; } + // Create a G1CollectedHeap with the specified policy. // Must call the initialize method afterwards. // May not return if something goes wrong. @@ -715,7 +724,9 @@ public: OrderAccess::fence(); } - void iterate_dirty_card_closure(bool concurrent, int worker_i); + void iterate_dirty_card_closure(CardTableEntryClosure* cl, + DirtyCardQueue* into_cset_dcq, + bool concurrent, int worker_i); // The shared block offset table array. G1BlockOffsetSharedArray* bot_shared() const { return _bot_shared; } @@ -1021,7 +1032,7 @@ public: virtual bool supports_tlab_allocation() const; virtual size_t tlab_capacity(Thread* thr) const; virtual size_t unsafe_max_tlab_alloc(Thread* thr) const; - virtual HeapWord* allocate_new_tlab(size_t size); + virtual HeapWord* allocate_new_tlab(size_t word_size); // Can a compiler initialize a new object without store barriers? // This permission only extends from the creation of a new object @@ -1564,9 +1575,6 @@ protected: CardTableModRefBS* _ct_bs; G1RemSet* _g1_rem; - typedef GrowableArray<StarTask> OverflowQueue; - OverflowQueue* _overflowed_refs; - G1ParGCAllocBuffer _surviving_alloc_buffer; G1ParGCAllocBuffer _tenured_alloc_buffer; G1ParGCAllocBuffer* _alloc_buffers[GCAllocPurposeCount]; @@ -1583,10 +1591,6 @@ protected: int _queue_num; size_t _term_attempts; -#if G1_DETAILED_STATS - int _pushes, _pops, _steals, _steal_attempts; - int _overflow_pushes; -#endif double _start; double _start_strong_roots; @@ -1600,7 +1604,7 @@ protected: // this points into the array, as we use the first few entries for padding size_t* _surviving_young_words; -#define PADDING_ELEM_NUM (64 / sizeof(size_t)) +#define PADDING_ELEM_NUM (DEFAULT_CACHE_LINE_SIZE / sizeof(size_t)) void add_to_alloc_buffer_waste(size_t waste) { _alloc_buffer_waste += waste; } @@ -1635,15 +1639,14 @@ public: } RefToScanQueue* refs() { return _refs; } - OverflowQueue* overflowed_refs() { return _overflowed_refs; } ageTable* age_table() { return &_age_table; } G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose) { return _alloc_buffers[purpose]; } - size_t alloc_buffer_waste() { return _alloc_buffer_waste; } - size_t undo_waste() { return _undo_waste; } + size_t alloc_buffer_waste() const { return _alloc_buffer_waste; } + size_t undo_waste() const { return _undo_waste; } template <class T> void push_on_queue(T* ref) { assert(ref != NULL, "invariant"); @@ -1656,12 +1659,7 @@ public: assert(_g1h->obj_in_cs(p), "Should be in CS"); } #endif - if (!refs()->push(ref)) { - overflowed_refs()->push(ref); - IF_G1_DETAILED_STATS(note_overflow_push()); - } else { - IF_G1_DETAILED_STATS(note_push()); - } + refs()->push(ref); } void pop_from_queue(StarTask& ref) { @@ -1672,7 +1670,6 @@ public: _g1h->is_in_g1_reserved(ref.is_narrow() ? oopDesc::load_decode_heap_oop((narrowOop*)ref) : oopDesc::load_decode_heap_oop((oop*)ref)), "invariant"); - IF_G1_DETAILED_STATS(note_pop()); } else { StarTask null_task; ref = null_task; @@ -1680,7 +1677,8 @@ public: } void pop_from_overflow_queue(StarTask& ref) { - StarTask new_ref = overflowed_refs()->pop(); + StarTask new_ref; + refs()->pop_overflow(new_ref); assert((oop*)new_ref != NULL, "pop() from a local non-empty stack"); assert(UseCompressedOops || !new_ref.is_narrow(), "Error"); assert(has_partial_array_mask((oop*)new_ref) || @@ -1690,8 +1688,8 @@ public: ref = new_ref; } - int refs_to_scan() { return refs()->size(); } - int overflowed_refs_to_scan() { return overflowed_refs()->length(); } + int refs_to_scan() { return refs()->size(); } + int overflowed_refs_to_scan() { return refs()->overflow_stack()->length(); } template <class T> void update_rs(HeapRegion* from, T* p, int tid) { if (G1DeferredRSUpdate) { @@ -1760,30 +1758,16 @@ public: int* hash_seed() { return &_hash_seed; } int queue_num() { return _queue_num; } - size_t term_attempts() { return _term_attempts; } + size_t term_attempts() const { return _term_attempts; } void note_term_attempt() { _term_attempts++; } -#if G1_DETAILED_STATS - int pushes() { return _pushes; } - int pops() { return _pops; } - int steals() { return _steals; } - int steal_attempts() { return _steal_attempts; } - int overflow_pushes() { return _overflow_pushes; } - - void note_push() { _pushes++; } - void note_pop() { _pops++; } - void note_steal() { _steals++; } - void note_steal_attempt() { _steal_attempts++; } - void note_overflow_push() { _overflow_pushes++; } -#endif - void start_strong_roots() { _start_strong_roots = os::elapsedTime(); } void end_strong_roots() { _strong_roots_time += (os::elapsedTime() - _start_strong_roots); } - double strong_roots_time() { return _strong_roots_time; } + double strong_roots_time() const { return _strong_roots_time; } void start_term_time() { note_term_attempt(); @@ -1792,12 +1776,17 @@ public: void end_term_time() { _term_time += (os::elapsedTime() - _start_term); } - double term_time() { return _term_time; } + double term_time() const { return _term_time; } - double elapsed() { + double elapsed_time() const { return os::elapsedTime() - _start; } + static void + print_termination_stats_hdr(outputStream* const st = gclog_or_tty); + void + print_termination_stats(int i, outputStream* const st = gclog_or_tty) const; + size_t* surviving_young_words() { // We add on to hide entry 0 which accumulates surviving words for // age -1 regions (i.e. non-young ones) diff --git a/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp b/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp index a2459d9d0..c4eb03884 100644 --- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp @@ -57,8 +57,9 @@ inline HeapWord* G1CollectedHeap::attempt_allocation(size_t word_size, assert( SafepointSynchronize::is_at_safepoint() || Heap_lock->owned_by_self(), "pre-condition of the call" ); - if (_cur_alloc_region != NULL) { - + // All humongous allocation requests should go through the slow path in + // attempt_allocation_slow(). + if (!isHumongous(word_size) && _cur_alloc_region != NULL) { // If this allocation causes a region to become non empty, // then we need to update our free_regions count. @@ -69,23 +70,23 @@ inline HeapWord* G1CollectedHeap::attempt_allocation(size_t word_size, } else { res = _cur_alloc_region->allocate(word_size); } - } - if (res != NULL) { - if (!SafepointSynchronize::is_at_safepoint()) { - assert( Heap_lock->owned_by_self(), "invariant" ); - Heap_lock->unlock(); + + if (res != NULL) { + if (!SafepointSynchronize::is_at_safepoint()) { + assert( Heap_lock->owned_by_self(), "invariant" ); + Heap_lock->unlock(); + } + return res; } - return res; } // attempt_allocation_slow will also unlock the heap lock when appropriate. return attempt_allocation_slow(word_size, permit_collection_pause); } -inline RefToScanQueue* G1CollectedHeap::task_queue(int i) { +inline RefToScanQueue* G1CollectedHeap::task_queue(int i) const { return _task_queues->queue(i); } - inline bool G1CollectedHeap::isMarkedPrev(oop obj) const { return _cm->prevMarkBitMap()->isMarked((HeapWord *)obj); } diff --git a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp index 0cb73eb0c..ef2128714 100644 --- a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp +++ b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp @@ -88,7 +88,6 @@ G1CollectorPolicy::G1CollectorPolicy() : _all_mod_union_times_ms(new NumberSeq()), _summary(new Summary()), - _abandoned_summary(new AbandonedSummary()), #ifndef PRODUCT _cur_clear_ct_time_ms(0.0), @@ -238,7 +237,6 @@ G1CollectorPolicy::G1CollectorPolicy() : _par_last_update_rs_processed_buffers = new double[_parallel_gc_threads]; _par_last_scan_rs_times_ms = new double[_parallel_gc_threads]; - _par_last_scan_new_refs_times_ms = new double[_parallel_gc_threads]; _par_last_obj_copy_times_ms = new double[_parallel_gc_threads]; @@ -842,7 +840,6 @@ void G1CollectorPolicy::record_collection_pause_start(double start_time_sec, _par_last_update_rs_times_ms[i] = -1234.0; _par_last_update_rs_processed_buffers[i] = -1234.0; _par_last_scan_rs_times_ms[i] = -1234.0; - _par_last_scan_new_refs_times_ms[i] = -1234.0; _par_last_obj_copy_times_ms[i] = -1234.0; _par_last_termination_times_ms[i] = -1234.0; _par_last_termination_attempts[i] = -1234.0; @@ -1126,7 +1123,7 @@ double G1CollectorPolicy::max_sum (double* data1, // Anything below that is considered to be zero #define MIN_TIMER_GRANULARITY 0.0000001 -void G1CollectorPolicy::record_collection_pause_end(bool abandoned) { +void G1CollectorPolicy::record_collection_pause_end() { double end_time_sec = os::elapsedTime(); double elapsed_ms = _last_pause_time_ms; bool parallel = ParallelGCThreads > 0; @@ -1136,7 +1133,7 @@ void G1CollectorPolicy::record_collection_pause_end(bool abandoned) { size_t cur_used_bytes = _g1->used(); assert(cur_used_bytes == _g1->recalculate_used(), "It should!"); bool last_pause_included_initial_mark = false; - bool update_stats = !abandoned && !_g1->evacuation_failed(); + bool update_stats = !_g1->evacuation_failed(); #ifndef PRODUCT if (G1YoungSurvRateVerbose) { @@ -1275,12 +1272,7 @@ void G1CollectorPolicy::record_collection_pause_end(bool abandoned) { gclog_or_tty->print_cr(" Recording collection pause(%d)", _n_pauses); } - PauseSummary* summary; - if (abandoned) { - summary = _abandoned_summary; - } else { - summary = _summary; - } + PauseSummary* summary = _summary; double ext_root_scan_time = avg_value(_par_last_ext_root_scan_times_ms); double mark_stack_scan_time = avg_value(_par_last_mark_stack_scan_times_ms); @@ -1348,61 +1340,58 @@ void G1CollectorPolicy::record_collection_pause_end(bool abandoned) { double other_time_ms = elapsed_ms; - if (!abandoned) { - if (_satb_drain_time_set) - other_time_ms -= _cur_satb_drain_time_ms; + if (_satb_drain_time_set) { + other_time_ms -= _cur_satb_drain_time_ms; + } - if (parallel) - other_time_ms -= _cur_collection_par_time_ms + _cur_clear_ct_time_ms; - else - other_time_ms -= - update_rs_time + - ext_root_scan_time + mark_stack_scan_time + - scan_rs_time + obj_copy_time; + if (parallel) { + other_time_ms -= _cur_collection_par_time_ms + _cur_clear_ct_time_ms; + } else { + other_time_ms -= + update_rs_time + + ext_root_scan_time + mark_stack_scan_time + + scan_rs_time + obj_copy_time; } if (PrintGCDetails) { - gclog_or_tty->print_cr("%s%s, %1.8lf secs]", - abandoned ? " (abandoned)" : "", + gclog_or_tty->print_cr("%s, %1.8lf secs]", (last_pause_included_initial_mark) ? " (initial-mark)" : "", elapsed_ms / 1000.0); - if (!abandoned) { - if (_satb_drain_time_set) { - print_stats(1, "SATB Drain Time", _cur_satb_drain_time_ms); - } - if (_last_satb_drain_processed_buffers >= 0) { - print_stats(2, "Processed Buffers", _last_satb_drain_processed_buffers); - } - if (parallel) { - print_stats(1, "Parallel Time", _cur_collection_par_time_ms); - print_par_stats(2, "GC Worker Start Time", - _par_last_gc_worker_start_times_ms, false); - print_par_stats(2, "Update RS", _par_last_update_rs_times_ms); - print_par_sizes(3, "Processed Buffers", - _par_last_update_rs_processed_buffers, true); - print_par_stats(2, "Ext Root Scanning", - _par_last_ext_root_scan_times_ms); - print_par_stats(2, "Mark Stack Scanning", - _par_last_mark_stack_scan_times_ms); - print_par_stats(2, "Scan RS", _par_last_scan_rs_times_ms); - print_par_stats(2, "Object Copy", _par_last_obj_copy_times_ms); - print_par_stats(2, "Termination", _par_last_termination_times_ms); - print_par_sizes(3, "Termination Attempts", - _par_last_termination_attempts, true); - print_par_stats(2, "GC Worker End Time", - _par_last_gc_worker_end_times_ms, false); - print_stats(2, "Other", parallel_other_time); - print_stats(1, "Clear CT", _cur_clear_ct_time_ms); - } else { - print_stats(1, "Update RS", update_rs_time); - print_stats(2, "Processed Buffers", - (int)update_rs_processed_buffers); - print_stats(1, "Ext Root Scanning", ext_root_scan_time); - print_stats(1, "Mark Stack Scanning", mark_stack_scan_time); - print_stats(1, "Scan RS", scan_rs_time); - print_stats(1, "Object Copying", obj_copy_time); - } + if (_satb_drain_time_set) { + print_stats(1, "SATB Drain Time", _cur_satb_drain_time_ms); + } + if (_last_satb_drain_processed_buffers >= 0) { + print_stats(2, "Processed Buffers", _last_satb_drain_processed_buffers); + } + if (parallel) { + print_stats(1, "Parallel Time", _cur_collection_par_time_ms); + print_par_stats(2, "GC Worker Start Time", + _par_last_gc_worker_start_times_ms, false); + print_par_stats(2, "Update RS", _par_last_update_rs_times_ms); + print_par_sizes(3, "Processed Buffers", + _par_last_update_rs_processed_buffers, true); + print_par_stats(2, "Ext Root Scanning", + _par_last_ext_root_scan_times_ms); + print_par_stats(2, "Mark Stack Scanning", + _par_last_mark_stack_scan_times_ms); + print_par_stats(2, "Scan RS", _par_last_scan_rs_times_ms); + print_par_stats(2, "Object Copy", _par_last_obj_copy_times_ms); + print_par_stats(2, "Termination", _par_last_termination_times_ms); + print_par_sizes(3, "Termination Attempts", + _par_last_termination_attempts, true); + print_par_stats(2, "GC Worker End Time", + _par_last_gc_worker_end_times_ms, false); + print_stats(2, "Other", parallel_other_time); + print_stats(1, "Clear CT", _cur_clear_ct_time_ms); + } else { + print_stats(1, "Update RS", update_rs_time); + print_stats(2, "Processed Buffers", + (int)update_rs_processed_buffers); + print_stats(1, "Ext Root Scanning", ext_root_scan_time); + print_stats(1, "Mark Stack Scanning", mark_stack_scan_time); + print_stats(1, "Scan RS", scan_rs_time); + print_stats(1, "Object Copying", obj_copy_time); } #ifndef PRODUCT print_stats(1, "Cur Clear CC", _cur_clear_cc_time_ms); @@ -2178,33 +2167,27 @@ void G1CollectorPolicy::print_summary(PauseSummary* summary) const { print_summary(1, "Other", summary->get_other_seq()); { NumberSeq calc_other_times_ms; - if (body_summary != NULL) { - // not abandoned - if (parallel) { - // parallel - NumberSeq* other_parts[] = { - body_summary->get_satb_drain_seq(), - body_summary->get_parallel_seq(), - body_summary->get_clear_ct_seq() - }; - calc_other_times_ms = NumberSeq(summary->get_total_seq(), - 3, other_parts); - } else { - // serial - NumberSeq* other_parts[] = { - body_summary->get_satb_drain_seq(), - body_summary->get_update_rs_seq(), - body_summary->get_ext_root_scan_seq(), - body_summary->get_mark_stack_scan_seq(), - body_summary->get_scan_rs_seq(), - body_summary->get_obj_copy_seq() - }; - calc_other_times_ms = NumberSeq(summary->get_total_seq(), - 7, other_parts); - } + if (parallel) { + // parallel + NumberSeq* other_parts[] = { + body_summary->get_satb_drain_seq(), + body_summary->get_parallel_seq(), + body_summary->get_clear_ct_seq() + }; + calc_other_times_ms = NumberSeq(summary->get_total_seq(), + 3, other_parts); } else { - // abandoned - calc_other_times_ms = NumberSeq(); + // serial + NumberSeq* other_parts[] = { + body_summary->get_satb_drain_seq(), + body_summary->get_update_rs_seq(), + body_summary->get_ext_root_scan_seq(), + body_summary->get_mark_stack_scan_seq(), + body_summary->get_scan_rs_seq(), + body_summary->get_obj_copy_seq() + }; + calc_other_times_ms = NumberSeq(summary->get_total_seq(), + 7, other_parts); } check_other_times(1, summary->get_other_seq(), &calc_other_times_ms); } @@ -2215,20 +2198,6 @@ void G1CollectorPolicy::print_summary(PauseSummary* summary) const { gclog_or_tty->print_cr(""); } -void -G1CollectorPolicy::print_abandoned_summary(PauseSummary* summary) const { - bool printed = false; - if (summary->get_total_seq()->num() > 0) { - printed = true; - print_summary(summary); - } - if (!printed) { - print_indent(0); - gclog_or_tty->print_cr("none"); - gclog_or_tty->print_cr(""); - } -} - void G1CollectorPolicy::print_tracing_info() const { if (TraceGen0Time) { gclog_or_tty->print_cr("ALL PAUSES"); @@ -2242,9 +2211,6 @@ void G1CollectorPolicy::print_tracing_info() const { gclog_or_tty->print_cr("EVACUATION PAUSES"); print_summary(_summary); - gclog_or_tty->print_cr("ABANDONED PAUSES"); - print_abandoned_summary(_abandoned_summary); - gclog_or_tty->print_cr("MISC"); print_summary_sd(0, "Stop World", _all_stop_world_times_ms); print_summary_sd(0, "Yields", _all_yield_times_ms); @@ -2870,19 +2836,12 @@ void G1CollectorPolicy::print_collection_set(HeapRegion* list_head, outputStream } #endif // !PRODUCT -bool +void G1CollectorPolicy_BestRegionsFirst::choose_collection_set( double target_pause_time_ms) { // Set this here - in case we're not doing young collections. double non_young_start_time_sec = os::elapsedTime(); - // The result that this routine will return. This will be set to - // false if: - // * we're doing a young or partially young collection and we - // have added the youg regions to collection set, or - // * we add old regions to the collection set. - bool abandon_collection = true; - start_recording_regions(); guarantee(target_pause_time_ms > 0.0, @@ -2986,10 +2945,6 @@ G1CollectorPolicy_BestRegionsFirst::choose_collection_set( } assert(_inc_cset_size == _g1->young_list()->length(), "Invariant"); - if (_inc_cset_size > 0) { - assert(_collection_set != NULL, "Invariant"); - abandon_collection = false; - } double young_end_time_sec = os::elapsedTime(); _recorded_young_cset_choice_time_ms = @@ -3011,10 +2966,6 @@ G1CollectorPolicy_BestRegionsFirst::choose_collection_set( NumberSeq seq; double avg_prediction = 100000000000000000.0; // something very large - // Save the current size of the collection set to detect - // if we actually added any old regions. - size_t n_young_regions = _collection_set_size; - do { hr = _collectionSetChooser->getNextMarkedRegion(time_remaining_ms, avg_prediction); @@ -3041,12 +2992,6 @@ G1CollectorPolicy_BestRegionsFirst::choose_collection_set( if (!adaptive_young_list_length() && _collection_set_size < _young_list_fixed_length) _should_revert_to_full_young_gcs = true; - - if (_collection_set_size > n_young_regions) { - // We actually added old regions to the collection set - // so we are not abandoning this collection. - abandon_collection = false; - } } choose_collection_set_end: @@ -3059,19 +3004,6 @@ choose_collection_set_end: double non_young_end_time_sec = os::elapsedTime(); _recorded_non_young_cset_choice_time_ms = (non_young_end_time_sec - non_young_start_time_sec) * 1000.0; - - // Here we are supposed to return whether the pause should be - // abandoned or not (i.e., whether the collection set is empty or - // not). However, this introduces a subtle issue when a pause is - // initiated explicitly with System.gc() and - // +ExplicitGCInvokesConcurrent (see Comment #2 in CR 6944166), it's - // supposed to start a marking cycle, and it's abandoned. So, by - // returning false here we are telling the caller never to consider - // a pause to be abandoned. We'll actually remove all the code - // associated with abandoned pauses as part of CR 6963209, but we are - // just disabling them this way for the moment to avoid increasing - // further the amount of changes for CR 6944166. - return false; } void G1CollectorPolicy_BestRegionsFirst::record_full_collection_end() { @@ -3086,7 +3018,7 @@ expand_if_possible(size_t numRegions) { } void G1CollectorPolicy_BestRegionsFirst:: -record_collection_pause_end(bool abandoned) { - G1CollectorPolicy::record_collection_pause_end(abandoned); +record_collection_pause_end() { + G1CollectorPolicy::record_collection_pause_end(); assert(assertMarkedBytesDataOK(), "Marked regions not OK at pause end."); } diff --git a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp index 61dbee6c0..33ee6ebc4 100644 --- a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp +++ b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp @@ -63,8 +63,6 @@ class MainBodySummary: public CHeapObj { define_num_seq(mark_stack_scan) define_num_seq(update_rs) define_num_seq(scan_rs) - define_num_seq(scan_new_refs) // Only for temp use; added to - // in parallel case. define_num_seq(obj_copy) define_num_seq(termination) // parallel only define_num_seq(parallel_other) // parallel only @@ -78,9 +76,6 @@ public: virtual MainBodySummary* main_body_summary() { return this; } }; -class AbandonedSummary: public PauseSummary { -}; - class G1CollectorPolicy: public CollectorPolicy { protected: // The number of pauses during the execution. @@ -150,7 +145,6 @@ protected: TruncatedSeq* _concurrent_mark_cleanup_times_ms; Summary* _summary; - AbandonedSummary* _abandoned_summary; NumberSeq* _all_pause_times_ms; NumberSeq* _all_full_gc_times_ms; @@ -177,7 +171,6 @@ protected: double* _par_last_update_rs_times_ms; double* _par_last_update_rs_processed_buffers; double* _par_last_scan_rs_times_ms; - double* _par_last_scan_new_refs_times_ms; double* _par_last_obj_copy_times_ms; double* _par_last_termination_times_ms; double* _par_last_termination_attempts; @@ -576,7 +569,6 @@ protected: NumberSeq* calc_other_times_ms) const; void print_summary (PauseSummary* stats) const; - void print_abandoned_summary(PauseSummary* summary) const; void print_summary (int level, const char* str, NumberSeq* seq) const; void print_summary_sd (int level, const char* str, NumberSeq* seq) const; @@ -889,7 +881,7 @@ public: virtual void record_collection_pause_end_CH_strong_roots(); virtual void record_collection_pause_end_G1_strong_roots(); - virtual void record_collection_pause_end(bool abandoned); + virtual void record_collection_pause_end(); // Record the fact that a full collection occurred. virtual void record_full_collection_start(); @@ -933,14 +925,6 @@ public: _par_last_scan_rs_times_ms[thread] = ms; } - void record_scan_new_refs_time(int thread, double ms) { - _par_last_scan_new_refs_times_ms[thread] = ms; - } - - double get_scan_new_refs_time(int thread) { - return _par_last_scan_new_refs_times_ms[thread]; - } - void reset_obj_copy_time(int thread) { _par_last_obj_copy_times_ms[thread] = 0.0; } @@ -1010,7 +994,7 @@ public: // Choose a new collection set. Marks the chosen regions as being // "in_collection_set", and links them together. The head and number of // the collection set are available via access methods. - virtual bool choose_collection_set(double target_pause_time_ms) = 0; + virtual void choose_collection_set(double target_pause_time_ms) = 0; // The head of the list (via "next_in_collection_set()") representing the // current collection set. @@ -1267,7 +1251,7 @@ class G1CollectorPolicy_BestRegionsFirst: public G1CollectorPolicy { // If the estimated is less then desirable, resize if possible. void expand_if_possible(size_t numRegions); - virtual bool choose_collection_set(double target_pause_time_ms); + virtual void choose_collection_set(double target_pause_time_ms); virtual void record_collection_pause_start(double start_time_sec, size_t start_used); virtual void record_concurrent_mark_cleanup_end(size_t freed_bytes, @@ -1278,7 +1262,7 @@ public: G1CollectorPolicy_BestRegionsFirst() { _collectionSetChooser = new CollectionSetChooser(); } - void record_collection_pause_end(bool abandoned); + void record_collection_pause_end(); bool should_do_collection_pause(size_t word_size); // This is not needed any more, after the CSet choosing code was // changed to use the pause prediction work. But let's leave the diff --git a/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp b/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp index 1037b83bd..28ec22b04 100644 --- a/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp +++ b/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2001, 2007, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -37,7 +37,8 @@ template <class T> inline void FilterIntoCSClosure::do_oop_nv(T* p) { _g1->obj_in_cs(oopDesc::decode_heap_oop_not_null(heap_oop))) { _oc->do_oop(p); #if FILTERINTOCSCLOSURE_DOHISTOGRAMCOUNT - _dcto_cl->incr_count(); + if (_dcto_cl != NULL) + _dcto_cl->incr_count(); #endif } } @@ -113,7 +114,10 @@ template <class T> inline void G1ParPushHeapRSClosure::do_oop_nv(T* p) { if (_g1->in_cset_fast_test(obj)) { Prefetch::write(obj->mark_addr(), 0); Prefetch::read(obj->mark_addr(), (HeapWordSize*2)); + + // Place on the references queue _par_scan_state->push_on_queue(p); } } } + diff --git a/src/share/vm/gc_implementation/g1/g1RemSet.cpp b/src/share/vm/gc_implementation/g1/g1RemSet.cpp index a3a77ed18..19ec341f9 100644 --- a/src/share/vm/gc_implementation/g1/g1RemSet.cpp +++ b/src/share/vm/gc_implementation/g1/g1RemSet.cpp @@ -122,23 +122,24 @@ public: HRInto_G1RemSet::HRInto_G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs) : G1RemSet(g1), _ct_bs(ct_bs), _g1p(_g1->g1_policy()), _cg1r(g1->concurrent_g1_refine()), - _par_traversal_in_progress(false), _new_refs(NULL), + _traversal_in_progress(false), + _cset_rs_update_cl(NULL), _cards_scanned(NULL), _total_cards_scanned(0) { _seq_task = new SubTasksDone(NumSeqTasks); guarantee(n_workers() > 0, "There should be some workers"); - _new_refs = NEW_C_HEAP_ARRAY(GrowableArray<OopOrNarrowOopStar>*, n_workers()); + _cset_rs_update_cl = NEW_C_HEAP_ARRAY(OopsInHeapRegionClosure*, n_workers()); for (uint i = 0; i < n_workers(); i++) { - _new_refs[i] = new (ResourceObj::C_HEAP) GrowableArray<OopOrNarrowOopStar>(8192,true); + _cset_rs_update_cl[i] = NULL; } } HRInto_G1RemSet::~HRInto_G1RemSet() { delete _seq_task; for (uint i = 0; i < n_workers(); i++) { - delete _new_refs[i]; + assert(_cset_rs_update_cl[i] == NULL, "it should be"); } - FREE_C_HEAP_ARRAY(GrowableArray<OopOrNarrowOopStar>*, _new_refs); + FREE_C_HEAP_ARRAY(OopsInHeapRegionClosure*, _cset_rs_update_cl); } void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) { @@ -306,12 +307,45 @@ void HRInto_G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) { _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0); } -void HRInto_G1RemSet::updateRS(int worker_i) { - ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine(); +// Closure used for updating RSets and recording references that +// point into the collection set. Only called during an +// evacuation pause. +class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure { + G1RemSet* _g1rs; + DirtyCardQueue* _into_cset_dcq; +public: + RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h, + DirtyCardQueue* into_cset_dcq) : + _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq) + {} + bool do_card_ptr(jbyte* card_ptr, int worker_i) { + // The only time we care about recording cards that + // contain references that point into the collection set + // is during RSet updating within an evacuation pause. + // In this case worker_i should be the id of a GC worker thread. + assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause"); + assert(worker_i < (int) DirtyCardQueueSet::num_par_ids(), "should be a GC worker"); + + if (_g1rs->concurrentRefineOneCard(card_ptr, worker_i, true)) { + // 'card_ptr' contains references that point into the collection + // set. We need to record the card in the DCQS + // (G1CollectedHeap::into_cset_dirty_card_queue_set()) + // that's used for that purpose. + // + // Enqueue the card + _into_cset_dcq->enqueue(card_ptr); + } + return true; + } +}; + +void HRInto_G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, int worker_i) { double start = os::elapsedTime(); - // Apply the appropriate closure to all remaining log entries. - _g1->iterate_dirty_card_closure(false, worker_i); + // Apply the given closure to all remaining log entries. + RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq); + _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, into_cset_dcq, false, worker_i); + // Now there should be no dirty cards. if (G1RSLogCheckCardTable) { CountNonCleanMemRegionClosure cl(_g1); @@ -405,33 +439,6 @@ public: } }; -template <class T> void -HRInto_G1RemSet::scanNewRefsRS_work(OopsInHeapRegionClosure* oc, - int worker_i) { - double scan_new_refs_start_sec = os::elapsedTime(); - G1CollectedHeap* g1h = G1CollectedHeap::heap(); - CardTableModRefBS* ct_bs = (CardTableModRefBS*) (g1h->barrier_set()); - for (int i = 0; i < _new_refs[worker_i]->length(); i++) { - T* p = (T*) _new_refs[worker_i]->at(i); - oop obj = oopDesc::load_decode_heap_oop(p); - // *p was in the collection set when p was pushed on "_new_refs", but - // another thread may have processed this location from an RS, so it - // might not point into the CS any longer. If so, it's obviously been - // processed, and we don't need to do anything further. - if (g1h->obj_in_cs(obj)) { - HeapRegion* r = g1h->heap_region_containing(p); - - DEBUG_ONLY(HeapRegion* to = g1h->heap_region_containing(obj)); - oc->set_region(r); - // If "p" has already been processed concurrently, this is - // idempotent. - oc->do_oop(p); - } - } - double scan_new_refs_time_ms = (os::elapsedTime() - scan_new_refs_start_sec) * 1000.0; - _g1p->record_scan_new_refs_time(worker_i, scan_new_refs_time_ms); -} - void HRInto_G1RemSet::cleanupHRRS() { HeapRegionRemSet::cleanup(); } @@ -457,32 +464,48 @@ HRInto_G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc, count_cl.print_histo(); } - if (ParallelGCThreads > 0) { - // The two flags below were introduced temporarily to serialize - // the updating and scanning of remembered sets. There are some - // race conditions when these two operations are done in parallel - // and they are causing failures. When we resolve said race - // conditions, we'll revert back to parallel remembered set - // updating and scanning. See CRs 6677707 and 6677708. - if (G1UseParallelRSetUpdating || (worker_i == 0)) { - updateRS(worker_i); - scanNewRefsRS(oc, worker_i); - } else { - _g1p->record_update_rs_processed_buffers(worker_i, 0.0); - _g1p->record_update_rs_time(worker_i, 0.0); - _g1p->record_scan_new_refs_time(worker_i, 0.0); - } - if (G1UseParallelRSetScanning || (worker_i == 0)) { - scanRS(oc, worker_i); - } else { - _g1p->record_scan_rs_time(worker_i, 0.0); - } + // We cache the value of 'oc' closure into the appropriate slot in the + // _cset_rs_update_cl for this worker + assert(worker_i < (int)n_workers(), "sanity"); + _cset_rs_update_cl[worker_i] = oc; + + // A DirtyCardQueue that is used to hold cards containing references + // that point into the collection set. This DCQ is associated with a + // special DirtyCardQueueSet (see g1CollectedHeap.hpp). Under normal + // circumstances (i.e. the pause successfully completes), these cards + // are just discarded (there's no need to update the RSets of regions + // that were in the collection set - after the pause these regions + // are wholly 'free' of live objects. In the event of an evacuation + // failure the cards/buffers in this queue set are: + // * passed to the DirtyCardQueueSet that is used to manage deferred + // RSet updates, or + // * scanned for references that point into the collection set + // and the RSet of the corresponding region in the collection set + // is updated immediately. + DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set()); + + assert((ParallelGCThreads > 0) || worker_i == 0, "invariant"); + + // The two flags below were introduced temporarily to serialize + // the updating and scanning of remembered sets. There are some + // race conditions when these two operations are done in parallel + // and they are causing failures. When we resolve said race + // conditions, we'll revert back to parallel remembered set + // updating and scanning. See CRs 6677707 and 6677708. + if (G1UseParallelRSetUpdating || (worker_i == 0)) { + updateRS(&into_cset_dcq, worker_i); } else { - assert(worker_i == 0, "invariant"); - updateRS(0); - scanNewRefsRS(oc, 0); - scanRS(oc, 0); + _g1p->record_update_rs_processed_buffers(worker_i, 0.0); + _g1p->record_update_rs_time(worker_i, 0.0); } + if (G1UseParallelRSetScanning || (worker_i == 0)) { + scanRS(oc, worker_i); + } else { + _g1p->record_scan_rs_time(worker_i, 0.0); + } + + // We now clear the cached values of _cset_rs_update_cl for this worker + _cset_rs_update_cl[worker_i] = NULL; } void HRInto_G1RemSet:: @@ -497,9 +520,9 @@ prepare_for_oops_into_collection_set_do() { DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); dcqs.concatenate_logs(); - assert(!_par_traversal_in_progress, "Invariant between iterations."); + assert(!_traversal_in_progress, "Invariant between iterations."); + set_traversal(true); if (ParallelGCThreads > 0) { - set_par_traversal(true); _seq_task->set_par_threads((int)n_workers()); } guarantee( _cards_scanned == NULL, "invariant" ); @@ -519,49 +542,65 @@ class cleanUpIteratorsClosure : public HeapRegionClosure { } }; -class UpdateRSetOopsIntoCSImmediate : public OopClosure { - G1CollectedHeap* _g1; -public: - UpdateRSetOopsIntoCSImmediate(G1CollectedHeap* g1) : _g1(g1) { } - virtual void do_oop(narrowOop* p) { do_oop_work(p); } - virtual void do_oop( oop* p) { do_oop_work(p); } - template <class T> void do_oop_work(T* p) { - HeapRegion* to = _g1->heap_region_containing(oopDesc::load_decode_heap_oop(p)); - if (to->in_collection_set()) { - to->rem_set()->add_reference(p, 0); - } - } -}; - -class UpdateRSetOopsIntoCSDeferred : public OopClosure { +// This closure, applied to a DirtyCardQueueSet, is used to immediately +// update the RSets for the regions in the CSet. For each card it iterates +// through the oops which coincide with that card. It scans the reference +// fields in each oop; when it finds an oop that points into the collection +// set, the RSet for the region containing the referenced object is updated. +// Note: _par_traversal_in_progress in the G1RemSet must be FALSE; otherwise +// the UpdateRSetImmediate closure will cause cards to be enqueued on to +// the DCQS that we're iterating over, causing an infinite loop. +class UpdateRSetCardTableEntryIntoCSetClosure: public CardTableEntryClosure { G1CollectedHeap* _g1; CardTableModRefBS* _ct_bs; - DirtyCardQueue* _dcq; public: - UpdateRSetOopsIntoCSDeferred(G1CollectedHeap* g1, DirtyCardQueue* dcq) : - _g1(g1), _ct_bs((CardTableModRefBS*)_g1->barrier_set()), _dcq(dcq) { } - virtual void do_oop(narrowOop* p) { do_oop_work(p); } - virtual void do_oop( oop* p) { do_oop_work(p); } - template <class T> void do_oop_work(T* p) { - oop obj = oopDesc::load_decode_heap_oop(p); - if (_g1->obj_in_cs(obj)) { - size_t card_index = _ct_bs->index_for(p); - if (_ct_bs->mark_card_deferred(card_index)) { - _dcq->enqueue((jbyte*)_ct_bs->byte_for_index(card_index)); - } - } + UpdateRSetCardTableEntryIntoCSetClosure(G1CollectedHeap* g1, + CardTableModRefBS* bs): + _g1(g1), _ct_bs(bs) + { } + + bool do_card_ptr(jbyte* card_ptr, int worker_i) { + // Construct the region representing the card. + HeapWord* start = _ct_bs->addr_for(card_ptr); + // And find the region containing it. + HeapRegion* r = _g1->heap_region_containing(start); + assert(r != NULL, "unexpected null"); + + // Scan oops in the card looking for references into the collection set + HeapWord* end = _ct_bs->addr_for(card_ptr + 1); + MemRegion scanRegion(start, end); + + UpdateRSetImmediate update_rs_cl(_g1->g1_rem_set()); + FilterIntoCSClosure update_rs_cset_oop_cl(NULL, _g1, &update_rs_cl); + FilterOutOfRegionClosure filter_then_update_rs_cset_oop_cl(r, &update_rs_cset_oop_cl); + + // We can pass false as the "filter_young" parameter here as: + // * we should be in a STW pause, + // * the DCQS to which this closure is applied is used to hold + // references that point into the collection set from the prior + // RSet updating, + // * the post-write barrier shouldn't be logging updates to young + // regions (but there is a situation where this can happen - see + // the comment in HRInto_G1RemSet::concurrentRefineOneCard below - + // that should not be applicable here), and + // * during actual RSet updating, the filtering of cards in young + // regions in HeapRegion::oops_on_card_seq_iterate_careful is + // employed. + // As a result, when this closure is applied to "refs into cset" + // DCQS, we shouldn't see any cards in young regions. + update_rs_cl.set_region(r); + HeapWord* stop_point = + r->oops_on_card_seq_iterate_careful(scanRegion, + &filter_then_update_rs_cset_oop_cl, + false /* filter_young */); + + // Since this is performed in the event of an evacuation failure, we + // we shouldn't see a non-null stop point + assert(stop_point == NULL, "saw an unallocated region"); + return true; } }; -template <class T> void HRInto_G1RemSet::new_refs_iterate_work(OopClosure* cl) { - for (size_t i = 0; i < n_workers(); i++) { - for (int j = 0; j < _new_refs[i]->length(); j++) { - T* p = (T*) _new_refs[i]->at(j); - cl->do_oop(p); - } - } -} - void HRInto_G1RemSet::cleanup_after_oops_into_collection_set_do() { guarantee( _cards_scanned != NULL, "invariant" ); _total_cards_scanned = 0; @@ -580,27 +619,42 @@ void HRInto_G1RemSet::cleanup_after_oops_into_collection_set_do() { // Set all cards back to clean. _g1->cleanUpCardTable(); - if (ParallelGCThreads > 0) { - set_par_traversal(false); - } + set_traversal(false); + + DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set(); + int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num(); if (_g1->evacuation_failed()) { - // Restore remembered sets for the regions pointing into - // the collection set. + // Restore remembered sets for the regions pointing into the collection set. + if (G1DeferredRSUpdate) { - DirtyCardQueue dcq(&_g1->dirty_card_queue_set()); - UpdateRSetOopsIntoCSDeferred deferred_update(_g1, &dcq); - new_refs_iterate(&deferred_update); + // If deferred RS updates are enabled then we just need to transfer + // the completed buffers from (a) the DirtyCardQueueSet used to hold + // cards that contain references that point into the collection set + // to (b) the DCQS used to hold the deferred RS updates + _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs); } else { - UpdateRSetOopsIntoCSImmediate immediate_update(_g1); - new_refs_iterate(&immediate_update); + + CardTableModRefBS* bs = (CardTableModRefBS*)_g1->barrier_set(); + UpdateRSetCardTableEntryIntoCSetClosure update_rs_cset_immediate(_g1, bs); + + int n_completed_buffers = 0; + while (into_cset_dcqs.apply_closure_to_completed_buffer(&update_rs_cset_immediate, + 0, 0, true)) { + n_completed_buffers++; + } + assert(n_completed_buffers == into_cset_n_buffers, "missed some buffers"); } } - for (uint i = 0; i < n_workers(); i++) { - _new_refs[i]->clear(); - } - assert(!_par_traversal_in_progress, "Invariant between iterations."); + // Free any completed buffers in the DirtyCardQueueSet used to hold cards + // which contain references that point into the collection. + _g1->into_cset_dirty_card_queue_set().clear(); + assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0, + "all buffers should be freed"); + _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers(); + + assert(!_traversal_in_progress, "Invariant between iterations."); } class UpdateRSObjectClosure: public ObjectClosure { @@ -652,7 +706,43 @@ void HRInto_G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm, static IntHistogram out_of_histo(50, 50); -void HRInto_G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i) { +class TriggerClosure : public OopClosure { + bool _trigger; +public: + TriggerClosure() : _trigger(false) { } + bool value() const { return _trigger; } + template <class T> void do_oop_nv(T* p) { _trigger = true; } + virtual void do_oop(oop* p) { do_oop_nv(p); } + virtual void do_oop(narrowOop* p) { do_oop_nv(p); } +}; + +class InvokeIfNotTriggeredClosure: public OopClosure { + TriggerClosure* _t; + OopClosure* _oc; +public: + InvokeIfNotTriggeredClosure(TriggerClosure* t, OopClosure* oc): + _t(t), _oc(oc) { } + template <class T> void do_oop_nv(T* p) { + if (!_t->value()) _oc->do_oop(p); + } + virtual void do_oop(oop* p) { do_oop_nv(p); } + virtual void do_oop(narrowOop* p) { do_oop_nv(p); } +}; + +class Mux2Closure : public OopClosure { + OopClosure* _c1; + OopClosure* _c2; +public: + Mux2Closure(OopClosure *c1, OopClosure *c2) : _c1(c1), _c2(c2) { } + template <class T> void do_oop_nv(T* p) { + _c1->do_oop(p); _c2->do_oop(p); + } + virtual void do_oop(oop* p) { do_oop_nv(p); } + virtual void do_oop(narrowOop* p) { do_oop_nv(p); } +}; + +bool HRInto_G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i, + bool check_for_refs_into_cset) { // Construct the region representing the card. HeapWord* start = _ct_bs->addr_for(card_ptr); // And find the region containing it. @@ -669,7 +759,16 @@ void HRInto_G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i UpdateRSOopClosure update_rs_oop_cl(this, worker_i); update_rs_oop_cl.set_from(r); - FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r, &update_rs_oop_cl); + + TriggerClosure trigger_cl; + FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl); + InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl); + Mux2Closure mux(&invoke_cl, &update_rs_oop_cl); + + FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r, + (check_for_refs_into_cset ? + (OopClosure*)&mux : + (OopClosure*)&update_rs_oop_cl)); // Undirty the card. *card_ptr = CardTableModRefBS::clean_card_val(); @@ -717,11 +816,18 @@ void HRInto_G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region()); _conc_refine_cards++; } + + return trigger_cl.value(); } -void HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i) { +bool HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i, + bool check_for_refs_into_cset) { // If the card is no longer dirty, nothing to do. - if (*card_ptr != CardTableModRefBS::dirty_card_val()) return; + if (*card_ptr != CardTableModRefBS::dirty_card_val()) { + // No need to return that this card contains refs that point + // into the collection set. + return false; + } // Construct the region representing the card. HeapWord* start = _ct_bs->addr_for(card_ptr); @@ -729,7 +835,9 @@ void HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i) { HeapRegion* r = _g1->heap_region_containing(start); if (r == NULL) { guarantee(_g1->is_in_permanent(start), "Or else where?"); - return; // Not in the G1 heap (might be in perm, for example.) + // Again no need to return that this card contains refs that + // point into the collection set. + return false; // Not in the G1 heap (might be in perm, for example.) } // Why do we have to check here whether a card is on a young region, // given that we dirty young regions and, as a result, the @@ -743,7 +851,7 @@ void HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i) { // and it doesn't happen often, but it can happen. So, the extra // check below filters out those cards. if (r->is_young()) { - return; + return false; } // While we are processing RSet buffers during the collection, we // actually don't want to scan any cards on the collection set, @@ -756,7 +864,7 @@ void HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i) { // however, that if evacuation fails, we have to scan any objects // that were not moved and create any missing entries. if (r->in_collection_set()) { - return; + return false; } // Should we defer processing the card? @@ -797,8 +905,14 @@ void HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i) { // cache. // Immediately process res; no need to process card_ptr. + jbyte* res = card_ptr; bool defer = false; + + // This gets set to true if the card being refined has references + // that point into the collection set. + bool oops_into_cset = false; + if (_cg1r->use_cache()) { jbyte* res = _cg1r->cache_insert(card_ptr, &defer); if (res != NULL && (res != card_ptr || defer)) { @@ -815,14 +929,31 @@ void HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i) { // Process card pointer we get back from the hot card cache. This // will check whether the region containing the card is young // _after_ checking that the region has been allocated from. - concurrentRefineOneCard_impl(res, worker_i); + oops_into_cset = concurrentRefineOneCard_impl(res, worker_i, + false /* check_for_refs_into_cset */); + // The above call to concurrentRefineOneCard_impl is only + // performed if the hot card cache is enabled. This cache is + // disabled during an evacuation pause - which is the only + // time when we need know if the card contains references + // that point into the collection set. Also when the hot card + // cache is enabled, this code is executed by the concurrent + // refine threads - rather than the GC worker threads - and + // concurrentRefineOneCard_impl will return false. + assert(!oops_into_cset, "should not see true here"); } } } if (!defer) { - concurrentRefineOneCard_impl(card_ptr, worker_i); + oops_into_cset = + concurrentRefineOneCard_impl(card_ptr, worker_i, check_for_refs_into_cset); + // We should only be detecting that the card contains references + // that point into the collection set if the current thread is + // a GC worker thread. + assert(!oops_into_cset || SafepointSynchronize::is_at_safepoint(), + "invalid result at non safepoint"); } + return oops_into_cset; } class HRRSStatsIter: public HeapRegionClosure { @@ -920,6 +1051,7 @@ void HRInto_G1RemSet::print_summary_info() { } } + void HRInto_G1RemSet::prepare_for_verify() { if (G1HRRSFlushLogBuffersOnVerify && (VerifyBeforeGC || VerifyAfterGC) @@ -932,7 +1064,9 @@ void HRInto_G1RemSet::prepare_for_verify() { } bool cg1r_use_cache = _cg1r->use_cache(); _cg1r->set_use_cache(false); - updateRS(0); + DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set()); + updateRS(&into_cset_dcq, 0); + _g1->into_cset_dirty_card_queue_set().clear(); _cg1r->set_use_cache(cg1r_use_cache); assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed"); diff --git a/src/share/vm/gc_implementation/g1/g1RemSet.hpp b/src/share/vm/gc_implementation/g1/g1RemSet.hpp index bc30a9699..fcb5ecd76 100644 --- a/src/share/vm/gc_implementation/g1/g1RemSet.hpp +++ b/src/share/vm/gc_implementation/g1/g1RemSet.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2001, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -83,7 +83,13 @@ public: // Refine the card corresponding to "card_ptr". If "sts" is non-NULL, // join and leave around parts that must be atomic wrt GC. (NULL means // being done at a safepoint.) - virtual void concurrentRefineOneCard(jbyte* card_ptr, int worker_i) {} + // With some implementations of this routine, when check_for_refs_into_cset + // is true, a true result may be returned if the given card contains oops + // that have references into the current collection set. + virtual bool concurrentRefineOneCard(jbyte* card_ptr, int worker_i, + bool check_for_refs_into_cset) { + return false; + } // Print any relevant summary info. virtual void print_summary_info() {} @@ -142,24 +148,22 @@ protected: size_t* _cards_scanned; size_t _total_cards_scanned; - // _par_traversal_in_progress is "true" iff a parallel traversal is in - // progress. If so, then cards added to remembered sets should also have - // their references into the collection summarized in "_new_refs". - bool _par_traversal_in_progress; - void set_par_traversal(bool b) { _par_traversal_in_progress = b; } - GrowableArray<OopOrNarrowOopStar>** _new_refs; - template <class T> void new_refs_iterate_work(OopClosure* cl); - void new_refs_iterate(OopClosure* cl) { - if (UseCompressedOops) { - new_refs_iterate_work<narrowOop>(cl); - } else { - new_refs_iterate_work<oop>(cl); - } - } + // _traversal_in_progress is "true" iff a traversal is in progress. + + bool _traversal_in_progress; + void set_traversal(bool b) { _traversal_in_progress = b; } + + // Used for caching the closure that is responsible for scanning + // references into the collection set. + OopsInHeapRegionClosure** _cset_rs_update_cl; // The routine that performs the actual work of refining a dirty // card. - void concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i); + // If check_for_refs_into_refs is true then a true result is returned + // if the card contains oops that have references into the current + // collection set. + bool concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i, + bool check_for_refs_into_cset); protected: template <class T> void write_ref_nv(HeapRegion* from, T* p); @@ -188,7 +192,7 @@ public: scanNewRefsRS_work<oop>(oc, worker_i); } } - void updateRS(int worker_i); + void updateRS(DirtyCardQueue* into_cset_dcq, int worker_i); HeapRegion* calculateStartRegion(int i); HRInto_G1RemSet* as_HRInto_G1RemSet() { return this; } @@ -219,7 +223,11 @@ public: void scrub_par(BitMap* region_bm, BitMap* card_bm, int worker_num, int claim_val); - virtual void concurrentRefineOneCard(jbyte* card_ptr, int worker_i); + // If check_for_refs_into_cset is true then a true result is returned + // if the card contains oops that have references into the current + // collection set. + virtual bool concurrentRefineOneCard(jbyte* card_ptr, int worker_i, + bool check_for_refs_into_cset); virtual void print_summary_info(); virtual void prepare_for_verify(); @@ -265,3 +273,16 @@ public: // bool idempotent() { return true; } bool apply_to_weak_ref_discovered_field() { return true; } }; + +class UpdateRSetImmediate: public OopsInHeapRegionClosure { +private: + G1RemSet* _g1_rem_set; + + template <class T> void do_oop_work(T* p); +public: + UpdateRSetImmediate(G1RemSet* rs) : + _g1_rem_set(rs) {} + + virtual void do_oop(narrowOop* p) { do_oop_work(p); } + virtual void do_oop( oop* p) { do_oop_work(p); } +}; diff --git a/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp b/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp index 2b916f866..ce6406552 100644 --- a/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp +++ b/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2001, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -56,19 +56,25 @@ template <class T> inline void HRInto_G1RemSet::par_write_ref_nv(HeapRegion* fro assert(Universe::heap()->is_in_reserved(obj), "must be in heap"); } #endif // ASSERT - assert(from == NULL || from->is_in_reserved(p), - "p is not in from"); + + assert(from == NULL || from->is_in_reserved(p), "p is not in from"); + HeapRegion* to = _g1->heap_region_containing(obj); // The test below could be optimized by applying a bit op to to and from. if (to != NULL && from != NULL && from != to) { - // There is a tricky infinite loop if we keep pushing - // self forwarding pointers onto our _new_refs list. - // The _par_traversal_in_progress flag is true during the collection pause, - // false during the evacuation failure handing. - if (_par_traversal_in_progress && + // The _traversal_in_progress flag is true during the collection pause, + // false during the evacuation failure handling. This should avoid a + // potential loop if we were to add the card containing 'p' to the DCQS + // that's used to regenerate the remembered sets for the collection set, + // in the event of an evacuation failure, here. The UpdateRSImmediate + // closure will eventally call this routine. + if (_traversal_in_progress && to->in_collection_set() && !self_forwarded(obj)) { - _new_refs[tid]->push((void*)p); - // Deferred updates to the Cset are either discarded (in the normal case), + + assert(_cset_rs_update_cl[tid] != NULL, "should have been set already"); + _cset_rs_update_cl[tid]->do_oop(p); + + // Deferred updates to the CSet are either discarded (in the normal case), // or processed (if an evacuation failure occurs) at the end // of the collection. // See HRInto_G1RemSet::cleanup_after_oops_into_collection_set_do(). @@ -89,3 +95,12 @@ template <class T> inline void UpdateRSOopClosure::do_oop_work(T* p) { assert(_from != NULL, "from region must be non-NULL"); _rs->par_write_ref(_from, p, _worker_i); } + +template <class T> inline void UpdateRSetImmediate::do_oop_work(T* p) { + assert(_from->is_in_reserved(p), "paranoia"); + T heap_oop = oopDesc::load_heap_oop(p); + if (!oopDesc::is_null(heap_oop) && !_from->is_survivor()) { + _g1_rem_set->par_write_ref(_from, p, 0); + } +} + diff --git a/src/share/vm/gc_implementation/g1/heapRegion.cpp b/src/share/vm/gc_implementation/g1/heapRegion.cpp index 4e5446e1f..58b55123a 100644 --- a/src/share/vm/gc_implementation/g1/heapRegion.cpp +++ b/src/share/vm/gc_implementation/g1/heapRegion.cpp @@ -683,6 +683,8 @@ oops_on_card_seq_iterate_careful(MemRegion mr, return NULL; } + assert(!is_young(), "check value of filter_young"); + // We used to use "block_start_careful" here. But we're actually happy // to update the BOT while we do this... HeapWord* cur = block_start(mr.start()); @@ -788,8 +790,18 @@ void HeapRegion::verify(bool allow_dirty, int objs = 0; int blocks = 0; VerifyLiveClosure vl_cl(g1, use_prev_marking); + bool is_humongous = isHumongous(); + size_t object_num = 0; while (p < top()) { size_t size = oop(p)->size(); + if (is_humongous != g1->isHumongous(size)) { + gclog_or_tty->print_cr("obj "PTR_FORMAT" is of %shumongous size (" + SIZE_FORMAT" words) in a %shumongous region", + p, g1->isHumongous(size) ? "" : "non-", + size, is_humongous ? "" : "non-"); + *failures = true; + } + object_num += 1; if (blocks == BLOCK_SAMPLE_INTERVAL) { HeapWord* res = block_start_const(p + (size/2)); if (p != res) { @@ -855,6 +867,13 @@ void HeapRegion::verify(bool allow_dirty, } } + if (is_humongous && object_num > 1) { + gclog_or_tty->print_cr("region ["PTR_FORMAT","PTR_FORMAT"] is humongous " + "but has "SIZE_FORMAT", objects", + bottom(), end(), object_num); + *failures = true; + } + if (p != top()) { gclog_or_tty->print_cr("end of last object "PTR_FORMAT" " "does not match top "PTR_FORMAT, p, top()); diff --git a/src/share/vm/gc_implementation/g1/sparsePRT.cpp b/src/share/vm/gc_implementation/g1/sparsePRT.cpp index 434836345..5bc3ab29c 100644 --- a/src/share/vm/gc_implementation/g1/sparsePRT.cpp +++ b/src/share/vm/gc_implementation/g1/sparsePRT.cpp @@ -424,7 +424,7 @@ void SparsePRT::cleanup_all() { SparsePRT::SparsePRT(HeapRegion* hr) : - _expanded(false), _next_expanded(NULL) + _hr(hr), _expanded(false), _next_expanded(NULL) { _cur = new RSHashTable(InitialCapacity); _next = _cur; diff --git a/src/share/vm/gc_implementation/includeDB_gc_concurrentMarkSweep b/src/share/vm/gc_implementation/includeDB_gc_concurrentMarkSweep index 7cda699d3..6efae46ed 100644 --- a/src/share/vm/gc_implementation/includeDB_gc_concurrentMarkSweep +++ b/src/share/vm/gc_implementation/includeDB_gc_concurrentMarkSweep @@ -1,5 +1,5 @@ // -// Copyright (c) 2004, 2009, Oracle and/or its affiliates. All rights reserved. +// Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved. // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. // // This code is free software; you can redistribute it and/or modify it @@ -34,6 +34,8 @@ binaryTreeDictionary.cpp spaceDecorator.hpp binaryTreeDictionary.hpp freeBlockDictionary.hpp binaryTreeDictionary.hpp freeList.hpp +blockOffsetTable.inline.hpp concurrentMarkSweepGeneration.hpp + cmsAdaptiveSizePolicy.cpp cmsAdaptiveSizePolicy.hpp cmsAdaptiveSizePolicy.cpp defNewGeneration.hpp cmsAdaptiveSizePolicy.cpp gcStats.hpp @@ -85,7 +87,7 @@ cmsOopClosures.hpp genOopClosures.hpp cmsOopClosures.inline.hpp cmsOopClosures.hpp cmsOopClosures.inline.hpp concurrentMarkSweepGeneration.hpp -cmsPermGen.cpp blockOffsetTable.hpp +cmsPermGen.cpp blockOffsetTable.inline.hpp cmsPermGen.cpp cSpaceCounters.hpp cmsPermGen.cpp cmsPermGen.hpp cmsPermGen.cpp collectedHeap.inline.hpp @@ -121,6 +123,7 @@ compactibleFreeListSpace.cpp universe.inline.hpp compactibleFreeListSpace.cpp vmThread.hpp compactibleFreeListSpace.hpp binaryTreeDictionary.hpp +compactibleFreeListSpace.hpp blockOffsetTable.inline.hpp compactibleFreeListSpace.hpp freeList.hpp compactibleFreeListSpace.hpp promotionInfo.hpp compactibleFreeListSpace.hpp space.hpp @@ -149,6 +152,7 @@ concurrentMarkSweepGeneration.cpp isGCActiveMark.hpp concurrentMarkSweepGeneration.cpp iterator.hpp concurrentMarkSweepGeneration.cpp java.hpp concurrentMarkSweepGeneration.cpp jvmtiExport.hpp +concurrentMarkSweepGeneration.cpp memoryService.hpp concurrentMarkSweepGeneration.cpp oop.inline.hpp concurrentMarkSweepGeneration.cpp parNewGeneration.hpp concurrentMarkSweepGeneration.cpp referencePolicy.hpp @@ -165,6 +169,7 @@ concurrentMarkSweepGeneration.hpp gSpaceCounters.hpp concurrentMarkSweepGeneration.hpp gcStats.hpp concurrentMarkSweepGeneration.hpp generation.hpp concurrentMarkSweepGeneration.hpp generationCounters.hpp +concurrentMarkSweepGeneration.hpp memoryService.hpp concurrentMarkSweepGeneration.hpp mutexLocker.hpp concurrentMarkSweepGeneration.hpp taskqueue.hpp concurrentMarkSweepGeneration.hpp virtualspace.hpp diff --git a/src/share/vm/gc_implementation/includeDB_gc_g1 b/src/share/vm/gc_implementation/includeDB_gc_g1 index 13e3464ed..d14769722 100644 --- a/src/share/vm/gc_implementation/includeDB_gc_g1 +++ b/src/share/vm/gc_implementation/includeDB_gc_g1 @@ -1,5 +1,5 @@ // -// Copyright (c) 2004, 2009, Oracle and/or its affiliates. All rights reserved. +// Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved. // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. // // This code is free software; you can redistribute it and/or modify it @@ -241,6 +241,7 @@ g1MMUTracker.cpp mutexLocker.hpp g1MMUTracker.hpp debug.hpp g1MMUTracker.hpp allocation.hpp + g1RemSet.cpp bufferingOopClosure.hpp g1RemSet.cpp concurrentG1Refine.hpp g1RemSet.cpp concurrentG1RefineThread.hpp diff --git a/src/share/vm/gc_implementation/includeDB_gc_parallelScavenge b/src/share/vm/gc_implementation/includeDB_gc_parallelScavenge index e06ed445a..83eae5eeb 100644 --- a/src/share/vm/gc_implementation/includeDB_gc_parallelScavenge +++ b/src/share/vm/gc_implementation/includeDB_gc_parallelScavenge @@ -330,7 +330,6 @@ psPromotionManager.cpp psPromotionManager.inline.hpp psPromotionManager.cpp psScavenge.inline.hpp psPromotionManager.hpp allocation.hpp -psPromotionManager.hpp prefetchQueue.hpp psPromotionManager.hpp psPromotionLAB.hpp psPromotionManager.hpp taskqueue.hpp diff --git a/src/share/vm/gc_implementation/parNew/parNewGeneration.cpp b/src/share/vm/gc_implementation/parNew/parNewGeneration.cpp index dc655aed1..e31e28546 100644 --- a/src/share/vm/gc_implementation/parNew/parNewGeneration.cpp +++ b/src/share/vm/gc_implementation/parNew/parNewGeneration.cpp @@ -51,9 +51,14 @@ ParScanThreadState::ParScanThreadState(Space* to_space_, _is_alive_closure(gen_), _scan_weak_ref_closure(gen_, this), _keep_alive_closure(&_scan_weak_ref_closure), _promotion_failure_size(0), - _pushes(0), _pops(0), _steals(0), _steal_attempts(0), _term_attempts(0), _strong_roots_time(0.0), _term_time(0.0) { + #if TASKQUEUE_STATS + _term_attempts = 0; + _overflow_refills = 0; + _overflow_refill_objs = 0; + #endif // TASKQUEUE_STATS + _survivor_chunk_array = (ChunkArray*) old_gen()->get_data_recorder(thread_num()); _hash_seed = 17; // Might want to take time-based random value. @@ -100,7 +105,6 @@ void ParScanThreadState::scan_partial_array_and_push_remainder(oop old) { // Push remainder. bool ok = work_queue()->push(old); assert(ok, "just popped, push must be okay"); - note_push(); } else { // Restore length so that it can be used if there // is a promotion failure and forwarding pointers @@ -126,7 +130,6 @@ void ParScanThreadState::trim_queues(int max_size) { while (queue->size() > (juint)max_size) { oop obj_to_scan; if (queue->pop_local(obj_to_scan)) { - note_pop(); if ((HeapWord *)obj_to_scan < young_old_boundary()) { if (obj_to_scan->is_objArray() && obj_to_scan->is_forwarded() && @@ -271,20 +274,28 @@ public: GrowableArray<oop>** overflow_stacks_, size_t desired_plab_sz, ParallelTaskTerminator& term); + + ~ParScanThreadStateSet() { TASKQUEUE_STATS_ONLY(reset_stats()); } + inline ParScanThreadState& thread_state(int i); - int pushes() { return _pushes; } - int pops() { return _pops; } - int steals() { return _steals; } + void reset(bool promotion_failed); void flush(); + + #if TASKQUEUE_STATS + static void + print_termination_stats_hdr(outputStream* const st = gclog_or_tty); + void print_termination_stats(outputStream* const st = gclog_or_tty); + static void + print_taskqueue_stats_hdr(outputStream* const st = gclog_or_tty); + void print_taskqueue_stats(outputStream* const st = gclog_or_tty); + void reset_stats(); + #endif // TASKQUEUE_STATS + private: ParallelTaskTerminator& _term; ParNewGeneration& _gen; Generation& _next_gen; - // staticstics - int _pushes; - int _pops; - int _steals; }; @@ -294,8 +305,7 @@ ParScanThreadStateSet::ParScanThreadStateSet( GrowableArray<oop>** overflow_stack_set_, size_t desired_plab_sz, ParallelTaskTerminator& term) : ResourceArray(sizeof(ParScanThreadState), num_threads), - _gen(gen), _next_gen(old_gen), _term(term), - _pushes(0), _pops(0), _steals(0) + _gen(gen), _next_gen(old_gen), _term(term) { assert(num_threads > 0, "sanity check!"); // Initialize states. @@ -323,6 +333,82 @@ void ParScanThreadStateSet::reset(bool promotion_failed) } } +#if TASKQUEUE_STATS +void +ParScanThreadState::reset_stats() +{ + taskqueue_stats().reset(); + _term_attempts = 0; + _overflow_refills = 0; + _overflow_refill_objs = 0; +} + +void ParScanThreadStateSet::reset_stats() +{ + for (int i = 0; i < length(); ++i) { + thread_state(i).reset_stats(); + } +} + +void +ParScanThreadStateSet::print_termination_stats_hdr(outputStream* const st) +{ + st->print_raw_cr("GC Termination Stats"); + st->print_raw_cr(" elapsed --strong roots-- " + "-------termination-------"); + st->print_raw_cr("thr ms ms % " + " ms % attempts"); + st->print_raw_cr("--- --------- --------- ------ " + "--------- ------ --------"); +} + +void ParScanThreadStateSet::print_termination_stats(outputStream* const st) +{ + print_termination_stats_hdr(st); + + for (int i = 0; i < length(); ++i) { + const ParScanThreadState & pss = thread_state(i); + const double elapsed_ms = pss.elapsed_time() * 1000.0; + const double s_roots_ms = pss.strong_roots_time() * 1000.0; + const double term_ms = pss.term_time() * 1000.0; + st->print_cr("%3d %9.2f %9.2f %6.2f " + "%9.2f %6.2f " SIZE_FORMAT_W(8), + i, elapsed_ms, s_roots_ms, s_roots_ms * 100 / elapsed_ms, + term_ms, term_ms * 100 / elapsed_ms, pss.term_attempts()); + } +} + +// Print stats related to work queue activity. +void ParScanThreadStateSet::print_taskqueue_stats_hdr(outputStream* const st) +{ + st->print_raw_cr("GC Task Stats"); + st->print_raw("thr "); TaskQueueStats::print_header(1, st); st->cr(); + st->print_raw("--- "); TaskQueueStats::print_header(2, st); st->cr(); +} + +void ParScanThreadStateSet::print_taskqueue_stats(outputStream* const st) +{ + print_taskqueue_stats_hdr(st); + + TaskQueueStats totals; + for (int i = 0; i < length(); ++i) { + const ParScanThreadState & pss = thread_state(i); + const TaskQueueStats & stats = pss.taskqueue_stats(); + st->print("%3d ", i); stats.print(st); st->cr(); + totals += stats; + + if (pss.overflow_refills() > 0) { + st->print_cr(" " SIZE_FORMAT_W(10) " overflow refills " + SIZE_FORMAT_W(10) " overflow objects", + pss.overflow_refills(), pss.overflow_refill_objs()); + } + } + st->print("tot "); totals.print(st); st->cr(); + + DEBUG_ONLY(totals.verify()); +} +#endif // TASKQUEUE_STATS + void ParScanThreadStateSet::flush() { // Work in this loop should be kept as lightweight as @@ -346,42 +432,8 @@ void ParScanThreadStateSet::flush() // Inform old gen that we're done. _next_gen.par_promote_alloc_done(i); _next_gen.par_oop_since_save_marks_iterate_done(i); - - // Flush stats related to work queue activity (push/pop/steal) - // This could conceivably become a bottleneck; if so, we'll put the - // stat's gathering under the flag. - if (PAR_STATS_ENABLED) { - _pushes += par_scan_state.pushes(); - _pops += par_scan_state.pops(); - _steals += par_scan_state.steals(); - if (ParallelGCVerbose) { - gclog_or_tty->print("Thread %d complete:\n" - " Pushes: %7d Pops: %7d Steals %7d (in %d attempts)\n", - i, par_scan_state.pushes(), par_scan_state.pops(), - par_scan_state.steals(), par_scan_state.steal_attempts()); - if (par_scan_state.overflow_pushes() > 0 || - par_scan_state.overflow_refills() > 0) { - gclog_or_tty->print(" Overflow pushes: %7d " - "Overflow refills: %7d for %d objs.\n", - par_scan_state.overflow_pushes(), - par_scan_state.overflow_refills(), - par_scan_state.overflow_refill_objs()); - } - - double elapsed = par_scan_state.elapsed(); - double strong_roots = par_scan_state.strong_roots_time(); - double term = par_scan_state.term_time(); - gclog_or_tty->print( - " Elapsed: %7.2f ms.\n" - " Strong roots: %7.2f ms (%6.2f%%)\n" - " Termination: %7.2f ms (%6.2f%%) (in %d entries)\n", - elapsed * 1000.0, - strong_roots * 1000.0, (strong_roots*100.0/elapsed), - term * 1000.0, (term*100.0/elapsed), - par_scan_state.term_attempts()); - } - } } + if (UseConcMarkSweepGC && ParallelGCThreads > 0) { // We need to call this even when ResizeOldPLAB is disabled // so as to avoid breaking some asserts. While we may be able @@ -456,15 +508,12 @@ void ParEvacuateFollowersClosure::do_void() { // We have no local work, attempt to steal from other threads. // attempt to steal work from promoted. - par_scan_state()->note_steal_attempt(); if (task_queues()->steal(par_scan_state()->thread_num(), par_scan_state()->hash_seed(), obj_to_scan)) { - par_scan_state()->note_steal(); bool res = work_q->push(obj_to_scan); assert(res, "Empty queue should have room for a push."); - par_scan_state()->note_push(); // if successful, goto Start. continue; @@ -842,17 +891,6 @@ void ParNewGeneration::collect(bool full, } thread_state_set.reset(promotion_failed()); - if (PAR_STATS_ENABLED && ParallelGCVerbose) { - gclog_or_tty->print("Thread totals:\n" - " Pushes: %7d Pops: %7d Steals %7d (sum = %7d).\n", - thread_state_set.pushes(), thread_state_set.pops(), - thread_state_set.steals(), - thread_state_set.pops()+thread_state_set.steals()); - } - assert(thread_state_set.pushes() == thread_state_set.pops() - + thread_state_set.steals(), - "Or else the queues are leaky."); - // Process (weak) reference objects found during scavenge. ReferenceProcessor* rp = ref_processor(); IsAliveClosure is_alive(this); @@ -932,6 +970,11 @@ void ParNewGeneration::collect(bool full, gch->print_heap_change(gch_prev_used); } + if (PrintGCDetails && ParallelGCVerbose) { + TASKQUEUE_STATS_ONLY(thread_state_set.print_termination_stats()); + TASKQUEUE_STATS_ONLY(thread_state_set.print_taskqueue_stats()); + } + if (UseAdaptiveSizePolicy) { size_policy->minor_collection_end(gch->gc_cause()); size_policy->avg_survived()->sample(from()->used()); @@ -1104,9 +1147,8 @@ oop ParNewGeneration::copy_to_survivor_space_avoiding_promotion_undo( gclog_or_tty->print("queue overflow!\n"); } push_on_overflow_list(old, par_scan_state); - par_scan_state->note_overflow_push(); + TASKQUEUE_STATS_ONLY(par_scan_state->taskqueue_stats().record_overflow(0)); } - par_scan_state->note_push(); return new_obj; } @@ -1227,9 +1269,8 @@ oop ParNewGeneration::copy_to_survivor_space_with_undo( if (simulate_overflow || !par_scan_state->work_queue()->push(obj_to_push)) { // Add stats for overflow pushes. push_on_overflow_list(old, par_scan_state); - par_scan_state->note_overflow_push(); + TASKQUEUE_STATS_ONLY(par_scan_state->taskqueue_stats().record_overflow(0)); } - par_scan_state->note_push(); return new_obj; } @@ -1466,7 +1507,7 @@ bool ParNewGeneration::take_from_overflow_list_work(ParScanThreadState* par_scan cur = next; n++; } - par_scan_state->note_overflow_refill(n); + TASKQUEUE_STATS_ONLY(par_scan_state->note_overflow_refill(n)); #ifndef PRODUCT assert(_num_par_pushes >= n, "Too many pops?"); Atomic::add_ptr(-(intptr_t)n, &_num_par_pushes); diff --git a/src/share/vm/gc_implementation/parNew/parNewGeneration.hpp b/src/share/vm/gc_implementation/parNew/parNewGeneration.hpp index 8196e6213..a3090ebf4 100644 --- a/src/share/vm/gc_implementation/parNew/parNewGeneration.hpp +++ b/src/share/vm/gc_implementation/parNew/parNewGeneration.hpp @@ -36,9 +36,6 @@ class ParEvacuateFollowersClosure; typedef Padded<OopTaskQueue> ObjToScanQueue; typedef GenericTaskQueueSet<ObjToScanQueue> ObjToScanQueueSet; -// Enable this to get push/pop/steal stats. -const int PAR_STATS_ENABLED = 0; - class ParKeepAliveClosure: public DefNewGeneration::KeepAliveClosure { private: ParScanWeakRefClosure* _par_cl; @@ -94,8 +91,11 @@ class ParScanThreadState { bool _to_space_full; - int _pushes, _pops, _steals, _steal_attempts, _term_attempts; - int _overflow_pushes, _overflow_refills, _overflow_refill_objs; +#if TASKQUEUE_STATS + size_t _term_attempts; + size_t _overflow_refills; + size_t _overflow_refill_objs; +#endif // TASKQUEUE_STATS // Stats for promotion failure size_t _promotion_failure_size; @@ -181,45 +181,38 @@ class ParScanThreadState { } void print_and_clear_promotion_failure_size(); - int pushes() { return _pushes; } - int pops() { return _pops; } - int steals() { return _steals; } - int steal_attempts() { return _steal_attempts; } - int term_attempts() { return _term_attempts; } - int overflow_pushes() { return _overflow_pushes; } - int overflow_refills() { return _overflow_refills; } - int overflow_refill_objs() { return _overflow_refill_objs; } - - void note_push() { if (PAR_STATS_ENABLED) _pushes++; } - void note_pop() { if (PAR_STATS_ENABLED) _pops++; } - void note_steal() { if (PAR_STATS_ENABLED) _steals++; } - void note_steal_attempt() { if (PAR_STATS_ENABLED) _steal_attempts++; } - void note_term_attempt() { if (PAR_STATS_ENABLED) _term_attempts++; } - void note_overflow_push() { if (PAR_STATS_ENABLED) _overflow_pushes++; } - void note_overflow_refill(int objs) { - if (PAR_STATS_ENABLED) { - _overflow_refills++; - _overflow_refill_objs += objs; - } +#if TASKQUEUE_STATS + TaskQueueStats & taskqueue_stats() const { return _work_queue->stats; } + + size_t term_attempts() const { return _term_attempts; } + size_t overflow_refills() const { return _overflow_refills; } + size_t overflow_refill_objs() const { return _overflow_refill_objs; } + + void note_term_attempt() { ++_term_attempts; } + void note_overflow_refill(size_t objs) { + ++_overflow_refills; _overflow_refill_objs += objs; } + void reset_stats(); +#endif // TASKQUEUE_STATS + void start_strong_roots() { _start_strong_roots = os::elapsedTime(); } void end_strong_roots() { _strong_roots_time += (os::elapsedTime() - _start_strong_roots); } - double strong_roots_time() { return _strong_roots_time; } + double strong_roots_time() const { return _strong_roots_time; } void start_term_time() { - note_term_attempt(); + TASKQUEUE_STATS_ONLY(note_term_attempt()); _start_term = os::elapsedTime(); } void end_term_time() { _term_time += (os::elapsedTime() - _start_term); } - double term_time() { return _term_time; } + double term_time() const { return _term_time; } - double elapsed() { + double elapsed_time() const { return os::elapsedTime() - _start; } }; diff --git a/src/share/vm/gc_implementation/parallelScavenge/cardTableExtension.cpp b/src/share/vm/gc_implementation/parallelScavenge/cardTableExtension.cpp index c87190d0c..e9da8d5fa 100644 --- a/src/share/vm/gc_implementation/parallelScavenge/cardTableExtension.cpp +++ b/src/share/vm/gc_implementation/parallelScavenge/cardTableExtension.cpp @@ -123,7 +123,6 @@ void CardTableExtension::scavenge_contents(ObjectStartArray* start_array, assert(start_array != NULL && sp != NULL && pm != NULL, "Sanity"); assert(start_array->covered_region().contains(sp->used_region()), "ObjectStartArray does not cover space"); - bool depth_first = pm->depth_first(); if (sp->not_empty()) { oop* sp_top = (oop*)space_top; @@ -201,21 +200,12 @@ void CardTableExtension::scavenge_contents(ObjectStartArray* start_array, *first_nonclean_card++ = clean_card; } // scan oops in objects - // hoisted the if (depth_first) check out of the loop - if (depth_first){ - do { - oop(bottom_obj)->push_contents(pm); - bottom_obj += oop(bottom_obj)->size(); - assert(bottom_obj <= sp_top, "just checking"); - } while (bottom_obj < top); - pm->drain_stacks_cond_depth(); - } else { - do { - oop(bottom_obj)->copy_contents(pm); - bottom_obj += oop(bottom_obj)->size(); - assert(bottom_obj <= sp_top, "just checking"); - } while (bottom_obj < top); - } + do { + oop(bottom_obj)->push_contents(pm); + bottom_obj += oop(bottom_obj)->size(); + assert(bottom_obj <= sp_top, "just checking"); + } while (bottom_obj < top); + pm->drain_stacks_cond_depth(); // remember top oop* scanned prev_top = top; } @@ -230,7 +220,6 @@ void CardTableExtension::scavenge_contents_parallel(ObjectStartArray* start_arra uint stripe_number) { int ssize = 128; // Naked constant! Work unit = 64k. int dirty_card_count = 0; - bool depth_first = pm->depth_first(); oop* sp_top = (oop*)space_top; jbyte* start_card = byte_for(sp->bottom()); @@ -363,43 +352,22 @@ void CardTableExtension::scavenge_contents_parallel(ObjectStartArray* start_arra const int interval = PrefetchScanIntervalInBytes; // scan all objects in the range if (interval != 0) { - // hoisted the if (depth_first) check out of the loop - if (depth_first) { - while (p < to) { - Prefetch::write(p, interval); - oop m = oop(p); - assert(m->is_oop_or_null(), "check for header"); - m->push_contents(pm); - p += m->size(); - } - pm->drain_stacks_cond_depth(); - } else { - while (p < to) { - Prefetch::write(p, interval); - oop m = oop(p); - assert(m->is_oop_or_null(), "check for header"); - m->copy_contents(pm); - p += m->size(); - } + while (p < to) { + Prefetch::write(p, interval); + oop m = oop(p); + assert(m->is_oop_or_null(), "check for header"); + m->push_contents(pm); + p += m->size(); } + pm->drain_stacks_cond_depth(); } else { - // hoisted the if (depth_first) check out of the loop - if (depth_first) { - while (p < to) { - oop m = oop(p); - assert(m->is_oop_or_null(), "check for header"); - m->push_contents(pm); - p += m->size(); - } - pm->drain_stacks_cond_depth(); - } else { - while (p < to) { - oop m = oop(p); - assert(m->is_oop_or_null(), "check for header"); - m->copy_contents(pm); - p += m->size(); - } + while (p < to) { + oop m = oop(p); + assert(m->is_oop_or_null(), "check for header"); + m->push_contents(pm); + p += m->size(); } + pm->drain_stacks_cond_depth(); } last_scanned = p; } diff --git a/src/share/vm/gc_implementation/parallelScavenge/prefetchQueue.hpp b/src/share/vm/gc_implementation/parallelScavenge/prefetchQueue.hpp deleted file mode 100644 index cd57b5ff8..000000000 --- a/src/share/vm/gc_implementation/parallelScavenge/prefetchQueue.hpp +++ /dev/null @@ -1,68 +0,0 @@ -/* - * Copyright (c) 2002, 2008, 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. - * - */ - -// -// PrefetchQueue is a FIFO queue of variable length (currently 8). -// -// We need to examine the performance penalty of variable lengths. -// We may also want to split this into cpu dependent bits. -// - -const int PREFETCH_QUEUE_SIZE = 8; - -class PrefetchQueue : public CHeapObj { - private: - void* _prefetch_queue[PREFETCH_QUEUE_SIZE]; - uint _prefetch_index; - - public: - int length() { return PREFETCH_QUEUE_SIZE; } - - inline void clear() { - for(int i=0; i<PREFETCH_QUEUE_SIZE; i++) { - _prefetch_queue[i] = NULL; - } - _prefetch_index = 0; - } - - template <class T> inline void* push_and_pop(T* p) { - oop o = oopDesc::load_decode_heap_oop_not_null(p); - Prefetch::write(o->mark_addr(), 0); - // This prefetch is intended to make sure the size field of array - // oops is in cache. It assumes the the object layout is - // mark -> klass -> size, and that mark and klass are heapword - // sized. If this should change, this prefetch will need updating! - Prefetch::write(o->mark_addr() + (HeapWordSize*2), 0); - _prefetch_queue[_prefetch_index++] = p; - _prefetch_index &= (PREFETCH_QUEUE_SIZE-1); - return _prefetch_queue[_prefetch_index]; - } - - // Stores a NULL pointer in the pop'd location. - inline void* pop() { - _prefetch_queue[_prefetch_index++] = NULL; - _prefetch_index &= (PREFETCH_QUEUE_SIZE-1); - return _prefetch_queue[_prefetch_index]; - } -}; diff --git a/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.cpp b/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.cpp index 2da32555d..1e73d11d1 100644 --- a/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.cpp +++ b/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.cpp @@ -27,7 +27,6 @@ PSPromotionManager** PSPromotionManager::_manager_array = NULL; OopStarTaskQueueSet* PSPromotionManager::_stack_array_depth = NULL; -OopTaskQueueSet* PSPromotionManager::_stack_array_breadth = NULL; PSOldGen* PSPromotionManager::_old_gen = NULL; MutableSpace* PSPromotionManager::_young_space = NULL; @@ -42,23 +41,14 @@ void PSPromotionManager::initialize() { _manager_array = NEW_C_HEAP_ARRAY(PSPromotionManager*, ParallelGCThreads+1 ); guarantee(_manager_array != NULL, "Could not initialize promotion manager"); - if (UseDepthFirstScavengeOrder) { - _stack_array_depth = new OopStarTaskQueueSet(ParallelGCThreads); - guarantee(_stack_array_depth != NULL, "Count not initialize promotion manager"); - } else { - _stack_array_breadth = new OopTaskQueueSet(ParallelGCThreads); - guarantee(_stack_array_breadth != NULL, "Count not initialize promotion manager"); - } + _stack_array_depth = new OopStarTaskQueueSet(ParallelGCThreads); + guarantee(_stack_array_depth != NULL, "Cound not initialize promotion manager"); // Create and register the PSPromotionManager(s) for the worker threads. for(uint i=0; i<ParallelGCThreads; i++) { _manager_array[i] = new PSPromotionManager(); guarantee(_manager_array[i] != NULL, "Could not create PSPromotionManager"); - if (UseDepthFirstScavengeOrder) { - stack_array_depth()->register_queue(i, _manager_array[i]->claimed_stack_depth()); - } else { - stack_array_breadth()->register_queue(i, _manager_array[i]->claimed_stack_breadth()); - } + stack_array_depth()->register_queue(i, _manager_array[i]->claimed_stack_depth()); } // The VMThread gets its own PSPromotionManager, which is not available @@ -93,11 +83,7 @@ void PSPromotionManager::post_scavenge() { TASKQUEUE_STATS_ONLY(if (PrintGCDetails && ParallelGCVerbose) print_stats()); for (uint i = 0; i < ParallelGCThreads + 1; i++) { PSPromotionManager* manager = manager_array(i); - if (UseDepthFirstScavengeOrder) { - assert(manager->claimed_stack_depth()->is_empty(), "should be empty"); - } else { - assert(manager->claimed_stack_breadth()->is_empty(), "should be empty"); - } + assert(manager->claimed_stack_depth()->is_empty(), "should be empty"); manager->flush_labs(); } } @@ -105,10 +91,8 @@ void PSPromotionManager::post_scavenge() { #if TASKQUEUE_STATS void PSPromotionManager::print_taskqueue_stats(uint i) const { - const TaskQueueStats& stats = depth_first() ? - _claimed_stack_depth.stats : _claimed_stack_breadth.stats; tty->print("%3u ", i); - stats.print(); + _claimed_stack_depth.stats.print(); tty->cr(); } @@ -128,8 +112,7 @@ static const char* const pm_stats_hdr[] = { void PSPromotionManager::print_stats() { - const bool df = UseDepthFirstScavengeOrder; - tty->print_cr("== GC Task Stats (%s-First), GC %3d", df ? "Depth" : "Breadth", + tty->print_cr("== GC Tasks Stats, GC %3d", Universe::heap()->total_collections()); tty->print("thr "); TaskQueueStats::print_header(1); tty->cr(); @@ -147,9 +130,7 @@ PSPromotionManager::print_stats() { void PSPromotionManager::reset_stats() { - TaskQueueStats& stats = depth_first() ? - claimed_stack_depth()->stats : claimed_stack_breadth()->stats; - stats.reset(); + claimed_stack_depth()->stats.reset(); _masked_pushes = _masked_steals = 0; _arrays_chunked = _array_chunks_processed = 0; } @@ -158,19 +139,13 @@ PSPromotionManager::reset_stats() { PSPromotionManager::PSPromotionManager() { ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap(); assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity"); - _depth_first = UseDepthFirstScavengeOrder; // We set the old lab's start array. _old_lab.set_start_array(old_gen()->start_array()); uint queue_size; - if (depth_first()) { - claimed_stack_depth()->initialize(); - queue_size = claimed_stack_depth()->max_elems(); - } else { - claimed_stack_breadth()->initialize(); - queue_size = claimed_stack_breadth()->max_elems(); - } + claimed_stack_depth()->initialize(); + queue_size = claimed_stack_depth()->max_elems(); _totally_drain = (ParallelGCThreads == 1) || (GCDrainStackTargetSize == 0); if (_totally_drain) { @@ -205,14 +180,11 @@ void PSPromotionManager::reset() { _old_lab.initialize(MemRegion(lab_base, (size_t)0)); _old_gen_is_full = false; - _prefetch_queue.clear(); - TASKQUEUE_STATS_ONLY(reset_stats()); } void PSPromotionManager::drain_stacks_depth(bool totally_drain) { - assert(depth_first(), "invariant"); assert(claimed_stack_depth()->overflow_stack() != NULL, "invariant"); totally_drain = totally_drain || _totally_drain; @@ -250,50 +222,6 @@ void PSPromotionManager::drain_stacks_depth(bool totally_drain) { assert(tq->overflow_empty(), "Sanity"); } -void PSPromotionManager::drain_stacks_breadth(bool totally_drain) { - assert(!depth_first(), "invariant"); - assert(claimed_stack_breadth()->overflow_stack() != NULL, "invariant"); - totally_drain = totally_drain || _totally_drain; - -#ifdef ASSERT - ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap(); - assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity"); - MutableSpace* to_space = heap->young_gen()->to_space(); - MutableSpace* old_space = heap->old_gen()->object_space(); - MutableSpace* perm_space = heap->perm_gen()->object_space(); -#endif /* ASSERT */ - - OverflowTaskQueue<oop>* const tq = claimed_stack_breadth(); - do { - oop obj; - - // Drain overflow stack first, so other threads can steal from - // claimed stack while we work. - while (tq->pop_overflow(obj)) { - obj->copy_contents(this); - } - - if (totally_drain) { - while (tq->pop_local(obj)) { - obj->copy_contents(this); - } - } else { - while (tq->size() > _target_stack_size && tq->pop_local(obj)) { - obj->copy_contents(this); - } - } - - // If we could not find any other work, flush the prefetch queue - if (tq->is_empty()) { - flush_prefetch_queue(); - } - } while (totally_drain && !tq->taskqueue_empty() || !tq->overflow_empty()); - - assert(!totally_drain || tq->taskqueue_empty(), "Sanity"); - assert(totally_drain || tq->size() <= _target_stack_size, "Sanity"); - assert(tq->overflow_empty(), "Sanity"); -} - void PSPromotionManager::flush_labs() { assert(stacks_empty(), "Attempt to flush lab with live stack"); @@ -319,7 +247,7 @@ void PSPromotionManager::flush_labs() { // performance. // -oop PSPromotionManager::copy_to_survivor_space(oop o, bool depth_first) { +oop PSPromotionManager::copy_to_survivor_space(oop o) { assert(PSScavenge::should_scavenge(&o), "Sanity"); oop new_obj = NULL; @@ -423,24 +351,20 @@ oop PSPromotionManager::copy_to_survivor_space(oop o, bool depth_first) { assert(young_space()->contains(new_obj), "Attempt to push non-promoted obj"); } - if (depth_first) { - // Do the size comparison first with new_obj_size, which we - // already have. Hopefully, only a few objects are larger than - // _min_array_size_for_chunking, and most of them will be arrays. - // So, the is->objArray() test would be very infrequent. - if (new_obj_size > _min_array_size_for_chunking && - new_obj->is_objArray() && - PSChunkLargeArrays) { - // we'll chunk it - oop* const masked_o = mask_chunked_array_oop(o); - push_depth(masked_o); - TASKQUEUE_STATS_ONLY(++_arrays_chunked; ++_masked_pushes); - } else { - // we'll just push its contents - new_obj->push_contents(this); - } + // Do the size comparison first with new_obj_size, which we + // already have. Hopefully, only a few objects are larger than + // _min_array_size_for_chunking, and most of them will be arrays. + // So, the is->objArray() test would be very infrequent. + if (new_obj_size > _min_array_size_for_chunking && + new_obj->is_objArray() && + PSChunkLargeArrays) { + // we'll chunk it + oop* const masked_o = mask_chunked_array_oop(o); + push_depth(masked_o); + TASKQUEUE_STATS_ONLY(++_arrays_chunked; ++_masked_pushes); } else { - push_breadth(new_obj); + // we'll just push its contents + new_obj->push_contents(this); } } else { // We lost, someone else "owns" this object @@ -537,13 +461,7 @@ oop PSPromotionManager::oop_promotion_failed(oop obj, markOop obj_mark) { // We won any races, we "own" this object. assert(obj == obj->forwardee(), "Sanity"); - if (depth_first()) { - obj->push_contents(this); - } else { - // Don't bother incrementing the age, just push - // onto the claimed_stack.. - push_breadth(obj); - } + obj->push_contents(this); // Save the mark if needed PSScavenge::oop_promotion_failed(obj, obj_mark); diff --git a/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.hpp b/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.hpp index ec89b9557..aa14478d4 100644 --- a/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.hpp +++ b/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.hpp @@ -48,7 +48,6 @@ class PSPromotionManager : public CHeapObj { private: static PSPromotionManager** _manager_array; static OopStarTaskQueueSet* _stack_array_depth; - static OopTaskQueueSet* _stack_array_breadth; static PSOldGen* _old_gen; static MutableSpace* _young_space; @@ -69,12 +68,10 @@ class PSPromotionManager : public CHeapObj { PSOldPromotionLAB _old_lab; bool _young_gen_is_full; bool _old_gen_is_full; - PrefetchQueue _prefetch_queue; OopStarTaskQueue _claimed_stack_depth; OverflowTaskQueue<oop> _claimed_stack_breadth; - bool _depth_first; bool _totally_drain; uint _target_stack_size; @@ -87,7 +84,6 @@ class PSPromotionManager : public CHeapObj { inline static PSPromotionManager* manager_array(int index); template <class T> inline void claim_or_forward_internal_depth(T* p); - template <class T> inline void claim_or_forward_internal_breadth(T* p); // On the task queues we push reference locations as well as // partially-scanned arrays (in the latter case, we push an oop to @@ -136,19 +132,11 @@ class PSPromotionManager : public CHeapObj { void process_array_chunk(oop old); template <class T> void push_depth(T* p) { - assert(depth_first(), "pre-condition"); claimed_stack_depth()->push(p); } - void push_breadth(oop o) { - assert(!depth_first(), "pre-condition"); - claimed_stack_breadth()->push(o); - } - protected: static OopStarTaskQueueSet* stack_array_depth() { return _stack_array_depth; } - static OopTaskQueueSet* stack_array_breadth() { return _stack_array_breadth; } - public: // Static static void initialize(); @@ -163,19 +151,12 @@ class PSPromotionManager : public CHeapObj { return stack_array_depth()->steal(queue_num, seed, t); } - static bool steal_breadth(int queue_num, int* seed, oop& t) { - return stack_array_breadth()->steal(queue_num, seed, t); - } - PSPromotionManager(); // Accessors OopStarTaskQueue* claimed_stack_depth() { return &_claimed_stack_depth; } - OverflowTaskQueue<oop>* claimed_stack_breadth() { - return &_claimed_stack_breadth; - } bool young_gen_is_full() { return _young_gen_is_full; } @@ -183,18 +164,14 @@ class PSPromotionManager : public CHeapObj { void set_old_gen_is_full(bool state) { _old_gen_is_full = state; } // Promotion methods - oop copy_to_survivor_space(oop o, bool depth_first); + oop copy_to_survivor_space(oop o); oop oop_promotion_failed(oop obj, markOop obj_mark); void reset(); void flush_labs(); void drain_stacks(bool totally_drain) { - if (depth_first()) { - drain_stacks_depth(totally_drain); - } else { - drain_stacks_breadth(totally_drain); - } + drain_stacks_depth(totally_drain); } public: void drain_stacks_cond_depth() { @@ -203,22 +180,14 @@ class PSPromotionManager : public CHeapObj { } } void drain_stacks_depth(bool totally_drain); - void drain_stacks_breadth(bool totally_drain); - bool depth_first() const { - return _depth_first; - } bool stacks_empty() { - return depth_first() ? - claimed_stack_depth()->is_empty() : - claimed_stack_breadth()->is_empty(); + return claimed_stack_depth()->is_empty(); } inline void process_popped_location_depth(StarTask p); - inline void flush_prefetch_queue(); template <class T> inline void claim_or_forward_depth(T* p); - template <class T> inline void claim_or_forward_breadth(T* p); TASKQUEUE_STATS_ONLY(inline void record_steal(StarTask& p);) }; diff --git a/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.inline.hpp b/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.inline.hpp index ea81c817b..decc5e99a 100644 --- a/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.inline.hpp +++ b/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.inline.hpp @@ -46,32 +46,7 @@ inline void PSPromotionManager::claim_or_forward_internal_depth(T* p) { } template <class T> -inline void PSPromotionManager::claim_or_forward_internal_breadth(T* p) { - if (p != NULL) { // XXX: error if p != NULL here - oop o = oopDesc::load_decode_heap_oop_not_null(p); - if (o->is_forwarded()) { - o = o->forwardee(); - } else { - o = copy_to_survivor_space(o, false); - } - // Card mark - if (PSScavenge::is_obj_in_young((HeapWord*) o)) { - PSScavenge::card_table()->inline_write_ref_field_gc(p, o); - } - oopDesc::encode_store_heap_oop_not_null(p, o); - } -} - -inline void PSPromotionManager::flush_prefetch_queue() { - assert(!depth_first(), "invariant"); - for (int i = 0; i < _prefetch_queue.length(); i++) { - claim_or_forward_internal_breadth((oop*)_prefetch_queue.pop()); - } -} - -template <class T> inline void PSPromotionManager::claim_or_forward_depth(T* p) { - assert(depth_first(), "invariant"); assert(PSScavenge::should_scavenge(p, true), "revisiting object?"); assert(Universe::heap()->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity"); @@ -80,36 +55,6 @@ inline void PSPromotionManager::claim_or_forward_depth(T* p) { claim_or_forward_internal_depth(p); } -template <class T> -inline void PSPromotionManager::claim_or_forward_breadth(T* p) { - assert(!depth_first(), "invariant"); - assert(PSScavenge::should_scavenge(p, true), "revisiting object?"); - assert(Universe::heap()->kind() == CollectedHeap::ParallelScavengeHeap, - "Sanity"); - assert(Universe::heap()->is_in(p), "pointer outside heap"); - - if (UsePrefetchQueue) { - claim_or_forward_internal_breadth((T*)_prefetch_queue.push_and_pop(p)); - } else { - // This option is used for testing. The use of the prefetch - // queue can delay the processing of the objects and thus - // change the order of object scans. For example, remembered - // set updates are typically the clearing of the remembered - // set (the cards) followed by updates of the remembered set - // for young-to-old pointers. In a situation where there - // is an error in the sequence of clearing and updating - // (e.g. clear card A, update card A, erroneously clear - // card A again) the error can be obscured by a delay - // in the update due to the use of the prefetch queue - // (e.g., clear card A, erroneously clear card A again, - // update card A that was pushed into the prefetch queue - // and thus delayed until after the erronous clear). The - // length of the delay is random depending on the objects - // in the queue and the delay can be zero. - claim_or_forward_internal_breadth(p); - } -} - inline void PSPromotionManager::process_popped_location_depth(StarTask p) { if (is_oop_masked(p)) { assert(PSChunkLargeArrays, "invariant"); diff --git a/src/share/vm/gc_implementation/parallelScavenge/psScavenge.cpp b/src/share/vm/gc_implementation/parallelScavenge/psScavenge.cpp index 9cc02479d..0ea076812 100644 --- a/src/share/vm/gc_implementation/parallelScavenge/psScavenge.cpp +++ b/src/share/vm/gc_implementation/parallelScavenge/psScavenge.cpp @@ -157,10 +157,8 @@ void PSRefProcTaskExecutor::execute(ProcessTask& task) q->enqueue(new PSRefProcTaskProxy(task, i)); } ParallelTaskTerminator terminator( - ParallelScavengeHeap::gc_task_manager()->workers(), - UseDepthFirstScavengeOrder ? - (TaskQueueSetSuper*) PSPromotionManager::stack_array_depth() - : (TaskQueueSetSuper*) PSPromotionManager::stack_array_breadth()); + ParallelScavengeHeap::gc_task_manager()->workers(), + (TaskQueueSetSuper*) PSPromotionManager::stack_array_depth()); if (task.marks_oops_alive() && ParallelGCThreads > 1) { for (uint j=0; j<ParallelGCThreads; j++) { q->enqueue(new StealTask(&terminator)); @@ -375,10 +373,8 @@ bool PSScavenge::invoke_no_policy() { q->enqueue(new ScavengeRootsTask(ScavengeRootsTask::code_cache)); ParallelTaskTerminator terminator( - gc_task_manager()->workers(), - promotion_manager->depth_first() ? - (TaskQueueSetSuper*) promotion_manager->stack_array_depth() - : (TaskQueueSetSuper*) promotion_manager->stack_array_breadth()); + gc_task_manager()->workers(), + (TaskQueueSetSuper*) promotion_manager->stack_array_depth()); if (ParallelGCThreads>1) { for (uint j=0; j<ParallelGCThreads; j++) { q->enqueue(new StealTask(&terminator)); diff --git a/src/share/vm/gc_implementation/parallelScavenge/psScavenge.inline.hpp b/src/share/vm/gc_implementation/parallelScavenge/psScavenge.inline.hpp index 7dcf93a95..3de87882f 100644 --- a/src/share/vm/gc_implementation/parallelScavenge/psScavenge.inline.hpp +++ b/src/share/vm/gc_implementation/parallelScavenge/psScavenge.inline.hpp @@ -65,7 +65,7 @@ inline void PSScavenge::copy_and_push_safe_barrier(PSPromotionManager* pm, oop o = oopDesc::load_decode_heap_oop_not_null(p); oop new_obj = o->is_forwarded() ? o->forwardee() - : pm->copy_to_survivor_space(o, pm->depth_first()); + : pm->copy_to_survivor_space(o); oopDesc::encode_store_heap_oop_not_null(p, new_obj); // We cannot mark without test, as some code passes us pointers diff --git a/src/share/vm/gc_implementation/parallelScavenge/psTasks.cpp b/src/share/vm/gc_implementation/parallelScavenge/psTasks.cpp index 6f72724bf..11774d941 100644 --- a/src/share/vm/gc_implementation/parallelScavenge/psTasks.cpp +++ b/src/share/vm/gc_implementation/parallelScavenge/psTasks.cpp @@ -144,29 +144,15 @@ void StealTask::do_it(GCTaskManager* manager, uint which) { "stacks should be empty at this point"); int random_seed = 17; - if (pm->depth_first()) { - while(true) { - StarTask p; - if (PSPromotionManager::steal_depth(which, &random_seed, p)) { - TASKQUEUE_STATS_ONLY(pm->record_steal(p)); - pm->process_popped_location_depth(p); - pm->drain_stacks_depth(true); - } else { - if (terminator()->offer_termination()) { - break; - } - } - } - } else { - while(true) { - oop obj; - if (PSPromotionManager::steal_breadth(which, &random_seed, obj)) { - obj->copy_contents(pm); - pm->drain_stacks_breadth(true); - } else { - if (terminator()->offer_termination()) { - break; - } + while(true) { + StarTask p; + if (PSPromotionManager::steal_depth(which, &random_seed, p)) { + TASKQUEUE_STATS_ONLY(pm->record_steal(p)); + pm->process_popped_location_depth(p); + pm->drain_stacks_depth(true); + } else { + if (terminator()->offer_termination()) { + break; } } } diff --git a/src/share/vm/includeDB_core b/src/share/vm/includeDB_core index fe4252dfc..07c40dc39 100644 --- a/src/share/vm/includeDB_core +++ b/src/share/vm/includeDB_core @@ -225,7 +225,6 @@ arrayOop.cpp oop.inline.hpp arrayOop.cpp symbolOop.hpp arrayOop.hpp oop.hpp -arrayOop.hpp universe.hpp arrayOop.hpp universe.inline.hpp assembler.cpp assembler.hpp @@ -236,7 +235,6 @@ assembler.cpp icache.hpp assembler.cpp os.hpp assembler.hpp allocation.hpp -assembler.hpp allocation.inline.hpp assembler.hpp debug.hpp assembler.hpp growableArray.hpp assembler.hpp oopRecorder.hpp @@ -330,7 +328,7 @@ blockOffsetTable.cpp collectedHeap.inline.hpp blockOffsetTable.cpp iterator.hpp blockOffsetTable.cpp java.hpp blockOffsetTable.cpp oop.inline.hpp -blockOffsetTable.cpp space.hpp +blockOffsetTable.cpp space.inline.hpp blockOffsetTable.cpp universe.hpp blockOffsetTable.hpp globalDefinitions.hpp @@ -338,6 +336,7 @@ blockOffsetTable.hpp memRegion.hpp blockOffsetTable.hpp virtualspace.hpp blockOffsetTable.inline.hpp blockOffsetTable.hpp +blockOffsetTable.inline.hpp safepoint.hpp blockOffsetTable.inline.hpp space.hpp bytecode.cpp bytecode.hpp @@ -1807,7 +1806,7 @@ generateOopMap.hpp signature.hpp generateOopMap.hpp universe.inline.hpp generation.cpp allocation.inline.hpp -generation.cpp blockOffsetTable.hpp +generation.cpp blockOffsetTable.inline.hpp generation.cpp cardTableRS.hpp generation.cpp collectedHeap.inline.hpp generation.cpp copy.hpp @@ -3436,7 +3435,7 @@ perfMemory_<os_family>.cpp perfMemory.hpp perfMemory_<os_family>.cpp resourceArea.hpp perfMemory_<os_family>.cpp vmSymbols.hpp -permGen.cpp blockOffsetTable.hpp +permGen.cpp blockOffsetTable.inline.hpp permGen.cpp cSpaceCounters.hpp permGen.cpp collectedHeap.inline.hpp permGen.cpp compactPermGen.hpp @@ -3805,7 +3804,7 @@ sizes.cpp sizes.hpp sizes.hpp allocation.hpp sizes.hpp globalDefinitions.hpp -space.cpp blockOffsetTable.hpp +space.cpp blockOffsetTable.inline.hpp space.cpp copy.hpp space.cpp defNewGeneration.hpp space.cpp genCollectedHeap.hpp @@ -3835,7 +3834,6 @@ space.hpp prefetch.hpp space.hpp watermark.hpp space.hpp workgroup.hpp -space.inline.hpp blockOffsetTable.inline.hpp space.inline.hpp collectedHeap.hpp space.inline.hpp safepoint.hpp space.inline.hpp space.hpp diff --git a/src/share/vm/interpreter/bytecodeInterpreter.cpp b/src/share/vm/interpreter/bytecodeInterpreter.cpp index 29b4ede2b..9581f24bc 100644 --- a/src/share/vm/interpreter/bytecodeInterpreter.cpp +++ b/src/share/vm/interpreter/bytecodeInterpreter.cpp @@ -421,7 +421,9 @@ BytecodeInterpreter::run(interpreterState istate) { #ifdef ASSERT if (istate->_msg != initialize) { assert(abs(istate->_stack_base - istate->_stack_limit) == (istate->_method->max_stack() + 1), "bad stack limit"); - IA32_ONLY(assert(istate->_stack_limit == istate->_thread->last_Java_sp() + 1, "wrong")); +#ifndef SHARK + IA32_ONLY(assert(istate->_stack_limit == istate->_thread->last_Java_sp() + 1, "wrong")); +#endif // !SHARK } // Verify linkages. interpreterState l = istate; diff --git a/src/share/vm/memory/allocation.cpp b/src/share/vm/memory/allocation.cpp index 6331bcefc..ada585932 100644 --- a/src/share/vm/memory/allocation.cpp +++ b/src/share/vm/memory/allocation.cpp @@ -58,7 +58,7 @@ void* ResourceObj::operator new(size_t size, allocation_type type) { void ResourceObj::operator delete(void* p) { assert(((ResourceObj *)p)->allocated_on_C_heap(), "delete only allowed for C_HEAP objects"); - DEBUG_ONLY(((ResourceObj *)p)->_allocation = badHeapOopVal;) + DEBUG_ONLY(((ResourceObj *)p)->_allocation = (uintptr_t) badHeapOopVal;) FreeHeap(p); } @@ -104,7 +104,7 @@ ResourceObj& ResourceObj::operator=(const ResourceObj& r) { // default copy assi ResourceObj::~ResourceObj() { // allocated_on_C_heap() also checks that encoded (in _allocation) address == this. if (!allocated_on_C_heap()) { // ResourceObj::delete() zaps _allocation for C_heap. - _allocation = badHeapOopVal; // zap type + _allocation = (uintptr_t) badHeapOopVal; // zap type } } #endif // ASSERT diff --git a/src/share/vm/memory/blockOffsetTable.cpp b/src/share/vm/memory/blockOffsetTable.cpp index 038702027..ee3ef125c 100644 --- a/src/share/vm/memory/blockOffsetTable.cpp +++ b/src/share/vm/memory/blockOffsetTable.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000, 2006, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -103,13 +103,13 @@ void BlockOffsetSharedArray::serialize(SerializeOopClosure* soc, ////////////////////////////////////////////////////////////////////// BlockOffsetArray::BlockOffsetArray(BlockOffsetSharedArray* array, - MemRegion mr, bool init_to_zero) : + MemRegion mr, bool init_to_zero_) : BlockOffsetTable(mr.start(), mr.end()), - _array(array), - _init_to_zero(init_to_zero) + _array(array) { assert(_bottom <= _end, "arguments out of order"); - if (!_init_to_zero) { + set_init_to_zero(init_to_zero_); + if (!init_to_zero_) { // initialize cards to point back to mr.start() set_remainder_to_point_to_start(mr.start() + N_words, mr.end()); _array->set_offset_array(0, 0); // set first card to 0 @@ -121,8 +121,9 @@ BlockOffsetArray::BlockOffsetArray(BlockOffsetSharedArray* array, // a right-open interval: [start, end) void BlockOffsetArray:: -set_remainder_to_point_to_start(HeapWord* start, HeapWord* end) { +set_remainder_to_point_to_start(HeapWord* start, HeapWord* end, bool reducing) { + check_reducing_assertion(reducing); if (start >= end) { // The start address is equal to the end address (or to // the right of the end address) so there are not cards @@ -167,7 +168,7 @@ set_remainder_to_point_to_start(HeapWord* start, HeapWord* end) { size_t end_card = _array->index_for(end-1); assert(start ==_array->address_for_index(start_card), "Precondition"); assert(end ==_array->address_for_index(end_card)+N_words, "Precondition"); - set_remainder_to_point_to_start_incl(start_card, end_card); // closed interval + set_remainder_to_point_to_start_incl(start_card, end_card, reducing); // closed interval } @@ -175,7 +176,9 @@ set_remainder_to_point_to_start(HeapWord* start, HeapWord* end) { // a closed, inclusive interval: [start_card, end_card], cf set_remainder_to_point_to_start() // above. void -BlockOffsetArray::set_remainder_to_point_to_start_incl(size_t start_card, size_t end_card) { +BlockOffsetArray::set_remainder_to_point_to_start_incl(size_t start_card, size_t end_card, bool reducing) { + + check_reducing_assertion(reducing); if (start_card > end_card) { return; } @@ -191,11 +194,11 @@ BlockOffsetArray::set_remainder_to_point_to_start_incl(size_t start_card, size_t size_t reach = start_card - 1 + (power_to_cards_back(i+1) - 1); offset = N_words + i; if (reach >= end_card) { - _array->set_offset_array(start_card_for_region, end_card, offset); + _array->set_offset_array(start_card_for_region, end_card, offset, reducing); start_card_for_region = reach + 1; break; } - _array->set_offset_array(start_card_for_region, reach, offset); + _array->set_offset_array(start_card_for_region, reach, offset, reducing); start_card_for_region = reach + 1; } assert(start_card_for_region > end_card, "Sanity check"); @@ -211,8 +214,10 @@ void BlockOffsetArray::check_all_cards(size_t start_card, size_t end_card) const return; } guarantee(_array->offset_array(start_card) == N_words, "Wrong value in second card"); + u_char last_entry = N_words; for (size_t c = start_card + 1; c <= end_card; c++ /* yeah! */) { u_char entry = _array->offset_array(c); + guarantee(entry >= last_entry, "Monotonicity"); if (c - start_card > power_to_cards_back(1)) { guarantee(entry > N_words, "Should be in logarithmic region"); } @@ -220,11 +225,13 @@ void BlockOffsetArray::check_all_cards(size_t start_card, size_t end_card) const size_t landing_card = c - backskip; guarantee(landing_card >= (start_card - 1), "Inv"); if (landing_card >= start_card) { - guarantee(_array->offset_array(landing_card) <= entry, "monotonicity"); + guarantee(_array->offset_array(landing_card) <= entry, "Monotonicity"); } else { - guarantee(landing_card == start_card - 1, "Tautology"); + guarantee(landing_card == (start_card - 1), "Tautology"); + // Note that N_words is the maximum offset value guarantee(_array->offset_array(landing_card) <= N_words, "Offset value"); } + last_entry = entry; // remember for monotonicity test } } @@ -243,7 +250,7 @@ BlockOffsetArray::alloc_block(HeapWord* blk_start, HeapWord* blk_end) { void BlockOffsetArray::do_block_internal(HeapWord* blk_start, HeapWord* blk_end, - Action action) { + Action action, bool reducing) { assert(Universe::heap()->is_in_reserved(blk_start), "reference must be into the heap"); assert(Universe::heap()->is_in_reserved(blk_end-1), @@ -275,18 +282,18 @@ BlockOffsetArray::do_block_internal(HeapWord* blk_start, switch (action) { case Action_mark: { if (init_to_zero()) { - _array->set_offset_array(start_index, boundary, blk_start); + _array->set_offset_array(start_index, boundary, blk_start, reducing); break; } // Else fall through to the next case } case Action_single: { - _array->set_offset_array(start_index, boundary, blk_start); + _array->set_offset_array(start_index, boundary, blk_start, reducing); // We have finished marking the "offset card". We need to now // mark the subsequent cards that this blk spans. if (start_index < end_index) { HeapWord* rem_st = _array->address_for_index(start_index) + N_words; HeapWord* rem_end = _array->address_for_index(end_index) + N_words; - set_remainder_to_point_to_start(rem_st, rem_end); + set_remainder_to_point_to_start(rem_st, rem_end, reducing); } break; } @@ -395,7 +402,7 @@ void BlockOffsetArrayNonContigSpace::split_block(HeapWord* blk, // Indices for starts of prefix block and suffix block. size_t pref_index = _array->index_for(pref_addr); if (_array->address_for_index(pref_index) != pref_addr) { - // pref_addr deos not begin pref_index + // pref_addr does not begin pref_index pref_index++; } @@ -430,18 +437,18 @@ void BlockOffsetArrayNonContigSpace::split_block(HeapWord* blk, if (num_suff_cards > 0) { HeapWord* boundary = _array->address_for_index(suff_index); // Set the offset card for suffix block - _array->set_offset_array(suff_index, boundary, suff_addr); + _array->set_offset_array(suff_index, boundary, suff_addr, true /* reducing */); // Change any further cards that need changing in the suffix if (num_pref_cards > 0) { if (num_pref_cards >= num_suff_cards) { // Unilaterally fix all of the suffix cards: closed card // index interval in args below. - set_remainder_to_point_to_start_incl(suff_index + 1, end_index - 1); + set_remainder_to_point_to_start_incl(suff_index + 1, end_index - 1, true /* reducing */); } else { // Unilaterally fix the first (num_pref_cards - 1) following // the "offset card" in the suffix block. set_remainder_to_point_to_start_incl(suff_index + 1, - suff_index + num_pref_cards - 1); + suff_index + num_pref_cards - 1, true /* reducing */); // Fix the appropriate cards in the remainder of the // suffix block -- these are the last num_pref_cards // cards in each power block of the "new" range plumbed @@ -461,7 +468,7 @@ void BlockOffsetArrayNonContigSpace::split_block(HeapWord* blk, // is non-null. if (left_index <= right_index) { _array->set_offset_array(left_index, right_index, - N_words + i - 1); + N_words + i - 1, true /* reducing */); } else { more = false; // we are done } @@ -482,7 +489,7 @@ void BlockOffsetArrayNonContigSpace::split_block(HeapWord* blk, more = false; } assert(left_index <= right_index, "Error"); - _array->set_offset_array(left_index, right_index, N_words + i - 1); + _array->set_offset_array(left_index, right_index, N_words + i - 1, true /* reducing */); i++; } } @@ -501,14 +508,13 @@ void BlockOffsetArrayNonContigSpace::split_block(HeapWord* blk, // any cards subsequent to the first one. void BlockOffsetArrayNonContigSpace::mark_block(HeapWord* blk_start, - HeapWord* blk_end) { - do_block_internal(blk_start, blk_end, Action_mark); + HeapWord* blk_end, bool reducing) { + do_block_internal(blk_start, blk_end, Action_mark, reducing); } HeapWord* BlockOffsetArrayNonContigSpace::block_start_unsafe( const void* addr) const { assert(_array->offset_array(0) == 0, "objects can't cross covered areas"); - assert(_bottom <= addr && addr < _end, "addr must be covered by this Array"); // Must read this exactly once because it can be modified by parallel @@ -542,9 +548,10 @@ HeapWord* BlockOffsetArrayNonContigSpace::block_start_unsafe( debug_only(HeapWord* last = q); // for debugging q = n; n += _sp->block_size(n); + assert(n > q, err_msg("Looping at: " INTPTR_FORMAT, n)); } - assert(q <= addr, "wrong order for current and arg"); - assert(addr <= n, "wrong order for arg and next"); + assert(q <= addr, err_msg("wrong order for current (" INTPTR_FORMAT ") <= arg (" INTPTR_FORMAT ")", q, addr)); + assert(addr <= n, err_msg("wrong order for arg (" INTPTR_FORMAT ") <= next (" INTPTR_FORMAT ")", addr, n)); return q; } @@ -727,9 +734,8 @@ void BlockOffsetArrayContigSpace::alloc_block_work(HeapWord* blk_start, _next_offset_index = end_index + 1; // Calculate _next_offset_threshold this way because end_index // may be the last valid index in the covered region. - _next_offset_threshold = _array->address_for_index(end_index) + - N_words; - assert(_next_offset_threshold >= blk_end, "Incorrent offset threshold"); + _next_offset_threshold = _array->address_for_index(end_index) + N_words; + assert(_next_offset_threshold >= blk_end, "Incorrect offset threshold"); #ifdef ASSERT // The offset can be 0 if the block starts on a boundary. That diff --git a/src/share/vm/memory/blockOffsetTable.hpp b/src/share/vm/memory/blockOffsetTable.hpp index db9461c5e..c0c78149a 100644 --- a/src/share/vm/memory/blockOffsetTable.hpp +++ b/src/share/vm/memory/blockOffsetTable.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -107,6 +107,8 @@ class BlockOffsetSharedArray: public CHeapObj { N_words = 1 << LogN_words }; + bool _init_to_zero; + // The reserved region covered by the shared array. MemRegion _reserved; @@ -125,17 +127,28 @@ class BlockOffsetSharedArray: public CHeapObj { assert(index < _vs.committed_size(), "index out of range"); return _offset_array[index]; } - void set_offset_array(size_t index, u_char offset) { + // An assertion-checking helper method for the set_offset_array() methods below. + void check_reducing_assertion(bool reducing); + + void set_offset_array(size_t index, u_char offset, bool reducing = false) { + check_reducing_assertion(reducing); assert(index < _vs.committed_size(), "index out of range"); + assert(!reducing || _offset_array[index] >= offset, "Not reducing"); _offset_array[index] = offset; } - void set_offset_array(size_t index, HeapWord* high, HeapWord* low) { + + void set_offset_array(size_t index, HeapWord* high, HeapWord* low, bool reducing = false) { + check_reducing_assertion(reducing); assert(index < _vs.committed_size(), "index out of range"); assert(high >= low, "addresses out of order"); assert(pointer_delta(high, low) <= N_words, "offset too large"); + assert(!reducing || _offset_array[index] >= (u_char)pointer_delta(high, low), + "Not reducing"); _offset_array[index] = (u_char)pointer_delta(high, low); } - void set_offset_array(HeapWord* left, HeapWord* right, u_char offset) { + + void set_offset_array(HeapWord* left, HeapWord* right, u_char offset, bool reducing = false) { + check_reducing_assertion(reducing); assert(index_for(right - 1) < _vs.committed_size(), "right address out of range"); assert(left < right, "Heap addresses out of order"); @@ -150,12 +163,15 @@ class BlockOffsetSharedArray: public CHeapObj { size_t i = index_for(left); const size_t end = i + num_cards; for (; i < end; i++) { + // Elided until CR 6977974 is fixed properly. + // assert(!reducing || _offset_array[i] >= offset, "Not reducing"); _offset_array[i] = offset; } } } - void set_offset_array(size_t left, size_t right, u_char offset) { + void set_offset_array(size_t left, size_t right, u_char offset, bool reducing = false) { + check_reducing_assertion(reducing); assert(right < _vs.committed_size(), "right address out of range"); assert(left <= right, "indexes out of order"); size_t num_cards = right - left + 1; @@ -169,6 +185,8 @@ class BlockOffsetSharedArray: public CHeapObj { size_t i = left; const size_t end = i + num_cards; for (; i < end; i++) { + // Elided until CR 6977974 is fixed properly. + // assert(!reducing || _offset_array[i] >= offset, "Not reducing"); _offset_array[i] = offset; } } @@ -212,6 +230,11 @@ public: void set_bottom(HeapWord* new_bottom); + // Whether entries should be initialized to zero. Used currently only for + // error checking. + void set_init_to_zero(bool val) { _init_to_zero = val; } + bool init_to_zero() { return _init_to_zero; } + // Updates all the BlockOffsetArray's sharing this shared array to // reflect the current "top"'s of their spaces. void update_offset_arrays(); // Not yet implemented! @@ -285,17 +308,23 @@ class BlockOffsetArray: public BlockOffsetTable { // initialized to point backwards to the beginning of the covered region. bool _init_to_zero; + // An assertion-checking helper method for the set_remainder*() methods below. + void check_reducing_assertion(bool reducing) { _array->check_reducing_assertion(reducing); } + // Sets the entries // corresponding to the cards starting at "start" and ending at "end" // to point back to the card before "start": the interval [start, end) - // is right-open. - void set_remainder_to_point_to_start(HeapWord* start, HeapWord* end); + // is right-open. The last parameter, reducing, indicates whether the + // updates to individual entries always reduce the entry from a higher + // to a lower value. (For example this would hold true during a temporal + // regime during which only block splits were updating the BOT. + void set_remainder_to_point_to_start(HeapWord* start, HeapWord* end, bool reducing = false); // Same as above, except that the args here are a card _index_ interval // that is closed: [start_index, end_index] - void set_remainder_to_point_to_start_incl(size_t start, size_t end); + void set_remainder_to_point_to_start_incl(size_t start, size_t end, bool reducing = false); // A helper function for BOT adjustment/verification work - void do_block_internal(HeapWord* blk_start, HeapWord* blk_end, Action action); + void do_block_internal(HeapWord* blk_start, HeapWord* blk_end, Action action, bool reducing = false); public: // The space may not have its bottom and top set yet, which is why the @@ -303,7 +332,7 @@ class BlockOffsetArray: public BlockOffsetTable { // elements of the array are initialized to zero. Otherwise, they are // initialized to point backwards to the beginning. BlockOffsetArray(BlockOffsetSharedArray* array, MemRegion mr, - bool init_to_zero); + bool init_to_zero_); // Note: this ought to be part of the constructor, but that would require // "this" to be passed as a parameter to a member constructor for @@ -358,6 +387,12 @@ class BlockOffsetArray: public BlockOffsetTable { // If true, initialize array slots with no allocated blocks to zero. // Otherwise, make them point back to the front. bool init_to_zero() { return _init_to_zero; } + // Corresponding setter + void set_init_to_zero(bool val) { + _init_to_zero = val; + assert(_array != NULL, "_array should be non-NULL"); + _array->set_init_to_zero(val); + } // Debugging // Return the index of the last entry in the "active" region. @@ -424,16 +459,16 @@ class BlockOffsetArrayNonContigSpace: public BlockOffsetArray { // of BOT is touched. It is assumed (and verified in the // non-product VM) that the remaining cards of the block // are correct. - void mark_block(HeapWord* blk_start, HeapWord* blk_end); - void mark_block(HeapWord* blk, size_t size) { - mark_block(blk, blk + size); + void mark_block(HeapWord* blk_start, HeapWord* blk_end, bool reducing = false); + void mark_block(HeapWord* blk, size_t size, bool reducing = false) { + mark_block(blk, blk + size, reducing); } // Adjust _unallocated_block to indicate that a particular // block has been newly allocated or freed. It is assumed (and // verified in the non-product VM) that the BOT is correct for // the given block. - void allocated(HeapWord* blk_start, HeapWord* blk_end) { + void allocated(HeapWord* blk_start, HeapWord* blk_end, bool reducing = false) { // Verify that the BOT shows [blk, blk + blk_size) to be one block. verify_single_block(blk_start, blk_end); if (BlockOffsetArrayUseUnallocatedBlock) { @@ -441,14 +476,12 @@ class BlockOffsetArrayNonContigSpace: public BlockOffsetArray { } } - void allocated(HeapWord* blk, size_t size) { - allocated(blk, blk + size); + void allocated(HeapWord* blk, size_t size, bool reducing = false) { + allocated(blk, blk + size, reducing); } void freed(HeapWord* blk_start, HeapWord* blk_end); - void freed(HeapWord* blk, size_t size) { - freed(blk, blk + size); - } + void freed(HeapWord* blk, size_t size); HeapWord* block_start_unsafe(const void* addr) const; @@ -456,7 +489,6 @@ class BlockOffsetArrayNonContigSpace: public BlockOffsetArray { // start of the block that contains the given address. HeapWord* block_start_careful(const void* addr) const; - // Verification & debugging: ensure that the offset table reflects // the fact that the block [blk_start, blk_end) or [blk, blk + size) // is a single block of storage. NOTE: can't const this because of diff --git a/src/share/vm/memory/blockOffsetTable.inline.hpp b/src/share/vm/memory/blockOffsetTable.inline.hpp index e9bee7907..39afcea87 100644 --- a/src/share/vm/memory/blockOffsetTable.inline.hpp +++ b/src/share/vm/memory/blockOffsetTable.inline.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000, 2002, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -55,10 +55,22 @@ inline HeapWord* BlockOffsetSharedArray::address_for_index(size_t index) const { return result; } +inline void BlockOffsetSharedArray::check_reducing_assertion(bool reducing) { + assert(reducing || !SafepointSynchronize::is_at_safepoint() || init_to_zero() || + Thread::current()->is_VM_thread() || + Thread::current()->is_ConcurrentGC_thread() || + ((!Thread::current()->is_ConcurrentGC_thread()) && + ParGCRareEvent_lock->owned_by_self()), "Crack"); +} ////////////////////////////////////////////////////////////////////////// // BlockOffsetArrayNonContigSpace inlines ////////////////////////////////////////////////////////////////////////// +inline void BlockOffsetArrayNonContigSpace::freed(HeapWord* blk, + size_t size) { + freed(blk, blk + size); +} + inline void BlockOffsetArrayNonContigSpace::freed(HeapWord* blk_start, HeapWord* blk_end) { // Verify that the BOT shows [blk_start, blk_end) to be one block. diff --git a/src/share/vm/oops/arrayKlassKlass.cpp b/src/share/vm/oops/arrayKlassKlass.cpp index 2128f61f9..b4b37b3ef 100644 --- a/src/share/vm/oops/arrayKlassKlass.cpp +++ b/src/share/vm/oops/arrayKlassKlass.cpp @@ -108,10 +108,6 @@ int arrayKlassKlass::oop_oop_iterate_m(oop obj, OopClosure* blk, MemRegion mr) { } #ifndef SERIALGC -void arrayKlassKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - assert(obj->blueprint()->oop_is_arrayKlass(),"must be an array klass"); -} - void arrayKlassKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { assert(obj->blueprint()->oop_is_arrayKlass(),"must be an array klass"); } diff --git a/src/share/vm/oops/compiledICHolderKlass.cpp b/src/share/vm/oops/compiledICHolderKlass.cpp index 800b7eba9..035d86e33 100644 --- a/src/share/vm/oops/compiledICHolderKlass.cpp +++ b/src/share/vm/oops/compiledICHolderKlass.cpp @@ -120,10 +120,6 @@ int compiledICHolderKlass::oop_adjust_pointers(oop obj) { } #ifndef SERIALGC -void compiledICHolderKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - assert(obj->is_compiledICHolder(), "must be compiledICHolder"); -} - void compiledICHolderKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { assert(obj->is_compiledICHolder(), "must be compiledICHolder"); } diff --git a/src/share/vm/oops/constMethodKlass.cpp b/src/share/vm/oops/constMethodKlass.cpp index b5f15e5f6..114dea85b 100644 --- a/src/share/vm/oops/constMethodKlass.cpp +++ b/src/share/vm/oops/constMethodKlass.cpp @@ -157,10 +157,6 @@ int constMethodKlass::oop_adjust_pointers(oop obj) { } #ifndef SERIALGC -void constMethodKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - assert(obj->is_constMethod(), "should be constMethod"); -} - void constMethodKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { assert(obj->is_constMethod(), "should be constMethod"); } diff --git a/src/share/vm/oops/constantPoolKlass.cpp b/src/share/vm/oops/constantPoolKlass.cpp index 7ca3c79b1..656ce16b6 100644 --- a/src/share/vm/oops/constantPoolKlass.cpp +++ b/src/share/vm/oops/constantPoolKlass.cpp @@ -268,21 +268,6 @@ constantPoolKlass::oop_update_pointers(ParCompactionManager* cm, oop obj, return cp->object_size(); } -void constantPoolKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - assert(obj->is_constantPool(), "should be constant pool"); - constantPoolOop cp = (constantPoolOop) obj; - if (AnonymousClasses && cp->has_pseudo_string() && cp->tags() != NULL) { - oop* base = (oop*)cp->base(); - for (int i = 0; i < cp->length(); ++i, ++base) { - if (cp->tag_at(i).is_string()) { - if (PSScavenge::should_scavenge(base)) { - pm->claim_or_forward_breadth(base); - } - } - } - } -} - void constantPoolKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { assert(obj->is_constantPool(), "should be constant pool"); constantPoolOop cp = (constantPoolOop) obj; diff --git a/src/share/vm/oops/cpCacheKlass.cpp b/src/share/vm/oops/cpCacheKlass.cpp index 3d7916ffa..216c62ee4 100644 --- a/src/share/vm/oops/cpCacheKlass.cpp +++ b/src/share/vm/oops/cpCacheKlass.cpp @@ -166,29 +166,6 @@ bool constantPoolCacheKlass::oop_is_conc_safe(oop obj) const { } #ifndef SERIALGC -void constantPoolCacheKlass::oop_copy_contents(PSPromotionManager* pm, - oop obj) { - assert(obj->is_constantPoolCache(), "should be constant pool"); - if (EnableInvokeDynamic) { - constantPoolCacheOop cache = (constantPoolCacheOop)obj; - // during a scavenge, it is safe to inspect my pool, since it is perm - constantPoolOop pool = cache->constant_pool(); - assert(pool->is_constantPool(), "should be constant pool"); - if (pool->has_invokedynamic()) { - for (int i = 0; i < cache->length(); i++) { - ConstantPoolCacheEntry* e = cache->entry_at(i); - oop* p = (oop*)&e->_f1; - if (e->is_secondary_entry()) { - if (PSScavenge::should_scavenge(p)) - pm->claim_or_forward_breadth(p); - assert(!(e->is_vfinal() && PSScavenge::should_scavenge((oop*)&e->_f2)), - "no live oops here"); - } - } - } - } -} - void constantPoolCacheKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { assert(obj->is_constantPoolCache(), "should be constant pool"); diff --git a/src/share/vm/oops/instanceKlass.cpp b/src/share/vm/oops/instanceKlass.cpp index 87bbddc5d..ff8077f1b 100644 --- a/src/share/vm/oops/instanceKlass.cpp +++ b/src/share/vm/oops/instanceKlass.cpp @@ -1809,18 +1809,7 @@ int instanceKlass::oop_adjust_pointers(oop obj) { } #ifndef SERIALGC -void instanceKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - assert(!pm->depth_first(), "invariant"); - InstanceKlass_OOP_MAP_REVERSE_ITERATE( \ - obj, \ - if (PSScavenge::should_scavenge(p)) { \ - pm->claim_or_forward_breadth(p); \ - }, \ - assert_nothing ) -} - void instanceKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { - assert(pm->depth_first(), "invariant"); InstanceKlass_OOP_MAP_REVERSE_ITERATE( \ obj, \ if (PSScavenge::should_scavenge(p)) { \ @@ -1846,18 +1835,7 @@ int instanceKlass::oop_update_pointers(ParCompactionManager* cm, oop obj, return size_helper(); } -void instanceKlass::copy_static_fields(PSPromotionManager* pm) { - assert(!pm->depth_first(), "invariant"); - InstanceKlass_OOP_ITERATE( \ - start_of_static_fields(), static_oop_field_size(), \ - if (PSScavenge::should_scavenge(p)) { \ - pm->claim_or_forward_breadth(p); \ - }, \ - assert_nothing ) -} - void instanceKlass::push_static_fields(PSPromotionManager* pm) { - assert(pm->depth_first(), "invariant"); InstanceKlass_OOP_ITERATE( \ start_of_static_fields(), static_oop_field_size(), \ if (PSScavenge::should_scavenge(p)) { \ diff --git a/src/share/vm/oops/instanceKlass.hpp b/src/share/vm/oops/instanceKlass.hpp index c25b84a36..d895ce108 100644 --- a/src/share/vm/oops/instanceKlass.hpp +++ b/src/share/vm/oops/instanceKlass.hpp @@ -711,7 +711,6 @@ class instanceKlass: public Klass { #ifndef SERIALGC // Parallel Scavenge - void copy_static_fields(PSPromotionManager* pm); void push_static_fields(PSPromotionManager* pm); // Parallel Old diff --git a/src/share/vm/oops/instanceKlassKlass.cpp b/src/share/vm/oops/instanceKlassKlass.cpp index 4184ce029..74268925e 100644 --- a/src/share/vm/oops/instanceKlassKlass.cpp +++ b/src/share/vm/oops/instanceKlassKlass.cpp @@ -292,41 +292,7 @@ int instanceKlassKlass::oop_adjust_pointers(oop obj) { } #ifndef SERIALGC -void instanceKlassKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - assert(!pm->depth_first(), "invariant"); - instanceKlass* ik = instanceKlass::cast(klassOop(obj)); - ik->copy_static_fields(pm); - - oop* loader_addr = ik->adr_class_loader(); - if (PSScavenge::should_scavenge(loader_addr)) { - pm->claim_or_forward_breadth(loader_addr); - } - - oop* pd_addr = ik->adr_protection_domain(); - if (PSScavenge::should_scavenge(pd_addr)) { - pm->claim_or_forward_breadth(pd_addr); - } - - oop* hk_addr = ik->adr_host_klass(); - if (PSScavenge::should_scavenge(hk_addr)) { - pm->claim_or_forward_breadth(hk_addr); - } - - oop* sg_addr = ik->adr_signers(); - if (PSScavenge::should_scavenge(sg_addr)) { - pm->claim_or_forward_breadth(sg_addr); - } - - oop* bsm_addr = ik->adr_bootstrap_method(); - if (PSScavenge::should_scavenge(bsm_addr)) { - pm->claim_or_forward_breadth(bsm_addr); - } - - klassKlass::oop_copy_contents(pm, obj); -} - void instanceKlassKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { - assert(pm->depth_first(), "invariant"); instanceKlass* ik = instanceKlass::cast(klassOop(obj)); ik->push_static_fields(pm); @@ -355,7 +321,7 @@ void instanceKlassKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { pm->claim_or_forward_depth(bsm_addr); } - klassKlass::oop_copy_contents(pm, obj); + klassKlass::oop_push_contents(pm, obj); } int instanceKlassKlass::oop_update_pointers(ParCompactionManager* cm, oop obj) { diff --git a/src/share/vm/oops/instanceRefKlass.cpp b/src/share/vm/oops/instanceRefKlass.cpp index f8670f992..022827f30 100644 --- a/src/share/vm/oops/instanceRefKlass.cpp +++ b/src/share/vm/oops/instanceRefKlass.cpp @@ -273,41 +273,8 @@ ALL_OOP_OOP_ITERATE_CLOSURES_2(InstanceRefKlass_OOP_OOP_ITERATE_DEFN_m) #ifndef SERIALGC template <class T> -void specialized_oop_copy_contents(instanceRefKlass *ref, - PSPromotionManager* pm, oop obj) { - assert(!pm->depth_first(), "invariant"); - T* referent_addr = (T*)java_lang_ref_Reference::referent_addr(obj); - if (PSScavenge::should_scavenge(referent_addr)) { - ReferenceProcessor* rp = PSScavenge::reference_processor(); - if (rp->discover_reference(obj, ref->reference_type())) { - // reference already enqueued, referent and next will be traversed later - ref->instanceKlass::oop_copy_contents(pm, obj); - return; - } else { - // treat referent as normal oop - pm->claim_or_forward_breadth(referent_addr); - } - } - // treat next as normal oop - T* next_addr = (T*)java_lang_ref_Reference::next_addr(obj); - if (PSScavenge::should_scavenge(next_addr)) { - pm->claim_or_forward_breadth(next_addr); - } - ref->instanceKlass::oop_copy_contents(pm, obj); -} - -void instanceRefKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - if (UseCompressedOops) { - specialized_oop_copy_contents<narrowOop>(this, pm, obj); - } else { - specialized_oop_copy_contents<oop>(this, pm, obj); - } -} - -template <class T> void specialized_oop_push_contents(instanceRefKlass *ref, PSPromotionManager* pm, oop obj) { - assert(pm->depth_first(), "invariant"); T* referent_addr = (T*)java_lang_ref_Reference::referent_addr(obj); if (PSScavenge::should_scavenge(referent_addr)) { ReferenceProcessor* rp = PSScavenge::reference_processor(); diff --git a/src/share/vm/oops/klassKlass.cpp b/src/share/vm/oops/klassKlass.cpp index 74a352fda..b1f0225d7 100644 --- a/src/share/vm/oops/klassKlass.cpp +++ b/src/share/vm/oops/klassKlass.cpp @@ -161,9 +161,6 @@ int klassKlass::oop_adjust_pointers(oop obj) { } #ifndef SERIALGC -void klassKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { -} - void klassKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { } diff --git a/src/share/vm/oops/klassPS.hpp b/src/share/vm/oops/klassPS.hpp index c53abf770..6c02905db 100644 --- a/src/share/vm/oops/klassPS.hpp +++ b/src/share/vm/oops/klassPS.hpp @@ -28,7 +28,6 @@ #ifndef SERIALGC #define PARALLEL_GC_DECLS \ - virtual void oop_copy_contents(PSPromotionManager* pm, oop obj); \ virtual void oop_push_contents(PSPromotionManager* pm, oop obj); \ /* Parallel Old GC support \ \ @@ -43,7 +42,6 @@ // Pure virtual version for klass.hpp #define PARALLEL_GC_DECLS_PV \ - virtual void oop_copy_contents(PSPromotionManager* pm, oop obj) = 0; \ virtual void oop_push_contents(PSPromotionManager* pm, oop obj) = 0; \ virtual void oop_follow_contents(ParCompactionManager* cm, oop obj) = 0; \ virtual int oop_update_pointers(ParCompactionManager* cm, oop obj) = 0; \ diff --git a/src/share/vm/oops/methodDataKlass.cpp b/src/share/vm/oops/methodDataKlass.cpp index 03b5bd887..e11f6afd6 100644 --- a/src/share/vm/oops/methodDataKlass.cpp +++ b/src/share/vm/oops/methodDataKlass.cpp @@ -154,13 +154,6 @@ int methodDataKlass::oop_adjust_pointers(oop obj) { #ifndef SERIALGC -void methodDataKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - assert (obj->is_methodData(), "object must be method data"); - methodDataOop m = methodDataOop(obj); - // This should never point into the young gen. - assert(!PSScavenge::should_scavenge(m->adr_method()), "Sanity"); -} - void methodDataKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { assert (obj->is_methodData(), "object must be method data"); methodDataOop m = methodDataOop(obj); diff --git a/src/share/vm/oops/methodKlass.cpp b/src/share/vm/oops/methodKlass.cpp index f0ba4ad84..40677d192 100644 --- a/src/share/vm/oops/methodKlass.cpp +++ b/src/share/vm/oops/methodKlass.cpp @@ -184,10 +184,6 @@ int methodKlass::oop_adjust_pointers(oop obj) { } #ifndef SERIALGC -void methodKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - assert(obj->is_method(), "should be method"); -} - void methodKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { assert(obj->is_method(), "should be method"); } diff --git a/src/share/vm/oops/objArrayKlass.cpp b/src/share/vm/oops/objArrayKlass.cpp index 96935b614..a932a117d 100644 --- a/src/share/vm/oops/objArrayKlass.cpp +++ b/src/share/vm/oops/objArrayKlass.cpp @@ -426,18 +426,7 @@ int objArrayKlass::oop_adjust_pointers(oop obj) { } #ifndef SERIALGC -void objArrayKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - assert(!pm->depth_first(), "invariant"); - assert(obj->is_objArray(), "obj must be obj array"); - ObjArrayKlass_OOP_ITERATE( \ - objArrayOop(obj), p, \ - if (PSScavenge::should_scavenge(p)) { \ - pm->claim_or_forward_breadth(p); \ - }) -} - void objArrayKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { - assert(pm->depth_first(), "invariant"); assert(obj->is_objArray(), "obj must be obj array"); ObjArrayKlass_OOP_ITERATE( \ objArrayOop(obj), p, \ diff --git a/src/share/vm/oops/objArrayKlassKlass.cpp b/src/share/vm/oops/objArrayKlassKlass.cpp index f486dc7db..7050d57b1 100644 --- a/src/share/vm/oops/objArrayKlassKlass.cpp +++ b/src/share/vm/oops/objArrayKlassKlass.cpp @@ -229,10 +229,6 @@ objArrayKlassKlass::oop_oop_iterate_m(oop obj, OopClosure* blk, MemRegion mr) { } #ifndef SERIALGC -void objArrayKlassKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - assert(obj->blueprint()->oop_is_objArrayKlass(),"must be an obj array klass"); -} - void objArrayKlassKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { assert(obj->blueprint()->oop_is_objArrayKlass(),"must be an obj array klass"); } diff --git a/src/share/vm/oops/oop.hpp b/src/share/vm/oops/oop.hpp index 952802c78..04d311804 100644 --- a/src/share/vm/oops/oop.hpp +++ b/src/share/vm/oops/oop.hpp @@ -306,7 +306,6 @@ class oopDesc { #ifndef SERIALGC // Parallel Scavenge - void copy_contents(PSPromotionManager* pm); void push_contents(PSPromotionManager* pm); // Parallel Old diff --git a/src/share/vm/oops/oop.psgc.inline.hpp b/src/share/vm/oops/oop.psgc.inline.hpp index 07ad5c5a6..40d40a372 100644 --- a/src/share/vm/oops/oop.psgc.inline.hpp +++ b/src/share/vm/oops/oop.psgc.inline.hpp @@ -24,15 +24,6 @@ // ParallelScavengeHeap methods -inline void oopDesc::copy_contents(PSPromotionManager* pm) { - Klass* klass = blueprint(); - if (!klass->oop_is_typeArray()) { - // It might contain oops beyond the header, so take the virtual call. - klass->oop_copy_contents(pm, this); - } - // Else skip it. The typeArrayKlass in the header never needs scavenging. -} - inline void oopDesc::push_contents(PSPromotionManager* pm) { Klass* klass = blueprint(); if (!klass->oop_is_typeArray()) { diff --git a/src/share/vm/oops/symbolKlass.cpp b/src/share/vm/oops/symbolKlass.cpp index 4fe4dc860..e3806db12 100644 --- a/src/share/vm/oops/symbolKlass.cpp +++ b/src/share/vm/oops/symbolKlass.cpp @@ -184,10 +184,6 @@ int symbolKlass::oop_adjust_pointers(oop obj) { #ifndef SERIALGC -void symbolKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - assert(obj->is_symbol(), "should be symbol"); -} - void symbolKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { assert(obj->is_symbol(), "should be symbol"); } diff --git a/src/share/vm/oops/typeArrayKlass.cpp b/src/share/vm/oops/typeArrayKlass.cpp index edb216e68..088960ce0 100644 --- a/src/share/vm/oops/typeArrayKlass.cpp +++ b/src/share/vm/oops/typeArrayKlass.cpp @@ -228,10 +228,6 @@ int typeArrayKlass::oop_oop_iterate_m(oop obj, OopClosure* blk, MemRegion mr) { } #ifndef SERIALGC -void typeArrayKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) { - assert(obj->is_typeArray(),"must be a type array"); -} - void typeArrayKlass::oop_push_contents(PSPromotionManager* pm, oop obj) { assert(obj->is_typeArray(),"must be a type array"); } diff --git a/src/share/vm/opto/c2_globals.hpp b/src/share/vm/opto/c2_globals.hpp index 3fdaf8427..e07e7fdec 100644 --- a/src/share/vm/opto/c2_globals.hpp +++ b/src/share/vm/opto/c2_globals.hpp @@ -178,6 +178,9 @@ product(bool, ReduceBulkZeroing, true, \ "When bulk-initializing, try to avoid needless zeroing") \ \ + product(bool, UseFPUForSpilling, false, \ + "Spill integer registers to FPU instead of stack when possible") \ + \ develop_pd(intx, RegisterCostAreaRatio, \ "Spill selection in reg allocator: scale area by (X/64K) before " \ "adding cost") \ diff --git a/src/share/vm/opto/coalesce.cpp b/src/share/vm/opto/coalesce.cpp index 311b55a71..2144b59ec 100644 --- a/src/share/vm/opto/coalesce.cpp +++ b/src/share/vm/opto/coalesce.cpp @@ -780,6 +780,14 @@ bool PhaseConservativeCoalesce::copy_copy( Node *dst_copy, Node *src_copy, Block // Number of bits free uint rm_size = rm.Size(); + if (UseFPUForSpilling && rm.is_AllStack() ) { + // Don't coalesce when frequency difference is large + Block *dst_b = _phc._cfg._bbs[dst_copy->_idx]; + Block *src_def_b = _phc._cfg._bbs[src_def->_idx]; + if (src_def_b->_freq > 10*dst_b->_freq ) + return false; + } + // If we can use any stack slot, then effective size is infinite if( rm.is_AllStack() ) rm_size += 1000000; // Incompatible masks, no way to coalesce diff --git a/src/share/vm/opto/matcher.cpp b/src/share/vm/opto/matcher.cpp index 78931baa4..94178813e 100644 --- a/src/share/vm/opto/matcher.cpp +++ b/src/share/vm/opto/matcher.cpp @@ -456,6 +456,23 @@ void Matcher::init_first_stack_mask() { *idealreg2spillmask[Op_RegP] = *idealreg2regmask[Op_RegP]; idealreg2spillmask[Op_RegP]->OR(C->FIRST_STACK_mask()); + if (UseFPUForSpilling) { + // This mask logic assumes that the spill operations are + // symmetric and that the registers involved are the same size. + // On sparc for instance we may have to use 64 bit moves will + // kill 2 registers when used with F0-F31. + idealreg2spillmask[Op_RegI]->OR(*idealreg2regmask[Op_RegF]); + idealreg2spillmask[Op_RegF]->OR(*idealreg2regmask[Op_RegI]); +#ifdef _LP64 + idealreg2spillmask[Op_RegN]->OR(*idealreg2regmask[Op_RegF]); + idealreg2spillmask[Op_RegL]->OR(*idealreg2regmask[Op_RegD]); + idealreg2spillmask[Op_RegD]->OR(*idealreg2regmask[Op_RegL]); + idealreg2spillmask[Op_RegP]->OR(*idealreg2regmask[Op_RegD]); +#else + idealreg2spillmask[Op_RegP]->OR(*idealreg2regmask[Op_RegF]); +#endif + } + // Make up debug masks. Any spill slot plus callee-save registers. // Caller-save registers are assumed to be trashable by the various // inline-cache fixup routines. diff --git a/src/share/vm/opto/reg_split.cpp b/src/share/vm/opto/reg_split.cpp index a4fee6f7c..9e3967e36 100644 --- a/src/share/vm/opto/reg_split.cpp +++ b/src/share/vm/opto/reg_split.cpp @@ -975,6 +975,19 @@ uint PhaseChaitin::Split( uint maxlrg ) { insidx++; // Reset iterator to skip USE side split continue; } + + if (UseFPUForSpilling && n->is_Call() && !uup && !dup ) { + // The use at the call can force the def down so insert + // a split before the use to allow the def more freedom. + maxlrg = split_USE(def,b,n,inpidx,maxlrg,dup,false, splits,slidx); + // If it wasn't split bail + if (!maxlrg) { + return 0; + } + insidx++; // Reset iterator to skip USE side split + continue; + } + // Here is the logic chart which describes USE Splitting: // 0 = false or DOWN, 1 = true or UP // diff --git a/src/share/vm/runtime/arguments.cpp b/src/share/vm/runtime/arguments.cpp index 48b612171..1e30d5022 100644 --- a/src/share/vm/runtime/arguments.cpp +++ b/src/share/vm/runtime/arguments.cpp @@ -184,6 +184,8 @@ static ObsoleteFlag obsolete_jvm_flags[] = { { "DefaultMaxRAM", JDK_Version::jdk_update(6,18), JDK_Version::jdk(7) }, { "DefaultInitialRAMFraction", JDK_Version::jdk_update(6,18), JDK_Version::jdk(7) }, + { "UseDepthFirstScavengeOrder", + JDK_Version::jdk_update(6,22), JDK_Version::jdk(7) }, { NULL, JDK_Version(0), JDK_Version(0) } }; @@ -3003,10 +3005,6 @@ jint Arguments::parse(const JavaVMInitArgs* args) { CommandLineFlags::printSetFlags(); } - if (PrintFlagsFinal) { - CommandLineFlags::printFlags(); - } - // Apply CPU specific policy for the BiasedLocking if (UseBiasedLocking) { if (!VM_Version::use_biased_locking() && diff --git a/src/share/vm/runtime/frame.cpp b/src/share/vm/runtime/frame.cpp index 89de3ca1a..69d287984 100644 --- a/src/share/vm/runtime/frame.cpp +++ b/src/share/vm/runtime/frame.cpp @@ -215,17 +215,15 @@ bool frame::can_be_deoptimized() const { return !nm->is_at_poll_return(pc()); } -void frame::deoptimize(JavaThread* thread, bool thread_is_known_safe) { -// Schedule deoptimization of an nmethod activation with this frame. - - // Store the original pc before an patch (or request to self-deopt) - // in the published location of the frame. - +void frame::deoptimize(JavaThread* thread) { + // Schedule deoptimization of an nmethod activation with this frame. assert(_cb != NULL && _cb->is_nmethod(), "must be"); nmethod* nm = (nmethod*)_cb; // This is a fix for register window patching race - if (NeedsDeoptSuspend && !thread_is_known_safe) { + if (NeedsDeoptSuspend && Thread::current() != thread) { + assert(SafepointSynchronize::is_at_safepoint(), + "patching other threads for deopt may only occur at a safepoint"); // It is possible especially with DeoptimizeALot/DeoptimizeRandom that // we could see the frame again and ask for it to be deoptimized since @@ -248,7 +246,11 @@ void frame::deoptimize(JavaThread* thread, bool thread_is_known_safe) { // whether to spin or block. It isn't worth it. Just treat it like // native and be done with it. // - JavaThreadState state = thread->thread_state(); + // Examine the state of the thread at the start of safepoint since + // threads that were in native at the start of the safepoint could + // come to a halt during the safepoint, changing the current value + // of the safepoint_state. + JavaThreadState state = thread->safepoint_state()->orig_thread_state(); if (state == _thread_in_native || state == _thread_in_native_trans) { // Since we are at a safepoint the target thread will stop itself // before it can return to java as long as we remain at the safepoint. diff --git a/src/share/vm/runtime/frame.hpp b/src/share/vm/runtime/frame.hpp index 4f2df2ad9..a2ac51f36 100644 --- a/src/share/vm/runtime/frame.hpp +++ b/src/share/vm/runtime/frame.hpp @@ -174,7 +174,7 @@ class frame VALUE_OBJ_CLASS_SPEC { address sender_pc() const; // Support for deoptimization - void deoptimize(JavaThread* thread, bool thread_is_known_safe = false); + void deoptimize(JavaThread* thread); // The frame's original SP, before any extension by an interpreted callee; // used for packing debug info into vframeArray objects and vframeArray lookup. diff --git a/src/share/vm/runtime/globals.hpp b/src/share/vm/runtime/globals.hpp index 28b26381f..92dde7768 100644 --- a/src/share/vm/runtime/globals.hpp +++ b/src/share/vm/runtime/globals.hpp @@ -1712,7 +1712,7 @@ class CommandLineFlags { develop(bool, VerifyBlockOffsetArray, false, \ "Do (expensive!) block offset array verification") \ \ - product(bool, BlockOffsetArrayUseUnallocatedBlock, trueInDebug, \ + product(bool, BlockOffsetArrayUseUnallocatedBlock, false, \ "Maintain _unallocated_block in BlockOffsetArray" \ " (currently applicable only to CMS collector)") \ \ @@ -3092,10 +3092,6 @@ class CommandLineFlags { \ product(intx, SafepointSpinBeforeYield, 2000, "(Unstable)") \ \ - product(bool, UseDepthFirstScavengeOrder, true, \ - "true: the scavenge order will be depth-first, " \ - "false: the scavenge order will be breadth-first") \ - \ product(bool, PSChunkLargeArrays, true, \ "true: process large arrays in chunks") \ \ diff --git a/src/share/vm/runtime/init.cpp b/src/share/vm/runtime/init.cpp index df3a8bd46..9eac44347 100644 --- a/src/share/vm/runtime/init.cpp +++ b/src/share/vm/runtime/init.cpp @@ -128,6 +128,12 @@ jint init_globals() { Universe::verify(); // make sure we're starting with a clean slate } + // All the flags that get adjusted by VM_Version_init and os::init_2 + // have been set so dump the flags now. + if (PrintFlagsFinal) { + CommandLineFlags::printFlags(); + } + return JNI_OK; } diff --git a/src/share/vm/runtime/orderAccess.cpp b/src/share/vm/runtime/orderAccess.cpp index b61c48160..12124f964 100644 --- a/src/share/vm/runtime/orderAccess.cpp +++ b/src/share/vm/runtime/orderAccess.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2003, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -25,8 +25,6 @@ # include "incls/_precompiled.incl" # include "incls/_orderAccess.cpp.incl" -volatile intptr_t OrderAccess::dummy = 0; - void OrderAccess::StubRoutines_fence() { // Use a stub if it exists. It may not exist during bootstrap so do // nothing in that case but assert if no fence code exists after threads have been created diff --git a/src/share/vm/runtime/orderAccess.hpp b/src/share/vm/runtime/orderAccess.hpp index 42c9227f5..28b049a43 100644 --- a/src/share/vm/runtime/orderAccess.hpp +++ b/src/share/vm/runtime/orderAccess.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2003, 2009, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -166,6 +166,12 @@ // and release must include a sequence point, usually via a volatile memory // access. Other ways to guarantee a sequence point are, e.g., use of // indirect calls and linux's __asm__ volatile. +// Note: as of 6973570, we have replaced the originally static "dummy" field +// (see above) by a volatile store to the stack. All of the versions of the +// compilers that we currently use (SunStudio, gcc and VC++) respect the +// semantics of volatile here. If you build HotSpot using other +// compilers, you may need to verify that no compiler reordering occurs +// across the sequence point respresented by the volatile access. // // // os::is_MP Considered Redundant @@ -297,10 +303,6 @@ class OrderAccess : AllStatic { static void release_store_ptr_fence(volatile intptr_t* p, intptr_t v); static void release_store_ptr_fence(volatile void* p, void* v); - // In order to force a memory access, implementations may - // need a volatile externally visible dummy variable. - static volatile intptr_t dummy; - private: // This is a helper that invokes the StubRoutines::fence_entry() // routine if it exists, It should only be used by platforms that diff --git a/src/share/vm/runtime/safepoint.cpp b/src/share/vm/runtime/safepoint.cpp index 65d130de1..2b51d3037 100644 --- a/src/share/vm/runtime/safepoint.cpp +++ b/src/share/vm/runtime/safepoint.cpp @@ -782,6 +782,9 @@ void ThreadSafepointState::examine_state_of_thread() { JavaThreadState state = _thread->thread_state(); + // Save the state at the start of safepoint processing. + _orig_thread_state = state; + // Check for a thread that is suspended. Note that thread resume tries // to grab the Threads_lock which we own here, so a thread cannot be // resumed during safepoint synchronization. diff --git a/src/share/vm/runtime/safepoint.hpp b/src/share/vm/runtime/safepoint.hpp index 46d7f3c3b..740cbd61c 100644 --- a/src/share/vm/runtime/safepoint.hpp +++ b/src/share/vm/runtime/safepoint.hpp @@ -185,6 +185,7 @@ class ThreadSafepointState: public CHeapObj { JavaThread * _thread; volatile suspend_type _type; + JavaThreadState _orig_thread_state; public: @@ -199,6 +200,7 @@ class ThreadSafepointState: public CHeapObj { JavaThread* thread() const { return _thread; } suspend_type type() const { return _type; } bool is_running() const { return (_type==_running); } + JavaThreadState orig_thread_state() const { return _orig_thread_state; } // Support for safepoint timeout (debugging) bool has_called_back() const { return _has_called_back; } diff --git a/src/share/vm/runtime/sharedRuntime.cpp b/src/share/vm/runtime/sharedRuntime.cpp index 4ff5aa0fa..5e9325399 100644 --- a/src/share/vm/runtime/sharedRuntime.cpp +++ b/src/share/vm/runtime/sharedRuntime.cpp @@ -2493,15 +2493,13 @@ nmethod *AdapterHandlerLibrary::create_native_wrapper(methodHandle method) { } // Must unlock before calling set_code + // Install the generated code. if (nm != NULL) { method->set_code(method, nm); nm->post_compiled_method_load_event(); } else { // CodeCache is full, disable compilation - // Ought to log this but compile log is only per compile thread - // and we're some non descript Java thread. - MutexUnlocker mu(AdapterHandlerLibrary_lock); CompileBroker::handle_full_code_cache(); } return nm; diff --git a/src/share/vm/runtime/thread.cpp b/src/share/vm/runtime/thread.cpp index 35afb6fe6..00e5e991e 100644 --- a/src/share/vm/runtime/thread.cpp +++ b/src/share/vm/runtime/thread.cpp @@ -2110,8 +2110,7 @@ void JavaThread::check_safepoint_and_suspend_for_native_trans(JavaThread *thread } if (f.id() == thread->must_deopt_id()) { thread->clear_must_deopt_id(); - // Since we know we're safe to deopt the current state is a safe state - f.deoptimize(thread, true); + f.deoptimize(thread); } else { fatal("missed deoptimization!"); } diff --git a/src/share/vm/services/management.cpp b/src/share/vm/services/management.cpp index dd6c3bc9b..c91a1acfc 100644 --- a/src/share/vm/services/management.cpp +++ b/src/share/vm/services/management.cpp @@ -1900,16 +1900,15 @@ JVM_ENTRY(void, jmm_GetLastGCStat(JNIEnv *env, jobject obj, jmmGCStat *gc_stat)) // Get the GCMemoryManager GCMemoryManager* mgr = get_gc_memory_manager_from_jobject(obj, CHECK); - if (mgr->last_gc_stat() == NULL) { - gc_stat->gc_index = 0; - return; - } // Make a copy of the last GC statistics // GC may occur while constructing the last GC information int num_pools = MemoryService::num_memory_pools(); GCStatInfo* stat = new GCStatInfo(num_pools); - stat->copy_stat(mgr->last_gc_stat()); + if (mgr->get_last_gc_stat(stat) == 0) { + gc_stat->gc_index = 0; + return; + } gc_stat->gc_index = stat->gc_index(); gc_stat->start_time = Management::ticks_to_ms(stat->start_time()); diff --git a/src/share/vm/services/memoryManager.cpp b/src/share/vm/services/memoryManager.cpp index 0266b331d..4921ae1f5 100644 --- a/src/share/vm/services/memoryManager.cpp +++ b/src/share/vm/services/memoryManager.cpp @@ -166,17 +166,6 @@ GCStatInfo::~GCStatInfo() { FREE_C_HEAP_ARRAY(MemoryUsage*, _after_gc_usage_array); } -void GCStatInfo::copy_stat(GCStatInfo* stat) { - set_index(stat->gc_index()); - set_start_time(stat->start_time()); - set_end_time(stat->end_time()); - assert(_usage_array_size == stat->usage_array_size(), "Must have same array size"); - for (int i = 0; i < _usage_array_size; i++) { - set_before_gc_usage(i, stat->before_gc_usage_for_pool(i)); - set_after_gc_usage(i, stat->after_gc_usage_for_pool(i)); - } -} - void GCStatInfo::set_gc_usage(int pool_index, MemoryUsage usage, bool before_gc) { MemoryUsage* gc_usage_array; if (before_gc) { @@ -187,67 +176,129 @@ void GCStatInfo::set_gc_usage(int pool_index, MemoryUsage usage, bool before_gc) gc_usage_array[pool_index] = usage; } +void GCStatInfo::clear() { + _index = 0; + _start_time = 0L; + _end_time = 0L; + size_t len = _usage_array_size * sizeof(MemoryUsage); + memset(_before_gc_usage_array, 0, len); + memset(_after_gc_usage_array, 0, len); +} + + GCMemoryManager::GCMemoryManager() : MemoryManager() { _num_collections = 0; _last_gc_stat = NULL; + _last_gc_lock = new Mutex(Mutex::leaf, "_last_gc_lock", true); + _current_gc_stat = NULL; _num_gc_threads = 1; } GCMemoryManager::~GCMemoryManager() { delete _last_gc_stat; + delete _last_gc_lock; + delete _current_gc_stat; } void GCMemoryManager::initialize_gc_stat_info() { assert(MemoryService::num_memory_pools() > 0, "should have one or more memory pools"); _last_gc_stat = new GCStatInfo(MemoryService::num_memory_pools()); + _current_gc_stat = new GCStatInfo(MemoryService::num_memory_pools()); + // tracking concurrent collections we need two objects: one to update, and one to + // hold the publicly available "last (completed) gc" information. } -void GCMemoryManager::gc_begin() { - assert(_last_gc_stat != NULL, "Just checking"); - _accumulated_timer.start(); - _num_collections++; - _last_gc_stat->set_index(_num_collections); - _last_gc_stat->set_start_time(Management::timestamp()); - - // Keep memory usage of all memory pools - for (int i = 0; i < MemoryService::num_memory_pools(); i++) { - MemoryPool* pool = MemoryService::get_memory_pool(i); - MemoryUsage usage = pool->get_memory_usage(); - _last_gc_stat->set_before_gc_usage(i, usage); - HS_DTRACE_PROBE8(hotspot, mem__pool__gc__begin, - name(), strlen(name()), - pool->name(), strlen(pool->name()), - usage.init_size(), usage.used(), - usage.committed(), usage.max_size()); +void GCMemoryManager::gc_begin(bool recordGCBeginTime, bool recordPreGCUsage, + bool recordAccumulatedGCTime) { + assert(_last_gc_stat != NULL && _current_gc_stat != NULL, "Just checking"); + if (recordAccumulatedGCTime) { + _accumulated_timer.start(); + } + // _num_collections now increases in gc_end, to count completed collections + if (recordGCBeginTime) { + _current_gc_stat->set_index(_num_collections+1); + _current_gc_stat->set_start_time(Management::timestamp()); + } + + if (recordPreGCUsage) { + // Keep memory usage of all memory pools + for (int i = 0; i < MemoryService::num_memory_pools(); i++) { + MemoryPool* pool = MemoryService::get_memory_pool(i); + MemoryUsage usage = pool->get_memory_usage(); + _current_gc_stat->set_before_gc_usage(i, usage); + HS_DTRACE_PROBE8(hotspot, mem__pool__gc__begin, + name(), strlen(name()), + pool->name(), strlen(pool->name()), + usage.init_size(), usage.used(), + usage.committed(), usage.max_size()); + } } } -void GCMemoryManager::gc_end() { - _accumulated_timer.stop(); - _last_gc_stat->set_end_time(Management::timestamp()); +// A collector MUST, even if it does not complete for some reason, +// make a TraceMemoryManagerStats object where countCollection is true, +// to ensure the current gc stat is placed in _last_gc_stat. +void GCMemoryManager::gc_end(bool recordPostGCUsage, + bool recordAccumulatedGCTime, + bool recordGCEndTime, bool countCollection) { + if (recordAccumulatedGCTime) { + _accumulated_timer.stop(); + } + if (recordGCEndTime) { + _current_gc_stat->set_end_time(Management::timestamp()); + } - int i; - // keep the last gc statistics for all memory pools - for (i = 0; i < MemoryService::num_memory_pools(); i++) { - MemoryPool* pool = MemoryService::get_memory_pool(i); - MemoryUsage usage = pool->get_memory_usage(); + if (recordPostGCUsage) { + int i; + // keep the last gc statistics for all memory pools + for (i = 0; i < MemoryService::num_memory_pools(); i++) { + MemoryPool* pool = MemoryService::get_memory_pool(i); + MemoryUsage usage = pool->get_memory_usage(); - HS_DTRACE_PROBE8(hotspot, mem__pool__gc__end, - name(), strlen(name()), - pool->name(), strlen(pool->name()), - usage.init_size(), usage.used(), - usage.committed(), usage.max_size()); + HS_DTRACE_PROBE8(hotspot, mem__pool__gc__end, + name(), strlen(name()), + pool->name(), strlen(pool->name()), + usage.init_size(), usage.used(), + usage.committed(), usage.max_size()); - _last_gc_stat->set_after_gc_usage(i, usage); - } + _current_gc_stat->set_after_gc_usage(i, usage); + } + + // Set last collection usage of the memory pools managed by this collector + for (i = 0; i < num_memory_pools(); i++) { + MemoryPool* pool = get_memory_pool(i); + MemoryUsage usage = pool->get_memory_usage(); - // Set last collection usage of the memory pools managed by this collector - for (i = 0; i < num_memory_pools(); i++) { - MemoryPool* pool = get_memory_pool(i); - MemoryUsage usage = pool->get_memory_usage(); + // Compare with GC usage threshold + pool->set_last_collection_usage(usage); + LowMemoryDetector::detect_after_gc_memory(pool); + } + } + if (countCollection) { + _num_collections++; + // alternately update two objects making one public when complete + { + MutexLockerEx ml(_last_gc_lock, Mutex::_no_safepoint_check_flag); + GCStatInfo *tmp = _last_gc_stat; + _last_gc_stat = _current_gc_stat; + _current_gc_stat = tmp; + // reset the current stat for diagnosability purposes + _current_gc_stat->clear(); + } + } +} - // Compare with GC usage threshold - pool->set_last_collection_usage(usage); - LowMemoryDetector::detect_after_gc_memory(pool); +size_t GCMemoryManager::get_last_gc_stat(GCStatInfo* dest) { + MutexLockerEx ml(_last_gc_lock, Mutex::_no_safepoint_check_flag); + if (_last_gc_stat->gc_index() != 0) { + dest->set_index(_last_gc_stat->gc_index()); + dest->set_start_time(_last_gc_stat->start_time()); + dest->set_end_time(_last_gc_stat->end_time()); + assert(dest->usage_array_size() == _last_gc_stat->usage_array_size(), + "Must have same array size"); + size_t len = dest->usage_array_size() * sizeof(MemoryUsage); + memcpy(dest->before_gc_usage_array(), _last_gc_stat->before_gc_usage_array(), len); + memcpy(dest->after_gc_usage_array(), _last_gc_stat->after_gc_usage_array(), len); } + return _last_gc_stat->gc_index(); } diff --git a/src/share/vm/services/memoryManager.hpp b/src/share/vm/services/memoryManager.hpp index 70b7eef71..7399141a8 100644 --- a/src/share/vm/services/memoryManager.hpp +++ b/src/share/vm/services/memoryManager.hpp @@ -131,6 +131,9 @@ public: return _after_gc_usage_array[pool_index]; } + MemoryUsage* before_gc_usage_array() { return _before_gc_usage_array; } + MemoryUsage* after_gc_usage_array() { return _after_gc_usage_array; } + void set_index(size_t index) { _index = index; } void set_start_time(jlong time) { _start_time = time; } void set_end_time(jlong time) { _end_time = time; } @@ -143,7 +146,7 @@ public: set_gc_usage(pool_index, usage, false /* after gc */); } - void copy_stat(GCStatInfo* stat); + void clear(); }; class GCMemoryManager : public MemoryManager { @@ -153,6 +156,8 @@ private: elapsedTimer _accumulated_timer; elapsedTimer _gc_timer; // for measuring every GC duration GCStatInfo* _last_gc_stat; + Mutex* _last_gc_lock; + GCStatInfo* _current_gc_stat; int _num_gc_threads; public: GCMemoryManager(); @@ -166,11 +171,16 @@ public: int num_gc_threads() { return _num_gc_threads; } void set_num_gc_threads(int count) { _num_gc_threads = count; } - void gc_begin(); - void gc_end(); + void gc_begin(bool recordGCBeginTime, bool recordPreGCUsage, + bool recordAccumulatedGCTime); + void gc_end(bool recordPostGCUsage, bool recordAccumulatedGCTime, + bool recordGCEndTime, bool countCollection); void reset_gc_stat() { _num_collections = 0; _accumulated_timer.reset(); } - GCStatInfo* last_gc_stat() { return _last_gc_stat; } + + // Copy out _last_gc_stat to the given destination, returning + // the collection count. Zero signifies no gc has taken place. + size_t get_last_gc_stat(GCStatInfo* dest); virtual MemoryManager::Name kind() = 0; }; diff --git a/src/share/vm/services/memoryService.cpp b/src/share/vm/services/memoryService.cpp index 3c55fffad..c50a4fefe 100644 --- a/src/share/vm/services/memoryService.cpp +++ b/src/share/vm/services/memoryService.cpp @@ -509,7 +509,10 @@ void MemoryService::track_memory_pool_usage(MemoryPool* pool) { } } -void MemoryService::gc_begin(bool fullGC) { +void MemoryService::gc_begin(bool fullGC, bool recordGCBeginTime, + bool recordAccumulatedGCTime, + bool recordPreGCUsage, bool recordPeakUsage) { + GCMemoryManager* mgr; if (fullGC) { mgr = _major_gc_manager; @@ -517,16 +520,21 @@ void MemoryService::gc_begin(bool fullGC) { mgr = _minor_gc_manager; } assert(mgr->is_gc_memory_manager(), "Sanity check"); - mgr->gc_begin(); + mgr->gc_begin(recordGCBeginTime, recordPreGCUsage, recordAccumulatedGCTime); // Track the peak memory usage when GC begins - for (int i = 0; i < _pools_list->length(); i++) { - MemoryPool* pool = _pools_list->at(i); - pool->record_peak_memory_usage(); + if (recordPeakUsage) { + for (int i = 0; i < _pools_list->length(); i++) { + MemoryPool* pool = _pools_list->at(i); + pool->record_peak_memory_usage(); + } } } -void MemoryService::gc_end(bool fullGC) { +void MemoryService::gc_end(bool fullGC, bool recordPostGCUsage, + bool recordAccumulatedGCTime, + bool recordGCEndTime, bool countCollection) { + GCMemoryManager* mgr; if (fullGC) { mgr = (GCMemoryManager*) _major_gc_manager; @@ -536,7 +544,8 @@ void MemoryService::gc_end(bool fullGC) { assert(mgr->is_gc_memory_manager(), "Sanity check"); // register the GC end statistics and memory usage - mgr->gc_end(); + mgr->gc_end(recordPostGCUsage, recordAccumulatedGCTime, recordGCEndTime, + countCollection); } void MemoryService::oops_do(OopClosure* f) { @@ -585,12 +594,12 @@ Handle MemoryService::create_MemoryUsage_obj(MemoryUsage usage, TRAPS) { return obj; } // -// GC manager type depends on the type of Generation. Depending the space -// availablity and vm option the gc uses major gc manager or minor gc +// GC manager type depends on the type of Generation. Depending on the space +// availablity and vm options the gc uses major gc manager or minor gc // manager or both. The type of gc manager depends on the generation kind. -// For DefNew, ParNew and ASParNew generation doing scavange gc uses minor -// gc manager (so _fullGC is set to false ) and for other generation kind -// DOing mark-sweep-compact uses major gc manager (so _fullGC is set +// For DefNew, ParNew and ASParNew generation doing scavenge gc uses minor +// gc manager (so _fullGC is set to false ) and for other generation kinds +// doing mark-sweep-compact uses major gc manager (so _fullGC is set // to true). TraceMemoryManagerStats::TraceMemoryManagerStats(Generation::Name kind) { switch (kind) { @@ -611,13 +620,48 @@ TraceMemoryManagerStats::TraceMemoryManagerStats(Generation::Name kind) { default: assert(false, "Unrecognized gc generation kind."); } - MemoryService::gc_begin(_fullGC); + // this has to be called in a stop the world pause and represent + // an entire gc pause, start to finish: + initialize(_fullGC, true, true, true, true, true, true, true); +} +TraceMemoryManagerStats::TraceMemoryManagerStats(bool fullGC, + bool recordGCBeginTime, + bool recordPreGCUsage, + bool recordPeakUsage, + bool recordPostGCUsage, + bool recordAccumulatedGCTime, + bool recordGCEndTime, + bool countCollection) { + initialize(fullGC, recordGCBeginTime, recordPreGCUsage, recordPeakUsage, + recordPostGCUsage, recordAccumulatedGCTime, recordGCEndTime, + countCollection); } -TraceMemoryManagerStats::TraceMemoryManagerStats(bool fullGC) { + +// for a subclass to create then initialize an instance before invoking +// the MemoryService +void TraceMemoryManagerStats::initialize(bool fullGC, + bool recordGCBeginTime, + bool recordPreGCUsage, + bool recordPeakUsage, + bool recordPostGCUsage, + bool recordAccumulatedGCTime, + bool recordGCEndTime, + bool countCollection) { _fullGC = fullGC; - MemoryService::gc_begin(_fullGC); + _recordGCBeginTime = recordGCBeginTime; + _recordPreGCUsage = recordPreGCUsage; + _recordPeakUsage = recordPeakUsage; + _recordPostGCUsage = recordPostGCUsage; + _recordAccumulatedGCTime = recordAccumulatedGCTime; + _recordGCEndTime = recordGCEndTime; + _countCollection = countCollection; + + MemoryService::gc_begin(_fullGC, _recordGCBeginTime, _recordAccumulatedGCTime, + _recordPreGCUsage, _recordPeakUsage); } TraceMemoryManagerStats::~TraceMemoryManagerStats() { - MemoryService::gc_end(_fullGC); + MemoryService::gc_end(_fullGC, _recordPostGCUsage, _recordAccumulatedGCTime, + _recordGCEndTime, _countCollection); } + diff --git a/src/share/vm/services/memoryService.hpp b/src/share/vm/services/memoryService.hpp index a21f3d1dc..cf26f1987 100644 --- a/src/share/vm/services/memoryService.hpp +++ b/src/share/vm/services/memoryService.hpp @@ -149,8 +149,13 @@ public: } static void track_memory_pool_usage(MemoryPool* pool); - static void gc_begin(bool fullGC); - static void gc_end(bool fullGC); + static void gc_begin(bool fullGC, bool recordGCBeginTime, + bool recordAccumulatedGCTime, + bool recordPreGCUsage, bool recordPeakUsage); + static void gc_end(bool fullGC, bool recordPostGCUsage, + bool recordAccumulatedGCTime, + bool recordGCEndTime, bool countCollection); + static void oops_do(OopClosure* f); @@ -164,8 +169,34 @@ public: class TraceMemoryManagerStats : public StackObj { private: bool _fullGC; + bool _recordGCBeginTime; + bool _recordPreGCUsage; + bool _recordPeakUsage; + bool _recordPostGCUsage; + bool _recordAccumulatedGCTime; + bool _recordGCEndTime; + bool _countCollection; + public: - TraceMemoryManagerStats(bool fullGC); + TraceMemoryManagerStats() {} + TraceMemoryManagerStats(bool fullGC, + bool recordGCBeginTime = true, + bool recordPreGCUsage = true, + bool recordPeakUsage = true, + bool recordPostGCUsage = true, + bool recordAccumulatedGCTime = true, + bool recordGCEndTime = true, + bool countCollection = true); + + void initialize(bool fullGC, + bool recordGCBeginTime, + bool recordPreGCUsage, + bool recordPeakUsage, + bool recordPostGCUsage, + bool recordAccumulatedGCTime, + bool recordGCEndTime, + bool countCollection); + TraceMemoryManagerStats(Generation::Name kind); ~TraceMemoryManagerStats(); }; diff --git a/src/share/vm/utilities/taskqueue.cpp b/src/share/vm/utilities/taskqueue.cpp index 7a001803d..2492b90d1 100644 --- a/src/share/vm/utilities/taskqueue.cpp +++ b/src/share/vm/utilities/taskqueue.cpp @@ -36,6 +36,14 @@ const char * const TaskQueueStats::_names[last_stat_id] = { "qpush", "qpop", "qpop-s", "qattempt", "qsteal", "opush", "omax" }; +TaskQueueStats & TaskQueueStats::operator +=(const TaskQueueStats & addend) +{ + for (unsigned int i = 0; i < last_stat_id; ++i) { + _stats[i] += addend._stats[i]; + } + return *this; +} + void TaskQueueStats::print_header(unsigned int line, outputStream* const stream, unsigned int width) { @@ -71,6 +79,29 @@ void TaskQueueStats::print(outputStream* stream, unsigned int width) const } #undef FMT } + +#ifdef ASSERT +// Invariants which should hold after a TaskQueue has been emptied and is +// quiescent; they do not hold at arbitrary times. +void TaskQueueStats::verify() const +{ + assert(get(push) == get(pop) + get(steal), + err_msg("push=" SIZE_FORMAT " pop=" SIZE_FORMAT " steal=" SIZE_FORMAT, + get(push), get(pop), get(steal))); + assert(get(pop_slow) <= get(pop), + err_msg("pop_slow=" SIZE_FORMAT " pop=" SIZE_FORMAT, + get(pop_slow), get(pop))); + assert(get(steal) <= get(steal_attempt), + err_msg("steal=" SIZE_FORMAT " steal_attempt=" SIZE_FORMAT, + get(steal), get(steal_attempt))); + assert(get(overflow) == 0 || get(push) != 0, + err_msg("overflow=" SIZE_FORMAT " push=" SIZE_FORMAT, + get(overflow), get(push))); + assert(get(overflow_max_len) == 0 || get(overflow) != 0, + err_msg("overflow_max_len=" SIZE_FORMAT " overflow=" SIZE_FORMAT, + get(overflow_max_len), get(overflow))); +} +#endif // ASSERT #endif // TASKQUEUE_STATS int TaskQueueSetSuper::randomParkAndMiller(int *seed0) { diff --git a/src/share/vm/utilities/taskqueue.hpp b/src/share/vm/utilities/taskqueue.hpp index 52e89fc80..1bd1ae723 100644 --- a/src/share/vm/utilities/taskqueue.hpp +++ b/src/share/vm/utilities/taskqueue.hpp @@ -59,15 +59,21 @@ public: inline void record_steal(bool success); inline void record_overflow(size_t new_length); + TaskQueueStats & operator +=(const TaskQueueStats & addend); + inline size_t get(StatId id) const { return _stats[id]; } inline const size_t* get() const { return _stats; } inline void reset(); + // Print the specified line of the header (does not include a line separator). static void print_header(unsigned int line, outputStream* const stream = tty, unsigned int width = 10); + // Print the statistics (does not include a line separator). void print(outputStream* const stream = tty, unsigned int width = 10) const; + DEBUG_ONLY(void verify() const;) + private: size_t _stats[last_stat_id]; static const char * const _names[last_stat_id]; diff --git a/test/compiler/6894807/IsInstanceTest.java b/test/compiler/6894807/IsInstanceTest.java new file mode 100644 index 000000000..ddc3fed36 --- /dev/null +++ b/test/compiler/6894807/IsInstanceTest.java @@ -0,0 +1,44 @@ +/* + * @test + * @bug 6894807 + * @summary No ClassCastException for HashAttributeSet constructors if run with -Xcomp + * @compile IsInstanceTest.java + * @run shell Test6894807.sh +*/ + +public class IsInstanceTest { + + public static void main(String[] args) { + BaseInterface baseInterfaceImpl = new BaseInterfaceImpl(); + for (int i = 0; i < 100000; i++) { + if (isInstanceOf(baseInterfaceImpl, ExtendedInterface.class)) { + System.out.println("Failed at index:" + i); + System.out.println("Arch: "+System.getProperty("os.arch", "")+ + " OS: "+System.getProperty("os.name", "")+ + " OSV: "+System.getProperty("os.version", "")+ + " Cores: "+Runtime.getRuntime().availableProcessors()+ + " JVM: "+System.getProperty("java.version", "")+" "+System.getProperty("sun.arch.data.model", "")); + break; + } + } + System.out.println("Done!"); + } + + public static boolean isInstanceOf(BaseInterface baseInterfaceImpl, Class... baseInterfaceClasses) { + for (Class baseInterfaceClass : baseInterfaceClasses) { + if (baseInterfaceClass.isInstance(baseInterfaceImpl)) { + return true; + } + } + return false; + } + + private interface BaseInterface { + } + + private interface ExtendedInterface extends BaseInterface { + } + + private static class BaseInterfaceImpl implements BaseInterface { + } +} diff --git a/test/compiler/6894807/Test6894807.sh b/test/compiler/6894807/Test6894807.sh new file mode 100644 index 000000000..da435cc38 --- /dev/null +++ b/test/compiler/6894807/Test6894807.sh @@ -0,0 +1,68 @@ +#!/bin/sh + +if [ "${TESTSRC}" = "" ] +then TESTSRC=. +fi + +if [ "${TESTJAVA}" = "" ] +then + PARENT=`dirname \`which java\`` + TESTJAVA=`dirname ${PARENT}` + echo "TESTJAVA not set, selecting " ${TESTJAVA} + echo "If this is incorrect, try setting the variable manually." +fi + +if [ "${TESTCLASSES}" = "" ] +then + echo "TESTCLASSES not set. Test cannot execute. Failed." + exit 1 +fi + +BIT_FLAG="" + +# set platform-dependent variables +OS=`uname -s` +case "$OS" in + SunOS | Linux ) + NULL=/dev/null + PS=":" + FS="/" + ## for solaris, linux it's HOME + FILE_LOCATION=$HOME + if [ -f ${FILE_LOCATION}${FS}JDK64BIT -a ${OS} = "SunOS" ] + then + BIT_FLAG=`cat ${FILE_LOCATION}${FS}JDK64BIT | grep -v '^#'` + fi + ;; + Windows_* ) + NULL=NUL + PS=";" + FS="\\" + ;; + * ) + echo "Unrecognized system!" + exit 1; + ;; +esac + +JEMMYPATH=${CPAPPEND} +CLASSPATH=.${PS}${TESTCLASSES}${PS}${JEMMYPATH} ; export CLASSPATH + +THIS_DIR=`pwd` + +${TESTJAVA}${FS}bin${FS}java ${BIT_FLAG} -version + +${TESTJAVA}${FS}bin${FS}java ${BIT_FLAG} -server IsInstanceTest > test.out 2>&1 + +cat test.out + +grep "Failed at index" test.out + +if [ $? = 0 ] +then + echo "Test Failed" + exit 1 +else + echo "Test Passed" + exit 0 +fi diff --git a/test/gc/6581734/Test6581734.java b/test/gc/6581734/Test6581734.java new file mode 100644 index 000000000..1967f0088 --- /dev/null +++ b/test/gc/6581734/Test6581734.java @@ -0,0 +1,149 @@ +/* + * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA + * or visit www.oracle.com if you need additional information or have any + * questions. + */ + +/* + * @test Test6581734.java + * @bug 6581734 + * @summary CMS Old Gen's collection usage is zero after GC which is incorrect + * @run main/othervm -Xmx512m -verbose:gc -XX:+UseConcMarkSweepGC Test6581734 + * + */ +import java.util.*; +import java.lang.management.*; + +// 6581734 states that memory pool usage via the mbean is wrong +// for CMS (zero, even after a collection). +// +// 6580448 states that the collection count similarly is wrong +// (stays at zero for CMS collections) +// -- closed as dup of 6581734 as the same fix resolves both. + + +public class Test6581734 { + + private String poolName = "CMS"; + private String collectorName = "ConcurrentMarkSweep"; + + public static void main(String [] args) { + + Test6581734 t = null; + if (args.length==2) { + t = new Test6581734(args[0], args[1]); + } else { + System.out.println("Defaulting to monitor CMS pool and collector."); + t = new Test6581734(); + } + t.run(); + } + + public Test6581734(String pool, String collector) { + poolName = pool; + collectorName = collector; + } + + public Test6581734() { + } + + public void run() { + // Use some memory, enough that we expect collections should + // have happened. + // Must run with options to ensure no stop the world full GC, + // but e.g. at least one CMS cycle. + allocationWork(300*1024*1024); + System.out.println("Done allocationWork"); + + // Verify some non-zero results are stored. + List<MemoryPoolMXBean> pools = ManagementFactory.getMemoryPoolMXBeans(); + int poolsFound = 0; + int poolsWithStats = 0; + for (int i=0; i<pools.size(); i++) { + MemoryPoolMXBean pool = pools.get(i); + String name = pool.getName(); + System.out.println("found pool: " + name); + + if (name.contains(poolName)) { + long usage = pool.getCollectionUsage().getUsed(); + System.out.println(name + ": usage after GC = " + usage); + poolsFound++; + if (usage > 0) { + poolsWithStats++; + } + } + } + if (poolsFound == 0) { + throw new RuntimeException("No matching memory pools found: test with -XX:+UseConcMarkSweepGC"); + } + + List<GarbageCollectorMXBean> collectors = ManagementFactory.getGarbageCollectorMXBeans(); + int collectorsFound = 0; + int collectorsWithTime= 0; + for (int i=0; i<collectors.size(); i++) { + GarbageCollectorMXBean collector = collectors.get(i); + String name = collector.getName(); + System.out.println("found collector: " + name); + if (name.contains(collectorName)) { + collectorsFound++; + System.out.println(name + ": collection count = " + + collector.getCollectionCount()); + System.out.println(name + ": collection time = " + + collector.getCollectionTime()); + if (collector.getCollectionCount() <= 0) { + throw new RuntimeException("collection count <= 0"); + } + if (collector.getCollectionTime() > 0) { + collectorsWithTime++; + } + } + } + // verify: + if (poolsWithStats < poolsFound) { + throw new RuntimeException("pools found with zero stats"); + } + + if (collectorsWithTime<collectorsFound) { + throw new RuntimeException("collectors found with zero time"; + } + System.out.println("Test passed."); + } + + public void allocationWork(long target) { + + long sizeAllocated = 0; + List list = new LinkedList(); + long delay = 50; + long count = 0; + + while (sizeAllocated < target) { + int size = 1024*1024; + byte [] alloc = new byte[size]; + if (count % 2 == 0) { + list.add(alloc); + sizeAllocated+=size; + System.out.print("."); + } + try { Thread.sleep(delay); } catch (InterruptedException ie) { } + count++; + } + } + +} diff --git a/test/runtime/6626217/IFace.java b/test/runtime/6626217/IFace.java new file mode 100644 index 000000000..1c2e86f78 --- /dev/null +++ b/test/runtime/6626217/IFace.java @@ -0,0 +1,6 @@ +// A simple interface, to allow an unknown foreign call from a class +// loaded with LOADER1 to a class loaded with LOADER2. +public interface IFace { + public many_loader[] gen(); +} + diff --git a/test/runtime/6626217/Loader2.java b/test/runtime/6626217/Loader2.java new file mode 100644 index 000000000..ea9e8ec1a --- /dev/null +++ b/test/runtime/6626217/Loader2.java @@ -0,0 +1,52 @@ +import java.io.ByteArrayInputStream; +import java.io.FileInputStream; +public class Loader2 extends ClassLoader { + int _recur; + public void print( String msg ) { + for( int i=0; i<_recur; i++ ) + System.out.print(" "); + System.out.println(">>Loader2>> "+msg); + } + + protected Class findClass2(String name) throws ClassNotFoundException { + print("Fetching the implementation of "+name); + int old = _recur; + try { + FileInputStream fi = new FileInputStream(name+".impl2"); + byte result[] = new byte[fi.available()]; + fi.read(result); + + print("DefineClass1 on "+name); + _recur++; + Class clazz = defineClass(name, result, 0, result.length); + _recur = old; + print("Returning newly loaded class."); + return clazz; + } catch (Exception e) { + _recur = old; + print("Not found on disk."); + // If we caught an exception, either the class was not found or + // it was unreadable by our process. + return null; + //throw new ClassNotFoundException(e.toString()); + } + } + + protected synchronized Class loadClass(String name, boolean resolve) throws ClassNotFoundException { + // Attempt a disk load first + Class c = findClass2(name); + if( c == null ) { + // check if the class has already been loaded + print("Checking for prior loaded class "+name); + c = findLoadedClass(name); + print("Letting super-loader load "+name); + int old = _recur; + _recur++; + c = super.loadClass(name, false); + _recur=old; + } + if (resolve) { print("Resolving class "+name); resolveClass(c); } + print("Returning clazz "+c.getClassLoader()+":"+name); + return c; + } +} diff --git a/test/runtime/6626217/Test6626217.sh b/test/runtime/6626217/Test6626217.sh new file mode 100644 index 000000000..6bc57a7be --- /dev/null +++ b/test/runtime/6626217/Test6626217.sh @@ -0,0 +1,101 @@ +# +# Copyright 2009 Sun Microsystems, Inc. All rights reserved. +# SUN PROPRIETARY/CONFIDENTIAL. Use is subject to license terms. +# + +# +# @test @(#)Test6626217.sh +# @bug 6626217 +# @summary Loader-constraint table allows arrays instead of only the base-classes +# @run shell Test6626217.sh +# + +if [ "${TESTSRC}" = "" ] + then TESTSRC=. +fi + +if [ "${TESTJAVA}" = "" ] +then + PARENT=`dirname \`which java\`` + TESTJAVA=`dirname ${PARENT}` + echo "TESTJAVA not set, selecting " ${TESTJAVA} + echo "If this is incorrect, try setting the variable manually." +fi + +if [ "${TESTCLASSES}" = "" ] +then + echo "TESTCLASSES not set. Test cannot execute. Failed." + exit 1 +fi + +BIT_FLAG="" + +# set platform-dependent variables +OS=`uname -s` +case "$OS" in + SunOS | Linux ) + NULL=/dev/null + PS=":" + FS="/" + RM=/bin/rm + CP=/bin/cp + MV=/bin/mv + ## for solaris, linux it's HOME + FILE_LOCATION=$HOME + if [ -f ${FILE_LOCATION}${FS}JDK64BIT -a ${OS} = "SunOS" ] + then + BIT_FLAG=`cat ${FILE_LOCATION}${FS}JDK64BIT` + fi + ;; + Windows_* ) + NULL=NUL + PS=";" + FS="\\" + RM=rm + CP=cp + MV=mv + ;; + * ) + echo "Unrecognized system!" + exit 1; + ;; +esac + +JEMMYPATH=${CPAPPEND} +CLASSPATH=.${PS}${TESTCLASSES}${PS}${JEMMYPATH} ; export CLASSPATH + +THIS_DIR=`pwd` + +JAVA=${TESTJAVA}${FS}bin${FS}java +JAVAC=${TESTJAVA}${FS}bin${FS}javac + +${JAVA} ${BIT_FLAG} -version + +# Current directory is scratch directory, copy all the test source there +# (for the subsequent moves to work). +${CP} ${TESTSRC}${FS}* ${THIS_DIR} + +# A Clean Compile: this line will probably fail within jtreg as have a clean dir: +${RM} -f *.class *.impl many_loader.java + +# Compile all the usual suspects, including the default 'many_loader' +${CP} many_loader1.java.foo many_loader.java +${JAVAC} -source 1.4 -target 1.4 -Xlint *.java + +# Rename the class files, so the custom loader (and not the system loader) will find it +${MV} from_loader2.class from_loader2.impl2 + +# Compile the next version of 'many_loader' +${MV} many_loader.class many_loader.impl1 +${CP} many_loader2.java.foo many_loader.java +${JAVAC} -source 1.4 -target 1.4 -Xlint many_loader.java + +# Rename the class file, so the custom loader (and not the system loader) will find it +${MV} many_loader.class many_loader.impl2 +${MV} many_loader.impl1 many_loader.class +${RM} many_loader.java + +${JAVA} ${BIT_FLAG} -Xverify -Xint -cp . bug_21227 >test.out 2>&1 +grep "violates loader constraints" test.out +exit $? + diff --git a/test/runtime/6626217/You_Have_Been_P0wned.java b/test/runtime/6626217/You_Have_Been_P0wned.java new file mode 100644 index 000000000..2bebbe059 --- /dev/null +++ b/test/runtime/6626217/You_Have_Been_P0wned.java @@ -0,0 +1,11 @@ + +// I can cast any old thing I want to this type object: +public class You_Have_Been_P0wned { + // Make a bunch of int-fields so I can peek & poke it + int _a; + int _b; + int _c; + int _d; + +} + diff --git a/test/runtime/6626217/bug_21227.java b/test/runtime/6626217/bug_21227.java new file mode 100644 index 000000000..18d30c41f --- /dev/null +++ b/test/runtime/6626217/bug_21227.java @@ -0,0 +1,61 @@ + +import java.lang.reflect.*; +import java.security.*; + +abstract public class bug_21227 { + + // Jam anything you want in here, it will be cast to a You_Have_Been_P0wned + public static Object _p0wnee; + + public static void main(String argv[]) throws ClassNotFoundException, InstantiationException, IllegalAccessException { + System.out.println("Warmup"); + + // Make a Class 'many_loader' under the default loader + bug_21227 bug = new many_loader(); + + // Some classes under a new Loader, LOADER2, including another version of 'many_loader' + ClassLoader LOADER2 = new Loader2(); + Class clazz2 = LOADER2.loadClass("from_loader2"); + IFace iface = (IFace)clazz2.newInstance(); + + // Set the victim, a String of length 6 + String s = "victim"; + _p0wnee = s; + + // Go cast '_p0wnee' to type You_Have_Been_P0wned + many_loader[] x2 = bug.make(iface); + + many_loader b = x2[0]; + + // Make it clear that the runtime type many_loader (what we get from the + // array X2) varies from the static type of many_loader. + Class cl1 = b.getClass(); + ClassLoader ld1 = cl1.getClassLoader(); + Class cl2 = many_loader.class; + ClassLoader ld2 = cl2.getClassLoader(); + System.out.println("bug.make() "+ld1+":"+cl1); + System.out.println("many_loader "+ld2+":"+cl2); + + // Read the victims guts out + You_Have_Been_P0wned q = b._p0wnee; + System.out.println("q._a = 0x"+Integer.toHexString(q._a)); + System.out.println("q._b = 0x"+Integer.toHexString(q._b)); + System.out.println("q._c = 0x"+Integer.toHexString(q._c)); + System.out.println("q._d = 0x"+Integer.toHexString(q._d)); + + System.out.println("I will now crash the VM:"); + // On 32-bit HotSpot Java6 this sets the victim String length shorter, then crashes the VM + //q._c = 3; + q._a = -1; + + System.out.println(s); + + } + + // I need to compile (hence call in a loop) a function which returns a value + // loaded from classloader other than the system one. The point of this + // call is to give me an abstract 'hook' into a function loaded with a + // foreign loader. + public abstract many_loader[] make( IFace iface ); // abstract factory +} + diff --git a/test/runtime/6626217/from_loader2.java b/test/runtime/6626217/from_loader2.java new file mode 100644 index 000000000..6421f2a3c --- /dev/null +++ b/test/runtime/6626217/from_loader2.java @@ -0,0 +1,9 @@ +// A simple class to extend an abstract class and get loaded with different +// loaders. This class is loaded via LOADER2. +public class from_loader2 implements IFace { + public many_loader[] gen() { + many_loader[] x = new many_loader[1]; + x[0] = new many_loader(); + return x; + } +} diff --git a/test/runtime/6626217/many_loader1.java.foo b/test/runtime/6626217/many_loader1.java.foo new file mode 100644 index 000000000..b5295a370 --- /dev/null +++ b/test/runtime/6626217/many_loader1.java.foo @@ -0,0 +1,23 @@ +// A simple class to extend an abstract class and get loaded with different +// loaders. This class is loaded via LOADER1. A similar named class will +// be loaded via LOADER2. +public class many_loader extends bug_21227 { + public You_Have_Been_P0wned _p0wnee; + + // I need to compile (hence call in a loop) a function which returns a value + // loaded from classloader other than the system one. The point of this + // call is to give me an abstract 'hook' into a function loaded with a + // foreign loader. + + // The original 'make(boolean)' returns a bug_21227. The VM will inject a + // synthetic method to up-cast the returned 'from_loader1' into a + // 'bug_21227'. + public many_loader[] make( IFace iface ) { + // This function needs to return a value known to be loaded from LOADER2. + // Since I need to use a yet different loader, I need to make an unknown + // foreign call. In this case I'll be using an interface to make the + // unknown call, with but a single implementor so the compiler can do the + // upcast statically. + return iface==null ? null : iface.gen(); + } +} diff --git a/test/runtime/6626217/many_loader2.java.foo b/test/runtime/6626217/many_loader2.java.foo new file mode 100644 index 000000000..1b72132a2 --- /dev/null +++ b/test/runtime/6626217/many_loader2.java.foo @@ -0,0 +1,19 @@ +// A simple class to extend an abstract class and get loaded with different +// loaders. This class is loaded via LOADER2. A similar named class will +// be loaded via LOADER1. +public class many_loader extends bug_21227 { + final Object _ref_to_be_p0wned; + + many_loader() { + _ref_to_be_p0wned = bug_21227._p0wnee; + System.out.println("Gonna hack this thing: " + _ref_to_be_p0wned.toString() ); + } + + // I need to compile (hence call in a loop) a function which returns a value + // loaded from classloader other than the system one. The point of this + // call is to give me an abstract 'hook' into a function loaded with a + // foreign loader. + public many_loader[] make( IFace iface ) { + throw new Error("do not call me"); + } +} |