Mercurial > hg > icedtea9-forest > hotspot
changeset 577:f9d5cfc2afa2
Merge
| author | xdono |
|---|---|
| date | Fri, 27 Feb 2009 15:13:00 -0800 |
| parents | 69c752d99841 (current diff) dae503d9f04c (diff) |
| children | f5eac45b1641 0fbdb4381b99 |
| files | |
| diffstat | 176 files changed, 3069 insertions(+), 995 deletions(-) [+] |
line wrap: on
line diff
--- a/.hgtags Sat Jan 31 17:19:42 2009 -0800 +++ b/.hgtags Fri Feb 27 15:13:00 2009 -0800 @@ -18,3 +18,9 @@ f9d938ede1960d18cb7cf23c645b026519c1a678 jdk7-b41 ad8c8ca4ab0f4c86e74c061958f44a8f4a930f2c jdk7-b42 fc6a5ae3fef5ebacfa896dbb3ae37715e388e282 jdk7-b43 +809e899c638bd9b21836abf9d09ab2a30ff3900b jdk7-b44 +945bf754069766e76873c53102fae48abf04cf5b jdk7-b45 +16bb38eeda35b46268eefa4c1f829eb086e0ca46 jdk7-b46 +fcb923bad68e2b10380a030ea83a723f4dc3d4d6 jdk7-b47 +bcb33806d186561c781992e5f4d8a90bb033f9f0 jdk7-b48 +8b22ccb5aba2c6c11bddf6488a7bb7ef5b4bf2be jdk7-b49
--- a/agent/src/os/linux/ps_core.c Sat Jan 31 17:19:42 2009 -0800 +++ b/agent/src/os/linux/ps_core.c Fri Feb 27 15:13:00 2009 -0800 @@ -238,8 +238,8 @@ // Ignore the rest of the FileMapHeader. We don't need those fields here. }; -static bool read_int(struct ps_prochandle* ph, uintptr_t addr, int* pvalue) { - int i; +static bool read_jboolean(struct ps_prochandle* ph, uintptr_t addr, jboolean* pvalue) { + jboolean i; if (ps_pdread(ph, (psaddr_t) addr, &i, sizeof(i)) == PS_OK) { *pvalue = i; return true; @@ -295,7 +295,7 @@ int fd = -1, m = 0; uintptr_t base = 0, useSharedSpacesAddr = 0; uintptr_t sharedArchivePathAddrAddr = 0, sharedArchivePathAddr = 0; - int useSharedSpaces = 0; + jboolean useSharedSpaces = 0; map_info* mi = 0; memset(classes_jsa, 0, sizeof(classes_jsa)); @@ -306,12 +306,15 @@ return false; } - if (read_int(ph, useSharedSpacesAddr, &useSharedSpaces) != true) { + // Hotspot vm types are not exported to build this library. So + // using equivalent type jboolean to read the value of + // UseSharedSpaces which is same as hotspot type "bool". + if (read_jboolean(ph, useSharedSpacesAddr, &useSharedSpaces) != true) { print_debug("can't read the value of 'UseSharedSpaces' flag\n"); return false; } - if (useSharedSpaces == 0) { + if ((int)useSharedSpaces == 0) { print_debug("UseSharedSpaces is false, assuming -Xshare:off!\n"); return true; }
--- a/agent/src/os/solaris/proc/saproc.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/agent/src/os/solaris/proc/saproc.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -502,8 +502,8 @@ }; static bool -read_int(struct ps_prochandle* ph, psaddr_t addr, int* pvalue) { - int i; +read_jboolean(struct ps_prochandle* ph, psaddr_t addr, jboolean* pvalue) { + jboolean i; if (ps_pread(ph, addr, &i, sizeof(i)) == PS_OK) { *pvalue = i; return true; @@ -575,10 +575,13 @@ } // read the value of the flag "UseSharedSpaces" - int value = 0; - if (read_int(ph, useSharedSpacesAddr, &value) != true) { + // Since hotspot types are not available to build this library. So + // equivalent type "jboolean" is used to read the value of "UseSharedSpaces" + // which is same as hotspot type "bool". + jboolean value = 0; + if (read_jboolean(ph, useSharedSpacesAddr, &value) != true) { THROW_NEW_DEBUGGER_EXCEPTION_("can't read 'UseSharedSpaces' flag", 1); - } else if (value == 0) { + } else if ((int)value == 0) { print_debug("UseSharedSpaces is false, assuming -Xshare:off!\n"); return 1; }
--- a/make/hotspot_version Sat Jan 31 17:19:42 2009 -0800 +++ b/make/hotspot_version Fri Feb 27 15:13:00 2009 -0800 @@ -33,9 +33,9 @@ # Don't put quotes (fail windows build). HOTSPOT_VM_COPYRIGHT=Copyright 2008 -HS_MAJOR_VER=14 +HS_MAJOR_VER=15 HS_MINOR_VER=0 -HS_BUILD_NUMBER=10 +HS_BUILD_NUMBER=02 JDK_MAJOR_VER=1 JDK_MINOR_VER=7
--- a/src/cpu/sparc/vm/jni_sparc.h Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/sparc/vm/jni_sparc.h Fri Feb 27 15:13:00 2009 -0800 @@ -28,5 +28,11 @@ #define JNICALL typedef int jint; -typedef long long jlong; + +#ifdef _LP64 + typedef long jlong; +#else + typedef long long jlong; +#endif + typedef signed char jbyte;
--- a/src/cpu/sparc/vm/sparc.ad Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/sparc/vm/sparc.ad Fri Feb 27 15:13:00 2009 -0800 @@ -762,7 +762,7 @@ case Assembler::stdf_op3: st_op = Op_StoreD; break; case Assembler::ldsb_op3: ld_op = Op_LoadB; break; - case Assembler::lduh_op3: ld_op = Op_LoadC; break; + case Assembler::lduh_op3: ld_op = Op_LoadUS; break; case Assembler::ldsh_op3: ld_op = Op_LoadS; break; case Assembler::ldx_op3: // may become LoadP or stay LoadI case Assembler::ldsw_op3: // may become LoadP or stay LoadI @@ -3869,6 +3869,8 @@ constraint(ALLOC_IN_RC(dflt_reg)); match(RegD); + match(regD_low); + format %{ %} interface(REG_INTER); %} @@ -3883,7 +3885,7 @@ operand regD_low() %{ constraint(ALLOC_IN_RC(dflt_low_reg)); - match(RegD); + match(regD); format %{ %} interface(REG_INTER); @@ -5314,9 +5316,9 @@ ins_pipe(iload_mask_mem); %} -// Load Char (16bit UNsigned) into a Long Register -instruct loadUCL(iRegL dst, memory mem, immL_FFFF bytemask) %{ - match(Set dst (AndL (ConvI2L (LoadC mem)) bytemask)); +// Load Unsigned Short/Char (16bit UNsigned) into a Long Register +instruct loadUS2L(iRegL dst, memory mem, immL_FFFF bytemask) %{ + match(Set dst (AndL (ConvI2L (LoadUS mem)) bytemask)); ins_cost(MEMORY_REF_COST); size(4); @@ -5326,9 +5328,9 @@ ins_pipe(iload_mask_mem); %} -// Load Char (16bit unsigned) -instruct loadC(iRegI dst, memory mem) %{ - match(Set dst (LoadC mem)); +// Load Unsigned Short/Char (16bit unsigned) +instruct loadUS(iRegI dst, memory mem) %{ + match(Set dst (LoadUS mem)); ins_cost(MEMORY_REF_COST); size(4);
--- a/src/cpu/x86/vm/assembler_x86.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/assembler_x86.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -5212,15 +5212,15 @@ void MacroAssembler::reset_last_Java_frame(bool clear_fp, bool clear_pc) { // we must set sp to zero to clear frame - movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), (int32_t)NULL_WORD); + movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD); // must clear fp, so that compiled frames are not confused; it is // possible that we need it only for debugging if (clear_fp) { - movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), (int32_t)NULL_WORD); + movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD); } if (clear_pc) { - movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), (int32_t)NULL_WORD); + movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD); } } @@ -5670,7 +5670,7 @@ // get oop result if there is one and reset the value in the thread if (oop_result->is_valid()) { movptr(oop_result, Address(java_thread, JavaThread::vm_result_offset())); - movptr(Address(java_thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD); + movptr(Address(java_thread, JavaThread::vm_result_offset()), NULL_WORD); verify_oop(oop_result, "broken oop in call_VM_base"); } } @@ -6426,13 +6426,13 @@ get_thread(java_thread); } // we must set sp to zero to clear frame - movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), (int32_t)NULL_WORD); + movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), NULL_WORD); if (clear_fp) { - movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), (int32_t)NULL_WORD); + movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), NULL_WORD); } if (clear_pc) - movptr(Address(java_thread, JavaThread::last_Java_pc_offset()), (int32_t)NULL_WORD); + movptr(Address(java_thread, JavaThread::last_Java_pc_offset()), NULL_WORD); } @@ -6943,29 +6943,32 @@ Label slow_case, done; - // x ?<= pi/4 - fld_d(ExternalAddress((address)&pi_4)); - fld_s(1); // Stack: X PI/4 X - fabs(); // Stack: |X| PI/4 X - fcmp(tmp); - jcc(Assembler::above, slow_case); - - // fastest case: -pi/4 <= x <= pi/4 - switch(trig) { - case 's': - fsin(); - break; - case 'c': - fcos(); - break; - case 't': - ftan(); - break; - default: - assert(false, "bad intrinsic"); - break; - } - jmp(done); + ExternalAddress pi4_adr = (address)&pi_4; + if (reachable(pi4_adr)) { + // x ?<= pi/4 + fld_d(pi4_adr); + fld_s(1); // Stack: X PI/4 X + fabs(); // Stack: |X| PI/4 X + fcmp(tmp); + jcc(Assembler::above, slow_case); + + // fastest case: -pi/4 <= x <= pi/4 + switch(trig) { + case 's': + fsin(); + break; + case 'c': + fcos(); + break; + case 't': + ftan(); + break; + default: + assert(false, "bad intrinsic"); + break; + } + jmp(done); + } // slow case: runtime call bind(slow_case);
--- a/src/cpu/x86/vm/bytecodeInterpreter_x86.inline.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/bytecodeInterpreter_x86.inline.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -213,7 +213,7 @@ inline jint BytecodeInterpreter::VMintDiv(jint op1, jint op2) { /* it's possible we could catch this special case implicitly */ - if (op1 == 0x80000000 && op2 == -1) return op1; + if ((juint)op1 == 0x80000000 && op2 == -1) return op1; else return op1 / op2; } @@ -231,7 +231,7 @@ inline jint BytecodeInterpreter::VMintRem(jint op1, jint op2) { /* it's possible we could catch this special case implicitly */ - if (op1 == 0x80000000 && op2 == -1) return 0; + if ((juint)op1 == 0x80000000 && op2 == -1) return 0; else return op1 % op2; }
--- a/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -779,7 +779,7 @@ case T_OBJECT: // fall through case T_ARRAY: if (c->as_jobject() == NULL) { - __ movptr(as_Address(addr), (int32_t)NULL_WORD); + __ movptr(as_Address(addr), NULL_WORD); } else { if (is_literal_address(addr)) { ShouldNotReachHere();
--- a/src/cpu/x86/vm/c1_Runtime1_x86.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/c1_Runtime1_x86.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -78,10 +78,10 @@ movptr(rax, Address(thread, Thread::pending_exception_offset())); // make sure that the vm_results are cleared if (oop_result1->is_valid()) { - movptr(Address(thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD); + movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD); } if (oop_result2->is_valid()) { - movptr(Address(thread, JavaThread::vm_result_2_offset()), (int32_t)NULL_WORD); + movptr(Address(thread, JavaThread::vm_result_2_offset()), NULL_WORD); } if (frame_size() == no_frame_size) { leave(); @@ -96,12 +96,12 @@ // get oop results if there are any and reset the values in the thread if (oop_result1->is_valid()) { movptr(oop_result1, Address(thread, JavaThread::vm_result_offset())); - movptr(Address(thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD); + movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD); verify_oop(oop_result1); } if (oop_result2->is_valid()) { movptr(oop_result2, Address(thread, JavaThread::vm_result_2_offset())); - movptr(Address(thread, JavaThread::vm_result_2_offset()), (int32_t)NULL_WORD); + movptr(Address(thread, JavaThread::vm_result_2_offset()), NULL_WORD); verify_oop(oop_result2); } return call_offset; @@ -728,8 +728,8 @@ // clear exception fields in JavaThread because they are no longer needed // (fields must be cleared because they are processed by GC otherwise) - __ movptr(Address(thread, JavaThread::exception_oop_offset()), (int32_t)NULL_WORD); - __ movptr(Address(thread, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD); + __ movptr(Address(thread, JavaThread::exception_oop_offset()), NULL_WORD); + __ movptr(Address(thread, JavaThread::exception_pc_offset()), NULL_WORD); // pop the stub frame off __ leave(); @@ -878,7 +878,7 @@ // load and clear pending exception __ movptr(rax, Address(thread, Thread::pending_exception_offset())); - __ movptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); + __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD); // check that there is really a valid exception __ verify_not_null_oop(rax); @@ -971,14 +971,14 @@ // load pending exception oop into rax, __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset())); // clear pending exception - __ movptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); + __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD); // load issuing PC (the return address for this stub) into rdx __ movptr(exception_pc, Address(rbp, 1*BytesPerWord)); // make sure that the vm_results are cleared (may be unnecessary) - __ movptr(Address(thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD); - __ movptr(Address(thread, JavaThread::vm_result_2_offset()), (int32_t)NULL_WORD); + __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD); + __ movptr(Address(thread, JavaThread::vm_result_2_offset()), NULL_WORD); // verify that that there is really a valid exception in rax, __ verify_not_null_oop(exception_oop); @@ -1393,7 +1393,7 @@ __ ret(0); __ bind(miss); - __ movptr(Address(rsp, (super_off) * VMRegImpl::stack_slot_size), 0); // result + __ movptr(Address(rsp, (super_off) * VMRegImpl::stack_slot_size), NULL_WORD); // result __ pop(rax); __ pop(rcx); __ pop(rsi);
--- a/src/cpu/x86/vm/cppInterpreter_x86.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/cppInterpreter_x86.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -594,7 +594,7 @@ __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax); // for c++ interpreter can rsi really be munged? - __ lea(state, Address(rbp, -sizeof(BytecodeInterpreter))); // restore state + __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); // restore state __ movptr(rbx, Address(state, byte_offset_of(BytecodeInterpreter, _method))); // restore method __ movptr(rdi, Address(state, byte_offset_of(BytecodeInterpreter, _locals))); // get locals pointer @@ -658,7 +658,7 @@ const Address size_of_stack (rbx, methodOopDesc::max_stack_offset()); // Always give one monitor to allow us to start interp if sync method. // Any additional monitors need a check when moving the expression stack - const one_monitor = frame::interpreter_frame_monitor_size() * wordSize; + const int one_monitor = frame::interpreter_frame_monitor_size() * wordSize; __ load_unsigned_word(rax, size_of_stack); // get size of expression stack in words __ lea(rax, Address(noreg, rax, Interpreter::stackElementScale(), one_monitor)); __ lea(rax, Address(rax, rdx, Interpreter::stackElementScale(), overhead_size)); @@ -1829,7 +1829,7 @@ Label unwind_and_forward; // restore state pointer. - __ lea(state, Address(rbp, -sizeof(BytecodeInterpreter))); + __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); __ movptr(rbx, STATE(_method)); // get method #ifdef _LP64 @@ -1877,14 +1877,14 @@ // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases if (UseSSE < 2) { - __ lea(state, Address(rbp, -sizeof(BytecodeInterpreter))); + __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); __ movptr(rbx, STATE(_result._to_call._callee)); // get method just executed __ movl(rcx, Address(rbx, methodOopDesc::result_index_offset())); __ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index __ jcc(Assembler::equal, do_float); __ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index __ jcc(Assembler::equal, do_double); -#ifdef COMPILER2 +#if !defined(_LP64) || defined(COMPILER1) || !defined(COMPILER2) __ empty_FPU_stack(); #endif // COMPILER2 __ jmp(done_conv); @@ -1928,7 +1928,7 @@ // Restore rsi/r13 as compiled code may not preserve it - __ lea(state, Address(rbp, -sizeof(BytecodeInterpreter))); + __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); // restore stack to what we had when we left (in case i2c extended it) @@ -1942,7 +1942,7 @@ #else __ movptr(rcx, STATE(_thread)); // get thread __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD); -#endif / __LP64 +#endif // _LP64 __ jcc(Assembler::notZero, return_with_exception); // get method just executed
--- a/src/cpu/x86/vm/frame_x86.inline.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/frame_x86.inline.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -139,7 +139,7 @@ #ifdef CC_INTERP inline interpreterState frame::get_interpreterState() const { - return ((interpreterState)addr_at( -sizeof(BytecodeInterpreter)/wordSize )); + return ((interpreterState)addr_at( -((int)sizeof(BytecodeInterpreter))/wordSize )); } inline intptr_t* frame::sender_sp() const {
--- a/src/cpu/x86/vm/interp_masm_x86_32.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/interp_masm_x86_32.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -133,7 +133,7 @@ + in_ByteSize(wordSize)); switch (state) { case atos: movptr(rax, oop_addr); - movptr(oop_addr, (int32_t)NULL_WORD); + movptr(oop_addr, NULL_WORD); verify_oop(rax, state); break; case ltos: movl(rdx, val_addr1); // fall through @@ -148,8 +148,8 @@ } // Clean up tos value in the thread object movl(tos_addr, (int32_t) ilgl); - movptr(val_addr, (int32_t)NULL_WORD); - NOT_LP64(movl(val_addr1, (int32_t)NULL_WORD)); + movptr(val_addr, NULL_WORD); + NOT_LP64(movptr(val_addr1, NULL_WORD)); } @@ -944,7 +944,7 @@ movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes ())); // Free entry - movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); + movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), NULL_WORD); if (UseBiasedLocking) { biased_locking_exit(obj_reg, header_reg, done);
--- a/src/cpu/x86/vm/interp_masm_x86_32.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/interp_masm_x86_32.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -120,7 +120,7 @@ void empty_expression_stack() { movptr(rsp, Address(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize)); // NULL last_sp until next java call - movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); + movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); } // Tagged stack helpers for swap and dup
--- a/src/cpu/x86/vm/interp_masm_x86_64.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/interp_masm_x86_64.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -30,7 +30,7 @@ #ifdef CC_INTERP void InterpreterMacroAssembler::get_method(Register reg) { - movptr(reg, Address(rbp, -(sizeof(BytecodeInterpreter) + 2 * wordSize))); + movptr(reg, Address(rbp, -((int)sizeof(BytecodeInterpreter) + 2 * wordSize))); movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method))); } #endif // CC_INTERP
--- a/src/cpu/x86/vm/interpreterRT_x86_32.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/interpreterRT_x86_32.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -54,7 +54,7 @@ __ cmpptr(Address(from(), Interpreter::local_offset_in_bytes(from_offset)), (int32_t)NULL_WORD); // do not use temp() to avoid AGI Label L; __ jcc(Assembler::notZero, L); - __ movptr(temp(), ((int32_t)NULL_WORD)); + __ movptr(temp(), NULL_WORD); __ bind(L); __ movptr(Address(to(), to_offset * wordSize), temp()); } @@ -110,7 +110,7 @@ virtual void pass_object() { // pass address of from intptr_t from_addr = (intptr_t)(_from + Interpreter::local_offset_in_bytes(0)); - *_to++ = (*(intptr_t*)from_addr == 0) ? NULL : from_addr; + *_to++ = (*(intptr_t*)from_addr == 0) ? NULL_WORD : from_addr; debug_only(verify_tag(frame::TagReference)); _from -= Interpreter::stackElementSize(); }
--- a/src/cpu/x86/vm/jni_x86.h Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/jni_x86.h Fri Feb 27 15:13:00 2009 -0800 @@ -32,7 +32,13 @@ #define JNICALL typedef int jint; + +#ifdef _LP64 + typedef long jlong; +#else typedef long long jlong; +#endif + #else #define JNIEXPORT __declspec(dllexport) #define JNIIMPORT __declspec(dllimport)
--- a/src/cpu/x86/vm/runtime_x86_32.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/runtime_x86_32.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -129,11 +129,11 @@ // Get the exception pc in case we are deoptimized __ movptr(rdx, Address(rcx, JavaThread::exception_pc_offset())); #ifdef ASSERT - __ movptr(Address(rcx, JavaThread::exception_handler_pc_offset()), (int32_t)NULL_WORD); - __ movptr(Address(rcx, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD); + __ movptr(Address(rcx, JavaThread::exception_handler_pc_offset()), NULL_WORD); + __ movptr(Address(rcx, JavaThread::exception_pc_offset()), NULL_WORD); #endif // Clear the exception oop so GC no longer processes it as a root. - __ movptr(Address(rcx, JavaThread::exception_oop_offset()), (int32_t)NULL_WORD); + __ movptr(Address(rcx, JavaThread::exception_oop_offset()), NULL_WORD); __ pop(rcx);
--- a/src/cpu/x86/vm/sharedRuntime_x86_32.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/sharedRuntime_x86_32.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -39,6 +39,8 @@ RuntimeStub* SharedRuntime::_resolve_virtual_call_blob; RuntimeStub* SharedRuntime::_resolve_static_call_blob; +const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size; + class RegisterSaver { enum { FPU_regs_live = 8 /*for the FPU stack*/+8/*eight more for XMM registers*/ }; // Capture info about frame layout @@ -1299,7 +1301,7 @@ // Now compute actual number of stack words we need rounding to make // stack properly aligned. - stack_slots = round_to(stack_slots, 2 * VMRegImpl::slots_per_word); + stack_slots = round_to(stack_slots, StackAlignmentInSlots); int stack_size = stack_slots * VMRegImpl::stack_slot_size; @@ -1793,7 +1795,7 @@ // reset handle block __ movptr(rcx, Address(thread, JavaThread::active_handles_offset())); - __ movptr(Address(rcx, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD); + __ movptr(Address(rcx, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD); // Any exception pending? __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD); @@ -1865,7 +1867,7 @@ // Save pending exception around call to VM (which contains an EXCEPTION_MARK) __ pushptr(Address(thread, in_bytes(Thread::pending_exception_offset()))); - __ movptr(Address(thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD); + __ movptr(Address(thread, in_bytes(Thread::pending_exception_offset())), NULL_WORD); // should be a peal @@ -2431,7 +2433,7 @@ __ get_thread(rdi); __ movptr(rdx, Address(rdi, JavaThread::exception_pc_offset())); __ movptr(Address(rbp, wordSize), rdx); - __ movptr(Address(rdi, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD); + __ movptr(Address(rdi, JavaThread::exception_pc_offset()), NULL_WORD); #ifdef ASSERT // verify that there is really an exception oop in JavaThread @@ -2489,8 +2491,8 @@ __ jcc(Assembler::notEqual, noException); __ movptr(rax, Address(rcx, JavaThread::exception_oop_offset())); __ movptr(rdx, Address(rcx, JavaThread::exception_pc_offset())); - __ movptr(Address(rcx, JavaThread::exception_oop_offset()), (int32_t)NULL_WORD); - __ movptr(Address(rcx, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD); + __ movptr(Address(rcx, JavaThread::exception_oop_offset()), NULL_WORD); + __ movptr(Address(rcx, JavaThread::exception_pc_offset()), NULL_WORD); __ verify_oop(rax); @@ -2582,7 +2584,7 @@ rbx); // Make it walkable #else /* CC_INTERP */ // This value is corrected by layout_activation_impl - __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD ); + __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), rbx); // Make it walkable #endif /* CC_INTERP */ __ movptr(sp_temp, rsp); // pass to next frame @@ -2802,7 +2804,7 @@ rbx); // Make it walkable #else /* CC_INTERP */ // This value is corrected by layout_activation_impl - __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD ); + __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD ); __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), rbx); // Make it walkable #endif /* CC_INTERP */ __ movptr(sp_temp, rsp); // pass to next frame @@ -3020,7 +3022,7 @@ // exception pending => remove activation and forward to exception handler __ get_thread(thread); - __ movptr(Address(thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD); + __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD); __ movptr(rax, Address(thread, Thread::pending_exception_offset())); __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
--- a/src/cpu/x86/vm/sharedRuntime_x86_64.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/sharedRuntime_x86_64.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -39,6 +39,8 @@ RuntimeStub* SharedRuntime::_resolve_virtual_call_blob; RuntimeStub* SharedRuntime::_resolve_static_call_blob; +const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size; + #define __ masm-> class SimpleRuntimeFrame { @@ -1286,7 +1288,7 @@ // Now compute actual number of stack words we need rounding to make // stack properly aligned. - stack_slots = round_to(stack_slots, 4 * VMRegImpl::slots_per_word); + stack_slots = round_to(stack_slots, StackAlignmentInSlots); int stack_size = stack_slots * VMRegImpl::stack_slot_size; @@ -2954,10 +2956,16 @@ __ pushptr(Address(rcx, 0)); // Save return address __ enter(); // Save old & set new rbp __ subptr(rsp, rbx); // Prolog +#ifdef CC_INTERP + __ movptr(Address(rbp, + -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))), + sender_sp); // Make it walkable +#else // CC_INTERP __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), sender_sp); // Make it walkable // This value is corrected by layout_activation_impl __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD ); +#endif // CC_INTERP __ mov(sender_sp, rsp); // Pass sender_sp to next frame __ addptr(rsi, wordSize); // Bump array pointer (sizes) __ addptr(rcx, wordSize); // Bump array pointer (pcs)
--- a/src/cpu/x86/vm/stubGenerator_x86_32.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/stubGenerator_x86_32.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -407,7 +407,7 @@ __ get_thread(rcx); __ pop(rdx); __ movptr(rax, Address(rcx, Thread::pending_exception_offset())); - __ movptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD); + __ movptr(Address(rcx, Thread::pending_exception_offset()), NULL_WORD); #ifdef ASSERT // make sure exception is set
--- a/src/cpu/x86/vm/stubGenerator_x86_64.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/stubGenerator_x86_64.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -472,7 +472,7 @@ // setup rax & rdx, remove return address & clear pending exception __ pop(rdx); __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset())); - __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int)NULL_WORD); + __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); #ifdef ASSERT // make sure exception is set @@ -954,9 +954,9 @@ __ jcc(Assembler::zero, exit); // if obj is NULL it is OK // Check if the oop is in the right area of memory __ movptr(c_rarg2, rax); - __ movptr(c_rarg3, (int64_t) Universe::verify_oop_mask()); + __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask()); __ andptr(c_rarg2, c_rarg3); - __ movptr(c_rarg3, (int64_t) Universe::verify_oop_bits()); + __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits()); __ cmpptr(c_rarg2, c_rarg3); __ jcc(Assembler::notZero, error); @@ -969,9 +969,9 @@ __ jcc(Assembler::zero, error); // if klass is NULL it is broken // Check if the klass is in the right area of memory __ mov(c_rarg2, rax); - __ movptr(c_rarg3, (int64_t) Universe::verify_klass_mask()); + __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask()); __ andptr(c_rarg2, c_rarg3); - __ movptr(c_rarg3, (int64_t) Universe::verify_klass_bits()); + __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits()); __ cmpptr(c_rarg2, c_rarg3); __ jcc(Assembler::notZero, error); @@ -980,9 +980,9 @@ __ testptr(rax, rax); __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken // Check if the klass' klass is in the right area of memory - __ movptr(c_rarg3, (int64_t) Universe::verify_klass_mask()); + __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask()); __ andptr(rax, c_rarg3); - __ movptr(c_rarg3, (int64_t) Universe::verify_klass_bits()); + __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits()); __ cmpptr(rax, c_rarg3); __ jcc(Assembler::notZero, error);
--- a/src/cpu/x86/vm/templateInterpreter_x86_32.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/templateInterpreter_x86_32.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -110,7 +110,7 @@ if (message != NULL) { __ lea(rbx, ExternalAddress((address)message)); } else { - __ movptr(rbx, (int32_t)NULL_WORD); + __ movptr(rbx, NULL_WORD); } __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx); } @@ -123,7 +123,7 @@ address TemplateInterpreterGenerator::generate_continuation_for(TosState state) { address entry = __ pc(); // NULL last_sp until next java call - __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); + __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); __ dispatch_next(state); return entry; } @@ -176,7 +176,7 @@ // Restore stack bottom in case i2c adjusted stack __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); // and NULL it as marker that rsp is now tos until next java call - __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); + __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); __ restore_bcp(); __ restore_locals(); @@ -211,7 +211,7 @@ // The stack is not extended by deopt but we must NULL last_sp as this // entry is like a "return". - __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); + __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); __ restore_bcp(); __ restore_locals(); // handle exceptions @@ -382,7 +382,7 @@ // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp). // The call returns the address of the verified entry point for the method or NULL // if the compilation did not complete (either went background or bailed out). - __ movptr(rax, (int32_t)false); + __ movptr(rax, (intptr_t)false); __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax); __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop @@ -1028,7 +1028,7 @@ // reset handle block __ movptr(t, Address(thread, JavaThread::active_handles_offset())); - __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD); + __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD); // If result was an oop then unbox and save it in the frame { Label L; @@ -1488,7 +1488,7 @@ // Restore sp to interpreter_frame_last_sp even though we are going // to empty the expression stack for the exception processing. - __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); + __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); // rax,: exception // rdx: return address/pc that threw exception __ restore_bcp(); // rsi points to call/send @@ -1608,7 +1608,7 @@ __ reset_last_Java_frame(rcx, true, true); // Restore the last_sp and null it out __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); - __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); + __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); __ restore_bcp(); __ restore_locals(); @@ -1636,7 +1636,7 @@ // restore exception __ get_thread(rcx); __ movptr(rax, Address(rcx, JavaThread::vm_result_offset())); - __ movptr(Address(rcx, JavaThread::vm_result_offset()), (int32_t)NULL_WORD); + __ movptr(Address(rcx, JavaThread::vm_result_offset()), NULL_WORD); __ verify_oop(rax); // Inbetween activations - previous activation type unknown yet
--- a/src/cpu/x86/vm/templateTable_x86_32.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/templateTable_x86_32.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -137,7 +137,7 @@ // Do the actual store // noreg means NULL if (val == noreg) { - __ movl(Address(rdx, 0), NULL_WORD); + __ movptr(Address(rdx, 0), NULL_WORD); // No post barrier for NULL } else { __ movl(Address(rdx, 0), val); @@ -152,7 +152,7 @@ case BarrierSet::CardTableExtension: { if (val == noreg) { - __ movl(obj, NULL_WORD); + __ movptr(obj, NULL_WORD); } else { __ movl(obj, val); // flatten object address if needed @@ -168,7 +168,7 @@ case BarrierSet::ModRef: case BarrierSet::Other: if (val == noreg) { - __ movl(obj, NULL_WORD); + __ movptr(obj, NULL_WORD); } else { __ movl(obj, val); }
--- a/src/cpu/x86/vm/x86_32.ad Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/x86_32.ad Fri Feb 27 15:13:00 2009 -0800 @@ -3371,7 +3371,7 @@ masm.movptr(Address(boxReg, 0), 3) ; // results in ST-before-CAS penalty masm.get_thread (scrReg) ; masm.movptr(boxReg, tmpReg); // consider: LEA box, [tmp-2] - masm.movptr(tmpReg, 0); // consider: xor vs mov + masm.movptr(tmpReg, NULL_WORD); // consider: xor vs mov if (os::is_MP()) { masm.lock(); } masm.cmpxchgptr(scrReg, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; } else @@ -3387,7 +3387,7 @@ if ((EmitSync & 64) == 0) { // Optimistic form: consider XORL tmpReg,tmpReg - masm.movptr(tmpReg, 0 ) ; + masm.movptr(tmpReg, NULL_WORD) ; } else { // Can suffer RTS->RTO upgrades on shared or cold $ lines // Test-And-CAS instead of CAS @@ -3587,7 +3587,7 @@ masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ; masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ; masm.jccb (Assembler::notZero, DONE_LABEL) ; - masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), 0) ; + masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), NULL_WORD) ; masm.jmpb (DONE_LABEL) ; } else { masm.xorptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; @@ -3596,7 +3596,7 @@ masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ; masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ; masm.jccb (Assembler::notZero, CheckSucc) ; - masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), 0) ; + masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), NULL_WORD) ; masm.jmpb (DONE_LABEL) ; } @@ -3644,7 +3644,7 @@ // We currently use (3), although it's likely that switching to (2) // is correct for the future. - masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), 0) ; + masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), NULL_WORD) ; if (os::is_MP()) { if (VM_Version::supports_sse2() && 1 == FenceInstruction) { masm.mfence(); @@ -6413,9 +6413,9 @@ ins_pipe( ialu_reg_mem ); %} -// Load Char (16bit unsigned) -instruct loadC(eRegI dst, memory mem) %{ - match(Set dst (LoadC mem)); +// Load Unsigned Short/Char (16bit unsigned) +instruct loadUS(eRegI dst, memory mem) %{ + match(Set dst (LoadUS mem)); ins_cost(125); format %{ "MOVZX $dst,$mem" %}
--- a/src/cpu/x86/vm/x86_64.ad Sat Jan 31 17:19:42 2009 -0800 +++ b/src/cpu/x86/vm/x86_64.ad Fri Feb 27 15:13:00 2009 -0800 @@ -6096,25 +6096,25 @@ // ins_pipe(ialu_reg_mem); // %} -// Load Char (16 bit UNsigned) -instruct loadC(rRegI dst, memory mem) -%{ - match(Set dst (LoadC mem)); +// Load Unsigned Short/Char (16 bit UNsigned) +instruct loadUS(rRegI dst, memory mem) +%{ + match(Set dst (LoadUS mem)); ins_cost(125); - format %{ "movzwl $dst, $mem\t# char" %} + format %{ "movzwl $dst, $mem\t# ushort/char" %} opcode(0x0F, 0xB7); ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem)); ins_pipe(ialu_reg_mem); %} -// Load Char (16 bit UNsigned) into long -// instruct loadC2L(rRegL dst, memory mem) +// Load Unsigned Short/Char (16 bit UNsigned) into long +// instruct loadUS2L(rRegL dst, memory mem) // %{ -// match(Set dst (ConvI2L (LoadC mem))); +// match(Set dst (ConvI2L (LoadUS mem))); // ins_cost(125); -// format %{ "movzwl $dst, $mem\t# char -> long" %} +// format %{ "movzwl $dst, $mem\t# ushort/char -> long" %} // opcode(0x0F, 0xB7); // ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem)); // ins_pipe(ialu_reg_mem); @@ -9490,14 +9490,14 @@ %{ match(Set dst (AndL dst src)); - format %{ "movzbq $dst, $src\t# long & 0xFF" %} + format %{ "movzbq $dst, $dst\t# long & 0xFF" %} opcode(0x0F, 0xB6); ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst)); ins_pipe(ialu_reg); %} // And Register with Immediate 65535 -instruct andL_rReg_imm65535(rRegI dst, immL_65535 src) +instruct andL_rReg_imm65535(rRegL dst, immL_65535 src) %{ match(Set dst (AndL dst src));
--- a/src/os/linux/vm/os_linux.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/os/linux/vm/os_linux.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -279,7 +279,11 @@ * ... * 7: The default directories, normally /lib and /usr/lib. */ +#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) +#define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" +#else #define DEFAULT_LIBPATH "/lib:/usr/lib" +#endif #define EXTENSIONS_DIR "/lib/ext" #define ENDORSED_DIR "/lib/endorsed" @@ -1160,7 +1164,10 @@ /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ /* 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 */ - i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld %lu %lu %ld %lu %lu %lu %lu", + i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " + UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT + " %lu " + UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT, &state, /* 3 %c */ &ppid, /* 4 %d */ &pgrp, /* 5 %d */ @@ -1180,13 +1187,13 @@ &nice, /* 19 %ld */ &junk, /* 20 %ld */ &it_real, /* 21 %ld */ - &start, /* 22 %lu */ - &vsize, /* 23 %lu */ - &rss, /* 24 %ld */ + &start, /* 22 UINTX_FORMAT */ + &vsize, /* 23 UINTX_FORMAT */ + &rss, /* 24 UINTX_FORMAT */ &rsslim, /* 25 %lu */ - &scodes, /* 26 %lu */ - &ecode, /* 27 %lu */ - &stack_start); /* 28 %lu */ + &scodes, /* 26 UINTX_FORMAT */ + &ecode, /* 27 UINTX_FORMAT */ + &stack_start); /* 28 UINTX_FORMAT */ } if (i != 28 - 2) { @@ -1425,6 +1432,10 @@ return buf; } +struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { + return localtime_r(clock, res); +} + //////////////////////////////////////////////////////////////////////////////// // runtime exit support @@ -2024,7 +2035,8 @@ CAST_FROM_FN_PTR(address, os::jvm_path), dli_fname, sizeof(dli_fname), NULL); assert(ret != 0, "cannot locate libjvm"); - realpath(dli_fname, buf); + if (realpath(dli_fname, buf) == NULL) + return; if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) { // Support for the gamma launcher. Typical value for buf is @@ -2048,7 +2060,8 @@ assert(strstr(p, "/libjvm") == p, "invalid library name"); p = strstr(p, "_g") ? "_g" : ""; - realpath(java_home_var, buf); + if (realpath(java_home_var, buf) == NULL) + return; sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch); if (0 == access(buf, F_OK)) { // Use current module name "libjvm[_g].so" instead of @@ -2059,7 +2072,8 @@ sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p); } else { // Go back to path of .so - realpath(dli_fname, buf); + if (realpath(dli_fname, buf) == NULL) + return; } } } @@ -4184,11 +4198,11 @@ // Skip blank chars do s++; while (isspace(*s)); - count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", - &idummy, &idummy, &idummy, &idummy, &idummy, &idummy, + count = sscanf(s,"%*c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", + &idummy, &idummy, &idummy, &idummy, &idummy, &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, &user_time, &sys_time); - if ( count != 13 ) return -1; + if ( count != 12 ) return -1; if (user_sys_cpu_time) { return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); } else {
--- a/src/os/solaris/vm/os_solaris.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/os/solaris/vm/os_solaris.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -323,6 +323,10 @@ return (size_t)(base - bottom); } +struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { + return localtime_r(clock, res); +} + // interruptible infrastructure // setup_interruptible saves the thread state before going into an
--- a/src/os/windows/vm/os_windows.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/os/windows/vm/os_windows.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -327,6 +327,14 @@ return sz; } +struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { + const struct tm* time_struct_ptr = localtime(clock); + if (time_struct_ptr != NULL) { + *res = *time_struct_ptr; + return res; + } + return NULL; +} LONG WINAPI topLevelExceptionFilter(struct _EXCEPTION_POINTERS* exceptionInfo);
--- a/src/share/vm/adlc/dict2.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/adlc/dict2.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -316,9 +316,12 @@ return strcmp((const char *)k1,(const char *)k2); } -// Slimey cheap key comparator. +// Cheap key comparator. int cmpkey(const void *key1, const void *key2) { - return (int)((intptr_t)key1 - (intptr_t)key2); + if (key1 == key2) return 0; + intptr_t delta = (intptr_t)key1 - (intptr_t)key2; + if (delta > 0) return 1; + return -1; } //=============================================================================
--- a/src/share/vm/adlc/forms.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/adlc/forms.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -248,7 +248,7 @@ // True if 'opType', an ideal name, loads or stores. Form::DataType Form::is_load_from_memory(const char *opType) const { if( strcmp(opType,"LoadB")==0 ) return Form::idealB; - if( strcmp(opType,"LoadC")==0 ) return Form::idealC; + if( strcmp(opType,"LoadUS")==0 ) return Form::idealC; if( strcmp(opType,"LoadD")==0 ) return Form::idealD; if( strcmp(opType,"LoadD_unaligned")==0 ) return Form::idealD; if( strcmp(opType,"LoadF")==0 ) return Form::idealF;
--- a/src/share/vm/adlc/formssel.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/adlc/formssel.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -3314,7 +3314,7 @@ "StoreI","StoreL","StoreP","StoreN","StoreD","StoreF" , "StoreB","StoreC","Store" ,"StoreFP", "LoadI" ,"LoadL", "LoadP" ,"LoadN", "LoadD" ,"LoadF" , - "LoadB" ,"LoadC" ,"LoadS" ,"Load" , + "LoadB" ,"LoadUS" ,"LoadS" ,"Load" , "Store4I","Store2I","Store2L","Store2D","Store4F","Store2F","Store16B", "Store8B","Store4B","Store8C","Store4C","Store2C", "Load4I" ,"Load2I" ,"Load2L" ,"Load2D" ,"Load4F" ,"Load2F" ,"Load16B" ,
--- a/src/share/vm/asm/codeBuffer.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/asm/codeBuffer.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -123,6 +123,10 @@ // addresses constructed before expansions will not be confused. cb->free_blob(); } + + // free any overflow storage + delete _overflow_arena; + #ifdef ASSERT Copy::fill_to_bytes(this, sizeof(*this), badResourceValue); #endif
--- a/src/share/vm/classfile/classFileParser.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/classfile/classFileParser.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -232,7 +232,9 @@ length >= 1, "Illegal constant pool size %u in class file %s", length, CHECK_(nullHandle)); constantPoolOop constant_pool = - oopFactory::new_constantPool(length, CHECK_(nullHandle)); + oopFactory::new_constantPool(length, + methodOopDesc::IsSafeConc, + CHECK_(nullHandle)); constantPoolHandle cp (THREAD, constant_pool); cp->set_partially_loaded(); // Enables heap verify to work on partial constantPoolOops @@ -1675,7 +1677,8 @@ // All sizing information for a methodOop is finally available, now create it methodOop m_oop = oopFactory::new_method( code_length, access_flags, linenumber_table_length, - total_lvt_length, checked_exceptions_length, CHECK_(nullHandle)); + total_lvt_length, checked_exceptions_length, + methodOopDesc::IsSafeConc, CHECK_(nullHandle)); methodHandle m (THREAD, m_oop); ClassLoadingService::add_class_method_size(m_oop->size()*HeapWordSize);
--- a/src/share/vm/classfile/javaClasses.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/classfile/javaClasses.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -441,6 +441,7 @@ bool java_lang_Class::offsets_computed = false; int java_lang_Class::classRedefinedCount_offset = -1; +int java_lang_Class::parallelCapable_offset = -1; void java_lang_Class::compute_offsets() { assert(!offsets_computed, "offsets should be initialized only once"); @@ -451,6 +452,23 @@ // so don't go fatal. compute_optional_offset(classRedefinedCount_offset, k, vmSymbols::classRedefinedCount_name(), vmSymbols::int_signature()); + + // The field indicating parallelCapable (parallelLockMap) is only present starting in 7, + klassOop k1 = SystemDictionary::classloader_klass(); + compute_optional_offset(parallelCapable_offset, + k1, vmSymbols::parallelCapable_name(), vmSymbols::concurrenthashmap_signature()); +} + +// For class loader classes, parallelCapable defined +// based on non-null field +// Written to by java.lang.ClassLoader, vm only reads this field, doesn't set it +bool java_lang_Class::parallelCapable(oop class_loader) { + if (!JDK_Version::is_gte_jdk17x_version() + || parallelCapable_offset == -1) { + // Default for backward compatibility is false + return false; + } + return (class_loader->obj_field(parallelCapable_offset) != NULL); } int java_lang_Class::classRedefinedCount(oop the_class_mirror) { @@ -866,7 +884,7 @@ } nmethod* nm = method->code(); if (WizardMode && nm != NULL) { - sprintf(buf + (int)strlen(buf), "(nmethod %#x)", nm); + sprintf(buf + (int)strlen(buf), "(nmethod " PTR_FORMAT ")", (intptr_t)nm); } }
--- a/src/share/vm/classfile/javaClasses.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/classfile/javaClasses.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -141,6 +141,7 @@ static void compute_offsets(); static bool offsets_computed; static int classRedefinedCount_offset; + static int parallelCapable_offset; public: // Instance creation @@ -168,6 +169,8 @@ // Support for classRedefinedCount field static int classRedefinedCount(oop the_class_mirror); static void set_classRedefinedCount(oop the_class_mirror, int value); + // Support for parallelCapable field + static bool parallelCapable(oop the_class_mirror); // Debugging friend class JavaClasses; friend class instanceKlass; // verification code accesses offsets
--- a/src/share/vm/classfile/systemDictionary.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/classfile/systemDictionary.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -90,6 +90,14 @@ #endif // ---------------------------------------------------------------------------- +// Parallel class loading check + +bool SystemDictionary::is_parallelCapable(Handle class_loader) { + if (UnsyncloadClass || class_loader.is_null()) return true; + if (AlwaysLockClassLoader) return false; + return java_lang_Class::parallelCapable(class_loader()); +} +// ---------------------------------------------------------------------------- // Resolving of classes // Forwards to resolve_or_null @@ -196,7 +204,8 @@ // super-class callers: // ClassFileParser - for defineClass & jvmtiRedefineClasses // load_shared_class - while loading a class from shared archive -// resolve_instance_class_or_fail: +// resolve_instance_class_or_null: +// via: handle_parallel_super_load // when resolving a class that has an existing placeholder with // a saved superclass [i.e. a defineClass is currently in progress] // if another thread is trying to resolve the class, it must do @@ -283,12 +292,9 @@ if (probe && probe->check_seen_thread(THREAD, PlaceholderTable::LOAD_SUPER)) { throw_circularity_error = true; } - - // add placeholder entry even if error - callers will remove on error + } + if (!throw_circularity_error) { PlaceholderEntry* newprobe = placeholders()->find_and_add(p_index, p_hash, child_name, class_loader, PlaceholderTable::LOAD_SUPER, class_name, THREAD); - if (throw_circularity_error) { - newprobe->remove_seen_thread(THREAD, PlaceholderTable::LOAD_SUPER); - } } } if (throw_circularity_error) { @@ -325,7 +331,6 @@ return superk_h(); } - void SystemDictionary::validate_protection_domain(instanceKlassHandle klass, Handle class_loader, Handle protection_domain, @@ -421,7 +426,7 @@ bool calledholdinglock = ObjectSynchronizer::current_thread_holds_lock((JavaThread*)THREAD, lockObject); assert(calledholdinglock,"must hold lock for notify"); - assert(!UnsyncloadClass, "unexpected double_lock_wait"); + assert((!(lockObject() == _system_loader_lock_obj) && !is_parallelCapable(lockObject)), "unexpected double_lock_wait"); ObjectSynchronizer::notifyall(lockObject, THREAD); intptr_t recursions = ObjectSynchronizer::complete_exit(lockObject, THREAD); SystemDictionary_lock->wait(); @@ -439,7 +444,7 @@ // even in non-circularity situations. // Note: only one thread can define the class, but multiple can resolve // Note: must call resolve_super_or_fail even if null super - -// to force placeholder entry creation for this class +// to force placeholder entry creation for this class for circularity detection // Caller must check for pending exception // Returns non-null klassOop if other thread has completed load // and we are done, @@ -477,9 +482,9 @@ SystemDictionary_lock->notify_all(); } - // UnsyncloadClass does NOT wait for parallel superclass loads to complete - // Bootstrap classloader does wait for parallel superclass loads - if (UnsyncloadClass) { + // parallelCapable class loaders do NOT wait for parallel superclass loads to complete + // Serial class loaders and bootstrap classloader do wait for superclass loads + if (!class_loader.is_null() && is_parallelCapable(class_loader)) { MutexLocker mu(SystemDictionary_lock, THREAD); // Check if classloading completed while we were loading superclass or waiting klassOop check = find_class(d_index, d_hash, name, class_loader); @@ -566,10 +571,10 @@ // This lock must be acquired here so the waiter will find // any successful result in the SystemDictionary and not attempt // the define - // Classloaders that support parallelism, e.g. bootstrap classloader, + // ParallelCapable Classloaders and the bootstrap classloader, // or all classloaders with UnsyncloadClass do not acquire lock here bool DoObjectLock = true; - if (UnsyncloadClass || (class_loader.is_null())) { + if (is_parallelCapable(class_loader)) { DoObjectLock = false; } @@ -627,6 +632,9 @@ // Five cases: // All cases need to prevent modifying bootclasssearchpath // in parallel with a classload of same classname + // Redefineclasses uses existence of the placeholder for the duration + // of the class load to prevent concurrent redefinition of not completely + // defined classes. // case 1. traditional classloaders that rely on the classloader object lock // - no other need for LOAD_INSTANCE // case 2. traditional classloaders that break the classloader object lock @@ -642,12 +650,13 @@ // This classloader supports parallelism at the classloader level, // but only allows a single load of a class/classloader pair. // No performance benefit and no deadlock issues. - // case 5. Future: parallel user level classloaders - without objectLocker + // case 5. parallelCapable user level classloaders - without objectLocker + // Allow parallel classloading of a class/classloader pair symbolHandle nullsymbolHandle; bool throw_circularity_error = false; { MutexLocker mu(SystemDictionary_lock, THREAD); - if (!UnsyncloadClass) { + if (class_loader.is_null() || !is_parallelCapable(class_loader)) { PlaceholderEntry* oldprobe = placeholders()->get_entry(p_index, p_hash, name, class_loader); if (oldprobe) { // only need check_seen_thread once, not on each loop @@ -681,25 +690,25 @@ } } // All cases: add LOAD_INSTANCE - // case 3: UnsyncloadClass: allow competing threads to try + // case 3: UnsyncloadClass || case 5: parallelCapable: allow competing threads to try // LOAD_INSTANCE in parallel // add placeholder entry even if error - callers will remove on error - if (!class_has_been_loaded) { + if (!throw_circularity_error && !class_has_been_loaded) { PlaceholderEntry* newprobe = placeholders()->find_and_add(p_index, p_hash, name, class_loader, PlaceholderTable::LOAD_INSTANCE, nullsymbolHandle, THREAD); - if (throw_circularity_error) { - newprobe->remove_seen_thread(THREAD, PlaceholderTable::LOAD_INSTANCE); - } // For class loaders that do not acquire the classloader object lock, // if they did not catch another thread holding LOAD_INSTANCE, // need a check analogous to the acquire ObjectLocker/find_class // i.e. now that we hold the LOAD_INSTANCE token on loading this class/CL // one final check if the load has already completed + // class loaders holding the ObjectLock shouldn't find the class here klassOop check = find_class(d_index, d_hash, name, class_loader); if (check != NULL) { // Klass is already loaded, so just return it k = instanceKlassHandle(THREAD, check); class_has_been_loaded = true; newprobe->remove_seen_thread(THREAD, PlaceholderTable::LOAD_INSTANCE); + placeholders()->find_and_remove(p_index, p_hash, name, class_loader, THREAD); + SystemDictionary_lock->notify_all(); } } } @@ -714,18 +723,14 @@ // Do actual loading k = load_instance_class(name, class_loader, THREAD); - // In custom class loaders, the usual findClass calls - // findLoadedClass, which directly searches the SystemDictionary, then - // defineClass. If these are not atomic with respect to other threads, - // the findLoadedClass can fail, but the defineClass can get a - // LinkageError:: duplicate class definition. + // For UnsyncloadClass and AllowParallelDefineClass only: // If they got a linkageError, check if a parallel class load succeeded. // If it did, then for bytecode resolution the specification requires // that we return the same result we did for the other thread, i.e. the // successfully loaded instanceKlass - // Note: Class can not be unloaded as long as any classloader refs exist // Should not get here for classloaders that support parallelism - // with the new cleaner mechanism, e.g. bootstrap classloader + // with the new cleaner mechanism + // Bootstrap goes through here to allow for an extra guarantee check if (UnsyncloadClass || (class_loader.is_null())) { if (k.is_null() && HAS_PENDING_EXCEPTION && PENDING_EXCEPTION->is_a(SystemDictionary::linkageError_klass())) { @@ -841,6 +846,12 @@ Handle protection_domain, TRAPS) { + // UseNewReflection + // The result of this call should be consistent with the result + // of the call to resolve_instance_class_or_null(). + // See evaluation 6790209 and 4474172 for more details. + class_loader = Handle(THREAD, java_lang_ClassLoader::non_reflection_class_loader(class_loader())); + unsigned int d_hash = dictionary()->compute_hash(class_name, class_loader); int d_index = dictionary()->hash_to_index(d_hash); @@ -955,10 +966,10 @@ instanceKlassHandle k = ClassFileParser(st).parseClassFile(class_name, class_loader, protection_domain, - cp_patches, parsed_name, THREAD); + // We don't redefine the class, so we just need to clean up whether there // was an error or not (don't want to modify any system dictionary // data structures). @@ -1013,11 +1024,17 @@ ClassFileStream* st, TRAPS) { - // Make sure we are synchronized on the class loader before we initiate - // loading. + // Classloaders that support parallelism, e.g. bootstrap classloader, + // or all classloaders with UnsyncloadClass do not acquire lock here + bool DoObjectLock = true; + if (is_parallelCapable(class_loader)) { + DoObjectLock = false; + } + + // Make sure we are synchronized on the class loader before we proceed Handle lockObject = compute_loader_lock_object(class_loader, THREAD); check_loader_lock_contention(lockObject, THREAD); - ObjectLocker ol(lockObject, THREAD); + ObjectLocker ol(lockObject, THREAD, DoObjectLock); symbolHandle parsed_name; @@ -1069,7 +1086,13 @@ "external class name format used internally"); // Add class just loaded - define_instance_class(k, THREAD); + // If a class loader supports parallel classloading handle parallel define requests + // find_or_define_instance_class may return a different instanceKlass + if (is_parallelCapable(class_loader)) { + k = find_or_define_instance_class(class_name, class_loader, k, THREAD); + } else { + define_instance_class(k, THREAD); + } } // If parsing the class file or define_instance_class failed, we @@ -1299,7 +1322,7 @@ } #endif // KERNEL - // find_or_define_instance_class may return a different k + // find_or_define_instance_class may return a different instanceKlass if (!k.is_null()) { k = find_or_define_instance_class(class_name, class_loader, k, CHECK_(nh)); } @@ -1316,14 +1339,24 @@ KlassHandle spec_klass (THREAD, SystemDictionary::classloader_klass()); - // UnsyncloadClass option means don't synchronize loadClass() calls. - // loadClassInternal() is synchronized and public loadClass(String) is not. - // This flag is for diagnostic purposes only. It is risky to call + // Call public unsynchronized loadClass(String) directly for all class loaders + // for parallelCapable class loaders. JDK >=7, loadClass(String, boolean) will + // acquire a class-name based lock rather than the class loader object lock. + // JDK < 7 already acquire the class loader lock in loadClass(String, boolean), + // so the call to loadClassInternal() was not required. + // + // UnsyncloadClass flag means both call loadClass(String) and do + // not acquire the class loader lock even for class loaders that are + // not parallelCapable. This was a risky transitional + // flag for diagnostic purposes only. It is risky to call // custom class loaders without synchronization. // WARNING If a custom class loader does NOT synchronizer findClass, or callers of - // findClass, this flag risks unexpected timing bugs in the field. + // findClass, the UnsyncloadClass flag risks unexpected timing bugs in the field. // Do NOT assume this will be supported in future releases. - if (!UnsyncloadClass && has_loadClassInternal()) { + // + // Added MustCallLoadClassInternal in case we discover in the field + // a customer that counts on this call + if (MustCallLoadClassInternal && has_loadClassInternal()) { JavaCalls::call_special(&result, class_loader, spec_klass, @@ -1365,14 +1398,17 @@ Handle class_loader_h(THREAD, k->class_loader()); - // for bootstrap classloader don't acquire lock - if (!class_loader_h.is_null()) { + // for bootstrap and other parallel classloaders don't acquire lock, + // use placeholder token + // If a parallelCapable class loader calls define_instance_class instead of + // find_or_define_instance_class to get here, we have a timing + // hole with systemDictionary updates and check_constraints + if (!class_loader_h.is_null() && !is_parallelCapable(class_loader_h)) { assert(ObjectSynchronizer::current_thread_holds_lock((JavaThread*)THREAD, compute_loader_lock_object(class_loader_h, THREAD)), "define called without lock"); } - // Check class-loading constraints. Throw exception if violation is detected. // Grabs and releases SystemDictionary_lock // The check_constraints/find_class call and update_dictionary sequence @@ -1427,59 +1463,63 @@ // Support parallel classloading // Initial implementation for bootstrap classloader -// For future: // For custom class loaders that support parallel classloading, -// in case they do not synchronize around -// FindLoadedClass/DefineClass calls, we check for parallel +// With AllowParallelDefine flag==true, in case they do not synchronize around +// FindLoadedClass/DefineClass, calls, we check for parallel // loading for them, wait if a defineClass is in progress // and return the initial requestor's results +// With AllowParallelDefine flag==false, call through to define_instance_class +// which will throw LinkageError: duplicate class definition. // For better performance, the class loaders should synchronize -// findClass(), i.e. FindLoadedClass/DefineClass or they +// findClass(), i.e. FindLoadedClass/DefineClassIfAbsent or they // potentially waste time reading and parsing the bytestream. // Note: VM callers should ensure consistency of k/class_name,class_loader instanceKlassHandle SystemDictionary::find_or_define_instance_class(symbolHandle class_name, Handle class_loader, instanceKlassHandle k, TRAPS) { instanceKlassHandle nh = instanceKlassHandle(); // null Handle + symbolHandle name_h(THREAD, k->name()); // passed in class_name may be null - unsigned int d_hash = dictionary()->compute_hash(class_name, class_loader); + unsigned int d_hash = dictionary()->compute_hash(name_h, class_loader); int d_index = dictionary()->hash_to_index(d_hash); // Hold SD lock around find_class and placeholder creation for DEFINE_CLASS - unsigned int p_hash = placeholders()->compute_hash(class_name, class_loader); + unsigned int p_hash = placeholders()->compute_hash(name_h, class_loader); int p_index = placeholders()->hash_to_index(p_hash); PlaceholderEntry* probe; { MutexLocker mu(SystemDictionary_lock, THREAD); // First check if class already defined - klassOop check = find_class(d_index, d_hash, class_name, class_loader); + klassOop check = find_class(d_index, d_hash, name_h, class_loader); if (check != NULL) { return(instanceKlassHandle(THREAD, check)); } // Acquire define token for this class/classloader symbolHandle nullsymbolHandle; - probe = placeholders()->find_and_add(p_index, p_hash, class_name, class_loader, PlaceholderTable::DEFINE_CLASS, nullsymbolHandle, THREAD); - // Check if another thread defining in parallel - if (probe->definer() == NULL) { - // Thread will define the class - probe->set_definer(THREAD); + probe = placeholders()->find_and_add(p_index, p_hash, name_h, class_loader, PlaceholderTable::DEFINE_CLASS, nullsymbolHandle, THREAD); + // Wait if another thread defining in parallel + // All threads wait - even those that will throw duplicate class: otherwise + // caller is surprised by LinkageError: duplicate, but findLoadedClass fails + // if other thread has not finished updating dictionary + while (probe->definer() != NULL) { + SystemDictionary_lock->wait(); + } + // Only special cases allow parallel defines and can use other thread's results + // Other cases fall through, and may run into duplicate defines + // caught by finding an entry in the SystemDictionary + if ((UnsyncloadClass || AllowParallelDefineClass) && (probe->instanceKlass() != NULL)) { + probe->remove_seen_thread(THREAD, PlaceholderTable::DEFINE_CLASS); + placeholders()->find_and_remove(p_index, p_hash, name_h, class_loader, THREAD); + SystemDictionary_lock->notify_all(); +#ifdef ASSERT + klassOop check = find_class(d_index, d_hash, name_h, class_loader); + assert(check != NULL, "definer missed recording success"); +#endif + return(instanceKlassHandle(THREAD, probe->instanceKlass())); } else { - // Wait for defining thread to finish and return results - while (probe->definer() != NULL) { - SystemDictionary_lock->wait(); - } - if (probe->instanceKlass() != NULL) { - probe->remove_seen_thread(THREAD, PlaceholderTable::DEFINE_CLASS); - return(instanceKlassHandle(THREAD, probe->instanceKlass())); - } else { - // If definer had an error, try again as any new thread would - probe->set_definer(THREAD); -#ifdef ASSERT - klassOop check = find_class(d_index, d_hash, class_name, class_loader); - assert(check == NULL, "definer missed recording success"); -#endif - } + // This thread will define the class (even if earlier thread tried and had an error) + probe->set_definer(THREAD); } } @@ -1490,7 +1530,7 @@ // definer must notify any waiting threads { MutexLocker mu(SystemDictionary_lock, THREAD); - PlaceholderEntry* probe = placeholders()->get_entry(p_index, p_hash, class_name, class_loader); + PlaceholderEntry* probe = placeholders()->get_entry(p_index, p_hash, name_h, class_loader); assert(probe != NULL, "DEFINE_CLASS placeholder lost?"); if (probe != NULL) { if (HAS_PENDING_EXCEPTION) { @@ -1501,6 +1541,7 @@ } probe->set_definer(NULL); probe->remove_seen_thread(THREAD, PlaceholderTable::DEFINE_CLASS); + placeholders()->find_and_remove(p_index, p_hash, name_h, class_loader, THREAD); SystemDictionary_lock->notify_all(); } } @@ -1512,7 +1553,6 @@ return k; } - Handle SystemDictionary::compute_loader_lock_object(Handle class_loader, TRAPS) { // If class_loader is NULL we synchronize on _system_loader_lock_obj if (class_loader.is_null()) { @@ -1902,11 +1942,11 @@ warning("Cannot find sun/jkernel/DownloadManager"); } #endif // KERNEL + { // Compute whether we should use loadClass or loadClassInternal when loading classes. methodOop method = instanceKlass::cast(classloader_klass())->find_method(vmSymbols::loadClassInternal_name(), vmSymbols::string_class_signature()); _has_loadClassInternal = (method != NULL); } - { // Compute whether we should use checkPackageAccess or NOT methodOop method = instanceKlass::cast(classloader_klass())->find_method(vmSymbols::checkPackageAccess_name(), vmSymbols::class_protectiondomain_signature()); _has_checkPackageAccess = (method != NULL);
--- a/src/share/vm/classfile/systemDictionary.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/classfile/systemDictionary.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -526,6 +526,7 @@ static instanceKlassHandle load_instance_class(symbolHandle class_name, Handle class_loader, TRAPS); static Handle compute_loader_lock_object(Handle class_loader, TRAPS); static void check_loader_lock_contention(Handle loader_lock, TRAPS); + static bool is_parallelCapable(Handle class_loader); static klassOop find_shared_class(symbolHandle class_name);
--- a/src/share/vm/classfile/vmSymbols.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/classfile/vmSymbols.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -362,6 +362,7 @@ template(class_signature, "Ljava/lang/Class;") \ template(string_signature, "Ljava/lang/String;") \ template(reference_signature, "Ljava/lang/ref/Reference;") \ + template(concurrenthashmap_signature, "Ljava/util/concurrent/ConcurrentHashMap;") \ /* signature symbols needed by intrinsics */ \ VM_INTRINSICS_DO(VM_INTRINSIC_IGNORE, VM_SYMBOL_IGNORE, VM_SYMBOL_IGNORE, template, VM_ALIAS_IGNORE) \ \ @@ -374,6 +375,9 @@ /* used by ClassFormatError when class name is not known yet */ \ template(unknown_class_name, "<Unknown>") \ \ + /* used to identify class loaders handling parallel class loading */ \ + template(parallelCapable_name, "parallelLockMap;") \ + \ /* JVM monitoring and management support */ \ template(java_lang_StackTraceElement_array, "[Ljava/lang/StackTraceElement;") \ template(java_lang_management_ThreadState, "java/lang/management/ThreadState") \
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -706,6 +706,30 @@ } } +// Apply the given closure to each live object in the space +// The usage of CompactibleFreeListSpace +// by the ConcurrentMarkSweepGeneration for concurrent GC's allows +// objects in the space with references to objects that are no longer +// valid. For example, an object may reference another object +// that has already been sweep up (collected). This method uses +// obj_is_alive() to determine whether it is safe to apply the closure to +// an object. See obj_is_alive() for details on how liveness of an +// object is decided. + +void CompactibleFreeListSpace::safe_object_iterate(ObjectClosure* blk) { + assert_lock_strong(freelistLock()); + NOT_PRODUCT(verify_objects_initialized()); + HeapWord *cur, *limit; + size_t curSize; + for (cur = bottom(), limit = end(); cur < limit; + cur += curSize) { + curSize = block_size(cur); + if (block_is_obj(cur) && obj_is_alive(cur)) { + blk->do_object(oop(cur)); + } + } +} + void CompactibleFreeListSpace::object_iterate_mem(MemRegion mr, UpwardsObjectClosure* cl) { assert_locked(); @@ -861,7 +885,9 @@ } else { // must read from what 'p' points to in each loop. klassOop k = ((volatile oopDesc*)p)->klass_or_null(); - if (k != NULL && ((oopDesc*)p)->is_parsable()) { + if (k != NULL && + ((oopDesc*)p)->is_parsable() && + ((oopDesc*)p)->is_conc_safe()) { assert(k->is_oop(), "Should really be klass oop."); oop o = (oop)p; assert(o->is_oop(), "Should be an oop");
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -481,6 +481,15 @@ void oop_iterate(OopClosure* cl); void object_iterate(ObjectClosure* blk); + // Apply the closure to each object in the space whose references + // point to objects in the heap. The usage of CompactibleFreeListSpace + // by the ConcurrentMarkSweepGeneration for concurrent GC's allows + // objects in the space with references to objects that are no longer + // valid. For example, an object may reference another object + // that has already been sweep up (collected). This method uses + // obj_is_alive() to determine whether it is safe to iterate of + // an object. + void safe_object_iterate(ObjectClosure* blk); void object_iterate_mem(MemRegion mr, UpwardsObjectClosure* cl); // Requires that "mr" be entirely within the space.
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -3018,6 +3018,16 @@ } void +ConcurrentMarkSweepGeneration::safe_object_iterate(ObjectClosure* cl) { + if (freelistLock()->owned_by_self()) { + Generation::safe_object_iterate(cl); + } else { + MutexLockerEx x(freelistLock(), Mutex::_no_safepoint_check_flag); + Generation::safe_object_iterate(cl); + } +} + +void ConcurrentMarkSweepGeneration::pre_adjust_pointers() { } @@ -6623,7 +6633,11 @@ if (_bitMap->isMarked(addr)) { // it's marked; is it potentially uninitialized? if (p->klass_or_null() != NULL) { - if (CMSPermGenPrecleaningEnabled && !p->is_parsable()) { + // If is_conc_safe is false, the object may be undergoing + // change by the VM outside a safepoint. Don't try to + // scan it, but rather leave it for the remark phase. + if (CMSPermGenPrecleaningEnabled && + (!p->is_conc_safe() || !p->is_parsable())) { // Signal precleaning to redirty the card since // the klass pointer is already installed. assert(size == 0, "Initial value"); @@ -7001,7 +7015,6 @@ _mut->clear_range(mr); } DEBUG_ONLY(}) - // Note: the finger doesn't advance while we drain // the stack below. PushOrMarkClosure pushOrMarkClosure(_collector, @@ -8062,9 +8075,13 @@ #ifdef DEBUG if (oop(addr)->klass_or_null() != NULL && ( !_collector->should_unload_classes() - || oop(addr)->is_parsable())) { + || (oop(addr)->is_parsable()) && + oop(addr)->is_conc_safe())) { // Ignore mark word because we are running concurrent with mutators assert(oop(addr)->is_oop(true), "live block should be an oop"); + // is_conc_safe is checked before performing this assertion + // because an object that is not is_conc_safe may yet have + // the return from size() correct. assert(size == CompactibleFreeListSpace::adjustObjectSize(oop(addr)->size()), "P-mark and computed size do not agree"); @@ -8077,6 +8094,13 @@ (!_collector->should_unload_classes() || oop(addr)->is_parsable()), "Should be an initialized object"); + // Note that there are objects used during class redefinition + // (e.g., merge_cp in VM_RedefineClasses::merge_cp_and_rewrite() + // which are discarded with their is_conc_safe state still + // false. These object may be floating garbage so may be + // seen here. If they are floating garbage their size + // should be attainable from their klass. Do not that + // is_conc_safe() is true for oop(addr). // Ignore mark word because we are running concurrent with mutators assert(oop(addr)->is_oop(true), "live block should be an oop"); // Verify that the bit map has no bits marked between @@ -8484,7 +8508,7 @@ size_t i = num; oop cur = _overflow_list; const markOop proto = markOopDesc::prototype(); - NOT_PRODUCT(size_t n = 0;) + NOT_PRODUCT(ssize_t n = 0;) for (oop next; i > 0 && cur != NULL; cur = next, i--) { next = oop(cur->mark()); cur->set_mark(proto); // until proven otherwise @@ -8501,45 +8525,131 @@ return !stack->isEmpty(); } -// Multi-threaded; use CAS to break off a prefix +#define BUSY (oop(0x1aff1aff)) +// (MT-safe) Get a prefix of at most "num" from the list. +// The overflow list is chained through the mark word of +// each object in the list. We fetch the entire list, +// break off a prefix of the right size and return the +// remainder. If other threads try to take objects from +// the overflow list at that time, they will wait for +// some time to see if data becomes available. If (and +// only if) another thread places one or more object(s) +// on the global list before we have returned the suffix +// to the global list, we will walk down our local list +// to find its end and append the global list to +// our suffix before returning it. This suffix walk can +// prove to be expensive (quadratic in the amount of traffic) +// when there are many objects in the overflow list and +// there is much producer-consumer contention on the list. +// *NOTE*: The overflow list manipulation code here and +// in ParNewGeneration:: are very similar in shape, +// except that in the ParNew case we use the old (from/eden) +// copy of the object to thread the list via its klass word. +// Because of the common code, if you make any changes in +// the code below, please check the ParNew version to see if +// similar changes might be needed. +// CR 6797058 has been filed to consolidate the common code. bool CMSCollector::par_take_from_overflow_list(size_t num, OopTaskQueue* work_q) { - assert(work_q->size() == 0, "That's the current policy"); + assert(work_q->size() == 0, "First empty local work queue"); assert(num < work_q->max_elems(), "Can't bite more than we can chew"); if (_overflow_list == NULL) { return false; } // Grab the entire list; we'll put back a suffix - oop prefix = (oop)Atomic::xchg_ptr(NULL, &_overflow_list); - if (prefix == NULL) { // someone grabbed it before we did ... - // ... we could spin for a short while, but for now we don't - return false; - } + oop prefix = (oop)Atomic::xchg_ptr(BUSY, &_overflow_list); + Thread* tid = Thread::current(); + size_t CMSOverflowSpinCount = (size_t)ParallelGCThreads; + size_t sleep_time_millis = MAX2((size_t)1, num/100); + // If the list is busy, we spin for a short while, + // sleeping between attempts to get the list. + for (size_t spin = 0; prefix == BUSY && spin < CMSOverflowSpinCount; spin++) { + os::sleep(tid, sleep_time_millis, false); + if (_overflow_list == NULL) { + // Nothing left to take + return false; + } else if (_overflow_list != BUSY) { + // Try and grab the prefix + prefix = (oop)Atomic::xchg_ptr(BUSY, &_overflow_list); + } + } + // If the list was found to be empty, or we spun long + // enough, we give up and return empty-handed. If we leave + // the list in the BUSY state below, it must be the case that + // some other thread holds the overflow list and will set it + // to a non-BUSY state in the future. + if (prefix == NULL || prefix == BUSY) { + // Nothing to take or waited long enough + if (prefix == NULL) { + // Write back the NULL in case we overwrote it with BUSY above + // and it is still the same value. + (void) Atomic::cmpxchg_ptr(NULL, &_overflow_list, BUSY); + } + return false; + } + assert(prefix != NULL && prefix != BUSY, "Error"); size_t i = num; oop cur = prefix; + // Walk down the first "num" objects, unless we reach the end. for (; i > 1 && cur->mark() != NULL; cur = oop(cur->mark()), i--); - if (cur->mark() != NULL) { + if (cur->mark() == NULL) { + // We have "num" or fewer elements in the list, so there + // is nothing to return to the global list. + // Write back the NULL in lieu of the BUSY we wrote + // above, if it is still the same value. + if (_overflow_list == BUSY) { + (void) Atomic::cmpxchg_ptr(NULL, &_overflow_list, BUSY); + } + } else { + // Chop off the suffix and rerturn it to the global list. + assert(cur->mark() != BUSY, "Error"); oop suffix_head = cur->mark(); // suffix will be put back on global list cur->set_mark(NULL); // break off suffix - // Find tail of suffix so we can prepend suffix to global list - for (cur = suffix_head; cur->mark() != NULL; cur = (oop)(cur->mark())); - oop suffix_tail = cur; - assert(suffix_tail != NULL && suffix_tail->mark() == NULL, - "Tautology"); + // It's possible that the list is still in the empty(busy) state + // we left it in a short while ago; in that case we may be + // able to place back the suffix without incurring the cost + // of a walk down the list. oop observed_overflow_list = _overflow_list; - do { - cur = observed_overflow_list; - suffix_tail->set_mark(markOop(cur)); + oop cur_overflow_list = observed_overflow_list; + bool attached = false; + while (observed_overflow_list == BUSY || observed_overflow_list == NULL) { observed_overflow_list = - (oop) Atomic::cmpxchg_ptr(suffix_head, &_overflow_list, cur); - } while (cur != observed_overflow_list); + (oop) Atomic::cmpxchg_ptr(suffix_head, &_overflow_list, cur_overflow_list); + if (cur_overflow_list == observed_overflow_list) { + attached = true; + break; + } else cur_overflow_list = observed_overflow_list; + } + if (!attached) { + // Too bad, someone else sneaked in (at least) an element; we'll need + // to do a splice. Find tail of suffix so we can prepend suffix to global + // list. + for (cur = suffix_head; cur->mark() != NULL; cur = (oop)(cur->mark())); + oop suffix_tail = cur; + assert(suffix_tail != NULL && suffix_tail->mark() == NULL, + "Tautology"); + observed_overflow_list = _overflow_list; + do { + cur_overflow_list = observed_overflow_list; + if (cur_overflow_list != BUSY) { + // Do the splice ... + suffix_tail->set_mark(markOop(cur_overflow_list)); + } else { // cur_overflow_list == BUSY + suffix_tail->set_mark(NULL); + } + // ... and try to place spliced list back on overflow_list ... + observed_overflow_list = + (oop) Atomic::cmpxchg_ptr(suffix_head, &_overflow_list, cur_overflow_list); + } while (cur_overflow_list != observed_overflow_list); + // ... until we have succeeded in doing so. + } } // Push the prefix elements on work_q assert(prefix != NULL, "control point invariant"); const markOop proto = markOopDesc::prototype(); oop next; - NOT_PRODUCT(size_t n = 0;) + NOT_PRODUCT(ssize_t n = 0;) for (cur = prefix; cur != NULL; cur = next) { next = oop(cur->mark()); cur->set_mark(proto); // until proven otherwise @@ -8573,11 +8683,16 @@ oop cur_overflow_list; do { cur_overflow_list = observed_overflow_list; - p->set_mark(markOop(cur_overflow_list)); + if (cur_overflow_list != BUSY) { + p->set_mark(markOop(cur_overflow_list)); + } else { + p->set_mark(NULL); + } observed_overflow_list = (oop) Atomic::cmpxchg_ptr(p, &_overflow_list, cur_overflow_list); } while (cur_overflow_list != observed_overflow_list); } +#undef BUSY // Single threaded // General Note on GrowableArray: pushes may silently fail @@ -8586,7 +8701,7 @@ // a lot of code in the JVM. The prudent thing for GrowableArray // to do (for now) is to exit with an error. However, that may // be too draconian in some cases because the caller may be -// able to recover without much harm. For suych cases, we +// able to recover without much harm. For such cases, we // should probably introduce a "soft_push" method which returns // an indication of success or failure with the assumption that // the caller may be able to recover from a failure; code in @@ -8594,8 +8709,6 @@ // failures where possible, thus, incrementally hardening the VM // in such low resource situations. void CMSCollector::preserve_mark_work(oop p, markOop m) { - int PreserveMarkStackSize = 128; - if (_preserved_oop_stack == NULL) { assert(_preserved_mark_stack == NULL, "bijection with preserved_oop_stack");
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -595,7 +595,7 @@ size_t _ser_kac_preclean_ovflw; size_t _ser_kac_ovflw; size_t _par_kac_ovflw; - NOT_PRODUCT(size_t _num_par_pushes;) + NOT_PRODUCT(ssize_t _num_par_pushes;) // ("Weak") Reference processing support ReferenceProcessor* _ref_processor; @@ -1212,6 +1212,7 @@ // More iteration support virtual void oop_iterate(MemRegion mr, OopClosure* cl); virtual void oop_iterate(OopClosure* cl); + virtual void safe_object_iterate(ObjectClosure* cl); virtual void object_iterate(ObjectClosure* cl); // Need to declare the full complement of closures, whether we'll
--- a/src/share/vm/gc_implementation/g1/collectionSetChooser.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/collectionSetChooser.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -24,7 +24,7 @@ // We need to sort heap regions by collection desirability. -class CSetChooserCache { +class CSetChooserCache VALUE_OBJ_CLASS_SPEC { private: enum { CacheLength = 16
--- a/src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -33,7 +33,7 @@ PYA_cancel // It's been completed by somebody else: cancel. }; -class ConcurrentG1Refine { +class ConcurrentG1Refine: public CHeapObj { ConcurrentG1RefineThread* _cg1rThread; volatile jint _pya;
--- a/src/share/vm/gc_implementation/g1/concurrentMark.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/concurrentMark.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -30,7 +30,7 @@ // A generic CM bit map. This is essentially a wrapper around the BitMap // class, with one bit per (1<<_shifter) HeapWords. -class CMBitMapRO { +class CMBitMapRO VALUE_OBJ_CLASS_SPEC { protected: HeapWord* _bmStartWord; // base address of range covered by map size_t _bmWordSize; // map size (in #HeapWords covered) @@ -139,7 +139,7 @@ // Represents a marking stack used by the CM collector. // Ideally this should be GrowableArray<> just like MSC's marking stack(s). -class CMMarkStack { +class CMMarkStack VALUE_OBJ_CLASS_SPEC { ConcurrentMark* _cm; oop* _base; // bottom of stack jint _index; // one more than last occupied index @@ -237,7 +237,7 @@ void oops_do(OopClosure* f); }; -class CMRegionStack { +class CMRegionStack VALUE_OBJ_CLASS_SPEC { MemRegion* _base; jint _capacity; jint _index; @@ -312,7 +312,7 @@ class ConcurrentMarkThread; -class ConcurrentMark { +class ConcurrentMark: public CHeapObj { friend class ConcurrentMarkThread; friend class CMTask; friend class CMBitMapClosure;
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -141,7 +141,7 @@ _scan_only_head(NULL), _scan_only_tail(NULL), _curr_scan_only(NULL), _length(0), _scan_only_length(0), _last_sampled_rs_lengths(0), - _survivor_head(NULL), _survivors_tail(NULL), _survivor_length(0) + _survivor_head(NULL), _survivor_tail(NULL), _survivor_length(0) { guarantee( check_list_empty(false), "just making sure..." ); } @@ -159,16 +159,15 @@ } void YoungList::add_survivor_region(HeapRegion* hr) { - assert(!hr->is_survivor(), "should not already be for survived"); + assert(hr->is_survivor(), "should be flagged as survivor region"); assert(hr->get_next_young_region() == NULL, "cause it should!"); hr->set_next_young_region(_survivor_head); if (_survivor_head == NULL) { - _survivors_tail = hr; + _survivor_tail = hr; } _survivor_head = hr; - hr->set_survivor(); ++_survivor_length; } @@ -239,7 +238,7 @@ empty_list(_survivor_head); _survivor_head = NULL; - _survivors_tail = NULL; + _survivor_tail = NULL; _survivor_length = 0; _last_sampled_rs_lengths = 0; @@ -391,6 +390,7 @@ // Add survivor regions to SurvRateGroup. _g1h->g1_policy()->note_start_adding_survivor_regions(); + _g1h->g1_policy()->finished_recalculating_age_indexes(true /* is_survivors */); for (HeapRegion* curr = _survivor_head; curr != NULL; curr = curr->get_next_young_region()) { @@ -401,7 +401,7 @@ if (_survivor_head != NULL) { _head = _survivor_head; _length = _survivor_length + _scan_only_length; - _survivors_tail->set_next_young_region(_scan_only_head); + _survivor_tail->set_next_young_region(_scan_only_head); } else { _head = _scan_only_head; _length = _scan_only_length; @@ -418,9 +418,9 @@ _curr_scan_only = NULL; _survivor_head = NULL; - _survivors_tail = NULL; + _survivor_tail = NULL; _survivor_length = 0; - _g1h->g1_policy()->finished_recalculating_age_indexes(); + _g1h->g1_policy()->finished_recalculating_age_indexes(false /* is_survivors */); assert(check_list_well_formed(), "young list should be well formed"); } @@ -553,7 +553,7 @@ if (_gc_alloc_region_counts[purpose] < g1_policy()->max_regions(purpose)) { alloc_region = newAllocRegion_work(word_size, true, zero_filled); if (purpose == GCAllocForSurvived && alloc_region != NULL) { - _young_list->add_survivor_region(alloc_region); + alloc_region->set_survivor(); } ++_gc_alloc_region_counts[purpose]; } else { @@ -949,6 +949,10 @@ GCOverheadReporter::recordSTWEnd(end); g1_policy()->record_full_collection_end(); +#ifdef TRACESPINNING + ParallelTaskTerminator::print_termination_counts(); +#endif + gc_epilogue(true); // Abandon concurrent refinement. This must happen last: in the @@ -1285,7 +1289,9 @@ _unclean_regions_coming(false), _young_list(new YoungList(this)), _gc_time_stamp(0), - _surviving_young_words(NULL) + _surviving_young_words(NULL), + _in_cset_fast_test(NULL), + _in_cset_fast_test_base(NULL) { _g1h = this; // To catch bugs. if (_process_strong_tasks == NULL || !_process_strong_tasks->valid()) { @@ -2485,6 +2491,19 @@ g1_policy()->record_collection_pause_start(start_time_sec, start_used_bytes); + guarantee(_in_cset_fast_test == NULL, "invariant"); + guarantee(_in_cset_fast_test_base == NULL, "invariant"); + _in_cset_fast_test_length = n_regions(); + _in_cset_fast_test_base = + NEW_C_HEAP_ARRAY(bool, _in_cset_fast_test_length); + memset(_in_cset_fast_test_base, false, + _in_cset_fast_test_length * sizeof(bool)); + // We're biasing _in_cset_fast_test to avoid subtracting the + // beginning of the heap every time we want to index; basically + // it's the same with what we do with the card table. + _in_cset_fast_test = _in_cset_fast_test_base - + ((size_t) _g1_reserved.start() >> HeapRegion::LogOfHRGrainBytes); + #if SCAN_ONLY_VERBOSE _young_list->print(); #endif // SCAN_ONLY_VERBOSE @@ -2553,6 +2572,12 @@ free_collection_set(g1_policy()->collection_set()); g1_policy()->clear_collection_set(); + FREE_C_HEAP_ARRAY(bool, _in_cset_fast_test_base); + // this is more for peace of mind; we're nulling them here and + // we're expecting them to be null at the beginning of the next GC + _in_cset_fast_test = NULL; + _in_cset_fast_test_base = NULL; + if (popular_region != NULL) { // We have to wait until now, because we don't want the region to // be rescheduled for pop-evac during RS update. @@ -2572,6 +2597,9 @@ _young_list->print(); #endif // SCAN_ONLY_VERBOSE + 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 { @@ -2598,7 +2626,9 @@ #endif // SCAN_ONLY_VERBOSE double end_time_sec = os::elapsedTime(); - g1_policy()->record_pause_time((end_time_sec - start_time_sec)*1000.0); + if (!evacuation_failed()) { + g1_policy()->record_pause_time((end_time_sec - start_time_sec)*1000.0); + } GCOverheadReporter::recordSTWEnd(end_time_sec); g1_policy()->record_collection_pause_end(popular_region != NULL, abandoned); @@ -2621,8 +2651,13 @@ } } - if (mark_in_progress()) + if (mark_in_progress()) { concurrent_mark()->update_g1_committed(); + } + +#ifdef TRACESPINNING + ParallelTaskTerminator::print_termination_counts(); +#endif gc_epilogue(false); } @@ -2733,6 +2768,13 @@ _gc_alloc_region_list = r->next_gc_alloc_region(); r->set_next_gc_alloc_region(NULL); r->set_is_gc_alloc_region(false); + if (r->is_survivor()) { + if (r->is_empty()) { + r->set_not_young(); + } else { + _young_list->add_survivor_region(r); + } + } if (r->is_empty()) { ++_free_regions; } @@ -3129,6 +3171,20 @@ return block; } +void G1CollectedHeap::retire_alloc_region(HeapRegion* alloc_region, + bool par) { + // Another thread might have obtained alloc_region for the given + // purpose, and might be attempting to allocate in it, and might + // succeed. Therefore, we can't do the "finalization" stuff on the + // region below until we're sure the last allocation has happened. + // We ensure this by allocating the remaining space with a garbage + // object. + if (par) par_allocate_remaining_space(alloc_region); + // Now we can do the post-GC stuff on the region. + alloc_region->note_end_of_copying(); + g1_policy()->record_after_bytes(alloc_region->used()); +} + HeapWord* G1CollectedHeap::allocate_during_gc_slow(GCAllocPurpose purpose, HeapRegion* alloc_region, @@ -3146,16 +3202,7 @@ // Otherwise, continue; this new region is empty, too. } assert(alloc_region != NULL, "We better have an allocation region"); - // Another thread might have obtained alloc_region for the given - // purpose, and might be attempting to allocate in it, and might - // succeed. Therefore, we can't do the "finalization" stuff on the - // region below until we're sure the last allocation has happened. - // We ensure this by allocating the remaining space with a garbage - // object. - if (par) par_allocate_remaining_space(alloc_region); - // Now we can do the post-GC stuff on the region. - alloc_region->note_end_of_copying(); - g1_policy()->record_after_bytes(alloc_region->used()); + retire_alloc_region(alloc_region, par); if (_gc_alloc_region_counts[purpose] >= g1_policy()->max_regions(purpose)) { // Cannot allocate more regions for the given purpose. @@ -3164,7 +3211,7 @@ if (purpose != alt_purpose) { HeapRegion* alt_region = _gc_alloc_regions[alt_purpose]; // Has not the alternative region been aliased? - if (alloc_region != alt_region) { + if (alloc_region != alt_region && alt_region != NULL) { // Try to allocate in the alternative region. if (par) { block = alt_region->par_allocate(word_size); @@ -3173,9 +3220,10 @@ } // Make an alias. _gc_alloc_regions[purpose] = _gc_alloc_regions[alt_purpose]; - } - if (block != NULL) { - return block; + if (block != NULL) { + return block; + } + retire_alloc_region(alt_region, par); } // Both the allocation region and the alternative one are full // and aliased, replace them with a new allocation region. @@ -3476,6 +3524,7 @@ OverflowQueue* _overflowed_refs; G1ParGCAllocBuffer _alloc_buffers[GCAllocPurposeCount]; + ageTable _age_table; size_t _alloc_buffer_waste; size_t _undo_waste; @@ -3517,6 +3566,7 @@ _refs(g1h->task_queue(queue_num)), _hash_seed(17), _queue_num(queue_num), _term_attempts(0), + _age_table(false), #if G1_DETAILED_STATS _pushes(0), _pops(0), _steals(0), _steal_attempts(0), _overflow_pushes(0), @@ -3551,8 +3601,9 @@ RefToScanQueue* refs() { return _refs; } OverflowQueue* overflowed_refs() { return _overflowed_refs; } - - inline G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose) { + ageTable* age_table() { return &_age_table; } + + G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose) { return &_alloc_buffers[purpose]; } @@ -3560,6 +3611,9 @@ size_t undo_waste() { return _undo_waste; } void push_on_queue(oop* ref) { + assert(ref != NULL, "invariant"); + assert(has_partial_array_mask(ref) || _g1h->obj_in_cs(*ref), "invariant"); + if (!refs()->push(ref)) { overflowed_refs()->push(ref); IF_G1_DETAILED_STATS(note_overflow_push()); @@ -3572,6 +3626,10 @@ if (!refs()->pop_local(ref)) { ref = NULL; } else { + assert(ref != NULL, "invariant"); + assert(has_partial_array_mask(ref) || _g1h->obj_in_cs(*ref), + "invariant"); + IF_G1_DETAILED_STATS(note_pop()); } } @@ -3601,8 +3659,7 @@ obj = alloc_buf->allocate(word_sz); assert(obj != NULL, "buffer was definitely big enough..."); - } - else { + } else { obj = _g1h->par_allocate_during_gc(purpose, word_sz); } return obj; @@ -3695,24 +3752,57 @@ } } +private: + void deal_with_reference(oop* ref_to_scan) { + if (has_partial_array_mask(ref_to_scan)) { + _partial_scan_cl->do_oop_nv(ref_to_scan); + } else { + // Note: we can use "raw" versions of "region_containing" because + // "obj_to_scan" is definitely in the heap, and is not in a + // humongous region. + HeapRegion* r = _g1h->heap_region_containing_raw(ref_to_scan); + _evac_cl->set_region(r); + _evac_cl->do_oop_nv(ref_to_scan); + } + } + +public: void trim_queue() { + // I've replicated the loop twice, first to drain the overflow + // queue, second to drain the task queue. This is better than + // having a single loop, which checks both conditions and, inside + // it, either pops the overflow queue or the task queue, as each + // loop is tighter. Also, the decision to drain the overflow queue + // first is not arbitrary, as the overflow queue is not visible + // to the other workers, whereas the task queue is. So, we want to + // drain the "invisible" entries first, while allowing the other + // workers to potentially steal the "visible" entries. + while (refs_to_scan() > 0 || overflowed_refs_to_scan() > 0) { - oop *ref_to_scan = NULL; - if (overflowed_refs_to_scan() == 0) { - pop_from_queue(ref_to_scan); - } else { + while (overflowed_refs_to_scan() > 0) { + oop *ref_to_scan = NULL; pop_from_overflow_queue(ref_to_scan); + assert(ref_to_scan != NULL, "invariant"); + // We shouldn't have pushed it on the queue if it was not + // pointing into the CSet. + assert(ref_to_scan != NULL, "sanity"); + assert(has_partial_array_mask(ref_to_scan) || + _g1h->obj_in_cs(*ref_to_scan), "sanity"); + + deal_with_reference(ref_to_scan); } - if (ref_to_scan != NULL) { - if ((intptr_t)ref_to_scan & G1_PARTIAL_ARRAY_MASK) { - _partial_scan_cl->do_oop_nv(ref_to_scan); - } else { - // Note: we can use "raw" versions of "region_containing" because - // "obj_to_scan" is definitely in the heap, and is not in a - // humongous region. - HeapRegion* r = _g1h->heap_region_containing_raw(ref_to_scan); - _evac_cl->set_region(r); - _evac_cl->do_oop_nv(ref_to_scan); + + while (refs_to_scan() > 0) { + oop *ref_to_scan = NULL; + pop_from_queue(ref_to_scan); + + if (ref_to_scan != NULL) { + // We shouldn't have pushed it on the queue if it was not + // pointing into the CSet. + assert(has_partial_array_mask(ref_to_scan) || + _g1h->obj_in_cs(*ref_to_scan), "sanity"); + + deal_with_reference(ref_to_scan); } } } @@ -3728,16 +3818,25 @@ // Should probably be made inline and moved in g1OopClosures.inline.hpp. void G1ParScanClosure::do_oop_nv(oop* p) { oop obj = *p; + if (obj != NULL) { - if (_g1->obj_in_cs(obj)) { - if (obj->is_forwarded()) { - *p = obj->forwardee(); - } else { - _par_scan_state->push_on_queue(p); - return; - } + if (_g1->in_cset_fast_test(obj)) { + // We're not going to even bother checking whether the object is + // already forwarded or not, as this usually causes an immediate + // stall. We'll try to prefetch the object (for write, given that + // we might need to install the forwarding reference) and we'll + // get back to it when pop it from the queue + Prefetch::write(obj->mark_addr(), 0); + Prefetch::read(obj->mark_addr(), (HeapWordSize*2)); + + // slightly paranoid test; I'm trying to catch potential + // problems before we go into push_on_queue to know where the + // problem is coming from + assert(obj == *p, "the value of *p should not have changed"); + _par_scan_state->push_on_queue(p); + } else { + _g1_rem->par_write_ref(_from, p, _par_scan_state->queue_num()); } - _g1_rem->par_write_ref(_from, p, _par_scan_state->queue_num()); } } @@ -3765,7 +3864,9 @@ (!from_region->is_young() && young_index == 0), "invariant" ); G1CollectorPolicy* g1p = _g1->g1_policy(); markOop m = old->mark(); - GCAllocPurpose alloc_purpose = g1p->evacuation_destination(from_region, m->age(), + int age = m->has_displaced_mark_helper() ? m->displaced_mark_helper()->age() + : m->age(); + GCAllocPurpose alloc_purpose = g1p->evacuation_destination(from_region, age, word_sz); HeapWord* obj_ptr = _par_scan_state->allocate(alloc_purpose, word_sz); oop obj = oop(obj_ptr); @@ -3777,13 +3878,39 @@ return _g1->handle_evacuation_failure_par(cl, old); } + // We're going to allocate linearly, so might as well prefetch ahead. + Prefetch::write(obj_ptr, PrefetchCopyIntervalInBytes); + oop forward_ptr = old->forward_to_atomic(obj); if (forward_ptr == NULL) { Copy::aligned_disjoint_words((HeapWord*) old, obj_ptr, word_sz); - obj->set_mark(m); if (g1p->track_object_age(alloc_purpose)) { - obj->incr_age(); + // We could simply do obj->incr_age(). However, this causes a + // performance issue. obj->incr_age() will first check whether + // the object has a displaced mark by checking its mark word; + // getting the mark word from the new location of the object + // stalls. So, given that we already have the mark word and we + // are about to install it anyway, it's better to increase the + // age on the mark word, when the object does not have a + // displaced mark word. We're not expecting many objects to have + // a displaced marked word, so that case is not optimized + // further (it could be...) and we simply call obj->incr_age(). + + if (m->has_displaced_mark_helper()) { + // in this case, we have to install the mark word first, + // otherwise obj looks to be forwarded (the old mark word, + // which contains the forward pointer, was copied) + obj->set_mark(m); + obj->incr_age(); + } else { + m = m->incr_age(); + obj->set_mark(m); + } + _par_scan_state->age_table()->add(obj, word_sz); + } else { + obj->set_mark(m); } + // preserve "next" mark bit if (_g1->mark_in_progress() && !_g1->is_obj_ill(old)) { if (!use_local_bitmaps || @@ -3805,9 +3932,11 @@ if (obj->is_objArray() && arrayOop(obj)->length() >= ParGCArrayScanChunk) { arrayOop(old)->set_length(0); - _par_scan_state->push_on_queue((oop*) ((intptr_t)old | G1_PARTIAL_ARRAY_MASK)); + _par_scan_state->push_on_queue(set_partial_array_mask(old)); } else { - _scanner->set_region(_g1->heap_region_containing(obj)); + // No point in using the slower heap_region_containing() method, + // given that we know obj is in the heap. + _scanner->set_region(_g1->heap_region_containing_raw(obj)); obj->oop_iterate_backwards(_scanner); } } else { @@ -3817,47 +3946,55 @@ return obj; } -template<bool do_gen_barrier, G1Barrier barrier, bool do_mark_forwardee> -void G1ParCopyClosure<do_gen_barrier, barrier, do_mark_forwardee>::do_oop_work(oop* p) { +template<bool do_gen_barrier, G1Barrier barrier, + bool do_mark_forwardee, bool skip_cset_test> +void G1ParCopyClosure<do_gen_barrier, barrier, + do_mark_forwardee, skip_cset_test>::do_oop_work(oop* p) { oop obj = *p; assert(barrier != G1BarrierRS || obj != NULL, "Precondition: G1BarrierRS implies obj is nonNull"); - if (obj != NULL) { - if (_g1->obj_in_cs(obj)) { + // The only time we skip the cset test is when we're scanning + // references popped from the queue. And we only push on the queue + // references that we know point into the cset, so no point in + // checking again. But we'll leave an assert here for peace of mind. + assert(!skip_cset_test || _g1->obj_in_cs(obj), "invariant"); + + // here the null check is implicit in the cset_fast_test() test + if (skip_cset_test || _g1->in_cset_fast_test(obj)) { #if G1_REM_SET_LOGGING - gclog_or_tty->print_cr("Loc "PTR_FORMAT" contains pointer "PTR_FORMAT" into CS.", - p, (void*) obj); + gclog_or_tty->print_cr("Loc "PTR_FORMAT" contains pointer "PTR_FORMAT" " + "into CS.", p, (void*) obj); #endif - if (obj->is_forwarded()) { - *p = obj->forwardee(); - } else { - *p = copy_to_survivor_space(obj); - } - // When scanning the RS, we only care about objs in CS. - if (barrier == G1BarrierRS) { - _g1_rem->par_write_ref(_from, p, _par_scan_state->queue_num()); - } + if (obj->is_forwarded()) { + *p = obj->forwardee(); + } else { + *p = copy_to_survivor_space(obj); } - // When scanning moved objs, must look at all oops. - if (barrier == G1BarrierEvac) { + // When scanning the RS, we only care about objs in CS. + if (barrier == G1BarrierRS) { _g1_rem->par_write_ref(_from, p, _par_scan_state->queue_num()); } - - if (do_gen_barrier) { - par_do_barrier(p); - } - } -} - -template void G1ParCopyClosure<false, G1BarrierEvac, false>::do_oop_work(oop* p); - -template <class T> void G1ParScanPartialArrayClosure::process_array_chunk( + } + + // When scanning moved objs, must look at all oops. + if (barrier == G1BarrierEvac && obj != NULL) { + _g1_rem->par_write_ref(_from, p, _par_scan_state->queue_num()); + } + + if (do_gen_barrier && obj != NULL) { + par_do_barrier(p); + } +} + +template void G1ParCopyClosure<false, G1BarrierEvac, false, true>::do_oop_work(oop* p); + +template<class T> void G1ParScanPartialArrayClosure::process_array_chunk( oop obj, int start, int end) { // process our set of indices (include header in first chunk) assert(start < end, "invariant"); T* const base = (T*)objArrayOop(obj)->base(); - T* const start_addr = base + start; + T* const start_addr = (start == 0) ? (T*) obj : base + start; T* const end_addr = base + end; MemRegion mr((HeapWord*)start_addr, (HeapWord*)end_addr); _scanner.set_region(_g1->heap_region_containing(obj)); @@ -3866,7 +4003,8 @@ void G1ParScanPartialArrayClosure::do_oop_nv(oop* p) { assert(!UseCompressedOops, "Needs to be fixed to work with compressed oops"); - oop old = oop((intptr_t)p & ~G1_PARTIAL_ARRAY_MASK); + assert(has_partial_array_mask(p), "invariant"); + oop old = clear_partial_array_mask(p); assert(old->is_objArray(), "must be obj array"); assert(old->is_forwarded(), "must be forwarded"); assert(Universe::heap()->is_in_reserved(old), "must be in heap."); @@ -3884,7 +4022,7 @@ end = start + ParGCArrayScanChunk; arrayOop(old)->set_length(end); // Push remainder. - _par_scan_state->push_on_queue((oop*) ((intptr_t) old | G1_PARTIAL_ARRAY_MASK)); + _par_scan_state->push_on_queue(set_partial_array_mask(old)); } else { // Restore length so that the heap remains parsable in // case of evacuation failure. @@ -3893,11 +4031,6 @@ // process our set of indices (include header in first chunk) process_array_chunk<oop>(obj, start, end); - oop* start_addr = start == 0 ? (oop*)obj : obj->obj_at_addr<oop>(start); - oop* end_addr = (oop*)(obj->base()) + end; // obj_at_addr(end) asserts end < length - MemRegion mr((HeapWord*)start_addr, (HeapWord*)end_addr); - _scanner.set_region(_g1->heap_region_containing(obj)); - obj->oop_iterate(&_scanner, mr); } int G1ScanAndBalanceClosure::_nq = 0; @@ -3931,6 +4064,13 @@ pss->hash_seed(), ref_to_scan)) { 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 + assert(ref_to_scan != NULL, "invariant"); + assert(has_partial_array_mask(ref_to_scan) || + _g1h->obj_in_cs(*ref_to_scan), "invariant"); pss->push_on_queue(ref_to_scan); continue; } @@ -3976,10 +4116,10 @@ ResourceMark rm; HandleMark hm; - G1ParScanThreadState pss(_g1h, i); - G1ParScanHeapEvacClosure scan_evac_cl(_g1h, &pss); - G1ParScanHeapEvacClosure evac_failure_cl(_g1h, &pss); - G1ParScanPartialArrayClosure partial_scan_cl(_g1h, &pss); + G1ParScanThreadState pss(_g1h, i); + G1ParScanHeapEvacClosure scan_evac_cl(_g1h, &pss); + G1ParScanHeapEvacFailureClosure evac_failure_cl(_g1h, &pss); + G1ParScanPartialArrayClosure partial_scan_cl(_g1h, &pss); pss.set_evac_closure(&scan_evac_cl); pss.set_evac_failure_closure(&evac_failure_cl); @@ -4024,6 +4164,9 @@ _g1h->g1_policy()->record_obj_copy_time(i, elapsed_ms-term_ms); _g1h->g1_policy()->record_termination_time(i, term_ms); } + if (G1UseSurvivorSpace) { + _g1h->g1_policy()->record_thread_age_table(pss.age_table()); + } _g1h->update_surviving_young_words(pss.surviving_young_words()+1); // Clean up any par-expanded rem sets. @@ -4263,7 +4406,7 @@ // Is this the right thing to do here? We don't save marks // on individual heap regions when we allocate from // them in parallel, so this seems like the correct place for this. - all_alloc_regions_note_end_of_copying(); + retire_all_alloc_regions(); { G1IsAliveClosure is_alive(this); G1KeepAliveClosure keep_alive(this); @@ -4903,7 +5046,7 @@ return no_allocs; } -void G1CollectedHeap::all_alloc_regions_note_end_of_copying() { +void G1CollectedHeap::retire_all_alloc_regions() { for (int ap = 0; ap < GCAllocPurposeCount; ++ap) { HeapRegion* r = _gc_alloc_regions[ap]; if (r != NULL) { @@ -4916,8 +5059,7 @@ } } if (!has_processed_alias) { - r->note_end_of_copying(); - g1_policy()->record_after_bytes(r->used()); + retire_alloc_region(r, false /* par */); } } }
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -90,7 +90,7 @@ HeapRegion* _curr_scan_only; HeapRegion* _survivor_head; - HeapRegion* _survivors_tail; + HeapRegion* _survivor_tail; size_t _survivor_length; void empty_list(HeapRegion* list); @@ -105,6 +105,7 @@ bool is_empty() { return _length == 0; } size_t length() { return _length; } size_t scan_only_length() { return _scan_only_length; } + size_t survivor_length() { return _survivor_length; } void rs_length_sampling_init(); bool rs_length_sampling_more(); @@ -120,6 +121,7 @@ HeapRegion* first_region() { return _head; } HeapRegion* first_scan_only_region() { return _scan_only_head; } HeapRegion* first_survivor_region() { return _survivor_head; } + HeapRegion* last_survivor_region() { return _survivor_tail; } HeapRegion* par_get_next_scan_only_region() { MutexLockerEx x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag); HeapRegion* ret = _curr_scan_only; @@ -219,7 +221,7 @@ // The to-space memory regions into which objects are being copied during // a GC. HeapRegion* _gc_alloc_regions[GCAllocPurposeCount]; - uint _gc_alloc_region_counts[GCAllocPurposeCount]; + size_t _gc_alloc_region_counts[GCAllocPurposeCount]; // A list of the regions that have been set to be alloc regions in the // current collection. @@ -247,6 +249,27 @@ NumberSeq _pop_obj_rc_at_copy; void print_popularity_summary_info() const; + // This is used for a quick test on whether a reference points into + // the collection set or not. Basically, we have an array, with one + // byte per region, and that byte denotes whether the corresponding + // region is in the collection set or not. The entry corresponding + // the bottom of the heap, i.e., region 0, is pointed to by + // _in_cset_fast_test_base. The _in_cset_fast_test field has been + // biased so that it actually points to address 0 of the address + // space, to make the test as fast as possible (we can simply shift + // the address to address into it, instead of having to subtract the + // bottom of the heap from the address before shifting it; basically + // it works in the same way the card table works). + bool* _in_cset_fast_test; + + // The allocated array used for the fast test on whether a reference + // points into the collection set or not. This field is also used to + // free the array. + bool* _in_cset_fast_test_base; + + // The length of the _in_cset_fast_test_base array. + size_t _in_cset_fast_test_length; + volatile unsigned _gc_time_stamp; size_t* _surviving_young_words; @@ -260,8 +283,8 @@ // Returns "true" iff none of the gc alloc regions have any allocations // since the last call to "save_marks". bool all_alloc_regions_no_allocs_since_save_marks(); - // Calls "note_end_of_copying on all gc alloc_regions. - void all_alloc_regions_note_end_of_copying(); + // Perform finalization stuff on all allocation regions. + void retire_all_alloc_regions(); // The number of regions allocated to hold humongous objects. int _num_humongous_regions; @@ -330,6 +353,10 @@ // that parallel threads might be attempting allocations. void par_allocate_remaining_space(HeapRegion* r); + // Retires an allocation region when it is full or at the end of a + // GC pause. + void retire_alloc_region(HeapRegion* alloc_region, bool par); + // Helper function for two callbacks below. // "full", if true, indicates that the GC is for a System.gc() request, // and should collect the entire heap. If "clear_all_soft_refs" is true, @@ -368,6 +395,38 @@ virtual void gc_prologue(bool full); virtual void gc_epilogue(bool full); + // We register a region with the fast "in collection set" test. We + // simply set to true the array slot corresponding to this region. + void register_region_with_in_cset_fast_test(HeapRegion* r) { + assert(_in_cset_fast_test_base != NULL, "sanity"); + assert(r->in_collection_set(), "invariant"); + int index = r->hrs_index(); + assert(0 <= (size_t) index && (size_t) index < _in_cset_fast_test_length, + "invariant"); + assert(!_in_cset_fast_test_base[index], "invariant"); + _in_cset_fast_test_base[index] = true; + } + + // This is a fast test on whether a reference points into the + // collection set or not. It does not assume that the reference + // points into the heap; if it doesn't, it will return false. + bool in_cset_fast_test(oop obj) { + assert(_in_cset_fast_test != NULL, "sanity"); + if (_g1_committed.contains((HeapWord*) obj)) { + // no need to subtract the bottom of the heap from obj, + // _in_cset_fast_test is biased + size_t index = ((size_t) obj) >> HeapRegion::LogOfHRGrainBytes; + bool ret = _in_cset_fast_test[index]; + // let's make sure the result is consistent with what the slower + // test returns + assert( ret || !obj_in_cs(obj), "sanity"); + assert(!ret || obj_in_cs(obj), "sanity"); + return ret; + } else { + return false; + } + } + protected: // Shrink the garbage-first heap by at most the given size (in bytes!). @@ -850,6 +909,7 @@ // Iterate over all objects, calling "cl.do_object" on each. virtual void object_iterate(ObjectClosure* cl); + virtual void safe_object_iterate(ObjectClosure* cl) { object_iterate(cl); } // Iterate over all objects allocated since the last collection, calling // "cl.do_object" on each. The heap must have been initialized properly
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -36,8 +36,11 @@ inline HeapRegion* G1CollectedHeap::heap_region_containing_raw(const void* addr) const { - HeapRegion* res = _hrs->addr_to_region(addr); - assert(res != NULL, "addr outside of heap?"); + assert(_g1_reserved.contains(addr), "invariant"); + size_t index = ((intptr_t) addr - (intptr_t) _g1_reserved.start()) + >> HeapRegion::LogOfHRGrainBytes; + HeapRegion* res = _hrs->at(index); + assert(res == _hrs->addr_to_region(addr), "sanity"); return res; }
--- a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -196,8 +196,13 @@ _short_lived_surv_rate_group(new SurvRateGroup(this, "Short Lived", G1YoungSurvRateNumRegionsSummary)), _survivor_surv_rate_group(new SurvRateGroup(this, "Survivor", - G1YoungSurvRateNumRegionsSummary)) + G1YoungSurvRateNumRegionsSummary)), // add here any more surv rate groups + _recorded_survivor_regions(0), + _recorded_survivor_head(NULL), + _recorded_survivor_tail(NULL), + _survivors_age_table(true) + { _recent_prev_end_times_for_all_gcs_sec->add(os::elapsedTime()); _prev_collection_pause_end_ms = os::elapsedTime() * 1000.0; @@ -272,6 +277,15 @@ _concurrent_mark_cleanup_times_ms->add(0.20); _tenuring_threshold = MaxTenuringThreshold; + if (G1UseSurvivorSpace) { + // if G1FixedSurvivorSpaceSize is 0 which means the size is not + // fixed, then _max_survivor_regions will be calculated at + // calculate_young_list_target_config during initialization + _max_survivor_regions = G1FixedSurvivorSpaceSize / HeapRegion::GrainBytes; + } else { + _max_survivor_regions = 0; + } + initialize_all(); } @@ -283,6 +297,9 @@ void G1CollectorPolicy::initialize_flags() { set_min_alignment(HeapRegion::GrainBytes); set_max_alignment(GenRemSet::max_alignment_constraint(rem_set_name())); + if (SurvivorRatio < 1) { + vm_exit_during_initialization("Invalid survivor ratio specified"); + } CollectorPolicy::initialize_flags(); } @@ -301,6 +318,8 @@ "-XX:+UseConcMarkSweepGC."); } + initialize_gc_policy_counters(); + if (G1Gen) { _in_young_gc_mode = true; @@ -322,6 +341,12 @@ } } +// Create the jstat counters for the policy. +void G1CollectorPolicy::initialize_gc_policy_counters() +{ + _gc_policy_counters = new GCPolicyCounters("GarbageFirst", 1, 2 + G1Gen); +} + void G1CollectorPolicy::calculate_young_list_min_length() { _young_list_min_length = 0; @@ -352,6 +377,7 @@ guarantee( so_length < _young_list_target_length, "invariant" ); _young_list_so_prefix_length = so_length; } + calculate_survivors_policy(); } // This method calculate the optimal scan-only set for a fixed young @@ -448,6 +474,9 @@ if (full_young_gcs() && _free_regions_at_end_of_collection > 0) { // we are in fully-young mode and there are free regions in the heap + double survivor_regions_evac_time = + predict_survivor_regions_evac_time(); + size_t min_so_length = 0; size_t max_so_length = 0; @@ -497,9 +526,8 @@ scanned_cards = predict_non_young_card_num(adj_rs_lengths); // calculate this once, so that we don't have to recalculate it in // the innermost loop - double base_time_ms = predict_base_elapsed_time_ms(pending_cards, - scanned_cards); - + double base_time_ms = predict_base_elapsed_time_ms(pending_cards, scanned_cards) + + survivor_regions_evac_time; // the result size_t final_young_length = 0; size_t final_so_length = 0; @@ -548,14 +576,14 @@ bool done = false; // this is the outermost loop while (!done) { -#if 0 +#ifdef TRACE_CALC_YOUNG_CONFIG // leave this in for debugging, just in case gclog_or_tty->print_cr("searching between " SIZE_FORMAT " and " SIZE_FORMAT ", incr " SIZE_FORMAT ", pass %s", from_so_length, to_so_length, so_length_incr, (pass == pass_type_coarse) ? "coarse" : (pass == pass_type_fine) ? "fine" : "final"); -#endif // 0 +#endif // TRACE_CALC_YOUNG_CONFIG size_t so_length = from_so_length; size_t init_free_regions = @@ -651,11 +679,11 @@ guarantee( so_length_incr == so_coarse_increments, "invariant" ); guarantee( final_so_length >= min_so_length, "invariant" ); -#if 0 +#ifdef TRACE_CALC_YOUNG_CONFIG // leave this in for debugging, just in case gclog_or_tty->print_cr(" coarse pass: SO length " SIZE_FORMAT, final_so_length); -#endif // 0 +#endif // TRACE_CALC_YOUNG_CONFIG from_so_length = (final_so_length - min_so_length > so_coarse_increments) ? @@ -687,12 +715,12 @@ // of the optimal size_t new_so_length = 950 * final_so_length / 1000; -#if 0 +#ifdef TRACE_CALC_YOUNG_CONFIG // leave this in for debugging, just in case gclog_or_tty->print_cr(" fine pass: SO length " SIZE_FORMAT ", setting it to " SIZE_FORMAT, final_so_length, new_so_length); -#endif // 0 +#endif // TRACE_CALC_YOUNG_CONFIG from_so_length = new_so_length; to_so_length = new_so_length; @@ -719,7 +747,8 @@ } // we should have at least one region in the target young length - _young_list_target_length = MAX2((size_t) 1, final_young_length); + _young_list_target_length = + MAX2((size_t) 1, final_young_length + _recorded_survivor_regions); if (final_so_length >= final_young_length) // and we need to ensure that the S-O length is not greater than // the target young length (this is being a bit careful) @@ -734,7 +763,7 @@ double end_time_sec = os::elapsedTime(); double elapsed_time_ms = (end_time_sec - start_time_sec) * 1000.0; -#if 0 +#ifdef TRACE_CALC_YOUNG_CONFIG // leave this in for debugging, just in case gclog_or_tty->print_cr("target = %1.1lf ms, young = " SIZE_FORMAT ", SO = " SIZE_FORMAT ", " @@ -747,9 +776,9 @@ calculations, full_young_gcs() ? "full" : "partial", should_initiate_conc_mark() ? " i-m" : "", - in_marking_window(), - in_marking_window_im()); -#endif // 0 + _in_marking_window, + _in_marking_window_im); +#endif // TRACE_CALC_YOUNG_CONFIG if (_young_list_target_length < _young_list_min_length) { // bummer; this means that, if we do a pause when the optimal @@ -768,14 +797,14 @@ // S-O length so_length = calculate_optimal_so_length(_young_list_min_length); -#if 0 +#ifdef TRACE_CALC_YOUNG_CONFIG // leave this in for debugging, just in case gclog_or_tty->print_cr("adjusted target length from " SIZE_FORMAT " to " SIZE_FORMAT ", SO " SIZE_FORMAT, _young_list_target_length, _young_list_min_length, so_length); -#endif // 0 +#endif // TRACE_CALC_YOUNG_CONFIG _young_list_target_length = MAX2(_young_list_min_length, (size_t)1); @@ -785,12 +814,12 @@ // we are in a partially-young mode or we've run out of regions (due // to evacuation failure) -#if 0 +#ifdef TRACE_CALC_YOUNG_CONFIG // leave this in for debugging, just in case gclog_or_tty->print_cr("(partial) setting target to " SIZE_FORMAT ", SO " SIZE_FORMAT, _young_list_min_length, 0); -#endif // 0 +#endif // TRACE_CALC_YOUNG_CONFIG // we'll do the pause as soon as possible and with no S-O prefix // (see above for the reasons behind the latter) @@ -884,6 +913,16 @@ return true; } +double G1CollectorPolicy::predict_survivor_regions_evac_time() { + double survivor_regions_evac_time = 0.0; + for (HeapRegion * r = _recorded_survivor_head; + r != NULL && r != _recorded_survivor_tail->get_next_young_region(); + r = r->get_next_young_region()) { + survivor_regions_evac_time += predict_region_elapsed_time_ms(r, true); + } + return survivor_regions_evac_time; +} + void G1CollectorPolicy::check_prediction_validity() { guarantee( adaptive_young_list_length(), "should not call this otherwise" ); @@ -995,11 +1034,15 @@ _short_lived_surv_rate_group->start_adding_regions(); // also call this on any additional surv rate groups + record_survivor_regions(0, NULL, NULL); + _prev_region_num_young = _region_num_young; _prev_region_num_tenured = _region_num_tenured; _free_regions_at_end_of_collection = _g1->free_regions(); _scan_only_regions_at_end_of_collection = 0; + // Reset survivors SurvRateGroup. + _survivor_surv_rate_group->reset(); calculate_young_list_min_length(); calculate_young_list_target_config(); } @@ -1104,6 +1147,10 @@ _short_lived_surv_rate_group->record_scan_only_prefix(short_lived_so_length); tag_scan_only(short_lived_so_length); + if (G1UseSurvivorSpace) { + _survivors_age_table.clear(); + } + assert( verify_young_ages(), "region age verification" ); } @@ -1965,9 +2012,6 @@ // </NEW PREDICTION> _target_pause_time_ms = -1.0; - - // TODO: calculate tenuring threshold - _tenuring_threshold = MaxTenuringThreshold; } // <NEW PREDICTION> @@ -2058,7 +2102,7 @@ guarantee( hr->is_young() && hr->age_in_surv_rate_group() != -1, "invariant" ); int age = hr->age_in_surv_rate_group(); - double yg_surv_rate = predict_yg_surv_rate(age); + double yg_surv_rate = predict_yg_surv_rate(age, hr->surv_rate_group()); bytes_to_copy = (size_t) ((double) hr->used() * yg_surv_rate); } @@ -2091,7 +2135,7 @@ } #if PREDICTIONS_VERBOSE if (young) { - _recorded_young_bytes += hr->asSpace()->used(); + _recorded_young_bytes += hr->used(); } else { _recorded_marked_bytes += hr->max_live_bytes(); } @@ -2119,11 +2163,6 @@ predict_non_young_card_num(_predicted_rs_lengths); _recorded_region_num = _recorded_young_regions + _recorded_non_young_regions; - _predicted_young_survival_ratio = 0.0; - for (int i = 0; i < _recorded_young_regions; ++i) - _predicted_young_survival_ratio += predict_yg_surv_rate(i); - _predicted_young_survival_ratio /= (double) _recorded_young_regions; - _predicted_scan_only_scan_time_ms = predict_scan_only_time_ms(_recorded_scan_only_regions); _predicted_rs_update_time_ms = @@ -2673,8 +2712,11 @@ assert(in_young_gc_mode(), "should be in young GC mode"); bool ret; size_t young_list_length = _g1->young_list_length(); - - if (young_list_length < _young_list_target_length) { + size_t young_list_max_length = _young_list_target_length; + if (G1FixedEdenSize) { + young_list_max_length -= _max_survivor_regions; + } + if (young_list_length < young_list_max_length) { ret = true; ++_region_num_young; } else { @@ -2710,17 +2752,39 @@ } -uint G1CollectorPolicy::max_regions(int purpose) { +size_t G1CollectorPolicy::max_regions(int purpose) { switch (purpose) { case GCAllocForSurvived: - return G1MaxSurvivorRegions; + return _max_survivor_regions; case GCAllocForTenured: - return UINT_MAX; + return REGIONS_UNLIMITED; default: - return UINT_MAX; + ShouldNotReachHere(); + return REGIONS_UNLIMITED; }; } +// Calculates survivor space parameters. +void G1CollectorPolicy::calculate_survivors_policy() +{ + if (!G1UseSurvivorSpace) { + return; + } + if (G1FixedSurvivorSpaceSize == 0) { + _max_survivor_regions = _young_list_target_length / SurvivorRatio; + } else { + _max_survivor_regions = G1FixedSurvivorSpaceSize / HeapRegion::GrainBytes; + } + + if (G1FixedTenuringThreshold) { + _tenuring_threshold = MaxTenuringThreshold; + } else { + _tenuring_threshold = _survivors_age_table.compute_tenuring_threshold( + HeapRegion::GrainWords * _max_survivor_regions); + } +} + + void G1CollectorPolicy_BestRegionsFirst:: set_single_region_collection_set(HeapRegion* hr) { @@ -2743,7 +2807,11 @@ double max_pause_time_ms = _mmu_tracker->max_gc_time() * 1000.0; size_t young_list_length = _g1->young_list_length(); - bool reached_target_length = young_list_length >= _young_list_target_length; + size_t young_list_max_length = _young_list_target_length; + if (G1FixedEdenSize) { + young_list_max_length -= _max_survivor_regions; + } + bool reached_target_length = young_list_length >= young_list_max_length; if (in_young_gc_mode()) { if (reached_target_length) { @@ -2985,6 +3053,7 @@ _collection_set = hr; _collection_set_size++; _collection_set_bytes_used_before += hr->used(); + _g1->register_region_with_in_cset_fast_test(hr); } void
--- a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -49,7 +49,7 @@ class MainBodySummary; class PopPreambleSummary; -class PauseSummary { +class PauseSummary: public CHeapObj { define_num_seq(total) define_num_seq(other) @@ -58,7 +58,7 @@ virtual PopPreambleSummary* pop_preamble_summary() { return NULL; } }; -class MainBodySummary { +class MainBodySummary: public CHeapObj { define_num_seq(satb_drain) // optional define_num_seq(parallel) // parallel only define_num_seq(ext_root_scan) @@ -75,7 +75,7 @@ define_num_seq(clear_ct) // parallel only }; -class PopPreambleSummary { +class PopPreambleSummary: public CHeapObj { define_num_seq(pop_preamble) define_num_seq(pop_update_rs) define_num_seq(pop_scan_rs) @@ -557,6 +557,8 @@ return get_new_neg_prediction(_young_gc_eff_seq); } + double predict_survivor_regions_evac_time(); + // </NEW PREDICTION> public: @@ -599,8 +601,8 @@ // Returns an estimate of the survival rate of the region at yg-age // "yg_age". - double predict_yg_surv_rate(int age) { - TruncatedSeq* seq = _short_lived_surv_rate_group->get_seq(age); + double predict_yg_surv_rate(int age, SurvRateGroup* surv_rate_group) { + TruncatedSeq* seq = surv_rate_group->get_seq(age); if (seq->num() == 0) gclog_or_tty->print("BARF! age is %d", age); guarantee( seq->num() > 0, "invariant" ); @@ -610,6 +612,10 @@ return pred; } + double predict_yg_surv_rate(int age) { + return predict_yg_surv_rate(age, _short_lived_surv_rate_group); + } + double accum_yg_surv_rate_pred(int age) { return _short_lived_surv_rate_group->accum_surv_rate_pred(age); } @@ -822,6 +828,9 @@ virtual void init(); + // Create jstat counters for the policy. + virtual void initialize_gc_policy_counters(); + virtual HeapWord* mem_allocate_work(size_t size, bool is_tlab, bool* gc_overhead_limit_was_exceeded); @@ -1047,8 +1056,12 @@ // Print stats on young survival ratio void print_yg_surv_rate_info() const; - void finished_recalculating_age_indexes() { - _short_lived_surv_rate_group->finished_recalculating_age_indexes(); + void finished_recalculating_age_indexes(bool is_survivors) { + if (is_survivors) { + _survivor_surv_rate_group->finished_recalculating_age_indexes(); + } else { + _short_lived_surv_rate_group->finished_recalculating_age_indexes(); + } // do that for any other surv rate groups } @@ -1097,6 +1110,17 @@ // maximum amount of suvivors regions. int _tenuring_threshold; + // The limit on the number of regions allocated for survivors. + size_t _max_survivor_regions; + + // The amount of survor regions after a collection. + size_t _recorded_survivor_regions; + // List of survivor regions. + HeapRegion* _recorded_survivor_head; + HeapRegion* _recorded_survivor_tail; + + ageTable _survivors_age_table; + public: inline GCAllocPurpose @@ -1116,7 +1140,9 @@ return GCAllocForTenured; } - uint max_regions(int purpose); + static const size_t REGIONS_UNLIMITED = ~(size_t)0; + + size_t max_regions(int purpose); // The limit on regions for a particular purpose is reached. void note_alloc_region_limit_reached(int purpose) { @@ -1132,6 +1158,23 @@ void note_stop_adding_survivor_regions() { _survivor_surv_rate_group->stop_adding_regions(); } + + void record_survivor_regions(size_t regions, + HeapRegion* head, + HeapRegion* tail) { + _recorded_survivor_regions = regions; + _recorded_survivor_head = head; + _recorded_survivor_tail = tail; + } + + void record_thread_age_table(ageTable* age_table) + { + _survivors_age_table.merge_par(age_table); + } + + // Calculates survivor space parameters. + void calculate_survivors_policy(); + }; // This encapsulates a particular strategy for a g1 Collector.
--- a/src/share/vm/gc_implementation/g1/g1MMUTracker.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1MMUTracker.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -28,7 +28,7 @@ /***** ALL TIMES ARE IN SECS!!!!!!! *****/ // this is the "interface" -class G1MMUTracker { +class G1MMUTracker: public CHeapObj { protected: double _time_slice; double _max_gc_time; // this is per time slice @@ -67,7 +67,7 @@ } }; -class G1MMUTrackerQueueElem { +class G1MMUTrackerQueueElem VALUE_OBJ_CLASS_SPEC { private: double _start_time; double _end_time;
--- a/src/share/vm/gc_implementation/g1/g1OopClosures.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1OopClosures.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -77,6 +77,18 @@ #define G1_PARTIAL_ARRAY_MASK 1 +inline bool has_partial_array_mask(oop* ref) { + return (intptr_t) ref & G1_PARTIAL_ARRAY_MASK; +} + +inline oop* set_partial_array_mask(oop obj) { + return (oop*) ((intptr_t) obj | G1_PARTIAL_ARRAY_MASK); +} + +inline oop clear_partial_array_mask(oop* ref) { + return oop((intptr_t) ref & ~G1_PARTIAL_ARRAY_MASK); +} + class G1ParScanPartialArrayClosure : public G1ParClosureSuper { G1ParScanClosure _scanner; template <class T> void process_array_chunk(oop obj, int start, int end); @@ -101,7 +113,8 @@ G1ParClosureSuper(g1, par_scan_state), _scanner(scanner) { } }; -template<bool do_gen_barrier, G1Barrier barrier, bool do_mark_forwardee> +template<bool do_gen_barrier, G1Barrier barrier, + bool do_mark_forwardee, bool skip_cset_test> class G1ParCopyClosure : public G1ParCopyHelper { G1ParScanClosure _scanner; void do_oop_work(oop* p); @@ -119,14 +132,22 @@ virtual void do_oop(narrowOop* p) { do_oop_nv(p); } }; -typedef G1ParCopyClosure<false, G1BarrierNone, false> G1ParScanExtRootClosure; -typedef G1ParCopyClosure<true, G1BarrierNone, false> G1ParScanPermClosure; -typedef G1ParCopyClosure<false, G1BarrierNone, true> G1ParScanAndMarkExtRootClosure; -typedef G1ParCopyClosure<true, G1BarrierNone, true> G1ParScanAndMarkPermClosure; -typedef G1ParCopyClosure<false, G1BarrierRS, false> G1ParScanHeapRSClosure; -typedef G1ParCopyClosure<false, G1BarrierRS, true> G1ParScanAndMarkHeapRSClosure; -typedef G1ParCopyClosure<false, G1BarrierEvac, false> G1ParScanHeapEvacClosure; - +typedef G1ParCopyClosure<false, G1BarrierNone, false, false> G1ParScanExtRootClosure; +typedef G1ParCopyClosure<true, G1BarrierNone, false, false> G1ParScanPermClosure; +typedef G1ParCopyClosure<false, G1BarrierNone, true, false> G1ParScanAndMarkExtRootClosure; +typedef G1ParCopyClosure<true, G1BarrierNone, true, false> G1ParScanAndMarkPermClosure; +typedef G1ParCopyClosure<false, G1BarrierRS, false, false> G1ParScanHeapRSClosure; +typedef G1ParCopyClosure<false, G1BarrierRS, true, false> G1ParScanAndMarkHeapRSClosure; +// This is the only case when we set skip_cset_test. Basically, this +// closure is (should?) only be called directly while we're draining +// the overflow and task queues. In that case we know that the +// reference in question points into the collection set, otherwise we +// would not have pushed it on the queue. +typedef G1ParCopyClosure<false, G1BarrierEvac, false, true> G1ParScanHeapEvacClosure; +// We need a separate closure to handle references during evacuation +// failure processing, as it cannot asume that the reference already + // points to the collection set (like G1ParScanHeapEvacClosure does). +typedef G1ParCopyClosure<false, G1BarrierEvac, false, false> G1ParScanHeapEvacFailureClosure; class FilterIntoCSClosure: public OopClosure { G1CollectedHeap* _g1;
--- a/src/share/vm/gc_implementation/g1/g1RemSet.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1RemSet.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -572,6 +572,9 @@ } guarantee( _cards_scanned == NULL, "invariant" ); _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers()); + for (uint i = 0; i < n_workers(); ++i) { + _cards_scanned[i] = 0; + } _total_cards_scanned = 0; }
--- a/src/share/vm/gc_implementation/g1/g1RemSet.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1RemSet.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -30,7 +30,7 @@ class HRInto_G1RemSet; class ConcurrentG1Refine; -class G1RemSet { +class G1RemSet: public CHeapObj { protected: G1CollectedHeap* _g1;
--- a/src/share/vm/gc_implementation/g1/g1_globals.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1_globals.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -28,7 +28,7 @@ #define G1_FLAGS(develop, develop_pd, product, product_pd, diagnostic, experimental, notproduct, manageable, product_rw) \ \ - product(intx, ParallelGCG1AllocBufferSize, 4*K, \ + product(intx, ParallelGCG1AllocBufferSize, 8*K, \ "Size of parallel G1 allocation buffers in to-space.") \ \ product(intx, G1TimeSliceMS, 500, \ @@ -281,7 +281,17 @@ develop(bool, G1HRRSFlushLogBuffersOnVerify, false, \ "Forces flushing of log buffers before verification.") \ \ - product(intx, G1MaxSurvivorRegions, 0, \ - "The maximum number of survivor regions") + product(bool, G1UseSurvivorSpace, true, \ + "When true, use survivor space.") \ + \ + product(bool, G1FixedTenuringThreshold, false, \ + "When set, G1 will not adjust the tenuring threshold") \ + \ + product(bool, G1FixedEdenSize, false, \ + "When set, G1 will not allocate unused survivor space regions") \ + \ + product(uintx, G1FixedSurvivorSpaceSize, 0, \ + "If non-0 is the size of the G1 survivor space, " \ + "otherwise SurvivorRatio is used to determine the size") G1_FLAGS(DECLARE_DEVELOPER_FLAG, DECLARE_PD_DEVELOPER_FLAG, DECLARE_PRODUCT_FLAG, DECLARE_PD_PRODUCT_FLAG, DECLARE_DIAGNOSTIC_FLAG, DECLARE_EXPERIMENTAL_FLAG, DECLARE_NOTPRODUCT_FLAG, DECLARE_MANAGEABLE_FLAG, DECLARE_PRODUCT_RW_FLAG)
--- a/src/share/vm/gc_implementation/g1/g1_specialized_oop_closures.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1_specialized_oop_closures.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -32,11 +32,13 @@ G1BarrierNone, G1BarrierRS, G1BarrierEvac }; -template<bool do_gen_barrier, G1Barrier barrier, bool do_mark_forwardee> +template<bool do_gen_barrier, G1Barrier barrier, + bool do_mark_forwardee, bool skip_cset_test> class G1ParCopyClosure; class G1ParScanClosure; -typedef G1ParCopyClosure<false, G1BarrierEvac, false> G1ParScanHeapEvacClosure; +typedef G1ParCopyClosure<false, G1BarrierEvac, false, true> + G1ParScanHeapEvacClosure; class FilterIntoCSClosure; class FilterOutOfRegionClosure;
--- a/src/share/vm/gc_implementation/g1/heapRegion.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/heapRegion.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -566,7 +566,11 @@ void note_end_of_copying() { assert(top() >= _next_top_at_mark_start, "Increase only"); - _next_top_at_mark_start = top(); + // Survivor regions will be scanned on the start of concurrent + // marking. + if (!is_survivor()) { + _next_top_at_mark_start = top(); + } } // Returns "false" iff no object in the region was allocated when the @@ -829,7 +833,7 @@ // A linked lists of heap regions. It leaves the "next" field // unspecified; that's up to subtypes. -class RegionList { +class RegionList VALUE_OBJ_CLASS_SPEC { protected: virtual HeapRegion* get_next(HeapRegion* chr) = 0; virtual void set_next(HeapRegion* chr,
--- a/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -65,9 +65,11 @@ // We need access in order to union things into the base table. BitMap* bm() { return &_bm; } +#if PRT_COUNT_OCCUPIED void recount_occupied() { _occupied = (jint) bm()->count_one_bits(); } +#endif PerRegionTable(HeapRegion* hr) : _hr(hr), @@ -1144,7 +1146,9 @@ size_t i = _outgoing_region_map.get_next_one_offset(0); while (i < _outgoing_region_map.size()) { HeapRegion* to_region = g1h->region_at(i); - to_region->rem_set()->clear_incoming_entry(hr()); + if (!to_region->in_collection_set()) { + to_region->rem_set()->clear_incoming_entry(hr()); + } i = _outgoing_region_map.get_next_one_offset(i+1); } }
--- a/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -58,7 +58,7 @@ // is represented. If a deleted PRT is re-used, a thread adding a bit, // thinking the PRT is for a different region, does no harm. -class OtherRegionsTable: public CHeapObj { +class OtherRegionsTable VALUE_OBJ_CLASS_SPEC { friend class HeapRegionRemSetIterator; G1CollectedHeap* _g1h;
--- a/src/share/vm/gc_implementation/g1/ptrQueue.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/ptrQueue.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -29,7 +29,7 @@ class PtrQueueSet; -class PtrQueue: public CHeapObj { +class PtrQueue VALUE_OBJ_CLASS_SPEC { protected: // The ptr queue set to which this queue belongs. @@ -130,7 +130,7 @@ // In particular, the individual queues allocate buffers from this shared // set, and return completed buffers to the set. // All these variables are are protected by the TLOQ_CBL_mon. XXX ??? -class PtrQueueSet: public CHeapObj { +class PtrQueueSet VALUE_OBJ_CLASS_SPEC { protected:
--- a/src/share/vm/gc_implementation/g1/sparsePRT.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/sparsePRT.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -33,7 +33,7 @@ // old versions synchronously. -class SparsePRTEntry { +class SparsePRTEntry: public CHeapObj { public: enum SomePublicConstants { CardsPerEntry = (short)4, @@ -167,7 +167,7 @@ }; // ValueObj because will be embedded in HRRS iterator. -class RSHashTableIter: public CHeapObj { +class RSHashTableIter VALUE_OBJ_CLASS_SPEC { short _tbl_ind; short _bl_ind; short _card_ind; @@ -213,7 +213,7 @@ class SparsePRTIter; -class SparsePRT : public CHeapObj { +class SparsePRT VALUE_OBJ_CLASS_SPEC { // Iterations are done on the _cur hash table, since they only need to // see entries visible at the start of a collection pause. // All other operations are done using the _next hash table.
--- a/src/share/vm/gc_implementation/g1/survRateGroup.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/survRateGroup.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -29,23 +29,14 @@ const char* name, size_t summary_surv_rates_len) : _g1p(g1p), _name(name), - _all_regions_allocated(0), - _curr_length(0), _scan_only_prefix(0), _setup_seq_num(0), - _array_length(0), _surv_rate(NULL), _accum_surv_rate_pred(NULL), - _accum_surv_rate(0.0), _surv_rate_pred(NULL), _last_pred(0.0), _summary_surv_rates_len(summary_surv_rates_len), _summary_surv_rates_max_len(0), - _summary_surv_rates(NULL) { - - // the following will set up the arrays with length 1 - _curr_length = 1; - stop_adding_regions(); - guarantee( _array_length == 1, "invariant" ); - guarantee( _surv_rate_pred[0] != NULL, "invariant" ); - _surv_rate_pred[0]->add(0.4); - all_surviving_words_recorded(false); - _curr_length = 0; - + _summary_surv_rates(NULL), + _surv_rate(NULL), + _accum_surv_rate_pred(NULL), + _surv_rate_pred(NULL) +{ + reset(); if (summary_surv_rates_len > 0) { size_t length = summary_surv_rates_len; _summary_surv_rates = NEW_C_HEAP_ARRAY(NumberSeq*, length); @@ -60,61 +51,80 @@ start_adding_regions(); } + +void SurvRateGroup::reset() +{ + _all_regions_allocated = 0; + _scan_only_prefix = 0; + _setup_seq_num = 0; + _stats_arrays_length = 0; + _accum_surv_rate = 0.0; + _last_pred = 0.0; + // the following will set up the arrays with length 1 + _region_num = 1; + stop_adding_regions(); + guarantee( _stats_arrays_length == 1, "invariant" ); + guarantee( _surv_rate_pred[0] != NULL, "invariant" ); + _surv_rate_pred[0]->add(0.4); + all_surviving_words_recorded(false); + _region_num = 0; +} + + void SurvRateGroup::start_adding_regions() { - _setup_seq_num = _array_length; - _curr_length = _scan_only_prefix; + _setup_seq_num = _stats_arrays_length; + _region_num = _scan_only_prefix; _accum_surv_rate = 0.0; #if 0 - gclog_or_tty->print_cr("start adding regions, seq num %d, length %d", - _setup_seq_num, _curr_length); + gclog_or_tty->print_cr("[%s] start adding regions, seq num %d, length %d", + _name, _setup_seq_num, _region_num); #endif // 0 } void SurvRateGroup::stop_adding_regions() { - size_t length = _curr_length; #if 0 - gclog_or_tty->print_cr("stop adding regions, length %d", length); + gclog_or_tty->print_cr("[%s] stop adding regions, length %d", _name, _region_num); #endif // 0 - if (length > _array_length) { + if (_region_num > _stats_arrays_length) { double* old_surv_rate = _surv_rate; double* old_accum_surv_rate_pred = _accum_surv_rate_pred; TruncatedSeq** old_surv_rate_pred = _surv_rate_pred; - _surv_rate = NEW_C_HEAP_ARRAY(double, length); + _surv_rate = NEW_C_HEAP_ARRAY(double, _region_num); if (_surv_rate == NULL) { - vm_exit_out_of_memory(sizeof(double) * length, + vm_exit_out_of_memory(sizeof(double) * _region_num, "Not enough space for surv rate array."); } - _accum_surv_rate_pred = NEW_C_HEAP_ARRAY(double, length); + _accum_surv_rate_pred = NEW_C_HEAP_ARRAY(double, _region_num); if (_accum_surv_rate_pred == NULL) { - vm_exit_out_of_memory(sizeof(double) * length, + vm_exit_out_of_memory(sizeof(double) * _region_num, "Not enough space for accum surv rate pred array."); } - _surv_rate_pred = NEW_C_HEAP_ARRAY(TruncatedSeq*, length); + _surv_rate_pred = NEW_C_HEAP_ARRAY(TruncatedSeq*, _region_num); if (_surv_rate == NULL) { - vm_exit_out_of_memory(sizeof(TruncatedSeq*) * length, + vm_exit_out_of_memory(sizeof(TruncatedSeq*) * _region_num, "Not enough space for surv rate pred array."); } - for (size_t i = 0; i < _array_length; ++i) + for (size_t i = 0; i < _stats_arrays_length; ++i) _surv_rate_pred[i] = old_surv_rate_pred[i]; #if 0 - gclog_or_tty->print_cr("stop adding regions, new seqs %d to %d", - _array_length, length - 1); + gclog_or_tty->print_cr("[%s] stop adding regions, new seqs %d to %d", + _name, _array_length, _region_num - 1); #endif // 0 - for (size_t i = _array_length; i < length; ++i) { + for (size_t i = _stats_arrays_length; i < _region_num; ++i) { _surv_rate_pred[i] = new TruncatedSeq(10); // _surv_rate_pred[i]->add(last_pred); } - _array_length = length; + _stats_arrays_length = _region_num; if (old_surv_rate != NULL) FREE_C_HEAP_ARRAY(double, old_surv_rate); @@ -124,7 +134,7 @@ FREE_C_HEAP_ARRAY(NumberSeq*, old_surv_rate_pred); } - for (size_t i = 0; i < _array_length; ++i) + for (size_t i = 0; i < _stats_arrays_length; ++i) _surv_rate[i] = 0.0; } @@ -135,7 +145,7 @@ double ret = _accum_surv_rate; if (adjustment > 0) { - TruncatedSeq* seq = get_seq(_curr_length+1); + TruncatedSeq* seq = get_seq(_region_num+1); double surv_rate = _g1p->get_new_prediction(seq); ret += surv_rate; } @@ -145,23 +155,23 @@ int SurvRateGroup::next_age_index() { - TruncatedSeq* seq = get_seq(_curr_length); + TruncatedSeq* seq = get_seq(_region_num); double surv_rate = _g1p->get_new_prediction(seq); _accum_surv_rate += surv_rate; - ++_curr_length; + ++_region_num; return (int) ++_all_regions_allocated; } void SurvRateGroup::record_scan_only_prefix(size_t scan_only_prefix) { - guarantee( scan_only_prefix <= _curr_length, "pre-condition" ); + guarantee( scan_only_prefix <= _region_num, "pre-condition" ); _scan_only_prefix = scan_only_prefix; } void SurvRateGroup::record_surviving_words(int age_in_group, size_t surv_words) { - guarantee( 0 <= age_in_group && (size_t) age_in_group < _curr_length, + guarantee( 0 <= age_in_group && (size_t) age_in_group < _region_num, "pre-condition" ); guarantee( _surv_rate[age_in_group] <= 0.00001, "should only update each slot once" ); @@ -178,15 +188,15 @@ void SurvRateGroup::all_surviving_words_recorded(bool propagate) { - if (propagate && _curr_length > 0) { // conservative - double surv_rate = _surv_rate_pred[_curr_length-1]->last(); + if (propagate && _region_num > 0) { // conservative + double surv_rate = _surv_rate_pred[_region_num-1]->last(); #if 0 gclog_or_tty->print_cr("propagating %1.2lf from %d to %d", surv_rate, _curr_length, _array_length - 1); #endif // 0 - for (size_t i = _curr_length; i < _array_length; ++i) { + for (size_t i = _region_num; i < _stats_arrays_length; ++i) { guarantee( _surv_rate[i] <= 0.00001, "the slot should not have been updated" ); _surv_rate_pred[i]->add(surv_rate); @@ -195,7 +205,7 @@ double accum = 0.0; double pred = 0.0; - for (size_t i = 0; i < _array_length; ++i) { + for (size_t i = 0; i < _stats_arrays_length; ++i) { pred = _g1p->get_new_prediction(_surv_rate_pred[i]); if (pred > 1.0) pred = 1.0; accum += pred; @@ -209,8 +219,8 @@ void SurvRateGroup::print() { gclog_or_tty->print_cr("Surv Rate Group: %s (%d entries, %d scan-only)", - _name, _curr_length, _scan_only_prefix); - for (size_t i = 0; i < _curr_length; ++i) { + _name, _region_num, _scan_only_prefix); + for (size_t i = 0; i < _region_num; ++i) { gclog_or_tty->print_cr(" age %4d surv rate %6.2lf %% pred %6.2lf %%%s", i, _surv_rate[i] * 100.0, _g1p->get_new_prediction(_surv_rate_pred[i]) * 100.0,
--- a/src/share/vm/gc_implementation/g1/survRateGroup.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/survRateGroup.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -29,7 +29,7 @@ G1CollectorPolicy* _g1p; const char* _name; - size_t _array_length; + size_t _stats_arrays_length; double* _surv_rate; double* _accum_surv_rate_pred; double _last_pred; @@ -40,7 +40,7 @@ size_t _summary_surv_rates_max_len; int _all_regions_allocated; - size_t _curr_length; + size_t _region_num; size_t _scan_only_prefix; size_t _setup_seq_num; @@ -48,6 +48,7 @@ SurvRateGroup(G1CollectorPolicy* g1p, const char* name, size_t summary_surv_rates_len); + void reset(); void start_adding_regions(); void stop_adding_regions(); void record_scan_only_prefix(size_t scan_only_prefix); @@ -55,22 +56,21 @@ void all_surviving_words_recorded(bool propagate); const char* name() { return _name; } - size_t region_num() { return _curr_length; } + size_t region_num() { return _region_num; } size_t scan_only_length() { return _scan_only_prefix; } double accum_surv_rate_pred(int age) { assert(age >= 0, "must be"); - if ((size_t)age < _array_length) + if ((size_t)age < _stats_arrays_length) return _accum_surv_rate_pred[age]; else { - double diff = (double) (age - _array_length + 1); - return _accum_surv_rate_pred[_array_length-1] + diff * _last_pred; + double diff = (double) (age - _stats_arrays_length + 1); + return _accum_surv_rate_pred[_stats_arrays_length-1] + diff * _last_pred; } } double accum_surv_rate(size_t adjustment); TruncatedSeq* get_seq(size_t age) { - guarantee( 0 <= age, "pre-condition" ); if (age >= _setup_seq_num) { guarantee( _setup_seq_num > 0, "invariant" ); age = _setup_seq_num-1;
--- a/src/share/vm/gc_implementation/includeDB_gc_concurrentMarkSweep Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/includeDB_gc_concurrentMarkSweep Fri Feb 27 15:13:00 2009 -0800 @@ -28,6 +28,7 @@ binaryTreeDictionary.cpp binaryTreeDictionary.hpp binaryTreeDictionary.cpp globals.hpp binaryTreeDictionary.cpp ostream.hpp +binaryTreeDictionary.cpp space.inline.hpp binaryTreeDictionary.cpp spaceDecorator.hpp binaryTreeDictionary.hpp freeBlockDictionary.hpp
--- a/src/share/vm/gc_implementation/includeDB_gc_g1 Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/includeDB_gc_g1 Fri Feb 27 15:13:00 2009 -0800 @@ -31,9 +31,10 @@ cardTableRS.cpp concurrentMark.hpp cardTableRS.cpp g1SATBCardTableModRefBS.hpp -collectionSetChooser.cpp g1CollectedHeap.hpp +collectionSetChooser.cpp g1CollectedHeap.inline.hpp collectionSetChooser.cpp g1CollectorPolicy.hpp collectionSetChooser.cpp collectionSetChooser.hpp +collectionSetChooser.cpp space.inline.hpp collectionSetChooser.hpp heapRegion.hpp collectionSetChooser.hpp growableArray.hpp @@ -42,14 +43,16 @@ concurrentG1Refine.cpp concurrentG1Refine.hpp concurrentG1Refine.cpp concurrentG1RefineThread.hpp concurrentG1Refine.cpp copy.hpp -concurrentG1Refine.cpp g1CollectedHeap.hpp +concurrentG1Refine.cpp g1CollectedHeap.inline.hpp concurrentG1Refine.cpp g1RemSet.hpp +concurrentG1Refine.cpp space.inline.hpp concurrentG1Refine.hpp globalDefinitions.hpp +concurrentG1Refine.hpp allocation.hpp concurrentG1RefineThread.cpp concurrentG1Refine.hpp concurrentG1RefineThread.cpp concurrentG1RefineThread.hpp -concurrentG1RefineThread.cpp g1CollectedHeap.hpp +concurrentG1RefineThread.cpp g1CollectedHeap.inline.hpp concurrentG1RefineThread.cpp g1CollectorPolicy.hpp concurrentG1RefineThread.cpp handles.inline.hpp concurrentG1RefineThread.cpp mutexLocker.hpp @@ -166,10 +169,11 @@ g1CollectorPolicy.cpp concurrentMarkThread.inline.hpp g1CollectorPolicy.cpp debug.hpp g1CollectorPolicy.cpp java.hpp -g1CollectorPolicy.cpp g1CollectedHeap.hpp +g1CollectorPolicy.cpp g1CollectedHeap.inline.hpp g1CollectorPolicy.cpp g1CollectorPolicy.hpp g1CollectorPolicy.cpp heapRegionRemSet.hpp g1CollectorPolicy.cpp mutexLocker.hpp +g1CollectorPolicy.cpp gcPolicyCounters.hpp g1CollectorPolicy.hpp collectorPolicy.hpp g1CollectorPolicy.hpp collectionSetChooser.hpp @@ -187,7 +191,7 @@ g1MarkSweep.cpp codeCache.hpp g1MarkSweep.cpp events.hpp g1MarkSweep.cpp fprofiler.hpp -g1MarkSweep.hpp g1CollectedHeap.hpp +g1MarkSweep.hpp g1CollectedHeap.inline.hpp g1MarkSweep.cpp g1MarkSweep.hpp g1MarkSweep.cpp gcLocker.hpp g1MarkSweep.cpp genCollectedHeap.hpp @@ -226,7 +230,7 @@ g1MMUTracker.cpp mutexLocker.hpp g1MMUTracker.hpp debug.hpp - +g1MMUTracker.hpp allocation.hpp g1RemSet.cpp bufferingOopClosure.hpp g1RemSet.cpp concurrentG1Refine.hpp g1RemSet.cpp concurrentG1RefineThread.hpp @@ -264,12 +268,13 @@ heapRegion.cpp iterator.hpp heapRegion.cpp oop.inline.hpp -heapRegion.hpp space.hpp +heapRegion.hpp space.inline.hpp heapRegion.hpp spaceDecorator.hpp heapRegion.hpp g1BlockOffsetTable.inline.hpp heapRegion.hpp watermark.hpp heapRegion.hpp g1_specialized_oop_closures.hpp heapRegion.hpp survRateGroup.hpp +heapRegion.hpp ageTable.hpp heapRegionRemSet.hpp sparsePRT.hpp @@ -283,7 +288,7 @@ heapRegionRemSet.cpp space.inline.hpp heapRegionSeq.cpp allocation.hpp -heapRegionSeq.cpp g1CollectedHeap.hpp +heapRegionSeq.cpp g1CollectedHeap.inline.hpp heapRegionSeq.cpp heapRegionSeq.hpp heapRegionSeq.hpp growableArray.hpp @@ -334,18 +339,18 @@ survRateGroup.hpp numberSeq.hpp survRateGroup.cpp allocation.hpp -survRateGroup.cpp g1CollectedHeap.hpp +survRateGroup.cpp g1CollectedHeap.inline.hpp survRateGroup.cpp g1CollectorPolicy.hpp survRateGroup.cpp heapRegion.hpp survRateGroup.cpp survRateGroup.hpp thread.cpp concurrentMarkThread.inline.hpp -universe.cpp g1CollectedHeap.hpp +universe.cpp g1CollectedHeap.inline.hpp universe.cpp g1CollectorPolicy.hpp vm_operations_g1.hpp vmGCOperations.hpp vm_operations_g1.cpp vm_operations_g1.hpp -vm_operations_g1.cpp g1CollectedHeap.hpp +vm_operations_g1.cpp g1CollectedHeap.inline.hpp vm_operations_g1.cpp isGCActiveMark.hpp
--- a/src/share/vm/gc_implementation/includeDB_gc_parNew Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/includeDB_gc_parNew Fri Feb 27 15:13:00 2009 -0800 @@ -29,6 +29,8 @@ asParNewGeneration.cpp cmsAdaptiveSizePolicy.hpp asParNewGeneration.cpp cmsGCAdaptivePolicyCounters.hpp asParNewGeneration.cpp defNewGeneration.inline.hpp +asParNewGeneration.cpp markOop.inline.hpp +asParNewGeneration.cpp markSweep.inline.hpp asParNewGeneration.cpp oop.pcgc.inline.hpp asParNewGeneration.cpp parNewGeneration.hpp asParNewGeneration.cpp referencePolicy.hpp @@ -40,7 +42,7 @@ parCardTableModRefBS.cpp java.hpp parCardTableModRefBS.cpp mutexLocker.hpp parCardTableModRefBS.cpp sharedHeap.hpp -parCardTableModRefBS.cpp space.hpp +parCardTableModRefBS.cpp space.inline.hpp parCardTableModRefBS.cpp universe.hpp parCardTableModRefBS.cpp virtualspace.hpp @@ -77,6 +79,7 @@ parNewGeneration.cpp sharedHeap.hpp parNewGeneration.cpp space.hpp parNewGeneration.cpp spaceDecorator.hpp +parNewGeneration.cpp thread.hpp parNewGeneration.cpp workgroup.hpp parNewGeneration.hpp defNewGeneration.hpp
--- a/src/share/vm/gc_implementation/includeDB_gc_parallelScavenge Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/includeDB_gc_parallelScavenge Fri Feb 27 15:13:00 2009 -0800 @@ -302,6 +302,8 @@ psOldGen.hpp spaceCounters.hpp psPermGen.cpp gcUtil.hpp +psPermGen.cpp markOop.inline.hpp +psPermGen.cpp markSweep.inline.hpp psPermGen.cpp parallelScavengeHeap.hpp psPermGen.cpp psMarkSweepDecorator.hpp psPermGen.cpp psParallelCompact.hpp
--- a/src/share/vm/gc_implementation/includeDB_gc_shared Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/includeDB_gc_shared Fri Feb 27 15:13:00 2009 -0800 @@ -100,4 +100,4 @@ spaceCounters.hpp perfData.hpp spaceCounters.hpp generationCounters.hpp -vmGCOperations.cpp g1CollectedHeap.hpp +vmGCOperations.cpp g1CollectedHeap.inline.hpp
--- a/src/share/vm/gc_implementation/parNew/parNewGeneration.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/parNew/parNewGeneration.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -404,6 +404,8 @@ if (terminator()->offer_termination()) break; par_scan_state()->end_term_time(); } + assert(par_gen()->_overflow_list == NULL && par_gen()->_num_par_pushes == 0, + "Broken overflow list?"); // Finish the last termination pause. par_scan_state()->end_term_time(); } @@ -456,6 +458,8 @@ _is_alive_closure(this), _plab_stats(YoungPLABSize, PLABWeight) { + NOT_PRODUCT(_overflow_counter = ParGCWorkQueueOverflowInterval;) + NOT_PRODUCT(_num_par_pushes = 0;) _task_queues = new ObjToScanQueueSet(ParallelGCThreads); guarantee(_task_queues != NULL, "task_queues allocation failure."); @@ -993,12 +997,19 @@ "push forwarded object"); } // Push it on one of the queues of to-be-scanned objects. - if (!par_scan_state->work_queue()->push(obj_to_push)) { + bool simulate_overflow = false; + NOT_PRODUCT( + if (ParGCWorkQueueOverflowALot && should_simulate_overflow()) { + // simulate a stack overflow + simulate_overflow = true; + } + ) + if (simulate_overflow || !par_scan_state->work_queue()->push(obj_to_push)) { // Add stats for overflow pushes. if (Verbose && PrintGCDetails) { gclog_or_tty->print("queue overflow!\n"); } - push_on_overflow_list(old); + push_on_overflow_list(old, par_scan_state); par_scan_state->note_overflow_push(); } par_scan_state->note_push(); @@ -1110,9 +1121,16 @@ "push forwarded object"); } // Push it on one of the queues of to-be-scanned objects. - if (!par_scan_state->work_queue()->push(obj_to_push)) { + bool simulate_overflow = false; + NOT_PRODUCT( + if (ParGCWorkQueueOverflowALot && should_simulate_overflow()) { + // simulate a stack overflow + simulate_overflow = true; + } + ) + if (simulate_overflow || !par_scan_state->work_queue()->push(obj_to_push)) { // Add stats for overflow pushes. - push_on_overflow_list(old); + push_on_overflow_list(old, par_scan_state); par_scan_state->note_overflow_push(); } par_scan_state->note_push(); @@ -1135,89 +1153,190 @@ return forward_ptr; } -void ParNewGeneration::push_on_overflow_list(oop from_space_obj) { - oop cur_overflow_list = _overflow_list; +#ifndef PRODUCT +// It's OK to call this multi-threaded; the worst thing +// that can happen is that we'll get a bunch of closely +// spaced simulated oveflows, but that's OK, in fact +// probably good as it would exercise the overflow code +// under contention. +bool ParNewGeneration::should_simulate_overflow() { + if (_overflow_counter-- <= 0) { // just being defensive + _overflow_counter = ParGCWorkQueueOverflowInterval; + return true; + } else { + return false; + } +} +#endif + +#define BUSY (oop(0x1aff1aff)) +void ParNewGeneration::push_on_overflow_list(oop from_space_obj, ParScanThreadState* par_scan_state) { // if the object has been forwarded to itself, then we cannot // use the klass pointer for the linked list. Instead we have // to allocate an oopDesc in the C-Heap and use that for the linked list. + // XXX This is horribly inefficient when a promotion failure occurs + // and should be fixed. XXX FIX ME !!! +#ifndef PRODUCT + Atomic::inc_ptr(&_num_par_pushes); + assert(_num_par_pushes > 0, "Tautology"); +#endif if (from_space_obj->forwardee() == from_space_obj) { oopDesc* listhead = NEW_C_HEAP_ARRAY(oopDesc, 1); listhead->forward_to(from_space_obj); from_space_obj = listhead; } - while (true) { - from_space_obj->set_klass_to_list_ptr(cur_overflow_list); - oop observed_overflow_list = + oop observed_overflow_list = _overflow_list; + oop cur_overflow_list; + do { + cur_overflow_list = observed_overflow_list; + if (cur_overflow_list != BUSY) { + from_space_obj->set_klass_to_list_ptr(cur_overflow_list); + } else { + from_space_obj->set_klass_to_list_ptr(NULL); + } + observed_overflow_list = (oop)Atomic::cmpxchg_ptr(from_space_obj, &_overflow_list, cur_overflow_list); - if (observed_overflow_list == cur_overflow_list) break; - // Otherwise... - cur_overflow_list = observed_overflow_list; - } + } while (cur_overflow_list != observed_overflow_list); } +// *NOTE*: The overflow list manipulation code here and +// in CMSCollector:: are very similar in shape, +// except that in the CMS case we thread the objects +// directly into the list via their mark word, and do +// not need to deal with special cases below related +// to chunking of object arrays and promotion failure +// handling. +// CR 6797058 has been filed to attempt consolidation of +// the common code. +// Because of the common code, if you make any changes in +// the code below, please check the CMS version to see if +// similar changes might be needed. +// See CMSCollector::par_take_from_overflow_list() for +// more extensive documentation comments. bool ParNewGeneration::take_from_overflow_list(ParScanThreadState* par_scan_state) { ObjToScanQueue* work_q = par_scan_state->work_queue(); + assert(work_q->size() == 0, "Should first empty local work queue"); // How many to take? - int objsFromOverflow = MIN2(work_q->max_elems()/4, - (juint)ParGCDesiredObjsFromOverflowList); + size_t objsFromOverflow = MIN2((size_t)work_q->max_elems()/4, + (size_t)ParGCDesiredObjsFromOverflowList); if (_overflow_list == NULL) return false; // Otherwise, there was something there; try claiming the list. - oop prefix = (oop)Atomic::xchg_ptr(NULL, &_overflow_list); - - if (prefix == NULL) { - return false; + oop prefix = (oop)Atomic::xchg_ptr(BUSY, &_overflow_list); + // Trim off a prefix of at most objsFromOverflow items + Thread* tid = Thread::current(); + size_t spin_count = (size_t)ParallelGCThreads; + size_t sleep_time_millis = MAX2((size_t)1, objsFromOverflow/100); + for (size_t spin = 0; prefix == BUSY && spin < spin_count; spin++) { + // someone grabbed it before we did ... + // ... we spin for a short while... + os::sleep(tid, sleep_time_millis, false); + if (_overflow_list == NULL) { + // nothing left to take + return false; + } else if (_overflow_list != BUSY) { + // try and grab the prefix + prefix = (oop)Atomic::xchg_ptr(BUSY, &_overflow_list); + } } - // Trim off a prefix of at most objsFromOverflow items - int i = 1; + if (prefix == NULL || prefix == BUSY) { + // Nothing to take or waited long enough + if (prefix == NULL) { + // Write back the NULL in case we overwrote it with BUSY above + // and it is still the same value. + (void) Atomic::cmpxchg_ptr(NULL, &_overflow_list, BUSY); + } + return false; + } + assert(prefix != NULL && prefix != BUSY, "Error"); + size_t i = 1; oop cur = prefix; while (i < objsFromOverflow && cur->klass_or_null() != NULL) { i++; cur = oop(cur->klass()); } // Reattach remaining (suffix) to overflow list - if (cur->klass_or_null() != NULL) { - oop suffix = oop(cur->klass()); - cur->set_klass_to_list_ptr(NULL); - - // Find last item of suffix list - oop last = suffix; - while (last->klass_or_null() != NULL) { - last = oop(last->klass()); + if (cur->klass_or_null() == NULL) { + // Write back the NULL in lieu of the BUSY we wrote + // above and it is still the same value. + if (_overflow_list == BUSY) { + (void) Atomic::cmpxchg_ptr(NULL, &_overflow_list, BUSY); } - // Atomically prepend suffix to current overflow list - oop cur_overflow_list = _overflow_list; - while (true) { - last->set_klass_to_list_ptr(cur_overflow_list); - oop observed_overflow_list = - (oop)Atomic::cmpxchg_ptr(suffix, &_overflow_list, cur_overflow_list); - if (observed_overflow_list == cur_overflow_list) break; - // Otherwise... - cur_overflow_list = observed_overflow_list; + } else { + assert(cur->klass_or_null() != BUSY, "Error"); + oop suffix = oop(cur->klass()); // suffix will be put back on global list + cur->set_klass_to_list_ptr(NULL); // break off suffix + // It's possible that the list is still in the empty(busy) state + // we left it in a short while ago; in that case we may be + // able to place back the suffix. + oop observed_overflow_list = _overflow_list; + oop cur_overflow_list = observed_overflow_list; + bool attached = false; + while (observed_overflow_list == BUSY || observed_overflow_list == NULL) { + observed_overflow_list = + (oop) Atomic::cmpxchg_ptr(suffix, &_overflow_list, cur_overflow_list); + if (cur_overflow_list == observed_overflow_list) { + attached = true; + break; + } else cur_overflow_list = observed_overflow_list; + } + if (!attached) { + // Too bad, someone else got in in between; we'll need to do a splice. + // Find the last item of suffix list + oop last = suffix; + while (last->klass_or_null() != NULL) { + last = oop(last->klass()); + } + // Atomically prepend suffix to current overflow list + observed_overflow_list = _overflow_list; + do { + cur_overflow_list = observed_overflow_list; + if (cur_overflow_list != BUSY) { + // Do the splice ... + last->set_klass_to_list_ptr(cur_overflow_list); + } else { // cur_overflow_list == BUSY + last->set_klass_to_list_ptr(NULL); + } + observed_overflow_list = + (oop)Atomic::cmpxchg_ptr(suffix, &_overflow_list, cur_overflow_list); + } while (cur_overflow_list != observed_overflow_list); } } // Push objects on prefix list onto this thread's work queue - assert(cur != NULL, "program logic"); + assert(prefix != NULL && prefix != BUSY, "program logic"); cur = prefix; - int n = 0; + ssize_t n = 0; while (cur != NULL) { oop obj_to_push = cur->forwardee(); oop next = oop(cur->klass_or_null()); cur->set_klass(obj_to_push->klass()); - if (par_scan_state->should_be_partially_scanned(obj_to_push, cur)) { - obj_to_push = cur; + // This may be an array object that is self-forwarded. In that case, the list pointer + // space, cur, is not in the Java heap, but rather in the C-heap and should be freed. + if (!is_in_reserved(cur)) { + // This can become a scaling bottleneck when there is work queue overflow coincident + // with promotion failure. + oopDesc* f = cur; + FREE_C_HEAP_ARRAY(oopDesc, f); + } else if (par_scan_state->should_be_partially_scanned(obj_to_push, cur)) { assert(arrayOop(cur)->length() == 0, "entire array remaining to be scanned"); + obj_to_push = cur; } - work_q->push(obj_to_push); + bool ok = work_q->push(obj_to_push); + assert(ok, "Should have succeeded"); cur = next; n++; } 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); +#endif return true; } +#undef BUSY void ParNewGeneration::ref_processor_init() {
--- a/src/share/vm/gc_implementation/parNew/parNewGeneration.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/parNew/parNewGeneration.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -278,6 +278,7 @@ friend class ParNewRefProcTask; friend class ParNewRefProcTaskExecutor; friend class ParScanThreadStateSet; + friend class ParEvacuateFollowersClosure; private: // XXX use a global constant instead of 64! @@ -296,6 +297,7 @@ // klass-pointers (klass information already copied to the forwarded // image.) Manipulated with CAS. oop _overflow_list; + NOT_PRODUCT(ssize_t _num_par_pushes;) // If true, older generation does not support promotion undo, so avoid. static bool _avoid_promotion_undo; @@ -372,8 +374,12 @@ oop copy_to_survivor_space_with_undo(ParScanThreadState* par_scan_state, oop obj, size_t obj_sz, markOop m); + // in support of testing overflow code + NOT_PRODUCT(int _overflow_counter;) + NOT_PRODUCT(bool should_simulate_overflow();) + // Push the given (from-space) object on the global overflow list. - void push_on_overflow_list(oop from_space_obj); + void push_on_overflow_list(oop from_space_obj, ParScanThreadState* par_scan_state); // If the global overflow list is non-empty, move some tasks from it // onto "work_q" (which must be empty). No more than 1/4 of the
--- a/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -200,6 +200,7 @@ void oop_iterate(OopClosure* cl); void object_iterate(ObjectClosure* cl); + void safe_object_iterate(ObjectClosure* cl) { object_iterate(cl); } void permanent_oop_iterate(OopClosure* cl); void permanent_object_iterate(ObjectClosure* cl);
--- a/src/share/vm/gc_implementation/parallelScavenge/psMarkSweep.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/parallelScavenge/psMarkSweep.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -362,6 +362,10 @@ if (PrintHeapAtGC) { Universe::print_heap_after_gc(); } + +#ifdef TRACESPINNING + ParallelTaskTerminator::print_termination_counts(); +#endif } bool PSMarkSweep::absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy,
--- a/src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -116,7 +116,7 @@ // ObjectSpace stuff // - _object_space = new MutableSpace(); + _object_space = new MutableSpace(virtual_space()->alignment()); if (_object_space == NULL) vm_exit_during_initialization("Could not allocate an old gen space"); @@ -385,10 +385,10 @@ start_array()->set_covered_region(new_memregion); Universe::heap()->barrier_set()->resize_covered_region(new_memregion); - HeapWord* const virtual_space_high = (HeapWord*) virtual_space()->high(); - // ALWAYS do this last!! - object_space()->set_end(virtual_space_high); + object_space()->initialize(new_memregion, + SpaceDecorator::DontClear, + SpaceDecorator::DontMangle); assert(new_word_size == heap_word_size(object_space()->capacity_in_bytes()), "Sanity");
--- a/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -2203,6 +2203,10 @@ collection_exit.ticks()); gc_task_manager()->print_task_time_stamps(); } + +#ifdef TRACESPINNING + ParallelTaskTerminator::print_termination_counts(); +#endif } bool PSParallelCompact::absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy,
--- a/src/share/vm/gc_implementation/parallelScavenge/psScavenge.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/parallelScavenge/psScavenge.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -615,6 +615,10 @@ gc_task_manager()->print_task_time_stamps(); } +#ifdef TRACESPINNING + ParallelTaskTerminator::print_termination_counts(); +#endif + return !promotion_failure_occurred; }
--- a/src/share/vm/gc_implementation/parallelScavenge/psVirtualspace.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/parallelScavenge/psVirtualspace.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -78,7 +78,7 @@ _special = false; } -bool PSVirtualSpace::expand_by(size_t bytes, bool pre_touch) { +bool PSVirtualSpace::expand_by(size_t bytes) { assert(is_aligned(bytes), "arg not aligned"); DEBUG_ONLY(PSVirtualSpaceVerifier this_verifier(this)); @@ -92,15 +92,6 @@ _committed_high_addr += bytes; } - if (pre_touch || AlwaysPreTouch) { - for (char* curr = base_addr; - curr < _committed_high_addr; - curr += os::vm_page_size()) { - char tmp = *curr; - *curr = 0; - } - } - return result; } @@ -255,7 +246,7 @@ DEBUG_ONLY(verify()); } -bool PSVirtualSpaceHighToLow::expand_by(size_t bytes, bool pre_touch) { +bool PSVirtualSpaceHighToLow::expand_by(size_t bytes) { assert(is_aligned(bytes), "arg not aligned"); DEBUG_ONLY(PSVirtualSpaceVerifier this_verifier(this)); @@ -269,15 +260,6 @@ _committed_low_addr -= bytes; } - if (pre_touch || AlwaysPreTouch) { - for (char* curr = base_addr; - curr < _committed_high_addr; - curr += os::vm_page_size()) { - char tmp = *curr; - *curr = 0; - } - } - return result; }
--- a/src/share/vm/gc_implementation/parallelScavenge/psVirtualspace.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/parallelScavenge/psVirtualspace.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -80,7 +80,7 @@ inline void set_reserved(char* low_addr, char* high_addr, bool special); inline void set_reserved(ReservedSpace rs); inline void set_committed(char* low_addr, char* high_addr); - virtual bool expand_by(size_t bytes, bool pre_touch = false); + virtual bool expand_by(size_t bytes); virtual bool shrink_by(size_t bytes); virtual size_t expand_into(PSVirtualSpace* space, size_t bytes); void release(); @@ -127,7 +127,7 @@ PSVirtualSpaceHighToLow(ReservedSpace rs, size_t alignment); PSVirtualSpaceHighToLow(ReservedSpace rs); - virtual bool expand_by(size_t bytes, bool pre_touch = false); + virtual bool expand_by(size_t bytes); virtual bool shrink_by(size_t bytes); virtual size_t expand_into(PSVirtualSpace* space, size_t bytes);
--- a/src/share/vm/gc_implementation/parallelScavenge/psYoungGen.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/parallelScavenge/psYoungGen.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -64,12 +64,12 @@ } if (UseNUMA) { - _eden_space = new MutableNUMASpace(); + _eden_space = new MutableNUMASpace(virtual_space()->alignment()); } else { - _eden_space = new MutableSpace(); + _eden_space = new MutableSpace(virtual_space()->alignment()); } - _from_space = new MutableSpace(); - _to_space = new MutableSpace(); + _from_space = new MutableSpace(virtual_space()->alignment()); + _to_space = new MutableSpace(virtual_space()->alignment()); if (_eden_space == NULL || _from_space == NULL || _to_space == NULL) { vm_exit_during_initialization("Could not allocate a young gen space");
--- a/src/share/vm/gc_implementation/shared/ageTable.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/shared/ageTable.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -67,6 +67,12 @@ } } +void ageTable::merge_par(ageTable* subTable) { + for (int i = 0; i < table_size; i++) { + Atomic::add_ptr(subTable->sizes[i], &sizes[i]); + } +} + int ageTable::compute_tenuring_threshold(size_t survivor_capacity) { size_t desired_survivor_size = (size_t)((((double) survivor_capacity)*TargetSurvivorRatio)/100); size_t total = 0;
--- a/src/share/vm/gc_implementation/shared/ageTable.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/shared/ageTable.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -56,6 +56,7 @@ // Merge another age table with the current one. Used // for parallel young generation gc. void merge(ageTable* subTable); + void merge_par(ageTable* subTable); // calculate new tenuring threshold based on age information int compute_tenuring_threshold(size_t survivor_capacity);
--- a/src/share/vm/gc_implementation/shared/mutableNUMASpace.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/shared/mutableNUMASpace.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -27,7 +27,7 @@ # include "incls/_mutableNUMASpace.cpp.incl" -MutableNUMASpace::MutableNUMASpace() { +MutableNUMASpace::MutableNUMASpace(size_t alignment) : MutableSpace(alignment) { _lgrp_spaces = new (ResourceObj::C_HEAP) GrowableArray<LGRPSpace*>(0, true); _page_size = os::vm_page_size(); _adaptation_cycles = 0; @@ -221,7 +221,7 @@ } } if (!found) { - lgrp_spaces()->append(new LGRPSpace(lgrp_ids[i])); + lgrp_spaces()->append(new LGRPSpace(lgrp_ids[i], alignment())); } } @@ -443,10 +443,10 @@ // Is there bottom? if (new_region.start() < intersection.start()) { // Yes // Try to coalesce small pages into a large one. - if (UseLargePages && page_size() >= os::large_page_size()) { - HeapWord* p = (HeapWord*)round_to((intptr_t) intersection.start(), os::large_page_size()); + if (UseLargePages && page_size() >= alignment()) { + HeapWord* p = (HeapWord*)round_to((intptr_t) intersection.start(), alignment()); if (new_region.contains(p) - && pointer_delta(p, new_region.start(), sizeof(char)) >= os::large_page_size()) { + && pointer_delta(p, new_region.start(), sizeof(char)) >= alignment()) { if (intersection.contains(p)) { intersection = MemRegion(p, intersection.end()); } else { @@ -462,10 +462,10 @@ // Is there top? if (intersection.end() < new_region.end()) { // Yes // Try to coalesce small pages into a large one. - if (UseLargePages && page_size() >= os::large_page_size()) { - HeapWord* p = (HeapWord*)round_down((intptr_t) intersection.end(), os::large_page_size()); + if (UseLargePages && page_size() >= alignment()) { + HeapWord* p = (HeapWord*)round_down((intptr_t) intersection.end(), alignment()); if (new_region.contains(p) - && pointer_delta(new_region.end(), p, sizeof(char)) >= os::large_page_size()) { + && pointer_delta(new_region.end(), p, sizeof(char)) >= alignment()) { if (intersection.contains(p)) { intersection = MemRegion(intersection.start(), p); } else { @@ -504,12 +504,12 @@ // That's the only case we have to make an additional bias_region() call. HeapWord* start = invalid_region->start(); HeapWord* end = invalid_region->end(); - if (UseLargePages && page_size() >= os::large_page_size()) { - HeapWord *p = (HeapWord*)round_down((intptr_t) start, os::large_page_size()); + if (UseLargePages && page_size() >= alignment()) { + HeapWord *p = (HeapWord*)round_down((intptr_t) start, alignment()); if (new_region.contains(p)) { start = p; } - p = (HeapWord*)round_to((intptr_t) end, os::large_page_size()); + p = (HeapWord*)round_to((intptr_t) end, alignment()); if (new_region.contains(end)) { end = p; } @@ -526,7 +526,8 @@ void MutableNUMASpace::initialize(MemRegion mr, bool clear_space, - bool mangle_space) { + bool mangle_space, + bool setup_pages) { assert(clear_space, "Reallocation will destory data!"); assert(lgrp_spaces()->length() > 0, "There should be at least one space"); @@ -538,7 +539,7 @@ // Compute chunk sizes size_t prev_page_size = page_size(); - set_page_size(UseLargePages ? os::large_page_size() : os::vm_page_size()); + set_page_size(UseLargePages ? alignment() : os::vm_page_size()); HeapWord* rounded_bottom = (HeapWord*)round_to((intptr_t) bottom(), page_size()); HeapWord* rounded_end = (HeapWord*)round_down((intptr_t) end(), page_size()); size_t base_space_size_pages = pointer_delta(rounded_end, rounded_bottom, sizeof(char)) / page_size(); @@ -666,7 +667,7 @@ } // Clear space (set top = bottom) but never mangle. - s->initialize(new_region, SpaceDecorator::Clear, SpaceDecorator::DontMangle); + s->initialize(new_region, SpaceDecorator::Clear, SpaceDecorator::DontMangle, MutableSpace::DontSetupPages); set_adaptation_cycles(samples_count()); }
--- a/src/share/vm/gc_implementation/shared/mutableNUMASpace.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/shared/mutableNUMASpace.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -82,8 +82,8 @@ char* last_page_scanned() { return _last_page_scanned; } void set_last_page_scanned(char* p) { _last_page_scanned = p; } public: - LGRPSpace(int l) : _lgrp_id(l), _last_page_scanned(NULL), _allocation_failed(false) { - _space = new MutableSpace(); + LGRPSpace(int l, size_t alignment) : _lgrp_id(l), _last_page_scanned(NULL), _allocation_failed(false) { + _space = new MutableSpace(alignment); _alloc_rate = new AdaptiveWeightedAverage(NUMAChunkResizeWeight); } ~LGRPSpace() { @@ -183,10 +183,10 @@ public: GrowableArray<LGRPSpace*>* lgrp_spaces() const { return _lgrp_spaces; } - MutableNUMASpace(); + MutableNUMASpace(size_t alignment); virtual ~MutableNUMASpace(); // Space initialization. - virtual void initialize(MemRegion mr, bool clear_space, bool mangle_space); + virtual void initialize(MemRegion mr, bool clear_space, bool mangle_space, bool setup_pages = SetupPages); // Update space layout if necessary. Do all adaptive resizing job. virtual void update(); // Update allocation rate averages.
--- a/src/share/vm/gc_implementation/shared/mutableSpace.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/shared/mutableSpace.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -25,7 +25,10 @@ # include "incls/_precompiled.incl" # include "incls/_mutableSpace.cpp.incl" -MutableSpace::MutableSpace(): ImmutableSpace(), _top(NULL) { +MutableSpace::MutableSpace(size_t alignment): ImmutableSpace(), _top(NULL), _alignment(alignment) { + assert(MutableSpace::alignment() >= 0 && + MutableSpace::alignment() % os::vm_page_size() == 0, + "Space should be aligned"); _mangler = new MutableSpaceMangler(this); } @@ -33,16 +36,88 @@ delete _mangler; } +void MutableSpace::numa_setup_pages(MemRegion mr, bool clear_space) { + if (!mr.is_empty()) { + size_t page_size = UseLargePages ? alignment() : os::vm_page_size(); + HeapWord *start = (HeapWord*)round_to((intptr_t) mr.start(), page_size); + HeapWord *end = (HeapWord*)round_down((intptr_t) mr.end(), page_size); + if (end > start) { + size_t size = pointer_delta(end, start, sizeof(char)); + if (clear_space) { + // Prefer page reallocation to migration. + os::free_memory((char*)start, size); + } + os::numa_make_global((char*)start, size); + } + } +} + +void MutableSpace::pretouch_pages(MemRegion mr) { + for (volatile char *p = (char*)mr.start(); p < (char*)mr.end(); p += os::vm_page_size()) { + char t = *p; *p = t; + } +} + void MutableSpace::initialize(MemRegion mr, bool clear_space, - bool mangle_space) { - HeapWord* bottom = mr.start(); - HeapWord* end = mr.end(); + bool mangle_space, + bool setup_pages) { + + assert(Universe::on_page_boundary(mr.start()) && Universe::on_page_boundary(mr.end()), + "invalid space boundaries"); - assert(Universe::on_page_boundary(bottom) && Universe::on_page_boundary(end), - "invalid space boundaries"); - set_bottom(bottom); - set_end(end); + if (setup_pages && (UseNUMA || AlwaysPreTouch)) { + // The space may move left and right or expand/shrink. + // We'd like to enforce the desired page placement. + MemRegion head, tail; + if (last_setup_region().is_empty()) { + // If it's the first initialization don't limit the amount of work. + head = mr; + tail = MemRegion(mr.end(), mr.end()); + } else { + // Is there an intersection with the address space? + MemRegion intersection = last_setup_region().intersection(mr); + if (intersection.is_empty()) { + intersection = MemRegion(mr.end(), mr.end()); + } + // All the sizes below are in words. + size_t head_size = 0, tail_size = 0; + if (mr.start() <= intersection.start()) { + head_size = pointer_delta(intersection.start(), mr.start()); + } + if(intersection.end() <= mr.end()) { + tail_size = pointer_delta(mr.end(), intersection.end()); + } + // Limit the amount of page manipulation if necessary. + if (NUMASpaceResizeRate > 0 && !AlwaysPreTouch) { + const size_t change_size = head_size + tail_size; + const float setup_rate_words = NUMASpaceResizeRate >> LogBytesPerWord; + head_size = MIN2((size_t)(setup_rate_words * head_size / change_size), + head_size); + tail_size = MIN2((size_t)(setup_rate_words * tail_size / change_size), + tail_size); + } + head = MemRegion(intersection.start() - head_size, intersection.start()); + tail = MemRegion(intersection.end(), intersection.end() + tail_size); + } + assert(mr.contains(head) && mr.contains(tail), "Sanity"); + + if (UseNUMA) { + numa_setup_pages(head, clear_space); + numa_setup_pages(tail, clear_space); + } + + if (AlwaysPreTouch) { + pretouch_pages(head); + pretouch_pages(tail); + } + + // Remember where we stopped so that we can continue later. + set_last_setup_region(MemRegion(head.start(), tail.end())); + } + + set_bottom(mr.start()); + set_end(mr.end()); if (clear_space) { clear(mangle_space);
--- a/src/share/vm/gc_implementation/shared/mutableSpace.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_implementation/shared/mutableSpace.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -25,7 +25,10 @@ // A MutableSpace is a subtype of ImmutableSpace that supports the // concept of allocation. This includes the concepts that a space may // be only partially full, and the querry methods that go with such -// an assumption. +// an assumption. MutableSpace is also responsible for minimizing the +// page allocation time by having the memory pretouched (with +// AlwaysPretouch) and for optimizing page placement on NUMA systems +// by make the underlying region interleaved (with UseNUMA). // // Invariant: (ImmutableSpace +) bottom() <= top() <= end() // top() is inclusive and end() is exclusive. @@ -37,15 +40,23 @@ // Helper for mangling unused space in debug builds MutableSpaceMangler* _mangler; - + // The last region which page had been setup to be interleaved. + MemRegion _last_setup_region; + size_t _alignment; protected: HeapWord* _top; MutableSpaceMangler* mangler() { return _mangler; } + void numa_setup_pages(MemRegion mr, bool clear_space); + void pretouch_pages(MemRegion mr); + + void set_last_setup_region(MemRegion mr) { _last_setup_region = mr; } + MemRegion last_setup_region() const { return _last_setup_region; } + public: virtual ~MutableSpace(); - MutableSpace(); + MutableSpace(size_t page_size); // Accessors HeapWord* top() const { return _top; } @@ -57,13 +68,20 @@ virtual void set_bottom(HeapWord* value) { _bottom = value; } virtual void set_end(HeapWord* value) { _end = value; } + size_t alignment() { return _alignment; } + // Returns a subregion containing all objects in this space. MemRegion used_region() { return MemRegion(bottom(), top()); } + static const bool SetupPages = true; + static const bool DontSetupPages = false; + // Initialization virtual void initialize(MemRegion mr, bool clear_space, - bool mangle_space); + bool mangle_space, + bool setup_pages = SetupPages); + virtual void clear(bool mangle_space); // Does the usual initialization but optionally resets top to bottom. #if 0 // MANGLE_SPACE
--- a/src/share/vm/gc_interface/collectedHeap.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/gc_interface/collectedHeap.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -42,6 +42,7 @@ class CollectedHeap : public CHeapObj { friend class VMStructs; friend class IsGCActiveMark; // Block structured external access to _is_gc_active + friend class constantPoolCacheKlass; // allocate() method inserts is_conc_safe #ifdef ASSERT static int _fire_out_of_memory_count; @@ -82,8 +83,6 @@ // Reinitialize tlabs before resuming mutators. virtual void resize_all_tlabs(); - debug_only(static void check_for_valid_allocation_state();) - protected: // Allocate from the current thread's TLAB, with broken-out slow path. inline static HeapWord* allocate_from_tlab(Thread* thread, size_t size); @@ -142,6 +141,7 @@ PRODUCT_RETURN; virtual void check_for_non_bad_heap_word_value(HeapWord* addr, size_t size) PRODUCT_RETURN; + debug_only(static void check_for_valid_allocation_state();) public: enum Name { @@ -466,6 +466,10 @@ // This includes objects in permanent memory. virtual void object_iterate(ObjectClosure* cl) = 0; + // Similar to object_iterate() except iterates only + // over live objects. + virtual void safe_object_iterate(ObjectClosure* cl) = 0; + // Behaves the same as oop_iterate, except only traverses // interior pointers contained in permanent memory. If there // is no permanent memory, does nothing.
--- a/src/share/vm/includeDB_compiler2 Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/includeDB_compiler2 Fri Feb 27 15:13:00 2009 -0800 @@ -140,6 +140,7 @@ c2_globals_<os_family>.hpp macros.hpp c2_init_<arch>.cpp compile.hpp +c2_init_<arch>.cpp node.hpp c2compiler.cpp ad_<arch_model>.hpp c2compiler.cpp c2compiler.hpp @@ -839,6 +840,7 @@ phase.cpp compile.hpp phase.cpp compileBroker.hpp phase.cpp nmethod.hpp +phase.cpp node.hpp phase.cpp phase.hpp phase.hpp port.hpp
--- a/src/share/vm/includeDB_core Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/includeDB_core Fri Feb 27 15:13:00 2009 -0800 @@ -1311,6 +1311,7 @@ cppInterpreter_<arch>.cpp debug.hpp cppInterpreter_<arch>.cpp deoptimization.hpp cppInterpreter_<arch>.cpp frame.inline.hpp +cppInterpreter_<arch>.cpp interfaceSupport.hpp cppInterpreter_<arch>.cpp interpreterRuntime.hpp cppInterpreter_<arch>.cpp interpreter.hpp cppInterpreter_<arch>.cpp interpreterGenerator.hpp @@ -2014,7 +2015,7 @@ instanceKlass.cpp vmSymbols.hpp instanceKlass.hpp accessFlags.hpp -instanceKlass.hpp bitMap.hpp +instanceKlass.hpp bitMap.inline.hpp instanceKlass.hpp constMethodOop.hpp instanceKlass.hpp constantPoolOop.hpp instanceKlass.hpp handles.hpp @@ -3771,6 +3772,7 @@ spaceDecorator.cpp copy.hpp spaceDecorator.cpp spaceDecorator.hpp +spaceDecorator.cpp space.inline.hpp specialized_oop_closures.cpp ostream.hpp specialized_oop_closures.cpp specialized_oop_closures.hpp
--- a/src/share/vm/includeDB_features Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/includeDB_features Fri Feb 27 15:13:00 2009 -0800 @@ -59,6 +59,8 @@ dump_<arch_model>.cpp assembler_<arch>.inline.hpp dump_<arch_model>.cpp compactingPermGenGen.hpp +dump_<arch_model>.cpp generation.inline.hpp +dump_<arch_model>.cpp space.inline.hpp forte.cpp collectedHeap.inline.hpp forte.cpp debugInfoRec.hpp
--- a/src/share/vm/interpreter/bytecodeInterpreter.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/interpreter/bytecodeInterpreter.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -163,7 +163,7 @@ #ifdef USELABELS // Have to do this dispatch this way in C++ because otherwise gcc complains about crossing an // initialization (which is is the initialization of the table pointer...) -#define DISPATCH(opcode) goto *dispatch_table[opcode] +#define DISPATCH(opcode) goto *(void*)dispatch_table[opcode] #define CONTINUE { \ opcode = *pc; \ DO_UPDATE_INSTRUCTION_COUNT(opcode); \ @@ -341,7 +341,7 @@ */ #undef CHECK_NULL #define CHECK_NULL(obj_) \ - if ((obj_) == 0) { \ + if ((obj_) == NULL) { \ VM_JAVA_ERROR(vmSymbols::java_lang_NullPointerException(), ""); \ } @@ -1362,7 +1362,7 @@ #define NULL_COMPARISON_NOT_OP(name) \ CASE(_if##name): { \ - int skip = (!(STACK_OBJECT(-1) == 0)) \ + int skip = (!(STACK_OBJECT(-1) == NULL)) \ ? (int16_t)Bytes::get_Java_u2(pc + 1) : 3; \ address branch_pc = pc; \ UPDATE_PC_AND_TOS(skip, -1); \ @@ -1372,7 +1372,7 @@ #define NULL_COMPARISON_OP(name) \ CASE(_if##name): { \ - int skip = ((STACK_OBJECT(-1) == 0)) \ + int skip = ((STACK_OBJECT(-1) == NULL)) \ ? (int16_t)Bytes::get_Java_u2(pc + 1) : 3; \ address branch_pc = pc; \ UPDATE_PC_AND_TOS(skip, -1); \
--- a/src/share/vm/interpreter/bytecodeInterpreter.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/interpreter/bytecodeInterpreter.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -66,7 +66,6 @@ friend class InterpreterGenerator; friend class InterpreterMacroAssembler; friend class frame; -friend class SharedRuntime; friend class VMStructs; public:
--- a/src/share/vm/interpreter/rewriter.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/interpreter/rewriter.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -48,9 +48,14 @@ // Creates a constant pool cache given an inverse_index_map +// This creates the constant pool cache initially in a state +// that is unsafe for concurrent GC processing but sets it to +// a safe mode before the constant pool cache is returned. constantPoolCacheHandle Rewriter::new_constant_pool_cache(intArray& inverse_index_map, TRAPS) { const int length = inverse_index_map.length(); - constantPoolCacheOop cache = oopFactory::new_constantPoolCache(length, CHECK_(constantPoolCacheHandle())); + constantPoolCacheOop cache = oopFactory::new_constantPoolCache(length, + methodOopDesc::IsUnsafeConc, + CHECK_(constantPoolCacheHandle())); cache->initialize(inverse_index_map); return constantPoolCacheHandle(THREAD, cache); }
--- a/src/share/vm/libadt/dict.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/libadt/dict.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -346,9 +346,12 @@ return strcmp((const char *)k1,(const char *)k2); } -// Slimey cheap key comparator. +// Cheap key comparator. int32 cmpkey(const void *key1, const void *key2) { - return (int32)((intptr_t)key1 - (intptr_t)key2); + if (key1 == key2) return 0; + intptr_t delta = (intptr_t)key1 - (intptr_t)key2; + if (delta > 0) return 1; + return -1; } //=============================================================================
--- a/src/share/vm/libadt/port.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/libadt/port.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -34,17 +34,6 @@ #include <stddef.h> #include <stdlib.h> #include <string.h> -#undef bzero -inline void bzero(void *b, int len) { memset(b,0,len); } -#undef bcopy -inline void bcopy(const void *s, void *d, size_t len) { memmove(d,s,len); } -#undef bcmp -inline int bcmp(const void *s,const void *t,int len) { return memcmp(s,t,len);} -extern "C" unsigned long strtoul(const char *s, char **end, int base); - -// Definition for sys_errlist varies from Sun 4.1 & Solaris. -// We use the new Solaris definition. -#include <string.h> // Access to the C++ class virtual function pointer // Put the class in the macro
--- a/src/share/vm/memory/genCollectedHeap.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/memory/genCollectedHeap.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -610,6 +610,10 @@ Universe::print_heap_after_gc(); } +#ifdef TRACESPINNING + ParallelTaskTerminator::print_termination_counts(); +#endif + if (ExitAfterGCNum > 0 && total_collections() == ExitAfterGCNum) { tty->print_cr("Stopping after GC #%d", ExitAfterGCNum); vm_exit(-1); @@ -910,6 +914,13 @@ perm_gen()->object_iterate(cl); } +void GenCollectedHeap::safe_object_iterate(ObjectClosure* cl) { + for (int i = 0; i < _n_gens; i++) { + _gens[i]->safe_object_iterate(cl); + } + perm_gen()->safe_object_iterate(cl); +} + void GenCollectedHeap::object_iterate_since_last_GC(ObjectClosure* cl) { for (int i = 0; i < _n_gens; i++) { _gens[i]->object_iterate_since_last_GC(cl);
--- a/src/share/vm/memory/genCollectedHeap.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/memory/genCollectedHeap.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -215,6 +215,7 @@ void oop_iterate(OopClosure* cl); void oop_iterate(MemRegion mr, OopClosure* cl); void object_iterate(ObjectClosure* cl); + void safe_object_iterate(ObjectClosure* cl); void object_iterate_since_last_GC(ObjectClosure* cl); Space* space_containing(const void* addr) const;
--- a/src/share/vm/memory/generation.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/memory/generation.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -319,6 +319,21 @@ space_iterate(&blk); } +class GenerationSafeObjIterateClosure : public SpaceClosure { + private: + ObjectClosure* _cl; + public: + virtual void do_space(Space* s) { + s->safe_object_iterate(_cl); + } + GenerationSafeObjIterateClosure(ObjectClosure* cl) : _cl(cl) {} +}; + +void Generation::safe_object_iterate(ObjectClosure* cl) { + GenerationSafeObjIterateClosure blk(cl); + space_iterate(&blk); +} + void Generation::prepare_for_compaction(CompactPoint* cp) { // Generic implementation, can be specialized CompactibleSpace* space = first_compaction_space();
--- a/src/share/vm/memory/generation.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/memory/generation.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -518,6 +518,11 @@ // each. virtual void object_iterate(ObjectClosure* cl); + // Iterate over all safe objects in the generation, calling "cl.do_object" on + // each. An object is safe if its references point to other objects in + // the heap. This defaults to object_iterate() unless overridden. + virtual void safe_object_iterate(ObjectClosure* cl); + // Iterate over all objects allocated in the generation since the last // collection, calling "cl.do_object" on each. The generation must have // been initialized properly to support this function, or else this call
--- a/src/share/vm/memory/heapInspection.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/memory/heapInspection.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -263,6 +263,9 @@ if (!cit.allocation_failed()) { // Iterate over objects in the heap RecordInstanceClosure ric(&cit); + // If this operation encounters a bad object when using CMS, + // consider using safe_object_iterate() which avoids perm gen + // objects that may contain bad references. Universe::heap()->object_iterate(&ric); // Report if certain classes are not counted because of @@ -317,5 +320,8 @@ // Iterate over objects in the heap FindInstanceClosure fic(k, result); + // If this operation encounters a bad object when using CMS, + // consider using safe_object_iterate() which avoids perm gen + // objects that may contain bad references. Universe::heap()->object_iterate(&fic); }
--- a/src/share/vm/memory/oopFactory.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/memory/oopFactory.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -82,15 +82,19 @@ } -constantPoolOop oopFactory::new_constantPool(int length, TRAPS) { +constantPoolOop oopFactory::new_constantPool(int length, + bool is_conc_safe, + TRAPS) { constantPoolKlass* ck = constantPoolKlass::cast(Universe::constantPoolKlassObj()); - return ck->allocate(length, CHECK_NULL); + return ck->allocate(length, is_conc_safe, CHECK_NULL); } -constantPoolCacheOop oopFactory::new_constantPoolCache(int length, TRAPS) { +constantPoolCacheOop oopFactory::new_constantPoolCache(int length, + bool is_conc_safe, + TRAPS) { constantPoolCacheKlass* ck = constantPoolCacheKlass::cast(Universe::constantPoolCacheKlassObj()); - return ck->allocate(length, CHECK_NULL); + return ck->allocate(length, is_conc_safe, CHECK_NULL); } @@ -105,11 +109,13 @@ int compressed_line_number_size, int localvariable_table_length, int checked_exceptions_length, + bool is_conc_safe, TRAPS) { klassOop cmkObj = Universe::constMethodKlassObj(); constMethodKlass* cmk = constMethodKlass::cast(cmkObj); return cmk->allocate(byte_code_size, compressed_line_number_size, localvariable_table_length, checked_exceptions_length, + is_conc_safe, CHECK_NULL); } @@ -117,14 +123,17 @@ methodOop oopFactory::new_method(int byte_code_size, AccessFlags access_flags, int compressed_line_number_size, int localvariable_table_length, - int checked_exceptions_length, TRAPS) { + int checked_exceptions_length, + bool is_conc_safe, + TRAPS) { methodKlass* mk = methodKlass::cast(Universe::methodKlassObj()); assert(!access_flags.is_native() || byte_code_size == 0, "native methods should not contain byte codes"); constMethodOop cm = new_constMethod(byte_code_size, compressed_line_number_size, localvariable_table_length, - checked_exceptions_length, CHECK_NULL); + checked_exceptions_length, + is_conc_safe, CHECK_NULL); constMethodHandle rw(THREAD, cm); return mk->allocate(rw, access_flags, CHECK_NULL); }
--- a/src/share/vm/memory/oopFactory.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/memory/oopFactory.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -81,8 +81,12 @@ static symbolHandle new_symbol_handle(const char* name, TRAPS) { return new_symbol_handle(name, (int)strlen(name), CHECK_(symbolHandle())); } // Constant pools - static constantPoolOop new_constantPool (int length, TRAPS); - static constantPoolCacheOop new_constantPoolCache(int length, TRAPS); + static constantPoolOop new_constantPool (int length, + bool is_conc_safe, + TRAPS); + static constantPoolCacheOop new_constantPoolCache(int length, + bool is_conc_safe, + TRAPS); // Instance classes static klassOop new_instanceKlass(int vtable_len, int itable_len, int static_field_size, @@ -93,9 +97,20 @@ static constMethodOop new_constMethod(int byte_code_size, int compressed_line_number_size, int localvariable_table_length, - int checked_exceptions_length, TRAPS); + int checked_exceptions_length, + bool is_conc_safe, + TRAPS); public: - static methodOop new_method(int byte_code_size, AccessFlags access_flags, int compressed_line_number_size, int localvariable_table_length, int checked_exceptions_length, TRAPS); + // Set is_conc_safe for methods which cannot safely be + // processed by concurrent GC even after the return of + // the method. + static methodOop new_method(int byte_code_size, + AccessFlags access_flags, + int compressed_line_number_size, + int localvariable_table_length, + int checked_exceptions_length, + bool is_conc_safe, + TRAPS); // Method Data containers static methodDataOop new_methodData(methodHandle method, TRAPS);
--- a/src/share/vm/memory/referenceProcessor.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/memory/referenceProcessor.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -721,12 +721,6 @@ iter.obj(), iter.obj()->blueprint()->internal_name()); } assert(iter.obj()->is_oop(UseConcMarkSweepGC), "Adding a bad reference"); - // If discovery is concurrent, we may have objects with null referents, - // being those that were concurrently cleared after they were discovered - // (and not subsequently precleaned). - assert( (discovery_is_atomic() && iter.referent()->is_oop()) - || (!discovery_is_atomic() && iter.referent()->is_oop_or_null(UseConcMarkSweepGC)), - "Adding a bad referent"); iter.next(); } // Remember to keep sentinel pointer around
--- a/src/share/vm/memory/space.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/memory/space.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -569,7 +569,15 @@ if (prev > mr.start()) { region_start_addr = prev; blk_start_addr = prev; - assert(blk_start_addr == block_start(region_start_addr), "invariant"); + // The previous invocation may have pushed "prev" beyond the + // last allocated block yet there may be still be blocks + // in this region due to a particular coalescing policy. + // Relax the assertion so that the case where the unallocated + // block is maintained and "prev" is beyond the unallocated + // block does not cause the assertion to fire. + assert((BlockOffsetArrayUseUnallocatedBlock && + (!is_in(prev))) || + (blk_start_addr == block_start(region_start_addr)), "invariant"); } else { region_start_addr = mr.start(); blk_start_addr = block_start(region_start_addr); @@ -705,6 +713,12 @@ object_iterate_from(bm, blk); } +// For a continguous space object_iterate() and safe_object_iterate() +// are the same. +void ContiguousSpace::safe_object_iterate(ObjectClosure* blk) { + object_iterate(blk); +} + void ContiguousSpace::object_iterate_from(WaterMark mark, ObjectClosure* blk) { assert(mark.space() == this, "Mark does not match space"); HeapWord* p = mark.point();
--- a/src/share/vm/memory/space.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/memory/space.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -193,6 +193,9 @@ // each. Objects allocated by applications of the closure are not // included in the iteration. virtual void object_iterate(ObjectClosure* blk) = 0; + // Similar to object_iterate() except only iterates over + // objects whose internal references point to objects in the space. + virtual void safe_object_iterate(ObjectClosure* blk) = 0; // Iterate over all objects that intersect with mr, calling "cl->do_object" // on each. There is an exception to this: if this closure has already @@ -843,6 +846,9 @@ void oop_iterate(OopClosure* cl); void oop_iterate(MemRegion mr, OopClosure* cl); void object_iterate(ObjectClosure* blk); + // For contiguous spaces this method will iterate safely over objects + // in the space (i.e., between bottom and top) when at a safepoint. + void safe_object_iterate(ObjectClosure* blk); void object_iterate_mem(MemRegion mr, UpwardsObjectClosure* cl); // iterates on objects up to the safe limit HeapWord* object_iterate_careful(ObjectClosureCareful* cl);
--- a/src/share/vm/oops/constMethodKlass.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/constMethodKlass.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -49,10 +49,16 @@ return constMethodOop(obj)->object_is_parsable(); } +bool constMethodKlass::oop_is_conc_safe(oop obj) const { + assert(obj->is_constMethod(), "must be constMethod oop"); + return constMethodOop(obj)->is_conc_safe(); +} + constMethodOop constMethodKlass::allocate(int byte_code_size, int compressed_line_number_size, int localvariable_table_length, int checked_exceptions_length, + bool is_conc_safe, TRAPS) { int size = constMethodOopDesc::object_size(byte_code_size, @@ -75,6 +81,7 @@ compressed_line_number_size, localvariable_table_length); assert(cm->size() == size, "wrong size for object"); + cm->set_is_conc_safe(is_conc_safe); cm->set_partially_loaded(); assert(cm->is_parsable(), "Is safely parsable by gc"); return cm;
--- a/src/share/vm/oops/constMethodKlass.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/constMethodKlass.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -32,12 +32,16 @@ // Testing bool oop_is_constMethod() const { return true; } virtual bool oop_is_parsable(oop obj) const; + virtual bool oop_is_conc_safe(oop obj) const; + // Allocation DEFINE_ALLOCATE_PERMANENT(constMethodKlass); constMethodOop allocate(int byte_code_size, int compressed_line_number_size, int localvariable_table_length, - int checked_exceptions_length, TRAPS); + int checked_exceptions_length, + bool is_conc_safe, + TRAPS); static klassOop create_klass(TRAPS); // Sizing
--- a/src/share/vm/oops/constMethodOop.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/constMethodOop.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -104,6 +104,7 @@ // loads and stores. This value may updated and read without a lock by // multiple threads, so is volatile. volatile uint64_t _fingerprint; + volatile bool _is_conc_safe; // if true, safe for concurrent GC processing public: oop* oop_block_beg() const { return adr_method(); } @@ -273,6 +274,8 @@ oop* adr_method() const { return (oop*)&_method; } oop* adr_stackmap_data() const { return (oop*)&_stackmap_data; } oop* adr_exception_table() const { return (oop*)&_exception_table; } + bool is_conc_safe() { return _is_conc_safe; } + void set_is_conc_safe(bool v) { _is_conc_safe = v; } // Unique id for the method static const u2 MAX_IDNUM;
--- a/src/share/vm/oops/constantPoolKlass.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/constantPoolKlass.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -25,7 +25,7 @@ # include "incls/_precompiled.incl" # include "incls/_constantPoolKlass.cpp.incl" -constantPoolOop constantPoolKlass::allocate(int length, TRAPS) { +constantPoolOop constantPoolKlass::allocate(int length, bool is_conc_safe, TRAPS) { int size = constantPoolOopDesc::object_size(length); KlassHandle klass (THREAD, as_klassOop()); constantPoolOop c = @@ -38,6 +38,9 @@ c->set_flags(0); // only set to non-zero if constant pool is merged by RedefineClasses c->set_orig_length(0); + // if constant pool may change during RedefineClasses, it is created + // unsafe for GC concurrent processing. + c->set_is_conc_safe(is_conc_safe); // all fields are initialized; needed for GC // initialize tag array @@ -207,6 +210,11 @@ return size; } +bool constantPoolKlass::oop_is_conc_safe(oop obj) const { + assert(obj->is_constantPool(), "must be constantPool"); + return constantPoolOop(obj)->is_conc_safe(); +} + #ifndef SERIALGC int constantPoolKlass::oop_update_pointers(ParCompactionManager* cm, oop obj) { assert (obj->is_constantPool(), "obj must be constant pool");
--- a/src/share/vm/oops/constantPoolKlass.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/constantPoolKlass.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -34,7 +34,7 @@ // Allocation DEFINE_ALLOCATE_PERMANENT(constantPoolKlass); - constantPoolOop allocate(int length, TRAPS); + constantPoolOop allocate(int length, bool is_conc_safe, TRAPS); static klassOop create_klass(TRAPS); // Casting from klassOop @@ -48,6 +48,8 @@ int object_size() const { return align_object_size(header_size()); } // Garbage collection + // Returns true is the object is safe for GC concurrent processing. + virtual bool oop_is_conc_safe(oop obj) const; void oop_follow_contents(oop obj); int oop_adjust_pointers(oop obj);
--- a/src/share/vm/oops/constantPoolOop.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/constantPoolOop.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -962,7 +962,7 @@ } case JVM_CONSTANT_Long: { u8 val = Bytes::get_Java_u8(bytes); - printf("long %lldl", *(jlong *) &val); + printf("long "INT64_FORMAT, *(jlong *) &val); ent_size = 8; idx++; // Long takes two cpool slots break;
--- a/src/share/vm/oops/constantPoolOop.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/constantPoolOop.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -43,6 +43,8 @@ klassOop _pool_holder; // the corresponding class int _flags; // a few header bits to describe contents for GC int _length; // number of elements in the array + volatile bool _is_conc_safe; // if true, safe for concurrent + // GC processing // only set to non-zero if constant pool is merged by RedefineClasses int _orig_length; @@ -379,6 +381,9 @@ static int object_size(int length) { return align_object_size(header_size() + length); } int object_size() { return object_size(length()); } + bool is_conc_safe() { return _is_conc_safe; } + void set_is_conc_safe(bool v) { _is_conc_safe = v; } + friend class constantPoolKlass; friend class ClassFileParser; friend class SystemDictionary;
--- a/src/share/vm/oops/cpCacheKlass.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/cpCacheKlass.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -32,13 +32,43 @@ } -constantPoolCacheOop constantPoolCacheKlass::allocate(int length, TRAPS) { +constantPoolCacheOop constantPoolCacheKlass::allocate(int length, + bool is_conc_safe, + TRAPS) { // allocate memory int size = constantPoolCacheOopDesc::object_size(length); + KlassHandle klass (THREAD, as_klassOop()); - constantPoolCacheOop cache = (constantPoolCacheOop) - CollectedHeap::permanent_obj_allocate(klass, size, CHECK_NULL); + + // This is the original code. The code from permanent_obj_allocate() + // was in-lined to allow the setting of is_conc_safe before the klass + // is installed. + // constantPoolCacheOop cache = (constantPoolCacheOop) + // CollectedHeap::permanent_obj_allocate(klass, size, CHECK_NULL); + + oop obj = CollectedHeap::permanent_obj_allocate_no_klass_install(klass, size, CHECK_NULL); + constantPoolCacheOop cache = (constantPoolCacheOop) obj; + cache->set_is_conc_safe(is_conc_safe); + // The store to is_conc_safe must be visible before the klass + // is set. This should be done safely because _is_conc_safe has + // been declared volatile. If there are any problems, consider adding + // OrderAccess::storestore(); + CollectedHeap::post_allocation_install_obj_klass(klass, obj, size); + NOT_PRODUCT(Universe::heap()->check_for_bad_heap_word_value((HeapWord*) obj, + size)); + + // The length field affects the size of the object. The allocation + // above allocates the correct size (see calculation of "size") but + // the size() method of the constant pool cache oop will not reflect + // that size until the correct length is set. cache->set_length(length); + + // The store of the length must be visible before is_conc_safe is + // set to a safe state. + // This should be done safely because _is_conc_safe has + // been declared volatile. If there are any problems, consider adding + // OrderAccess::storestore(); + cache->set_is_conc_safe(methodOopDesc::IsSafeConc); cache->set_constant_pool(NULL); return cache; } @@ -114,7 +144,6 @@ return size; } - int constantPoolCacheKlass::oop_adjust_pointers(oop obj) { assert(obj->is_constantPoolCache(), "obj must be constant pool cache"); constantPoolCacheOop cache = (constantPoolCacheOop)obj; @@ -131,6 +160,11 @@ return size; } +bool constantPoolCacheKlass::oop_is_conc_safe(oop obj) const { + assert(obj->is_constantPoolCache(), "should be constant pool"); + return constantPoolCacheOop(obj)->is_conc_safe(); +} + #ifndef SERIALGC void constantPoolCacheKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) {
--- a/src/share/vm/oops/cpCacheKlass.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/cpCacheKlass.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -32,7 +32,7 @@ // Allocation DEFINE_ALLOCATE_PERMANENT(constantPoolCacheKlass); - constantPoolCacheOop allocate(int length, TRAPS); + constantPoolCacheOop allocate(int length, bool is_conc_safe, TRAPS); static klassOop create_klass(TRAPS); // Casting from klassOop @@ -48,6 +48,7 @@ // Garbage collection void oop_follow_contents(oop obj); int oop_adjust_pointers(oop obj); + virtual bool oop_is_conc_safe(oop obj) const; // Parallel Scavenge and Parallel Old PARALLEL_GC_DECLS
--- a/src/share/vm/oops/cpCacheOop.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/cpCacheOop.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -291,6 +291,9 @@ private: int _length; constantPoolOop _constant_pool; // the corresponding constant pool + // If true, safe for concurrent GC processing, + // Set unconditionally in constantPoolCacheKlass::allocate() + volatile bool _is_conc_safe; // Sizing debug_only(friend class ClassVerifier;) @@ -316,6 +319,12 @@ constantPoolOop constant_pool() const { return _constant_pool; } ConstantPoolCacheEntry* entry_at(int i) const { assert(0 <= i && i < length(), "index out of bounds"); return base() + i; } + // GC support + // If the _length field has not been set, the size of the + // constantPoolCache cannot be correctly calculated. + bool is_conc_safe() { return _is_conc_safe; } + void set_is_conc_safe(bool v) { _is_conc_safe = v; } + // Code generation static ByteSize base_offset() { return in_ByteSize(sizeof(constantPoolCacheOopDesc)); }
--- a/src/share/vm/oops/klass.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/klass.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -606,8 +606,19 @@ #undef assert_same_query // Unless overridden, oop is parsable if it has a klass pointer. + // Parsability of an object is object specific. virtual bool oop_is_parsable(oop obj) const { return true; } + // Unless overridden, oop is safe for concurrent GC processing + // after its allocation is complete. The exception to + // this is the case where objects are changed after allocation. + // Class redefinition is one of the known exceptions. During + // class redefinition, an allocated class can changed in order + // order to create a merged class (the combiniation of the + // old class definition that has to be perserved and the new class + // definition which is being created. + virtual bool oop_is_conc_safe(oop obj) const { return true; } + // Access flags AccessFlags access_flags() const { return _access_flags; } void set_access_flags(AccessFlags flags) { _access_flags = flags; }
--- a/src/share/vm/oops/methodOop.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/methodOop.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -792,15 +792,34 @@ AccessFlags flags = m->access_flags(); int checked_exceptions_len = m->checked_exceptions_length(); int localvariable_len = m->localvariable_table_length(); - methodOop newm_oop = oopFactory::new_method(new_code_length, flags, new_compressed_linenumber_size, localvariable_len, checked_exceptions_len, CHECK_(methodHandle())); + // Allocate newm_oop with the is_conc_safe parameter set + // to IsUnsafeConc to indicate that newm_oop is not yet + // safe for concurrent processing by a GC. + methodOop newm_oop = oopFactory::new_method(new_code_length, + flags, + new_compressed_linenumber_size, + localvariable_len, + checked_exceptions_len, + IsUnsafeConc, + CHECK_(methodHandle())); methodHandle newm (THREAD, newm_oop); int new_method_size = newm->method_size(); // Create a shallow copy of methodOopDesc part, but be careful to preserve the new constMethodOop constMethodOop newcm = newm->constMethod(); int new_const_method_size = newm->constMethod()->object_size(); + memcpy(newm(), m(), sizeof(methodOopDesc)); // Create shallow copy of constMethodOopDesc, but be careful to preserve the methodOop + // is_conc_safe is set to false because that is the value of + // is_conc_safe initialzied into newcm and the copy should + // not overwrite that value. During the window during which it is + // tagged as unsafe, some extra work could be needed during precleaning + // or concurrent marking but those phases will be correct. Setting and + // resetting is done in preference to a careful copying into newcm to + // avoid having to know the precise layout of a constMethodOop. + m->constMethod()->set_is_conc_safe(false); memcpy(newcm, m->constMethod(), sizeof(constMethodOopDesc)); + m->constMethod()->set_is_conc_safe(true); // Reset correct method/const method, method size, and parameter info newcm->set_method(newm()); newm->set_constMethod(newcm); @@ -831,6 +850,10 @@ m->localvariable_table_start(), localvariable_len * sizeof(LocalVariableTableElement)); } + + // Only set is_conc_safe to true when changes to newcm are + // complete. + newcm->set_is_conc_safe(true); return newm; }
--- a/src/share/vm/oops/methodOop.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/methodOop.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -129,6 +129,10 @@ volatile address _from_interpreted_entry; // Cache of _code ? _adapter->i2c_entry() : _i2i_entry public: + + static const bool IsUnsafeConc = false; + static const bool IsSafeConc = true; + // accessors for instance variables constMethodOop constMethod() const { return _constMethod; } void set_constMethod(constMethodOop xconst) { oop_store_without_check((oop*)&_constMethod, (oop)xconst); }
--- a/src/share/vm/oops/oop.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/oop.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -108,6 +108,13 @@ // installation of their klass pointer. bool is_parsable(); + // Some perm gen objects that have been allocated and initialized + // can be changed by the VM when not at a safe point (class rededfinition + // is an example). Such objects should not be examined by the + // concurrent processing of a garbage collector if is_conc_safe() + // returns false. + bool is_conc_safe(); + // type test operations (inlined in oop.inline.h) bool is_instance() const; bool is_instanceRef() const;
--- a/src/share/vm/oops/oop.inline.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/oop.inline.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -435,6 +435,10 @@ return blueprint()->oop_is_parsable(this); } +inline bool oopDesc::is_conc_safe() { + return blueprint()->oop_is_conc_safe(this); +} + inline void update_barrier_set(void* p, oop v) { assert(oopDesc::bs() != NULL, "Uninitialized bs in oop!"); oopDesc::bs()->write_ref_field(p, v);
--- a/src/share/vm/oops/oopsHierarchy.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/oops/oopsHierarchy.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -126,8 +126,11 @@ operator jobject () const { return (jobject)obj(); } // from javaClasses.cpp operator JavaThread* () const { return (JavaThread*)obj(); } + +#ifndef _LP64 // from jvm.cpp operator jlong* () const { return (jlong*)obj(); } +#endif // from parNewGeneration and other things that want to get to the end of // an oop for stuff (like constMethodKlass.cpp, objArrayKlass.cpp)
--- a/src/share/vm/opto/block.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/block.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -880,6 +880,7 @@ } void PhaseCFG::verify( ) const { +#ifdef ASSERT // Verify sane CFG for( uint i = 0; i < _num_blocks; i++ ) { Block *b = _blocks[i]; @@ -894,10 +895,20 @@ "CreateEx must be first instruction in block" ); } for( uint k = 0; k < n->req(); k++ ) { - Node *use = n->in(k); - if( use && use != n ) { - assert( _bbs[use->_idx] || use->is_Con(), + Node *def = n->in(k); + if( def && def != n ) { + assert( _bbs[def->_idx] || def->is_Con(), "must have block; constants for debug info ok" ); + // Verify that instructions in the block is in correct order. + // Uses must follow their definition if they are at the same block. + // Mostly done to check that MachSpillCopy nodes are placed correctly + // when CreateEx node is moved in build_ifg_physical(). + if( _bbs[def->_idx] == b && + !(b->head()->is_Loop() && n->is_Phi()) && + // See (+++) comment in reg_split.cpp + !(n->jvms() != NULL && n->jvms()->is_monitor_use(k)) ) { + assert( b->find_node(def) < j, "uses must follow definitions" ); + } } } } @@ -914,6 +925,7 @@ assert( b->_num_succs == 2, "Conditional branch must have two targets"); } } +#endif } #endif
--- a/src/share/vm/opto/c2_globals.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/c2_globals.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -191,6 +191,9 @@ notproduct(bool, VerifyHashTableKeys, true, \ "Verify the immutability of keys in the VN hash tables") \ \ + notproduct(bool, VerifyRegisterAllocator , false, \ + "Verify Register Allocator") \ + \ develop_pd(intx, FLOATPRESSURE, \ "Number of float LRG's that constitute high register pressure") \ \
--- a/src/share/vm/opto/cfgnode.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/cfgnode.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -858,12 +858,18 @@ // convert the one to the other. const TypePtr* ttp = _type->make_ptr(); const TypeInstPtr* ttip = (ttp != NULL) ? ttp->isa_instptr() : NULL; + const TypeKlassPtr* ttkp = (ttp != NULL) ? ttp->isa_klassptr() : NULL; bool is_intf = false; if (ttip != NULL) { ciKlass* k = ttip->klass(); if (k->is_loaded() && k->is_interface()) is_intf = true; } + if (ttkp != NULL) { + ciKlass* k = ttkp->klass(); + if (k->is_loaded() && k->is_interface()) + is_intf = true; + } // Default case: merge all inputs const Type *t = Type::TOP; // Merged type starting value @@ -921,6 +927,8 @@ // uplift the type. if( !t->empty() && ttip && ttip->is_loaded() && ttip->klass()->is_interface() ) { assert(ft == _type, ""); } // Uplift to interface + else if( !t->empty() && ttkp && ttkp->is_loaded() && ttkp->klass()->is_interface() ) + { assert(ft == _type, ""); } // Uplift to interface // Otherwise it's something stupid like non-overlapping int ranges // found on dying counted loops. else @@ -936,6 +944,7 @@ // because the type system doesn't interact well with interfaces. const TypePtr *jtp = jt->make_ptr(); const TypeInstPtr *jtip = (jtp != NULL) ? jtp->isa_instptr() : NULL; + const TypeKlassPtr *jtkp = (jtp != NULL) ? jtp->isa_klassptr() : NULL; if( jtip && ttip ) { if( jtip->is_loaded() && jtip->klass()->is_interface() && ttip->is_loaded() && !ttip->klass()->is_interface() ) { @@ -945,6 +954,14 @@ jt = ft; } } + if( jtkp && ttkp ) { + if( jtkp->is_loaded() && jtkp->klass()->is_interface() && + ttkp->is_loaded() && !ttkp->klass()->is_interface() ) { + assert(ft == ttkp->cast_to_ptr_type(jtkp->ptr()) || + ft->isa_narrowoop() && ft->make_ptr() == ttkp->cast_to_ptr_type(jtkp->ptr()), ""); + jt = ft; + } + } if (jt != ft && jt->base() == ft->base()) { if (jt->isa_int() && jt->is_int()->_lo == ft->is_int()->_lo &&
--- a/src/share/vm/opto/chaitin.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/chaitin.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -228,6 +228,11 @@ // them for real. de_ssa(); +#ifdef ASSERT + // Veify the graph before RA. + verify(&live_arena); +#endif + { NOT_PRODUCT( Compile::TracePhase t3("computeLive", &_t_computeLive, TimeCompiler); ) _live = NULL; // Mark live as being not available @@ -306,12 +311,6 @@ C->check_node_count(2*NodeLimitFudgeFactor, "out of nodes after physical split"); if (C->failing()) return; -#ifdef ASSERT - if( VerifyOpto ) { - _cfg.verify(); - verify_base_ptrs(&live_arena); - } -#endif NOT_PRODUCT( C->verify_graph_edges(); ) compact(); // Compact LRGs; return new lower max lrg @@ -340,7 +339,7 @@ compress_uf_map_for_nodes(); #ifdef ASSERT - if( VerifyOpto ) _ifg->verify(this); + verify(&live_arena, true); #endif } else { ifg.SquareUp(); @@ -376,12 +375,6 @@ // Bail out if unique gets too large (ie - unique > MaxNodeLimit - 2*NodeLimitFudgeFactor) C->check_node_count(2*NodeLimitFudgeFactor, "out of nodes after split"); if (C->failing()) return; -#ifdef ASSERT - if( VerifyOpto ) { - _cfg.verify(); - verify_base_ptrs(&live_arena); - } -#endif compact(); // Compact LRGs; return new lower max lrg @@ -412,7 +405,7 @@ } compress_uf_map_for_nodes(); #ifdef ASSERT - if( VerifyOpto ) _ifg->verify(this); + verify(&live_arena, true); #endif cache_lrg_info(); // Count degree of LRGs @@ -432,6 +425,11 @@ // Peephole remove copies post_allocate_copy_removal(); +#ifdef ASSERT + // Veify the graph after RA. + verify(&live_arena); +#endif + // max_reg is past the largest *register* used. // Convert that to a frame_slot number. if( _max_reg <= _matcher._new_SP ) @@ -956,7 +954,7 @@ while ((neighbor = elements.next()) != 0) { LRG *n = &lrgs(neighbor); #ifdef ASSERT - if( VerifyOpto ) { + if( VerifyOpto || VerifyRegisterAllocator ) { assert( _ifg->effective_degree(neighbor) == n->degree(), "" ); } #endif
--- a/src/share/vm/opto/chaitin.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/chaitin.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -491,6 +491,8 @@ // Verify that base pointers and derived pointers are still sane void verify_base_ptrs( ResourceArea *a ) const; + void verify( ResourceArea *a, bool verify_ifg = false ) const; + void dump_for_spill_split_recycle() const; public:
--- a/src/share/vm/opto/classes.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/classes.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -129,7 +129,7 @@ macro(LShiftI) macro(LShiftL) macro(LoadB) -macro(LoadC) +macro(LoadUS) macro(LoadD) macro(LoadD_unaligned) macro(LoadF)
--- a/src/share/vm/opto/compile.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/compile.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -2005,7 +2005,7 @@ case Op_StoreP: case Op_StoreN: case Op_LoadB: - case Op_LoadC: + case Op_LoadUS: case Op_LoadI: case Op_LoadKlass: case Op_LoadNKlass:
--- a/src/share/vm/opto/divnode.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/divnode.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2009 Sun Microsystems, Inc. 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 @@ -244,42 +244,73 @@ //---------------------long_by_long_mulhi-------------------------------------- // Generate ideal node graph for upper half of a 64 bit x 64 bit multiplication -static Node *long_by_long_mulhi( PhaseGVN *phase, Node *dividend, jlong magic_const) { +static Node* long_by_long_mulhi(PhaseGVN* phase, Node* dividend, jlong magic_const) { // If the architecture supports a 64x64 mulhi, there is // no need to synthesize it in ideal nodes. if (Matcher::has_match_rule(Op_MulHiL)) { - Node *v = phase->longcon(magic_const); + Node* v = phase->longcon(magic_const); return new (phase->C, 3) MulHiLNode(dividend, v); } + // Taken from Hacker's Delight, Fig. 8-2. Multiply high signed. + // (http://www.hackersdelight.org/HDcode/mulhs.c) + // + // int mulhs(int u, int v) { + // unsigned u0, v0, w0; + // int u1, v1, w1, w2, t; + // + // u0 = u & 0xFFFF; u1 = u >> 16; + // v0 = v & 0xFFFF; v1 = v >> 16; + // w0 = u0*v0; + // t = u1*v0 + (w0 >> 16); + // w1 = t & 0xFFFF; + // w2 = t >> 16; + // w1 = u0*v1 + w1; + // return u1*v1 + w2 + (w1 >> 16); + // } + // + // Note: The version above is for 32x32 multiplications, while the + // following inline comments are adapted to 64x64. + const int N = 64; - Node *u_hi = phase->transform(new (phase->C, 3) RShiftLNode(dividend, phase->intcon(N / 2))); - Node *u_lo = phase->transform(new (phase->C, 3) AndLNode(dividend, phase->longcon(0xFFFFFFFF))); + // u0 = u & 0xFFFFFFFF; u1 = u >> 32; + Node* u0 = phase->transform(new (phase->C, 3) AndLNode(dividend, phase->longcon(0xFFFFFFFF))); + Node* u1 = phase->transform(new (phase->C, 3) RShiftLNode(dividend, phase->intcon(N / 2))); + + // v0 = v & 0xFFFFFFFF; v1 = v >> 32; + Node* v0 = phase->longcon(magic_const & 0xFFFFFFFF); + Node* v1 = phase->longcon(magic_const >> (N / 2)); - Node *v_hi = phase->longcon(magic_const >> N/2); - Node *v_lo = phase->longcon(magic_const & 0XFFFFFFFF); + // w0 = u0*v0; + Node* w0 = phase->transform(new (phase->C, 3) MulLNode(u0, v0)); - Node *hihi_product = phase->transform(new (phase->C, 3) MulLNode(u_hi, v_hi)); - Node *hilo_product = phase->transform(new (phase->C, 3) MulLNode(u_hi, v_lo)); - Node *lohi_product = phase->transform(new (phase->C, 3) MulLNode(u_lo, v_hi)); - Node *lolo_product = phase->transform(new (phase->C, 3) MulLNode(u_lo, v_lo)); + // t = u1*v0 + (w0 >> 32); + Node* u1v0 = phase->transform(new (phase->C, 3) MulLNode(u1, v0)); + Node* temp = phase->transform(new (phase->C, 3) URShiftLNode(w0, phase->intcon(N / 2))); + Node* t = phase->transform(new (phase->C, 3) AddLNode(u1v0, temp)); + + // w1 = t & 0xFFFFFFFF; + Node* w1 = new (phase->C, 3) AndLNode(t, phase->longcon(0xFFFFFFFF)); - Node *t1 = phase->transform(new (phase->C, 3) URShiftLNode(lolo_product, phase->intcon(N / 2))); - Node *t2 = phase->transform(new (phase->C, 3) AddLNode(hilo_product, t1)); + // w2 = t >> 32; + Node* w2 = new (phase->C, 3) RShiftLNode(t, phase->intcon(N / 2)); + + // 6732154: Construct both w1 and w2 before transforming, so t + // doesn't go dead prematurely. + w1 = phase->transform(w1); + w2 = phase->transform(w2); - // Construct both t3 and t4 before transforming so t2 doesn't go dead - // prematurely. - Node *t3 = new (phase->C, 3) RShiftLNode(t2, phase->intcon(N / 2)); - Node *t4 = new (phase->C, 3) AndLNode(t2, phase->longcon(0xFFFFFFFF)); - t3 = phase->transform(t3); - t4 = phase->transform(t4); + // w1 = u0*v1 + w1; + Node* u0v1 = phase->transform(new (phase->C, 3) MulLNode(u0, v1)); + w1 = phase->transform(new (phase->C, 3) AddLNode(u0v1, w1)); - Node *t5 = phase->transform(new (phase->C, 3) AddLNode(t4, lohi_product)); - Node *t6 = phase->transform(new (phase->C, 3) RShiftLNode(t5, phase->intcon(N / 2))); - Node *t7 = phase->transform(new (phase->C, 3) AddLNode(t3, hihi_product)); + // return u1*v1 + w2 + (w1 >> 32); + Node* u1v1 = phase->transform(new (phase->C, 3) MulLNode(u1, v1)); + Node* temp1 = phase->transform(new (phase->C, 3) AddLNode(u1v1, w2)); + Node* temp2 = phase->transform(new (phase->C, 3) RShiftLNode(w1, phase->intcon(N / 2))); - return new (phase->C, 3) AddLNode(t7, t6); + return new (phase->C, 3) AddLNode(temp1, temp2); } @@ -976,7 +1007,7 @@ // Expand mod if( con >= 0 && con < max_jlong && is_power_of_2_long(con+1) ) { - uint k = log2_long(con); // Extract k + uint k = exact_log2_long(con+1); // Extract k // Basic algorithm by David Detlefs. See fastmod_long.java for gory details. // Used to help a popular random number generator which does a long-mod
--- a/src/share/vm/opto/gcm.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/gcm.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -29,6 +29,9 @@ #include "incls/_precompiled.incl" #include "incls/_gcm.cpp.incl" +// To avoid float value underflow +#define MIN_BLOCK_FREQUENCY 1.e-35f + //----------------------------schedule_node_into_block------------------------- // Insert node n into block b. Look for projections of n and make sure they // are in b also. @@ -1380,6 +1383,13 @@ } } +#ifdef ASSERT + for (uint i = 0; i < _num_blocks; i++ ) { + Block *b = _blocks[i]; + assert(b->_freq >= MIN_BLOCK_FREQUENCY, "Register Allocator requiers meaningful block frequency"); + } +#endif + #ifndef PRODUCT if (PrintCFGBlockFreq) { tty->print_cr("CFG Block Frequencies"); @@ -1877,7 +1887,9 @@ float loop_freq = _freq * trip_count(); for (int i = 0; i < _members.length(); i++) { CFGElement* s = _members.at(i); - s->_freq *= loop_freq; + float block_freq = s->_freq * loop_freq; + if (block_freq < MIN_BLOCK_FREQUENCY) block_freq = MIN_BLOCK_FREQUENCY; + s->_freq = block_freq; } CFGLoop* ch = _child; while (ch != NULL) {
--- a/src/share/vm/opto/graphKit.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/graphKit.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -1836,10 +1836,7 @@ (CardTableModRefBS*)(Universe::heap()->barrier_set()); Node *b = _gvn.transform(new (C, 3) URShiftXNode( cast, _gvn.intcon(CardTableModRefBS::card_shift) )); // We store into a byte array, so do not bother to left-shift by zero - // Get base of card map - assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), - "adjust this code"); - Node *c = makecon(TypeRawPtr::make((address)ct->byte_map_base)); + Node *c = byte_map_base_node(); // Combine Node *sb_ctl = control(); Node *sb_adr = _gvn.transform(new (C, 4) AddPNode( top()/*no base ptr*/, c, b )); @@ -2945,16 +2942,10 @@ // Now generate allocation code - // With escape analysis, the entire memory state is needed to be able to - // eliminate the allocation. If the allocations cannot be eliminated, this - // will be optimized to the raw slice when the allocation is expanded. - Node *mem; - if (C->do_escape_analysis()) { - mem = reset_memory(); - set_all_memory(mem); - } else { - mem = memory(Compile::AliasIdxRaw); - } + // The entire memory state is needed for slow path of the allocation + // since GC and deoptimization can happened. + Node *mem = reset_memory(); + set_all_memory(mem); // Create new memory state AllocateNode* alloc = new (C, AllocateNode::ParmLimit) @@ -3091,16 +3082,10 @@ // Now generate allocation code - // With escape analysis, the entire memory state is needed to be able to - // eliminate the allocation. If the allocations cannot be eliminated, this - // will be optimized to the raw slice when the allocation is expanded. - Node *mem; - if (C->do_escape_analysis()) { - mem = reset_memory(); - set_all_memory(mem); - } else { - mem = memory(Compile::AliasIdxRaw); - } + // The entire memory state is needed for slow path of the allocation + // since GC and deoptimization can happened. + Node *mem = reset_memory(); + set_all_memory(mem); // Create new memory state // Create the AllocateArrayNode and its result projections AllocateArrayNode* alloc @@ -3233,12 +3218,11 @@ // Now some of the values - Node* marking = __ load(no_ctrl, marking_adr, TypeInt::INT, active_type, Compile::AliasIdxRaw); - Node* index = __ load(no_ctrl, index_adr, TypeInt::INT, T_INT, Compile::AliasIdxRaw); - Node* buffer = __ load(no_ctrl, buffer_adr, TypeRawPtr::NOTNULL, T_ADDRESS, Compile::AliasIdxRaw); + Node* marking = __ load(__ ctrl(), marking_adr, TypeInt::INT, active_type, Compile::AliasIdxRaw); // if (!marking) __ if_then(marking, BoolTest::ne, zero); { + Node* index = __ load(__ ctrl(), index_adr, TypeInt::INT, T_INT, Compile::AliasIdxRaw); const Type* t1 = adr->bottom_type(); const Type* t2 = val->bottom_type(); @@ -3246,6 +3230,7 @@ Node* orig = __ load(no_ctrl, adr, val_type, bt, alias_idx); // if (orig != NULL) __ if_then(orig, BoolTest::ne, null()); { + Node* buffer = __ load(__ ctrl(), buffer_adr, TypeRawPtr::NOTNULL, T_ADDRESS, Compile::AliasIdxRaw); // load original value // alias_idx correct?? @@ -3365,14 +3350,6 @@ const TypeFunc *tf = OptoRuntime::g1_wb_post_Type(); - // Get the address of the card table - CardTableModRefBS* ct = - (CardTableModRefBS*)(Universe::heap()->barrier_set()); - Node *card_table = __ makecon(TypeRawPtr::make((address)ct->byte_map_base)); - // Get base of card map - assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); - - // Offsets into the thread const int index_offset = in_bytes(JavaThread::dirty_card_queue_offset() + PtrQueue::byte_offset_of_index()); @@ -3402,7 +3379,7 @@ Node* card_offset = __ URShiftX( cast, __ ConI(CardTableModRefBS::card_shift) ); // Combine card table base and card offset - Node *card_adr = __ AddP(no_base, card_table, card_offset ); + Node *card_adr = __ AddP(no_base, byte_map_base_node(), card_offset ); // If we know the value being stored does it cross regions?
--- a/src/share/vm/opto/graphKit.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/graphKit.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -83,6 +83,18 @@ Node* zerocon(BasicType bt) const { return _gvn.zerocon(bt); } // (See also macro MakeConX in type.hpp, which uses intcon or longcon.) + // Helper for byte_map_base + Node* byte_map_base_node() { + // Get base of card map + CardTableModRefBS* ct = (CardTableModRefBS*)(Universe::heap()->barrier_set()); + assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust users of this code"); + if (ct->byte_map_base != NULL) { + return makecon(TypeRawPtr::make((address)ct->byte_map_base)); + } else { + return null(); + } + } + jint find_int_con(Node* n, jint value_if_unknown) { return _gvn.find_int_con(n, value_if_unknown); }
--- a/src/share/vm/opto/idealGraphPrinter.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/idealGraphPrinter.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -557,7 +557,7 @@ // max. 2 chars allowed if (value >= -9 && value <= 99) { - sprintf(buffer, "%d", value); + sprintf(buffer, INT64_FORMAT, value); print_prop(short_name, buffer); } else { print_prop(short_name, "L");
--- a/src/share/vm/opto/ifg.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/ifg.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -471,12 +471,28 @@ // for the "collect_gc_info" phase later. IndexSet liveout(_live->live(b)); uint last_inst = b->end_idx(); - // Compute last phi index - uint last_phi; - for( last_phi = 1; last_phi < last_inst; last_phi++ ) - if( !b->_nodes[last_phi]->is_Phi() ) + // Compute first nonphi node index + uint first_inst; + for( first_inst = 1; first_inst < last_inst; first_inst++ ) + if( !b->_nodes[first_inst]->is_Phi() ) break; + // Spills could be inserted before CreateEx node which should be + // first instruction in block after Phis. Move CreateEx up. + for( uint insidx = first_inst; insidx < last_inst; insidx++ ) { + Node *ex = b->_nodes[insidx]; + if( ex->is_SpillCopy() ) continue; + if( insidx > first_inst && ex->is_Mach() && + ex->as_Mach()->ideal_Opcode() == Op_CreateEx ) { + // If the CreateEx isn't above all the MachSpillCopies + // then move it to the top. + b->_nodes.remove(insidx); + b->_nodes.insert(first_inst, ex); + } + // Stop once a CreateEx or any other node is found + break; + } + // Reset block's register pressure values for each ifg construction uint pressure[2], hrp_index[2]; pressure[0] = pressure[1] = 0; @@ -485,7 +501,7 @@ // Liveout things are presumed live for the whole block. We accumulate // 'area' accordingly. If they get killed in the block, we'll subtract // the unused part of the block from the area. - int inst_count = last_inst - last_phi; + int inst_count = last_inst - first_inst; double cost = (inst_count <= 0) ? 0.0 : b->_freq * double(inst_count); assert(!(cost < 0.0), "negative spill cost" ); IndexSetIterator elements(&liveout);
--- a/src/share/vm/opto/lcm.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/lcm.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -107,7 +107,7 @@ was_store = false; switch( mach->ideal_Opcode() ) { case Op_LoadB: - case Op_LoadC: + case Op_LoadUS: case Op_LoadD: case Op_LoadF: case Op_LoadI:
--- a/src/share/vm/opto/live.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/live.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -271,9 +271,9 @@ //------------------------------verify_base_ptrs------------------------------- // Verify that base pointers and derived pointers are still sane. -// Basically, if a derived pointer is live at a safepoint, then its -// base pointer must be live also. void PhaseChaitin::verify_base_ptrs( ResourceArea *a ) const { +#ifdef ASSERT + Unique_Node_List worklist(a); for( uint i = 0; i < _cfg._num_blocks; i++ ) { Block *b = _cfg._blocks[i]; for( uint j = b->end_idx() + 1; j > 1; j-- ) { @@ -287,28 +287,81 @@ // Now scan for a live derived pointer if (jvms->oopoff() < sfpt->req()) { // Check each derived/base pair - for (uint idx = jvms->oopoff(); idx < sfpt->req(); idx += 2) { + for (uint idx = jvms->oopoff(); idx < sfpt->req(); idx++) { Node *check = sfpt->in(idx); - uint j = 0; + bool is_derived = ((idx - jvms->oopoff()) & 1) == 0; // search upwards through spills and spill phis for AddP - while(true) { - if( !check ) break; - int idx = check->is_Copy(); - if( idx ) { - check = check->in(idx); - } else if( check->is_Phi() && check->_idx >= _oldphi ) { - check = check->in(1); - } else - break; - j++; - assert(j < 100000,"Derived pointer checking in infinite loop"); + worklist.clear(); + worklist.push(check); + uint k = 0; + while( k < worklist.size() ) { + check = worklist.at(k); + assert(check,"Bad base or derived pointer"); + // See PhaseChaitin::find_base_for_derived() for all cases. + int isc = check->is_Copy(); + if( isc ) { + worklist.push(check->in(isc)); + } else if( check->is_Phi() ) { + for (uint m = 1; m < check->req(); m++) + worklist.push(check->in(m)); + } else if( check->is_Con() ) { + if (is_derived) { + // Derived is NULL+offset + assert(!is_derived || check->bottom_type()->is_ptr()->ptr() == TypePtr::Null,"Bad derived pointer"); + } else { + assert(check->bottom_type()->is_ptr()->_offset == 0,"Bad base pointer"); + // Base either ConP(NULL) or loadConP + if (check->is_Mach()) { + assert(check->as_Mach()->ideal_Opcode() == Op_ConP,"Bad base pointer"); + } else { + assert(check->Opcode() == Op_ConP && + check->bottom_type()->is_ptr()->ptr() == TypePtr::Null,"Bad base pointer"); + } + } + } else if( check->bottom_type()->is_ptr()->_offset == 0 ) { + if(check->is_Proj() || check->is_Mach() && + (check->as_Mach()->ideal_Opcode() == Op_CreateEx || + check->as_Mach()->ideal_Opcode() == Op_ThreadLocal || + check->as_Mach()->ideal_Opcode() == Op_CMoveP || + check->as_Mach()->ideal_Opcode() == Op_CheckCastPP || +#ifdef _LP64 + UseCompressedOops && check->as_Mach()->ideal_Opcode() == Op_CastPP || + UseCompressedOops && check->as_Mach()->ideal_Opcode() == Op_DecodeN || +#endif + check->as_Mach()->ideal_Opcode() == Op_LoadP || + check->as_Mach()->ideal_Opcode() == Op_LoadKlass)) { + // Valid nodes + } else { + check->dump(); + assert(false,"Bad base or derived pointer"); + } + } else { + assert(is_derived,"Bad base pointer"); + assert(check->is_Mach() && check->as_Mach()->ideal_Opcode() == Op_AddP,"Bad derived pointer"); + } + k++; + assert(k < 100000,"Derived pointer checking in infinite loop"); } // End while - assert(check->is_Mach() && check->as_Mach()->ideal_Opcode() == Op_AddP,"Bad derived pointer") } } // End of check for derived pointers } // End of Kcheck for debug info } // End of if found a safepoint } // End of forall instructions in block } // End of forall blocks +#endif } + +//------------------------------verify------------------------------------- +// Verify that graphs and base pointers are still sane. +void PhaseChaitin::verify( ResourceArea *a, bool verify_ifg ) const { +#ifdef ASSERT + if( VerifyOpto || VerifyRegisterAllocator ) { + _cfg.verify(); + verify_base_ptrs(a); + if(verify_ifg) + _ifg->verify(this); + } #endif +} + +#endif
--- a/src/share/vm/opto/loopnode.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/loopnode.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -2654,7 +2654,7 @@ case Op_ModF: case Op_ModD: case Op_LoadB: // Same with Loads; they can sink - case Op_LoadC: // during loop optimizations. + case Op_LoadUS: // during loop optimizations. case Op_LoadD: case Op_LoadF: case Op_LoadI:
--- a/src/share/vm/opto/macro.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/macro.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -952,13 +952,6 @@ Node* klass_node = alloc->in(AllocateNode::KlassNode); Node* initial_slow_test = alloc->in(AllocateNode::InitialTest); - // With escape analysis, the entire memory state was needed to be able to - // eliminate the allocation. Since the allocations cannot be eliminated, - // optimize it to the raw slice. - if (mem->is_MergeMem()) { - mem = mem->as_MergeMem()->memory_at(Compile::AliasIdxRaw); - } - assert(ctrl != NULL, "must have control"); // We need a Region and corresponding Phi's to merge the slow-path and fast-path results. // they will not be used if "always_slow" is set @@ -1016,6 +1009,11 @@ Node *slow_mem = mem; // save the current memory state for slow path // generate the fast allocation code unless we know that the initial test will always go slow if (!always_slow) { + // Fast path modifies only raw memory. + if (mem->is_MergeMem()) { + mem = mem->as_MergeMem()->memory_at(Compile::AliasIdxRaw); + } + Node* eden_top_adr; Node* eden_end_adr; @@ -1239,8 +1237,6 @@ } } - mem = result_phi_rawmem; - // An allocate node has separate i_o projections for the uses on the control and i_o paths // Replace uses of the control i_o projection with result_phi_i_o (unless we are only generating a slow call) if (_ioproj_fallthrough == NULL) {
--- a/src/share/vm/opto/matcher.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/matcher.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -1824,7 +1824,7 @@ mem_op = true; break; case Op_LoadB: - case Op_LoadC: + case Op_LoadUS: case Op_LoadD: case Op_LoadF: case Op_LoadI:
--- a/src/share/vm/opto/memnode.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/memnode.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -779,14 +779,14 @@ "use LoadRangeNode instead"); switch (bt) { case T_BOOLEAN: - case T_BYTE: return new (C, 3) LoadBNode(ctl, mem, adr, adr_type, rt->is_int() ); - case T_INT: return new (C, 3) LoadINode(ctl, mem, adr, adr_type, rt->is_int() ); - case T_CHAR: return new (C, 3) LoadCNode(ctl, mem, adr, adr_type, rt->is_int() ); - case T_SHORT: return new (C, 3) LoadSNode(ctl, mem, adr, adr_type, rt->is_int() ); - case T_LONG: return new (C, 3) LoadLNode(ctl, mem, adr, adr_type, rt->is_long() ); - case T_FLOAT: return new (C, 3) LoadFNode(ctl, mem, adr, adr_type, rt ); - case T_DOUBLE: return new (C, 3) LoadDNode(ctl, mem, adr, adr_type, rt ); - case T_ADDRESS: return new (C, 3) LoadPNode(ctl, mem, adr, adr_type, rt->is_ptr() ); + case T_BYTE: return new (C, 3) LoadBNode (ctl, mem, adr, adr_type, rt->is_int() ); + case T_INT: return new (C, 3) LoadINode (ctl, mem, adr, adr_type, rt->is_int() ); + case T_CHAR: return new (C, 3) LoadUSNode(ctl, mem, adr, adr_type, rt->is_int() ); + case T_SHORT: return new (C, 3) LoadSNode (ctl, mem, adr, adr_type, rt->is_int() ); + case T_LONG: return new (C, 3) LoadLNode (ctl, mem, adr, adr_type, rt->is_long() ); + case T_FLOAT: return new (C, 3) LoadFNode (ctl, mem, adr, adr_type, rt ); + case T_DOUBLE: return new (C, 3) LoadDNode (ctl, mem, adr, adr_type, rt ); + case T_ADDRESS: return new (C, 3) LoadPNode (ctl, mem, adr, adr_type, rt->is_ptr() ); case T_OBJECT: #ifdef _LP64 if (adr->bottom_type()->is_ptr_to_narrowoop()) { @@ -1076,13 +1076,14 @@ // of the original value. Node* mem_phi = in(Memory); Node* offset = in(Address)->in(AddPNode::Offset); + Node* region = base->in(0); Node* in1 = clone(); Node* in1_addr = in1->in(Address)->clone(); in1_addr->set_req(AddPNode::Base, base->in(allocation_index)); in1_addr->set_req(AddPNode::Address, base->in(allocation_index)); in1_addr->set_req(AddPNode::Offset, offset); - in1->set_req(0, base->in(allocation_index)); + in1->set_req(0, region->in(allocation_index)); in1->set_req(Address, in1_addr); in1->set_req(Memory, mem_phi->in(allocation_index)); @@ -1091,7 +1092,7 @@ in2_addr->set_req(AddPNode::Base, base->in(load_index)); in2_addr->set_req(AddPNode::Address, base->in(load_index)); in2_addr->set_req(AddPNode::Offset, offset); - in2->set_req(0, base->in(load_index)); + in2->set_req(0, region->in(load_index)); in2->set_req(Address, in2_addr); in2->set_req(Memory, mem_phi->in(load_index)); @@ -1100,7 +1101,7 @@ in2_addr = phase->transform(in2_addr); in2 = phase->transform(in2); - PhiNode* result = PhiNode::make_blank(base->in(0), this); + PhiNode* result = PhiNode::make_blank(region, this); result->set_req(allocation_index, in1); result->set_req(load_index, in2); return result; @@ -1303,6 +1304,7 @@ Node* base = AddPNode::Ideal_base_and_offset(address, phase, ignore); if (base != NULL && phase->type(base)->higher_equal(TypePtr::NOTNULL) + && phase->C->get_alias_index(phase->type(address)->is_ptr()) != Compile::AliasIdxRaw && all_controls_dominate(base, phase->C->start())) { // A method-invariant, non-null address (constant or 'this' argument). set_req(MemNode::Control, NULL); @@ -1356,7 +1358,7 @@ // Steps (a), (b): Walk past independent stores to find an exact match. if (prev_mem != NULL && prev_mem != in(MemNode::Memory)) { // (c) See if we can fold up on the spot, but don't fold up here. - // Fold-up might require truncation (for LoadB/LoadS/LoadC) or + // Fold-up might require truncation (for LoadB/LoadS/LoadUS) or // just return a prior value, which is done by Identity calls. if (can_see_stored_value(prev_mem, phase)) { // Make ready for step (d): @@ -1605,14 +1607,14 @@ return LoadNode::Ideal(phase, can_reshape); } -//--------------------------LoadCNode::Ideal-------------------------------------- +//--------------------------LoadUSNode::Ideal------------------------------------- // // If the previous store is to the same address as this load, // and the value stored was larger than a char, replace this load // with the value stored truncated to a char. If no truncation is // needed, the replacement is done in LoadNode::Identity(). // -Node *LoadCNode::Ideal(PhaseGVN *phase, bool can_reshape) { +Node *LoadUSNode::Ideal(PhaseGVN *phase, bool can_reshape) { Node* mem = in(MemNode::Memory); Node* value = can_see_stored_value(mem,phase); if( value && !phase->type(value)->higher_equal( _type ) )
--- a/src/share/vm/opto/memnode.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/memnode.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -207,11 +207,11 @@ virtual BasicType memory_type() const { return T_BYTE; } }; -//------------------------------LoadCNode-------------------------------------- -// Load a char (16bits unsigned) from memory -class LoadCNode : public LoadNode { +//------------------------------LoadUSNode------------------------------------- +// Load an unsigned short/char (16bits unsigned) from memory +class LoadUSNode : public LoadNode { public: - LoadCNode( Node *c, Node *mem, Node *adr, const TypePtr* at, const TypeInt *ti = TypeInt::CHAR ) + LoadUSNode( Node *c, Node *mem, Node *adr, const TypePtr* at, const TypeInt *ti = TypeInt::CHAR ) : LoadNode(c,mem,adr,at,ti) {} virtual int Opcode() const; virtual uint ideal_reg() const { return Op_RegI; }
--- a/src/share/vm/opto/mulnode.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/mulnode.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2009 Sun Microsystems, Inc. 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 @@ -442,16 +442,17 @@ return load; } uint lop = load->Opcode(); - if( lop == Op_LoadC && + if( lop == Op_LoadUS && con == 0x0000FFFF ) // Already zero-extended return load; // Masking off the high bits of a unsigned-shift-right is not // needed either. if( lop == Op_URShiftI ) { const TypeInt *t12 = phase->type( load->in(2) )->isa_int(); - if( t12 && t12->is_con() ) { - int shift_con = t12->get_con(); - int mask = max_juint >> shift_con; + if( t12 && t12->is_con() ) { // Shift is by a constant + int shift = t12->get_con(); + shift &= BitsPerJavaInteger - 1; // semantics of Java shifts + int mask = max_juint >> shift; if( (mask&con) == mask ) // If AND is useless, skip it return load; } @@ -470,19 +471,19 @@ uint lop = load->Opcode(); // Masking bits off of a Character? Hi bits are already zero. - if( lop == Op_LoadC && + if( lop == Op_LoadUS && (mask & 0xFFFF0000) ) // Can we make a smaller mask? return new (phase->C, 3) AndINode(load,phase->intcon(mask&0xFFFF)); // Masking bits off of a Short? Loading a Character does some masking if( lop == Op_LoadS && (mask & 0xFFFF0000) == 0 ) { - Node *ldc = new (phase->C, 3) LoadCNode(load->in(MemNode::Control), + Node *ldus = new (phase->C, 3) LoadUSNode(load->in(MemNode::Control), load->in(MemNode::Memory), load->in(MemNode::Address), load->adr_type()); - ldc = phase->transform(ldc); - return new (phase->C, 3) AndINode(ldc,phase->intcon(mask&0xFFFF)); + ldus = phase->transform(ldus); + return new (phase->C, 3) AndINode(ldus, phase->intcon(mask&0xFFFF)); } // Masking sign bits off of a Byte? Let the matcher use an unsigned load @@ -579,9 +580,10 @@ // needed either. if( lop == Op_URShiftL ) { const TypeInt *t12 = phase->type( usr->in(2) )->isa_int(); - if( t12 && t12->is_con() ) { - int shift_con = t12->get_con(); - jlong mask = max_julong >> shift_con; + if( t12 && t12->is_con() ) { // Shift is by a constant + int shift = t12->get_con(); + shift &= BitsPerJavaLong - 1; // semantics of Java shifts + jlong mask = max_julong >> shift; if( (mask&con) == mask ) // If AND is useless, skip it return usr; } @@ -605,8 +607,8 @@ const TypeInt *t12 = phase->type(rsh->in(2))->isa_int(); if( t12 && t12->is_con() ) { // Shift is by a constant int shift = t12->get_con(); - shift &= (BitsPerJavaInteger*2)-1; // semantics of Java shifts - const jlong sign_bits_mask = ~(((jlong)CONST64(1) << (jlong)(BitsPerJavaInteger*2 - shift)) -1); + shift &= BitsPerJavaLong - 1; // semantics of Java shifts + const jlong sign_bits_mask = ~(((jlong)CONST64(1) << (jlong)(BitsPerJavaLong - shift)) -1); // If the AND'ing of the 2 masks has no bits, then only original shifted // bits survive. NO sign-extension bits survive the maskings. if( (sign_bits_mask & mask) == 0 ) { @@ -786,7 +788,7 @@ // Check for ((x & ((CONST64(1)<<(64-c0))-1)) << c0) which ANDs off high bits // before shifting them away. - const jlong bits_mask = ((jlong)CONST64(1) << (jlong)(BitsPerJavaInteger*2 - con)) - CONST64(1); + const jlong bits_mask = ((jlong)CONST64(1) << (jlong)(BitsPerJavaLong - con)) - CONST64(1); if( add1_op == Op_AndL && phase->type(add1->in(2)) == TypeLong::make( bits_mask ) ) return new (phase->C, 3) LShiftLNode( add1->in(1), in(2) ); @@ -820,7 +822,7 @@ return TypeLong::LONG; uint shift = r2->get_con(); - shift &= (BitsPerJavaInteger*2)-1; // semantics of Java shifts + shift &= BitsPerJavaLong - 1; // semantics of Java shifts // Shift by a multiple of 64 does nothing: if (shift == 0) return t1; @@ -913,7 +915,7 @@ set_req(2, phase->intcon(0)); return this; } - else if( ld->Opcode() == Op_LoadC ) + else if( ld->Opcode() == Op_LoadUS ) // Replace zero-extension-load with sign-extension-load return new (phase->C, 3) LoadSNode( ld->in(MemNode::Control), ld->in(MemNode::Memory), @@ -1235,7 +1237,7 @@ if ( con == 0 ) return NULL; // let Identity() handle a 0 shift count // note: mask computation below does not work for 0 shift count // We'll be wanting the right-shift amount as a mask of that many bits - const jlong mask = (((jlong)CONST64(1) << (jlong)(BitsPerJavaInteger*2 - con)) -1); + const jlong mask = (((jlong)CONST64(1) << (jlong)(BitsPerJavaLong - con)) -1); // Check for ((x << z) + Y) >>> z. Replace with x + con>>>z // The idiom for rounding to a power of 2 is "(Q+(2^z-1)) >>> z". @@ -1302,7 +1304,7 @@ if (r2->is_con()) { uint shift = r2->get_con(); - shift &= (2*BitsPerJavaInteger)-1; // semantics of Java shifts + shift &= BitsPerJavaLong - 1; // semantics of Java shifts // Shift by a multiple of 64 does nothing: if (shift == 0) return t1; // Calculate reasonably aggressive bounds for the result. @@ -1325,7 +1327,7 @@ const TypeLong* tl = TypeLong::make(lo, hi, MAX2(r1->_widen,r2->_widen)); #ifdef ASSERT // Make sure we get the sign-capture idiom correct. - if (shift == (2*BitsPerJavaInteger)-1) { + if (shift == BitsPerJavaLong - 1) { if (r1->_lo >= 0) assert(tl == TypeLong::ZERO, ">>>63 of + is 0"); if (r1->_hi < 0) assert(tl == TypeLong::ONE, ">>>63 of - is +1"); }
--- a/src/share/vm/opto/superword.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/superword.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -1444,7 +1444,7 @@ // (Start, end] half-open range defining which operands are vector void SuperWord::vector_opd_range(Node* n, uint* start, uint* end) { switch (n->Opcode()) { - case Op_LoadB: case Op_LoadC: + case Op_LoadB: case Op_LoadUS: case Op_LoadI: case Op_LoadL: case Op_LoadF: case Op_LoadD: case Op_LoadP:
--- a/src/share/vm/opto/type.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/type.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -2471,6 +2471,8 @@ const Type* ft = join(kills); const TypeInstPtr* ftip = ft->isa_instptr(); const TypeInstPtr* ktip = kills->isa_instptr(); + const TypeKlassPtr* ftkp = ft->isa_klassptr(); + const TypeKlassPtr* ktkp = kills->isa_klassptr(); if (ft->empty()) { // Check for evil case of 'this' being a class and 'kills' expecting an @@ -2484,6 +2486,8 @@ // uplift the type. if (!empty() && ktip != NULL && ktip->is_loaded() && ktip->klass()->is_interface()) return kills; // Uplift to interface + if (!empty() && ktkp != NULL && ktkp->klass()->is_loaded() && ktkp->klass()->is_interface()) + return kills; // Uplift to interface return Type::TOP; // Canonical empty value } @@ -2499,6 +2503,12 @@ // Happens in a CTW of rt.jar, 320-341, no extra flags return ktip->cast_to_ptr_type(ftip->ptr()); } + if (ftkp != NULL && ktkp != NULL && + ftkp->is_loaded() && ftkp->klass()->is_interface() && + ktkp->is_loaded() && !ktkp->klass()->is_interface()) { + // Happens in a CTW of rt.jar, 320-341, no extra flags + return ktkp->cast_to_ptr_type(ftkp->ptr()); + } return ft; } @@ -3657,7 +3667,7 @@ //------------------------------cast_to_ptr_type------------------------------- const Type *TypeKlassPtr::cast_to_ptr_type(PTR ptr) const { - assert(_base == OopPtr, "subclass must override cast_to_ptr_type"); + assert(_base == KlassPtr, "subclass must override cast_to_ptr_type"); if( ptr == _ptr ) return this; return make(ptr, _klass, _offset); }
--- a/src/share/vm/opto/type.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/type.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -882,6 +882,8 @@ public: ciSymbol* name() const { return _klass->name(); } + bool is_loaded() const { return _klass->is_loaded(); } + // ptr to klass 'k' static const TypeKlassPtr *make( ciKlass* k ) { return make( TypePtr::Constant, k, 0); } // ptr to klass 'k' with offset
--- a/src/share/vm/opto/vectornode.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/opto/vectornode.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -239,7 +239,7 @@ return Op_XorV; case Op_LoadB: - case Op_LoadC: + case Op_LoadUS: case Op_LoadS: case Op_LoadI: case Op_LoadL: @@ -269,7 +269,7 @@ case 16: return Op_Load16B; } break; - case Op_LoadC: + case Op_LoadUS: switch (vlen) { case 2: return Op_Load2C; case 4: return Op_Load4C;
--- a/src/share/vm/prims/jni.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/prims/jni.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -2691,8 +2691,13 @@ directBufferSupportInitializeEnded = 1; } else { - ThreadInVMfromNative tivn(thread); // set state as yield_all can call os:sleep while (!directBufferSupportInitializeEnded && !directBufferSupportInitializeFailed) { + // Set state as yield_all can call os:sleep. On Solaris, yield_all calls + // os::sleep which requires the VM state transition. On other platforms, it + // is not necessary. The following call to change the VM state is purposely + // put inside the loop to avoid potential deadlock when multiple threads + // try to call this method. See 6791815 for more details. + ThreadInVMfromNative tivn(thread); os::yield_all(); } }
--- a/src/share/vm/prims/jvm.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/prims/jvm.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -2475,7 +2475,8 @@ if (Arguments::vfprintf_hook() != NULL) { jio_fprintf(defaultStream::output_stream(), "%s", s); } else { - ::write(defaultStream::output_fd(), s, (int)strlen(s)); + // Make an unused local variable to avoid warning from gcc 4.x compiler. + size_t count = ::write(defaultStream::output_fd(), s, (int)strlen(s)); } }
--- a/src/share/vm/prims/jvmtiRedefineClasses.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/prims/jvmtiRedefineClasses.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -1230,8 +1230,14 @@ // Constant pools are not easily reused so we allocate a new one // each time. + // merge_cp is created unsafe for concurrent GC processing. It + // should be marked safe before discarding it because, even if + // garbage. If it crosses a card boundary, it may be scanned + // in order to find the start of the first complete object on the card. constantPoolHandle merge_cp(THREAD, - oopFactory::new_constantPool(merge_cp_length, THREAD)); + oopFactory::new_constantPool(merge_cp_length, + methodOopDesc::IsUnsafeConc, + THREAD)); int orig_length = old_cp->orig_length(); if (orig_length == 0) { // This old_cp is an actual original constant pool. We save @@ -1274,6 +1280,7 @@ // rewriting so we can't use the old constant pool with the new // class. + merge_cp()->set_is_conc_safe(true); merge_cp = constantPoolHandle(); // toss the merged constant pool } else if (old_cp->length() < scratch_cp->length()) { // The old constant pool has fewer entries than the new constant @@ -1283,6 +1290,7 @@ // rewriting so we can't use the new constant pool with the old // class. + merge_cp()->set_is_conc_safe(true); merge_cp = constantPoolHandle(); // toss the merged constant pool } else { // The old constant pool has more entries than the new constant @@ -1296,6 +1304,7 @@ set_new_constant_pool(scratch_class, merge_cp, merge_cp_length, true, THREAD); // drop local ref to the merged constant pool + merge_cp()->set_is_conc_safe(true); merge_cp = constantPoolHandle(); } } else { @@ -1325,7 +1334,10 @@ // GCed. set_new_constant_pool(scratch_class, merge_cp, merge_cp_length, true, THREAD); + merge_cp()->set_is_conc_safe(true); } + assert(old_cp()->is_conc_safe(), "Just checking"); + assert(scratch_cp()->is_conc_safe(), "Just checking"); return JVMTI_ERROR_NONE; } // end merge_cp_and_rewrite() @@ -2314,13 +2326,16 @@ // worst case merge situation. We want to associate the minimum // sized constant pool with the klass to save space. constantPoolHandle smaller_cp(THREAD, - oopFactory::new_constantPool(scratch_cp_length, THREAD)); + oopFactory::new_constantPool(scratch_cp_length, + methodOopDesc::IsUnsafeConc, + THREAD)); // preserve orig_length() value in the smaller copy int orig_length = scratch_cp->orig_length(); assert(orig_length != 0, "sanity check"); smaller_cp->set_orig_length(orig_length); scratch_cp->copy_cp_to(1, scratch_cp_length - 1, smaller_cp, 1, THREAD); scratch_cp = smaller_cp; + smaller_cp()->set_is_conc_safe(true); } // attach new constant pool to klass @@ -2516,6 +2531,7 @@ rewrite_cp_refs_in_stack_map_table(method, THREAD); } // end for each method + assert(scratch_cp()->is_conc_safe(), "Just checking"); } // end set_new_constant_pool()
--- a/src/share/vm/prims/jvmtiTagMap.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/prims/jvmtiTagMap.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -1320,6 +1320,9 @@ } // do the iteration + // If this operation encounters a bad object when using CMS, + // consider using safe_object_iterate() which avoids perm gen + // objects that may contain bad references. Universe::heap()->object_iterate(_blk); // when sharing is enabled we must iterate over the shared spaces
--- a/src/share/vm/runtime/arguments.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/runtime/arguments.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -229,6 +229,7 @@ inline void set_base(const char* base); inline void add_prefix(const char* prefix); + inline void add_suffix_to_prefix(const char* suffix); inline void add_suffix(const char* suffix); inline void reset_path(const char* base); @@ -290,6 +291,10 @@ _items[_scp_prefix] = add_to_path(_items[_scp_prefix], prefix, true); } +inline void SysClassPath::add_suffix_to_prefix(const char* suffix) { + _items[_scp_prefix] = add_to_path(_items[_scp_prefix], suffix, false); +} + inline void SysClassPath::add_suffix(const char* suffix) { _items[_scp_suffix] = add_to_path(_items[_scp_suffix], suffix, false); } @@ -512,7 +517,6 @@ return CommandLineFlags::boolAtPut(name, &value, origin); } - static bool set_fp_numeric_flag(char* name, char* value, FlagValueOrigin origin) { double v; if (sscanf(value, "%lf", &v) != 1) { @@ -525,7 +529,6 @@ return false; } - static bool set_numeric_flag(char* name, char* value, FlagValueOrigin origin) { julong v; intx intx_v; @@ -555,7 +558,6 @@ return false; } - static bool set_string_flag(char* name, const char* value, FlagValueOrigin origin) { if (!CommandLineFlags::ccstrAtPut(name, &value, origin)) return false; // Contract: CommandLineFlags always returns a pointer that needs freeing. @@ -591,7 +593,6 @@ return true; } - bool Arguments::parse_argument(const char* arg, FlagValueOrigin origin) { // range of acceptable characters spelled out for portability reasons @@ -652,7 +653,6 @@ return false; } - void Arguments::add_string(char*** bldarray, int* count, const char* arg) { assert(bldarray != NULL, "illegal argument"); @@ -756,7 +756,6 @@ return true; } - bool Arguments::process_settings_file(const char* file_name, bool should_exist, jboolean ignore_unrecognized) { FILE* stream = fopen(file_name, "rb"); if (stream == NULL) { @@ -932,7 +931,6 @@ } } - // Conflict: required to use shared spaces (-Xshare:on), but // incompatible command line options were chosen. @@ -946,7 +944,6 @@ } } - // If the user has chosen ParallelGCThreads > 0, we set UseParNewGC // if it's not explictly set or unset. If the user has chosen // UseParNewGC and not explicitly set ParallelGCThreads we @@ -1361,7 +1358,7 @@ // Feed the cache size setting into the JDK char buffer[1024]; - sprintf(buffer, "java.lang.Integer.IntegerCache.high=%d", AutoBoxCacheMax); + sprintf(buffer, "java.lang.Integer.IntegerCache.high=" INTX_FORMAT, AutoBoxCacheMax); add_property(buffer); } if (AggressiveOpts && FLAG_IS_DEFAULT(DoEscapeAnalysis)) { @@ -1714,6 +1711,21 @@ return result; } + if (AggressiveOpts) { + // Insert alt-rt.jar between user-specified bootclasspath + // prefix and the default bootclasspath. os::set_boot_path() + // uses meta_index_dir as the default bootclasspath directory. + const char* altclasses_jar = "alt-rt.jar"; + size_t altclasses_path_len = strlen(get_meta_index_dir()) + 1 + + strlen(altclasses_jar); + char* altclasses_path = NEW_C_HEAP_ARRAY(char, altclasses_path_len); + strcpy(altclasses_path, get_meta_index_dir()); + strcat(altclasses_path, altclasses_jar); + scp.add_suffix_to_prefix(altclasses_path); + scp_assembly_required = true; + FREE_C_HEAP_ARRAY(char, altclasses_path); + } + // Parse _JAVA_OPTIONS environment variable (if present) (mimics classic VM) result = parse_java_options_environment_variable(&scp, &scp_assembly_required); if (result != JNI_OK) { @@ -1729,7 +1741,6 @@ return JNI_OK; } - jint Arguments::parse_each_vm_init_arg(const JavaVMInitArgs* args, SysClassPath* scp_p, bool* scp_assembly_required_p, @@ -1795,7 +1806,7 @@ *scp_assembly_required_p = true; // -Xrun } else if (match_option(option, "-Xrun", &tail)) { - if(tail != NULL) { + if (tail != NULL) { const char* pos = strchr(tail, ':'); size_t len = (pos == NULL) ? strlen(tail) : pos - tail; char* name = (char*)memcpy(NEW_C_HEAP_ARRAY(char, len + 1), tail, len); @@ -2478,7 +2489,7 @@ vm_args.version = JNI_VERSION_1_2; vm_args.options = options; vm_args.nOptions = i; - vm_args.ignoreUnrecognized = false; + vm_args.ignoreUnrecognized = IgnoreUnrecognizedVMOptions; if (PrintVMOptions) { const char* tail; @@ -2525,13 +2536,12 @@ // If flag "-XX:Flags=flags-file" is used it will be the first option to be processed. bool settings_file_specified = false; + const char* flags_file; int index; for (index = 0; index < args->nOptions; index++) { const JavaVMOption *option = args->options + index; if (match_option(option, "-XX:Flags=", &tail)) { - if (!process_settings_file(tail, true, args->ignoreUnrecognized)) { - return JNI_EINVAL; - } + flags_file = tail; settings_file_specified = true; } if (match_option(option, "-XX:+PrintVMOptions", &tail)) { @@ -2540,6 +2550,24 @@ if (match_option(option, "-XX:-PrintVMOptions", &tail)) { PrintVMOptions = false; } + if (match_option(option, "-XX:+IgnoreUnrecognizedVMOptions", &tail)) { + IgnoreUnrecognizedVMOptions = true; + } + if (match_option(option, "-XX:-IgnoreUnrecognizedVMOptions", &tail)) { + IgnoreUnrecognizedVMOptions = false; + } + } + + if (IgnoreUnrecognizedVMOptions) { + // uncast const to modify the flag args->ignoreUnrecognized + *(jboolean*)(&args->ignoreUnrecognized) = true; + } + + // Parse specified settings file + if (settings_file_specified) { + if (!process_settings_file(flags_file, true, args->ignoreUnrecognized)) { + return JNI_EINVAL; + } } // Parse default .hotspotrc settings file @@ -2558,7 +2586,6 @@ } } - // Parse JavaVMInitArgs structure passed in, as well as JAVA_TOOL_OPTIONS and _JAVA_OPTIONS jint result = parse_vm_init_args(args); if (result != JNI_OK) {
--- a/src/share/vm/runtime/globals.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/runtime/globals.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -835,8 +835,21 @@ "Prints the system dictionary at exit") \ \ diagnostic(bool, UnsyncloadClass, false, \ - "Unstable: VM calls loadClass unsynchronized. Custom classloader "\ - "must call VM synchronized for findClass & defineClass") \ + "Unstable: VM calls loadClass unsynchronized. Custom " \ + "class loader must call VM synchronized for findClass " \ + "and defineClass.") \ + \ + product(bool, AlwaysLockClassLoader, false, \ + "Require the VM to acquire the class loader lock before calling " \ + "loadClass() even for class loaders registering " \ + "as parallel capable. Default false. ") \ + \ + product(bool, AllowParallelDefineClass, false, \ + "Allow parallel defineClass requests for class loaders " \ + "registering as parallel capable. Default false") \ + \ + product(bool, MustCallLoadClassInternal, false, \ + "Call loadClassInternal() rather than loadClass().Default false") \ \ product_pd(bool, DontYieldALot, \ "Throw away obvious excess yield calls (for SOLARIS only)") \ @@ -1294,7 +1307,14 @@ product(intx, ParGCArrayScanChunk, 50, \ "Scan a subset and push remainder, if array is bigger than this") \ \ - product(intx, ParGCDesiredObjsFromOverflowList, 20, \ + notproduct(bool, ParGCWorkQueueOverflowALot, false, \ + "Whether we should simulate work queue overflow in ParNew") \ + \ + notproduct(uintx, ParGCWorkQueueOverflowInterval, 1000, \ + "An `interval' counter that determines how frequently" \ + " we simulate overflow; a smaller number increases frequency") \ + \ + product(uintx, ParGCDesiredObjsFromOverflowList, 20, \ "The desired number of objects to claim from the overflow list") \ \ product(uintx, CMSParPromoteBlocksToClaim, 50, \ @@ -1406,18 +1426,18 @@ develop(bool, CMSOverflowEarlyRestoration, false, \ "Whether preserved marks should be restored early") \ \ - product(uintx, CMSMarkStackSize, 32*K, \ + product(uintx, CMSMarkStackSize, NOT_LP64(32*K) LP64_ONLY(4*M), \ "Size of CMS marking stack") \ \ - product(uintx, CMSMarkStackSizeMax, 4*M, \ + product(uintx, CMSMarkStackSizeMax, NOT_LP64(4*M) LP64_ONLY(512*M), \ "Max size of CMS marking stack") \ \ notproduct(bool, CMSMarkStackOverflowALot, false, \ "Whether we should simulate frequent marking stack / work queue" \ " overflow") \ \ - notproduct(intx, CMSMarkStackOverflowInterval, 1000, \ - "A per-thread `interval' counter that determines how frequently" \ + notproduct(uintx, CMSMarkStackOverflowInterval, 1000, \ + "An `interval' counter that determines how frequently" \ " we simulate overflow; a smaller number increases frequency") \ \ product(uintx, CMSMaxAbortablePrecleanLoops, 0, \ @@ -1635,7 +1655,14 @@ develop(uintx, WorkStealingYieldsBeforeSleep, 1000, \ "Number of yields before a sleep is done during workstealing") \ \ - product(uintx, PreserveMarkStackSize, 40, \ + develop(uintx, WorkStealingHardSpins, 4096, \ + "Number of iterations in a spin loop between checks on " \ + "time out of hard spin") \ + \ + develop(uintx, WorkStealingSpinToYieldRatio, 10, \ + "Ratio of hard spins to calls to yield") \ + \ + product(uintx, PreserveMarkStackSize, 1024, \ "Size for stack used in promotion failure handling") \ \ product_pd(bool, UseTLAB, "Use thread-local object allocation") \ @@ -2167,6 +2194,9 @@ product(bool, PrintVMOptions, trueInDebug, \ "print VM flag settings") \ \ + product(bool, IgnoreUnrecognizedVMOptions, false, \ + "Ignore unrecognized VM options") \ + \ diagnostic(bool, SerializeVMOutput, true, \ "Use a mutex to serialize output to tty and hotspot.log") \ \
--- a/src/share/vm/runtime/memprofiler.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/runtime/memprofiler.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -104,21 +104,22 @@ } // Print trace line in log - fprintf(_log_fp, "%6.1f,%5d,%5d,%6ld,%6ld,%6ld,%6ld,", - os::elapsedTime(), - Threads::number_of_threads(), - SystemDictionary::number_of_classes(), - Universe::heap()->used() / K, - Universe::heap()->capacity() / K, - Universe::heap()->permanent_used() / HWperKB, - Universe::heap()->permanent_capacity() / HWperKB); + fprintf(_log_fp, "%6.1f,%5d,%5d," UINTX_FORMAT_W(6) "," UINTX_FORMAT_W(6) "," + UINTX_FORMAT_W(6) "," UINTX_FORMAT_W(6) ",", + os::elapsedTime(), + Threads::number_of_threads(), + SystemDictionary::number_of_classes(), + Universe::heap()->used() / K, + Universe::heap()->capacity() / K, + Universe::heap()->permanent_used() / HWperKB, + Universe::heap()->permanent_capacity() / HWperKB); - fprintf(_log_fp, "%6ld,", CodeCache::capacity() / K); + fprintf(_log_fp, UINTX_FORMAT_W(6) ",", CodeCache::capacity() / K); - fprintf(_log_fp, "%6ld,%6ld,%6ld\n", - handles_memory_usage / K, - resource_memory_usage / K, - OopMapCache::memory_usage() / K); + fprintf(_log_fp, UINTX_FORMAT_W(6) "," UINTX_FORMAT_W(6) ",%6ld\n", + handles_memory_usage / K, + resource_memory_usage / K, + OopMapCache::memory_usage() / K); fflush(_log_fp); }
--- a/src/share/vm/runtime/os.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/runtime/os.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -74,13 +74,11 @@ const int milliseconds_after_second = milliseconds_since_19700101 % milliseconds_per_microsecond; // Convert the time value to a tm and timezone variable - const struct tm *time_struct_temp = localtime(&seconds_since_19700101); - if (time_struct_temp == NULL) { - assert(false, "Failed localtime"); + struct tm time_struct; + if (localtime_pd(&seconds_since_19700101, &time_struct) == NULL) { + assert(false, "Failed localtime_pd"); return NULL; } - // Save the results of localtime - const struct tm time_struct = *time_struct_temp; const time_t zone = timezone; // If daylight savings time is in effect, @@ -93,10 +91,10 @@ UTC_to_local = UTC_to_local - seconds_per_hour; } // Compute the time zone offset. - // localtime(3C) sets timezone to the difference (in seconds) + // localtime_pd() sets timezone to the difference (in seconds) // between UTC and and local time. // ISO 8601 says we need the difference between local time and UTC, - // we change the sign of the localtime(3C) result. + // we change the sign of the localtime_pd() result. const time_t local_to_UTC = -(UTC_to_local); // Then we have to figure out if if we are ahead (+) or behind (-) UTC. char sign_local_to_UTC = '+';
--- a/src/share/vm/runtime/os.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/runtime/os.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -120,7 +120,8 @@ // Return current local time in a string (YYYY-MM-DD HH:MM:SS). // It is MT safe, but not async-safe, as reading time zone // information may require a lock on some platforms. - static char* local_time_string(char *buf, size_t buflen); + static char* local_time_string(char *buf, size_t buflen); + static struct tm* localtime_pd (const time_t* clock, struct tm* res); // Fill in buffer with current local time as an ISO-8601 string. // E.g., YYYY-MM-DDThh:mm:ss.mmm+zzzz. // Returns buffer, or NULL if it failed.
--- a/src/share/vm/runtime/safepoint.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/runtime/safepoint.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -730,7 +730,7 @@ if (DieOnSafepointTimeout) { char msg[1024]; VM_Operation *op = VMThread::vm_operation(); - sprintf(msg, "Safepoint sync time longer than %d ms detected when executing %s.", + sprintf(msg, "Safepoint sync time longer than " INTX_FORMAT "ms detected when executing %s.", SafepointTimeoutDelay, op != NULL ? op->name() : "no vm operation"); fatal(msg);
--- a/src/share/vm/runtime/sharedRuntime.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/runtime/sharedRuntime.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -192,64 +192,46 @@ JRT_LEAF(jint, SharedRuntime::f2i(jfloat x)) - if (g_isnan(x)) {return 0;} - jlong lltmp = (jlong)x; - jint ltmp = (jint)lltmp; - if (ltmp == lltmp) { - return ltmp; - } else { - if (x < 0) { - return min_jint; - } else { - return max_jint; - } - } + if (g_isnan(x)) + return 0; + if (x >= (jfloat) max_jint) + return max_jint; + if (x <= (jfloat) min_jint) + return min_jint; + return (jint) x; JRT_END JRT_LEAF(jlong, SharedRuntime::f2l(jfloat x)) - if (g_isnan(x)) {return 0;} - jlong lltmp = (jlong)x; - if (lltmp != min_jlong) { - return lltmp; - } else { - if (x < 0) { - return min_jlong; - } else { - return max_jlong; - } - } + if (g_isnan(x)) + return 0; + if (x >= (jfloat) max_jlong) + return max_jlong; + if (x <= (jfloat) min_jlong) + return min_jlong; + return (jlong) x; JRT_END JRT_LEAF(jint, SharedRuntime::d2i(jdouble x)) - if (g_isnan(x)) {return 0;} - jlong lltmp = (jlong)x; - jint ltmp = (jint)lltmp; - if (ltmp == lltmp) { - return ltmp; - } else { - if (x < 0) { - return min_jint; - } else { - return max_jint; - } - } + if (g_isnan(x)) + return 0; + if (x >= (jdouble) max_jint) + return max_jint; + if (x <= (jdouble) min_jint) + return min_jint; + return (jint) x; JRT_END JRT_LEAF(jlong, SharedRuntime::d2l(jdouble x)) - if (g_isnan(x)) {return 0;} - jlong lltmp = (jlong)x; - if (lltmp != min_jlong) { - return lltmp; - } else { - if (x < 0) { - return min_jlong; - } else { - return max_jlong; - } - } + if (g_isnan(x)) + return 0; + if (x >= (jdouble) max_jlong) + return max_jlong; + if (x <= (jdouble) min_jlong) + return min_jlong; + return (jlong) x; JRT_END
--- a/src/share/vm/runtime/synchronizer.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/runtime/synchronizer.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -424,7 +424,7 @@ // asserts is that error message -- often something about negative array // indices -- is opaque. -#define CTASSERT(x) { int tag[1-(2*!(x))]; printf ("Tag @%X\n", tag); } +#define CTASSERT(x) { int tag[1-(2*!(x))]; printf ("Tag @" INTPTR_FORMAT "\n", (intptr_t)tag); } void ObjectMonitor::ctAsserts() { CTASSERT(offset_of (ObjectMonitor, _header) == 0);
--- a/src/share/vm/services/heapDumper.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/services/heapDumper.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -1700,7 +1700,7 @@ // The HPROF_GC_CLASS_DUMP and HPROF_GC_INSTANCE_DUMP are the vast bulk // of the heap dump. HeapObjectDumper obj_dumper(this, writer()); - Universe::heap()->object_iterate(&obj_dumper); + Universe::heap()->safe_object_iterate(&obj_dumper); // HPROF_GC_ROOT_THREAD_OBJ + frames + jni locals do_threads();
--- a/src/share/vm/utilities/globalDefinitions.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/utilities/globalDefinitions.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2009 Sun Microsystems, Inc. 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 @@ -74,6 +74,7 @@ extern int BitsPerHeapOop; const int BitsPerJavaInteger = 32; +const int BitsPerJavaLong = 64; const int BitsPerSize_t = size_tSize * BitsPerByte; // Size of a char[] needed to represent a jint as a string in decimal. @@ -906,6 +907,14 @@ return log2_intptr(x); } +//* the argument must be exactly a power of 2 +inline int exact_log2_long(jlong x) { + #ifdef ASSERT + if (!is_power_of_2_long(x)) basic_fatal("x must be a power of 2"); + #endif + return log2_long(x); +} + // returns integer round-up to the nearest multiple of s (s must be a power of two) inline intptr_t round_to(intptr_t x, uintx s) { @@ -1087,15 +1096,24 @@ // Format macros that allow the field width to be specified. The width must be // a string literal (e.g., "8") or a macro that evaluates to one. #ifdef _LP64 +#define UINTX_FORMAT_W(width) UINT64_FORMAT_W(width) #define SSIZE_FORMAT_W(width) INT64_FORMAT_W(width) #define SIZE_FORMAT_W(width) UINT64_FORMAT_W(width) #else +#define UINTX_FORMAT_W(width) UINT32_FORMAT_W(width) #define SSIZE_FORMAT_W(width) INT32_FORMAT_W(width) #define SIZE_FORMAT_W(width) UINT32_FORMAT_W(width) #endif // _LP64 // Format pointers and size_t (or size_t-like integer types) which change size -// between 32- and 64-bit. +// between 32- and 64-bit. The pointer format theoretically should be "%p", +// however, it has different output on different platforms. On Windows, the data +// will be padded with zeros automatically. On Solaris, we can use "%016p" & +// "%08p" on 64 bit & 32 bit platforms to make the data padded with extra zeros. +// On Linux, "%016p" or "%08p" is not be allowed, at least on the latest GCC +// 4.3.2. So we have to use "%016x" or "%08x" to simulate the printing format. +// GCC 4.3.2, however requires the data to be converted to "intptr_t" when +// using "%x". #ifdef _LP64 #define PTR_FORMAT PTR64_FORMAT #define UINTX_FORMAT UINT64_FORMAT
--- a/src/share/vm/utilities/globalDefinitions_gcc.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/utilities/globalDefinitions_gcc.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -116,7 +116,9 @@ #ifdef _LP64 #define NULL_WORD 0L #else - #define NULL_WORD 0 + // Cast 0 to intptr_t rather than int32_t since they are not the same type + // on platforms such as Mac OS X. + #define NULL_WORD ((intptr_t)0) #endif #else #define NULL_WORD NULL
--- a/src/share/vm/utilities/globalDefinitions_sparcWorks.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/utilities/globalDefinitions_sparcWorks.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -115,7 +115,9 @@ #ifdef _LP64 #define NULL_WORD 0L #else - #define NULL_WORD 0 + // Cast 0 to intptr_t rather than int32_t since they are not the same type + // on some platforms. + #define NULL_WORD ((intptr_t)0) #endif #else #define NULL_WORD NULL
--- a/src/share/vm/utilities/ostream.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/utilities/ostream.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -300,7 +300,10 @@ } void fileStream::write(const char* s, size_t len) { - if (_file != NULL) fwrite(s, 1, len, _file); + if (_file != NULL) { + // Make an unused local variable to avoid warning from gcc 4.x compiler. + size_t count = fwrite(s, 1, len, _file); + } update_position(s, len); } @@ -328,7 +331,10 @@ } void fdStream::write(const char* s, size_t len) { - if (_fd != -1) ::write(_fd, s, (int)len); + if (_fd != -1) { + // Make an unused local variable to avoid warning from gcc 4.x compiler. + size_t count = ::write(_fd, s, (int)len); + } update_position(s, len); }
--- a/src/share/vm/utilities/taskqueue.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/utilities/taskqueue.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -25,6 +25,12 @@ # include "incls/_precompiled.incl" # include "incls/_taskqueue.cpp.incl" +#ifdef TRACESPINNING +uint ParallelTaskTerminator::_total_yields = 0; +uint ParallelTaskTerminator::_total_spins = 0; +uint ParallelTaskTerminator::_total_peeks = 0; +#endif + bool TaskQueueSuper::peek() { return _bottom != _age.top(); } @@ -69,15 +75,62 @@ ParallelTaskTerminator::offer_termination(TerminatorTerminator* terminator) { Atomic::inc(&_offered_termination); - juint yield_count = 0; + uint yield_count = 0; + // Number of hard spin loops done since last yield + uint hard_spin_count = 0; + // Number of iterations in the hard spin loop. + uint hard_spin_limit = WorkStealingHardSpins; + + // If WorkStealingSpinToYieldRatio is 0, no hard spinning is done. + // If it is greater than 0, then start with a small number + // of spins and increase number with each turn at spinning until + // the count of hard spins exceeds WorkStealingSpinToYieldRatio. + // Then do a yield() call and start spinning afresh. + if (WorkStealingSpinToYieldRatio > 0) { + hard_spin_limit = WorkStealingHardSpins >> WorkStealingSpinToYieldRatio; + hard_spin_limit = MAX2(hard_spin_limit, 1U); + } + // Remember the initial spin limit. + uint hard_spin_start = hard_spin_limit; + + // Loop waiting for all threads to offer termination or + // more work. while (true) { + // Are all threads offering termination? if (_offered_termination == _n_threads) { - //inner_termination_loop(); return true; } else { + // Look for more work. + // Periodically sleep() instead of yield() to give threads + // waiting on the cores the chance to grab this code if (yield_count <= WorkStealingYieldsBeforeSleep) { + // Do a yield or hardspin. For purposes of deciding whether + // to sleep, count this as a yield. yield_count++; - yield(); + + // Periodically call yield() instead spinning + // After WorkStealingSpinToYieldRatio spins, do a yield() call + // and reset the counts and starting limit. + if (hard_spin_count > WorkStealingSpinToYieldRatio) { + yield(); + hard_spin_count = 0; + hard_spin_limit = hard_spin_start; +#ifdef TRACESPINNING + _total_yields++; +#endif + } else { + // Hard spin this time + // Increase the hard spinning period but only up to a limit. + hard_spin_limit = MIN2(2*hard_spin_limit, + (uint) WorkStealingHardSpins); + for (uint j = 0; j < hard_spin_limit; j++) { + SpinPause(); + } + hard_spin_count++; +#ifdef TRACESPINNING + _total_spins++; +#endif + } } else { if (PrintGCDetails && Verbose) { gclog_or_tty->print_cr("ParallelTaskTerminator::offer_termination() " @@ -92,6 +145,9 @@ sleep(WorkStealingSleepMillis); } +#ifdef TRACESPINNING + _total_peeks++; +#endif if (peek_in_queue_set() || (terminator != NULL && terminator->should_exit_termination())) { Atomic::dec(&_offered_termination); @@ -101,6 +157,16 @@ } } +#ifdef TRACESPINNING +void ParallelTaskTerminator::print_termination_counts() { + gclog_or_tty->print_cr("ParallelTaskTerminator Total yields: %lld " + "Total spins: %lld Total peeks: %lld", + total_yields(), + total_spins(), + total_peeks()); +} +#endif + void ParallelTaskTerminator::reset_for_reuse() { if (_offered_termination != 0) { assert(_offered_termination == _n_threads,
--- a/src/share/vm/utilities/taskqueue.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/utilities/taskqueue.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -22,67 +22,76 @@ * */ +#ifdef LP64 +typedef juint TAG_TYPE; +// for a taskqueue size of 4M +#define LOG_TASKQ_SIZE 22 +#else +typedef jushort TAG_TYPE; +// for a taskqueue size of 16K +#define LOG_TASKQ_SIZE 14 +#endif + class TaskQueueSuper: public CHeapObj { protected: // The first free element after the last one pushed (mod _n). - // (For now we'll assume only 32-bit CAS). - volatile juint _bottom; + volatile uint _bottom; // log2 of the size of the queue. enum SomeProtectedConstants { - Log_n = 14 + Log_n = LOG_TASKQ_SIZE }; +#undef LOG_TASKQ_SIZE // Size of the queue. - juint n() { return (1 << Log_n); } + uint n() { return (1 << Log_n); } // For computing "x mod n" efficiently. - juint n_mod_mask() { return n() - 1; } + uint n_mod_mask() { return n() - 1; } struct Age { - jushort _top; - jushort _tag; + TAG_TYPE _top; + TAG_TYPE _tag; - jushort tag() const { return _tag; } - jushort top() const { return _top; } + TAG_TYPE tag() const { return _tag; } + TAG_TYPE top() const { return _top; } Age() { _tag = 0; _top = 0; } friend bool operator ==(const Age& a1, const Age& a2) { return a1.tag() == a2.tag() && a1.top() == a2.top(); } - }; Age _age; // These make sure we do single atomic reads and writes. Age get_age() { - jint res = *(volatile jint*)(&_age); + uint res = *(volatile uint*)(&_age); return *(Age*)(&res); } void set_age(Age a) { - *(volatile jint*)(&_age) = *(int*)(&a); + *(volatile uint*)(&_age) = *(uint*)(&a); } - jushort get_top() { + TAG_TYPE get_top() { return get_age().top(); } // These both operate mod _n. - juint increment_index(juint ind) { + uint increment_index(uint ind) { return (ind + 1) & n_mod_mask(); } - juint decrement_index(juint ind) { + uint decrement_index(uint ind) { return (ind - 1) & n_mod_mask(); } // Returns a number in the range [0.._n). If the result is "n-1", it // should be interpreted as 0. - juint dirty_size(juint bot, juint top) { - return ((jint)bot - (jint)top) & n_mod_mask(); + uint dirty_size(uint bot, uint top) { + return ((int)bot - (int)top) & n_mod_mask(); } // Returns the size corresponding to the given "bot" and "top". - juint size(juint bot, juint top) { - juint sz = dirty_size(bot, top); + uint size(uint bot, uint top) { + uint sz = dirty_size(bot, top); // Has the queue "wrapped", so that bottom is less than top? // There's a complicated special case here. A pair of threads could // perform pop_local and pop_global operations concurrently, starting @@ -94,7 +103,7 @@ // owner performs pop_local's, and several concurrent threads // attempting to perform the pop_global will all perform the same CAS, // and only one can succeed. Any stealing thread that reads after - // either the increment or decrement will seen an empty queue, and will + // either the increment or decrement will see an empty queue, and will // not join the competitors. The "sz == -1 || sz == _n-1" state will // not be modified by concurrent queues, so the owner thread can reset // the state to _bottom == top so subsequent pushes will be performed @@ -112,11 +121,11 @@ // Return an estimate of the number of elements in the queue. // The "careful" version admits the possibility of pop_local/pop_global // races. - juint size() { + uint size() { return size(_bottom, get_top()); } - juint dirty_size() { + uint dirty_size() { return dirty_size(_bottom, get_top()); } @@ -127,15 +136,15 @@ // Maximum number of elements allowed in the queue. This is two less // than the actual queue size, for somewhat complicated reasons. - juint max_elems() { return n() - 2; } + uint max_elems() { return n() - 2; } }; template<class E> class GenericTaskQueue: public TaskQueueSuper { private: // Slow paths for push, pop_local. (pop_global has no fast path.) - bool push_slow(E t, juint dirty_n_elems); - bool pop_local_slow(juint localBot, Age oldAge); + bool push_slow(E t, uint dirty_n_elems); + bool pop_local_slow(uint localBot, Age oldAge); public: // Initializes the queue to empty. @@ -170,7 +179,7 @@ template<class E> GenericTaskQueue<E>::GenericTaskQueue():TaskQueueSuper() { - assert(sizeof(Age) == sizeof(jint), "Depends on this."); + assert(sizeof(Age) == sizeof(int), "Depends on this."); } template<class E> @@ -182,9 +191,9 @@ template<class E> void GenericTaskQueue<E>::oops_do(OopClosure* f) { // tty->print_cr("START OopTaskQueue::oops_do"); - int iters = size(); - juint index = _bottom; - for (int i = 0; i < iters; ++i) { + uint iters = size(); + uint index = _bottom; + for (uint i = 0; i < iters; ++i) { index = decrement_index(index); // tty->print_cr(" doing entry %d," INTPTR_T " -> " INTPTR_T, // index, &_elems[index], _elems[index]); @@ -198,10 +207,10 @@ template<class E> -bool GenericTaskQueue<E>::push_slow(E t, juint dirty_n_elems) { +bool GenericTaskQueue<E>::push_slow(E t, uint dirty_n_elems) { if (dirty_n_elems == n() - 1) { // Actually means 0, so do the push. - juint localBot = _bottom; + uint localBot = _bottom; _elems[localBot] = t; _bottom = increment_index(localBot); return true; @@ -211,7 +220,7 @@ template<class E> bool GenericTaskQueue<E>:: -pop_local_slow(juint localBot, Age oldAge) { +pop_local_slow(uint localBot, Age oldAge) { // This queue was observed to contain exactly one element; either this // thread will claim it, or a competing "pop_global". In either case, // the queue will be logically empty afterwards. Create a new Age value @@ -230,9 +239,8 @@ Age tempAge; // No competing pop_global has yet incremented "top"; we'll try to // install new_age, thus claiming the element. - assert(sizeof(Age) == sizeof(jint) && sizeof(jint) == sizeof(juint), - "Assumption about CAS unit."); - *(jint*)&tempAge = Atomic::cmpxchg(*(jint*)&newAge, (volatile jint*)&_age, *(jint*)&oldAge); + assert(sizeof(Age) == sizeof(int), "Assumption about CAS unit."); + *(uint*)&tempAge = Atomic::cmpxchg(*(uint*)&newAge, (volatile uint*)&_age, *(uint*)&oldAge); if (tempAge == oldAge) { // We win. assert(dirty_size(localBot, get_top()) != n() - 1, @@ -253,8 +261,8 @@ bool GenericTaskQueue<E>::pop_global(E& t) { Age newAge; Age oldAge = get_age(); - juint localBot = _bottom; - juint n_elems = size(localBot, oldAge.top()); + uint localBot = _bottom; + uint n_elems = size(localBot, oldAge.top()); if (n_elems == 0) { return false; } @@ -263,7 +271,7 @@ newAge._top = increment_index(newAge.top()); if ( newAge._top == 0 ) newAge._tag++; Age resAge; - *(jint*)&resAge = Atomic::cmpxchg(*(jint*)&newAge, (volatile jint*)&_age, *(jint*)&oldAge); + *(uint*)&resAge = Atomic::cmpxchg(*(uint*)&newAge, (volatile uint*)&_age, *(uint*)&oldAge); // Note that using "_bottom" here might fail, since a pop_local might // have decremented it. assert(dirty_size(localBot, newAge._top) != n() - 1, @@ -287,7 +295,7 @@ template<class E> class GenericTaskQueueSet: public TaskQueueSetSuper { private: - int _n; + uint _n; GenericTaskQueue<E>** _queues; public: @@ -300,51 +308,51 @@ } } - bool steal_1_random(int queue_num, int* seed, E& t); - bool steal_best_of_2(int queue_num, int* seed, E& t); - bool steal_best_of_all(int queue_num, int* seed, E& t); + bool steal_1_random(uint queue_num, int* seed, E& t); + bool steal_best_of_2(uint queue_num, int* seed, E& t); + bool steal_best_of_all(uint queue_num, int* seed, E& t); - void register_queue(int i, GenericTaskQueue<E>* q); + void register_queue(uint i, GenericTaskQueue<E>* q); - GenericTaskQueue<E>* queue(int n); + GenericTaskQueue<E>* queue(uint n); // The thread with queue number "queue_num" (and whose random number seed // is at "seed") is trying to steal a task from some other queue. (It // may try several queues, according to some configuration parameter.) // If some steal succeeds, returns "true" and sets "t" the stolen task, // otherwise returns false. - bool steal(int queue_num, int* seed, E& t); + bool steal(uint queue_num, int* seed, E& t); bool peek(); }; template<class E> -void GenericTaskQueueSet<E>::register_queue(int i, GenericTaskQueue<E>* q) { - assert(0 <= i && i < _n, "index out of range."); +void GenericTaskQueueSet<E>::register_queue(uint i, GenericTaskQueue<E>* q) { + assert(i < _n, "index out of range."); _queues[i] = q; } template<class E> -GenericTaskQueue<E>* GenericTaskQueueSet<E>::queue(int i) { +GenericTaskQueue<E>* GenericTaskQueueSet<E>::queue(uint i) { return _queues[i]; } template<class E> -bool GenericTaskQueueSet<E>::steal(int queue_num, int* seed, E& t) { - for (int i = 0; i < 2 * _n; i++) +bool GenericTaskQueueSet<E>::steal(uint queue_num, int* seed, E& t) { + for (uint i = 0; i < 2 * _n; i++) if (steal_best_of_2(queue_num, seed, t)) return true; return false; } template<class E> -bool GenericTaskQueueSet<E>::steal_best_of_all(int queue_num, int* seed, E& t) { +bool GenericTaskQueueSet<E>::steal_best_of_all(uint queue_num, int* seed, E& t) { if (_n > 2) { int best_k; - jint best_sz = 0; - for (int k = 0; k < _n; k++) { + uint best_sz = 0; + for (uint k = 0; k < _n; k++) { if (k == queue_num) continue; - jint sz = _queues[k]->size(); + uint sz = _queues[k]->size(); if (sz > best_sz) { best_sz = sz; best_k = k; @@ -362,9 +370,9 @@ } template<class E> -bool GenericTaskQueueSet<E>::steal_1_random(int queue_num, int* seed, E& t) { +bool GenericTaskQueueSet<E>::steal_1_random(uint queue_num, int* seed, E& t) { if (_n > 2) { - int k = queue_num; + uint k = queue_num; while (k == queue_num) k = randomParkAndMiller(seed) % _n; return _queues[2]->pop_global(t); } else if (_n == 2) { @@ -378,20 +386,20 @@ } template<class E> -bool GenericTaskQueueSet<E>::steal_best_of_2(int queue_num, int* seed, E& t) { +bool GenericTaskQueueSet<E>::steal_best_of_2(uint queue_num, int* seed, E& t) { if (_n > 2) { - int k1 = queue_num; + uint k1 = queue_num; while (k1 == queue_num) k1 = randomParkAndMiller(seed) % _n; - int k2 = queue_num; + uint k2 = queue_num; while (k2 == queue_num || k2 == k1) k2 = randomParkAndMiller(seed) % _n; // Sample both and try the larger. - juint sz1 = _queues[k1]->size(); - juint sz2 = _queues[k2]->size(); + uint sz1 = _queues[k1]->size(); + uint sz2 = _queues[k2]->size(); if (sz2 > sz1) return _queues[k2]->pop_global(t); else return _queues[k1]->pop_global(t); } else if (_n == 2) { // Just try the other one. - int k = (queue_num + 1) % 2; + uint k = (queue_num + 1) % 2; return _queues[k]->pop_global(t); } else { assert(_n == 1, "can't be zero."); @@ -402,7 +410,7 @@ template<class E> bool GenericTaskQueueSet<E>::peek() { // Try all the queues. - for (int j = 0; j < _n; j++) { + for (uint j = 0; j < _n; j++) { if (_queues[j]->peek()) return true; } @@ -418,11 +426,19 @@ // A class to aid in the termination of a set of parallel tasks using // TaskQueueSet's for work stealing. +#undef TRACESPINNING + class ParallelTaskTerminator: public StackObj { private: int _n_threads; TaskQueueSetSuper* _queue_set; - jint _offered_termination; + int _offered_termination; + +#ifdef TRACESPINNING + static uint _total_yields; + static uint _total_spins; + static uint _total_peeks; +#endif bool peek_in_queue_set(); protected: @@ -454,13 +470,19 @@ // the terminator is finished. void reset_for_reuse(); +#ifdef TRACESPINNING + static uint total_yields() { return _total_yields; } + static uint total_spins() { return _total_spins; } + static uint total_peeks() { return _total_peeks; } + static void print_termination_counts(); +#endif }; #define SIMPLE_STACK 0 template<class E> inline bool GenericTaskQueue<E>::push(E t) { #if SIMPLE_STACK - juint localBot = _bottom; + uint localBot = _bottom; if (_bottom < max_elems()) { _elems[localBot] = t; _bottom = localBot + 1; @@ -469,10 +491,10 @@ return false; } #else - juint localBot = _bottom; + uint localBot = _bottom; assert((localBot >= 0) && (localBot < n()), "_bottom out of range."); - jushort top = get_top(); - juint dirty_n_elems = dirty_size(localBot, top); + TAG_TYPE top = get_top(); + uint dirty_n_elems = dirty_size(localBot, top); assert((dirty_n_elems >= 0) && (dirty_n_elems < n()), "n_elems out of range."); if (dirty_n_elems < max_elems()) { @@ -487,19 +509,19 @@ template<class E> inline bool GenericTaskQueue<E>::pop_local(E& t) { #if SIMPLE_STACK - juint localBot = _bottom; + uint localBot = _bottom; assert(localBot > 0, "precondition."); localBot--; t = _elems[localBot]; _bottom = localBot; return true; #else - juint localBot = _bottom; + uint localBot = _bottom; // This value cannot be n-1. That can only occur as a result of // the assignment to bottom in this method. If it does, this method // resets the size( to 0 before the next call (which is sequential, // since this is pop_local.) - juint dirty_n_elems = dirty_size(localBot, get_top()); + uint dirty_n_elems = dirty_size(localBot, get_top()); assert(dirty_n_elems != n() - 1, "Shouldn't be possible..."); if (dirty_n_elems == 0) return false; localBot = decrement_index(localBot); @@ -512,7 +534,7 @@ // If there's still at least one element in the queue, based on the // "_bottom" and "age" we've read, then there can be no interference with // a "pop_global" operation, and we're done. - juint tp = get_top(); + TAG_TYPE tp = get_top(); // XXX if (size(localBot, tp) > 0) { assert(dirty_size(localBot, tp) != n() - 1, "Shouldn't be possible..."); @@ -581,7 +603,7 @@ bool is_empty(); bool stealable_is_empty(); bool overflow_is_empty(); - juint stealable_size() { return _region_queue.size(); } + uint stealable_size() { return _region_queue.size(); } RegionTaskQueue* task_queue() { return &_region_queue; } };
--- a/src/share/vm/utilities/vmError.cpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/utilities/vmError.cpp Fri Feb 27 15:13:00 2009 -0800 @@ -674,6 +674,11 @@ reset_signal_handlers(); } else { + // If UseOsErrorReporting we call this for each level of the call stack + // while searching for the exception handler. Only the first level needs + // to be reported. + if (UseOSErrorReporting && log_done) return; + // This is not the first error, see if it happened in a different thread // or in the same thread during error reporting. if (first_error_tid != mytid) {
--- a/src/share/vm/utilities/vmError.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/utilities/vmError.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -50,7 +50,7 @@ // additional info for VM internal errors const char * _filename; - int _lineno; + size_t _lineno; // used by fatal error handler int _current_step;
--- a/src/share/vm/utilities/workgroup.hpp Sat Jan 31 17:19:42 2009 -0800 +++ b/src/share/vm/utilities/workgroup.hpp Fri Feb 27 15:13:00 2009 -0800 @@ -32,7 +32,7 @@ // An abstract task to be worked on by a gang. // You subclass this to supply your own work() method -class AbstractGangTask: public CHeapObj { +class AbstractGangTask VALUE_OBJ_CLASS_SPEC { public: // The abstract work method. // The argument tells you which member of the gang you are.
--- a/test/Makefile Sat Jan 31 17:19:42 2009 -0800 +++ b/test/Makefile Fri Feb 27 15:13:00 2009 -0800 @@ -28,9 +28,9 @@ # Get OS/ARCH specifics OSNAME = $(shell uname -s) -SLASH_JAVA = /java ifeq ($(OSNAME), SunOS) PLATFORM = solaris + SLASH_JAVA = /java ARCH = $(shell uname -p) ifeq ($(ARCH), i386) ARCH=i586 @@ -38,6 +38,7 @@ endif ifeq ($(OSNAME), Linux) PLATFORM = linux + SLASH_JAVA = /java ARCH = $(shell uname -m) ifeq ($(ARCH), i386) ARCH = i586 @@ -62,6 +63,10 @@ EXESUFFIX = .exe endif +ifdef ALT_SLASH_JAVA + SLASH_JAVA = $(ALT_SLASH_JAVA) +endif + # Utilities used CD = cd CP = cp
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6603011/Test.java Fri Feb 27 15:13:00 2009 -0800 @@ -0,0 +1,220 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + */ + +/** + * @test + * @bug 6603011 + * @summary long/int division by constant + * + * @run main/othervm -Xcomp -Xbatch -XX:-Inline Test + */ + +// +// -XX:-Inline is essential to this test so that verification functions +// divi, modi, divl and modl generate "plain" divides. +// -Xcomp -Xbatch are also useful to ensure the full range of +// dividend and divisor combinations are tested +// + +import java.net.*; + +class s { + static int divi(int dividend, int divisor) { return dividend / divisor; } + static int modi(int dividend, int divisor) { return dividend % divisor; } + static long divl(long dividend, long divisor) { return dividend / divisor; } + static long modl(long dividend, long divisor) { return dividend % divisor; } +} + +public class Test implements Runnable { + // Report verbose messages on failure; turn off to suppress + // too much output with gross numbers of failures. + static final boolean VERBOSE = true; + + // Initailize DIVISOR so that it is final in this class. + static final int DIVISOR; + static { + int value = 0; + try { + value = Integer.decode(System.getProperty("divisor")); + } catch (Throwable e) { + } + DIVISOR = value; + } + + // The methods of interest. We want the JIT to compile these + // and convert the divide into a multiply. + public int divbyI (int dividend) { return dividend / DIVISOR; } + public int modbyI (int dividend) { return dividend % DIVISOR; } + public long divbyL (long dividend) { return dividend / DIVISOR; } + public long modbyL (long dividend) { return dividend % DIVISOR; } + + public int divisor() { return DIVISOR; } + + public boolean checkI (int dividend) { + int quo = divbyI(dividend); + int rem = modbyI(dividend); + int quo0 = s.divi(dividend, divisor()); + int rem0 = s.modi(dividend, divisor()); + + if (quo != quo0 || rem != rem0) { + if (VERBOSE) { + System.out.println("Computed: " + dividend + " / " + divisor() + " = " + + quo + ", " + dividend + " % " + divisor() + " = " + rem ); + System.out.println("expected: " + dividend + " / " + divisor() + " = " + + quo0 + ", " + dividend + " % " + divisor() + " = " + rem0); + // Report sign of rem failure + if (rem != 0 && (rem ^ dividend) < 0) { + System.out.println(" rem & dividend have different signs"); + } + // Report range of rem failure + if (java.lang.Math.abs(rem) >= java.lang.Math.abs(divisor())) { + System.out.println(" remainder out of range"); + } + // Report quo/rem identity relationship failure + if ((quo * divisor()) + rem != dividend) { + System.out.println(" quotien/remainder invariant broken"); + } + } + return false; + } + return true; + } + + public boolean checkL (long dividend) { + long quo = divbyL(dividend); + long rem = modbyL(dividend); + long quo0 = s.divl(dividend, divisor()); + long rem0 = s.modl(dividend, divisor()); + + if (quo != quo0 || rem != rem0) { + if (VERBOSE) { + System.out.println(" " + dividend + " / " + divisor() + " = " + + quo + ", " + dividend + " % " + divisor() + " = " + rem); + // Report sign of rem failure + if (rem != 0 && (rem ^ dividend) < 0) { + System.out.println(" rem & dividend have different signs"); + } + // Report range of rem failure + if (java.lang.Math.abs(rem) >= java.lang.Math.abs(divisor())) { + System.out.println(" remainder out of range"); + } + // Report quo/rem identity relationship failure + if ((quo * divisor()) + rem != dividend) { + System.out.println(" (" + quo + " * " + divisor() + ") + " + rem + " != " + + dividend); + } + } + return false; + } + return true; + } + + public void run() { + // Don't try to divide by zero + if (divisor() == 0) return; + + // Range of dividends to check. Try dividends from start to end + // inclusive, as well as variations on those values as shifted + // left. + int start = -1024; + int end = 1024; + + // Test int division using a variety of dividends. + int wrong = 0; + int total = 0; + + outerloop: + for (int i = start; i <= end; i++) { + for (int s = 0; s < 32; s += 4) { + total++; + int dividend = i << s; + if (!checkI(dividend)) { + wrong++; + // Stop on the first failure + // break outerloop; + } + } + } + if (wrong > 0) { + System.out.println("divisor " + divisor() + ": " + + wrong + "/" + total + " wrong int divisions"); + } + + // Test long division using a variety of dividends. + wrong = 0; + total = 0; + + outerloop: + for (int i = start; i <= end; i++) { + for (int s = 0; s < 64; s += 4) { + total++; + long dividend = i << s; + if (!checkL(dividend)) { + wrong++; + // Stop on the first failure + // break outerloop; + } + } + } + if (wrong > 0) { + System.out.println("divisor " + divisor() + ": " + + wrong + "/" + total + " wrong long divisions"); + } + + } + + // Reload this class with the "divisor" property set to the input parameter. + // This allows the JIT to see q.DIVISOR as a final constant, and change + // any divisions or mod operations into multiplies. + public static void test_divisor(int divisor, + URLClassLoader apploader) throws Exception { + System.setProperty("divisor", "" + divisor); + ClassLoader loader = new URLClassLoader(apploader.getURLs(), + apploader.getParent()); + Class c = loader.loadClass("Test"); + Runnable r = (Runnable)c.newInstance(); + r.run(); + } + + public static void main(String[] args) throws Exception { + Class cl = Class.forName("Test"); + URLClassLoader apploader = (URLClassLoader)cl.getClassLoader(); + + + // Test every divisor between -100 and 100. + for (int i = -100; i <= 100; i++) { + test_divisor(i, apploader); + } + + // Try a few divisors outside the typical range. + // The values below have been observed in rt.jar. + test_divisor(101, apploader); + test_divisor(400, apploader); + test_divisor(1000, apploader); + test_divisor(3600, apploader); + test_divisor(9973, apploader); + test_divisor(86400, apploader); + test_divisor(1000000, apploader); + } + +}
--- a/test/compiler/6775880/Test.java Sat Jan 31 17:19:42 2009 -0800 +++ b/test/compiler/6775880/Test.java Fri Feb 27 15:13:00 2009 -0800 @@ -27,7 +27,7 @@ * @bug 6775880 * @summary EA +DeoptimizeALot: assert(mon_info->owner()->is_locked(),"object must be locked now") * @compile -source 1.4 -target 1.4 Test.java - * @run main/othervm -server -Xbatch -XX:+DoEscapeAnalysis -XX:+DeoptimizeALot -XX:CompileCommand=exclude,java.lang.AbstractStringBuilder::append Test + * @run main/othervm -XX:+IgnoreUnrecognizedVMOptions -Xbatch -XX:+DoEscapeAnalysis -XX:+DeoptimizeALot -XX:CompileCommand=exclude,java.lang.AbstractStringBuilder::append Test */ public class Test {
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6778657/Test.java Fri Feb 27 15:13:00 2009 -0800 @@ -0,0 +1,75 @@ +/* + * Copyright 2008 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/* + * @test + * @bug 6778657 + * @summary Casts in SharedRuntime::f2i, f2l, d2i and d2l rely on undefined C++ behaviour + */ + +public class Test { + public static void check_f2i(int expect) { + float check = expect; + check *= 2; + int actual = (int) check; + if (actual != expect) + throw new RuntimeException("expecting " + expect + ", got " + actual); + } + + public static void check_f2l(long expect) { + float check = expect; + check *= 2; + long actual = (long) check; + if (actual != expect) + throw new RuntimeException("expecting " + expect + ", got " + actual); + } + + public static void check_d2i(int expect) { + double check = expect; + check *= 2; + int actual = (int) check; + if (actual != expect) + throw new RuntimeException("expecting " + expect + ", got " + actual); + } + + public static void check_d2l(long expect) { + double check = expect; + check *= 2; + long actual = (long) check; + if (actual != expect) + throw new RuntimeException("expecting " + expect + ", got " + actual); + } + + public static void main(String[] args) { + check_f2i(Integer.MAX_VALUE); + check_f2i(Integer.MIN_VALUE); + check_f2l(Long.MAX_VALUE); + check_f2l(Long.MIN_VALUE); + check_d2i(Integer.MAX_VALUE); + check_d2i(Integer.MIN_VALUE); + check_d2l(Long.MAX_VALUE); + check_d2l(Long.MIN_VALUE); + } +} +
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6795161/Test.java Fri Feb 27 15:13:00 2009 -0800 @@ -0,0 +1,60 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/* + * @test + * @bug 6795161 + * @summary Escape analysis leads to data corruption + * @run main/othervm -server -Xcomp -XX:CompileOnly=Test -XX:+DoEscapeAnalysis Test + */ + +class Test_Class_1 { + static String var_1; + + static void badFunc(int size) + { + try { + for (int i = 0; i < 1; (new byte[size-i])[0] = 0, i++) {} + } catch (Exception e) { + // don't comment it out, it will lead to correct results ;) + //System.out.println("Got exception: " + e); + } + } +} + +public class Test { + static String var_1_copy = Test_Class_1.var_1; + + static byte var_check; + + public static void main(String[] args) + { + var_check = 1; + + Test_Class_1.badFunc(-1); + + System.out.println("EATester.var_check = " + Test.var_check + " (expected 1)\n"); + } +} +
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6795362/Test6795362.java Fri Feb 27 15:13:00 2009 -0800 @@ -0,0 +1,48 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + */ + +/** + * @test + * @bug 6795362 + * @summary 32bit server compiler leads to wrong results on solaris-x86 + * + * @run main/othervm -Xcomp -XX:CompileOnly=Test6795362.sub Test6795362 + */ + +public class Test6795362 { + public static void main(String[] args) + { + sub(); + + if (var_bad != 0) + throw new InternalError(var_bad + " != 0"); + } + + static long var_bad = -1L; + + static void sub() + { + var_bad >>= 65; + var_bad /= 65; + } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6799693/Test.java Fri Feb 27 15:13:00 2009 -0800 @@ -0,0 +1,47 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/* + * @test + * @bug 6799693 + * @summary Server compiler leads to data corruption when expression throws an Exception + * @run main/othervm -Xcomp -XX:CompileOnly=Test Test + */ + +public class Test { + static int var_bad = 1; + + public static void main(String[] args) + { + var_bad++; + + try { + for (int i = 0; i < 10; i++) (new byte[((byte)-1 << i)])[0] = 0; + } + catch (Exception e) { System.out.println("Got " + e); } + + System.out.println("Test.var_bad = " + var_bad + " (expected 2)\n"); + } +} +
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6800154/Test6800154.java Fri Feb 27 15:13:00 2009 -0800 @@ -0,0 +1,109 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + */ + +/** + * @test + * @bug 6800154 + * @summary Add comments to long_by_long_mulhi() for better understandability + * + * @run main/othervm -Xcomp -XX:CompileOnly=Test6800154.divcomp Test6800154 + */ + +import java.net.URLClassLoader; + +public class Test6800154 implements Runnable { + static final long[] DIVIDENDS = { + 0, + 1, + 2, + 1423487, + 4444441, + 4918923241323L, + -1, + -24351, + 0x3333, + 0x0000000080000000L, + 0x7fffffffffffffffL, + 0x8000000000000000L + }; + + static final long[] DIVISORS = { + 1, + 2, + 17, + 12342, + 24123, + 143444, + 123444442344L, + -1, + -2, + -4423423234231423L, + 0x0000000080000000L, + 0x7fffffffffffffffL, + 0x8000000000000000L + }; + + // Initialize DIVISOR so that it is final in this class. + static final long DIVISOR; + + static { + long value = 0; + try { + value = Long.decode(System.getProperty("divisor")); + } catch (Throwable e) { + } + DIVISOR = value; + } + + public static void main(String[] args) throws Exception + { + Class cl = Class.forName("Test6800154"); + URLClassLoader apploader = (URLClassLoader) cl.getClassLoader(); + + // Iterate over all divisors. + for (int i = 0; i < DIVISORS.length; i++) { + System.setProperty("divisor", "" + DIVISORS[i]); + ClassLoader loader = new URLClassLoader(apploader.getURLs(), apploader.getParent()); + Class c = loader.loadClass("Test6800154"); + Runnable r = (Runnable) c.newInstance(); + r.run(); + } + } + + public void run() + { + // Iterate over all dividends. + for (int i = 0; i < DIVIDENDS.length; i++) { + long dividend = DIVIDENDS[i]; + + long expected = divint(dividend); + long result = divcomp(dividend); + + if (result != expected) + throw new InternalError(dividend + " / " + DIVISOR + " failed: " + result + " != " + expected); + } + } + + static long divint(long a) { return a / DIVISOR; } + static long divcomp(long a) { return a / DIVISOR; } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6805724/Test6805724.java Fri Feb 27 15:13:00 2009 -0800 @@ -0,0 +1,80 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + */ + +/** + * @test + * @bug 6805724 + * @summary ModLNode::Ideal() generates functionally incorrect graph when divisor is any (2^k-1) constant. + * + * @run main/othervm -Xcomp -XX:CompileOnly=Test6805724.fcomp Test6805724 + */ + +import java.net.URLClassLoader; + +public class Test6805724 implements Runnable { + // Initialize DIVISOR so that it is final in this class. + static final long DIVISOR; // 2^k-1 constant + + static { + long value = 0; + try { + value = Long.decode(System.getProperty("divisor")); + } catch (Throwable t) { + // This one is required for the Class.forName() in main. + } + DIVISOR = value; + } + + static long fint(long x) { + return x % DIVISOR; + } + + static long fcomp(long x) { + return x % DIVISOR; + } + + public void run() { + long a = 0x617981E1L; + + long expected = fint(a); + long result = fcomp(a); + + if (result != expected) + throw new InternalError(result + " != " + expected); + } + + public static void main(String args[]) throws Exception { + Class cl = Class.forName("Test6805724"); + URLClassLoader apploader = (URLClassLoader) cl.getClassLoader(); + + // Iterate over all 2^k-1 divisors. + for (int k = 1; k < Long.SIZE; k++) { + long divisor = (1L << k) - 1; + System.setProperty("divisor", "" + divisor); + ClassLoader loader = new URLClassLoader(apploader.getURLs(), apploader.getParent()); + Class c = loader.loadClass("Test6805724"); + Runnable r = (Runnable) c.newInstance(); + r.run(); + } + } +}