duke@435: /* jiangli@3826: * Copyright (c) 2003, 2012, Oracle and/or its affiliates. All rights reserved. duke@435: * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. duke@435: * duke@435: * This code is free software; you can redistribute it and/or modify it duke@435: * under the terms of the GNU General Public License version 2 only, as duke@435: * published by the Free Software Foundation. duke@435: * duke@435: * This code is distributed in the hope that it will be useful, but WITHOUT duke@435: * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or duke@435: * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License duke@435: * version 2 for more details (a copy is included in the LICENSE file that duke@435: * accompanied this code). duke@435: * duke@435: * You should have received a copy of the GNU General Public License version duke@435: * 2 along with this work; if not, write to the Free Software Foundation, duke@435: * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. duke@435: * trims@1907: * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA trims@1907: * or visit www.oracle.com if you need additional information or have any trims@1907: * questions. duke@435: * duke@435: */ duke@435: stefank@2314: #include "precompiled.hpp" twisti@4318: #include "asm/macroAssembler.hpp" stefank@2314: #include "interpreter/bytecodeHistogram.hpp" stefank@2314: #include "interpreter/interpreter.hpp" stefank@2314: #include "interpreter/interpreterGenerator.hpp" stefank@2314: #include "interpreter/interpreterRuntime.hpp" stefank@2314: #include "interpreter/templateTable.hpp" stefank@2314: #include "oops/arrayOop.hpp" coleenp@4037: #include "oops/methodData.hpp" coleenp@4037: #include "oops/method.hpp" stefank@2314: #include "oops/oop.inline.hpp" stefank@2314: #include "prims/jvmtiExport.hpp" stefank@2314: #include "prims/jvmtiThreadState.hpp" stefank@2314: #include "runtime/arguments.hpp" stefank@2314: #include "runtime/deoptimization.hpp" stefank@2314: #include "runtime/frame.inline.hpp" stefank@2314: #include "runtime/sharedRuntime.hpp" stefank@2314: #include "runtime/stubRoutines.hpp" stefank@2314: #include "runtime/synchronizer.hpp" stefank@2314: #include "runtime/timer.hpp" stefank@2314: #include "runtime/vframeArray.hpp" stefank@2314: #include "utilities/debug.hpp" jprovino@4542: #include "utilities/macros.hpp" duke@435: duke@435: #define __ _masm-> duke@435: never@739: #ifndef CC_INTERP never@739: duke@435: const int method_offset = frame::interpreter_frame_method_offset * wordSize; duke@435: const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize; duke@435: const int locals_offset = frame::interpreter_frame_locals_offset * wordSize; duke@435: duke@435: //----------------------------------------------------------------------------- duke@435: duke@435: address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { duke@435: address entry = __ pc(); duke@435: duke@435: #ifdef ASSERT duke@435: { duke@435: Label L; never@739: __ lea(rax, Address(rbp, never@739: frame::interpreter_frame_monitor_block_top_offset * never@739: wordSize)); never@739: __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack never@739: // grows negative) duke@435: __ jcc(Assembler::aboveEqual, L); // check if frame is complete duke@435: __ stop ("interpreter frame not set up"); duke@435: __ bind(L); duke@435: } duke@435: #endif // ASSERT duke@435: // Restore bcp under the assumption that the current frame is still duke@435: // interpreted duke@435: __ restore_bcp(); duke@435: duke@435: // expression stack must be empty before entering the VM if an duke@435: // exception happened duke@435: __ empty_expression_stack(); duke@435: // throw exception duke@435: __ call_VM(noreg, duke@435: CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime::throw_StackOverflowError)); duke@435: return entry; duke@435: } duke@435: duke@435: address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler( duke@435: const char* name) { duke@435: address entry = __ pc(); duke@435: // expression stack must be empty before entering the VM if an duke@435: // exception happened duke@435: __ empty_expression_stack(); duke@435: // setup parameters duke@435: // ??? convention: expect aberrant index in register ebx duke@435: __ lea(c_rarg1, ExternalAddress((address)name)); duke@435: __ call_VM(noreg, duke@435: CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime:: duke@435: throw_ArrayIndexOutOfBoundsException), duke@435: c_rarg1, rbx); duke@435: return entry; duke@435: } duke@435: duke@435: address TemplateInterpreterGenerator::generate_ClassCastException_handler() { duke@435: address entry = __ pc(); duke@435: duke@435: // object is at TOS never@739: __ pop(c_rarg1); duke@435: duke@435: // expression stack must be empty before entering the VM if an duke@435: // exception happened duke@435: __ empty_expression_stack(); duke@435: duke@435: __ call_VM(noreg, duke@435: CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime:: duke@435: throw_ClassCastException), duke@435: c_rarg1); duke@435: return entry; duke@435: } duke@435: duke@435: address TemplateInterpreterGenerator::generate_exception_handler_common( duke@435: const char* name, const char* message, bool pass_oop) { duke@435: assert(!pass_oop || message == NULL, "either oop or message but not both"); duke@435: address entry = __ pc(); duke@435: if (pass_oop) { duke@435: // object is at TOS never@739: __ pop(c_rarg2); duke@435: } duke@435: // expression stack must be empty before entering the VM if an duke@435: // exception happened duke@435: __ empty_expression_stack(); duke@435: // setup parameters duke@435: __ lea(c_rarg1, ExternalAddress((address)name)); duke@435: if (pass_oop) { duke@435: __ call_VM(rax, CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime:: duke@435: create_klass_exception), duke@435: c_rarg1, c_rarg2); duke@435: } else { duke@435: // kind of lame ExternalAddress can't take NULL because duke@435: // external_word_Relocation will assert. duke@435: if (message != NULL) { duke@435: __ lea(c_rarg2, ExternalAddress((address)message)); duke@435: } else { duke@435: __ movptr(c_rarg2, NULL_WORD); duke@435: } duke@435: __ call_VM(rax, duke@435: CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), duke@435: c_rarg1, c_rarg2); duke@435: } duke@435: // throw exception duke@435: __ jump(ExternalAddress(Interpreter::throw_exception_entry())); duke@435: return entry; duke@435: } duke@435: duke@435: duke@435: address TemplateInterpreterGenerator::generate_continuation_for(TosState state) { duke@435: address entry = __ pc(); duke@435: // NULL last_sp until next java call never@739: __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); duke@435: __ dispatch_next(state); duke@435: return entry; duke@435: } duke@435: duke@435: twisti@2552: address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) { duke@435: address entry = __ pc(); duke@435: duke@435: // Restore stack bottom in case i2c adjusted stack never@739: __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); duke@435: // and NULL it as marker that esp is now tos until next java call never@739: __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); duke@435: duke@435: __ restore_bcp(); duke@435: __ restore_locals(); never@739: twisti@1543: Label L_got_cache, L_giant_index; twisti@1543: if (EnableInvokeDynamic) { twisti@1543: __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic); twisti@1543: __ jcc(Assembler::equal, L_giant_index); twisti@1543: } jrose@1920: __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2)); twisti@1543: __ bind(L_got_cache); duke@435: __ movl(rbx, Address(rbx, rcx, twisti@1543: Address::times_ptr, coleenp@4037: in_bytes(ConstantPoolCache::base_offset()) + duke@435: 3 * wordSize)); duke@435: __ andl(rbx, 0xFF); never@739: __ lea(rsp, Address(rsp, rbx, Address::times_8)); duke@435: __ dispatch_next(state, step); twisti@1543: twisti@1543: // out of the main line of code... twisti@1543: if (EnableInvokeDynamic) { twisti@1543: __ bind(L_giant_index); jrose@1920: __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4)); twisti@1543: __ jmp(L_got_cache); twisti@1543: } twisti@1543: duke@435: return entry; duke@435: } duke@435: duke@435: duke@435: address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, duke@435: int step) { duke@435: address entry = __ pc(); duke@435: // NULL last_sp until next java call never@739: __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); duke@435: __ restore_bcp(); duke@435: __ restore_locals(); duke@435: // handle exceptions duke@435: { duke@435: Label L; never@739: __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD); duke@435: __ jcc(Assembler::zero, L); duke@435: __ call_VM(noreg, duke@435: CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime::throw_pending_exception)); duke@435: __ should_not_reach_here(); duke@435: __ bind(L); duke@435: } duke@435: __ dispatch_next(state, step); duke@435: return entry; duke@435: } duke@435: duke@435: int AbstractInterpreter::BasicType_as_index(BasicType type) { duke@435: int i = 0; duke@435: switch (type) { duke@435: case T_BOOLEAN: i = 0; break; duke@435: case T_CHAR : i = 1; break; duke@435: case T_BYTE : i = 2; break; duke@435: case T_SHORT : i = 3; break; duke@435: case T_INT : i = 4; break; duke@435: case T_LONG : i = 5; break; duke@435: case T_VOID : i = 6; break; duke@435: case T_FLOAT : i = 7; break; duke@435: case T_DOUBLE : i = 8; break; duke@435: case T_OBJECT : i = 9; break; duke@435: case T_ARRAY : i = 9; break; duke@435: default : ShouldNotReachHere(); duke@435: } duke@435: assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, duke@435: "index out of bounds"); duke@435: return i; duke@435: } duke@435: duke@435: duke@435: address TemplateInterpreterGenerator::generate_result_handler_for( duke@435: BasicType type) { duke@435: address entry = __ pc(); duke@435: switch (type) { duke@435: case T_BOOLEAN: __ c2bool(rax); break; duke@435: case T_CHAR : __ movzwl(rax, rax); break; duke@435: case T_BYTE : __ sign_extend_byte(rax); break; duke@435: case T_SHORT : __ sign_extend_short(rax); break; duke@435: case T_INT : /* nothing to do */ break; duke@435: case T_LONG : /* nothing to do */ break; duke@435: case T_VOID : /* nothing to do */ break; duke@435: case T_FLOAT : /* nothing to do */ break; duke@435: case T_DOUBLE : /* nothing to do */ break; duke@435: case T_OBJECT : duke@435: // retrieve result from frame never@739: __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize)); duke@435: // and verify it duke@435: __ verify_oop(rax); duke@435: break; duke@435: default : ShouldNotReachHere(); duke@435: } duke@435: __ ret(0); // return from result handler duke@435: return entry; duke@435: } duke@435: duke@435: address TemplateInterpreterGenerator::generate_safept_entry_for( duke@435: TosState state, duke@435: address runtime_entry) { duke@435: address entry = __ pc(); duke@435: __ push(state); duke@435: __ call_VM(noreg, runtime_entry); duke@435: __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); duke@435: return entry; duke@435: } duke@435: duke@435: duke@435: duke@435: // Helpers for commoning out cases in the various type of method entries. duke@435: // duke@435: duke@435: duke@435: // increment invocation count & check for overflow duke@435: // duke@435: // Note: checking for negative value instead of overflow duke@435: // so we have a 'sticky' overflow test duke@435: // duke@435: // rbx: method duke@435: // ecx: invocation counter duke@435: // duke@435: void InterpreterGenerator::generate_counter_incr( duke@435: Label* overflow, duke@435: Label* profile_method, duke@435: Label* profile_method_continue) { coleenp@4037: const Address invocation_counter(rbx, in_bytes(Method::invocation_counter_offset()) + iveresov@2138: in_bytes(InvocationCounter::counter_offset())); coleenp@4037: // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not. iveresov@2138: if (TieredCompilation) { iveresov@2138: int increment = InvocationCounter::count_increment; iveresov@2138: int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift; iveresov@2138: Label no_mdo, done; iveresov@2138: if (ProfileInterpreter) { iveresov@2138: // Are we profiling? coleenp@4037: __ movptr(rax, Address(rbx, Method::method_data_offset())); iveresov@2138: __ testptr(rax, rax); iveresov@2138: __ jccb(Assembler::zero, no_mdo); iveresov@2138: // Increment counter in the MDO coleenp@4037: const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) + iveresov@2138: in_bytes(InvocationCounter::counter_offset())); iveresov@2138: __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow); iveresov@2138: __ jmpb(done); iveresov@2138: } iveresov@2138: __ bind(no_mdo); coleenp@4037: // Increment counter in Method* (we don't need to load it, it's in ecx). iveresov@2138: __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow); iveresov@2138: __ bind(done); iveresov@2138: } else { iveresov@2138: const Address backedge_counter(rbx, coleenp@4037: Method::backedge_counter_offset() + iveresov@2138: InvocationCounter::counter_offset()); duke@435: coleenp@4037: if (ProfileInterpreter) { // %%% Merge this into MethodData* iveresov@2138: __ incrementl(Address(rbx, coleenp@4037: Method::interpreter_invocation_counter_offset())); iveresov@2138: } iveresov@2138: // Update standard invocation counters iveresov@2138: __ movl(rax, backedge_counter); // load backedge counter duke@435: iveresov@2138: __ incrementl(rcx, InvocationCounter::count_increment); iveresov@2138: __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits iveresov@2138: iveresov@2138: __ movl(invocation_counter, rcx); // save invocation count iveresov@2138: __ addl(rcx, rax); // add both counters iveresov@2138: iveresov@2138: // profile_method is non-null only for interpreted method so iveresov@2138: // profile_method != NULL == !native_call iveresov@2138: iveresov@2138: if (ProfileInterpreter && profile_method != NULL) { iveresov@2138: // Test to see if we should create a method data oop iveresov@2138: __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit)); iveresov@2138: __ jcc(Assembler::less, *profile_method_continue); iveresov@2138: iveresov@2138: // if no method data exists, go to profile_method iveresov@2138: __ test_method_data_pointer(rax, *profile_method); iveresov@2138: } iveresov@2138: iveresov@2138: __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit)); iveresov@2138: __ jcc(Assembler::aboveEqual, *overflow); duke@435: } duke@435: } duke@435: duke@435: void InterpreterGenerator::generate_counter_overflow(Label* do_continue) { duke@435: duke@435: // Asm interpreter on entry duke@435: // r14 - locals duke@435: // r13 - bcp duke@435: // rbx - method duke@435: // edx - cpool --- DOES NOT APPEAR TO BE TRUE duke@435: // rbp - interpreter frame duke@435: duke@435: // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] duke@435: // Everything as it was on entry duke@435: // rdx is not restored. Doesn't appear to really be set. duke@435: duke@435: // InterpreterRuntime::frequency_counter_overflow takes two duke@435: // arguments, the first (thread) is passed by call_VM, the second duke@435: // indicates if the counter overflow occurs at a backwards branch duke@435: // (NULL bcp). We pass zero for it. The call returns the address duke@435: // of the verified entry point for the method or NULL if the duke@435: // compilation did not complete (either went background or bailed duke@435: // out). duke@435: __ movl(c_rarg1, 0); duke@435: __ call_VM(noreg, duke@435: CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime::frequency_counter_overflow), duke@435: c_rarg1); duke@435: coleenp@4037: __ movptr(rbx, Address(rbp, method_offset)); // restore Method* duke@435: // Preserve invariant that r13/r14 contain bcp/locals of sender frame duke@435: // and jump to the interpreted entry. duke@435: __ jmp(*do_continue, relocInfo::none); duke@435: } duke@435: duke@435: // See if we've got enough room on the stack for locals plus overhead. duke@435: // The expression stack grows down incrementally, so the normal guard duke@435: // page mechanism will work for that. duke@435: // duke@435: // NOTE: Since the additional locals are also always pushed (wasn't duke@435: // obvious in generate_method_entry) so the guard should work for them duke@435: // too. duke@435: // duke@435: // Args: duke@435: // rdx: number of additional locals this frame needs (what we must check) coleenp@4037: // rbx: Method* duke@435: // duke@435: // Kills: duke@435: // rax duke@435: void InterpreterGenerator::generate_stack_overflow_check(void) { duke@435: duke@435: // monitor entry size: see picture of stack set duke@435: // (generate_method_entry) and frame_amd64.hpp duke@435: const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; duke@435: duke@435: // total overhead size: entry_size + (saved rbp through expr stack duke@435: // bottom). be sure to change this if you add/subtract anything duke@435: // to/from the overhead area duke@435: const int overhead_size = duke@435: -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size; duke@435: duke@435: const int page_size = os::vm_page_size(); duke@435: duke@435: Label after_frame_check; duke@435: duke@435: // see if the frame is greater than one page in size. If so, duke@435: // then we need to verify there is enough stack space remaining duke@435: // for the additional locals. twisti@1861: __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize); duke@435: __ jcc(Assembler::belowEqual, after_frame_check); duke@435: duke@435: // compute rsp as if this were going to be the last frame on duke@435: // the stack before the red zone duke@435: duke@435: const Address stack_base(r15_thread, Thread::stack_base_offset()); duke@435: const Address stack_size(r15_thread, Thread::stack_size_offset()); duke@435: duke@435: // locals + overhead, in bytes never@739: __ mov(rax, rdx); twisti@1861: __ shlptr(rax, Interpreter::logStackElementSize); // 2 slots per parameter. never@739: __ addptr(rax, overhead_size); duke@435: duke@435: #ifdef ASSERT duke@435: Label stack_base_okay, stack_size_okay; duke@435: // verify that thread stack base is non-zero never@739: __ cmpptr(stack_base, (int32_t)NULL_WORD); duke@435: __ jcc(Assembler::notEqual, stack_base_okay); duke@435: __ stop("stack base is zero"); duke@435: __ bind(stack_base_okay); duke@435: // verify that thread stack size is non-zero never@739: __ cmpptr(stack_size, 0); duke@435: __ jcc(Assembler::notEqual, stack_size_okay); duke@435: __ stop("stack size is zero"); duke@435: __ bind(stack_size_okay); duke@435: #endif duke@435: duke@435: // Add stack base to locals and subtract stack size never@739: __ addptr(rax, stack_base); never@739: __ subptr(rax, stack_size); duke@435: twisti@1570: // Use the maximum number of pages we might bang. twisti@1570: const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages : twisti@1570: (StackRedPages+StackYellowPages); twisti@1570: duke@435: // add in the red and yellow zone sizes twisti@1570: __ addptr(rax, max_pages * page_size); duke@435: duke@435: // check against the current stack bottom never@739: __ cmpptr(rsp, rax); duke@435: __ jcc(Assembler::above, after_frame_check); duke@435: bdelsart@3372: // Restore sender's sp as SP. This is necessary if the sender's bdelsart@3372: // frame is an extended compiled frame (see gen_c2i_adapter()) bdelsart@3372: // and safer anyway in case of JSR292 adaptations. bdelsart@3372: bdelsart@3372: __ pop(rax); // return address must be moved if SP is changed bdelsart@3372: __ mov(rsp, r13); bdelsart@3372: __ push(rax); bdelsart@3372: bdelsart@3372: // Note: the restored frame is not necessarily interpreted. bdelsart@3372: // Use the shared runtime version of the StackOverflowError. bdelsart@3372: assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated"); bdelsart@3372: __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry())); duke@435: duke@435: // all done with frame size check duke@435: __ bind(after_frame_check); duke@435: } duke@435: duke@435: // Allocate monitor and lock method (asm interpreter) duke@435: // duke@435: // Args: coleenp@4037: // rbx: Method* duke@435: // r14: locals duke@435: // duke@435: // Kills: duke@435: // rax duke@435: // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs) duke@435: // rscratch1, rscratch2 (scratch regs) duke@435: void InterpreterGenerator::lock_method(void) { duke@435: // synchronize method coleenp@4037: const Address access_flags(rbx, Method::access_flags_offset()); duke@435: const Address monitor_block_top( duke@435: rbp, duke@435: frame::interpreter_frame_monitor_block_top_offset * wordSize); duke@435: const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; duke@435: duke@435: #ifdef ASSERT duke@435: { duke@435: Label L; duke@435: __ movl(rax, access_flags); duke@435: __ testl(rax, JVM_ACC_SYNCHRONIZED); duke@435: __ jcc(Assembler::notZero, L); duke@435: __ stop("method doesn't need synchronization"); duke@435: __ bind(L); duke@435: } duke@435: #endif // ASSERT duke@435: duke@435: // get synchronization object duke@435: { stefank@3391: const int mirror_offset = in_bytes(Klass::java_mirror_offset()); duke@435: Label done; duke@435: __ movl(rax, access_flags); duke@435: __ testl(rax, JVM_ACC_STATIC); duke@435: // get receiver (assume this is frequent case) never@739: __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0))); duke@435: __ jcc(Assembler::zero, done); coleenp@4037: __ movptr(rax, Address(rbx, Method::const_offset())); coleenp@4037: __ movptr(rax, Address(rax, ConstMethod::constants_offset())); never@739: __ movptr(rax, Address(rax, coleenp@4037: ConstantPool::pool_holder_offset_in_bytes())); never@739: __ movptr(rax, Address(rax, mirror_offset)); duke@435: duke@435: #ifdef ASSERT duke@435: { duke@435: Label L; never@739: __ testptr(rax, rax); duke@435: __ jcc(Assembler::notZero, L); duke@435: __ stop("synchronization object is NULL"); duke@435: __ bind(L); duke@435: } duke@435: #endif // ASSERT duke@435: duke@435: __ bind(done); duke@435: } duke@435: duke@435: // add space for monitor & lock never@739: __ subptr(rsp, entry_size); // add space for a monitor entry never@739: __ movptr(monitor_block_top, rsp); // set new monitor block top duke@435: // store object never@739: __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); never@739: __ movptr(c_rarg1, rsp); // object address duke@435: __ lock_object(c_rarg1); duke@435: } duke@435: duke@435: // Generate a fixed interpreter frame. This is identical setup for duke@435: // interpreted methods and for native methods hence the shared code. duke@435: // duke@435: // Args: duke@435: // rax: return address coleenp@4037: // rbx: Method* duke@435: // r14: pointer to locals duke@435: // r13: sender sp duke@435: // rdx: cp cache duke@435: void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { duke@435: // initialize fixed part of activation frame never@739: __ push(rax); // save return address duke@435: __ enter(); // save old & set new rbp never@739: __ push(r13); // set sender sp never@739: __ push((int)NULL_WORD); // leave last_sp as null coleenp@4037: __ movptr(r13, Address(rbx, Method::const_offset())); // get ConstMethod* coleenp@4037: __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase coleenp@4037: __ push(rbx); // save Method* duke@435: if (ProfileInterpreter) { duke@435: Label method_data_continue; coleenp@4037: __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset()))); never@739: __ testptr(rdx, rdx); duke@435: __ jcc(Assembler::zero, method_data_continue); coleenp@4037: __ addptr(rdx, in_bytes(MethodData::data_offset())); duke@435: __ bind(method_data_continue); never@739: __ push(rdx); // set the mdp (method data pointer) duke@435: } else { never@739: __ push(0); duke@435: } duke@435: coleenp@4037: __ movptr(rdx, Address(rbx, Method::const_offset())); coleenp@4037: __ movptr(rdx, Address(rdx, ConstMethod::constants_offset())); coleenp@4037: __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes())); never@739: __ push(rdx); // set constant pool cache never@739: __ push(r14); // set locals pointer duke@435: if (native_call) { never@739: __ push(0); // no bcp duke@435: } else { never@739: __ push(r13); // set bcp duke@435: } never@739: __ push(0); // reserve word for pointer to expression stack bottom never@739: __ movptr(Address(rsp, 0), rsp); // set expression stack bottom duke@435: } duke@435: duke@435: // End of helpers duke@435: never@739: // Various method entries never@739: //------------------------------------------------------------------------------------------------------------------------ never@739: // never@739: // never@739: never@739: // Call an accessor method (assuming it is resolved, otherwise drop never@739: // into vanilla (slow path) entry never@739: address InterpreterGenerator::generate_accessor_entry(void) { coleenp@4037: // rbx: Method* never@739: never@739: // r13: senderSP must preserver for slow path, set SP to it on fast path never@739: never@739: address entry_point = __ pc(); never@739: Label xreturn_path; never@739: never@739: // do fastpath for resolved accessor methods never@739: if (UseFastAccessorMethods) { never@739: // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites never@739: // thereof; parameter size = 1 never@739: // Note: We can only use this code if the getfield has been resolved never@739: // and if we don't have a null-pointer exception => check for never@739: // these conditions first and use slow path if necessary. never@739: Label slow_path; never@739: // If we need a safepoint check, generate full interpreter entry. never@739: __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), never@739: SafepointSynchronize::_not_synchronized); never@739: never@739: __ jcc(Assembler::notEqual, slow_path); never@739: // rbx: method never@739: __ movptr(rax, Address(rsp, wordSize)); never@739: never@739: // check if local 0 != NULL and read field never@739: __ testptr(rax, rax); never@739: __ jcc(Assembler::zero, slow_path); never@739: never@739: // read first instruction word and extract bytecode @ 1 and index @ 2 coleenp@4037: __ movptr(rdx, Address(rbx, Method::const_offset())); coleenp@4037: __ movptr(rdi, Address(rdx, ConstMethod::constants_offset())); coleenp@4037: __ movl(rdx, Address(rdx, ConstMethod::codes_offset())); never@739: // Shift codes right to get the index on the right. never@739: // The bytecode fetched looks like <0xb4><0x2a> never@739: __ shrl(rdx, 2 * BitsPerByte); never@739: __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); coleenp@4037: __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes())); never@739: never@739: // rax: local 0 never@739: // rbx: method never@739: // rdx: constant pool cache index never@739: // rdi: constant pool cache never@739: never@739: // check if getfield has been resolved and read constant pool cache entry never@739: // check the validity of the cache entry by testing whether _indices field never@739: // contains Bytecode::_getfield in b1 byte. never@739: assert(in_words(ConstantPoolCacheEntry::size()) == 4, never@739: "adjust shift below"); never@739: __ movl(rcx, never@739: Address(rdi, never@739: rdx, never@739: Address::times_8, coleenp@4037: ConstantPoolCache::base_offset() + never@739: ConstantPoolCacheEntry::indices_offset())); never@739: __ shrl(rcx, 2 * BitsPerByte); never@739: __ andl(rcx, 0xFF); never@739: __ cmpl(rcx, Bytecodes::_getfield); never@739: __ jcc(Assembler::notEqual, slow_path); never@739: never@739: // Note: constant pool entry is not valid before bytecode is resolved never@739: __ movptr(rcx, never@739: Address(rdi, never@739: rdx, never@739: Address::times_8, coleenp@4037: ConstantPoolCache::base_offset() + never@739: ConstantPoolCacheEntry::f2_offset())); never@739: // edx: flags never@739: __ movl(rdx, never@739: Address(rdi, never@739: rdx, never@739: Address::times_8, coleenp@4037: ConstantPoolCache::base_offset() + never@739: ConstantPoolCacheEntry::flags_offset())); never@739: never@739: Label notObj, notInt, notByte, notShort; never@739: const Address field_address(rax, rcx, Address::times_1); never@739: never@739: // Need to differentiate between igetfield, agetfield, bgetfield etc. never@739: // because they are different sizes. never@739: // Use the type from the constant pool cache twisti@3969: __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift); twisti@3969: // Make sure we don't need to mask edx after the above shift twisti@3969: ConstantPoolCacheEntry::verify_tos_state_shift(); never@739: never@739: __ cmpl(rdx, atos); never@739: __ jcc(Assembler::notEqual, notObj); never@739: // atos never@739: __ load_heap_oop(rax, field_address); never@739: __ jmp(xreturn_path); never@739: never@739: __ bind(notObj); never@739: __ cmpl(rdx, itos); never@739: __ jcc(Assembler::notEqual, notInt); never@739: // itos never@739: __ movl(rax, field_address); never@739: __ jmp(xreturn_path); never@739: never@739: __ bind(notInt); never@739: __ cmpl(rdx, btos); never@739: __ jcc(Assembler::notEqual, notByte); never@739: // btos never@739: __ load_signed_byte(rax, field_address); never@739: __ jmp(xreturn_path); never@739: never@739: __ bind(notByte); never@739: __ cmpl(rdx, stos); never@739: __ jcc(Assembler::notEqual, notShort); never@739: // stos jrose@1057: __ load_signed_short(rax, field_address); never@739: __ jmp(xreturn_path); never@739: never@739: __ bind(notShort); never@739: #ifdef ASSERT never@739: Label okay; never@739: __ cmpl(rdx, ctos); never@739: __ jcc(Assembler::equal, okay); never@739: __ stop("what type is this?"); never@739: __ bind(okay); never@739: #endif never@739: // ctos jrose@1057: __ load_unsigned_short(rax, field_address); never@739: never@739: __ bind(xreturn_path); never@739: never@739: // _ireturn/_areturn never@739: __ pop(rdi); never@739: __ mov(rsp, r13); never@739: __ jmp(rdi); never@739: __ ret(0); never@739: never@739: // generate a vanilla interpreter entry as the slow path never@739: __ bind(slow_path); never@739: (void) generate_normal_entry(false); never@739: } else { never@739: (void) generate_normal_entry(false); never@739: } never@739: never@739: return entry_point; never@739: } never@739: johnc@2781: // Method entry for java.lang.ref.Reference.get. johnc@2781: address InterpreterGenerator::generate_Reference_get_entry(void) { jprovino@4542: #if INCLUDE_ALL_GCS johnc@2781: // Code: _aload_0, _getfield, _areturn johnc@2781: // parameter size = 1 johnc@2781: // johnc@2781: // The code that gets generated by this routine is split into 2 parts: johnc@2781: // 1. The "intrinsified" code for G1 (or any SATB based GC), johnc@2781: // 2. The slow path - which is an expansion of the regular method entry. johnc@2781: // johnc@2781: // Notes:- johnc@2781: // * In the G1 code we do not check whether we need to block for johnc@2781: // a safepoint. If G1 is enabled then we must execute the specialized johnc@2781: // code for Reference.get (except when the Reference object is null) johnc@2781: // so that we can log the value in the referent field with an SATB johnc@2781: // update buffer. johnc@2781: // If the code for the getfield template is modified so that the johnc@2781: // G1 pre-barrier code is executed when the current method is johnc@2781: // Reference.get() then going through the normal method entry johnc@2781: // will be fine. johnc@2781: // * The G1 code can, however, check the receiver object (the instance johnc@2781: // of java.lang.Reference) and jump to the slow path if null. If the johnc@2781: // Reference object is null then we obviously cannot fetch the referent johnc@2781: // and so we don't need to call the G1 pre-barrier. Thus we can use the johnc@2781: // regular method entry code to generate the NPE. johnc@2781: // johnc@2781: // This code is based on generate_accessor_enty. johnc@2781: // coleenp@4037: // rbx: Method* johnc@2781: johnc@2781: // r13: senderSP must preserve for slow path, set SP to it on fast path johnc@2781: johnc@2781: address entry = __ pc(); johnc@2781: johnc@2781: const int referent_offset = java_lang_ref_Reference::referent_offset; johnc@2781: guarantee(referent_offset > 0, "referent offset not initialized"); johnc@2781: johnc@2781: if (UseG1GC) { johnc@2781: Label slow_path; johnc@2781: // rbx: method johnc@2781: johnc@2781: // Check if local 0 != NULL johnc@2781: // If the receiver is null then it is OK to jump to the slow path. johnc@2781: __ movptr(rax, Address(rsp, wordSize)); johnc@2781: johnc@2781: __ testptr(rax, rax); johnc@2781: __ jcc(Assembler::zero, slow_path); johnc@2781: johnc@2781: // rax: local 0 johnc@2781: // rbx: method (but can be used as scratch now) johnc@2781: // rdx: scratch johnc@2781: // rdi: scratch johnc@2781: johnc@2781: // Generate the G1 pre-barrier code to log the value of johnc@2781: // the referent field in an SATB buffer. johnc@2781: johnc@2781: // Load the value of the referent field. johnc@2781: const Address field_address(rax, referent_offset); johnc@2781: __ load_heap_oop(rax, field_address); johnc@2781: johnc@2781: // Generate the G1 pre-barrier code to log the value of johnc@2781: // the referent field in an SATB buffer. johnc@2781: __ g1_write_barrier_pre(noreg /* obj */, johnc@2781: rax /* pre_val */, johnc@2781: r15_thread /* thread */, johnc@2781: rbx /* tmp */, johnc@2781: true /* tosca_live */, johnc@2781: true /* expand_call */); johnc@2781: johnc@2781: // _areturn johnc@2781: __ pop(rdi); // get return address johnc@2781: __ mov(rsp, r13); // set sp to sender sp johnc@2781: __ jmp(rdi); johnc@2781: __ ret(0); johnc@2781: johnc@2781: // generate a vanilla interpreter entry as the slow path johnc@2781: __ bind(slow_path); johnc@2781: (void) generate_normal_entry(false); johnc@2781: johnc@2781: return entry; johnc@2781: } jprovino@4542: #endif // INCLUDE_ALL_GCS johnc@2781: johnc@2781: // If G1 is not enabled then attempt to go through the accessor entry point johnc@2781: // Reference.get is an accessor johnc@2781: return generate_accessor_entry(); johnc@2781: } johnc@2781: johnc@2781: duke@435: // Interpreter stub for calling a native method. (asm interpreter) duke@435: // This sets up a somewhat different looking stack for calling the duke@435: // native method than the typical interpreter frame setup. duke@435: address InterpreterGenerator::generate_native_entry(bool synchronized) { duke@435: // determine code generation flags duke@435: bool inc_counter = UseCompiler || CountCompiledCalls; duke@435: coleenp@4037: // rbx: Method* duke@435: // r13: sender sp duke@435: duke@435: address entry_point = __ pc(); duke@435: jiangli@4338: const Address constMethod (rbx, Method::const_offset()); coleenp@4037: const Address invocation_counter(rbx, Method:: duke@435: invocation_counter_offset() + duke@435: InvocationCounter::counter_offset()); coleenp@4037: const Address access_flags (rbx, Method::access_flags_offset()); jiangli@4338: const Address size_of_parameters(rcx, ConstMethod:: jiangli@4338: size_of_parameters_offset()); jiangli@4338: duke@435: duke@435: // get parameter size (always needed) jiangli@4338: __ movptr(rcx, constMethod); jrose@1057: __ load_unsigned_short(rcx, size_of_parameters); duke@435: duke@435: // native calls don't need the stack size check since they have no duke@435: // expression stack and the arguments are already on the stack and duke@435: // we only add a handful of words to the stack duke@435: coleenp@4037: // rbx: Method* duke@435: // rcx: size of parameters duke@435: // r13: sender sp never@739: __ pop(rax); // get return address duke@435: duke@435: // for natives the size of locals is zero duke@435: duke@435: // compute beginning of parameters (r14) never@739: __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize)); duke@435: duke@435: // add 2 zero-initialized slots for native calls duke@435: // initialize result_handler slot never@739: __ push((int) NULL_WORD); duke@435: // slot for oop temp duke@435: // (static native method holder mirror/jni oop result) never@739: __ push((int) NULL_WORD); duke@435: duke@435: if (inc_counter) { duke@435: __ movl(rcx, invocation_counter); // (pre-)fetch invocation count duke@435: } duke@435: duke@435: // initialize fixed part of activation frame duke@435: generate_fixed_frame(true); duke@435: duke@435: // make sure method is native & not abstract duke@435: #ifdef ASSERT duke@435: __ movl(rax, access_flags); duke@435: { duke@435: Label L; duke@435: __ testl(rax, JVM_ACC_NATIVE); duke@435: __ jcc(Assembler::notZero, L); duke@435: __ stop("tried to execute non-native method as native"); duke@435: __ bind(L); duke@435: } duke@435: { duke@435: Label L; duke@435: __ testl(rax, JVM_ACC_ABSTRACT); duke@435: __ jcc(Assembler::zero, L); duke@435: __ stop("tried to execute abstract method in interpreter"); duke@435: __ bind(L); duke@435: } duke@435: #endif duke@435: duke@435: // Since at this point in the method invocation the exception handler duke@435: // would try to exit the monitor of synchronized methods which hasn't duke@435: // been entered yet, we set the thread local variable duke@435: // _do_not_unlock_if_synchronized to true. The remove_activation will duke@435: // check this flag. duke@435: duke@435: const Address do_not_unlock_if_synchronized(r15_thread, duke@435: in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); duke@435: __ movbool(do_not_unlock_if_synchronized, true); duke@435: duke@435: // increment invocation count & check for overflow duke@435: Label invocation_counter_overflow; duke@435: if (inc_counter) { duke@435: generate_counter_incr(&invocation_counter_overflow, NULL, NULL); duke@435: } duke@435: duke@435: Label continue_after_compile; duke@435: __ bind(continue_after_compile); duke@435: duke@435: bang_stack_shadow_pages(true); duke@435: duke@435: // reset the _do_not_unlock_if_synchronized flag duke@435: __ movbool(do_not_unlock_if_synchronized, false); duke@435: duke@435: // check for synchronized methods duke@435: // Must happen AFTER invocation_counter check and stack overflow check, duke@435: // so method is not locked if overflows. duke@435: if (synchronized) { duke@435: lock_method(); duke@435: } else { duke@435: // no synchronization necessary duke@435: #ifdef ASSERT duke@435: { duke@435: Label L; duke@435: __ movl(rax, access_flags); duke@435: __ testl(rax, JVM_ACC_SYNCHRONIZED); duke@435: __ jcc(Assembler::zero, L); duke@435: __ stop("method needs synchronization"); duke@435: __ bind(L); duke@435: } duke@435: #endif duke@435: } duke@435: duke@435: // start execution duke@435: #ifdef ASSERT duke@435: { duke@435: Label L; duke@435: const Address monitor_block_top(rbp, duke@435: frame::interpreter_frame_monitor_block_top_offset * wordSize); never@739: __ movptr(rax, monitor_block_top); never@739: __ cmpptr(rax, rsp); duke@435: __ jcc(Assembler::equal, L); duke@435: __ stop("broken stack frame setup in interpreter"); duke@435: __ bind(L); duke@435: } duke@435: #endif duke@435: duke@435: // jvmti support duke@435: __ notify_method_entry(); duke@435: duke@435: // work registers duke@435: const Register method = rbx; coleenp@548: const Register t = r11; duke@435: duke@435: // allocate space for parameters duke@435: __ get_method(method); jiangli@4338: __ movptr(t, Address(method, Method::const_offset())); jiangli@4338: __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset())); twisti@1861: __ shll(t, Interpreter::logStackElementSize); duke@435: never@739: __ subptr(rsp, t); never@739: __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows twisti@1040: __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI) duke@435: duke@435: // get signature handler duke@435: { duke@435: Label L; coleenp@4037: __ movptr(t, Address(method, Method::signature_handler_offset())); never@739: __ testptr(t, t); duke@435: __ jcc(Assembler::notZero, L); duke@435: __ call_VM(noreg, duke@435: CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime::prepare_native_call), duke@435: method); duke@435: __ get_method(method); coleenp@4037: __ movptr(t, Address(method, Method::signature_handler_offset())); duke@435: __ bind(L); duke@435: } duke@435: duke@435: // call signature handler duke@435: assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14, duke@435: "adjust this code"); duke@435: assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp, duke@435: "adjust this code"); duke@435: assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1, duke@435: "adjust this code"); duke@435: duke@435: // The generated handlers do not touch RBX (the method oop). duke@435: // However, large signatures cannot be cached and are generated duke@435: // each time here. The slow-path generator can do a GC on return, duke@435: // so we must reload it after the call. duke@435: __ call(t); duke@435: __ get_method(method); // slow path can do a GC, reload RBX duke@435: duke@435: duke@435: // result handler is in rax duke@435: // set result handler never@739: __ movptr(Address(rbp, never@739: (frame::interpreter_frame_result_handler_offset) * wordSize), never@739: rax); duke@435: duke@435: // pass mirror handle if static call duke@435: { duke@435: Label L; stefank@3391: const int mirror_offset = in_bytes(Klass::java_mirror_offset()); coleenp@4037: __ movl(t, Address(method, Method::access_flags_offset())); duke@435: __ testl(t, JVM_ACC_STATIC); duke@435: __ jcc(Assembler::zero, L); duke@435: // get mirror coleenp@4037: __ movptr(t, Address(method, Method::const_offset())); coleenp@4037: __ movptr(t, Address(t, ConstMethod::constants_offset())); coleenp@4037: __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes())); never@739: __ movptr(t, Address(t, mirror_offset)); duke@435: // copy mirror into activation frame never@739: __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), duke@435: t); duke@435: // pass handle to mirror never@739: __ lea(c_rarg1, never@739: Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize)); duke@435: __ bind(L); duke@435: } duke@435: duke@435: // get native function entry point duke@435: { duke@435: Label L; coleenp@4037: __ movptr(rax, Address(method, Method::native_function_offset())); duke@435: ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry()); duke@435: __ movptr(rscratch2, unsatisfied.addr()); never@739: __ cmpptr(rax, rscratch2); duke@435: __ jcc(Assembler::notEqual, L); duke@435: __ call_VM(noreg, duke@435: CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime::prepare_native_call), duke@435: method); duke@435: __ get_method(method); coleenp@4037: __ movptr(rax, Address(method, Method::native_function_offset())); duke@435: __ bind(L); duke@435: } duke@435: duke@435: // pass JNIEnv never@739: __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset())); duke@435: duke@435: // It is enough that the pc() points into the right code duke@435: // segment. It does not have to be the correct return pc. duke@435: __ set_last_Java_frame(rsp, rbp, (address) __ pc()); duke@435: duke@435: // change thread state duke@435: #ifdef ASSERT duke@435: { duke@435: Label L; duke@435: __ movl(t, Address(r15_thread, JavaThread::thread_state_offset())); duke@435: __ cmpl(t, _thread_in_Java); duke@435: __ jcc(Assembler::equal, L); duke@435: __ stop("Wrong thread state in native stub"); duke@435: __ bind(L); duke@435: } duke@435: #endif duke@435: duke@435: // Change state to native duke@435: duke@435: __ movl(Address(r15_thread, JavaThread::thread_state_offset()), duke@435: _thread_in_native); duke@435: duke@435: // Call the native method. duke@435: __ call(rax); duke@435: // result potentially in rax or xmm0 duke@435: duke@435: // Depending on runtime options, either restore the MXCSR duke@435: // register after returning from the JNI Call or verify that duke@435: // it wasn't changed during -Xcheck:jni. duke@435: if (RestoreMXCSROnJNICalls) { never@739: __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std())); duke@435: } duke@435: else if (CheckJNICalls) { never@739: __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry()))); duke@435: } duke@435: duke@435: // NOTE: The order of these pushes is known to frame::interpreter_frame_result duke@435: // in order to extract the result of a method call. If the order of these duke@435: // pushes change or anything else is added to the stack then the code in duke@435: // interpreter_frame_result must also change. duke@435: duke@435: __ push(dtos); duke@435: __ push(ltos); duke@435: duke@435: // change thread state duke@435: __ movl(Address(r15_thread, JavaThread::thread_state_offset()), duke@435: _thread_in_native_trans); duke@435: duke@435: if (os::is_MP()) { duke@435: if (UseMembar) { duke@435: // Force this write out before the read below duke@435: __ membar(Assembler::Membar_mask_bits( duke@435: Assembler::LoadLoad | Assembler::LoadStore | duke@435: Assembler::StoreLoad | Assembler::StoreStore)); duke@435: } else { duke@435: // Write serialization page so VM thread can do a pseudo remote membar. duke@435: // We use the current thread pointer to calculate a thread specific duke@435: // offset to write to within the page. This minimizes bus traffic duke@435: // due to cache line collision. duke@435: __ serialize_memory(r15_thread, rscratch2); duke@435: } duke@435: } duke@435: duke@435: // check for safepoint operation in progress and/or pending suspend requests duke@435: { duke@435: Label Continue; duke@435: __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), duke@435: SafepointSynchronize::_not_synchronized); duke@435: duke@435: Label L; duke@435: __ jcc(Assembler::notEqual, L); duke@435: __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0); duke@435: __ jcc(Assembler::equal, Continue); duke@435: __ bind(L); duke@435: duke@435: // Don't use call_VM as it will see a possible pending exception duke@435: // and forward it and never return here preventing us from duke@435: // clearing _last_native_pc down below. Also can't use duke@435: // call_VM_leaf either as it will check to see if r13 & r14 are duke@435: // preserved and correspond to the bcp/locals pointers. So we do a duke@435: // runtime call by hand. duke@435: // never@739: __ mov(c_rarg0, r15_thread); coleenp@2318: __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM) never@739: __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows never@739: __ andptr(rsp, -16); // align stack as required by ABI duke@435: __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans))); never@739: __ mov(rsp, r12); // restore sp coleenp@548: __ reinit_heapbase(); duke@435: __ bind(Continue); duke@435: } duke@435: duke@435: // change thread state duke@435: __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java); duke@435: duke@435: // reset_last_Java_frame duke@435: __ reset_last_Java_frame(true, true); duke@435: duke@435: // reset handle block never@739: __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset())); never@739: __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD); duke@435: duke@435: // If result is an oop unbox and store it in frame where gc will see it duke@435: // and result handler will pick it up duke@435: duke@435: { duke@435: Label no_oop, store_result; duke@435: __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT))); never@739: __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize)); duke@435: __ jcc(Assembler::notEqual, no_oop); duke@435: // retrieve result duke@435: __ pop(ltos); never@739: __ testptr(rax, rax); duke@435: __ jcc(Assembler::zero, store_result); never@739: __ movptr(rax, Address(rax, 0)); duke@435: __ bind(store_result); never@739: __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax); duke@435: // keep stack depth as expected by pushing oop which will eventually be discarde duke@435: __ push(ltos); duke@435: __ bind(no_oop); duke@435: } duke@435: duke@435: duke@435: { duke@435: Label no_reguard; duke@435: __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()), duke@435: JavaThread::stack_guard_yellow_disabled); duke@435: __ jcc(Assembler::notEqual, no_reguard); duke@435: never@739: __ pusha(); // XXX only save smashed registers coleenp@2318: __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM) never@739: __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows never@739: __ andptr(rsp, -16); // align stack as required by ABI duke@435: __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); never@739: __ mov(rsp, r12); // restore sp never@739: __ popa(); // XXX only restore smashed registers coleenp@548: __ reinit_heapbase(); duke@435: duke@435: __ bind(no_reguard); duke@435: } duke@435: duke@435: duke@435: // The method register is junk from after the thread_in_native transition duke@435: // until here. Also can't call_VM until the bcp has been duke@435: // restored. Need bcp for throwing exception below so get it now. duke@435: __ get_method(method); duke@435: duke@435: // restore r13 to have legal interpreter frame, i.e., bci == 0 <=> duke@435: // r13 == code_base() coleenp@4037: __ movptr(r13, Address(method, Method::const_offset())); // get ConstMethod* coleenp@4037: __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase duke@435: // handle exceptions (exception handling will handle unlocking!) duke@435: { duke@435: Label L; never@739: __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD); duke@435: __ jcc(Assembler::zero, L); duke@435: // Note: At some point we may want to unify this with the code duke@435: // used in call_VM_base(); i.e., we should use the duke@435: // StubRoutines::forward_exception code. For now this doesn't work duke@435: // here because the rsp is not correctly set at this point. duke@435: __ MacroAssembler::call_VM(noreg, duke@435: CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime::throw_pending_exception)); duke@435: __ should_not_reach_here(); duke@435: __ bind(L); duke@435: } duke@435: duke@435: // do unlocking if necessary duke@435: { duke@435: Label L; coleenp@4037: __ movl(t, Address(method, Method::access_flags_offset())); duke@435: __ testl(t, JVM_ACC_SYNCHRONIZED); duke@435: __ jcc(Assembler::zero, L); duke@435: // the code below should be shared with interpreter macro duke@435: // assembler implementation duke@435: { duke@435: Label unlock; duke@435: // BasicObjectLock will be first in list, since this is a duke@435: // synchronized method. However, need to check that the object duke@435: // has not been unlocked by an explicit monitorexit bytecode. duke@435: const Address monitor(rbp, duke@435: (intptr_t)(frame::interpreter_frame_initial_sp_offset * duke@435: wordSize - sizeof(BasicObjectLock))); duke@435: duke@435: // monitor expect in c_rarg1 for slow unlock path never@739: __ lea(c_rarg1, monitor); // address of first monitor duke@435: never@739: __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); never@739: __ testptr(t, t); duke@435: __ jcc(Assembler::notZero, unlock); duke@435: duke@435: // Entry already unlocked, need to throw exception duke@435: __ MacroAssembler::call_VM(noreg, duke@435: CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime::throw_illegal_monitor_state_exception)); duke@435: __ should_not_reach_here(); duke@435: duke@435: __ bind(unlock); duke@435: __ unlock_object(c_rarg1); duke@435: } duke@435: __ bind(L); duke@435: } duke@435: duke@435: // jvmti support duke@435: // Note: This must happen _after_ handling/throwing any exceptions since duke@435: // the exception handler code notifies the runtime of method exits duke@435: // too. If this happens before, method entry/exit notifications are duke@435: // not properly paired (was bug - gri 11/22/99). duke@435: __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI); duke@435: duke@435: // restore potential result in edx:eax, call result handler to duke@435: // restore potential result in ST0 & handle result duke@435: duke@435: __ pop(ltos); duke@435: __ pop(dtos); duke@435: never@739: __ movptr(t, Address(rbp, never@739: (frame::interpreter_frame_result_handler_offset) * wordSize)); duke@435: __ call(t); duke@435: duke@435: // remove activation never@739: __ movptr(t, Address(rbp, never@739: frame::interpreter_frame_sender_sp_offset * never@739: wordSize)); // get sender sp duke@435: __ leave(); // remove frame anchor never@739: __ pop(rdi); // get return address never@739: __ mov(rsp, t); // set sp to sender sp duke@435: __ jmp(rdi); duke@435: duke@435: if (inc_counter) { duke@435: // Handle overflow of counter and compile method duke@435: __ bind(invocation_counter_overflow); duke@435: generate_counter_overflow(&continue_after_compile); duke@435: } duke@435: duke@435: return entry_point; duke@435: } duke@435: duke@435: // duke@435: // Generic interpreted method entry to (asm) interpreter duke@435: // duke@435: address InterpreterGenerator::generate_normal_entry(bool synchronized) { duke@435: // determine code generation flags duke@435: bool inc_counter = UseCompiler || CountCompiledCalls; duke@435: coleenp@4037: // ebx: Method* duke@435: // r13: sender sp duke@435: address entry_point = __ pc(); duke@435: jiangli@4338: const Address constMethod(rbx, Method::const_offset()); duke@435: const Address invocation_counter(rbx, coleenp@4037: Method::invocation_counter_offset() + duke@435: InvocationCounter::counter_offset()); coleenp@4037: const Address access_flags(rbx, Method::access_flags_offset()); jiangli@4338: const Address size_of_parameters(rdx, jiangli@4338: ConstMethod::size_of_parameters_offset()); jiangli@4338: const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset()); jiangli@4338: duke@435: duke@435: // get parameter size (always needed) jiangli@4338: __ movptr(rdx, constMethod); jrose@1057: __ load_unsigned_short(rcx, size_of_parameters); duke@435: coleenp@4037: // rbx: Method* duke@435: // rcx: size of parameters duke@435: // r13: sender_sp (could differ from sp+wordSize if we were called via c2i ) duke@435: jrose@1057: __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words duke@435: __ subl(rdx, rcx); // rdx = no. of additional locals duke@435: duke@435: // YYY duke@435: // __ incrementl(rdx); duke@435: // __ andl(rdx, -2); duke@435: duke@435: // see if we've got enough room on the stack for locals plus overhead. duke@435: generate_stack_overflow_check(); duke@435: duke@435: // get return address never@739: __ pop(rax); duke@435: duke@435: // compute beginning of parameters (r14) never@739: __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize)); duke@435: duke@435: // rdx - # of additional locals duke@435: // allocate space for locals duke@435: // explicitly initialize locals duke@435: { duke@435: Label exit, loop; duke@435: __ testl(rdx, rdx); duke@435: __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0 duke@435: __ bind(loop); never@739: __ push((int) NULL_WORD); // initialize local variables duke@435: __ decrementl(rdx); // until everything initialized duke@435: __ jcc(Assembler::greater, loop); duke@435: __ bind(exit); duke@435: } duke@435: duke@435: // (pre-)fetch invocation count duke@435: if (inc_counter) { duke@435: __ movl(rcx, invocation_counter); duke@435: } duke@435: // initialize fixed part of activation frame duke@435: generate_fixed_frame(false); duke@435: duke@435: // make sure method is not native & not abstract duke@435: #ifdef ASSERT duke@435: __ movl(rax, access_flags); duke@435: { duke@435: Label L; duke@435: __ testl(rax, JVM_ACC_NATIVE); duke@435: __ jcc(Assembler::zero, L); duke@435: __ stop("tried to execute native method as non-native"); duke@435: __ bind(L); duke@435: } duke@435: { duke@435: Label L; duke@435: __ testl(rax, JVM_ACC_ABSTRACT); duke@435: __ jcc(Assembler::zero, L); duke@435: __ stop("tried to execute abstract method in interpreter"); duke@435: __ bind(L); duke@435: } duke@435: #endif duke@435: duke@435: // Since at this point in the method invocation the exception duke@435: // handler would try to exit the monitor of synchronized methods duke@435: // which hasn't been entered yet, we set the thread local variable duke@435: // _do_not_unlock_if_synchronized to true. The remove_activation duke@435: // will check this flag. duke@435: duke@435: const Address do_not_unlock_if_synchronized(r15_thread, duke@435: in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); duke@435: __ movbool(do_not_unlock_if_synchronized, true); duke@435: duke@435: // increment invocation count & check for overflow duke@435: Label invocation_counter_overflow; duke@435: Label profile_method; duke@435: Label profile_method_continue; duke@435: if (inc_counter) { duke@435: generate_counter_incr(&invocation_counter_overflow, duke@435: &profile_method, duke@435: &profile_method_continue); duke@435: if (ProfileInterpreter) { duke@435: __ bind(profile_method_continue); duke@435: } duke@435: } duke@435: duke@435: Label continue_after_compile; duke@435: __ bind(continue_after_compile); duke@435: duke@435: // check for synchronized interpreted methods duke@435: bang_stack_shadow_pages(false); duke@435: duke@435: // reset the _do_not_unlock_if_synchronized flag duke@435: __ movbool(do_not_unlock_if_synchronized, false); duke@435: duke@435: // check for synchronized methods duke@435: // Must happen AFTER invocation_counter check and stack overflow check, duke@435: // so method is not locked if overflows. duke@435: if (synchronized) { duke@435: // Allocate monitor and lock method duke@435: lock_method(); duke@435: } else { duke@435: // no synchronization necessary duke@435: #ifdef ASSERT duke@435: { duke@435: Label L; duke@435: __ movl(rax, access_flags); duke@435: __ testl(rax, JVM_ACC_SYNCHRONIZED); duke@435: __ jcc(Assembler::zero, L); duke@435: __ stop("method needs synchronization"); duke@435: __ bind(L); duke@435: } duke@435: #endif duke@435: } duke@435: duke@435: // start execution duke@435: #ifdef ASSERT duke@435: { duke@435: Label L; duke@435: const Address monitor_block_top (rbp, duke@435: frame::interpreter_frame_monitor_block_top_offset * wordSize); never@739: __ movptr(rax, monitor_block_top); never@739: __ cmpptr(rax, rsp); duke@435: __ jcc(Assembler::equal, L); duke@435: __ stop("broken stack frame setup in interpreter"); duke@435: __ bind(L); duke@435: } duke@435: #endif duke@435: duke@435: // jvmti support duke@435: __ notify_method_entry(); duke@435: duke@435: __ dispatch_next(vtos); duke@435: duke@435: // invocation counter overflow duke@435: if (inc_counter) { duke@435: if (ProfileInterpreter) { duke@435: // We have decided to profile this method in the interpreter duke@435: __ bind(profile_method); iveresov@2438: __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); iveresov@2438: __ set_method_data_pointer_for_bcp(); iveresov@2461: __ get_method(rbx); duke@435: __ jmp(profile_method_continue); duke@435: } duke@435: // Handle overflow of counter and compile method duke@435: __ bind(invocation_counter_overflow); duke@435: generate_counter_overflow(&continue_after_compile); duke@435: } duke@435: duke@435: return entry_point; duke@435: } duke@435: duke@435: // Entry points duke@435: // duke@435: // Here we generate the various kind of entries into the interpreter. duke@435: // The two main entry type are generic bytecode methods and native duke@435: // call method. These both come in synchronized and non-synchronized duke@435: // versions but the frame layout they create is very similar. The duke@435: // other method entry types are really just special purpose entries duke@435: // that are really entry and interpretation all in one. These are for duke@435: // trivial methods like accessor, empty, or special math methods. duke@435: // duke@435: // When control flow reaches any of the entry types for the interpreter duke@435: // the following holds -> duke@435: // duke@435: // Arguments: duke@435: // coleenp@4037: // rbx: Method* duke@435: // duke@435: // Stack layout immediately at entry duke@435: // duke@435: // [ return address ] <--- rsp duke@435: // [ parameter n ] duke@435: // ... duke@435: // [ parameter 1 ] duke@435: // [ expression stack ] (caller's java expression stack) duke@435: duke@435: // Assuming that we don't go to one of the trivial specialized entries duke@435: // the stack will look like below when we are ready to execute the duke@435: // first bytecode (or call the native routine). The register usage duke@435: // will be as the template based interpreter expects (see duke@435: // interpreter_amd64.hpp). duke@435: // duke@435: // local variables follow incoming parameters immediately; i.e. duke@435: // the return address is moved to the end of the locals). duke@435: // duke@435: // [ monitor entry ] <--- rsp duke@435: // ... duke@435: // [ monitor entry ] duke@435: // [ expr. stack bottom ] duke@435: // [ saved r13 ] duke@435: // [ current r14 ] coleenp@4037: // [ Method* ] duke@435: // [ saved ebp ] <--- rbp duke@435: // [ return address ] duke@435: // [ local variable m ] duke@435: // ... duke@435: // [ local variable 1 ] duke@435: // [ parameter n ] duke@435: // ... duke@435: // [ parameter 1 ] <--- r14 duke@435: duke@435: address AbstractInterpreterGenerator::generate_method_entry( duke@435: AbstractInterpreter::MethodKind kind) { duke@435: // determine code generation flags duke@435: bool synchronized = false; duke@435: address entry_point = NULL; duke@435: duke@435: switch (kind) { duke@435: case Interpreter::zerolocals : break; duke@435: case Interpreter::zerolocals_synchronized: synchronized = true; break; twisti@3969: case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break; twisti@3969: case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break; twisti@3969: case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break; twisti@3969: case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break; twisti@3969: case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break; never@1141: never@1141: case Interpreter::java_lang_math_sin : // fall thru never@1141: case Interpreter::java_lang_math_cos : // fall thru never@1141: case Interpreter::java_lang_math_tan : // fall thru never@1141: case Interpreter::java_lang_math_abs : // fall thru never@1141: case Interpreter::java_lang_math_log : // fall thru never@1141: case Interpreter::java_lang_math_log10 : // fall thru roland@3787: case Interpreter::java_lang_math_sqrt : // fall thru roland@3787: case Interpreter::java_lang_math_pow : // fall thru twisti@3969: case Interpreter::java_lang_math_exp : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break; johnc@2781: case Interpreter::java_lang_ref_reference_get johnc@2781: : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break; twisti@3969: default: twisti@3969: fatal(err_msg("unexpected method kind: %d", kind)); twisti@3969: break; duke@435: } duke@435: duke@435: if (entry_point) { duke@435: return entry_point; duke@435: } duke@435: duke@435: return ((InterpreterGenerator*) this)-> duke@435: generate_normal_entry(synchronized); duke@435: } duke@435: never@1609: // These should never be compiled since the interpreter will prefer never@1609: // the compiled version to the intrinsic version. never@1609: bool AbstractInterpreter::can_be_compiled(methodHandle m) { never@1609: switch (method_kind(m)) { never@1609: case Interpreter::java_lang_math_sin : // fall thru never@1609: case Interpreter::java_lang_math_cos : // fall thru never@1609: case Interpreter::java_lang_math_tan : // fall thru never@1609: case Interpreter::java_lang_math_abs : // fall thru never@1609: case Interpreter::java_lang_math_log : // fall thru never@1609: case Interpreter::java_lang_math_log10 : // fall thru roland@3787: case Interpreter::java_lang_math_sqrt : // fall thru roland@3787: case Interpreter::java_lang_math_pow : // fall thru roland@3787: case Interpreter::java_lang_math_exp : never@1609: return false; never@1609: default: never@1609: return true; never@1609: } never@1609: } never@1609: duke@435: // How much stack a method activation needs in words. coleenp@4037: int AbstractInterpreter::size_top_interpreter_activation(Method* method) { duke@435: const int entry_size = frame::interpreter_frame_monitor_size(); duke@435: duke@435: // total overhead size: entry_size + (saved rbp thru expr stack duke@435: // bottom). be sure to change this if you add/subtract anything duke@435: // to/from the overhead area duke@435: const int overhead_size = duke@435: -(frame::interpreter_frame_initial_sp_offset) + entry_size; duke@435: duke@435: const int stub_code = frame::entry_frame_after_call_words; coleenp@4037: const int extra_stack = Method::extra_stack_entries(); jrose@1145: const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) * twisti@1861: Interpreter::stackElementWords; duke@435: return (overhead_size + method_stack + stub_code); duke@435: } duke@435: coleenp@4037: int AbstractInterpreter::layout_activation(Method* method, duke@435: int tempcount, duke@435: int popframe_extra_args, duke@435: int moncount, never@2901: int caller_actual_parameters, duke@435: int callee_param_count, duke@435: int callee_locals, duke@435: frame* caller, duke@435: frame* interpreter_frame, duke@435: bool is_top_frame) { duke@435: // Note: This calculation must exactly parallel the frame setup duke@435: // in AbstractInterpreterGenerator::generate_method_entry. duke@435: // If interpreter_frame!=NULL, set up the method, locals, and monitors. duke@435: // The frame interpreter_frame, if not NULL, is guaranteed to be the duke@435: // right size, as determined by a previous call to this method. duke@435: // It is also guaranteed to be walkable even though it is in a skeletal state duke@435: duke@435: // fixed size of an interpreter frame: twisti@1861: int max_locals = method->max_locals() * Interpreter::stackElementWords; duke@435: int extra_locals = (method->max_locals() - method->size_of_parameters()) * twisti@1861: Interpreter::stackElementWords; duke@435: duke@435: int overhead = frame::sender_sp_offset - duke@435: frame::interpreter_frame_initial_sp_offset; duke@435: // Our locals were accounted for by the caller (or last_frame_adjust duke@435: // on the transistion) Since the callee parameters already account duke@435: // for the callee's params we only need to account for the extra duke@435: // locals. duke@435: int size = overhead + twisti@1861: (callee_locals - callee_param_count)*Interpreter::stackElementWords + duke@435: moncount * frame::interpreter_frame_monitor_size() + twisti@1861: tempcount* Interpreter::stackElementWords + popframe_extra_args; duke@435: if (interpreter_frame != NULL) { duke@435: #ifdef ASSERT twisti@2698: if (!EnableInvokeDynamic) twisti@1570: // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences? twisti@1570: // Probably, since deoptimization doesn't work yet. twisti@1570: assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable"); duke@435: assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)"); duke@435: #endif duke@435: duke@435: interpreter_frame->interpreter_frame_set_method(method); duke@435: // NOTE the difference in using sender_sp and duke@435: // interpreter_frame_sender_sp interpreter_frame_sender_sp is duke@435: // the original sp of the caller (the unextended_sp) and duke@435: // sender_sp is fp+16 XXX duke@435: intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1; duke@435: twisti@3238: #ifdef ASSERT twisti@3238: if (caller->is_interpreted_frame()) { twisti@3238: assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement"); twisti@3238: } twisti@3238: #endif twisti@3238: duke@435: interpreter_frame->interpreter_frame_set_locals(locals); duke@435: BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin(); duke@435: BasicObjectLock* monbot = montop - moncount; duke@435: interpreter_frame->interpreter_frame_set_monitor_end(monbot); duke@435: duke@435: // Set last_sp duke@435: intptr_t* esp = (intptr_t*) monbot - twisti@1861: tempcount*Interpreter::stackElementWords - duke@435: popframe_extra_args; duke@435: interpreter_frame->interpreter_frame_set_last_sp(esp); duke@435: duke@435: // All frames but the initial (oldest) interpreter frame we fill in have duke@435: // a value for sender_sp that allows walking the stack but isn't duke@435: // truly correct. Correct the value here. duke@435: if (extra_locals != 0 && duke@435: interpreter_frame->sender_sp() == duke@435: interpreter_frame->interpreter_frame_sender_sp()) { duke@435: interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + duke@435: extra_locals); duke@435: } duke@435: *interpreter_frame->interpreter_frame_cache_addr() = duke@435: method->constants()->cache(); duke@435: } duke@435: return size; duke@435: } duke@435: duke@435: //----------------------------------------------------------------------------- duke@435: // Exceptions duke@435: duke@435: void TemplateInterpreterGenerator::generate_throw_exception() { duke@435: // Entry point in previous activation (i.e., if the caller was duke@435: // interpreted) duke@435: Interpreter::_rethrow_exception_entry = __ pc(); duke@435: // Restore sp to interpreter_frame_last_sp even though we are going duke@435: // to empty the expression stack for the exception processing. never@739: __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); duke@435: // rax: exception duke@435: // rdx: return address/pc that threw exception duke@435: __ restore_bcp(); // r13 points to call/send duke@435: __ restore_locals(); coleenp@548: __ reinit_heapbase(); // restore r12 as heapbase. duke@435: // Entry point for exceptions thrown within interpreter code duke@435: Interpreter::_throw_exception_entry = __ pc(); duke@435: // expression stack is undefined here duke@435: // rax: exception duke@435: // r13: exception bcp duke@435: __ verify_oop(rax); never@739: __ mov(c_rarg1, rax); duke@435: duke@435: // expression stack must be empty before entering the VM in case of duke@435: // an exception duke@435: __ empty_expression_stack(); duke@435: // find exception handler address and preserve exception oop duke@435: __ call_VM(rdx, duke@435: CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime::exception_handler_for_exception), duke@435: c_rarg1); duke@435: // rax: exception handler entry point duke@435: // rdx: preserved exception oop duke@435: // r13: bcp for exception handler duke@435: __ push_ptr(rdx); // push exception which is now the only value on the stack duke@435: __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!) duke@435: duke@435: // If the exception is not handled in the current frame the frame is duke@435: // removed and the exception is rethrown (i.e. exception duke@435: // continuation is _rethrow_exception). duke@435: // duke@435: // Note: At this point the bci is still the bxi for the instruction duke@435: // which caused the exception and the expression stack is duke@435: // empty. Thus, for any VM calls at this point, GC will find a legal duke@435: // oop map (with empty expression stack). duke@435: duke@435: // In current activation duke@435: // tos: exception duke@435: // esi: exception bcp duke@435: duke@435: // duke@435: // JVMTI PopFrame support duke@435: // duke@435: duke@435: Interpreter::_remove_activation_preserving_args_entry = __ pc(); duke@435: __ empty_expression_stack(); duke@435: // Set the popframe_processing bit in pending_popframe_condition duke@435: // indicating that we are currently handling popframe, so that duke@435: // call_VMs that may happen later do not trigger new popframe duke@435: // handling cycles. duke@435: __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset())); duke@435: __ orl(rdx, JavaThread::popframe_processing_bit); duke@435: __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx); duke@435: duke@435: { duke@435: // Check to see whether we are returning to a deoptimized frame. duke@435: // (The PopFrame call ensures that the caller of the popped frame is duke@435: // either interpreted or compiled and deoptimizes it if compiled.) duke@435: // In this case, we can't call dispatch_next() after the frame is duke@435: // popped, but instead must save the incoming arguments and restore duke@435: // them after deoptimization has occurred. duke@435: // duke@435: // Note that we don't compare the return PC against the duke@435: // deoptimization blob's unpack entry because of the presence of duke@435: // adapter frames in C2. duke@435: Label caller_not_deoptimized; never@739: __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize)); duke@435: __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime::interpreter_contains), c_rarg1); duke@435: __ testl(rax, rax); duke@435: __ jcc(Assembler::notZero, caller_not_deoptimized); duke@435: duke@435: // Compute size of arguments for saving when returning to duke@435: // deoptimized caller duke@435: __ get_method(rax); jiangli@4338: __ movptr(rax, Address(rax, Method::const_offset())); jiangli@4338: __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod:: duke@435: size_of_parameters_offset()))); twisti@1861: __ shll(rax, Interpreter::logStackElementSize); duke@435: __ restore_locals(); // XXX do we need this? never@739: __ subptr(r14, rax); never@739: __ addptr(r14, wordSize); duke@435: // Save these arguments duke@435: __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, duke@435: Deoptimization:: duke@435: popframe_preserve_args), duke@435: r15_thread, rax, r14); duke@435: duke@435: __ remove_activation(vtos, rdx, duke@435: /* throw_monitor_exception */ false, duke@435: /* install_monitor_exception */ false, duke@435: /* notify_jvmdi */ false); duke@435: duke@435: // Inform deoptimization that it is responsible for restoring duke@435: // these arguments duke@435: __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), duke@435: JavaThread::popframe_force_deopt_reexecution_bit); duke@435: duke@435: // Continue in deoptimization handler duke@435: __ jmp(rdx); duke@435: duke@435: __ bind(caller_not_deoptimized); duke@435: } duke@435: duke@435: __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */ duke@435: /* throw_monitor_exception */ false, duke@435: /* install_monitor_exception */ false, duke@435: /* notify_jvmdi */ false); duke@435: duke@435: // Finish with popframe handling duke@435: // A previous I2C followed by a deoptimization might have moved the duke@435: // outgoing arguments further up the stack. PopFrame expects the duke@435: // mutations to those outgoing arguments to be preserved and other duke@435: // constraints basically require this frame to look exactly as duke@435: // though it had previously invoked an interpreted activation with duke@435: // no space between the top of the expression stack (current duke@435: // last_sp) and the top of stack. Rather than force deopt to duke@435: // maintain this kind of invariant all the time we call a small duke@435: // fixup routine to move the mutated arguments onto the top of our duke@435: // expression stack if necessary. never@739: __ mov(c_rarg1, rsp); never@739: __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); duke@435: // PC must point into interpreter here duke@435: __ set_last_Java_frame(noreg, rbp, __ pc()); duke@435: __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2); duke@435: __ reset_last_Java_frame(true, true); duke@435: // Restore the last_sp and null it out never@739: __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); never@739: __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); duke@435: duke@435: __ restore_bcp(); // XXX do we need this? duke@435: __ restore_locals(); // XXX do we need this? duke@435: // The method data pointer was incremented already during duke@435: // call profiling. We have to restore the mdp for the current bcp. duke@435: if (ProfileInterpreter) { duke@435: __ set_method_data_pointer_for_bcp(); duke@435: } duke@435: duke@435: // Clear the popframe condition flag duke@435: __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), duke@435: JavaThread::popframe_inactive); duke@435: duke@435: __ dispatch_next(vtos); duke@435: // end of PopFrame support duke@435: duke@435: Interpreter::_remove_activation_entry = __ pc(); duke@435: duke@435: // preserve exception over this code sequence duke@435: __ pop_ptr(rax); never@739: __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax); duke@435: // remove the activation (without doing throws on illegalMonitorExceptions) duke@435: __ remove_activation(vtos, rdx, false, true, false); duke@435: // restore exception coleenp@4037: __ get_vm_result(rax, r15_thread); duke@435: duke@435: // In between activations - previous activation type unknown yet duke@435: // compute continuation point - the continuation point expects the duke@435: // following registers set up: duke@435: // duke@435: // rax: exception duke@435: // rdx: return address/pc that threw exception duke@435: // rsp: expression stack of caller duke@435: // rbp: ebp of caller never@739: __ push(rax); // save exception never@739: __ push(rdx); // save return address duke@435: __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, duke@435: SharedRuntime::exception_handler_for_return_address), twisti@1730: r15_thread, rdx); never@739: __ mov(rbx, rax); // save exception handler never@739: __ pop(rdx); // restore return address never@739: __ pop(rax); // restore exception duke@435: // Note that an "issuing PC" is actually the next PC after the call duke@435: __ jmp(rbx); // jump to exception duke@435: // handler of caller duke@435: } duke@435: duke@435: duke@435: // duke@435: // JVMTI ForceEarlyReturn support duke@435: // duke@435: address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { duke@435: address entry = __ pc(); duke@435: duke@435: __ restore_bcp(); duke@435: __ restore_locals(); duke@435: __ empty_expression_stack(); duke@435: __ load_earlyret_value(state); duke@435: never@739: __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset())); duke@435: Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset()); duke@435: duke@435: // Clear the earlyret state duke@435: __ movl(cond_addr, JvmtiThreadState::earlyret_inactive); duke@435: duke@435: __ remove_activation(state, rsi, duke@435: false, /* throw_monitor_exception */ duke@435: false, /* install_monitor_exception */ duke@435: true); /* notify_jvmdi */ duke@435: __ jmp(rsi); duke@435: duke@435: return entry; duke@435: } // end of ForceEarlyReturn support duke@435: duke@435: duke@435: //----------------------------------------------------------------------------- duke@435: // Helper for vtos entry point generation duke@435: duke@435: void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, duke@435: address& bep, duke@435: address& cep, duke@435: address& sep, duke@435: address& aep, duke@435: address& iep, duke@435: address& lep, duke@435: address& fep, duke@435: address& dep, duke@435: address& vep) { duke@435: assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); duke@435: Label L; duke@435: aep = __ pc(); __ push_ptr(); __ jmp(L); duke@435: fep = __ pc(); __ push_f(); __ jmp(L); duke@435: dep = __ pc(); __ push_d(); __ jmp(L); duke@435: lep = __ pc(); __ push_l(); __ jmp(L); duke@435: bep = cep = sep = duke@435: iep = __ pc(); __ push_i(); duke@435: vep = __ pc(); duke@435: __ bind(L); duke@435: generate_and_dispatch(t); duke@435: } duke@435: duke@435: duke@435: //----------------------------------------------------------------------------- duke@435: // Generation of individual instructions duke@435: duke@435: // helpers for generate_and_dispatch duke@435: duke@435: duke@435: InterpreterGenerator::InterpreterGenerator(StubQueue* code) duke@435: : TemplateInterpreterGenerator(code) { duke@435: generate_all(); // down here so it can be "virtual" duke@435: } duke@435: duke@435: //----------------------------------------------------------------------------- duke@435: duke@435: // Non-product code duke@435: #ifndef PRODUCT duke@435: address TemplateInterpreterGenerator::generate_trace_code(TosState state) { duke@435: address entry = __ pc(); duke@435: duke@435: __ push(state); never@739: __ push(c_rarg0); never@739: __ push(c_rarg1); never@739: __ push(c_rarg2); never@739: __ push(c_rarg3); never@739: __ mov(c_rarg2, rax); // Pass itos duke@435: #ifdef _WIN64 duke@435: __ movflt(xmm3, xmm0); // Pass ftos duke@435: #endif duke@435: __ call_VM(noreg, duke@435: CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), duke@435: c_rarg1, c_rarg2, c_rarg3); never@739: __ pop(c_rarg3); never@739: __ pop(c_rarg2); never@739: __ pop(c_rarg1); never@739: __ pop(c_rarg0); duke@435: __ pop(state); duke@435: __ ret(0); // return from result handler duke@435: duke@435: return entry; duke@435: } duke@435: duke@435: void TemplateInterpreterGenerator::count_bytecode() { duke@435: __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value)); duke@435: } duke@435: duke@435: void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { duke@435: __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()])); duke@435: } duke@435: duke@435: void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { duke@435: __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index)); duke@435: __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes); duke@435: __ orl(rbx, duke@435: ((int) t->bytecode()) << duke@435: BytecodePairHistogram::log2_number_of_codes); duke@435: __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx); duke@435: __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters)); duke@435: __ incrementl(Address(rscratch1, rbx, Address::times_4)); duke@435: } duke@435: duke@435: duke@435: void TemplateInterpreterGenerator::trace_bytecode(Template* t) { duke@435: // Call a little run-time stub to avoid blow-up for each bytecode. duke@435: // The run-time runtime saves the right registers, depending on duke@435: // the tosca in-state for the given template. duke@435: duke@435: assert(Interpreter::trace_code(t->tos_in()) != NULL, duke@435: "entry must have been generated"); coleenp@2318: __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM) never@739: __ andptr(rsp, -16); // align stack as required by ABI duke@435: __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in()))); never@739: __ mov(rsp, r12); // restore sp coleenp@548: __ reinit_heapbase(); duke@435: } duke@435: duke@435: duke@435: void TemplateInterpreterGenerator::stop_interpreter_at() { duke@435: Label L; duke@435: __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value), duke@435: StopInterpreterAt); duke@435: __ jcc(Assembler::notEqual, L); duke@435: __ int3(); duke@435: __ bind(L); duke@435: } duke@435: #endif // !PRODUCT never@739: #endif // ! CC_INTERP