duke@435: /* phh@9669: * Copyright (c) 1997, 2017, 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: duke@435: duke@435: #ifndef CC_INTERP 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: // Note: There should be a minimal interpreter frame set up when stack duke@435: // overflow occurs since we check explicitly for it now. duke@435: // duke@435: #ifdef ASSERT duke@435: { Label L; never@739: __ lea(rax, Address(rbp, duke@435: frame::interpreter_frame_monitor_block_top_offset * wordSize)); never@739: __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp, duke@435: // (stack 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 exception duke@435: // happened duke@435: __ empty_expression_stack(); duke@435: __ empty_FPU_stack(); duke@435: // throw exception duke@435: __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError)); duke@435: return entry; duke@435: } duke@435: duke@435: address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) { duke@435: address entry = __ pc(); duke@435: // expression stack must be empty before entering the VM if an exception happened duke@435: __ empty_expression_stack(); duke@435: __ empty_FPU_stack(); duke@435: // setup parameters duke@435: // ??? convention: expect aberrant index in register rbx, duke@435: __ lea(rax, ExternalAddress((address)name)); duke@435: __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx); duke@435: return entry; duke@435: } duke@435: duke@435: address TemplateInterpreterGenerator::generate_ClassCastException_handler() { duke@435: address entry = __ pc(); duke@435: // object is at TOS never@739: __ pop(rax); duke@435: // expression stack must be empty before entering the VM if an exception duke@435: // happened duke@435: __ empty_expression_stack(); duke@435: __ empty_FPU_stack(); duke@435: __ call_VM(noreg, duke@435: CAST_FROM_FN_PTR(address, duke@435: InterpreterRuntime::throw_ClassCastException), duke@435: rax); duke@435: return entry; duke@435: } duke@435: duke@435: address TemplateInterpreterGenerator::generate_exception_handler_common(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(rbx); duke@435: } duke@435: // expression stack must be empty before entering the VM if an exception happened duke@435: __ empty_expression_stack(); duke@435: __ empty_FPU_stack(); duke@435: // setup parameters duke@435: __ lea(rax, ExternalAddress((address)name)); duke@435: if (pass_oop) { duke@435: __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx); duke@435: } else { duke@435: if (message != NULL) { duke@435: __ lea(rbx, ExternalAddress((address)message)); duke@435: } else { xlu@947: __ movptr(rbx, NULL_WORD); duke@435: } duke@435: __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx); 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 xlu@947: __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); duke@435: __ dispatch_next(state); duke@435: return entry; duke@435: } duke@435: duke@435: twisti@6039: address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) { twisti@2552: address entry = __ pc(); duke@435: duke@435: #ifdef COMPILER2 duke@435: // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases twisti@6039: if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) { duke@435: for (int i = 1; i < 8; i++) { duke@435: __ ffree(i); duke@435: } duke@435: } else if (UseSSE < 2) { duke@435: __ empty_FPU_stack(); duke@435: } duke@435: #endif twisti@6039: if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) { duke@435: __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled"); duke@435: } else { duke@435: __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled"); duke@435: } duke@435: duke@435: // In SSE mode, interpreter returns FP results in xmm0 but they need duke@435: // to end up back on the FPU so it can operate on them. twisti@6039: if (state == ftos && UseSSE >= 1) { never@739: __ subptr(rsp, wordSize); duke@435: __ movflt(Address(rsp, 0), xmm0); duke@435: __ fld_s(Address(rsp, 0)); never@739: __ addptr(rsp, wordSize); twisti@6039: } else if (state == dtos && UseSSE >= 2) { never@739: __ subptr(rsp, 2*wordSize); duke@435: __ movdbl(Address(rsp, 0), xmm0); duke@435: __ fld_d(Address(rsp, 0)); never@739: __ addptr(rsp, 2*wordSize); duke@435: } duke@435: duke@435: __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter"); 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 rsp is now tos until next java call xlu@947: __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); duke@435: duke@435: __ restore_bcp(); duke@435: __ restore_locals(); jrose@1161: twisti@6039: if (state == atos) { roland@5921: Register mdp = rbx; roland@5921: Register tmp = rcx; roland@5921: __ profile_return_type(mdp, rax, tmp); roland@5921: } roland@5921: twisti@6039: const Register cache = rbx; twisti@6039: const Register index = rcx; twisti@6039: __ get_cache_and_index_at_bcp(cache, index, 1, index_size); twisti@6039: twisti@6039: const Register flags = cache; twisti@6039: __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset())); twisti@6039: __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask); twisti@6039: __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale())); duke@435: __ dispatch_next(state, step); jrose@1161: duke@435: return entry; duke@435: } duke@435: duke@435: duke@435: address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) { duke@435: address entry = __ pc(); duke@435: duke@435: // In SSE mode, FP results are in xmm0 duke@435: if (state == ftos && UseSSE > 0) { never@739: __ subptr(rsp, wordSize); duke@435: __ movflt(Address(rsp, 0), xmm0); duke@435: __ fld_s(Address(rsp, 0)); never@739: __ addptr(rsp, wordSize); duke@435: } else if (state == dtos && UseSSE >= 2) { never@739: __ subptr(rsp, 2*wordSize); duke@435: __ movdbl(Address(rsp, 0), xmm0); duke@435: __ fld_d(Address(rsp, 0)); never@739: __ addptr(rsp, 2*wordSize); duke@435: } duke@435: duke@435: __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter"); duke@435: duke@435: // The stack is not extended by deopt but we must NULL last_sp as this duke@435: // entry is like a "return". xlu@947: __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); duke@435: __ restore_bcp(); duke@435: __ restore_locals(); duke@435: // handle exceptions duke@435: { Label L; duke@435: const Register thread = rcx; duke@435: __ get_thread(thread); never@739: __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); duke@435: __ jcc(Assembler::zero, L); duke@435: __ call_VM(noreg, CAST_FROM_FN_PTR(address, 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: 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 : // fall through duke@435: case T_LONG : // fall through duke@435: case T_VOID : i = 4; break; duke@435: case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE duke@435: case T_DOUBLE : i = 6; break; duke@435: case T_OBJECT : // fall through duke@435: case T_ARRAY : i = 7; break; duke@435: default : ShouldNotReachHere(); duke@435: } duke@435: assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds"); duke@435: return i; duke@435: } duke@435: duke@435: duke@435: address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) { duke@435: address entry = __ pc(); duke@435: switch (type) { duke@435: case T_BOOLEAN: __ c2bool(rax); break; never@739: case T_CHAR : __ andptr(rax, 0xFFFF); 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_DOUBLE : duke@435: case T_FLOAT : duke@435: { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); never@739: __ pop(t); // remove return address first duke@435: // Must return a result for interpreter or compiler. In SSE duke@435: // mode, results are returned in xmm0 and the FPU stack must duke@435: // be empty. duke@435: if (type == T_FLOAT && UseSSE >= 1) { duke@435: // Load ST0 duke@435: __ fld_d(Address(rsp, 0)); duke@435: // Store as float and empty fpu stack duke@435: __ fstp_s(Address(rsp, 0)); duke@435: // and reload duke@435: __ movflt(xmm0, Address(rsp, 0)); duke@435: } else if (type == T_DOUBLE && UseSSE >= 2 ) { duke@435: __ movdbl(xmm0, Address(rsp, 0)); duke@435: } else { duke@435: // restore ST0 duke@435: __ fld_d(Address(rsp, 0)); duke@435: } duke@435: // and pop the temp never@739: __ addptr(rsp, 2 * wordSize); never@739: __ push(t); // restore return address duke@435: } duke@435: 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(TosState state, 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: // Helpers for commoning out cases in the various type of method entries. 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: // rcx: invocation counter duke@435: // duke@435: void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) { jiangli@4936: Label done; jiangli@4936: // Note: In tiered we increment either counters in MethodCounters* or in MDO jiangli@4936: // 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; jiangli@4936: Label no_mdo; 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); jiangli@4973: __ jmp(done); iveresov@2138: } iveresov@2138: __ bind(no_mdo); jiangli@4936: // Increment counter in MethodCounters jiangli@4936: const Address invocation_counter(rax, jiangli@4936: MethodCounters::invocation_counter_offset() + jiangli@4936: InvocationCounter::counter_offset()); jiangli@4936: jiangli@4936: __ get_method_counters(rbx, rax, done); jiangli@4936: __ increment_mask_and_jump(invocation_counter, increment, mask, jiangli@4936: rcx, false, Assembler::zero, overflow); iveresov@2138: __ bind(done); iveresov@2138: } else { jiangli@4936: const Address backedge_counter (rax, jiangli@4936: MethodCounters::backedge_counter_offset() + jiangli@4936: InvocationCounter::counter_offset()); jiangli@4936: const Address invocation_counter(rax, jiangli@4936: MethodCounters::invocation_counter_offset() + jiangli@4936: InvocationCounter::counter_offset()); duke@435: jiangli@4936: __ get_method_counters(rbx, rax, done); jiangli@4936: jiangli@4936: if (ProfileInterpreter) { jiangli@4936: __ incrementl(Address(rax, jiangli@4936: MethodCounters::interpreter_invocation_counter_offset())); iveresov@2138: } jiangli@4936: iveresov@2138: // Update standard invocation counters jiangli@4936: __ movl(rcx, invocation_counter); jiangli@4936: __ incrementl(rcx, InvocationCounter::count_increment); jiangli@4936: __ movl(invocation_counter, rcx); // save invocation count jiangli@4936: iveresov@2138: __ movl(rax, backedge_counter); // load backedge counter iveresov@2138: __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits iveresov@2138: 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: // BytecodeInterpreter only calls for native so code is elided. 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, iveresov@2138: 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, iveresov@2138: ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit)); iveresov@2138: __ jcc(Assembler::aboveEqual, *overflow); jiangli@4936: __ bind(done); 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: // rdi - locals duke@435: // rsi - bcp duke@435: // rbx, - method duke@435: // rdx - cpool duke@435: // rbp, - interpreter frame duke@435: duke@435: // C++ interpreter on entry duke@435: // rsi - new interpreter state pointer duke@435: // rbp - interpreter frame pointer duke@435: // rbx - method duke@435: duke@435: // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] duke@435: // rbx, - method duke@435: // rcx - rcvr (assuming there is one) duke@435: // top of stack return address of interpreter caller duke@435: // rsp - sender_sp duke@435: duke@435: // C++ interpreter only duke@435: // rsi - previous interpreter state pointer duke@435: duke@435: // InterpreterRuntime::frequency_counter_overflow takes one argument duke@435: // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp). duke@435: // The call returns the address of the verified entry point for the method or NULL duke@435: // if the compilation did not complete (either went background or bailed out). xlu@968: __ movptr(rax, (intptr_t)false); duke@435: __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax); duke@435: coleenp@4037: __ movptr(rbx, Address(rbp, method_offset)); // restore Method* duke@435: duke@435: // Preserve invariant that rsi/rdi 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: duke@435: void InterpreterGenerator::generate_stack_overflow_check(void) { 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: // Registers live on entry: duke@435: // duke@435: // Asm interpreter duke@435: // rdx: number of additional locals this frame needs (what we must check) coleenp@4037: // rbx,: Method* duke@435: duke@435: // destroyed on exit duke@435: // rax, duke@435: duke@435: // NOTE: since the additional locals are also always pushed (wasn't obvious in duke@435: // generate_method_entry) so the guard should work for them too. duke@435: // duke@435: duke@435: // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp duke@435: const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; duke@435: duke@435: // total overhead size: entry_size + (saved rbp, thru expr stack bottom). duke@435: // be sure to change this if you add/subtract anything to/from the overhead area duke@435: const int overhead_size = -(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: Label after_frame_check_pop; duke@435: never@739: __ push(rsi); duke@435: duke@435: const Register thread = rsi; duke@435: duke@435: __ get_thread(thread); duke@435: duke@435: const Address stack_base(thread, Thread::stack_base_offset()); duke@435: const Address stack_size(thread, Thread::stack_size_offset()); duke@435: duke@435: // locals + overhead, in bytes never@739: __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), 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: duke@435: // Use the maximum number of pages we might bang. duke@435: const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages : duke@435: (StackRedPages+StackYellowPages); never@739: __ 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_pop); duke@435: never@739: __ pop(rsi); // get saved bcp / (c++ prev state ). 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. duke@435: bdelsart@3372: __ pop(rax); // return address must be moved if SP is changed bdelsart@3372: __ mov(rsp, rsi); 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: // all done with frame size check duke@435: __ bind(after_frame_check_pop); never@739: __ pop(rsi); duke@435: duke@435: __ bind(after_frame_check); duke@435: } duke@435: duke@435: // Allocate monitor and lock method (asm interpreter) coleenp@4037: // rbx, - Method* duke@435: // 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 (rbp, 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: { 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: // get synchronization object duke@435: { Label done; stefank@3391: const int mirror_offset = in_bytes(Klass::java_mirror_offset()); duke@435: __ movl(rax, access_flags); duke@435: __ testl(rax, JVM_ACC_STATIC); never@739: __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case) duke@435: __ jcc(Assembler::zero, done); coleenp@4037: __ movptr(rax, Address(rbx, Method::const_offset())); coleenp@4037: __ movptr(rax, Address(rax, ConstMethod::constants_offset())); coleenp@4037: __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes())); never@739: __ movptr(rax, Address(rax, mirror_offset)); duke@435: __ bind(done); 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 never@739: __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object never@739: __ mov(rdx, rsp); // object address duke@435: __ lock_object(rdx); duke@435: } duke@435: duke@435: // duke@435: // Generate a fixed interpreter frame. This is identical setup for interpreted methods duke@435: // and for native methods hence the shared code. duke@435: 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, duke@435: duke@435: never@739: __ push(rsi); // set sender sp never@739: __ push((int32_t)NULL_WORD); // leave last_sp as null coleenp@4037: __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod* coleenp@4037: __ lea(rsi, Address(rsi,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(rdi); // set locals pointer duke@435: if (native_call) { never@739: __ push(0); // no bcp duke@435: } else { never@739: __ push(rsi); // 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: duke@435: // duke@435: // Various method entries duke@435: //------------------------------------------------------------------------------------------------------------------------ duke@435: // duke@435: // duke@435: duke@435: // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry duke@435: duke@435: address InterpreterGenerator::generate_accessor_entry(void) { duke@435: coleenp@4037: // rbx,: Method* duke@435: // rcx: receiver (preserve for slow entry into asm interpreter) duke@435: duke@435: // rsi: senderSP must preserved for slow path, set SP to it on fast path duke@435: duke@435: address entry_point = __ pc(); duke@435: Label xreturn_path; duke@435: duke@435: // do fastpath for resolved accessor methods duke@435: if (UseFastAccessorMethods) { duke@435: Label slow_path; duke@435: // If we need a safepoint check, generate full interpreter entry. duke@435: ExternalAddress state(SafepointSynchronize::address_of_state()); duke@435: __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), duke@435: SafepointSynchronize::_not_synchronized); duke@435: duke@435: __ jcc(Assembler::notEqual, slow_path); duke@435: // ASM/C++ Interpreter duke@435: // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1 duke@435: // Note: We can only use this code if the getfield has been resolved duke@435: // and if we don't have a null-pointer exception => check for duke@435: // these conditions first and use slow path if necessary. duke@435: // rbx,: method duke@435: // rcx: receiver never@739: __ movptr(rax, Address(rsp, wordSize)); duke@435: duke@435: // check if local 0 != NULL and read field never@739: __ testptr(rax, rax); duke@435: __ jcc(Assembler::zero, slow_path); duke@435: duke@435: // 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())); duke@435: // Shift codes right to get the index on the right. duke@435: // The bytecode fetched looks like <0xb4><0x2a> duke@435: __ shrl(rdx, 2*BitsPerByte); duke@435: __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); coleenp@4037: __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes())); duke@435: duke@435: // rax,: local 0 duke@435: // rbx,: method duke@435: // rcx: receiver - do not destroy since it is needed for slow path! duke@435: // rcx: scratch duke@435: // rdx: constant pool cache index duke@435: // rdi: constant pool cache duke@435: // rsi: sender sp duke@435: duke@435: // check if getfield has been resolved and read constant pool cache entry duke@435: // check the validity of the cache entry by testing whether _indices field duke@435: // contains Bytecode::_getfield in b1 byte. duke@435: assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below"); duke@435: __ movl(rcx, duke@435: Address(rdi, duke@435: rdx, coleenp@4037: Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset())); duke@435: __ shrl(rcx, 2*BitsPerByte); duke@435: __ andl(rcx, 0xFF); duke@435: __ cmpl(rcx, Bytecodes::_getfield); duke@435: __ jcc(Assembler::notEqual, slow_path); duke@435: duke@435: // Note: constant pool entry is not valid before bytecode is resolved never@739: __ movptr(rcx, never@739: Address(rdi, never@739: rdx, coleenp@4037: Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset())); duke@435: __ movl(rdx, duke@435: Address(rdi, duke@435: rdx, coleenp@4037: Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset())); duke@435: kevinw@8368: Label notByte, notBool, notShort, notChar; duke@435: const Address field_address (rax, rcx, Address::times_1); duke@435: duke@435: // Need to differentiate between igetfield, agetfield, bgetfield etc. duke@435: // because they are different sizes. duke@435: // 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 rdx after the above shift twisti@3969: ConstantPoolCacheEntry::verify_tos_state_shift(); duke@435: __ cmpl(rdx, btos); duke@435: __ jcc(Assembler::notEqual, notByte); duke@435: __ load_signed_byte(rax, field_address); duke@435: __ jmp(xreturn_path); duke@435: duke@435: __ bind(notByte); kevinw@8368: __ cmpl(rdx, ztos); kevinw@8368: __ jcc(Assembler::notEqual, notBool); kevinw@8368: __ load_signed_byte(rax, field_address); kevinw@8368: __ jmp(xreturn_path); kevinw@8368: kevinw@8368: __ bind(notBool); duke@435: __ cmpl(rdx, stos); duke@435: __ jcc(Assembler::notEqual, notShort); jrose@1057: __ load_signed_short(rax, field_address); duke@435: __ jmp(xreturn_path); duke@435: duke@435: __ bind(notShort); duke@435: __ cmpl(rdx, ctos); duke@435: __ jcc(Assembler::notEqual, notChar); jrose@1057: __ load_unsigned_short(rax, field_address); duke@435: __ jmp(xreturn_path); duke@435: duke@435: __ bind(notChar); duke@435: #ifdef ASSERT duke@435: Label okay; duke@435: __ cmpl(rdx, atos); duke@435: __ jcc(Assembler::equal, okay); duke@435: __ cmpl(rdx, itos); duke@435: __ jcc(Assembler::equal, okay); duke@435: __ stop("what type is this?"); duke@435: __ bind(okay); duke@435: #endif // ASSERT duke@435: // All the rest are a 32 bit wordsize never@739: // This is ok for now. Since fast accessors should be going away never@739: __ movptr(rax, field_address); duke@435: duke@435: __ bind(xreturn_path); duke@435: duke@435: // _ireturn/_areturn never@739: __ pop(rdi); // get return address never@739: __ mov(rsp, rsi); // set sp to sender sp duke@435: __ jmp(rdi); duke@435: duke@435: // generate a vanilla interpreter entry as the slow path duke@435: __ bind(slow_path); duke@435: duke@435: (void) generate_normal_entry(false); duke@435: return entry_point; duke@435: } duke@435: return NULL; duke@435: duke@435: } duke@435: 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 below 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: // rcx: receiver (preserve for slow entry into asm interpreter) johnc@2781: johnc@2781: // rsi: senderSP must preserved 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: 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: __ testptr(rax, rax); johnc@2781: __ jcc(Assembler::zero, slow_path); johnc@2781: johnc@2781: // rax: local 0 (must be preserved across the G1 barrier call) johnc@2781: // johnc@2781: // rbx: method (at this point it's scratch) johnc@2781: // rcx: receiver (at this point it's scratch) johnc@2781: // rdx: scratch johnc@2781: // rdi: scratch johnc@2781: // johnc@2781: // rsi: sender sp johnc@2781: johnc@2781: // Preserve the sender sp in case the pre-barrier johnc@2781: // calls the runtime johnc@2781: __ push(rsi); johnc@2781: johnc@2781: // Load the value of the referent field. johnc@2781: const Address field_address(rax, referent_offset); johnc@2781: __ movptr(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: __ get_thread(rcx); johnc@2781: __ g1_write_barrier_pre(noreg /* obj */, johnc@2781: rax /* pre_val */, johnc@2781: rcx /* thread */, johnc@2781: rbx /* tmp */, johnc@2781: true /* tosca_save */, johnc@2781: true /* expand_call */); johnc@2781: johnc@2781: // _areturn johnc@2781: __ pop(rsi); // get sender sp johnc@2781: __ pop(rdi); // get return address johnc@2781: __ mov(rsp, rsi); // set sp to sender sp johnc@2781: __ jmp(rdi); johnc@2781: 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: drchase@5353: /** drchase@5353: * Method entry for static native methods: drchase@5353: * int java.util.zip.CRC32.update(int crc, int b) drchase@5353: */ drchase@5353: address InterpreterGenerator::generate_CRC32_update_entry() { drchase@5353: if (UseCRC32Intrinsics) { drchase@5353: address entry = __ pc(); drchase@5353: drchase@5353: // rbx,: Method* drchase@5353: // rsi: senderSP must preserved for slow path, set SP to it on fast path drchase@5353: // rdx: scratch drchase@5353: // rdi: scratch drchase@5353: drchase@5353: Label slow_path; drchase@5353: // If we need a safepoint check, generate full interpreter entry. drchase@5353: ExternalAddress state(SafepointSynchronize::address_of_state()); drchase@5353: __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), drchase@5353: SafepointSynchronize::_not_synchronized); drchase@5353: __ jcc(Assembler::notEqual, slow_path); drchase@5353: drchase@5353: // We don't generate local frame and don't align stack because drchase@5353: // we call stub code and there is no safepoint on this path. drchase@5353: drchase@5353: // Load parameters drchase@5353: const Register crc = rax; // crc drchase@5353: const Register val = rdx; // source java byte value drchase@5353: const Register tbl = rdi; // scratch drchase@5353: drchase@5353: // Arguments are reversed on java expression stack drchase@5353: __ movl(val, Address(rsp, wordSize)); // byte value drchase@5353: __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC drchase@5353: drchase@5353: __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr())); drchase@5353: __ notl(crc); // ~crc drchase@5353: __ update_byte_crc32(crc, val, tbl); drchase@5353: __ notl(crc); // ~crc drchase@5353: // result in rax drchase@5353: drchase@5353: // _areturn drchase@5353: __ pop(rdi); // get return address drchase@5353: __ mov(rsp, rsi); // set sp to sender sp drchase@5353: __ jmp(rdi); drchase@5353: drchase@5353: // generate a vanilla native entry as the slow path drchase@5353: __ bind(slow_path); drchase@5353: drchase@5353: (void) generate_native_entry(false); drchase@5353: drchase@5353: return entry; drchase@5353: } drchase@5353: return generate_native_entry(false); drchase@5353: } drchase@5353: drchase@5353: /** drchase@5353: * Method entry for static native methods: drchase@5353: * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len) drchase@5353: * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) drchase@5353: */ drchase@5353: address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { drchase@5353: if (UseCRC32Intrinsics) { drchase@5353: address entry = __ pc(); drchase@5353: drchase@5353: // rbx,: Method* drchase@5353: // rsi: senderSP must preserved for slow path, set SP to it on fast path drchase@5353: // rdx: scratch drchase@5353: // rdi: scratch drchase@5353: drchase@5353: Label slow_path; drchase@5353: // If we need a safepoint check, generate full interpreter entry. drchase@5353: ExternalAddress state(SafepointSynchronize::address_of_state()); drchase@5353: __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), drchase@5353: SafepointSynchronize::_not_synchronized); drchase@5353: __ jcc(Assembler::notEqual, slow_path); drchase@5353: drchase@5353: // We don't generate local frame and don't align stack because drchase@5353: // we call stub code and there is no safepoint on this path. drchase@5353: drchase@5353: // Load parameters drchase@5353: const Register crc = rax; // crc drchase@5353: const Register buf = rdx; // source java byte array address drchase@5353: const Register len = rdi; // length drchase@5353: drchase@5353: // Arguments are reversed on java expression stack drchase@5353: __ movl(len, Address(rsp, wordSize)); // Length drchase@5353: // Calculate address of start element drchase@5353: if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { drchase@5353: __ movptr(buf, Address(rsp, 3*wordSize)); // long buf drchase@5353: __ addptr(buf, Address(rsp, 2*wordSize)); // + offset drchase@5353: __ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC drchase@5353: } else { drchase@5353: __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array drchase@5353: __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size drchase@5353: __ addptr(buf, Address(rsp, 2*wordSize)); // + offset drchase@5353: __ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC drchase@5353: } drchase@5353: drchase@5353: __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len); drchase@5353: // result in rax drchase@5353: drchase@5353: // _areturn drchase@5353: __ pop(rdi); // get return address drchase@5353: __ mov(rsp, rsi); // set sp to sender sp drchase@5353: __ jmp(rdi); drchase@5353: drchase@5353: // generate a vanilla native entry as the slow path drchase@5353: __ bind(slow_path); drchase@5353: drchase@5353: (void) generate_native_entry(false); drchase@5353: drchase@5353: return entry; drchase@5353: } drchase@5353: return generate_native_entry(false); drchase@5353: } drchase@5353: duke@435: // duke@435: // Interpreter stub for calling a native method. (asm interpreter) duke@435: // This sets up a somewhat different looking stack for calling the native method duke@435: // than the typical interpreter frame setup. duke@435: // duke@435: 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: // rsi: sender sp duke@435: // rsi: previous interpreter state (C++ interpreter) must preserve duke@435: address entry_point = __ pc(); duke@435: jiangli@4338: const Address constMethod (rbx, Method::const_offset()); coleenp@4037: const Address access_flags (rbx, Method::access_flags_offset()); jiangli@4338: const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset()); 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 expression stack duke@435: // and the arguments are already on the stack and we only add a handful of words duke@435: // to the stack duke@435: coleenp@4037: // rbx,: Method* duke@435: // rcx: size of parameters duke@435: // rsi: sender sp duke@435: never@739: __ pop(rax); // get return address duke@435: // for natives the size of locals is zero duke@435: duke@435: // compute beginning of parameters (rdi) never@739: __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize)); duke@435: duke@435: duke@435: // add 2 zero-initialized slots for native calls duke@435: // NULL result handler never@739: __ push((int32_t)NULL_WORD); duke@435: // NULL oop temp (mirror or jni oop result) never@739: __ push((int32_t)NULL_WORD); 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: { 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: __ get_thread(rax); duke@435: const Address do_not_unlock_if_synchronized(rax, 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: __ get_thread(rax); 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: // duke@435: if (synchronized) { duke@435: lock_method(); duke@435: } else { duke@435: // no synchronization necessary duke@435: #ifdef ASSERT 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: { 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/dtrace support duke@435: __ notify_method_entry(); duke@435: duke@435: // work registers duke@435: const Register method = rbx; duke@435: const Register thread = rdi; duke@435: const Register t = rcx; 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())); jiangli@4338: twisti@1861: __ shlptr(t, Interpreter::logStackElementSize); never@739: __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror never@739: __ subptr(rsp, t); never@739: __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics duke@435: duke@435: // get signature handler 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, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), 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() == rdi, "adjust this code"); duke@435: assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code"); duke@435: assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code"); 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 will blow RBX duke@435: // sometime, so we must reload it after the call. duke@435: __ call(t); duke@435: __ get_method(method); // slow path call blows RBX on DevStudio 5.0 duke@435: duke@435: // result handler is in rax, duke@435: // set result handler never@739: __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax); duke@435: duke@435: // pass mirror handle if static call 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), t); duke@435: // pass handle to mirror never@739: __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize)); never@739: __ movptr(Address(rsp, wordSize), t); duke@435: __ bind(L); duke@435: } duke@435: duke@435: // get native function entry point 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()); never@739: __ cmpptr(rax, unsatisfied.addr()); duke@435: __ jcc(Assembler::notEqual, L); duke@435: __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), 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 duke@435: __ get_thread(thread); never@739: __ lea(t, Address(thread, JavaThread::jni_environment_offset())); never@739: __ movptr(Address(rsp, 0), t); duke@435: duke@435: // set_last_Java_frame_before_call duke@435: // It is enough that the pc() duke@435: // points into the right code segment. It does not have to be the correct return pc. duke@435: __ set_last_Java_frame(thread, noreg, rbp, __ pc()); duke@435: duke@435: // change thread state duke@435: #ifdef ASSERT duke@435: { Label L; duke@435: __ movl(t, Address(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: __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native); duke@435: __ call(rax); duke@435: duke@435: // result potentially in rdx:rax or ST0 duke@435: kvn@4873: // Verify or restore cpu control state after JNI call kvn@4873: __ restore_cpu_control_state_after_jni(); duke@435: duke@435: // save potential result in ST(0) & rdx:rax duke@435: // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 - duke@435: // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers) duke@435: // It is safe to do this push because state is _thread_in_native and return address will be found duke@435: // via _last_native_pc and not via _last_jave_sp duke@435: duke@435: // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result. duke@435: // If the order changes or anything else is added to the stack the code in duke@435: // interpreter_frame_result will have to be changed. duke@435: duke@435: { Label L; duke@435: Label push_double; duke@435: ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT)); duke@435: ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE)); duke@435: __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize), duke@435: float_handler.addr()); duke@435: __ jcc(Assembler::equal, push_double); duke@435: __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize), duke@435: double_handler.addr()); duke@435: __ jcc(Assembler::notEqual, L); duke@435: __ bind(push_double); duke@435: __ push(dtos); duke@435: __ bind(L); duke@435: } duke@435: __ push(ltos); duke@435: duke@435: // change thread state duke@435: __ get_thread(thread); duke@435: __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans); duke@435: if(os::is_MP()) { duke@435: if (UseMembar) { never@739: // Force this write out before the read below never@739: __ membar(Assembler::Membar_mask_bits( never@739: Assembler::LoadLoad | Assembler::LoadStore | never@739: 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(thread, rcx); duke@435: } duke@435: } duke@435: duke@435: if (AlwaysRestoreFPU) { duke@435: // Make sure the control word is correct. duke@435: __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); duke@435: } duke@435: duke@435: // check for safepoint operation in progress and/or pending suspend requests duke@435: { Label Continue; duke@435: 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(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 and forward it duke@435: // and never return here preventing us from clearing _last_native_pc down below. duke@435: // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are duke@435: // preserved and correspond to the bcp/locals pointers. So we do a runtime call duke@435: // by hand. duke@435: // never@739: __ push(thread); duke@435: __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, duke@435: JavaThread::check_special_condition_for_native_trans))); duke@435: __ increment(rsp, wordSize); duke@435: __ get_thread(thread); duke@435: duke@435: __ bind(Continue); duke@435: } duke@435: duke@435: // change thread state duke@435: __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java); duke@435: kevinw@8877: __ reset_last_Java_frame(thread, true); duke@435: duke@435: // reset handle block never@739: __ movptr(t, Address(thread, JavaThread::active_handles_offset())); goetz@6558: __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD); duke@435: duke@435: // If result was an oop then unbox and save it in the frame phh@9669: { phh@9669: Label no_oop; duke@435: ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT)); duke@435: __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), duke@435: handler.addr()); duke@435: __ jcc(Assembler::notEqual, no_oop); never@739: __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD); duke@435: __ pop(ltos); phh@9669: // Unbox oop result, e.g. JNIHandles::resolve value. phh@9669: __ resolve_jobject(rax /* value */, phh@9669: thread /* thread */, phh@9669: t /* tmp */); 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 discarded duke@435: __ push(ltos); duke@435: __ bind(no_oop); duke@435: } duke@435: duke@435: { duke@435: Label no_reguard; duke@435: __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled); duke@435: __ jcc(Assembler::notEqual, no_reguard); duke@435: never@739: __ pusha(); duke@435: __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); never@739: __ popa(); duke@435: duke@435: __ bind(no_reguard); duke@435: } duke@435: duke@435: // restore rsi to have legal interpreter frame, duke@435: // i.e., bci == 0 <=> rsi == code_base() duke@435: // Can't call_VM until bcp is within reasonable. duke@435: __ get_method(method); // method is junk from thread_in_native to now. coleenp@4037: __ movptr(rsi, Address(method,Method::const_offset())); // get ConstMethod* coleenp@4037: __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase duke@435: duke@435: // handle exceptions (exception handling will handle unlocking!) duke@435: { Label L; never@739: __ cmpptr(Address(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 used in call_VM_base(); duke@435: // i.e., we should use the StubRoutines::forward_exception code. For now this duke@435: // doesn't work here because the rsp is not correctly set at this point. duke@435: __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, 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: { 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 assembler implementation duke@435: { Label unlock; duke@435: // BasicObjectLock will be first in list, since this is a synchronized method. However, need duke@435: // to check that the object has not been unlocked by an explicit monitorexit bytecode. duke@435: const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock)); duke@435: never@739: __ lea(rdx, monitor); // address of first monitor duke@435: never@739: __ movptr(t, Address(rdx, 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, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); duke@435: __ should_not_reach_here(); duke@435: duke@435: __ bind(unlock); duke@435: __ unlock_object(rdx); duke@435: } duke@435: __ bind(L); duke@435: } duke@435: duke@435: // jvmti/dtrace 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 rdx:rax, call result handler to restore potential result in ST0 & handle result duke@435: __ pop(ltos); never@739: __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize)); duke@435: __ call(t); duke@435: duke@435: // remove activation never@739: __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * 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: // rbx,: Method* duke@435: // rsi: sender sp duke@435: address entry_point = __ pc(); duke@435: jiangli@4338: const Address constMethod (rbx, Method::const_offset()); coleenp@4037: const Address access_flags (rbx, Method::access_flags_offset()); jiangli@4338: const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset()); jiangli@4338: const Address size_of_locals (rdx, ConstMethod::size_of_locals_offset()); 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: duke@435: // rsi: 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: // 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 (rdi) never@739: __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -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((int32_t)NULL_WORD); // initialize local variables duke@435: __ decrement(rdx); // until everything initialized duke@435: __ jcc(Assembler::greater, loop); duke@435: __ bind(exit); duke@435: } 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: { 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: __ get_thread(rax); duke@435: const Address do_not_unlock_if_synchronized(rax, duke@435: in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); duke@435: __ movbool(do_not_unlock_if_synchronized, true); duke@435: roland@5987: __ profile_parameters_type(rax, rcx, rdx); 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, &profile_method, &profile_method_continue); duke@435: if (ProfileInterpreter) { duke@435: __ bind(profile_method_continue); duke@435: } duke@435: } duke@435: Label continue_after_compile; duke@435: __ bind(continue_after_compile); duke@435: duke@435: bang_stack_shadow_pages(false); duke@435: duke@435: // reset the _do_not_unlock_if_synchronized flag duke@435: __ get_thread(rax); 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: // 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: { 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: { 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: //------------------------------------------------------------------------------------------------------------------------ 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 call method. duke@435: // These both come in synchronized and non-synchronized versions but the duke@435: // frame layout they create is very similar. The other method entry duke@435: // types are really just special purpose entries that are really entry duke@435: // and interpretation all in one. These are for trivial methods like duke@435: // 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: // rcx: receiver duke@435: // 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 duke@435: // entries the stack will look like below when we are ready to execute duke@435: // the first bytecode (or call the native routine). The register usage duke@435: // will be as the template based interpreter expects (see interpreter_x86.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 rsi ] duke@435: // [ current rdi ] coleenp@4037: // [ Method* ] duke@435: // [ saved rbp, ] <--- 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 ] <--- rdi duke@435: duke@435: address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) { duke@435: // determine code generation flags duke@435: bool synchronized = false; duke@435: address entry_point = NULL; drchase@5353: InterpreterGenerator* ig_this = (InterpreterGenerator*)this; duke@435: duke@435: switch (kind) { drchase@5353: case Interpreter::zerolocals : break; drchase@5353: case Interpreter::zerolocals_synchronized: synchronized = true; break; drchase@5353: case Interpreter::native : entry_point = ig_this->generate_native_entry(false); break; drchase@5353: case Interpreter::native_synchronized : entry_point = ig_this->generate_native_entry(true); break; drchase@5353: case Interpreter::empty : entry_point = ig_this->generate_empty_entry(); break; drchase@5353: case Interpreter::accessor : entry_point = ig_this->generate_accessor_entry(); break; drchase@5353: case Interpreter::abstract : entry_point = ig_this->generate_abstract_entry(); break; duke@435: duke@435: case Interpreter::java_lang_math_sin : // fall thru duke@435: case Interpreter::java_lang_math_cos : // fall thru duke@435: case Interpreter::java_lang_math_tan : // fall thru duke@435: case Interpreter::java_lang_math_abs : // fall thru duke@435: case Interpreter::java_lang_math_log : // fall thru duke@435: 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 drchase@5353: case Interpreter::java_lang_math_exp : entry_point = ig_this->generate_math_entry(kind); break; johnc@2781: case Interpreter::java_lang_ref_reference_get drchase@5353: : entry_point = ig_this->generate_Reference_get_entry(); break; drchase@5353: case Interpreter::java_util_zip_CRC32_update drchase@5353: : entry_point = ig_this->generate_CRC32_update_entry(); break; drchase@5353: case Interpreter::java_util_zip_CRC32_updateBytes drchase@5353: : // fall thru drchase@5353: case Interpreter::java_util_zip_CRC32_updateByteBuffer drchase@5353: : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); 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) return entry_point; duke@435: drchase@5353: return ig_this->generate_normal_entry(synchronized); duke@435: 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: duke@435: const int stub_code = 4; // see generate_call_stub duke@435: // Save space for one monitor to get into the interpreted method in case duke@435: // the method is synchronized duke@435: int monitor_size = method->is_synchronized() ? duke@435: 1*frame::interpreter_frame_monitor_size() : 0; duke@435: duke@435: // total overhead size: entry_size + (saved rbp, thru expr stack bottom). duke@435: // be sure to change this if you add/subtract anything to/from the overhead area duke@435: const int overhead_size = -frame::interpreter_frame_initial_sp_offset; duke@435: roland@5225: const int method_stack = (method->max_locals() + method->max_stack()) * twisti@1861: Interpreter::stackElementWords; duke@435: return overhead_size + method_stack + stub_code; 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 interpreted) duke@435: Interpreter::_rethrow_exception_entry = __ pc(); twisti@1730: const Register thread = rcx; duke@435: 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. xlu@947: __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); duke@435: // rax,: exception duke@435: // rdx: return address/pc that threw exception duke@435: __ restore_bcp(); // rsi points to call/send duke@435: __ restore_locals(); duke@435: 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: // rsi: exception bcp duke@435: __ verify_oop(rax); duke@435: duke@435: // expression stack must be empty before entering the VM in case of an exception duke@435: __ empty_expression_stack(); duke@435: __ empty_FPU_stack(); duke@435: // find exception handler address and preserve exception oop duke@435: __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax); duke@435: // rax,: exception handler entry point duke@435: // rdx: preserved exception oop duke@435: // rsi: 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 removed and duke@435: // the exception is rethrown (i.e. exception continuation is _rethrow_exception). duke@435: // duke@435: // Note: At this point the bci is still the bxi for the instruction which caused duke@435: // the exception and the expression stack is empty. Thus, for any VM calls duke@435: // at this point, GC will find a legal oop map (with empty expression stack). duke@435: duke@435: // In current activation duke@435: // tos: exception duke@435: // rsi: 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: __ empty_FPU_stack(); duke@435: // Set the popframe_processing bit in pending_popframe_condition indicating that we are duke@435: // currently handling popframe, so that call_VMs that may happen later do not trigger new duke@435: // popframe handling cycles. twisti@1730: __ get_thread(thread); twisti@1730: __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset())); duke@435: __ orl(rdx, JavaThread::popframe_processing_bit); twisti@1730: __ movl(Address(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(rdx, Address(rbp, frame::return_addr_offset * wordSize)); duke@435: __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx); 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 deoptimized caller duke@435: __ get_method(rax); jiangli@4338: __ movptr(rax, Address(rax, Method::const_offset())); jiangli@4338: __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset())); twisti@1861: __ shlptr(rax, Interpreter::logStackElementSize); duke@435: __ restore_locals(); never@739: __ subptr(rdi, rax); never@739: __ addptr(rdi, wordSize); duke@435: // Save these arguments twisti@1730: __ get_thread(thread); twisti@1730: __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi); 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 these arguments twisti@1730: __ get_thread(thread); twisti@1730: __ movl(Address(thread, JavaThread::popframe_condition_offset()), 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, 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(rax, rsp); never@739: __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); twisti@1730: __ get_thread(thread); duke@435: // PC must point into interpreter here twisti@1730: __ set_last_Java_frame(thread, noreg, rbp, __ pc()); twisti@1730: __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx); twisti@1730: __ get_thread(thread); kevinw@8877: __ reset_last_Java_frame(thread, true); duke@435: // Restore the last_sp and null it out never@739: __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); xlu@947: __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); duke@435: duke@435: __ restore_bcp(); duke@435: __ restore_locals(); 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 twisti@1730: __ get_thread(thread); twisti@1730: __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive); duke@435: sspitsyn@5496: #if INCLUDE_JVMTI sspitsyn@5496: if (EnableInvokeDynamic) { sspitsyn@5496: Label L_done; sspitsyn@5496: const Register local0 = rdi; sspitsyn@5496: sspitsyn@5496: __ cmpb(Address(rsi, 0), Bytecodes::_invokestatic); sspitsyn@5496: __ jcc(Assembler::notEqual, L_done); sspitsyn@5496: sspitsyn@5496: // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. sspitsyn@5496: // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL. sspitsyn@5496: sspitsyn@5496: __ get_method(rdx); sspitsyn@5496: __ movptr(rax, Address(local0, 0)); sspitsyn@5496: __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rsi); sspitsyn@5496: sspitsyn@5496: __ testptr(rax, rax); sspitsyn@5496: __ jcc(Assembler::zero, L_done); sspitsyn@5496: sspitsyn@5496: __ movptr(Address(rbx, 0), rax); sspitsyn@5496: __ bind(L_done); sspitsyn@5496: } sspitsyn@5496: #endif // INCLUDE_JVMTI sspitsyn@5496: 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); twisti@1730: __ get_thread(thread); twisti@1730: __ movptr(Address(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 twisti@1730: __ get_thread(thread); coleenp@4037: __ get_vm_result(rax, thread); duke@435: duke@435: // Inbetween activations - previous activation type unknown yet duke@435: // compute continuation point - the continuation point expects duke@435: // the following registers set up: duke@435: // twisti@1730: // rax: exception duke@435: // rdx: return address/pc that threw exception duke@435: // rsp: expression stack of caller twisti@1730: // rbp: rbp, of caller never@739: __ push(rax); // save exception never@739: __ push(rdx); // save return address twisti@1730: __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), 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 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(); twisti@1730: const Register thread = rcx; duke@435: duke@435: __ restore_bcp(); duke@435: __ restore_locals(); duke@435: __ empty_expression_stack(); duke@435: __ empty_FPU_stack(); duke@435: __ load_earlyret_value(state); duke@435: twisti@1730: __ get_thread(thread); twisti@1730: __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset())); duke@435: const Address cond_addr(rcx, 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: 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, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) { duke@435: assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); duke@435: Label L; duke@435: fep = __ pc(); __ push(ftos); __ jmp(L); duke@435: dep = __ pc(); __ push(dtos); __ jmp(L); duke@435: lep = __ pc(); __ push(ltos); __ jmp(L); duke@435: aep = __ pc(); __ push(atos); __ jmp(L); duke@435: bep = cep = sep = // fall through duke@435: iep = __ pc(); __ push(itos); // fall through duke@435: vep = __ pc(); __ bind(L); // fall through duke@435: generate_and_dispatch(t); 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: 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: // prepare expression stack never@739: __ pop(rcx); // pop return address so expression stack is 'pure' duke@435: __ push(state); // save tosca duke@435: duke@435: // pass tosca registers as arguments & call tracer duke@435: __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx); never@739: __ mov(rcx, rax); // make sure return address is not destroyed by pop(state) duke@435: __ pop(state); // restore tosca duke@435: duke@435: // return duke@435: __ jmp(rcx); duke@435: duke@435: return entry; duke@435: } duke@435: duke@435: duke@435: void TemplateInterpreterGenerator::count_bytecode() { never@739: __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value)); duke@435: } duke@435: duke@435: duke@435: void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { never@739: __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()])); duke@435: } duke@435: duke@435: duke@435: void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { duke@435: __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx); duke@435: __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes); duke@435: __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes); duke@435: ExternalAddress table((address) BytecodePairHistogram::_counters); duke@435: Address index(noreg, rbx, Address::times_4); never@739: __ incrementl(ArrayAddress(table, index)); 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: assert(Interpreter::trace_code(t->tos_in()) != NULL, duke@435: "entry must have been generated"); duke@435: __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in()))); 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 duke@435: #endif // CC_INTERP