duke@435: /* duke@435: * Copyright 1999-2007 Sun Microsystems, Inc. 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: * duke@435: * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, duke@435: * CA 95054 USA or visit www.sun.com if you need additional information or duke@435: * have any questions. duke@435: * duke@435: */ duke@435: duke@435: #include "incls/_precompiled.incl" duke@435: #include "incls/_c1_Runtime1_sparc.cpp.incl" duke@435: duke@435: // Implementation of StubAssembler duke@435: duke@435: int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry_point, int number_of_arguments) { duke@435: // for sparc changing the number of arguments doesn't change duke@435: // anything about the frame size so we'll always lie and claim that duke@435: // we are only passing 1 argument. duke@435: set_num_rt_args(1); duke@435: duke@435: assert_not_delayed(); duke@435: // bang stack before going to runtime duke@435: set(-os::vm_page_size() + STACK_BIAS, G3_scratch); duke@435: st(G0, SP, G3_scratch); duke@435: duke@435: // debugging support duke@435: assert(number_of_arguments >= 0 , "cannot have negative number of arguments"); duke@435: duke@435: set_last_Java_frame(SP, noreg); duke@435: if (VerifyThread) mov(G2_thread, O0); // about to be smashed; pass early duke@435: save_thread(L7_thread_cache); duke@435: // do the call duke@435: call(entry_point, relocInfo::runtime_call_type); duke@435: if (!VerifyThread) { duke@435: delayed()->mov(G2_thread, O0); // pass thread as first argument duke@435: } else { duke@435: delayed()->nop(); // (thread already passed) duke@435: } duke@435: int call_offset = offset(); // offset of return address duke@435: restore_thread(L7_thread_cache); duke@435: reset_last_Java_frame(); duke@435: duke@435: // check for pending exceptions duke@435: { Label L; duke@435: Address exception_addr(G2_thread, 0, in_bytes(Thread::pending_exception_offset())); duke@435: ld_ptr(exception_addr, Gtemp); duke@435: br_null(Gtemp, false, pt, L); duke@435: delayed()->nop(); duke@435: Address vm_result_addr(G2_thread, 0, in_bytes(JavaThread::vm_result_offset())); duke@435: st_ptr(G0, vm_result_addr); duke@435: Address vm_result_addr_2(G2_thread, 0, in_bytes(JavaThread::vm_result_2_offset())); duke@435: st_ptr(G0, vm_result_addr_2); duke@435: duke@435: if (frame_size() == no_frame_size) { duke@435: // we use O7 linkage so that forward_exception_entry has the issuing PC duke@435: call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type); duke@435: delayed()->restore(); duke@435: } else if (_stub_id == Runtime1::forward_exception_id) { duke@435: should_not_reach_here(); duke@435: } else { duke@435: Address exc(G4, Runtime1::entry_for(Runtime1::forward_exception_id)); duke@435: jump_to(exc, 0); duke@435: delayed()->nop(); duke@435: } duke@435: bind(L); duke@435: } duke@435: duke@435: // get oop result if there is one and reset the value in the thread duke@435: if (oop_result1->is_valid()) { // get oop result if there is one and reset it in the thread duke@435: get_vm_result (oop_result1); duke@435: } else { duke@435: // be a little paranoid and clear the result duke@435: Address vm_result_addr(G2_thread, 0, in_bytes(JavaThread::vm_result_offset())); duke@435: st_ptr(G0, vm_result_addr); duke@435: } duke@435: duke@435: if (oop_result2->is_valid()) { duke@435: get_vm_result_2(oop_result2); duke@435: } else { duke@435: // be a little paranoid and clear the result duke@435: Address vm_result_addr_2(G2_thread, 0, in_bytes(JavaThread::vm_result_2_offset())); duke@435: st_ptr(G0, vm_result_addr_2); duke@435: } duke@435: duke@435: return call_offset; duke@435: } duke@435: duke@435: duke@435: int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1) { duke@435: // O0 is reserved for the thread duke@435: mov(arg1, O1); duke@435: return call_RT(oop_result1, oop_result2, entry, 1); duke@435: } duke@435: duke@435: duke@435: int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2) { duke@435: // O0 is reserved for the thread duke@435: mov(arg1, O1); duke@435: mov(arg2, O2); assert(arg2 != O1, "smashed argument"); duke@435: return call_RT(oop_result1, oop_result2, entry, 2); duke@435: } duke@435: duke@435: duke@435: int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2, Register arg3) { duke@435: // O0 is reserved for the thread duke@435: mov(arg1, O1); duke@435: mov(arg2, O2); assert(arg2 != O1, "smashed argument"); duke@435: mov(arg3, O3); assert(arg3 != O1 && arg3 != O2, "smashed argument"); duke@435: return call_RT(oop_result1, oop_result2, entry, 3); duke@435: } duke@435: duke@435: duke@435: // Implementation of Runtime1 duke@435: duke@435: #define __ sasm-> duke@435: duke@435: static int cpu_reg_save_offsets[FrameMap::nof_cpu_regs]; duke@435: static int fpu_reg_save_offsets[FrameMap::nof_fpu_regs]; duke@435: static int reg_save_size_in_words; duke@435: static int frame_size_in_bytes = -1; duke@435: duke@435: static OopMap* generate_oop_map(StubAssembler* sasm, bool save_fpu_registers) { duke@435: assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words), duke@435: " mismatch in calculation"); duke@435: sasm->set_frame_size(frame_size_in_bytes / BytesPerWord); duke@435: int frame_size_in_slots = frame_size_in_bytes / sizeof(jint); duke@435: OopMap* oop_map = new OopMap(frame_size_in_slots, 0); duke@435: duke@435: int i; duke@435: for (i = 0; i < FrameMap::nof_cpu_regs; i++) { duke@435: Register r = as_Register(i); duke@435: if (r == G1 || r == G3 || r == G4 || r == G5) { duke@435: int sp_offset = cpu_reg_save_offsets[i]; duke@435: oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset), duke@435: r->as_VMReg()); duke@435: } duke@435: } duke@435: duke@435: if (save_fpu_registers) { duke@435: for (i = 0; i < FrameMap::nof_fpu_regs; i++) { duke@435: FloatRegister r = as_FloatRegister(i); duke@435: int sp_offset = fpu_reg_save_offsets[i]; duke@435: oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset), duke@435: r->as_VMReg()); duke@435: } duke@435: } duke@435: return oop_map; duke@435: } duke@435: duke@435: static OopMap* save_live_registers(StubAssembler* sasm, bool save_fpu_registers = true) { duke@435: assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words), duke@435: " mismatch in calculation"); duke@435: __ save_frame_c1(frame_size_in_bytes); duke@435: sasm->set_frame_size(frame_size_in_bytes / BytesPerWord); duke@435: duke@435: // Record volatile registers as callee-save values in an OopMap so their save locations will be duke@435: // propagated to the caller frame's RegisterMap during StackFrameStream construction (needed for duke@435: // deoptimization; see compiledVFrame::create_stack_value). The caller's I, L and O registers duke@435: // are saved in register windows - I's and L's in the caller's frame and O's in the stub frame duke@435: // (as the stub's I's) when the runtime routine called by the stub creates its frame. duke@435: // OopMap frame sizes are in c2 stack slot sizes (sizeof(jint)) duke@435: duke@435: int i; duke@435: for (i = 0; i < FrameMap::nof_cpu_regs; i++) { duke@435: Register r = as_Register(i); duke@435: if (r == G1 || r == G3 || r == G4 || r == G5) { duke@435: int sp_offset = cpu_reg_save_offsets[i]; duke@435: __ st_ptr(r, SP, (sp_offset * BytesPerWord) + STACK_BIAS); duke@435: } duke@435: } duke@435: duke@435: if (save_fpu_registers) { duke@435: for (i = 0; i < FrameMap::nof_fpu_regs; i++) { duke@435: FloatRegister r = as_FloatRegister(i); duke@435: int sp_offset = fpu_reg_save_offsets[i]; duke@435: __ stf(FloatRegisterImpl::S, r, SP, (sp_offset * BytesPerWord) + STACK_BIAS); duke@435: } duke@435: } duke@435: duke@435: return generate_oop_map(sasm, save_fpu_registers); duke@435: } duke@435: duke@435: static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) { duke@435: for (int i = 0; i < FrameMap::nof_cpu_regs; i++) { duke@435: Register r = as_Register(i); duke@435: if (r == G1 || r == G3 || r == G4 || r == G5) { duke@435: __ ld_ptr(SP, (cpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r); duke@435: } duke@435: } duke@435: duke@435: if (restore_fpu_registers) { duke@435: for (int i = 0; i < FrameMap::nof_fpu_regs; i++) { duke@435: FloatRegister r = as_FloatRegister(i); duke@435: __ ldf(FloatRegisterImpl::S, SP, (fpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r); duke@435: } duke@435: } duke@435: } duke@435: duke@435: duke@435: void Runtime1::initialize_pd() { duke@435: // compute word offsets from SP at which live (non-windowed) registers are captured by stub routines duke@435: // duke@435: // A stub routine will have a frame that is at least large enough to hold duke@435: // a register window save area (obviously) and the volatile g registers duke@435: // and floating registers. A user of save_live_registers can have a frame duke@435: // that has more scratch area in it (although typically they will use L-regs). duke@435: // in that case the frame will look like this (stack growing down) duke@435: // duke@435: // FP -> | | duke@435: // | scratch mem | duke@435: // | " " | duke@435: // -------------- duke@435: // | float regs | duke@435: // | " " | duke@435: // --------------- duke@435: // | G regs | duke@435: // | " " | duke@435: // --------------- duke@435: // | abi reg. | duke@435: // | window save | duke@435: // | area | duke@435: // SP -> --------------- duke@435: // duke@435: int i; duke@435: int sp_offset = round_to(frame::register_save_words, 2); // start doubleword aligned duke@435: duke@435: // only G int registers are saved explicitly; others are found in register windows duke@435: for (i = 0; i < FrameMap::nof_cpu_regs; i++) { duke@435: Register r = as_Register(i); duke@435: if (r == G1 || r == G3 || r == G4 || r == G5) { duke@435: cpu_reg_save_offsets[i] = sp_offset; duke@435: sp_offset++; duke@435: } duke@435: } duke@435: duke@435: // all float registers are saved explicitly duke@435: assert(FrameMap::nof_fpu_regs == 32, "double registers not handled here"); duke@435: for (i = 0; i < FrameMap::nof_fpu_regs; i++) { duke@435: fpu_reg_save_offsets[i] = sp_offset; duke@435: sp_offset++; duke@435: } duke@435: reg_save_size_in_words = sp_offset - frame::memory_parameter_word_sp_offset; duke@435: // this should match assembler::total_frame_size_in_bytes, which duke@435: // isn't callable from this context. It's checked by an assert when duke@435: // it's used though. duke@435: frame_size_in_bytes = align_size_up(sp_offset * wordSize, 8); duke@435: } duke@435: duke@435: duke@435: OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) { duke@435: // make a frame and preserve the caller's caller-save registers duke@435: OopMap* oop_map = save_live_registers(sasm); duke@435: int call_offset; duke@435: if (!has_argument) { duke@435: call_offset = __ call_RT(noreg, noreg, target); duke@435: } else { duke@435: call_offset = __ call_RT(noreg, noreg, target, G4); duke@435: } duke@435: OopMapSet* oop_maps = new OopMapSet(); duke@435: oop_maps->add_gc_map(call_offset, oop_map); duke@435: duke@435: __ should_not_reach_here(); duke@435: return oop_maps; duke@435: } duke@435: duke@435: duke@435: OopMapSet* Runtime1::generate_stub_call(StubAssembler* sasm, Register result, address target, duke@435: Register arg1, Register arg2, Register arg3) { duke@435: // make a frame and preserve the caller's caller-save registers duke@435: OopMap* oop_map = save_live_registers(sasm); duke@435: duke@435: int call_offset; duke@435: if (arg1 == noreg) { duke@435: call_offset = __ call_RT(result, noreg, target); duke@435: } else if (arg2 == noreg) { duke@435: call_offset = __ call_RT(result, noreg, target, arg1); duke@435: } else if (arg3 == noreg) { duke@435: call_offset = __ call_RT(result, noreg, target, arg1, arg2); duke@435: } else { duke@435: call_offset = __ call_RT(result, noreg, target, arg1, arg2, arg3); duke@435: } duke@435: OopMapSet* oop_maps = NULL; duke@435: duke@435: oop_maps = new OopMapSet(); duke@435: oop_maps->add_gc_map(call_offset, oop_map); duke@435: restore_live_registers(sasm); duke@435: duke@435: __ ret(); duke@435: __ delayed()->restore(); duke@435: duke@435: return oop_maps; duke@435: } duke@435: duke@435: duke@435: OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) { duke@435: // make a frame and preserve the caller's caller-save registers duke@435: OopMap* oop_map = save_live_registers(sasm); duke@435: duke@435: // call the runtime patching routine, returns non-zero if nmethod got deopted. duke@435: int call_offset = __ call_RT(noreg, noreg, target); duke@435: OopMapSet* oop_maps = new OopMapSet(); duke@435: oop_maps->add_gc_map(call_offset, oop_map); duke@435: duke@435: // re-execute the patched instruction or, if the nmethod was deoptmized, return to the duke@435: // deoptimization handler entry that will cause re-execution of the current bytecode duke@435: DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob(); duke@435: assert(deopt_blob != NULL, "deoptimization blob must have been created"); duke@435: duke@435: Label no_deopt; duke@435: __ tst(O0); duke@435: __ brx(Assembler::equal, false, Assembler::pt, no_deopt); duke@435: __ delayed()->nop(); duke@435: duke@435: // return to the deoptimization handler entry for unpacking and rexecute duke@435: // if we simply returned the we'd deopt as if any call we patched had just duke@435: // returned. duke@435: duke@435: restore_live_registers(sasm); duke@435: __ restore(); duke@435: __ br(Assembler::always, false, Assembler::pt, deopt_blob->unpack_with_reexecution(), relocInfo::runtime_call_type); duke@435: __ delayed()->nop(); duke@435: duke@435: __ bind(no_deopt); duke@435: restore_live_registers(sasm); duke@435: __ ret(); duke@435: __ delayed()->restore(); duke@435: duke@435: return oop_maps; duke@435: } duke@435: duke@435: OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) { duke@435: duke@435: OopMapSet* oop_maps = NULL; duke@435: // for better readability duke@435: const bool must_gc_arguments = true; duke@435: const bool dont_gc_arguments = false; duke@435: duke@435: // stub code & info for the different stubs duke@435: switch (id) { duke@435: case forward_exception_id: duke@435: { duke@435: // we're handling an exception in the context of a compiled duke@435: // frame. The registers have been saved in the standard duke@435: // places. Perform an exception lookup in the caller and duke@435: // dispatch to the handler if found. Otherwise unwind and duke@435: // dispatch to the callers exception handler. duke@435: duke@435: oop_maps = new OopMapSet(); duke@435: OopMap* oop_map = generate_oop_map(sasm, true); duke@435: duke@435: // transfer the pending exception to the exception_oop duke@435: __ ld_ptr(G2_thread, in_bytes(JavaThread::pending_exception_offset()), Oexception); duke@435: __ ld_ptr(Oexception, 0, G0); duke@435: __ st_ptr(G0, G2_thread, in_bytes(JavaThread::pending_exception_offset())); duke@435: __ add(I7, frame::pc_return_offset, Oissuing_pc); duke@435: duke@435: generate_handle_exception(sasm, oop_maps, oop_map); duke@435: __ should_not_reach_here(); duke@435: } duke@435: break; duke@435: duke@435: case new_instance_id: duke@435: case fast_new_instance_id: duke@435: case fast_new_instance_init_check_id: duke@435: { duke@435: Register G5_klass = G5; // Incoming duke@435: Register O0_obj = O0; // Outgoing duke@435: duke@435: if (id == new_instance_id) { duke@435: __ set_info("new_instance", dont_gc_arguments); duke@435: } else if (id == fast_new_instance_id) { duke@435: __ set_info("fast new_instance", dont_gc_arguments); duke@435: } else { duke@435: assert(id == fast_new_instance_init_check_id, "bad StubID"); duke@435: __ set_info("fast new_instance init check", dont_gc_arguments); duke@435: } duke@435: duke@435: if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) && duke@435: UseTLAB && FastTLABRefill) { duke@435: Label slow_path; duke@435: Register G1_obj_size = G1; duke@435: Register G3_t1 = G3; duke@435: Register G4_t2 = G4; duke@435: assert_different_registers(G5_klass, G1_obj_size, G3_t1, G4_t2); duke@435: duke@435: // Push a frame since we may do dtrace notification for the duke@435: // allocation which requires calling out and we don't want duke@435: // to stomp the real return address. duke@435: __ save_frame(0); duke@435: duke@435: if (id == fast_new_instance_init_check_id) { duke@435: // make sure the klass is initialized duke@435: __ ld(G5_klass, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc), G3_t1); duke@435: __ cmp(G3_t1, instanceKlass::fully_initialized); duke@435: __ br(Assembler::notEqual, false, Assembler::pn, slow_path); duke@435: __ delayed()->nop(); duke@435: } duke@435: #ifdef ASSERT duke@435: // assert object can be fast path allocated duke@435: { duke@435: Label ok, not_ok; duke@435: __ ld(G5_klass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), G1_obj_size); duke@435: __ cmp(G1_obj_size, 0); // make sure it's an instance (LH > 0) duke@435: __ br(Assembler::lessEqual, false, Assembler::pn, not_ok); duke@435: __ delayed()->nop(); duke@435: __ btst(Klass::_lh_instance_slow_path_bit, G1_obj_size); duke@435: __ br(Assembler::zero, false, Assembler::pn, ok); duke@435: __ delayed()->nop(); duke@435: __ bind(not_ok); duke@435: __ stop("assert(can be fast path allocated)"); duke@435: __ should_not_reach_here(); duke@435: __ bind(ok); duke@435: } duke@435: #endif // ASSERT duke@435: // if we got here then the TLAB allocation failed, so try duke@435: // refilling the TLAB or allocating directly from eden. duke@435: Label retry_tlab, try_eden; duke@435: __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G5_klass duke@435: duke@435: __ bind(retry_tlab); duke@435: duke@435: // get the instance size duke@435: __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size); duke@435: __ tlab_allocate(O0_obj, G1_obj_size, 0, G3_t1, slow_path); duke@435: __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2); duke@435: __ verify_oop(O0_obj); duke@435: __ mov(O0, I0); duke@435: __ ret(); duke@435: __ delayed()->restore(); duke@435: duke@435: __ bind(try_eden); duke@435: // get the instance size duke@435: __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size); duke@435: __ eden_allocate(O0_obj, G1_obj_size, 0, G3_t1, G4_t2, slow_path); duke@435: __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2); duke@435: __ verify_oop(O0_obj); duke@435: __ mov(O0, I0); duke@435: __ ret(); duke@435: __ delayed()->restore(); duke@435: duke@435: __ bind(slow_path); duke@435: duke@435: // pop this frame so generate_stub_call can push it's own duke@435: __ restore(); duke@435: } duke@435: duke@435: oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_instance), G5_klass); duke@435: // I0->O0: new instance duke@435: } duke@435: duke@435: break; duke@435: duke@435: #ifdef TIERED duke@435: case counter_overflow_id: duke@435: // G4 contains bci duke@435: oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, counter_overflow), G4); duke@435: break; duke@435: #endif // TIERED duke@435: duke@435: case new_type_array_id: duke@435: case new_object_array_id: duke@435: { duke@435: Register G5_klass = G5; // Incoming duke@435: Register G4_length = G4; // Incoming duke@435: Register O0_obj = O0; // Outgoing duke@435: duke@435: Address klass_lh(G5_klass, 0, ((klassOopDesc::header_size() * HeapWordSize) duke@435: + Klass::layout_helper_offset_in_bytes())); duke@435: assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise"); duke@435: assert(Klass::_lh_header_size_mask == 0xFF, "bytewise"); duke@435: // Use this offset to pick out an individual byte of the layout_helper: duke@435: const int klass_lh_header_size_offset = ((BytesPerInt - 1) // 3 - 2 selects byte {0,1,0,0} duke@435: - Klass::_lh_header_size_shift / BitsPerByte); duke@435: duke@435: if (id == new_type_array_id) { duke@435: __ set_info("new_type_array", dont_gc_arguments); duke@435: } else { duke@435: __ set_info("new_object_array", dont_gc_arguments); duke@435: } duke@435: duke@435: #ifdef ASSERT duke@435: // assert object type is really an array of the proper kind duke@435: { duke@435: Label ok; duke@435: Register G3_t1 = G3; duke@435: __ ld(klass_lh, G3_t1); duke@435: __ sra(G3_t1, Klass::_lh_array_tag_shift, G3_t1); duke@435: int tag = ((id == new_type_array_id) duke@435: ? Klass::_lh_array_tag_type_value duke@435: : Klass::_lh_array_tag_obj_value); duke@435: __ cmp(G3_t1, tag); duke@435: __ brx(Assembler::equal, false, Assembler::pt, ok); duke@435: __ delayed()->nop(); duke@435: __ stop("assert(is an array klass)"); duke@435: __ should_not_reach_here(); duke@435: __ bind(ok); duke@435: } duke@435: #endif // ASSERT duke@435: duke@435: if (UseTLAB && FastTLABRefill) { duke@435: Label slow_path; duke@435: Register G1_arr_size = G1; duke@435: Register G3_t1 = G3; duke@435: Register O1_t2 = O1; duke@435: assert_different_registers(G5_klass, G4_length, G1_arr_size, G3_t1, O1_t2); duke@435: duke@435: // check that array length is small enough for fast path duke@435: __ set(C1_MacroAssembler::max_array_allocation_length, G3_t1); duke@435: __ cmp(G4_length, G3_t1); duke@435: __ br(Assembler::greaterUnsigned, false, Assembler::pn, slow_path); duke@435: __ delayed()->nop(); duke@435: duke@435: // if we got here then the TLAB allocation failed, so try duke@435: // refilling the TLAB or allocating directly from eden. duke@435: Label retry_tlab, try_eden; duke@435: __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G4_length and G5_klass duke@435: duke@435: __ bind(retry_tlab); duke@435: duke@435: // get the allocation size: (length << (layout_helper & 0x1F)) + header_size duke@435: __ ld(klass_lh, G3_t1); duke@435: __ sll(G4_length, G3_t1, G1_arr_size); duke@435: __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1); duke@435: __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1); duke@435: __ add(G1_arr_size, G3_t1, G1_arr_size); duke@435: __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size); // align up duke@435: __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size); duke@435: duke@435: __ tlab_allocate(O0_obj, G1_arr_size, 0, G3_t1, slow_path); // preserves G1_arr_size duke@435: duke@435: __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2); duke@435: __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset); duke@435: __ sub(G1_arr_size, G3_t1, O1_t2); // body length duke@435: __ add(O0_obj, G3_t1, G3_t1); // body start duke@435: __ initialize_body(G3_t1, O1_t2); duke@435: __ verify_oop(O0_obj); duke@435: __ retl(); duke@435: __ delayed()->nop(); duke@435: duke@435: __ bind(try_eden); duke@435: // get the allocation size: (length << (layout_helper & 0x1F)) + header_size duke@435: __ ld(klass_lh, G3_t1); duke@435: __ sll(G4_length, G3_t1, G1_arr_size); duke@435: __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1); duke@435: __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1); duke@435: __ add(G1_arr_size, G3_t1, G1_arr_size); duke@435: __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size); duke@435: __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size); duke@435: duke@435: __ eden_allocate(O0_obj, G1_arr_size, 0, G3_t1, O1_t2, slow_path); // preserves G1_arr_size duke@435: duke@435: __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2); duke@435: __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset); duke@435: __ sub(G1_arr_size, G3_t1, O1_t2); // body length duke@435: __ add(O0_obj, G3_t1, G3_t1); // body start duke@435: __ initialize_body(G3_t1, O1_t2); duke@435: __ verify_oop(O0_obj); duke@435: __ retl(); duke@435: __ delayed()->nop(); duke@435: duke@435: __ bind(slow_path); duke@435: } duke@435: duke@435: if (id == new_type_array_id) { duke@435: oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_type_array), G5_klass, G4_length); duke@435: } else { duke@435: oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_object_array), G5_klass, G4_length); duke@435: } duke@435: // I0 -> O0: new array duke@435: } duke@435: break; duke@435: duke@435: case new_multi_array_id: duke@435: { // O0: klass duke@435: // O1: rank duke@435: // O2: address of 1st dimension duke@435: __ set_info("new_multi_array", dont_gc_arguments); duke@435: oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_multi_array), I0, I1, I2); duke@435: // I0 -> O0: new multi array duke@435: } duke@435: break; duke@435: duke@435: case register_finalizer_id: duke@435: { duke@435: __ set_info("register_finalizer", dont_gc_arguments); duke@435: duke@435: // load the klass and check the has finalizer flag duke@435: Label register_finalizer; duke@435: Register t = O1; duke@435: __ ld_ptr(O0, oopDesc::klass_offset_in_bytes(), t); duke@435: __ ld(t, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc), t); duke@435: __ set(JVM_ACC_HAS_FINALIZER, G3); duke@435: __ andcc(G3, t, G0); duke@435: __ br(Assembler::notZero, false, Assembler::pt, register_finalizer); duke@435: __ delayed()->nop(); duke@435: duke@435: // do a leaf return duke@435: __ retl(); duke@435: __ delayed()->nop(); duke@435: duke@435: __ bind(register_finalizer); duke@435: OopMap* oop_map = save_live_registers(sasm); duke@435: int call_offset = __ call_RT(noreg, noreg, duke@435: CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), I0); duke@435: oop_maps = new OopMapSet(); duke@435: oop_maps->add_gc_map(call_offset, oop_map); duke@435: duke@435: // Now restore all the live registers duke@435: restore_live_registers(sasm); duke@435: duke@435: __ ret(); duke@435: __ delayed()->restore(); duke@435: } duke@435: break; duke@435: duke@435: case throw_range_check_failed_id: duke@435: { __ set_info("range_check_failed", dont_gc_arguments); // arguments will be discarded duke@435: // G4: index duke@435: oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true); duke@435: } duke@435: break; duke@435: duke@435: case throw_index_exception_id: duke@435: { __ set_info("index_range_check_failed", dont_gc_arguments); // arguments will be discarded duke@435: // G4: index duke@435: oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true); duke@435: } duke@435: break; duke@435: duke@435: case throw_div0_exception_id: duke@435: { __ set_info("throw_div0_exception", dont_gc_arguments); duke@435: oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false); duke@435: } duke@435: break; duke@435: duke@435: case throw_null_pointer_exception_id: duke@435: { __ set_info("throw_null_pointer_exception", dont_gc_arguments); duke@435: oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false); duke@435: } duke@435: break; duke@435: duke@435: case handle_exception_id: duke@435: { duke@435: __ set_info("handle_exception", dont_gc_arguments); duke@435: // make a frame and preserve the caller's caller-save registers duke@435: duke@435: oop_maps = new OopMapSet(); duke@435: OopMap* oop_map = save_live_registers(sasm); duke@435: __ mov(Oexception->after_save(), Oexception); duke@435: __ mov(Oissuing_pc->after_save(), Oissuing_pc); duke@435: generate_handle_exception(sasm, oop_maps, oop_map); duke@435: } duke@435: break; duke@435: duke@435: case unwind_exception_id: duke@435: { duke@435: // O0: exception duke@435: // I7: address of call to this method duke@435: duke@435: __ set_info("unwind_exception", dont_gc_arguments); duke@435: __ mov(Oexception, Oexception->after_save()); duke@435: __ add(I7, frame::pc_return_offset, Oissuing_pc->after_save()); duke@435: duke@435: __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), duke@435: Oissuing_pc->after_save()); duke@435: __ verify_not_null_oop(Oexception->after_save()); duke@435: __ jmp(O0, 0); duke@435: __ delayed()->restore(); duke@435: } duke@435: break; duke@435: duke@435: case throw_array_store_exception_id: duke@435: { duke@435: __ set_info("throw_array_store_exception", dont_gc_arguments); duke@435: oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), false); duke@435: } duke@435: break; duke@435: duke@435: case throw_class_cast_exception_id: duke@435: { duke@435: // G4: object duke@435: __ set_info("throw_class_cast_exception", dont_gc_arguments); duke@435: oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true); duke@435: } duke@435: break; duke@435: duke@435: case throw_incompatible_class_change_error_id: duke@435: { duke@435: __ set_info("throw_incompatible_class_cast_exception", dont_gc_arguments); duke@435: oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false); duke@435: } duke@435: break; duke@435: duke@435: case slow_subtype_check_id: duke@435: { // Support for uint StubRoutine::partial_subtype_check( Klass sub, Klass super ); duke@435: // Arguments : duke@435: // duke@435: // ret : G3 duke@435: // sub : G3, argument, destroyed duke@435: // super: G1, argument, not changed duke@435: // raddr: O7, blown by call jrose@1079: Label miss; duke@435: duke@435: __ save_frame(0); // Blow no registers! duke@435: jrose@1079: __ check_klass_subtype_slow_path(G3, G1, L0, L1, L2, L4, NULL, &miss); duke@435: duke@435: __ mov(1, G3); jrose@1079: __ ret(); // Result in G5 is 'true' duke@435: __ delayed()->restore(); // free copy or add can go here duke@435: duke@435: __ bind(miss); duke@435: __ mov(0, G3); jrose@1079: __ ret(); // Result in G5 is 'false' duke@435: __ delayed()->restore(); // free copy or add can go here duke@435: } duke@435: duke@435: case monitorenter_nofpu_id: duke@435: case monitorenter_id: duke@435: { // G4: object duke@435: // G5: lock address duke@435: __ set_info("monitorenter", dont_gc_arguments); duke@435: duke@435: int save_fpu_registers = (id == monitorenter_id); duke@435: // make a frame and preserve the caller's caller-save registers duke@435: OopMap* oop_map = save_live_registers(sasm, save_fpu_registers); duke@435: duke@435: int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), G4, G5); duke@435: duke@435: oop_maps = new OopMapSet(); duke@435: oop_maps->add_gc_map(call_offset, oop_map); duke@435: restore_live_registers(sasm, save_fpu_registers); duke@435: duke@435: __ ret(); duke@435: __ delayed()->restore(); duke@435: } duke@435: break; duke@435: duke@435: case monitorexit_nofpu_id: duke@435: case monitorexit_id: duke@435: { // G4: lock address duke@435: // note: really a leaf routine but must setup last java sp duke@435: // => use call_RT for now (speed can be improved by duke@435: // doing last java sp setup manually) duke@435: __ set_info("monitorexit", dont_gc_arguments); duke@435: duke@435: int save_fpu_registers = (id == monitorexit_id); duke@435: // make a frame and preserve the caller's caller-save registers duke@435: OopMap* oop_map = save_live_registers(sasm, save_fpu_registers); duke@435: duke@435: int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), G4); duke@435: duke@435: oop_maps = new OopMapSet(); duke@435: oop_maps->add_gc_map(call_offset, oop_map); duke@435: restore_live_registers(sasm, save_fpu_registers); duke@435: duke@435: __ ret(); duke@435: __ delayed()->restore(); duke@435: duke@435: } duke@435: break; duke@435: duke@435: case access_field_patching_id: duke@435: { __ set_info("access_field_patching", dont_gc_arguments); duke@435: oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching)); duke@435: } duke@435: break; duke@435: duke@435: case load_klass_patching_id: duke@435: { __ set_info("load_klass_patching", dont_gc_arguments); duke@435: oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching)); duke@435: } duke@435: break; duke@435: duke@435: case jvmti_exception_throw_id: duke@435: { // Oexception : exception duke@435: __ set_info("jvmti_exception_throw", dont_gc_arguments); duke@435: oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, Runtime1::post_jvmti_exception_throw), I0); duke@435: } duke@435: break; duke@435: duke@435: case dtrace_object_alloc_id: duke@435: { // O0: object duke@435: __ set_info("dtrace_object_alloc", dont_gc_arguments); duke@435: // we can't gc here so skip the oopmap but make sure that all duke@435: // the live registers get saved. duke@435: save_live_registers(sasm); duke@435: duke@435: __ save_thread(L7_thread_cache); duke@435: __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), duke@435: relocInfo::runtime_call_type); duke@435: __ delayed()->mov(I0, O0); duke@435: __ restore_thread(L7_thread_cache); duke@435: duke@435: restore_live_registers(sasm); duke@435: __ ret(); duke@435: __ delayed()->restore(); duke@435: } duke@435: break; duke@435: ysr@777: #ifndef SERIALGC ysr@777: case g1_pre_barrier_slow_id: ysr@777: { // G4: previous value of memory ysr@777: BarrierSet* bs = Universe::heap()->barrier_set(); ysr@777: if (bs->kind() != BarrierSet::G1SATBCTLogging) { ysr@777: __ save_frame(0); ysr@777: __ set((int)id, O1); ysr@777: __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0); ysr@777: __ should_not_reach_here(); ysr@777: break; ysr@777: } ysr@777: ysr@777: __ set_info("g1_pre_barrier_slow_id", dont_gc_arguments); ysr@777: ysr@777: Register pre_val = G4; ysr@777: Register tmp = G1_scratch; ysr@777: Register tmp2 = G3_scratch; ysr@777: ysr@777: Label refill, restart; ysr@777: bool with_frame = false; // I don't know if we can do with-frame. ysr@777: int satb_q_index_byte_offset = ysr@777: in_bytes(JavaThread::satb_mark_queue_offset() + ysr@777: PtrQueue::byte_offset_of_index()); ysr@777: int satb_q_buf_byte_offset = ysr@777: in_bytes(JavaThread::satb_mark_queue_offset() + ysr@777: PtrQueue::byte_offset_of_buf()); ysr@777: __ bind(restart); ysr@777: __ ld_ptr(G2_thread, satb_q_index_byte_offset, tmp); ysr@777: ysr@777: __ br_on_reg_cond(Assembler::rc_z, /*annul*/false, ysr@777: Assembler::pn, tmp, refill); ysr@777: ysr@777: // If the branch is taken, no harm in executing this in the delay slot. ysr@777: __ delayed()->ld_ptr(G2_thread, satb_q_buf_byte_offset, tmp2); ysr@777: __ sub(tmp, oopSize, tmp); ysr@777: ysr@777: __ st_ptr(pre_val, tmp2, tmp); // [_buf + index] := ysr@777: // Use return-from-leaf ysr@777: __ retl(); ysr@777: __ delayed()->st_ptr(tmp, G2_thread, satb_q_index_byte_offset); ysr@777: ysr@777: __ bind(refill); ysr@777: __ save_frame(0); ysr@777: ysr@777: __ mov(pre_val, L0); ysr@777: __ mov(tmp, L1); ysr@777: __ mov(tmp2, L2); ysr@777: ysr@777: __ call_VM_leaf(L7_thread_cache, ysr@777: CAST_FROM_FN_PTR(address, ysr@777: SATBMarkQueueSet::handle_zero_index_for_thread), ysr@777: G2_thread); ysr@777: ysr@777: __ mov(L0, pre_val); ysr@777: __ mov(L1, tmp); ysr@777: __ mov(L2, tmp2); ysr@777: ysr@777: __ br(Assembler::always, /*annul*/false, Assembler::pt, restart); ysr@777: __ delayed()->restore(); ysr@777: } ysr@777: break; ysr@777: ysr@777: case g1_post_barrier_slow_id: ysr@777: { ysr@777: BarrierSet* bs = Universe::heap()->barrier_set(); ysr@777: if (bs->kind() != BarrierSet::G1SATBCTLogging) { ysr@777: __ save_frame(0); ysr@777: __ set((int)id, O1); ysr@777: __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0); ysr@777: __ should_not_reach_here(); ysr@777: break; ysr@777: } ysr@777: ysr@777: __ set_info("g1_post_barrier_slow_id", dont_gc_arguments); ysr@777: ysr@777: Register addr = G4; ysr@777: Register cardtable = G5; ysr@777: Register tmp = G1_scratch; ysr@777: Register tmp2 = G3_scratch; ysr@777: jbyte* byte_map_base = ((CardTableModRefBS*)bs)->byte_map_base; ysr@777: ysr@777: Label not_already_dirty, restart, refill; ysr@777: ysr@777: #ifdef _LP64 ysr@777: __ srlx(addr, CardTableModRefBS::card_shift, addr); ysr@777: #else ysr@777: __ srl(addr, CardTableModRefBS::card_shift, addr); ysr@777: #endif ysr@777: ysr@777: Address rs(cardtable, (address)byte_map_base); ysr@777: __ load_address(rs); // cardtable := ysr@777: __ ldub(addr, cardtable, tmp); // tmp := [addr + cardtable] ysr@777: ysr@777: __ br_on_reg_cond(Assembler::rc_nz, /*annul*/false, Assembler::pt, ysr@777: tmp, not_already_dirty); ysr@777: // Get cardtable + tmp into a reg by itself -- useful in the take-the-branch ysr@777: // case, harmless if not. ysr@777: __ delayed()->add(addr, cardtable, tmp2); ysr@777: ysr@777: // We didn't take the branch, so we're already dirty: return. ysr@777: // Use return-from-leaf ysr@777: __ retl(); ysr@777: __ delayed()->nop(); ysr@777: ysr@777: // Not dirty. ysr@777: __ bind(not_already_dirty); ysr@777: // First, dirty it. ysr@777: __ stb(G0, tmp2, 0); // [cardPtr] := 0 (i.e., dirty). ysr@777: ysr@777: Register tmp3 = cardtable; ysr@777: Register tmp4 = tmp; ysr@777: ysr@777: // these registers are now dead ysr@777: addr = cardtable = tmp = noreg; ysr@777: ysr@777: int dirty_card_q_index_byte_offset = ysr@777: in_bytes(JavaThread::dirty_card_queue_offset() + ysr@777: PtrQueue::byte_offset_of_index()); ysr@777: int dirty_card_q_buf_byte_offset = ysr@777: in_bytes(JavaThread::dirty_card_queue_offset() + ysr@777: PtrQueue::byte_offset_of_buf()); ysr@777: __ bind(restart); ysr@777: __ ld_ptr(G2_thread, dirty_card_q_index_byte_offset, tmp3); ysr@777: ysr@777: __ br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pn, ysr@777: tmp3, refill); ysr@777: // If the branch is taken, no harm in executing this in the delay slot. ysr@777: __ delayed()->ld_ptr(G2_thread, dirty_card_q_buf_byte_offset, tmp4); ysr@777: __ sub(tmp3, oopSize, tmp3); ysr@777: ysr@777: __ st_ptr(tmp2, tmp4, tmp3); // [_buf + index] := ysr@777: // Use return-from-leaf ysr@777: __ retl(); ysr@777: __ delayed()->st_ptr(tmp3, G2_thread, dirty_card_q_index_byte_offset); ysr@777: ysr@777: __ bind(refill); ysr@777: __ save_frame(0); ysr@777: ysr@777: __ mov(tmp2, L0); ysr@777: __ mov(tmp3, L1); ysr@777: __ mov(tmp4, L2); ysr@777: ysr@777: __ call_VM_leaf(L7_thread_cache, ysr@777: CAST_FROM_FN_PTR(address, ysr@777: DirtyCardQueueSet::handle_zero_index_for_thread), ysr@777: G2_thread); ysr@777: ysr@777: __ mov(L0, tmp2); ysr@777: __ mov(L1, tmp3); ysr@777: __ mov(L2, tmp4); ysr@777: ysr@777: __ br(Assembler::always, /*annul*/false, Assembler::pt, restart); ysr@777: __ delayed()->restore(); ysr@777: } ysr@777: break; ysr@777: #endif // !SERIALGC ysr@777: duke@435: default: duke@435: { __ set_info("unimplemented entry", dont_gc_arguments); duke@435: __ save_frame(0); duke@435: __ set((int)id, O1); duke@435: __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), O1); duke@435: __ should_not_reach_here(); duke@435: } duke@435: break; duke@435: } duke@435: return oop_maps; duke@435: } duke@435: duke@435: duke@435: void Runtime1::generate_handle_exception(StubAssembler* sasm, OopMapSet* oop_maps, OopMap* oop_map, bool) { duke@435: Label no_deopt; duke@435: Label no_handler; duke@435: duke@435: __ verify_not_null_oop(Oexception); duke@435: duke@435: // save the exception and issuing pc in the thread duke@435: __ st_ptr(Oexception, G2_thread, in_bytes(JavaThread::exception_oop_offset())); duke@435: __ st_ptr(Oissuing_pc, G2_thread, in_bytes(JavaThread::exception_pc_offset())); duke@435: duke@435: // save the real return address and use the throwing pc as the return address to lookup (has bci & oop map) duke@435: __ mov(I7, L0); duke@435: __ mov(Oissuing_pc, I7); duke@435: __ sub(I7, frame::pc_return_offset, I7); duke@435: int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc)); duke@435: duke@435: // Note: if nmethod has been deoptimized then regardless of duke@435: // whether it had a handler or not we will deoptimize duke@435: // by entering the deopt blob with a pending exception. duke@435: duke@435: __ tst(O0); duke@435: __ br(Assembler::zero, false, Assembler::pn, no_handler); duke@435: __ delayed()->nop(); duke@435: duke@435: // restore the registers that were saved at the beginning and jump to the exception handler. duke@435: restore_live_registers(sasm); duke@435: duke@435: __ jmp(O0, 0); duke@435: __ delayed()->restore(); duke@435: duke@435: __ bind(no_handler); duke@435: __ mov(L0, I7); // restore return address duke@435: duke@435: // restore exception oop duke@435: __ ld_ptr(G2_thread, in_bytes(JavaThread::exception_oop_offset()), Oexception->after_save()); duke@435: __ st_ptr(G0, G2_thread, in_bytes(JavaThread::exception_oop_offset())); duke@435: duke@435: __ restore(); duke@435: duke@435: Address exc(G4, Runtime1::entry_for(Runtime1::unwind_exception_id)); duke@435: __ jump_to(exc, 0); duke@435: __ delayed()->nop(); duke@435: duke@435: duke@435: oop_maps->add_gc_map(call_offset, oop_map); duke@435: } duke@435: duke@435: duke@435: #undef __ duke@435: duke@435: #define __ masm->