/* * Copyright (c) 2003, 2026, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved. * Copyright (c) 2020, 2023, Huawei Technologies Co., Ltd. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License version 2 only, as * published by the Free Software Foundation. * * This code is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * version 2 for more details (a copy is included in the LICENSE file that * accompanied this code). * * You should have received a copy of the GNU General Public License version * 2 along with this work; if not, write to the Free Software Foundation, * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. * * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA * or visit www.oracle.com if you need additional information or have any * questions. * */ #include "asm/macroAssembler.inline.hpp" #include "gc/shared/barrierSet.hpp" #include "gc/shared/barrierSetAssembler.hpp" #include "interp_masm_riscv.hpp" #include "interpreter/interpreter.hpp" #include "interpreter/interpreterRuntime.hpp" #include "logging/log.hpp" #include "oops/arrayOop.hpp" #include "oops/markWord.hpp" #include "oops/method.hpp" #include "oops/methodData.hpp" #include "oops/resolvedFieldEntry.hpp" #include "oops/resolvedIndyEntry.hpp" #include "oops/resolvedMethodEntry.hpp" #include "prims/jvmtiExport.hpp" #include "prims/jvmtiThreadState.hpp" #include "runtime/basicLock.hpp" #include "runtime/frame.inline.hpp" #include "runtime/javaThread.hpp" #include "runtime/safepointMechanism.hpp" #include "runtime/sharedRuntime.hpp" #include "utilities/powerOfTwo.hpp" void InterpreterMacroAssembler::narrow(Register result) { // Get method->_constMethod->_result_type ld(t0, Address(fp, frame::interpreter_frame_method_offset * wordSize)); ld(t0, Address(t0, Method::const_offset())); lbu(t0, Address(t0, ConstMethod::result_type_offset())); Label done, notBool, notByte, notChar; // common case first mv(t1, T_INT); beq(t0, t1, done); // mask integer result to narrower return type. mv(t1, T_BOOLEAN); bne(t0, t1, notBool); andi(result, result, 0x1); j(done); bind(notBool); mv(t1, T_BYTE); bne(t0, t1, notByte); sext(result, result, 8); j(done); bind(notByte); mv(t1, T_CHAR); bne(t0, t1, notChar); zext(result, result, 16); j(done); bind(notChar); sext(result, result, 16); bind(done); sext(result, result, 32); } void InterpreterMacroAssembler::jump_to_entry(address entry) { assert(entry != nullptr, "Entry must have been generated by now"); j(entry); } void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) { if (JvmtiExport::can_pop_frame()) { Label L; // Initiate popframe handling only if it is not already being // processed. If the flag has the popframe_processing bit set, // it means that this code is called *during* popframe handling - we // don't want to reenter. // This method is only called just after the call into the vm in // call_VM_base, so the arg registers are available. lwu(t1, Address(xthread, JavaThread::popframe_condition_offset())); test_bit(t0, t1, exact_log2(JavaThread::popframe_pending_bit)); beqz(t0, L); test_bit(t0, t1, exact_log2(JavaThread::popframe_processing_bit)); bnez(t0, L); // Call Interpreter::remove_activation_preserving_args_entry() to get the // address of the same-named entrypoint in the generated interpreter code. call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); jr(x10); bind(L); } } void InterpreterMacroAssembler::load_earlyret_value(TosState state) { ld(x12, Address(xthread, JavaThread::jvmti_thread_state_offset())); const Address tos_addr(x12, JvmtiThreadState::earlyret_tos_offset()); const Address oop_addr(x12, JvmtiThreadState::earlyret_oop_offset()); const Address val_addr(x12, JvmtiThreadState::earlyret_value_offset()); switch (state) { case atos: ld(x10, oop_addr); sd(zr, oop_addr); verify_oop(x10); break; case ltos: ld(x10, val_addr); break; case btos: // fall through case ztos: // fall through case ctos: // fall through case stos: // fall through case itos: lwu(x10, val_addr); break; case ftos: flw(f10, val_addr); break; case dtos: fld(f10, val_addr); break; case vtos: /* nothing to do */ break; default: ShouldNotReachHere(); } // Clean up tos value in the thread object mv(t0, (int)ilgl); sw(t0, tos_addr); sw(zr, val_addr); } void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) { if (JvmtiExport::can_force_early_return()) { Label L; ld(t0, Address(xthread, JavaThread::jvmti_thread_state_offset())); beqz(t0, L); // if thread->jvmti_thread_state() is null then exit // Initiate earlyret handling only if it is not already being processed. // If the flag has the earlyret_processing bit set, it means that this code // is called *during* earlyret handling - we don't want to reenter. lwu(t0, Address(t0, JvmtiThreadState::earlyret_state_offset())); mv(t1, JvmtiThreadState::earlyret_pending); bne(t0, t1, L); // Call Interpreter::remove_activation_early_entry() to get the address of the // same-named entrypoint in the generated interpreter code. ld(t0, Address(xthread, JavaThread::jvmti_thread_state_offset())); lwu(t0, Address(t0, JvmtiThreadState::earlyret_tos_offset())); call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), t0); jr(x10); bind(L); } } void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) { assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode"); lbu(t1, Address(xbcp, bcp_offset)); lbu(reg, Address(xbcp, bcp_offset + 1)); slli(t1, t1, 8); add(reg, reg, t1); } void InterpreterMacroAssembler::get_dispatch() { la(xdispatch, ExternalAddress((address)Interpreter::dispatch_table())); } void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, Register tmp, int bcp_offset, size_t index_size) { assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); if (index_size == sizeof(u2)) { load_short_misaligned(index, Address(xbcp, bcp_offset), tmp, false); } else if (index_size == sizeof(u4)) { load_int_misaligned(index, Address(xbcp, bcp_offset), tmp, false); } else if (index_size == sizeof(u1)) { load_unsigned_byte(index, Address(xbcp, bcp_offset)); } else { ShouldNotReachHere(); } } // Load object from cpool->resolved_references(index) void InterpreterMacroAssembler::load_resolved_reference_at_index( Register result, Register index, Register tmp) { assert_different_registers(result, index); get_constant_pool(result); // Load pointer for resolved_references[] objArray ld(result, Address(result, ConstantPool::cache_offset())); ld(result, Address(result, ConstantPoolCache::resolved_references_offset())); resolve_oop_handle(result, tmp, t1); // Add in the index addi(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop); shadd(result, index, result, index, LogBytesPerHeapOop); load_heap_oop(result, Address(result, 0), tmp, t1); } void InterpreterMacroAssembler::load_resolved_klass_at_offset( Register cpool, Register index, Register klass, Register temp) { shadd(temp, index, cpool, temp, LogBytesPerWord); lhu(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index ld(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses shadd(klass, temp, klass, temp, LogBytesPerWord); ld(klass, Address(klass, Array::base_offset_in_bytes())); } // Generate a subtype check: branch to ok_is_subtype if sub_klass is a // subtype of super_klass. // // Args: // x10: superklass // Rsub_klass: subklass // // Kills: // x12 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Label& ok_is_subtype) { assert(Rsub_klass != x10, "x10 holds superklass"); assert(Rsub_klass != x12, "x12 holds 2ndary super array length"); // Profile the not-null value's klass. profile_typecheck(x12, Rsub_klass); // blows x12 // Do the check. check_klass_subtype(Rsub_klass, x10, x12, ok_is_subtype); // blows x12 } // Java Expression Stack void InterpreterMacroAssembler::pop_ptr(Register r) { ld(r, Address(esp, 0)); addi(esp, esp, wordSize); } void InterpreterMacroAssembler::pop_i(Register r) { lw(r, Address(esp, 0)); // lw do signed extended addi(esp, esp, wordSize); } void InterpreterMacroAssembler::pop_l(Register r) { ld(r, Address(esp, 0)); addi(esp, esp, 2 * Interpreter::stackElementSize); } void InterpreterMacroAssembler::push_ptr(Register r) { subi(esp, esp, wordSize); sd(r, Address(esp, 0)); } void InterpreterMacroAssembler::push_i(Register r) { subi(esp, esp, wordSize); sext(r, r, 32); sd(r, Address(esp, 0)); } void InterpreterMacroAssembler::push_l(Register r) { subi(esp, esp, 2 * wordSize); sd(zr, Address(esp, wordSize)); sd(r, Address(esp)); } void InterpreterMacroAssembler::pop_f(FloatRegister r) { flw(r, Address(esp, 0)); addi(esp, esp, wordSize); } void InterpreterMacroAssembler::pop_d(FloatRegister r) { fld(r, Address(esp, 0)); addi(esp, esp, 2 * Interpreter::stackElementSize); } void InterpreterMacroAssembler::push_f(FloatRegister r) { subi(esp, esp, wordSize); fsw(r, Address(esp, 0)); } void InterpreterMacroAssembler::push_d(FloatRegister r) { subi(esp, esp, 2 * wordSize); fsd(r, Address(esp, 0)); } void InterpreterMacroAssembler::pop(TosState state) { switch (state) { case atos: pop_ptr(); verify_oop(x10); break; case btos: // fall through case ztos: // fall through case ctos: // fall through case stos: // fall through case itos: pop_i(); break; case ltos: pop_l(); break; case ftos: pop_f(); break; case dtos: pop_d(); break; case vtos: /* nothing to do */ break; default: ShouldNotReachHere(); } } void InterpreterMacroAssembler::push(TosState state) { switch (state) { case atos: verify_oop(x10); push_ptr(); break; case btos: // fall through case ztos: // fall through case ctos: // fall through case stos: // fall through case itos: push_i(); break; case ltos: push_l(); break; case ftos: push_f(); break; case dtos: push_d(); break; case vtos: /* nothing to do */ break; default: ShouldNotReachHere(); } } // Helpers for swap and dup void InterpreterMacroAssembler::load_ptr(int n, Register val) { ld(val, Address(esp, Interpreter::expr_offset_in_bytes(n))); } void InterpreterMacroAssembler::store_ptr(int n, Register val) { sd(val, Address(esp, Interpreter::expr_offset_in_bytes(n))); } void InterpreterMacroAssembler::load_float(Address src) { flw(f10, src); } void InterpreterMacroAssembler::load_double(Address src) { fld(f10, src); } void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() { // set sender sp mv(x19_sender_sp, sp); // record last_sp sub(t0, esp, fp); srai(t0, t0, Interpreter::logStackElementSize); sd(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize)); } // Jump to from_interpreted entry of a call unless single stepping is possible // in this thread in which case we must call the i2i entry void InterpreterMacroAssembler::jump_from_interpreted(Register method) { prepare_to_jump_from_interpreted(); if (JvmtiExport::can_post_interpreter_events()) { Label run_compiled_code; // JVMTI events, such as single-stepping, are implemented partly by avoiding running // compiled code in threads for which the event is enabled. Check here for // interp_only_mode if these events CAN be enabled. lwu(t0, Address(xthread, JavaThread::interp_only_mode_offset())); beqz(t0, run_compiled_code); ld(t1, Address(method, Method::interpreter_entry_offset())); jr(t1); bind(run_compiled_code); } ld(t1, Address(method, Method::from_interpreted_offset())); jr(t1); } // The following two routines provide a hook so that an implementation // can schedule the dispatch in two parts. amd64 does not do this. void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) { } void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) { dispatch_next(state, step); } void InterpreterMacroAssembler::dispatch_base(TosState state, address* table, bool verifyoop, bool generate_poll, Register Rs) { // Pay attention to the argument Rs, which is acquiesce in t0. if (VerifyActivationFrameSize) { Label L; sub(t1, fp, esp); int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset + frame::metadata_words) * wordSize; sub(t1, t1, min_frame_size); bgez(t1, L); stop("broken stack frame"); bind(L); } if (verifyoop && state == atos) { verify_oop(x10); } Label safepoint; address* const safepoint_table = Interpreter::safept_table(state); bool needs_thread_local_poll = generate_poll && table != safepoint_table; if (needs_thread_local_poll) { NOT_PRODUCT(block_comment("Thread-local Safepoint poll")); ld(t1, Address(xthread, JavaThread::polling_word_offset())); test_bit(t1, t1, exact_log2(SafepointMechanism::poll_bit())); bnez(t1, safepoint); } if (table == Interpreter::dispatch_table(state)) { mv(t1, Interpreter::distance_from_dispatch_table(state)); add(t1, Rs, t1); shadd(t1, t1, xdispatch, t1, 3); } else { mv(t1, (address)table); shadd(t1, Rs, t1, Rs, 3); } ld(t1, Address(t1)); jr(t1); if (needs_thread_local_poll) { bind(safepoint); la(t1, ExternalAddress((address)safepoint_table)); shadd(t1, Rs, t1, Rs, 3); ld(t1, Address(t1)); jr(t1); } } void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll, Register Rs) { dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll, Rs); } void InterpreterMacroAssembler::dispatch_only_normal(TosState state, Register Rs) { dispatch_base(state, Interpreter::normal_table(state), true, false, Rs); } void InterpreterMacroAssembler::dispatch_only_noverify(TosState state, Register Rs) { dispatch_base(state, Interpreter::normal_table(state), false, false, Rs); } void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) { // load next bytecode load_unsigned_byte(t0, Address(xbcp, step)); add(xbcp, xbcp, step); dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); } void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { // load current bytecode lbu(t0, Address(xbcp, 0)); dispatch_base(state, table); } // remove activation // // Unlock the receiver if this is a synchronized method. // Unlock any Java monitors from synchronized blocks. // Apply stack watermark barrier. // Notify JVMTI. // Remove the activation from the stack. // // If there are locked Java monitors // If throw_monitor_exception // throws IllegalMonitorStateException // Else if install_monitor_exception // installs IllegalMonitorStateException // Else // no error processing void InterpreterMacroAssembler::remove_activation(TosState state, bool throw_monitor_exception, bool install_monitor_exception, bool notify_jvmdi) { // Note: Registers x13 may be in use for the // result check if synchronized method Label unlocked, unlock, no_unlock; #ifdef ASSERT Label not_preempted; ld(t0, Address(xthread, JavaThread::preempt_alternate_return_offset())); beqz(t0, not_preempted); stop("remove_activation: should not have alternate return address set"); bind(not_preempted); #endif /* ASSERT */ // get the value of _do_not_unlock_if_synchronized into x13 const Address do_not_unlock_if_synchronized(xthread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); lbu(x13, do_not_unlock_if_synchronized); sb(zr, do_not_unlock_if_synchronized); // reset the flag // get method access flags ld(x11, Address(fp, frame::interpreter_frame_method_offset * wordSize)); load_unsigned_short(x12, Address(x11, Method::access_flags_offset())); test_bit(t0, x12, exact_log2(JVM_ACC_SYNCHRONIZED)); beqz(t0, unlocked); // Don't unlock anything if the _do_not_unlock_if_synchronized flag // is set. bnez(x13, no_unlock); // unlock monitor push(state); // save result // BasicObjectLock will be first in list, since this is a // synchronized method. However, need to check that the object has // not been unlocked by an explicit monitorexit bytecode. const Address monitor(fp, frame::interpreter_frame_initial_sp_offset * wordSize - (int) sizeof(BasicObjectLock)); // We use c_rarg1 so that if we go slow path it will be the correct // register for unlock_object to pass to VM directly la(c_rarg1, monitor); // address of first monitor ld(x10, Address(c_rarg1, BasicObjectLock::obj_offset())); bnez(x10, unlock); pop(state); if (throw_monitor_exception) { // Entry already unlocked, need to throw exception call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); should_not_reach_here(); } else { // Monitor already unlocked during a stack unroll. If requested, // install an illegal_monitor_state_exception. Continue with // stack unrolling. if (install_monitor_exception) { call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); } j(unlocked); } bind(unlock); unlock_object(c_rarg1); pop(state); // Check that for block-structured locking (i.e., that all locked // objects has been unlocked) bind(unlocked); // x10: Might contain return value // Check that all monitors are unlocked { Label loop, exception, entry, restart; const int entry_size = frame::interpreter_frame_monitor_size_in_bytes(); const Address monitor_block_top( fp, frame::interpreter_frame_monitor_block_top_offset * wordSize); const Address monitor_block_bot( fp, frame::interpreter_frame_initial_sp_offset * wordSize); bind(restart); // We use c_rarg1 so that if we go slow path it will be the correct // register for unlock_object to pass to VM directly ld(c_rarg1, monitor_block_top); // derelativize pointer shadd(c_rarg1, c_rarg1, fp, c_rarg1, LogBytesPerWord); // c_rarg1 points to current entry, starting with top-most entry la(x9, monitor_block_bot); // points to word before bottom of // monitor block j(entry); // Entry already locked, need to throw exception bind(exception); if (throw_monitor_exception) { // Throw exception MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime:: throw_illegal_monitor_state_exception)); should_not_reach_here(); } else { // Stack unrolling. Unlock object and install illegal_monitor_exception. // Unlock does not block, so don't have to worry about the frame. // We don't have to preserve c_rarg1 since we are going to throw an exception. push(state); unlock_object(c_rarg1); pop(state); if (install_monitor_exception) { call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime:: new_illegal_monitor_state_exception)); } j(restart); } bind(loop); // check if current entry is used add(t0, c_rarg1, in_bytes(BasicObjectLock::obj_offset())); ld(t0, Address(t0, 0)); bnez(t0, exception); add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry bind(entry); bne(c_rarg1, x9, loop); // check if bottom reached if not at bottom then check this entry } bind(no_unlock); JFR_ONLY(enter_jfr_critical_section();) // The below poll is for the stack watermark barrier. It allows fixing up frames lazily, // that would normally not be safe to use. Such bad returns into unsafe territory of // the stack, will call InterpreterRuntime::at_unwind. Label slow_path; Label fast_path; safepoint_poll(slow_path, true /* at_return */, false /* in_nmethod */); j(fast_path); bind(slow_path); push(state); set_last_Java_frame(esp, fp, pc(), t0); super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), xthread); reset_last_Java_frame(true); pop(state); bind(fast_path); // JVMTI support. Make sure the safepoint poll test is issued prior. if (notify_jvmdi) { notify_method_exit(state, NotifyJVMTI); // preserve TOSCA } else { notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA } // remove activation // get sender esp ld(t1, Address(fp, frame::interpreter_frame_sender_sp_offset * wordSize)); if (StackReservedPages > 0) { // testing if reserved zone needs to be re-enabled Label no_reserved_zone_enabling; // check if already enabled - if so no re-enabling needed assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size"); lw(t0, Address(xthread, JavaThread::stack_guard_state_offset())); subw(t0, t0, StackOverflow::stack_guard_enabled); beqz(t0, no_reserved_zone_enabling); // look for an overflow into the stack reserved zone, i.e. // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation ld(t0, Address(xthread, JavaThread::reserved_stack_activation_offset())); ble(t1, t0, no_reserved_zone_enabling); JFR_ONLY(leave_jfr_critical_section();) call_VM_leaf( CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), xthread); call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError)); should_not_reach_here(); bind(no_reserved_zone_enabling); } // remove frame anchor leave(); JFR_ONLY(leave_jfr_critical_section();) // restore sender esp mv(esp, t1); // If we're returning to interpreted code we will shortly be // adjusting SP to allow some space for ESP. If we're returning to // compiled code the saved sender SP was saved in sender_sp, so this // restores it. andi(sp, esp, -16); } #if INCLUDE_JFR void InterpreterMacroAssembler::enter_jfr_critical_section() { const Address sampling_critical_section(xthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR)); mv(t0, true); sb(t0, sampling_critical_section); } void InterpreterMacroAssembler::leave_jfr_critical_section() { const Address sampling_critical_section(xthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR)); sb(zr, sampling_critical_section); } #endif // INCLUDE_JFR // Lock object // // Args: // c_rarg1: BasicObjectLock to be used for locking // // Kills: // x10 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, c_rarg5, .. (param regs) // t0, t1 (temp regs) void InterpreterMacroAssembler::lock_object(Register lock_reg) { assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1"); const Register tmp = c_rarg2; const Register obj_reg = c_rarg3; // Will contain the oop const Register tmp2 = c_rarg4; const Register tmp3 = c_rarg5; // Load object pointer into obj_reg (c_rarg3) ld(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); Label done, slow_case; fast_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case); j(done); bind(slow_case); // Call the runtime routine for slow case call_VM_preemptable(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); bind(done); } // Unlocks an object. Used in monitorexit bytecode and // remove_activation. Throws an IllegalMonitorException if object is // not locked by current thread. // // Args: // c_rarg1: BasicObjectLock for lock // // Kills: // x10 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, ... (param regs) // t0, t1 (temp regs) void InterpreterMacroAssembler::unlock_object(Register lock_reg) { assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1"); const Register swap_reg = x10; const Register header_reg = c_rarg2; // Will contain the old oopMark const Register obj_reg = c_rarg3; // Will contain the oop const Register tmp_reg = c_rarg4; // Temporary used by fast_unlock save_bcp(); // Save in case of exception // Load oop into obj_reg (c_rarg3) ld(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // Free entry sd(zr, Address(lock_reg, BasicObjectLock::obj_offset())); Label done, slow_case; fast_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case); j(done); bind(slow_case); // Call the runtime routine for slow case. sd(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); bind(done); restore_bcp(); } void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) { assert(ProfileInterpreter, "must be profiling interpreter"); ld(mdp, Address(fp, frame::interpreter_frame_mdp_offset * wordSize)); beqz(mdp, zero_continue); } // Set the method data pointer for the current bcp. void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { assert(ProfileInterpreter, "must be profiling interpreter"); Label set_mdp; push_reg(RegSet::of(x10, x11), sp); // save x10, x11 // Test MDO to avoid the call if it is null. ld(x10, Address(xmethod, in_bytes(Method::method_data_offset()))); beqz(x10, set_mdp); call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), xmethod, xbcp); // x10: mdi // mdo is guaranteed to be non-zero here, we checked for it before the call. ld(x11, Address(xmethod, in_bytes(Method::method_data_offset()))); la(x11, Address(x11, in_bytes(MethodData::data_offset()))); add(x10, x11, x10); sd(x10, Address(fp, frame::interpreter_frame_mdp_offset * wordSize)); bind(set_mdp); pop_reg(RegSet::of(x10, x11), sp); } void InterpreterMacroAssembler::verify_method_data_pointer() { assert(ProfileInterpreter, "must be profiling interpreter"); #ifdef ASSERT Label verify_continue; subi(sp, sp, 4 * wordSize); sd(x10, Address(sp, 0)); sd(x11, Address(sp, wordSize)); sd(x12, Address(sp, 2 * wordSize)); sd(x13, Address(sp, 3 * wordSize)); test_method_data_pointer(x13, verify_continue); // If mdp is zero, continue get_method(x11); // If the mdp is valid, it will point to a DataLayout header which is // consistent with the bcp. The converse is highly probable also. lh(x12, Address(x13, in_bytes(DataLayout::bci_offset()))); ld(t0, Address(x11, Method::const_offset())); add(x12, x12, t0); la(x12, Address(x12, ConstMethod::codes_offset())); beq(x12, xbcp, verify_continue); // x10: method // xbcp: bcp // xbcp == 22 // x13: mdp call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), x11, xbcp, x13); bind(verify_continue); ld(x10, Address(sp, 0)); ld(x11, Address(sp, wordSize)); ld(x12, Address(sp, 2 * wordSize)); ld(x13, Address(sp, 3 * wordSize)); addi(sp, sp, 4 * wordSize); #endif // ASSERT } void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) { assert(ProfileInterpreter, "must be profiling interpreter"); Address data(mdp_in, constant); sd(value, data); } void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, int constant) { increment_mdp_data_at(mdp_in, noreg, constant); } void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, Register index, int constant) { assert(ProfileInterpreter, "must be profiling interpreter"); assert_different_registers(t1, t0, mdp_in, index); Address addr1(mdp_in, constant); Address addr2(t1, 0); Address &addr = addr1; if (index != noreg) { la(t1, addr1); add(t1, t1, index); addr = addr2; } ld(t0, addr); addi(t0, t0, DataLayout::counter_increment); sd(t0, addr); } void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, int flag_byte_constant) { assert(ProfileInterpreter, "must be profiling interpreter"); int flags_offset = in_bytes(DataLayout::flags_offset()); // Set the flag lbu(t1, Address(mdp_in, flags_offset)); ori(t1, t1, flag_byte_constant); sb(t1, Address(mdp_in, flags_offset)); } void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, int offset, Register value, Register test_value_out, Label& not_equal_continue) { assert(ProfileInterpreter, "must be profiling interpreter"); if (test_value_out == noreg) { ld(t1, Address(mdp_in, offset)); bne(value, t1, not_equal_continue); } else { // Put the test value into a register, so caller can use it: ld(test_value_out, Address(mdp_in, offset)); bne(value, test_value_out, not_equal_continue); } } void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) { assert(ProfileInterpreter, "must be profiling interpreter"); ld(t1, Address(mdp_in, offset_of_disp)); add(mdp_in, mdp_in, t1); sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize)); } void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, Register reg, int offset_of_disp) { assert(ProfileInterpreter, "must be profiling interpreter"); add(t1, mdp_in, reg); ld(t1, Address(t1, offset_of_disp)); add(mdp_in, mdp_in, t1); sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize)); } void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) { assert(ProfileInterpreter, "must be profiling interpreter"); add(mdp_in, mdp_in, (unsigned)constant); sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize)); } void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { assert(ProfileInterpreter, "must be profiling interpreter"); // save/restore across call_VM subi(sp, sp, 2 * wordSize); sd(zr, Address(sp, 0)); sd(return_bci, Address(sp, wordSize)); call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); ld(zr, Address(sp, 0)); ld(return_bci, Address(sp, wordSize)); addi(sp, sp, 2 * wordSize); } void InterpreterMacroAssembler::profile_taken_branch(Register mdp) { if (ProfileInterpreter) { Label profile_continue; // If no method data exists, go to profile_continue. test_method_data_pointer(mdp, profile_continue); // We are taking a branch. Increment the taken count. increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); // The method data pointer needs to be updated to reflect the new target. update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); bind(profile_continue); } } void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { if (ProfileInterpreter) { Label profile_continue; // If no method data exists, go to profile_continue. test_method_data_pointer(mdp, profile_continue); // We are not taking a branch. Increment the not taken count. increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset())); // The method data pointer needs to be updated to correspond to // the next bytecode update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); bind(profile_continue); } } void InterpreterMacroAssembler::profile_call(Register mdp) { if (ProfileInterpreter) { Label profile_continue; // If no method data exists, go to profile_continue. test_method_data_pointer(mdp, profile_continue); // We are making a call. Increment the count. increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); // The method data pointer needs to be updated to reflect the new target. update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); bind(profile_continue); } } void InterpreterMacroAssembler::profile_final_call(Register mdp) { if (ProfileInterpreter) { Label profile_continue; // If no method data exists, go to profile_continue. test_method_data_pointer(mdp, profile_continue); // We are making a call. Increment the count. increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); // The method data pointer needs to be updated to reflect the new target. update_mdp_by_constant(mdp, in_bytes(VirtualCallData:: virtual_call_data_size())); bind(profile_continue); } } void InterpreterMacroAssembler::profile_virtual_call(Register receiver, Register mdp) { if (ProfileInterpreter) { Label profile_continue; // If no method data exists, go to profile_continue. test_method_data_pointer(mdp, profile_continue); // Record the receiver type. profile_receiver_type(receiver, mdp, 0); // The method data pointer needs to be updated to reflect the new target. update_mdp_by_constant(mdp, in_bytes(VirtualCallData:: virtual_call_data_size())); bind(profile_continue); } } void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) { if (ProfileInterpreter) { Label profile_continue; // If no method data exists, go to profile_continue. test_method_data_pointer(mdp, profile_continue); // Update the total ret count. increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); for (uint row = 0; row < RetData::row_limit(); row++) { Label next_test; // See if return_bci is equal to bci[n]: test_mdp_data_at(mdp, in_bytes(RetData::bci_offset(row)), return_bci, noreg, next_test); // return_bci is equal to bci[n]. Increment the count. increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); // The method data pointer needs to be updated to reflect the new target. update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row))); j(profile_continue); bind(next_test); } update_mdp_for_ret(return_bci); bind(profile_continue); } } void InterpreterMacroAssembler::profile_null_seen(Register mdp) { if (ProfileInterpreter) { Label profile_continue; // If no method data exists, go to profile_continue. test_method_data_pointer(mdp, profile_continue); set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); // The method data pointer needs to be updated. int mdp_delta = in_bytes(BitData::bit_data_size()); if (TypeProfileCasts) { mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); } update_mdp_by_constant(mdp, mdp_delta); bind(profile_continue); } } void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass) { if (ProfileInterpreter) { Label profile_continue; // If no method data exists, go to profile_continue. test_method_data_pointer(mdp, profile_continue); // The method data pointer needs to be updated. int mdp_delta = in_bytes(BitData::bit_data_size()); if (TypeProfileCasts) { mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); // Record the object type. profile_receiver_type(klass, mdp, 0); } update_mdp_by_constant(mdp, mdp_delta); bind(profile_continue); } } void InterpreterMacroAssembler::profile_switch_default(Register mdp) { if (ProfileInterpreter) { Label profile_continue; // If no method data exists, go to profile_continue. test_method_data_pointer(mdp, profile_continue); // Update the default case count increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset())); // The method data pointer needs to be updated. update_mdp_by_offset(mdp, in_bytes(MultiBranchData:: default_displacement_offset())); bind(profile_continue); } } void InterpreterMacroAssembler::profile_switch_case(Register index, Register mdp, Register reg2) { if (ProfileInterpreter) { Label profile_continue; // If no method data exists, go to profile_continue. test_method_data_pointer(mdp, profile_continue); // Build the base (index * per_case_size_in_bytes()) + // case_array_offset_in_bytes() mv(reg2, in_bytes(MultiBranchData::per_case_size())); mv(t0, in_bytes(MultiBranchData::case_array_offset())); Assembler::mul(index, index, reg2); Assembler::add(index, index, t0); // Update the case count increment_mdp_data_at(mdp, index, in_bytes(MultiBranchData::relative_count_offset())); // The method data pointer need to be updated. update_mdp_by_offset(mdp, index, in_bytes(MultiBranchData:: relative_displacement_offset())); bind(profile_continue); } } void InterpreterMacroAssembler::notify_method_entry() { // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to // track stack depth. If it is possible to enter interp_only_mode we add // the code to check if the event should be sent. if (JvmtiExport::can_post_interpreter_events()) { Label L; lwu(x13, Address(xthread, JavaThread::interp_only_mode_offset())); beqz(x13, L); call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); bind(L); } if (DTraceMethodProbes) { get_method(c_rarg1); call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), xthread, c_rarg1); } // RedefineClasses() tracing support for obsolete method entry if (log_is_enabled(Trace, redefine, class, obsolete)) { get_method(c_rarg1); call_VM_leaf( CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), xthread, c_rarg1); } } void InterpreterMacroAssembler::notify_method_exit( TosState state, NotifyMethodExitMode mode) { // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to // track stack depth. If it is possible to enter interp_only_mode we add // the code to check if the event should be sent. if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { Label L; // Note: frame::interpreter_frame_result has a dependency on how the // method result is saved across the call to post_method_exit. If this // is changed then the interpreter_frame_result implementation will // need to be updated too. // template interpreter will leave the result on the top of the stack. push(state); lwu(x13, Address(xthread, JavaThread::interp_only_mode_offset())); beqz(x13, L); call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); bind(L); pop(state); } if (DTraceMethodProbes) { push(state); get_method(c_rarg1); call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), xthread, c_rarg1); pop(state); } } // Jump if ((*counter_addr += increment) & mask) satisfies the condition. void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, int increment, Address mask, Register tmp1, Register tmp2, bool preloaded, Label* where) { Label done; if (!preloaded) { lwu(tmp1, counter_addr); } add(tmp1, tmp1, increment); sw(tmp1, counter_addr); lwu(tmp2, mask); andr(tmp1, tmp1, tmp2); bnez(tmp1, done); j(*where); // offset is too large so we have to use j instead of beqz here bind(done); } void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, int number_of_arguments) { // interpreter specific // // Note: No need to save/restore xbcp & xlocals pointer since these // are callee saved registers and no blocking/ GC can happen // in leaf calls. #ifdef ASSERT { Label L; ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize)); beqz(t0, L); stop("InterpreterMacroAssembler::call_VM_leaf_base:" " last_sp isn't null"); bind(L); } #endif /* ASSERT */ // super call MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments); } void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register java_thread, Register last_java_sp, Label* return_pc, address entry_point, int number_of_arguments, bool check_exceptions) { // interpreter specific // // Note: Could avoid restoring locals ptr (callee saved) - however doesn't // really make a difference for these runtime calls, since they are // slow anyway. Btw., bcp must be saved/restored since it may change // due to GC. save_bcp(); #ifdef ASSERT { Label L; ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize)); beqz(t0, L); stop("InterpreterMacroAssembler::call_VM_base:" " last_sp isn't null"); bind(L); } #endif /* ASSERT */ // super call MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp, return_pc, entry_point, number_of_arguments, check_exceptions); // interpreter specific restore_bcp(); restore_locals(); } void InterpreterMacroAssembler::call_VM_preemptable_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) { assert(InterpreterRuntime::is_preemptable_call(entry_point), "VM call not preemptable, should use call_VM()"); Label resume_pc, not_preempted; #ifdef ASSERT { Label L1, L2; ld(t0, Address(xthread, JavaThread::preempt_alternate_return_offset())); beqz(t0, L1); stop("call_VM_preemptable_helper: Should not have alternate return address set"); bind(L1); // We check this counter in patch_return_pc_with_preempt_stub() during freeze. incrementw(Address(xthread, JavaThread::interp_at_preemptable_vmcall_cnt_offset())); lw(t0, Address(xthread, JavaThread::interp_at_preemptable_vmcall_cnt_offset())); bgtz(t0, L2); stop("call_VM_preemptable_helper: should be > 0"); bind(L2); } #endif /* ASSERT */ // Force freeze slow path. push_cont_fastpath(); // Make VM call. In case of preemption set last_pc to the one we want to resume to. // Note: call_VM_base will use resume_pc label to set last_Java_pc. call_VM_base(noreg, noreg, noreg, &resume_pc, entry_point, number_of_arguments, false /*check_exceptions*/); pop_cont_fastpath(); #ifdef ASSERT { Label L; decrementw(Address(xthread, JavaThread::interp_at_preemptable_vmcall_cnt_offset())); lw(t0, Address(xthread, JavaThread::interp_at_preemptable_vmcall_cnt_offset())); bgez(t0, L); stop("call_VM_preemptable_helper: should be >= 0"); bind(L); } #endif /* ASSERT */ // Check if preempted. ld(t1, Address(xthread, JavaThread::preempt_alternate_return_offset())); beqz(t1, not_preempted); sd(zr, Address(xthread, JavaThread::preempt_alternate_return_offset())); jr(t1); // In case of preemption, this is where we will resume once we finally acquire the monitor. bind(resume_pc); restore_after_resume(false /* is_native */); bind(not_preempted); if (check_exceptions) { // check for pending exceptions ld(t0, Address(xthread, in_bytes(Thread::pending_exception_offset()))); Label ok; beqz(t0, ok); la(t1, RuntimeAddress(StubRoutines::forward_exception_entry())); jr(t1); bind(ok); } // get oop result if there is one and reset the value in the thread if (oop_result->is_valid()) { get_vm_result_oop(oop_result, xthread); } } static void pass_arg1(MacroAssembler* masm, Register arg) { if (c_rarg1 != arg) { masm->mv(c_rarg1, arg); } } static void pass_arg2(MacroAssembler* masm, Register arg) { if (c_rarg2 != arg) { masm->mv(c_rarg2, arg); } } void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result, address entry_point, Register arg_1, bool check_exceptions) { pass_arg1(this, arg_1); call_VM_preemptable_helper(oop_result, entry_point, 1, check_exceptions); } void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) { LP64_ONLY(assert_different_registers(arg_1, c_rarg2)); pass_arg2(this, arg_2); pass_arg1(this, arg_1); call_VM_preemptable_helper(oop_result, entry_point, 2, check_exceptions); } void InterpreterMacroAssembler::restore_after_resume(bool is_native) { la(t1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter())); jalr(t1); if (is_native) { // On resume we need to set up stack as expected push(dtos); push(ltos); } } void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) { assert_different_registers(obj, tmp, t0, mdo_addr.base()); Label update, next, none; verify_oop(obj); bnez(obj, update); orptr(mdo_addr, TypeEntries::null_seen, t0, tmp); j(next); bind(update); load_klass(obj, obj); ld(tmp, mdo_addr); xorr(obj, obj, tmp); andi(t0, obj, TypeEntries::type_klass_mask); beqz(t0, next); // klass seen before, nothing to // do. The unknown bit may have been // set already but no need to check. test_bit(t0, obj, exact_log2(TypeEntries::type_unknown)); bnez(t0, next); // already unknown. Nothing to do anymore. beqz(tmp, none); mv(t0, (u1)TypeEntries::null_seen); beq(tmp, t0, none); // There is a chance that the checks above // fail if another thread has just set the // profiling to this obj's klass xorr(obj, obj, tmp); // get back original value before XOR ld(tmp, mdo_addr); xorr(obj, obj, tmp); andi(t0, obj, TypeEntries::type_klass_mask); beqz(t0, next); // different than before. Cannot keep accurate profile. orptr(mdo_addr, TypeEntries::type_unknown, t0, tmp); j(next); bind(none); // first time here. Set profile type. sd(obj, mdo_addr); #ifdef ASSERT andi(obj, obj, TypeEntries::type_mask); verify_klass_ptr(obj); #endif bind(next); } void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) { if (!ProfileInterpreter) { return; } if (MethodData::profile_arguments() || MethodData::profile_return()) { Label profile_continue; test_method_data_pointer(mdp, profile_continue); int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); lbu(t0, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start)); if (is_virtual) { mv(tmp, (u1)DataLayout::virtual_call_type_data_tag); bne(t0, tmp, profile_continue); } else { mv(tmp, (u1)DataLayout::call_type_data_tag); bne(t0, tmp, profile_continue); } // calculate slot step static int stack_slot_offset0 = in_bytes(TypeEntriesAtCall::stack_slot_offset(0)); static int slot_step = in_bytes(TypeEntriesAtCall::stack_slot_offset(1)) - stack_slot_offset0; // calculate type step static int argument_type_offset0 = in_bytes(TypeEntriesAtCall::argument_type_offset(0)); static int type_step = in_bytes(TypeEntriesAtCall::argument_type_offset(1)) - argument_type_offset0; if (MethodData::profile_arguments()) { Label done, loop, loopEnd, profileArgument, profileReturnType; RegSet pushed_registers; pushed_registers += x15; pushed_registers += x16; pushed_registers += x17; Register mdo_addr = x15; Register index = x16; Register off_to_args = x17; push_reg(pushed_registers, sp); mv(off_to_args, in_bytes(TypeEntriesAtCall::args_data_offset())); mv(t0, TypeProfileArgsLimit); beqz(t0, loopEnd); mv(index, zr); // index < TypeProfileArgsLimit bind(loop); bgtz(index, profileReturnType); mv(t0, (int)MethodData::profile_return()); beqz(t0, profileArgument); // (index > 0 || MethodData::profile_return()) == false bind(profileReturnType); // If return value type is profiled we may have no argument to profile ld(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset()))); mv(t1, - TypeStackSlotEntries::per_arg_count()); mul(t1, index, t1); add(tmp, tmp, t1); mv(t1, TypeStackSlotEntries::per_arg_count()); add(t0, mdp, off_to_args); blt(tmp, t1, done); bind(profileArgument); ld(tmp, Address(callee, Method::const_offset())); load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset())); // stack offset o (zero based) from the start of the argument // list, for n arguments translates into offset n - o - 1 from // the end of the argument list mv(t0, stack_slot_offset0); mv(t1, slot_step); mul(t1, index, t1); add(t0, t0, t1); add(t0, mdp, t0); ld(t0, Address(t0)); sub(tmp, tmp, t0); subi(tmp, tmp, 1); Address arg_addr = argument_address(tmp); ld(tmp, arg_addr); mv(t0, argument_type_offset0); mv(t1, type_step); mul(t1, index, t1); add(t0, t0, t1); add(mdo_addr, mdp, t0); Address mdo_arg_addr(mdo_addr, 0); profile_obj_type(tmp, mdo_arg_addr, t1); int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); addi(off_to_args, off_to_args, to_add); // increment index by 1 addi(index, index, 1); mv(t1, TypeProfileArgsLimit); blt(index, t1, loop); bind(loopEnd); if (MethodData::profile_return()) { ld(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset()))); sub(tmp, tmp, TypeProfileArgsLimit * TypeStackSlotEntries::per_arg_count()); } add(t0, mdp, off_to_args); bind(done); mv(mdp, t0); // unspill the clobbered registers pop_reg(pushed_registers, sp); if (MethodData::profile_return()) { // We're right after the type profile for the last // argument. tmp is the number of cells left in the // CallTypeData/VirtualCallTypeData to reach its end. Non null // if there's a return to profile. assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); shadd(mdp, tmp, mdp, tmp, exact_log2(DataLayout::cell_size)); } sd(mdp, Address(fp, frame::interpreter_frame_mdp_offset * wordSize)); } else { assert(MethodData::profile_return(), "either profile call args or call ret"); update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size())); } // mdp points right after the end of the // CallTypeData/VirtualCallTypeData, right after the cells for the // return value type if there's one bind(profile_continue); } } void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) { assert_different_registers(mdp, ret, tmp, xbcp, t0, t1); if (ProfileInterpreter && MethodData::profile_return()) { Label profile_continue, done; test_method_data_pointer(mdp, profile_continue); if (MethodData::profile_return_jsr292_only()) { assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2"); // If we don't profile all invoke bytecodes we must make sure // it's a bytecode we indeed profile. We can't go back to the // beginning of the ProfileData we intend to update to check its // type because we're right after it and we don't known its // length Label do_profile; lbu(t0, Address(xbcp, 0)); mv(tmp, (u1)Bytecodes::_invokedynamic); beq(t0, tmp, do_profile); mv(tmp, (u1)Bytecodes::_invokehandle); beq(t0, tmp, do_profile); get_method(tmp); lhu(t0, Address(tmp, Method::intrinsic_id_offset())); mv(t1, static_cast(vmIntrinsics::_compiledLambdaForm)); bne(t0, t1, profile_continue); bind(do_profile); } Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size())); mv(tmp, ret); profile_obj_type(tmp, mdo_ret_addr, t1); bind(profile_continue); } } void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2, Register tmp3) { assert_different_registers(t0, t1, mdp, tmp1, tmp2, tmp3); if (ProfileInterpreter && MethodData::profile_parameters()) { Label profile_continue, done; test_method_data_pointer(mdp, profile_continue); // Load the offset of the area within the MDO used for // parameters. If it's negative we're not profiling any parameters lwu(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()))); srli(tmp2, tmp1, 31); bnez(tmp2, profile_continue); // i.e. sign bit set // Compute a pointer to the area for parameters from the offset // and move the pointer to the slot for the last // parameters. Collect profiling from last parameter down. // mdo start + parameters offset + array length - 1 add(mdp, mdp, tmp1); ld(tmp1, Address(mdp, ArrayData::array_len_offset())); subi(tmp1, tmp1, TypeStackSlotEntries::per_arg_count()); Label loop; bind(loop); int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); int type_base = in_bytes(ParametersTypeData::type_offset(0)); int per_arg_scale = exact_log2(DataLayout::cell_size); add(t0, mdp, off_base); add(t1, mdp, type_base); shadd(tmp2, tmp1, t0, tmp2, per_arg_scale); // load offset on the stack from the slot for this parameter ld(tmp2, Address(tmp2, 0)); neg(tmp2, tmp2); // read the parameter from the local area shadd(tmp2, tmp2, xlocals, tmp2, Interpreter::logStackElementSize); ld(tmp2, Address(tmp2, 0)); // profile the parameter shadd(t1, tmp1, t1, t0, per_arg_scale); Address arg_type(t1, 0); profile_obj_type(tmp2, arg_type, tmp3); // go to next parameter subi(tmp1, tmp1, TypeStackSlotEntries::per_arg_count()); bgez(tmp1, loop); bind(profile_continue); } } void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) { // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp // register "cache" is trashed in next ld, so lets use it as a temporary register get_cache_index_at_bcp(index, cache, 1, sizeof(u4)); // Get address of invokedynamic array ld(cache, Address(xcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset()))); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry) slli(index, index, log2i_exact(sizeof(ResolvedIndyEntry))); addi(cache, cache, Array::base_offset_in_bytes()); add(cache, cache, index); } void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) { // Get index out of bytecode pointer get_cache_index_at_bcp(index, cache, bcp_offset, sizeof(u2)); // Take shortcut if the size is a power of 2 if (is_power_of_2(sizeof(ResolvedFieldEntry))) { slli(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2 } else { mv(cache, sizeof(ResolvedFieldEntry)); mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry) } // Get address of field entries array ld(cache, Address(xcpool, ConstantPoolCache::field_entries_offset())); addi(cache, cache, Array::base_offset_in_bytes()); add(cache, cache, index); // Prevents stale data from being read after the bytecode is patched to the fast bytecode membar(MacroAssembler::LoadLoad); } void InterpreterMacroAssembler::get_method_counters(Register method, Register mcs, Label& skip) { Label has_counters; ld(mcs, Address(method, Method::method_counters_offset())); bnez(mcs, has_counters); call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), method); ld(mcs, Address(method, Method::method_counters_offset())); beqz(mcs, skip); // No MethodCounters allocated, OutOfMemory bind(has_counters); } void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) { // Get index out of bytecode pointer get_cache_index_at_bcp(index, cache, bcp_offset, sizeof(u2)); mv(cache, sizeof(ResolvedMethodEntry)); mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry) // Get address of field entries array ld(cache, Address(xcpool, ConstantPoolCache::method_entries_offset())); addi(cache, cache, Array::base_offset_in_bytes()); add(cache, cache, index); } #ifdef ASSERT void InterpreterMacroAssembler::verify_field_offset(Register reg) { // Verify the field offset is not in the header, implicitly checks for 0 Label L; mv(t0, oopDesc::base_offset_in_bytes()); bge(reg, t0, L); stop("bad field offset"); bind(L); } void InterpreterMacroAssembler::verify_access_flags(Register access_flags, uint32_t flag, const char* msg, bool stop_by_hit) { Label L; test_bit(t0, access_flags, exact_log2(flag)); if (stop_by_hit) { beqz(t0, L); } else { bnez(t0, L); } stop(msg); bind(L); } void InterpreterMacroAssembler::verify_frame_setup() { Label L; const Address monitor_block_top(fp, frame::interpreter_frame_monitor_block_top_offset * wordSize); ld(t0, monitor_block_top); shadd(t0, t0, fp, t0, LogBytesPerWord); beq(esp, t0, L); stop("broken stack frame setup in interpreter"); bind(L); } #endif