Google Git Google Git
Sign in
third-party-mirror/openjdk/jdk25u/HEAD/./src/hotspot/cpu/x86/interp_masm_x86.cpp
blob: 92233ee0d079168a4c759a17b4bb7e9b484be806 [file]
/*
* Copyright (c) 1997, 2025, Oracle and/or its affiliates. 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 "compiler/compiler_globals.hpp"
#include "interp_masm_x86.hpp"
#include "interpreter/interpreter.hpp"
#include "interpreter/interpreterRuntime.hpp"
#include "logging/log.hpp"
#include "oops/arrayOop.hpp"
#include "oops/markWord.hpp"
#include "oops/methodData.hpp"
#include "oops/method.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"
// Implementation of InterpreterMacroAssembler
void InterpreterMacroAssembler::jump_to_entry(address entry) {
assert(entry, "Entry must have been generated by now");
jump(RuntimeAddress(entry));
}
void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
Label update, next, none;
assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
interp_verify_oop(obj, atos);
testptr(obj, obj);
jccb(Assembler::notZero, update);
testptr(mdo_addr, TypeEntries::null_seen);
jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore.
// atomic update to prevent overwriting Klass* with 0
lock();
orptr(mdo_addr, TypeEntries::null_seen);
jmpb(next);
bind(update);
load_klass(obj, obj, rscratch1);
mov(rscratch1, obj);
xorptr(obj, mdo_addr);
testptr(obj, TypeEntries::type_klass_mask);
jccb(Assembler::zero, next); // klass seen before, nothing to
// do. The unknown bit may have been
// set already but no need to check.
testptr(obj, TypeEntries::type_unknown);
jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
cmpptr(mdo_addr, 0);
jccb(Assembler::equal, none);
cmpptr(mdo_addr, TypeEntries::null_seen);
jccb(Assembler::equal, none);
// There is a chance that the checks above (re-reading profiling
// data from memory) fail if another thread has just set the
// profiling to this obj's klass
mov(obj, rscratch1);
xorptr(obj, mdo_addr);
testptr(obj, TypeEntries::type_klass_mask);
jccb(Assembler::zero, next);
// different than before. Cannot keep accurate profile.
orptr(mdo_addr, TypeEntries::type_unknown);
jmpb(next);
bind(none);
// first time here. Set profile type.
movptr(mdo_addr, obj);
#ifdef ASSERT
andptr(obj, TypeEntries::type_klass_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());
cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
jcc(Assembler::notEqual, profile_continue);
if (MethodData::profile_arguments()) {
Label done;
int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
addptr(mdp, off_to_args);
for (int i = 0; i < TypeProfileArgsLimit; i++) {
if (i > 0 || MethodData::profile_return()) {
// If return value type is profiled we may have no argument to profile
movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
subl(tmp, i*TypeStackSlotEntries::per_arg_count());
cmpl(tmp, TypeStackSlotEntries::per_arg_count());
jcc(Assembler::less, done);
}
movptr(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
subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
subl(tmp, 1);
Address arg_addr = argument_address(tmp);
movptr(tmp, arg_addr);
Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
profile_obj_type(tmp, mdo_arg_addr);
int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
addptr(mdp, to_add);
off_to_args += to_add;
}
if (MethodData::profile_return()) {
movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
}
bind(done);
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(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
shll(tmp, log2i_exact((int)DataLayout::cell_size));
addptr(mdp, tmp);
}
movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
} 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, _bcp_register);
if (ProfileInterpreter && MethodData::profile_return()) {
Label profile_continue;
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;
cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
jcc(Assembler::equal, do_profile);
cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
jcc(Assembler::equal, do_profile);
get_method(tmp);
cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
jcc(Assembler::notEqual, profile_continue);
bind(do_profile);
}
Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
mov(tmp, ret);
profile_obj_type(tmp, mdo_ret_addr);
bind(profile_continue);
}
}
void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
if (ProfileInterpreter && MethodData::profile_parameters()) {
Label profile_continue;
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
movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
testl(tmp1, tmp1);
jcc(Assembler::negative, profile_continue);
// 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
addptr(mdp, tmp1);
movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
decrement(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));
Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
Address arg_off(mdp, tmp1, per_arg_scale, off_base);
Address arg_type(mdp, tmp1, per_arg_scale, type_base);
// load offset on the stack from the slot for this parameter
movptr(tmp2, arg_off);
negptr(tmp2);
// read the parameter from the local area
movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
// profile the parameter
profile_obj_type(tmp2, arg_type);
// go to next parameter
decrement(tmp1, TypeStackSlotEntries::per_arg_count());
jcc(Assembler::positive, loop);
bind(profile_continue);
}
}
void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
int number_of_arguments) {
// interpreter specific
//
// Note: No need to save/restore bcp & locals registers
// since these are callee saved registers and no blocking/
// GC can happen in leaf calls.
// Further Note: DO NOT save/restore bcp/locals. If a caller has
// already saved them so that it can use rsi/rdi as temporaries
// then a save/restore here will DESTROY the copy the caller
// saved! There used to be a save_bcp() that only happened in
// the ASSERT path (no restore_bcp). Which caused bizarre failures
// when jvm built with ASSERTs.
#ifdef ASSERT
{
Label L;
cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
jcc(Assembler::equal, L);
stop("InterpreterMacroAssembler::call_VM_leaf_base:"
" last_sp != null");
bind(L);
}
#endif
// super call
MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
// interpreter specific
// LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
// but since they may not have been saved (and we don't want to
// save them here (see note above) the assert is invalid.
}
void InterpreterMacroAssembler::call_VM_base(Register oop_result,
Register last_java_sp,
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;
cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
jcc(Assembler::equal, L);
stop("InterpreterMacroAssembler::call_VM_base:"
" last_sp isn't null");
bind(L);
}
#endif /* ASSERT */
// super call
MacroAssembler::call_VM_base(oop_result, last_java_sp,
entry_point, number_of_arguments,
check_exceptions);
// interpreter specific
restore_bcp();
restore_locals();
}
void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
address entry_point,
Register arg_1) {
assert(arg_1 == c_rarg1, "");
Label resume_pc, not_preempted;
#ifdef ASSERT
{
Label L;
cmpptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
jcc(Assembler::equal, L);
stop("Should not have alternate return address set");
bind(L);
}
#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_helper requires last_Java_pc for anchor to be at the top of the stack.
lea(rscratch1, resume_pc);
push(rscratch1);
MacroAssembler::call_VM_helper(oop_result, entry_point, 1, false /*check_exceptions*/);
pop(rscratch1);
pop_cont_fastpath();
// Check if preempted.
movptr(rscratch1, Address(r15_thread, JavaThread::preempt_alternate_return_offset()));
cmpptr(rscratch1, NULL_WORD);
jccb(Assembler::zero, not_preempted);
movptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
jmp(rscratch1);
// 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);
}
void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
call(rscratch1);
if (is_native) {
// On resume we need to set up stack as expected.
push(dtos);
push(ltos);
}
}
void InterpreterMacroAssembler::check_and_handle_popframe() {
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.
Register pop_cond = c_rarg0;
movl(pop_cond, Address(r15_thread, JavaThread::popframe_condition_offset()));
testl(pop_cond, JavaThread::popframe_pending_bit);
jcc(Assembler::zero, L);
testl(pop_cond, JavaThread::popframe_processing_bit);
jcc(Assembler::notZero, 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));
jmp(rax);
bind(L);
}
}
void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
switch (state) {
case atos: movptr(rax, oop_addr);
movptr(oop_addr, NULL_WORD);
interp_verify_oop(rax, state); break;
case ltos: movptr(rax, val_addr); break;
case btos: // fall through
case ztos: // fall through
case ctos: // fall through
case stos: // fall through
case itos: movl(rax, val_addr); break;
case ftos: movflt(xmm0, val_addr); break;
case dtos: movdbl(xmm0, val_addr); break;
case vtos: /* nothing to do */ break;
default : ShouldNotReachHere();
}
// Clean up tos value in the thread object
movl(tos_addr, ilgl);
movptr(val_addr, NULL_WORD);
}
void InterpreterMacroAssembler::check_and_handle_earlyret() {
if (JvmtiExport::can_force_early_return()) {
Label L;
Register tmp = c_rarg0;
Register rthread = r15_thread;
movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
testptr(tmp, tmp);
jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) 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.
movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
cmpl(tmp, JvmtiThreadState::earlyret_pending);
jcc(Assembler::notEqual, L);
// Call Interpreter::remove_activation_early_entry() to get the address of the
// same-named entrypoint in the generated interpreter code.
movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
jmp(rax);
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");
load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
bswapl(reg);
shrl(reg, 16);
}
void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
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_unsigned_short(index, Address(_bcp_register, bcp_offset));
} else if (index_size == sizeof(u4)) {
movl(index, Address(_bcp_register, bcp_offset));
} else if (index_size == sizeof(u1)) {
load_unsigned_byte(index, Address(_bcp_register, 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
movptr(result, Address(result, ConstantPool::cache_offset()));
movptr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
resolve_oop_handle(result, tmp);
load_heap_oop(result, Address(result, index,
UseCompressedOops ? Address::times_4 : Address::times_ptr,
arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
}
// load cpool->resolved_klass_at(index)
void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
Register cpool,
Register index) {
assert_different_registers(cpool, index);
movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
Register resolved_klasses = cpool;
movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset()));
movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
}
// Generate a subtype check: branch to ok_is_subtype if sub_klass is a
// subtype of super_klass.
//
// Args:
// rax: superklass
// Rsub_klass: subklass
//
// Kills:
// rcx, rdi
void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
Label& ok_is_subtype) {
assert(Rsub_klass != rax, "rax holds superklass");
LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
// Profile the not-null value's klass.
profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
// Do the check.
check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
}
// Java Expression Stack
void InterpreterMacroAssembler::pop_ptr(Register r) {
pop(r);
}
void InterpreterMacroAssembler::push_ptr(Register r) {
push(r);
}
void InterpreterMacroAssembler::push_i(Register r) {
push(r);
}
void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
push(r);
}
void InterpreterMacroAssembler::push_f(XMMRegister r) {
subptr(rsp, wordSize);
movflt(Address(rsp, 0), r);
}
void InterpreterMacroAssembler::pop_f(XMMRegister r) {
movflt(r, Address(rsp, 0));
addptr(rsp, wordSize);
}
void InterpreterMacroAssembler::push_d(XMMRegister r) {
subptr(rsp, 2 * wordSize);
movdbl(Address(rsp, 0), r);
}
void InterpreterMacroAssembler::pop_d(XMMRegister r) {
movdbl(r, Address(rsp, 0));
addptr(rsp, 2 * Interpreter::stackElementSize);
}
void InterpreterMacroAssembler::pop_i(Register r) {
// XXX can't use pop currently, upper half non clean
movl(r, Address(rsp, 0));
addptr(rsp, wordSize);
}
void InterpreterMacroAssembler::pop_l(Register r) {
movq(r, Address(rsp, 0));
addptr(rsp, 2 * Interpreter::stackElementSize);
}
void InterpreterMacroAssembler::push_l(Register r) {
subptr(rsp, 2 * wordSize);
movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
}
void InterpreterMacroAssembler::pop(TosState state) {
switch (state) {
case atos: pop_ptr(); break;
case btos:
case ztos:
case ctos:
case stos:
case itos: pop_i(); break;
case ltos: pop_l(); break;
case ftos: pop_f(xmm0); break;
case dtos: pop_d(xmm0); break;
case vtos: /* nothing to do */ break;
default: ShouldNotReachHere();
}
interp_verify_oop(rax, state);
}
void InterpreterMacroAssembler::push(TosState state) {
interp_verify_oop(rax, state);
switch (state) {
case atos: push_ptr(); break;
case btos:
case ztos:
case ctos:
case stos:
case itos: push_i(); break;
case ltos: push_l(); break;
case ftos: push_f(xmm0); break;
case dtos: push_d(xmm0); break;
case vtos: /* nothing to do */ break;
default : ShouldNotReachHere();
}
}
// Helpers for swap and dup
void InterpreterMacroAssembler::load_ptr(int n, Register val) {
movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
}
void InterpreterMacroAssembler::store_ptr(int n, Register val) {
movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
}
void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
// set sender sp
lea(_bcp_register, Address(rsp, wordSize));
// record last_sp
mov(rcx, _bcp_register);
subptr(rcx, rbp);
sarptr(rcx, LogBytesPerWord);
movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx);
}
// 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, Register temp) {
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.
// interp_only is an int, on little endian it is sufficient to test the byte only
// Is a cmpl faster?
cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0);
jccb(Assembler::zero, run_compiled_code);
jmp(Address(method, Method::interpreter_entry_offset()));
bind(run_compiled_code);
}
jmp(Address(method, Method::from_interpreted_offset()));
}
// The following two routines provide a hook so that an implementation
// can schedule the dispatch in two parts. x86 does not do this.
void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
// Nothing x86 specific to be done here
}
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) {
if (VerifyActivationFrameSize) {
Label L;
mov(rcx, rbp);
subptr(rcx, rsp);
int32_t min_frame_size =
(frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
wordSize;
cmpptr(rcx, min_frame_size);
jcc(Assembler::greaterEqual, L);
stop("broken stack frame");
bind(L);
}
if (verifyoop) {
interp_verify_oop(rax, state);
}
address* const safepoint_table = Interpreter::safept_table(state);
Label no_safepoint, dispatch;
if (table != safepoint_table && generate_poll) {
NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
jccb(Assembler::zero, no_safepoint);
lea(rscratch1, ExternalAddress((address)safepoint_table));
jmpb(dispatch);
}
bind(no_safepoint);
lea(rscratch1, ExternalAddress((address)table));
bind(dispatch);
jmp(Address(rscratch1, rbx, Address::times_8));
}
void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
}
void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
dispatch_base(state, Interpreter::normal_table(state));
}
void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
dispatch_base(state, Interpreter::normal_table(state), false);
}
void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
// load next bytecode (load before advancing _bcp_register to prevent AGI)
load_unsigned_byte(rbx, Address(_bcp_register, step));
// advance _bcp_register
increment(_bcp_register, step);
dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
}
void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
// load current bytecode
load_unsigned_byte(rbx, Address(_bcp_register, 0));
dispatch_base(state, table);
}
void InterpreterMacroAssembler::narrow(Register result) {
// Get method->_constMethod->_result_type
movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
movptr(rcx, Address(rcx, Method::const_offset()));
load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
Label done, notBool, notByte, notChar;
// common case first
cmpl(rcx, T_INT);
jcc(Assembler::equal, done);
// mask integer result to narrower return type.
cmpl(rcx, T_BOOLEAN);
jcc(Assembler::notEqual, notBool);
andl(result, 0x1);
jmp(done);
bind(notBool);
cmpl(rcx, T_BYTE);
jcc(Assembler::notEqual, notByte);
movsbl(result, result);
jmp(done);
bind(notByte);
cmpl(rcx, T_CHAR);
jcc(Assembler::notEqual, notChar);
movzwl(result, result);
jmp(done);
bind(notChar);
// cmpl(rcx, T_SHORT); // all that's left
// jcc(Assembler::notEqual, done);
movswl(result, result);
// Nothing to do for T_INT
bind(done);
}
// 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,
Register ret_addr,
bool throw_monitor_exception,
bool install_monitor_exception,
bool notify_jvmdi) {
// Note: Registers rdx xmm0 may be in use for the
// result check if synchronized method
Label unlocked, unlock, no_unlock;
const Register rthread = r15_thread;
const Register robj = c_rarg1;
const Register rmon = c_rarg1;
// get the value of _do_not_unlock_if_synchronized into rdx
const Address do_not_unlock_if_synchronized(rthread,
in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
movbool(rbx, do_not_unlock_if_synchronized);
movbool(do_not_unlock_if_synchronized, false); // reset the flag
// get method access flags
movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
load_unsigned_short(rcx, Address(rcx, Method::access_flags_offset()));
testl(rcx, JVM_ACC_SYNCHRONIZED);
jcc(Assembler::zero, unlocked);
// Don't unlock anything if the _do_not_unlock_if_synchronized flag
// is set.
testbool(rbx);
jcc(Assembler::notZero, 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(rbp, frame::interpreter_frame_initial_sp_offset *
wordSize - (int) sizeof(BasicObjectLock));
// We use c_rarg1/rdx so that if we go slow path it will be the correct
// register for unlock_object to pass to VM directly
lea(robj, monitor); // address of first monitor
movptr(rax, Address(robj, BasicObjectLock::obj_offset()));
testptr(rax, rax);
jcc(Assembler::notZero, 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));
}
jmp(unlocked);
}
bind(unlock);
unlock_object(robj);
pop(state);
// Check that for block-structured locking (i.e., that all locked
// objects has been unlocked)
bind(unlocked);
// rax, rdx: 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(
rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
const Address monitor_block_bot(
rbp, 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
movptr(rmon, monitor_block_top); // derelativize pointer
lea(rmon, Address(rbp, rmon, Address::times_ptr));
// c_rarg1 points to current entry, starting with top-most entry
lea(rbx, monitor_block_bot); // points to word before bottom of
// monitor block
jmp(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);
mov(robj, rmon); // nop if robj and rmon are the same
unlock_object(robj);
pop(state);
if (install_monitor_exception) {
call_VM(noreg, CAST_FROM_FN_PTR(address,
InterpreterRuntime::
new_illegal_monitor_state_exception));
}
jmp(restart);
}
bind(loop);
// check if current entry is used
cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD);
jcc(Assembler::notEqual, exception);
addptr(rmon, entry_size); // otherwise advance to next entry
bind(entry);
cmpptr(rmon, rbx); // check if bottom reached
jcc(Assembler::notEqual, loop); // 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 */);
jmp(fast_path);
bind(slow_path);
push(state);
set_last_Java_frame(noreg, rbp, (address)pc(), rscratch1);
super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), r15_thread);
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 sp
movptr(rbx,
Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
if (StackReservedPages > 0) {
// testing if reserved zone needs to be re-enabled
Register rthread = r15_thread;
Label no_reserved_zone_enabling;
// check if already enabled - if so no re-enabling needed
assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
jcc(Assembler::equal, no_reserved_zone_enabling);
cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
jcc(Assembler::lessEqual, no_reserved_zone_enabling);
JFR_ONLY(leave_jfr_critical_section();)
call_VM_leaf(
CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
call_VM(noreg, CAST_FROM_FN_PTR(address,
InterpreterRuntime::throw_delayed_StackOverflowError));
should_not_reach_here();
bind(no_reserved_zone_enabling);
}
leave(); // remove frame anchor
JFR_ONLY(leave_jfr_critical_section();)
pop(ret_addr); // get return address
mov(rsp, rbx); // set sp to sender sp
pop_cont_fastpath();
}
#if INCLUDE_JFR
void InterpreterMacroAssembler::enter_jfr_critical_section() {
const Address sampling_critical_section(r15_thread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
movbool(sampling_critical_section, true);
}
void InterpreterMacroAssembler::leave_jfr_critical_section() {
const Address sampling_critical_section(r15_thread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
movbool(sampling_critical_section, false);
}
#endif // INCLUDE_JFR
void InterpreterMacroAssembler::get_method_counters(Register method,
Register mcs, Label& skip) {
Label has_counters;
movptr(mcs, Address(method, Method::method_counters_offset()));
testptr(mcs, mcs);
jcc(Assembler::notZero, has_counters);
call_VM(noreg, CAST_FROM_FN_PTR(address,
InterpreterRuntime::build_method_counters), method);
movptr(mcs, Address(method,Method::method_counters_offset()));
testptr(mcs, mcs);
jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
bind(has_counters);
}
// Lock object
//
// Args:
// rdx, c_rarg1: BasicObjectLock to be used for locking
//
// Kills:
// rax, rbx
void InterpreterMacroAssembler::lock_object(Register lock_reg) {
assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
if (LockingMode == LM_MONITOR) {
call_VM_preemptable(noreg,
CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
lock_reg);
} else {
Label count_locking, done, slow_case;
const Register swap_reg = rax; // Must use rax for cmpxchg instruction
const Register tmp_reg = rbx;
const Register obj_reg = c_rarg3; // Will contain the oop
const Register rklass_decode_tmp = rscratch1;
const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
const int mark_offset = lock_offset +
BasicLock::displaced_header_offset_in_bytes();
// Load object pointer into obj_reg
movptr(obj_reg, Address(lock_reg, obj_offset));
if (LockingMode == LM_LIGHTWEIGHT) {
lightweight_lock(lock_reg, obj_reg, swap_reg, tmp_reg, slow_case);
} else if (LockingMode == LM_LEGACY) {
if (DiagnoseSyncOnValueBasedClasses != 0) {
load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
testb(Address(tmp_reg, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
jcc(Assembler::notZero, slow_case);
}
// Load immediate 1 into swap_reg %rax
movl(swap_reg, 1);
// Load (object->mark() | 1) into swap_reg %rax
orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
// Save (object->mark() | 1) into BasicLock's displaced header
movptr(Address(lock_reg, mark_offset), swap_reg);
assert(lock_offset == 0,
"displaced header must be first word in BasicObjectLock");
lock();
cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
jcc(Assembler::zero, count_locking);
const int zero_bits = 7;
// Fast check for recursive lock.
//
// Can apply the optimization only if this is a stack lock
// allocated in this thread. For efficiency, we can focus on
// recently allocated stack locks (instead of reading the stack
// base and checking whether 'mark' points inside the current
// thread stack):
// 1) (mark & zero_bits) == 0, and
// 2) rsp <= mark < mark + os::pagesize()
//
// Warning: rsp + os::pagesize can overflow the stack base. We must
// neither apply the optimization for an inflated lock allocated
// just above the thread stack (this is why condition 1 matters)
// nor apply the optimization if the stack lock is inside the stack
// of another thread. The latter is avoided even in case of overflow
// because we have guard pages at the end of all stacks. Hence, if
// we go over the stack base and hit the stack of another thread,
// this should not be in a writeable area that could contain a
// stack lock allocated by that thread. As a consequence, a stack
// lock less than page size away from rsp is guaranteed to be
// owned by the current thread.
//
// These 3 tests can be done by evaluating the following
// expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
// assuming both stack pointer and pagesize have their
// least significant bits clear.
// NOTE: the mark is in swap_reg %rax as the result of cmpxchg
subptr(swap_reg, rsp);
andptr(swap_reg, zero_bits - (int)os::vm_page_size());
// Save the test result, for recursive case, the result is zero
movptr(Address(lock_reg, mark_offset), swap_reg);
jcc(Assembler::notZero, slow_case);
bind(count_locking);
inc_held_monitor_count();
}
jmp(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:
// rdx, c_rarg1: BasicObjectLock for lock
//
// Kills:
// rax
// c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
// rscratch1 (scratch reg)
// rax, rbx, rcx, rdx
void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
if (LockingMode == LM_MONITOR) {
call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
} else {
Label count_locking, done, slow_case;
const Register swap_reg = rax; // Must use rax for cmpxchg instruction
const Register header_reg = c_rarg2; // Will contain the old oopMark
const Register obj_reg = c_rarg3; // Will contain the oop
save_bcp(); // Save in case of exception
if (LockingMode != LM_LIGHTWEIGHT) {
// Convert from BasicObjectLock structure to object and BasicLock
// structure Store the BasicLock address into %rax
lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
}
// Load oop into obj_reg(%c_rarg3)
movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
// Free entry
movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD);
if (LockingMode == LM_LIGHTWEIGHT) {
lightweight_unlock(obj_reg, swap_reg, header_reg, slow_case);
} else if (LockingMode == LM_LEGACY) {
// Load the old header from BasicLock structure
movptr(header_reg, Address(swap_reg,
BasicLock::displaced_header_offset_in_bytes()));
// Test for recursion
testptr(header_reg, header_reg);
// zero for recursive case
jcc(Assembler::zero, count_locking);
// Atomic swap back the old header
lock();
cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
// zero for simple unlock of a stack-lock case
jcc(Assembler::notZero, slow_case);
bind(count_locking);
dec_held_monitor_count();
}
jmp(done);
bind(slow_case);
// Call the runtime routine for slow case.
movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // 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");
movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
testptr(mdp, mdp);
jcc(Assembler::zero, 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(rax);
push(rbx);
get_method(rbx);
// Test MDO to avoid the call if it is null.
movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
testptr(rax, rax);
jcc(Assembler::zero, set_mdp);
// rbx: method
// _bcp_register: bcp
call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
// rax: mdi
// mdo is guaranteed to be non-zero here, we checked for it before the call.
movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
addptr(rbx, in_bytes(MethodData::data_offset()));
addptr(rax, rbx);
bind(set_mdp);
movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
pop(rbx);
pop(rax);
}
void InterpreterMacroAssembler::verify_method_data_pointer() {
assert(ProfileInterpreter, "must be profiling interpreter");
#ifdef ASSERT
Label verify_continue;
push(rax);
push(rbx);
Register arg3_reg = c_rarg3;
Register arg2_reg = c_rarg2;
push(arg3_reg);
push(arg2_reg);
test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
get_method(rbx);
// If the mdp is valid, it will point to a DataLayout header which is
// consistent with the bcp. The converse is highly probable also.
load_unsigned_short(arg2_reg,
Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
addptr(arg2_reg, Address(rbx, Method::const_offset()));
lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
cmpptr(arg2_reg, _bcp_register);
jcc(Assembler::equal, verify_continue);
// rbx: method
// _bcp_register: bcp
// c_rarg3: mdp
call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
rbx, _bcp_register, arg3_reg);
bind(verify_continue);
pop(arg2_reg);
pop(arg3_reg);
pop(rbx);
pop(rax);
#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);
movptr(data, value);
}
void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
int constant) {
assert(ProfileInterpreter, "must be profiling interpreter");
Address data(mdp_in, constant);
addptr(data, DataLayout::counter_increment);
}
void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
Register index,
int constant) {
assert(ProfileInterpreter, "must be profiling interpreter");
Address data(mdp_in, index, Address::times_1, constant);
addptr(data, DataLayout::counter_increment);
}
void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
int flag_byte_constant) {
assert(ProfileInterpreter, "must be profiling interpreter");
int header_offset = in_bytes(DataLayout::flags_offset());
int header_bits = flag_byte_constant;
// Set the flag
orb(Address(mdp_in, header_offset), header_bits);
}
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) {
cmpptr(value, Address(mdp_in, offset));
} else {
// Put the test value into a register, so caller can use it:
movptr(test_value_out, Address(mdp_in, offset));
cmpptr(test_value_out, value);
}
jcc(Assembler::notEqual, not_equal_continue);
}
void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
int offset_of_disp) {
assert(ProfileInterpreter, "must be profiling interpreter");
Address disp_address(mdp_in, offset_of_disp);
addptr(mdp_in, disp_address);
movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
}
void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
Register reg,
int offset_of_disp) {
assert(ProfileInterpreter, "must be profiling interpreter");
Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
addptr(mdp_in, disp_address);
movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
}
void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
int constant) {
assert(ProfileInterpreter, "must be profiling interpreter");
addptr(mdp_in, constant);
movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
}
void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
assert(ProfileInterpreter, "must be profiling interpreter");
push(return_bci); // save/restore across call_VM
call_VM(noreg,
CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
return_bci);
pop(return_bci);
}
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,
Register reg2,
bool receiver_can_be_null) {
if (ProfileInterpreter) {
Label profile_continue;
// If no method data exists, go to profile_continue.
test_method_data_pointer(mdp, profile_continue);
Label skip_receiver_profile;
if (receiver_can_be_null) {
Label not_null;
testptr(receiver, receiver);
jccb(Assembler::notZero, not_null);
// We are making a call. Increment the count for null receiver.
increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
jmp(skip_receiver_profile);
bind(not_null);
}
// Record the receiver type.
record_klass_in_profile(receiver, mdp, reg2, true);
bind(skip_receiver_profile);
// 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);
}
}
// This routine creates a state machine for updating the multi-row
// type profile at a virtual call site (or other type-sensitive bytecode).
// The machine visits each row (of receiver/count) until the receiver type
// is found, or until it runs out of rows. At the same time, it remembers
// the location of the first empty row. (An empty row records null for its
// receiver, and can be allocated for a newly-observed receiver type.)
// Because there are two degrees of freedom in the state, a simple linear
// search will not work; it must be a decision tree. Hence this helper
// function is recursive, to generate the required tree structured code.
// It's the interpreter, so we are trading off code space for speed.
// See below for example code.
void InterpreterMacroAssembler::record_klass_in_profile_helper(
Register receiver, Register mdp,
Register reg2, int start_row,
Label& done, bool is_virtual_call) {
if (TypeProfileWidth == 0) {
increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
} else {
record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
&VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
}
}
void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row,
Label& done, int total_rows,
OffsetFunction item_offset_fn,
OffsetFunction item_count_offset_fn) {
int last_row = total_rows - 1;
assert(start_row <= last_row, "must be work left to do");
// Test this row for both the item and for null.
// Take any of three different outcomes:
// 1. found item => increment count and goto done
// 2. found null => keep looking for case 1, maybe allocate this cell
// 3. found something else => keep looking for cases 1 and 2
// Case 3 is handled by a recursive call.
for (int row = start_row; row <= last_row; row++) {
Label next_test;
bool test_for_null_also = (row == start_row);
// See if the item is item[n].
int item_offset = in_bytes(item_offset_fn(row));
test_mdp_data_at(mdp, item_offset, item,
(test_for_null_also ? reg2 : noreg),
next_test);
// (Reg2 now contains the item from the CallData.)
// The item is item[n]. Increment count[n].
int count_offset = in_bytes(item_count_offset_fn(row));
increment_mdp_data_at(mdp, count_offset);
jmp(done);
bind(next_test);
if (test_for_null_also) {
// Failed the equality check on item[n]... Test for null.
testptr(reg2, reg2);
if (start_row == last_row) {
// The only thing left to do is handle the null case.
Label found_null;
jccb(Assembler::zero, found_null);
// Item did not match any saved item and there is no empty row for it.
// Increment total counter to indicate polymorphic case.
increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
jmp(done);
bind(found_null);
break;
}
Label found_null;
// Since null is rare, make it be the branch-taken case.
jcc(Assembler::zero, found_null);
// Put all the "Case 3" tests here.
record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
item_offset_fn, item_count_offset_fn);
// Found a null. Keep searching for a matching item,
// but remember that this is an empty (unused) slot.
bind(found_null);
}
}
// In the fall-through case, we found no matching item, but we
// observed the item[start_row] is null.
// Fill in the item field and increment the count.
int item_offset = in_bytes(item_offset_fn(start_row));
set_mdp_data_at(mdp, item_offset, item);
int count_offset = in_bytes(item_count_offset_fn(start_row));
movl(reg2, DataLayout::counter_increment);
set_mdp_data_at(mdp, count_offset, reg2);
if (start_row > 0) {
jmp(done);
}
}
// Example state machine code for three profile rows:
// // main copy of decision tree, rooted at row[1]
// if (row[0].rec == rec) { row[0].incr(); goto done; }
// if (row[0].rec != nullptr) {
// // inner copy of decision tree, rooted at row[1]
// if (row[1].rec == rec) { row[1].incr(); goto done; }
// if (row[1].rec != nullptr) {
// // degenerate decision tree, rooted at row[2]
// if (row[2].rec == rec) { row[2].incr(); goto done; }
// if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
// row[2].init(rec); goto done;
// } else {
// // remember row[1] is empty
// if (row[2].rec == rec) { row[2].incr(); goto done; }
// row[1].init(rec); goto done;
// }
// } else {
// // remember row[0] is empty
// if (row[1].rec == rec) { row[1].incr(); goto done; }
// if (row[2].rec == rec) { row[2].incr(); goto done; }
// row[0].init(rec); goto done;
// }
// done:
void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
Register mdp, Register reg2,
bool is_virtual_call) {
assert(ProfileInterpreter, "must be profiling");
Label done;
record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
bind (done);
}
void InterpreterMacroAssembler::profile_ret(Register return_bci,
Register mdp) {
if (ProfileInterpreter) {
Label profile_continue;
uint row;
// 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 (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)));
jmp(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, Register reg2) {
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.
record_klass_in_profile(klass, mdp, reg2, false);
}
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()
movl(reg2, in_bytes(MultiBranchData::per_case_size()));
imulptr(index, reg2); // XXX l ?
addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
// Update the case count
increment_mdp_data_at(mdp,
index,
in_bytes(MultiBranchData::relative_count_offset()));
// The method data pointer needs to be updated.
update_mdp_by_offset(mdp,
index,
in_bytes(MultiBranchData::
relative_displacement_offset()));
bind(profile_continue);
}
}
void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
if (state == atos) {
MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
}
}
// Jump if ((*counter_addr += increment) & mask) == 0
void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
Register scratch, Label* where) {
// This update is actually not atomic and can lose a number of updates
// under heavy contention, but the alternative of using the (contended)
// atomic update here penalizes profiling paths too much.
movl(scratch, counter_addr);
incrementl(scratch, InvocationCounter::count_increment);
movl(counter_addr, scratch);
andl(scratch, mask);
if (where != nullptr) {
jcc(Assembler::zero, *where);
}
}
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.
Register rthread = r15_thread;
Register rarg = c_rarg1;
if (JvmtiExport::can_post_interpreter_events()) {
Label L;
movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
testl(rdx, rdx);
jcc(Assembler::zero, L);
call_VM(noreg, CAST_FROM_FN_PTR(address,
InterpreterRuntime::post_method_entry));
bind(L);
}
if (DTraceMethodProbes) {
get_method(rarg);
call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
rthread, rarg);
}
// RedefineClasses() tracing support for obsolete method entry
if (log_is_enabled(Trace, redefine, class, obsolete)) {
get_method(rarg);
call_VM_leaf(
CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
rthread, rarg);
}
}
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.
Register rthread = r15_thread;
Register rarg = c_rarg1;
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);
movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
testl(rdx, rdx);
jcc(Assembler::zero, L);
call_VM(noreg,
CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
bind(L);
pop(state);
}
if (DTraceMethodProbes) {
push(state);
get_method(rarg);
call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
rthread, rarg);
pop(state);
}
}
void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
// Get index out of bytecode pointer
get_cache_index_at_bcp(index, 1, sizeof(u4));
// Get address of invokedynamic array
movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
if (is_power_of_2(sizeof(ResolvedIndyEntry))) {
shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2
} else {
imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
}
lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes()));
}
void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
// Get index out of bytecode pointer
movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset()));
// Take shortcut if the size is a power of 2
if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
} else {
imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
}
lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes()));
}
void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
// Get index out of bytecode pointer
movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset()));
imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes()));
}