inline void MacroAssembler::load_heap_oop(Register d, RegisterOrConstant offs, Register s1) {
if (UseCompressedOops) {
lwz(d, offs, s1);
decode_heap_oop(d);
} else {
ld(d, offs, s1);
}
}
void CppInterpreterGenerator::generate_convert_result(address* converter_array)
{
__ load (r5, (intptr_t) converter_array);
__ lwz (r0, Address(Rmethod, methodOopDesc::result_index_offset()));
__ shift_left (r0, r0, LogBytesPerWord);
__ load_indexed (r0, r5, r0);
__ mtctr (r0);
__ bctrl ();
}
inline void MacroAssembler::load_heap_oop_not_null(Register d, RegisterOrConstant offs, Register s1) {
if (UseCompressedOops) {
lwz(d, offs, s1);
// Attention: no null check here!
decode_heap_oop_not_null(d);
} else {
ld(d, offs, s1);
}
}
address CppInterpreterGenerator::generate_stack_to_stack_converter(
BasicType type)
{
const Register stack = r3;
address start = __ pc();
switch (type) {
case T_VOID:
break;
case T_BOOLEAN:
case T_CHAR:
case T_BYTE:
case T_SHORT:
case T_INT:
case T_FLOAT:
__ load (stack, STATE(_stack));
__ lwz (r0, Address(stack, wordSize));
__ stw (r0, Address(Rlocals, 0));
__ subi (Rlocals, Rlocals, wordSize);
break;
case T_LONG:
case T_DOUBLE:
__ load (stack, STATE(_stack));
__ load (r0, Address(stack, wordSize));
__ store (r0, Address(Rlocals, -wordSize));
#ifdef PPC32
__ load (r0, Address(stack, wordSize * 2));
__ store (r0, Address(Rlocals, 0));
#endif
__ subi (Rlocals, Rlocals, wordSize * 2);
break;
case T_OBJECT:
__ load (stack, STATE(_stack));
__ load (r0, Address(stack, wordSize));
__ verify_oop (r0);
__ store (r0, Address(Rlocals, 0));
__ subi (Rlocals, Rlocals, wordSize);
break;
default:
ShouldNotReachHere();
}
__ blr ();
return start;
}
void C1_MacroAssembler::initialize_object(
Register obj, // result: pointer to object after successful allocation
Register klass, // object klass
Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
int con_size_in_bytes, // object size in bytes if known at compile time
Register t1, // temp register
Register t2 // temp register
) {
const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
initialize_header(obj, klass, noreg, t1, t2);
#ifdef ASSERT
{
lwz(t1, in_bytes(Klass::layout_helper_offset()), klass);
if (var_size_in_bytes != noreg) {
cmpw(CCR0, t1, var_size_in_bytes);
} else {
cmpwi(CCR0, t1, con_size_in_bytes);
}
asm_assert_eq("bad size in initialize_object", 0x753);
}
#endif
// Initialize body.
if (var_size_in_bytes != noreg) {
// Use a loop.
addi(t1, obj, hdr_size_in_bytes); // Compute address of first element.
addi(t2, var_size_in_bytes, -hdr_size_in_bytes); // Compute size of body.
initialize_body(t1, t2);
} else if (con_size_in_bytes > hdr_size_in_bytes) {
// Use a loop.
initialize_body(obj, t1, t2, con_size_in_bytes, hdr_size_in_bytes);
}
if (CURRENT_ENV->dtrace_alloc_probes()) {
Unimplemented();
// assert(obj == O0, "must be");
// call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
// relocInfo::runtime_call_type);
}
verify_oop(obj);
}
void MethodHandles::jump_from_method_handle(MacroAssembler* _masm, Register method, Register target, Register temp,
bool for_compiler_entry) {
Label L_no_such_method;
assert(method == R19_method, "interpreter calling convention");
assert_different_registers(method, target, temp);
if (!for_compiler_entry && 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.
__ verify_thread();
__ lwz(temp, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
__ cmplwi(CCR0, temp, 0);
__ beq(CCR0, run_compiled_code);
// Null method test is replicated below in compiled case,
// it might be able to address across the verify_thread()
__ cmplwi(CCR0, R19_method, 0);
__ beq(CCR0, L_no_such_method);
__ ld(target, in_bytes(Method::interpreter_entry_offset()), R19_method);
__ mtctr(target);
__ bctr();
__ BIND(run_compiled_code);
}
// Compiled case, either static or fall-through from runtime conditional
__ cmplwi(CCR0, R19_method, 0);
__ beq(CCR0, L_no_such_method);
const ByteSize entry_offset = for_compiler_entry ? Method::from_compiled_offset() :
Method::from_interpreted_offset();
__ ld(target, in_bytes(entry_offset), R19_method);
__ mtctr(target);
__ bctr();
__ bind(L_no_such_method);
assert(StubRoutines::throw_AbstractMethodError_entry() != NULL, "not yet generated!");
__ load_const_optimized(target, StubRoutines::throw_AbstractMethodError_entry());
__ mtctr(target);
__ bctr();
}
void
emit_trampoline(struct compilation_unit *cu, void *target_addr, struct jit_trampoline *t)
{
struct buffer *b = t->objcode;
jit_text_lock();
b->buf = jit_text_ptr();
/* Allocate memory on the stack */
emit(b, stwu(1, -16, 1));
/* Save LR on stack */
emit(b, mflr(0));
emit(b, stw(0, 0, 1));
/* Pass pointer to 'struct compilation_unit' as first argument */
emit(b, lis(3, ptr_high(cu)));
emit(b, ori(3, 3, ptr_low(cu)));
/* Then call 'target_addr' */
emit(b, lis(0, ptr_high(target_addr)));
emit(b, ori(0, 0, ptr_low(target_addr)));
emit(b, mtctr(0));
emit(b, bctrl());
/* Restore LR from stack */
emit(b, lwz(0, 0, 1));
emit(b, mtlr(0));
/* Free memory on stack */
emit(b, addi(1, 1, 16));
/* Finally jump to the compiled method */
emit(b, mtctr(3));
emit(b, bctr());
jit_text_reserve(buffer_offset(b));
jit_text_unlock();
}
address InterpreterGenerator::generate_normal_entry(bool synchronized)
{
assert_different_registers(Rmethod, Rlocals, Rthread, Rstate, Rmonitor);
Label re_dispatch;
Label call_interpreter;
Label call_method;
Label call_non_interpreted_method;
Label return_with_exception;
Label return_from_method;
Label resume_interpreter;
Label return_to_initial_caller;
Label more_monitors;
Label throwing_exception;
// We use the same code for synchronized and not
if (normal_entry)
return normal_entry;
address start = __ pc();
// There are two ways in which we can arrive at this entry.
// There is the special case where a normal interpreted method
// calls another normal interpreted method, and there is the
// general case of when we enter from somewhere else: from
// call_stub, from C1 or C2, or from a fast accessor which
// deferred. In the special case we're already in frame manager
// code: we arrive at re_dispatch with Rstate containing the
// previous interpreter state. In the general case we arrive
// at start with no previous interpreter state so we set Rstate
// to NULL to indicate this.
__ bind (fast_accessor_slow_entry_path);
__ load (Rstate, 0);
__ bind (re_dispatch);
// Adjust the caller's stack frame to accomodate any additional
// local variables we have contiguously with our parameters.
generate_adjust_callers_stack();
// Allocate and initialize our stack frame.
generate_compute_interpreter_state(false);
// Call the interpreter ==============================================
__ bind (call_interpreter);
// We can setup the frame anchor with everything we want at
// this point as we are thread_in_Java and no safepoints can
// occur until we go to vm mode. We do have to clear flags
// on return from vm but that is it
__ set_last_Java_frame ();
// Call interpreter
address interpreter = JvmtiExport::can_post_interpreter_events() ?
CAST_FROM_FN_PTR(address, BytecodeInterpreter::runWithChecks) :
CAST_FROM_FN_PTR(address, BytecodeInterpreter::run);
__ mr (r3, Rstate);
__ call (interpreter);
__ fixup_after_potential_safepoint ();
// Clear the frame anchor
__ reset_last_Java_frame ();
// Examine the message from the interpreter to decide what to do
__ lwz (r4, STATE(_msg));
__ compare (r4, BytecodeInterpreter::call_method);
__ beq (call_method);
__ compare (r4, BytecodeInterpreter::return_from_method);
__ beq (return_from_method);
__ compare (r4, BytecodeInterpreter::more_monitors);
__ beq (more_monitors);
__ compare (r4, BytecodeInterpreter::throwing_exception);
__ beq (throwing_exception);
__ load (r3, (intptr_t) "error: bad message from interpreter: %d\n");
__ call (CAST_FROM_FN_PTR(address, printf));
__ should_not_reach_here (__FILE__, __LINE__);
// Handle a call_method message ======================================
__ bind (call_method);
__ load (Rmethod, STATE(_result._to_call._callee));
__ verify_oop(Rmethod);
__ load (Rlocals, STATE(_stack));
__ lhz (r0, Address(Rmethod, methodOopDesc::size_of_parameters_offset()));
__ shift_left (r0, r0, LogBytesPerWord);
__ add (Rlocals, Rlocals, r0);
__ load (r0, STATE(_result._to_call._callee_entry_point));
__ load (r3, (intptr_t) start);
__ compare (r0, r3);
__ bne (call_non_interpreted_method);
// Interpreted methods are intercepted and re-dispatched -----------
__ load (r0, CAST_FROM_FN_PTR(intptr_t, RecursiveInterpreterActivation));
__ mtlr (r0);
__ b (re_dispatch);
// Non-interpreted methods are dispatched normally -----------------
__ bind (call_non_interpreted_method);
__ mtctr (r0);
__ bctrl ();
// Restore Rstate
__ load (Rstate, Address(r1, StackFrame::back_chain_offset * wordSize));
__ subi (Rstate, Rstate, sizeof(BytecodeInterpreter));
// Check for pending exceptions
__ load (r0, Address(Rthread, Thread::pending_exception_offset()));
__ compare (r0, 0);
__ bne (return_with_exception);
// Convert the result and resume
generate_convert_result(CppInterpreter::_tosca_to_stack);
__ b (resume_interpreter);
// Handle a return_from_method message ===============================
__ bind (return_from_method);
__ load (r0, STATE(_prev_link));
__ compare (r0, 0);
__ beq (return_to_initial_caller);
// "Return" from a re-dispatch -------------------------------------
generate_convert_result(CppInterpreter::_stack_to_stack);
generate_unwind_interpreter_state();
// Resume the interpreter
__ bind (resume_interpreter);
__ store (Rlocals, STATE(_stack));
__ load (Rlocals, STATE(_locals));
__ load (Rmethod, STATE(_method));
__ verify_oop(Rmethod);
__ load (r0, BytecodeInterpreter::method_resume);
__ stw (r0, STATE(_msg));
__ b (call_interpreter);
// Return to the initial caller (call_stub etc) --------------------
__ bind (return_to_initial_caller);
generate_convert_result(CppInterpreter::_stack_to_native_abi);
generate_unwind_interpreter_state();
__ blr ();
// Handle a more_monitors message ====================================
__ bind (more_monitors);
generate_more_monitors();
__ load (r0, BytecodeInterpreter::got_monitors);
__ stw (r0, STATE(_msg));
__ b (call_interpreter);
// Handle a throwing_exception message ===============================
__ bind (throwing_exception);
// Check we actually have an exception
#ifdef ASSERT
{
Label ok;
__ load (r0, Address(Rthread, Thread::pending_exception_offset()));
__ compare (r0, 0);
__ bne (ok);
__ should_not_reach_here (__FILE__, __LINE__);
__ bind (ok);
}
#endif
// Return to wherever
generate_unwind_interpreter_state();
__ bind (return_with_exception);
__ compare (Rstate, 0);
__ bne (resume_interpreter);
__ blr ();
normal_entry = start;
return start;
}
address InterpreterGenerator::generate_native_entry(bool synchronized)
{
const Register handler = r14;
const Register function = r15;
assert_different_registers(Rmethod, Rlocals, Rthread, Rstate, Rmonitor,
handler, function);
// We use the same code for synchronized and not
if (native_entry)
return native_entry;
address start = __ pc();
// Allocate and initialize our stack frame.
__ load (Rstate, 0);
generate_compute_interpreter_state(true);
// Make sure method is native and not abstract
#ifdef ASSERT
{
Label ok;
__ lwz (r0, Address(Rmethod, methodOopDesc::access_flags_offset()));
__ andi_ (r0, r0, JVM_ACC_NATIVE | JVM_ACC_ABSTRACT);
__ compare (r0, JVM_ACC_NATIVE);
__ beq (ok);
__ should_not_reach_here (__FILE__, __LINE__);
__ bind (ok);
}
#endif
// Lock if necessary
Label not_synchronized_1;
__ bne (CRsync, not_synchronized_1);
__ lock_object (Rmonitor);
__ bind (not_synchronized_1);
// Get signature handler
const Address signature_handler_addr(
Rmethod, methodOopDesc::signature_handler_offset());
Label return_to_caller, got_signature_handler;
__ load (handler, signature_handler_addr);
__ compare (handler, 0);
__ bne (got_signature_handler);
__ call_VM (noreg,
CAST_FROM_FN_PTR(address,
InterpreterRuntime::prepare_native_call),
Rmethod,
CALL_VM_NO_EXCEPTION_CHECKS);
__ load (r0, Address(Rthread, Thread::pending_exception_offset()));
__ compare (r0, 0);
__ bne (return_to_caller);
__ load (handler, signature_handler_addr);
__ bind (got_signature_handler);
// Get the native function entry point
const Address native_function_addr(
Rmethod, methodOopDesc::native_function_offset());
Label got_function;
__ load (function, native_function_addr);
#ifdef ASSERT
{
// InterpreterRuntime::prepare_native_call() sets the mirror
// handle and native function address first and the signature
// handler last, so function should always be set here.
Label ok;
__ compare (function, 0);
__ bne (ok);
__ should_not_reach_here (__FILE__, __LINE__);
__ bind (ok);
}
#endif
// Call signature handler
__ mtctr (handler);
__ bctrl ();
__ mr (handler, r0);
// Pass JNIEnv
__ la (r3, Address(Rthread, JavaThread::jni_environment_offset()));
// Pass mirror handle if static
const Address oop_temp_addr = STATE(_oop_temp);
Label not_static;
__ bne (CRstatic, not_static);
__ get_mirror_handle (r4);
__ store (r4, oop_temp_addr);
__ la (r4, oop_temp_addr);
__ bind (not_static);
// Set up the Java frame anchor
__ set_last_Java_frame ();
// Change the thread state to native
const Address thread_state_addr(Rthread, JavaThread::thread_state_offset());
#ifdef ASSERT
{
Label ok;
__ lwz (r0, thread_state_addr);
__ compare (r0, _thread_in_Java);
__ beq (ok);
__ should_not_reach_here (__FILE__, __LINE__);
__ bind (ok);
}
#endif
__ load (r0, _thread_in_native);
__ stw (r0, thread_state_addr);
// Make the call
__ call (function);
__ fixup_after_potential_safepoint ();
// The result will be in r3 (and maybe r4 on 32-bit) or f1.
// Wherever it is, we need to store it before calling anything
const Register r3_save = r16;
#ifdef PPC32
const Register r4_save = r17;
#endif
const FloatRegister f1_save = f14;
__ mr (r3_save, r3);
#ifdef PPC32
__ mr (r4_save, r4);
#endif
__ fmr (f1_save, f1);
// Switch thread to "native transition" state before reading the
// synchronization state. This additional state is necessary
// because reading and testing the synchronization state is not
// atomic with respect to garbage collection.
__ load (r0, _thread_in_native_trans);
__ stw (r0, thread_state_addr);
// Ensure the new state is visible to the VM thread.
if(os::is_MP()) {
if (UseMembar)
__ sync ();
else
__ serialize_memory (r3, r4);
}
// Check for safepoint operation in progress and/or pending
// suspend requests. We use a leaf call in order to leave
// the last_Java_frame setup undisturbed.
Label block, no_block;
__ load (r3, (intptr_t) SafepointSynchronize::address_of_state());
__ lwz (r0, Address(r3, 0));
__ compare (r0, SafepointSynchronize::_not_synchronized);
__ bne (block);
__ lwz (r0, Address(Rthread, JavaThread::suspend_flags_offset()));
__ compare (r0, 0);
__ beq (no_block);
__ bind (block);
__ call_VM_leaf (
CAST_FROM_FN_PTR(address,
JavaThread::check_special_condition_for_native_trans));
__ fixup_after_potential_safepoint ();
__ bind (no_block);
// Change the thread state
__ load (r0, _thread_in_Java);
__ stw (r0, thread_state_addr);
// Reset the frame anchor
__ reset_last_Java_frame ();
// If the result was an OOP then unbox it and store it in the frame
// (where it will be safe from garbage collection) before we release
// the handle it might be protected by
Label non_oop, store_oop;
__ load (r0, (intptr_t) AbstractInterpreter::result_handler(T_OBJECT));
__ compare (r0, handler);
__ bne (non_oop);
__ compare (r3_save, 0);
__ beq (store_oop);
__ load (r3_save, Address(r3_save, 0));
__ bind (store_oop);
__ store (r3_save, STATE(_oop_temp));
__ bind (non_oop);
// Reset handle block
__ load (r3, Address(Rthread, JavaThread::active_handles_offset()));
__ load (r0, 0);
__ stw (r0, Address(r3, JNIHandleBlock::top_offset_in_bytes()));
// If there is an exception we skip the result handler and return.
// Note that this also skips unlocking which seems totally wrong,
// but apparently this is what the asm interpreter does so we do
// too.
__ load (r0, Address(Rthread, Thread::pending_exception_offset()));
__ compare (r0, 0);
__ bne (return_to_caller);
// Unlock if necessary
Label not_synchronized_2;
__ bne (CRsync, not_synchronized_2);
__ unlock_object (Rmonitor);
__ bind (not_synchronized_2);
// Restore saved result and call the result handler
__ mr (r3, r3_save);
#ifdef PPC32
__ mr (r4, r4_save);
#endif
__ fmr (f1, f1_save);
__ mtctr (handler);
__ bctrl ();
// Unwind the current activation and return
__ bind (return_to_caller);
generate_unwind_interpreter_state();
__ blr ();
native_entry = start;
return start;
}
address InterpreterGenerator::generate_accessor_entry()
{
if (!UseFastAccessorMethods)
return NULL;
Label& slow_path = fast_accessor_slow_entry_path;
address start = __ pc();
// Drop into the slow path if we need a safepoint check.
__ load (r3, (intptr_t) SafepointSynchronize::address_of_state());
__ load (r0, Address(r3, 0));
__ compare (r0, SafepointSynchronize::_not_synchronized);
__ bne (slow_path);
// Load the object pointer and drop into the slow path
// if we have a NullPointerException.
const Register object = r4;
__ load (object, Address(Rlocals, 0));
__ compare (object, 0);
__ beq (slow_path);
// Read the field index from the bytecode, which looks like this:
// 0: 0x2a: aload_0
// 1: 0xb4: getfield
// 2: index (high byte)
// 3: index (low byte)
// 4: 0xac/b0: ireturn/areturn
const Register index = r5;
__ load (index, Address(Rmethod, methodOopDesc::const_offset()));
__ lwz (index, Address(index, constMethodOopDesc::codes_offset()));
#ifdef ASSERT
{
Label ok;
__ shift_right (r0, index, 16);
__ compare (r0, (Bytecodes::_aload_0 << 8) | Bytecodes::_getfield);
__ beq (ok);
__ should_not_reach_here (__FILE__, __LINE__);
__ bind (ok);
}
#endif
__ andi_ (index, index, 0xffff);
// Locate the entry in the constant pool cache
const Register entry = r6;
__ load (entry, Address(Rmethod, methodOopDesc::constants_offset()));
__ load (entry, Address(entry,constantPoolOopDesc::cache_offset_in_bytes()));
__ la (entry, Address(entry, constantPoolCacheOopDesc::base_offset()));
__ shift_left(r0, index,
exact_log2(in_words(ConstantPoolCacheEntry::size())) + LogBytesPerWord);
__ add (entry, entry, r0);
// Check the validity of the cache entry by testing whether the
// _indices field contains Bytecode::_getfield in b1 byte.
__ load (r0, Address(entry, ConstantPoolCacheEntry::indices_offset()));
__ shift_right (r0, r0, 16);
__ andi_ (r0, r0, 0xff);
__ compare (r0, Bytecodes::_getfield);
__ bne (slow_path);
// Calculate the type and offset of the field
const Register offset = r7;
const Register type = r8;
__ load (offset, Address(entry, ConstantPoolCacheEntry::f2_offset()));
__ load (type, Address(entry, ConstantPoolCacheEntry::flags_offset()));
ConstantPoolCacheEntry::verify_tosBits();
__ shift_right (type, type, ConstantPoolCacheEntry::tosBits);
// Load the value
Label is_object, is_int, is_byte, is_short, is_char;
__ compare (type, atos);
__ beq (is_object);
__ compare (type, itos);
__ beq (is_int);
__ compare (type, btos);
__ beq (is_byte);
__ compare (type, stos);
__ beq (is_short);
__ compare (type, ctos);
__ beq (is_char);
__ load (r3, (intptr_t) "error: unknown type: %d\n");
__ mr (r4, type);
__ call (CAST_FROM_FN_PTR(address, printf));
__ should_not_reach_here (__FILE__, __LINE__);
__ bind (is_object);
__ load_indexed (r3, object, offset);
__ blr ();
__ bind (is_int);
__ lwax (r3, object, offset);
__ blr ();
__ bind (is_byte);
__ lbax (r3, object, offset);
__ blr ();
__ bind (is_short);
__ lhax (r3, object, offset);
__ blr ();
__ bind (is_char);
__ lhzx (r3, object, offset);
__ blr ();
return start;
}
void CppInterpreterGenerator::generate_compute_interpreter_state(bool native)
{
StackFrame frame;
const Address stack_words_addr(
Rmethod, methodOopDesc::max_stack_offset());
const Address access_flags_addr(
Rmethod, methodOopDesc::access_flags_offset());
Label not_synchronized_1, not_synchronized_2, not_synchronized_3;
Label not_static, init_monitor;
const int monitor_size = frame::interpreter_frame_monitor_size() * wordSize;
// Calculate the access flags conditions
const Register access_flags = r3;
__ lwz (access_flags, access_flags_addr);
__ andi_ (r0, access_flags, JVM_ACC_SYNCHRONIZED);
__ compare (CRsync, r0, JVM_ACC_SYNCHRONIZED);
__ andi_ (r0, access_flags, JVM_ACC_STATIC);
__ compare (CRstatic, r0, JVM_ACC_STATIC);
const int basic_frame_size =
frame.unaligned_size() + sizeof(BytecodeInterpreter) + slop_factor;
// Calculate the frame size
const Register stack_size = r3;
const Register frame_size = r4;
const Register padding = r5;
if (native) {
__ load (frame_size, basic_frame_size);
}
else {
__ lhz (stack_size, stack_words_addr);
__ shift_left (stack_size, stack_size, LogBytesPerWord);
__ addi (frame_size, stack_size, basic_frame_size);
}
__ bne (CRsync, not_synchronized_1);
__ addi (frame_size, frame_size, monitor_size);
__ bind (not_synchronized_1);
__ calc_padding_for_alignment (padding, frame_size, StackAlignmentInBytes);
__ add (frame_size, frame_size, padding);
// Save the link register and create the new frame
__ mflr (r0);
__ store (r0, Address(r1, StackFrame::lr_save_offset * wordSize));
__ neg (r0, frame_size);
__ store_update_indexed (r1, r1, r0);
// Calculate everything's addresses
const Register stack_limit = r6;
const Register stack = r7;
const Register stack_base = Rmonitor;
const Register monitor_base = r8;
__ addi (stack_limit, r1, frame.start_of_locals() + slop_factor - wordSize);
__ add (stack_limit, stack_limit, padding);
if (native)
__ mr (stack, stack_limit);
else
__ add (stack, stack_limit, stack_size);
__ addi (stack_base, stack, wordSize);
__ mr (monitor_base, stack_base);
__ bne (CRsync, not_synchronized_2);
__ addi (monitor_base, monitor_base, monitor_size);
__ bind (not_synchronized_2);
__ mr (r0, Rstate);
__ mr (Rstate, monitor_base);
// Initialise the interpreter state object
__ store (Rlocals, STATE(_locals));
__ store (Rmethod, STATE(_method));
__ store (Rstate, STATE(_self_link));
__ store (r0, STATE(_prev_link));
__ store (stack_limit, STATE(_stack_limit));
__ store (stack, STATE(_stack));
__ store (stack_base, STATE(_stack_base));
__ store (monitor_base, STATE(_monitor_base));
__ store (Rthread, STATE(_thread));
#ifdef ASSERT
{
Label ok;
__ load (r3, ThreadLocalStorage::thread_index());
__ call (CAST_FROM_FN_PTR(address, pthread_getspecific));
__ compare (Rthread, r3);
__ beq (ok);
__ should_not_reach_here (__FILE__, __LINE__);
__ bind (ok);
}
#endif
if (!native) {
__ load (r3, Address(Rmethod, methodOopDesc::const_offset()));
__ addi (r3, r3, in_bytes(constMethodOopDesc::codes_offset()));
__ store (r3, STATE(_bcp));
}
__ load (r3, Address(Rmethod, methodOopDesc::constants_offset()));
__ load (r3, Address(r3, constantPoolOopDesc::cache_offset_in_bytes()));
__ store (r3, STATE(_constants));
__ load (r3, BytecodeInterpreter::method_entry);
__ stw (r3, STATE(_msg));
__ load (r3, 0);
if (native)
__ store (r3, STATE(_bcp));
__ store (r3, STATE(_oop_temp));
__ store (r3, STATE(_mdx));
__ store (r3, STATE(_result._to_call._callee));
// Initialise the monitor if synchronized
__ bne (CRsync, not_synchronized_3);
__ bne (CRstatic, not_static);
__ get_mirror_handle (r3);
__ b (init_monitor);
__ bind (not_static);
__ load (r3, Address(Rlocals, 0));
__ bind (init_monitor);
__ store (r3, Address(Rmonitor, BasicObjectLock::obj_offset_in_bytes()));
__ bind (not_synchronized_3);
}
// Used by compiler only; may use only caller saved, non-argument
// registers.
VtableStub* VtableStubs::create_vtable_stub(int vtable_index) {
// PPC port: use fixed size.
const int code_length = VtableStub::pd_code_size_limit(true);
VtableStub* s = new (code_length) VtableStub(true, vtable_index);
// Can be NULL if there is no free space in the code cache.
if (s == NULL) {
return NULL;
}
ResourceMark rm;
CodeBuffer cb(s->entry_point(), code_length);
MacroAssembler* masm = new MacroAssembler(&cb);
#ifndef PRODUCT
if (CountCompiledCalls) {
int offs = __ load_const_optimized(R11_scratch1, SharedRuntime::nof_megamorphic_calls_addr(), R12_scratch2, true);
__ lwz(R12_scratch2, offs, R11_scratch1);
__ addi(R12_scratch2, R12_scratch2, 1);
__ stw(R12_scratch2, offs, R11_scratch1);
}
#endif
assert(VtableStub::receiver_location() == R3_ARG1->as_VMReg(), "receiver expected in R3_ARG1");
// Get receiver klass.
const Register rcvr_klass = R11_scratch1;
// We might implicit NULL fault here.
address npe_addr = __ pc(); // npe = null pointer exception
__ load_klass_with_trap_null_check(rcvr_klass, R3);
// Set method (in case of interpreted method), and destination address.
int entry_offset = InstanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
#ifndef PRODUCT
if (DebugVtables) {
Label L;
// Check offset vs vtable length.
const Register vtable_len = R12_scratch2;
__ lwz(vtable_len, InstanceKlass::vtable_length_offset()*wordSize, rcvr_klass);
__ cmpwi(CCR0, vtable_len, vtable_index*vtableEntry::size());
__ bge(CCR0, L);
__ li(R12_scratch2, vtable_index);
__ call_VM(noreg, CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), R3_ARG1, R12_scratch2, false);
__ bind(L);
}
#endif
int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes();
__ ld(R19_method, v_off, rcvr_klass);
#ifndef PRODUCT
if (DebugVtables) {
Label L;
__ cmpdi(CCR0, R19_method, 0);
__ bne(CCR0, L);
__ stop("Vtable entry is ZERO", 102);
__ bind(L);
}
#endif
// If the vtable entry is null, the method is abstract.
address ame_addr = __ pc(); // ame = abstract method error
__ load_with_trap_null_check(R12_scratch2, in_bytes(Method::from_compiled_offset()), R19_method);
__ mtctr(R12_scratch2);
__ bctr();
masm->flush();
guarantee(__ pc() <= s->code_end(), "overflowed buffer");
s->set_exception_points(npe_addr, ame_addr);
return s;
}
VtableStub* VtableStubs::create_itable_stub(int itable_index) {
// PPC port: use fixed size.
const int code_length = VtableStub::pd_code_size_limit(false);
VtableStub* s = new (code_length) VtableStub(false, itable_index);
// Can be NULL if there is no free space in the code cache.
if (s == NULL) {
return NULL;
}
ResourceMark rm;
CodeBuffer cb(s->entry_point(), code_length);
MacroAssembler* masm = new MacroAssembler(&cb);
address start_pc;
#ifndef PRODUCT
if (CountCompiledCalls) {
int offs = __ load_const_optimized(R11_scratch1, SharedRuntime::nof_megamorphic_calls_addr(), R12_scratch2, true);
__ lwz(R12_scratch2, offs, R11_scratch1);
__ addi(R12_scratch2, R12_scratch2, 1);
__ stw(R12_scratch2, offs, R11_scratch1);
}
#endif
assert(VtableStub::receiver_location() == R3_ARG1->as_VMReg(), "receiver expected in R3_ARG1");
// Entry arguments:
// R19_method: Interface
// R3_ARG1: Receiver
Label L_no_such_interface;
const Register rcvr_klass = R11_scratch1,
interface = R12_scratch2,
tmp1 = R21_tmp1,
tmp2 = R22_tmp2;
address npe_addr = __ pc(); // npe = null pointer exception
__ load_klass_with_trap_null_check(rcvr_klass, R3_ARG1);
// Receiver subtype check against REFC.
__ ld(interface, CompiledICHolder::holder_klass_offset(), R19_method);
__ lookup_interface_method(rcvr_klass, interface, noreg,
R0, tmp1, tmp2,
L_no_such_interface, /*return_method=*/ false);
// Get Method* and entrypoint for compiler
__ ld(interface, CompiledICHolder::holder_metadata_offset(), R19_method);
__ lookup_interface_method(rcvr_klass, interface, itable_index,
R19_method, tmp1, tmp2,
L_no_such_interface, /*return_method=*/ true);
#ifndef PRODUCT
if (DebugVtables) {
Label ok;
__ cmpd(CCR0, R19_method, 0);
__ bne(CCR0, ok);
__ stop("method is null", 103);
__ bind(ok);
}
#endif
// If the vtable entry is null, the method is abstract.
address ame_addr = __ pc(); // ame = abstract method error
// Must do an explicit check if implicit checks are disabled.
assert(!MacroAssembler::needs_explicit_null_check(in_bytes(Method::from_compiled_offset())), "sanity");
if (!ImplicitNullChecks || !os::zero_page_read_protected()) {
if (TrapBasedNullChecks) {
__ trap_null_check(R19_method);
} else {
__ cmpdi(CCR0, R19_method, 0);
__ beq(CCR0, L_no_such_interface);
}
}
__ ld(R12_scratch2, in_bytes(Method::from_compiled_offset()), R19_method);
__ mtctr(R12_scratch2);
__ bctr();
// Handle IncompatibleClassChangeError in itable stubs.
// More detailed error message.
// We force resolving of the call site by jumping to the "handle
// wrong method" stub, and so let the interpreter runtime do all the
// dirty work.
__ bind(L_no_such_interface);
__ load_const_optimized(R11_scratch1, SharedRuntime::get_handle_wrong_method_stub(), R12_scratch2);
__ mtctr(R11_scratch1);
__ bctr();
masm->flush();
guarantee(__ pc() <= s->code_end(), "overflowed buffer");
s->set_exception_points(npe_addr, ame_addr);
return s;
}
// Call an accessor method (assuming it is resolved, otherwise drop into
// vanilla (slow path) entry.
address InterpreterGenerator::generate_accessor_entry(void) {
if (!UseFastAccessorMethods && (!FLAG_IS_ERGO(UseFastAccessorMethods))) {
return NULL;
}
Label Lslow_path, Lacquire;
const Register
Rclass_or_obj = R3_ARG1,
Rconst_method = R4_ARG2,
Rcodes = Rconst_method,
Rcpool_cache = R5_ARG3,
Rscratch = R11_scratch1,
Rjvmti_mode = Rscratch,
Roffset = R12_scratch2,
Rflags = R6_ARG4,
Rbtable = R7_ARG5;
static address branch_table[number_of_states];
address entry = __ pc();
// Check for safepoint:
// Ditch this, real man don't need safepoint checks.
// Also check for JVMTI mode
// Check for null obj, take slow path if so.
__ ld(Rclass_or_obj, Interpreter::stackElementSize, CC_INTERP_ONLY(R17_tos) NOT_CC_INTERP(R15_esp));
__ lwz(Rjvmti_mode, thread_(interp_only_mode));
__ cmpdi(CCR1, Rclass_or_obj, 0);
__ cmpwi(CCR0, Rjvmti_mode, 0);
__ crorc(/*CCR0 eq*/2, /*CCR1 eq*/4+2, /*CCR0 eq*/2);
__ beq(CCR0, Lslow_path); // this==null or jvmti_mode!=0
// Do 2 things in parallel:
// 1. Load the index out of the first instruction word, which looks like this:
// <0x2a><0xb4><index (2 byte, native endianess)>.
// 2. Load constant pool cache base.
__ ld(Rconst_method, in_bytes(Method::const_offset()), R19_method);
__ ld(Rcpool_cache, in_bytes(ConstMethod::constants_offset()), Rconst_method);
__ lhz(Rcodes, in_bytes(ConstMethod::codes_offset()) + 2, Rconst_method); // Lower half of 32 bit field.
__ ld(Rcpool_cache, ConstantPool::cache_offset_in_bytes(), Rcpool_cache);
// Get the const pool entry by means of <index>.
const int codes_shift = exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord);
__ slwi(Rscratch, Rcodes, codes_shift); // (codes&0xFFFF)<<codes_shift
__ add(Rcpool_cache, Rscratch, Rcpool_cache);
// Check if cpool cache entry is resolved.
// We are resolved if the indices offset contains the current bytecode.
ByteSize cp_base_offset = ConstantPoolCache::base_offset();
// Big Endian:
__ lbz(Rscratch, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::indices_offset()) + 7 - 2, Rcpool_cache);
__ cmpwi(CCR0, Rscratch, Bytecodes::_getfield);
__ bne(CCR0, Lslow_path);
__ isync(); // Order succeeding loads wrt. load of _indices field from cpool_cache.
// Finally, start loading the value: Get cp cache entry into regs.
__ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcpool_cache);
__ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcpool_cache);
// Following code is from templateTable::getfield_or_static
// Load pointer to branch table
__ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
// Get volatile flag
__ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // extract volatile bit
// note: sync is needed before volatile load on PPC64
// Check field type
__ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
#ifdef ASSERT
Label LFlagInvalid;
__ cmpldi(CCR0, Rflags, number_of_states);
__ bge(CCR0, LFlagInvalid);
__ ld(R9_ARG7, 0, R1_SP);
__ ld(R10_ARG8, 0, R21_sender_SP);
__ cmpd(CCR0, R9_ARG7, R10_ARG8);
__ asm_assert_eq("backlink", 0x543);
#endif // ASSERT
__ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
// Load from branch table and dispatch (volatile case: one instruction ahead)
__ sldi(Rflags, Rflags, LogBytesPerWord);
__ cmpwi(CCR6, Rscratch, 1); // volatile?
if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
__ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // volatile ? size of 1 instruction : 0
}
__ ldx(Rbtable, Rbtable, Rflags);
if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
__ subf(Rbtable, Rscratch, Rbtable); // point to volatile/non-volatile entry point
}
__ mtctr(Rbtable);
__ bctr();
#ifdef ASSERT
__ bind(LFlagInvalid);
__ stop("got invalid flag", 0x6541);
bool all_uninitialized = true,
all_initialized = true;
for (int i = 0; i<number_of_states; ++i) {
all_uninitialized = all_uninitialized && (branch_table[i] == NULL);
all_initialized = all_initialized && (branch_table[i] != NULL);
}
assert(all_uninitialized != all_initialized, "consistency"); // either or
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
if (branch_table[vtos] == 0) branch_table[vtos] = __ pc(); // non-volatile_entry point
if (branch_table[dtos] == 0) branch_table[dtos] = __ pc(); // non-volatile_entry point
if (branch_table[ftos] == 0) branch_table[ftos] = __ pc(); // non-volatile_entry point
__ stop("unexpected type", 0x6551);
#endif
if (branch_table[itos] == 0) { // generate only once
__ align(32, 28, 28); // align load
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
branch_table[itos] = __ pc(); // non-volatile_entry point
__ lwax(R3_RET, Rclass_or_obj, Roffset);
__ beq(CCR6, Lacquire);
__ blr();
}
if (branch_table[ltos] == 0) { // generate only once
__ align(32, 28, 28); // align load
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
branch_table[ltos] = __ pc(); // non-volatile_entry point
__ ldx(R3_RET, Rclass_or_obj, Roffset);
__ beq(CCR6, Lacquire);
__ blr();
}
if (branch_table[btos] == 0) { // generate only once
__ align(32, 28, 28); // align load
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
branch_table[btos] = __ pc(); // non-volatile_entry point
__ lbzx(R3_RET, Rclass_or_obj, Roffset);
__ extsb(R3_RET, R3_RET);
__ beq(CCR6, Lacquire);
__ blr();
}
if (branch_table[ctos] == 0) { // generate only once
__ align(32, 28, 28); // align load
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
branch_table[ctos] = __ pc(); // non-volatile_entry point
__ lhzx(R3_RET, Rclass_or_obj, Roffset);
__ beq(CCR6, Lacquire);
__ blr();
}
if (branch_table[stos] == 0) { // generate only once
__ align(32, 28, 28); // align load
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
branch_table[stos] = __ pc(); // non-volatile_entry point
__ lhax(R3_RET, Rclass_or_obj, Roffset);
__ beq(CCR6, Lacquire);
__ blr();
}
if (branch_table[atos] == 0) { // generate only once
__ align(32, 28, 28); // align load
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
branch_table[atos] = __ pc(); // non-volatile_entry point
__ load_heap_oop(R3_RET, (RegisterOrConstant)Roffset, Rclass_or_obj);
__ verify_oop(R3_RET);
//__ dcbt(R3_RET); // prefetch
__ beq(CCR6, Lacquire);
__ blr();
}
__ align(32, 12);
__ bind(Lacquire);
__ twi_0(R3_RET);
__ isync(); // acquire
__ blr();
#ifdef ASSERT
for (int i = 0; i<number_of_states; ++i) {
assert(branch_table[i], "accessor_entry initialization");
//tty->print_cr("accessor_entry: branch_table[%d] = 0x%llx (opcode 0x%llx)", i, branch_table[i], *((unsigned int*)branch_table[i]));
}
#endif
__ bind(Lslow_path);
__ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), Rscratch);
__ flush();
return entry;
}
