int MethodComparator::check_stack_and_locals_size(methodOop old_method, methodOop new_method) {
if (old_method->max_stack() != new_method->max_stack()) {
return 1;
} else if (old_method->max_locals() != new_method->max_locals()) {
return 2;
} else if (old_method->size_of_parameters() != new_method->size_of_parameters()) {
return 3;
} else return 0;
}
// for hashing into the table
static int hash(methodOop method) {
// The point here is to try to make something fairly unique
// out of the fields we can read without grabbing any locks
// since the method may be locked when we need the hash.
return (
method->code_size() ^
method->max_stack() ^
method->max_locals() ^
method->size_of_parameters());
}
int AbstractInterpreter::layout_activation(methodOop method,
int tempcount,
int popframe_extra_args,
int moncount,
int caller_actual_parameters,
int callee_param_count,
int callee_locals,
frame* caller,
frame* interpreter_frame,
bool is_top_frame,
bool is_bottom_frame) {
assert(popframe_extra_args == 0, "what to do?");
assert(!is_top_frame || (!callee_locals && !callee_param_count),
"top frame should have no caller");
// This code must exactly match what InterpreterFrame::build
// does (the full InterpreterFrame::build, that is, not the
// one that creates empty frames for the deoptimizer).
//
// If interpreter_frame is not NULL then it will be filled in.
// It's size is determined by a previous call to this method,
// so it should be correct.
//
// Note that tempcount is the current size of the expression
// stack. For top most frames we will allocate a full sized
// expression stack and not the trimmed version that non-top
// frames have.
int header_words = InterpreterFrame::header_words;
int monitor_words = moncount * frame::interpreter_frame_monitor_size();
int stack_words = is_top_frame ? method->max_stack() : tempcount;
int callee_extra_locals = callee_locals - callee_param_count;
if (interpreter_frame) {
intptr_t *locals = interpreter_frame->fp() + method->max_locals();
interpreterState istate = interpreter_frame->get_interpreterState();
intptr_t *monitor_base = (intptr_t*) istate;
intptr_t *stack_base = monitor_base - monitor_words;
intptr_t *stack = stack_base - tempcount - 1;
BytecodeInterpreter::layout_interpreterState(istate,
caller,
NULL,
method,
locals,
stack,
stack_base,
monitor_base,
NULL,
is_top_frame);
}
return header_words + monitor_words + stack_words + callee_extra_locals;
}
int AbstractInterpreter::size_top_interpreter_activation(methodOop method)
{
#ifdef PPC
StackFrame frame;
int call_stub_frame = round_to(
StubRoutines::call_stub_base_size() +
method->max_locals() * wordSize, StackAlignmentInBytes);
int interpreter_frame = round_to(
frame.unaligned_size() +
slop_factor +
method->max_stack() * wordSize +
(method->is_synchronized() ?
frame::interpreter_frame_monitor_size() * wordSize : 0) +
sizeof(BytecodeInterpreter), StackAlignmentInBytes);
return (call_stub_frame + interpreter_frame) / wordSize;
#else
Unimplemented();
#endif // PPC
}
void vframe::print_xml_on(JavaThread *thread, xmlBuffer *xb) const {
ResourceMark rm;
frame fr = get_frame();
const char *frame_style = "UNKNOWN";
CodeBlob*cb=CodeCache::find_blob(fr.pc());
if( fr.is_interpreted_frame() )
frame_style = "INTERPRETER";
else if( cb && cb->is_native_method() )
frame_style = "NATIVE";
else if( cb && cb->is_c1_method() )
frame_style = "COMPILER1";
else if( cb && cb->is_c2_method() )
frame_style = "COMPILER2";
// Frame header
xb->name_value_item("frame_style",frame_style);
// Java Locals
if (!fr.is_java_frame())
return; // no more printing to be done
const methodOop moop = method();
if (moop->is_native())
return; // no more printing to be done
const int max_locals = moop->max_locals();
if(fr.is_compiled_frame()){
for( int i=0; i<max_locals; i++ ) {
xmlElement je(xb, "java_element");
{ xmlElement n(xb, "name", xmlElement::delayed_LF);
bool out_of_scope = true;
const char* name = moop->localvariable_name(i, bci(), out_of_scope);
if( name ) xb->print("%s%s%s", out_of_scope ? "{": "", name, out_of_scope ? "}": "");
else xb->print("JL%-2d", i);
}
DebugScopeValue::Name vreg = _ds->get_local(i);
if( !DebugScopeValue::is_valid(vreg) ) {
xb->name_value_item("type","invalid");
xb->name_ptr_item("value",0);
} else {
intptr_t *data_ptr = fr.reg_to_addr(DebugScopeValue::to_vreg(vreg));
const int rel_pc = cb->rel_pc(fr.pc());
const bool isoop = cb->oop_maps()->is_oop( rel_pc, VReg::as_VOopReg(DebugScopeValue::to_vreg(vreg)) );
if( isoop ) {
xb->name_value_item("type", "oop");
oop o = ((objectRef*)data_ptr)->as_oop();
o->print_xml_on(xb, true);
} else {
xb->name_value_item("type", "normal");
xb->name_ptr_item("value", (void*)*data_ptr);
}
}
}
}else if(fr.is_interpreted_frame()){
for( int i = 0; i < max_locals; i++ ) {
frame::InterpreterStackSlotType tag = fr.tag_at_address(fr.interpreter_frame_local_addr(i));
if (tag == frame::double_slot_primitive_type)
i++;//skip the tag slot
xmlElement je(xb, "java_element");
{ xmlElement n(xb, "name", xmlElement::delayed_LF);
bool out_of_scope = true;
const char* name = moop->localvariable_name(i, bci(), out_of_scope);
if (name != NULL) {
xb->print("%s%s%s", out_of_scope ? "{": "", name, out_of_scope ? "}": "");
} else {
xb->print("JL%-2d", i);
}
}
intptr_t local_value = fr.interpreter_frame_local_at(i);
switch (tag) {
case frame::single_slot_primitive_type: {
xb->name_value_item("type", "single_slot_primitive_type");
xb->name_ptr_item("value", (void*)local_value);
break;
}
case frame::double_slot_primitive_type: {
xb->name_value_item("type", "double_slot_primitive_type");
xb->name_ptr_item("value", (void*)local_value);
break;
}
case frame::single_slot_ref_type: {
xb->name_value_item("type", "oop");
oop o = objectRef((uint64_t)local_value).as_oop();
o->print_xml_on(xb, true);
break;
}
default: ShouldNotReachHere(); break;
}
}
}
}
// --- build_repack_buffer ---------------------------------------------------
// Build a IFrame structure to help ASM code repack the 1 compiled frame into
// many interpreter (or C1) frames. Takes in the current thread and a vframe;
// the vframe is pointing and the virtual Java frame needing to be repacked.
// It takes in the callee (which this frame is busy trying to call in it's
// inlined code), and an array of IFrames. It returns the updated IFrame
// buffer filled in for this frame.
void Deoptimization::build_repack_buffer( JavaThread *thread, frame fr, IFrame *buf, const DebugMap *dm, const DebugScope *ds, intptr_t *jexstk, objectRef *lckstk, bool is_deopt, bool is_c1, bool is_youngest) {
assert( thread->_deopt_buffer->contains((char*)(buf+1)), "over-ran large deopt buffer?" );
int bci=ds->bci();
if(bci==InvocationEntryBci){
// We deoptimized while hanging in prologue code for a synchronized
// method. We got the lock (after all, deopt happens after returning
// from the blocking call). We want to begin execution in the
// interpreter at BCI 0, and after taking the lock.
// Also it is possilble to enter the deopt code through the br_s on method
// entry before the first byte code.
bci = 0;
}
const methodOop moop = ds->method().as_methodOop();
if( ds->caller() ) { // Do I have a caller? Am I mid-call?
// Initialize the constant pool entry for caller-parameter size. It
// might be the case that we inlined and compiled a callee, and are busy
// calling it in the compiled code, and get deoptimized with that callee
// in-progress AND we've never executed it in the interpreter - which
// would have filled in the constant pool cache before making the call.
// Fill it in now.
const methodOop caller = ds->caller()->method().as_methodOop();
int index = Bytes::get_native_u2(caller->bcp_from(ds->caller()->bci())+1);
ConstantPoolCacheEntry *cpe = caller->constants()->cache()->entry_at(index);
// Since we are setting the constant pool entry here, and another thread
// could be busy resolving here we have a race condition setting the
// flags. Use a CAS to only set the flags if they are currently 0.
intx *flags_adr = (intx*)((intptr_t)cpe + in_bytes(ConstantPoolCacheEntry::flags_offset()));
if( !*flags_adr ) { // Flags currently 0?
// Set the flags, because the interpreter-return-entry points need some
// info from them. Not all fields are set, because it's too complex to
// do it here... and not needed. The cpCacheEntry is left "unresolved"
// such that the next real use of it from the interpreter will be forced
// to do a proper resolve, which will fill in the missing fields.
// Compute new flags needed by the interpreter-return-entry
intx flags =
(moop->size_of_parameters() & 0xFF) |
(1 << ConstantPoolCacheEntry::hotSwapBit) |
(moop->result_type() << ConstantPoolCacheEntry::tosBits);
// CAS 'em in, but only if there is currently a 0 flags
assert0( sizeof(jlong)==sizeof(intx) );
Atomic::cmpxchg((jlong)flags, (jlong*)flags_adr, 0);
// We don't care about the result, because the cache is monomorphic.
// Either our CAS succeeded and jammed the right parameter count, or
// another thread succeeded and jammed in the right parameter count.
}
}
if (TraceDeoptimization) {
BufferedLoggerMark m(NOTAG, Log::M_DEOPT, TraceDeoptimization, true);
m.out("DEOPT REPACK c%d: ", is_c1 ? 1 : 2);
moop->print_short_name(m.stream());
m.out(" @ bci %d %s", bci, ds->caller() ? "called by...": " (oldest frame)" );
}
// If there was a suitable C1 frame, use it.
// Otherwise, use an interpreter frame.
if( 1 ) {
// Build an interpreter-style IFrame. Naked oops abound.
assert0( !objectRef(moop).is_stack() );
buf->_mref = objectRef(moop);
buf->_cpc = moop->constants()->cacheRef();
// Compute monitor list length. If we have coarsened a lock we will end
// up unlocking it and the repack buffer will not need to see it.
uint mons_len = ds->numlocks();
if( ds->is_extra_lock() ) { mons_len--; assert0( mons_len >= 0 ); }
assert0( mons_len < (256*sizeof(buf->_numlck)) );
buf->_numlck = mons_len;
// Set up the return pc for the next frame: the next frame is a younger
// frame which will return to this older frame. All middle frames return
// back into the interpreter, just after a call with proper TOS state.
// Youngest frames always start in vtos state because the uncommon-trap
// blob sets them up that way.
const address bcp = moop->bcp_from(bci);
Bytecodes::Code c = Bytecodes::java_code(Bytecodes::cast(*bcp));
BasicType return_type=T_VOID;
bool handle_popframe = is_youngest && JvmtiExport::can_pop_frame() && thread->popframe_forcing_deopt_reexecution();
int bci_bump = 0;
if( !is_youngest ) { // Middle-frame?
bool from_call = (c == Bytecodes::_invokevirtual ||
c==Bytecodes::_invokespecial||
c==Bytecodes::_invokestatic||
c == Bytecodes::_invokeinterface );
assert(from_call,"Middle frame is in the middle of a call");
bci_bump = Bytecodes::length_at(bcp); // But need to know how much it will be bumped for the return address
buf->_bci = bci; // Save bci without bumping it; normal interpreter call returns bump the bci as needed
buf[-1]._retadr = Interpreter::return_entry(vtos, bci_bump);
} else if( thread->pending_exception() ) {
// Deopt-with-pending. Throw up on return to interpreter, which is
// handled by unpack_and_go.
buf->_bci=bci;
buf[-1]._retadr = Interpreter::unpack_and_go();
} else if( !is_deopt ) { // It is a C2-style uncommon-trap.
// Do NOT increment the BCP! We are re-executing the current bytecode.
buf->_bci=bci;
buf[-1]._retadr = Interpreter::unpack_and_go();
} else { // It is a plain deopt
// It is a deopt without exception. See if we are C1 in mid-patch.
// If so, we always need to re-execute the bytecode.
bool is_C1_mid_patch = false;
if( is_c1 ) { // C1 codeblob?
address caller_pc=fr.pc();
if(NativeCall::is_call_before(caller_pc)){
address target = nativeCall_at(caller_pc)->destination();
is_C1_mid_patch = target == Runtime1::entry_for(Runtime1::load_klass_patching_id);
}
}
if( is_C1_mid_patch ) {
Untested("");
// Do NOT increment the BCP! We are re-executing the current bytecode.
} else if( ds->bci() == InvocationEntryBci ) {
// It is deopt while hanging on a method-entry lock.
// Do not advance BCP, as we have not executed bci 0 yet.
} else { // Else C2 or C1-not-mid-patch
// It is a deopt. Whether we re-execute the current bytecode or
// assume it has completed depends on the bytecode.
switch( c ) {
case Bytecodes::_lookupswitch:
case Bytecodes::_tableswitch:
case Bytecodes::_fast_binaryswitch:
case Bytecodes::_fast_linearswitch:
// recompute condtional expression folded into _if<cond>
case Bytecodes::_lcmp :
case Bytecodes::_fcmpl :
case Bytecodes::_fcmpg :
case Bytecodes::_dcmpl :
case Bytecodes::_dcmpg :
case Bytecodes::_ifnull :
case Bytecodes::_ifnonnull :
case Bytecodes::_goto :
case Bytecodes::_goto_w :
case Bytecodes::_ifeq :
case Bytecodes::_ifne :
case Bytecodes::_iflt :
case Bytecodes::_ifge :
case Bytecodes::_ifgt :
case Bytecodes::_ifle :
case Bytecodes::_if_icmpeq :
case Bytecodes::_if_icmpne :
case Bytecodes::_if_icmplt :
case Bytecodes::_if_icmpge :
case Bytecodes::_if_icmpgt :
case Bytecodes::_if_icmple :
case Bytecodes::_if_acmpeq :
case Bytecodes::_if_acmpne :
// special cases
case Bytecodes::_aastore:
// We are re-executing the current bytecode.
Untested("");
break;
// special cases
case Bytecodes::_putstatic:
case Bytecodes::_getstatic:
case Bytecodes::_getfield:
case Bytecodes::_putfield:
// We are re-executing the current bytecode.
break;
case Bytecodes::_athrow :
break; // Must be deopt-w-exception
case Bytecodes::_invokevirtual:
case Bytecodes::_invokespecial:
case Bytecodes::_invokestatic:{
methodHandle mh(thread,moop);
return_type=Bytecode_invoke_at(mh,bci)->result_type(thread);
if( !handle_popframe &&
!ds->should_reexecute())
bci_bump = 3; // Increment the BCP to post-call!!! See below!
break;
}
case Bytecodes::_invokeinterface:{
methodHandle mh(thread,moop);
return_type=Bytecode_invoke_at(mh,bci)->result_type(thread);
if( !handle_popframe &&
!ds->should_reexecute())
bci_bump = 5; // Increment the BCP to post-call!!! See below!
break;
}
case Bytecodes::_ldc :
Untested("");
return_type=constant_pool_type(moop,*(bcp+1));
if( !ds->should_reexecute()) bci_bump = 2; // Increment the BCP to post-call!!! See below!
break;
case Bytecodes::_ldc_w : // fall through
case Bytecodes::_ldc2_w:
return_type=constant_pool_type(moop,Bytes::get_Java_u2(bcp+1));
if( !ds->should_reexecute()) bci_bump = 3; // Increment the BCP to post-call!!! See below!
break;
default:
return_type=Bytecodes::result_type(c);
if( !ds->should_reexecute()) bci_bump = Bytecodes::length_at(bcp); // Increment the BCP to post-call!!! See below!
break;
}
if (ds->should_reexecute()) return_type = T_VOID;
}
// Save (possibly advanced) bci
buf->_bci = bci+bci_bump;
buf[-1]._retadr = Interpreter::unpack_and_go(); // Interpreter::return_entry(vtos, bci_bump);
}
// ---
// Now all the Java locals.
// First set the start of locals for the interpreter frame we are building.
buf->_loc = (intptr_t)jexstk;
uint loc_len = moop->max_locals();
for(uint i=0;i<loc_len;i++){
*jexstk++ = dm->get_value(ds->get_local(i),fr);
}
// Now that the locals have been unpacked if we have any deferred local writes
// added by jvmti then we can free up that structure as the data is now in the
// buffer
GrowableArray<jvmtiDeferredLocalVariableSet*>* list = thread->deferred_locals();
if( list ) {
// Because of inlining we could have multiple vframes for a single frame
// and several of the vframes could have deferred writes. Find them all.
Unimplemented();
}
// ---
// Now all the Java Expressions
uint expr_len = ds->numstk();
for(uint i=0;i<expr_len;i++)
*jexstk++ = dm->get_value(ds->get_expr(i),fr);
// If returning from a deoptimized call, we will have return values in
// registers that need to end up on the Java execution stack. They are
// not recorded in the debug info, since they did not exist at the time
// the call began.
if( is_youngest && is_deopt ) {
if( type2size[return_type] > 0 ) {
if( type2size[return_type]==2 ) {
*jexstk++ = (intptr_t)frame::double_slot_primitive_type_empty_slot_id << 32;
}
*jexstk++ = pd_fetch_return_values( thread, return_type );
// Need to adjust the final jexstk_top for the youngest frame
// returning values. These returned values are not accounted for in
// the standard debug info.
thread->_jexstk_top = jexstk;
}
}
// JVMTI PopFrame support
// Add the number of words of popframe preserved args to expr_len
int popframe_preserved_args_size_in_bytes = in_bytes(thread->popframe_preserved_args_size());
int popframe_preserved_args_size_in_words = in_words(thread->popframe_preserved_args_size_in_words());
if (handle_popframe) {
Unimplemented();
expr_len += popframe_preserved_args_size_in_words;
// An interpreted frame was popped but it returns to a deoptimized
// frame. The incoming arguments to the interpreted activation
// were preserved in thread-local storage by the
// remove_activation_preserving_args_entry in the interpreter; now
// we put them back into the just-unpacked interpreter frame.
// Note that this assumes that the locals arena grows toward lower
// addresses.
}
// Set the JEX stk top
buf->_stk = (intptr_t)jexstk;
// ---
// Now move locked objects to the interpreters lock-stack.
// No need to inflate anything, as we're moving standard oops.
int numlcks = ds->numlocks();
if( ds->is_extra_lock() ) { // coarsened a lock
Untested("");
// The last lock is "coarsened" - kept locked when it should have been
// unlocked and relocked. With no deopt, keeping it locked saves the 2
// sets of back-to-back CAS's and fences. However, here we need to
// unlock it to match the proper Java state.
ObjectSynchronizer::unlock(ALWAYS_POISON_OBJECTREF((objectRef)dm->get_value(ds->get_lock(numlcks-1),fr)).as_oop());
numlcks--;
}
for(int i=0;i<numlcks;i++){
*lckstk++ = ALWAYS_POISON_OBJECTREF((objectRef)dm->get_value(ds->get_lock(i),fr));
}
} else { // Make a C1 frame
Unimplemented();
}
}