1 /*
2 * Copyright (c) 1997, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "interp_masm_x86.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "logging/log.hpp"
30 #include "oops/arrayOop.hpp"
31 #include "oops/markOop.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/method.hpp"
34 #include "prims/jvmtiExport.hpp"
35 #include "prims/jvmtiThreadState.hpp"
36 #include "runtime/basicLock.hpp"
37 #include "runtime/biasedLocking.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/thread.inline.hpp"
40
41 // Implementation of InterpreterMacroAssembler
42
43 void InterpreterMacroAssembler::jump_to_entry(address entry) {
44 assert(entry, "Entry must have been generated by now");
45 jump(RuntimeAddress(entry));
46 }
47
48 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
49 Label update, next, none;
50
51 verify_oop(obj);
52
53 testptr(obj, obj);
54 jccb(Assembler::notZero, update);
55 orptr(mdo_addr, TypeEntries::null_seen);
56 jmpb(next);
57
58 bind(update);
59 load_klass(obj, obj);
60
61 xorptr(obj, mdo_addr);
62 testptr(obj, TypeEntries::type_klass_mask);
63 jccb(Assembler::zero, next); // klass seen before, nothing to
64 // do. The unknown bit may have been
65 // set already but no need to check.
66
67 testptr(obj, TypeEntries::type_unknown);
68 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
69
70 cmpptr(mdo_addr, 0);
71 jccb(Assembler::equal, none);
72 cmpptr(mdo_addr, TypeEntries::null_seen);
73 jccb(Assembler::equal, none);
74 // There is a chance that the checks above (re-reading profiling
75 // data from memory) fail if another thread has just set the
76 // profiling to this obj's klass
77 xorptr(obj, mdo_addr);
78 testptr(obj, TypeEntries::type_klass_mask);
79 jccb(Assembler::zero, next);
80
81 // different than before. Cannot keep accurate profile.
82 orptr(mdo_addr, TypeEntries::type_unknown);
83 jmpb(next);
84
85 bind(none);
86 // first time here. Set profile type.
87 movptr(mdo_addr, obj);
88
89 bind(next);
90 }
91
92 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
93 if (!ProfileInterpreter) {
94 return;
95 }
96
97 if (MethodData::profile_arguments() || MethodData::profile_return()) {
98 Label profile_continue;
99
100 test_method_data_pointer(mdp, profile_continue);
101
102 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
103
104 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
105 jcc(Assembler::notEqual, profile_continue);
106
107 if (MethodData::profile_arguments()) {
108 Label done;
109 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
110 addptr(mdp, off_to_args);
111
112 for (int i = 0; i < TypeProfileArgsLimit; i++) {
113 if (i > 0 || MethodData::profile_return()) {
114 // If return value type is profiled we may have no argument to profile
115 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
116 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
117 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
118 jcc(Assembler::less, done);
119 }
120 movptr(tmp, Address(callee, Method::const_offset()));
121 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
122 // stack offset o (zero based) from the start of the argument
123 // list, for n arguments translates into offset n - o - 1 from
124 // the end of the argument list
125 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
126 subl(tmp, 1);
127 Address arg_addr = argument_address(tmp);
128 movptr(tmp, arg_addr);
129
130 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
131 profile_obj_type(tmp, mdo_arg_addr);
132
133 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
134 addptr(mdp, to_add);
135 off_to_args += to_add;
136 }
137
138 if (MethodData::profile_return()) {
139 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
140 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
141 }
142
143 bind(done);
144
145 if (MethodData::profile_return()) {
146 // We're right after the type profile for the last
147 // argument. tmp is the number of cells left in the
148 // CallTypeData/VirtualCallTypeData to reach its end. Non null
149 // if there's a return to profile.
150 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
151 shll(tmp, exact_log2(DataLayout::cell_size));
152 addptr(mdp, tmp);
153 }
154 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
155 } else {
156 assert(MethodData::profile_return(), "either profile call args or call ret");
157 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
158 }
159
160 // mdp points right after the end of the
161 // CallTypeData/VirtualCallTypeData, right after the cells for the
162 // return value type if there's one
163
164 bind(profile_continue);
165 }
166 }
167
168 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
169 assert_different_registers(mdp, ret, tmp, _bcp_register);
170 if (ProfileInterpreter && MethodData::profile_return()) {
171 Label profile_continue, done;
172
173 test_method_data_pointer(mdp, profile_continue);
174
175 if (MethodData::profile_return_jsr292_only()) {
176 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
177
178 // If we don't profile all invoke bytecodes we must make sure
179 // it's a bytecode we indeed profile. We can't go back to the
180 // begining of the ProfileData we intend to update to check its
181 // type because we're right after it and we don't known its
182 // length
183 Label do_profile;
184 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
185 jcc(Assembler::equal, do_profile);
186 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
187 jcc(Assembler::equal, do_profile);
188 get_method(tmp);
189 cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm);
190 jcc(Assembler::notEqual, profile_continue);
191
192 bind(do_profile);
193 }
194
195 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
196 mov(tmp, ret);
197 profile_obj_type(tmp, mdo_ret_addr);
198
199 bind(profile_continue);
200 }
201 }
202
203 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
204 if (ProfileInterpreter && MethodData::profile_parameters()) {
205 Label profile_continue, done;
206
207 test_method_data_pointer(mdp, profile_continue);
208
209 // Load the offset of the area within the MDO used for
210 // parameters. If it's negative we're not profiling any parameters
211 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
212 testl(tmp1, tmp1);
213 jcc(Assembler::negative, profile_continue);
214
215 // Compute a pointer to the area for parameters from the offset
216 // and move the pointer to the slot for the last
217 // parameters. Collect profiling from last parameter down.
218 // mdo start + parameters offset + array length - 1
219 addptr(mdp, tmp1);
220 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
221 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
222
223 Label loop;
224 bind(loop);
225
226 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
227 int type_base = in_bytes(ParametersTypeData::type_offset(0));
228 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
229 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
230 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
231
232 // load offset on the stack from the slot for this parameter
233 movptr(tmp2, arg_off);
234 negptr(tmp2);
235 // read the parameter from the local area
236 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
237
238 // profile the parameter
239 profile_obj_type(tmp2, arg_type);
240
241 // go to next parameter
242 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
243 jcc(Assembler::positive, loop);
244
245 bind(profile_continue);
246 }
247 }
248
249 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
250 int number_of_arguments) {
251 // interpreter specific
252 //
253 // Note: No need to save/restore bcp & locals registers
254 // since these are callee saved registers and no blocking/
255 // GC can happen in leaf calls.
256 // Further Note: DO NOT save/restore bcp/locals. If a caller has
257 // already saved them so that it can use rsi/rdi as temporaries
258 // then a save/restore here will DESTROY the copy the caller
259 // saved! There used to be a save_bcp() that only happened in
260 // the ASSERT path (no restore_bcp). Which caused bizarre failures
261 // when jvm built with ASSERTs.
262 #ifdef ASSERT
263 {
264 Label L;
265 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
266 jcc(Assembler::equal, L);
267 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
268 " last_sp != NULL");
269 bind(L);
270 }
271 #endif
272 // super call
273 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
274 // interpreter specific
275 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
276 // but since they may not have been saved (and we don't want to
277 // save them here (see note above) the assert is invalid.
278 }
279
280 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
281 Register java_thread,
282 Register last_java_sp,
283 address entry_point,
284 int number_of_arguments,
285 bool check_exceptions) {
286 // interpreter specific
287 //
288 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
289 // really make a difference for these runtime calls, since they are
290 // slow anyway. Btw., bcp must be saved/restored since it may change
291 // due to GC.
292 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
293 save_bcp();
294 #ifdef ASSERT
295 {
296 Label L;
297 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
298 jcc(Assembler::equal, L);
299 stop("InterpreterMacroAssembler::call_VM_base:"
300 " last_sp != NULL");
301 bind(L);
302 }
303 #endif /* ASSERT */
304 // super call
305 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
306 entry_point, number_of_arguments,
307 check_exceptions);
308 // interpreter specific
309 restore_bcp();
310 restore_locals();
311 }
312
313 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
314 if (JvmtiExport::can_pop_frame()) {
315 Label L;
316 // Initiate popframe handling only if it is not already being
317 // processed. If the flag has the popframe_processing bit set, it
318 // means that this code is called *during* popframe handling - we
319 // don't want to reenter.
320 // This method is only called just after the call into the vm in
321 // call_VM_base, so the arg registers are available.
322 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
323 LP64_ONLY(c_rarg0);
324 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
325 testl(pop_cond, JavaThread::popframe_pending_bit);
326 jcc(Assembler::zero, L);
327 testl(pop_cond, JavaThread::popframe_processing_bit);
328 jcc(Assembler::notZero, L);
329 // Call Interpreter::remove_activation_preserving_args_entry() to get the
330 // address of the same-named entrypoint in the generated interpreter code.
331 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
332 jmp(rax);
333 bind(L);
334 NOT_LP64(get_thread(java_thread);)
335 }
336 }
337
338 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
339 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
340 NOT_LP64(get_thread(thread);)
341 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
342 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
343 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
344 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
345 #ifdef _LP64
346 switch (state) {
347 case atos: movptr(rax, oop_addr);
348 movptr(oop_addr, (int32_t)NULL_WORD);
349 verify_oop(rax, state); break;
350 case ltos: movptr(rax, val_addr); break;
351 case btos: // fall through
352 case ztos: // fall through
353 case ctos: // fall through
354 case stos: // fall through
355 case itos: movl(rax, val_addr); break;
356 case ftos: load_float(val_addr); break;
357 case dtos: load_double(val_addr); break;
358 case vtos: /* nothing to do */ break;
359 default : ShouldNotReachHere();
360 }
361 // Clean up tos value in the thread object
362 movl(tos_addr, (int) ilgl);
363 movl(val_addr, (int32_t) NULL_WORD);
364 #else
365 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
366 + in_ByteSize(wordSize));
367 switch (state) {
368 case atos: movptr(rax, oop_addr);
369 movptr(oop_addr, NULL_WORD);
370 verify_oop(rax, state); break;
371 case ltos:
372 movl(rdx, val_addr1); // fall through
373 case btos: // fall through
374 case ztos: // fall through
375 case ctos: // fall through
376 case stos: // fall through
377 case itos: movl(rax, val_addr); break;
378 case ftos: load_float(val_addr); break;
379 case dtos: load_double(val_addr); break;
380 case vtos: /* nothing to do */ break;
381 default : ShouldNotReachHere();
382 }
383 #endif // _LP64
384 // Clean up tos value in the thread object
385 movl(tos_addr, (int32_t) ilgl);
386 movptr(val_addr, NULL_WORD);
387 NOT_LP64(movptr(val_addr1, NULL_WORD);)
388 }
389
390
391 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
392 if (JvmtiExport::can_force_early_return()) {
393 Label L;
394 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
395 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
396
397 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
398 testptr(tmp, tmp);
399 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
400
401 // Initiate earlyret handling only if it is not already being processed.
402 // If the flag has the earlyret_processing bit set, it means that this code
403 // is called *during* earlyret handling - we don't want to reenter.
404 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
405 cmpl(tmp, JvmtiThreadState::earlyret_pending);
406 jcc(Assembler::notEqual, L);
407
408 // Call Interpreter::remove_activation_early_entry() to get the address of the
409 // same-named entrypoint in the generated interpreter code.
410 NOT_LP64(get_thread(java_thread);)
411 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
412 #ifdef _LP64
413 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
414 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
415 #else
416 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
417 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
418 #endif // _LP64
419 jmp(rax);
420 bind(L);
421 NOT_LP64(get_thread(java_thread);)
422 }
423 }
424
425 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
426 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
427 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
428 bswapl(reg);
429 shrl(reg, 16);
430 }
431
432 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
433 int bcp_offset,
434 size_t index_size) {
435 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
436 if (index_size == sizeof(u2)) {
437 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
438 } else if (index_size == sizeof(u4)) {
439 movl(index, Address(_bcp_register, bcp_offset));
440 // Check if the secondary index definition is still ~x, otherwise
441 // we have to change the following assembler code to calculate the
442 // plain index.
443 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
444 notl(index); // convert to plain index
445 } else if (index_size == sizeof(u1)) {
446 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
447 } else {
448 ShouldNotReachHere();
449 }
450 }
451
452 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
453 Register index,
454 int bcp_offset,
455 size_t index_size) {
456 assert_different_registers(cache, index);
457 get_cache_index_at_bcp(index, bcp_offset, index_size);
458 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
459 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
460 // convert from field index to ConstantPoolCacheEntry index
461 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
462 shll(index, 2);
463 }
464
465 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
466 Register index,
467 Register bytecode,
468 int byte_no,
469 int bcp_offset,
470 size_t index_size) {
471 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
472 // We use a 32-bit load here since the layout of 64-bit words on
473 // little-endian machines allow us that.
474 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
475 const int shift_count = (1 + byte_no) * BitsPerByte;
476 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
477 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
478 "correct shift count");
479 shrl(bytecode, shift_count);
480 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
481 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
482 }
483
484 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
485 Register tmp,
486 int bcp_offset,
487 size_t index_size) {
488 assert(cache != tmp, "must use different register");
489 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
490 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
491 // convert from field index to ConstantPoolCacheEntry index
492 // and from word offset to byte offset
493 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
494 shll(tmp, 2 + LogBytesPerWord);
495 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
496 // skip past the header
497 addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
498 addptr(cache, tmp); // construct pointer to cache entry
499 }
500
501 // Load object from cpool->resolved_references(index)
502 void InterpreterMacroAssembler::load_resolved_reference_at_index(
503 Register result, Register index) {
504 assert_different_registers(result, index);
505 // convert from field index to resolved_references() index and from
506 // word index to byte offset. Since this is a java object, it can be compressed
507 Register tmp = index; // reuse
508 shll(tmp, LogBytesPerHeapOop);
509
510 get_constant_pool(result);
511 // load pointer for resolved_references[] objArray
512 movptr(result, Address(result, ConstantPool::resolved_references_offset_in_bytes()));
513 // JNIHandles::resolve(obj);
514 movptr(result, Address(result, 0));
515 // Add in the index
516 addptr(result, tmp);
517 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
518 }
519
520
521 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
522 // subtype of super_klass.
523 //
524 // Args:
525 // rax: superklass
526 // Rsub_klass: subklass
527 //
528 // Kills:
529 // rcx, rdi
530 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
531 Label& ok_is_subtype) {
532 assert(Rsub_klass != rax, "rax holds superklass");
533 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
534 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
535 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
536 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
537
538 // Profile the not-null value's klass.
539 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
540
541 // Do the check.
542 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
543
544 // Profile the failure of the check.
545 profile_typecheck_failed(rcx); // blows rcx
546 }
547
548
549 #ifndef _LP64
550 void InterpreterMacroAssembler::f2ieee() {
551 if (IEEEPrecision) {
552 fstp_s(Address(rsp, 0));
553 fld_s(Address(rsp, 0));
554 }
555 }
556
557
558 void InterpreterMacroAssembler::d2ieee() {
559 if (IEEEPrecision) {
560 fstp_d(Address(rsp, 0));
561 fld_d(Address(rsp, 0));
562 }
563 }
564 #endif // _LP64
565
566 // Java Expression Stack
567
568 void InterpreterMacroAssembler::pop_ptr(Register r) {
569 pop(r);
570 }
571
572 void InterpreterMacroAssembler::push_ptr(Register r) {
573 push(r);
574 }
575
576 void InterpreterMacroAssembler::push_i(Register r) {
577 push(r);
578 }
579
580 void InterpreterMacroAssembler::push_f(XMMRegister r) {
581 subptr(rsp, wordSize);
582 movflt(Address(rsp, 0), r);
583 }
584
585 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
586 movflt(r, Address(rsp, 0));
587 addptr(rsp, wordSize);
588 }
589
590 void InterpreterMacroAssembler::push_d(XMMRegister r) {
591 subptr(rsp, 2 * wordSize);
592 movdbl(Address(rsp, 0), r);
593 }
594
595 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
596 movdbl(r, Address(rsp, 0));
597 addptr(rsp, 2 * Interpreter::stackElementSize);
598 }
599
600 #ifdef _LP64
601 void InterpreterMacroAssembler::pop_i(Register r) {
602 // XXX can't use pop currently, upper half non clean
603 movl(r, Address(rsp, 0));
604 addptr(rsp, wordSize);
605 }
606
607 void InterpreterMacroAssembler::pop_l(Register r) {
608 movq(r, Address(rsp, 0));
609 addptr(rsp, 2 * Interpreter::stackElementSize);
610 }
611
612 void InterpreterMacroAssembler::push_l(Register r) {
613 subptr(rsp, 2 * wordSize);
614 movq(Address(rsp, 0), r);
615 }
616
617 void InterpreterMacroAssembler::pop(TosState state) {
618 switch (state) {
619 case atos: pop_ptr(); break;
620 case btos:
621 case ztos:
622 case ctos:
623 case stos:
624 case itos: pop_i(); break;
625 case ltos: pop_l(); break;
626 case ftos: pop_f(xmm0); break;
627 case dtos: pop_d(xmm0); break;
628 case vtos: /* nothing to do */ break;
629 default: ShouldNotReachHere();
630 }
631 verify_oop(rax, state);
632 }
633
634 void InterpreterMacroAssembler::push(TosState state) {
635 verify_oop(rax, state);
636 switch (state) {
637 case atos: push_ptr(); break;
638 case btos:
639 case ztos:
640 case ctos:
641 case stos:
642 case itos: push_i(); break;
643 case ltos: push_l(); break;
644 case ftos: push_f(xmm0); break;
645 case dtos: push_d(xmm0); break;
646 case vtos: /* nothing to do */ break;
647 default : ShouldNotReachHere();
648 }
649 }
650 #else
651 void InterpreterMacroAssembler::pop_i(Register r) {
652 pop(r);
653 }
654
655 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
656 pop(lo);
657 pop(hi);
658 }
659
660 void InterpreterMacroAssembler::pop_f() {
661 fld_s(Address(rsp, 0));
662 addptr(rsp, 1 * wordSize);
663 }
664
665 void InterpreterMacroAssembler::pop_d() {
666 fld_d(Address(rsp, 0));
667 addptr(rsp, 2 * wordSize);
668 }
669
670
671 void InterpreterMacroAssembler::pop(TosState state) {
672 switch (state) {
673 case atos: pop_ptr(rax); break;
674 case btos: // fall through
675 case ztos: // fall through
676 case ctos: // fall through
677 case stos: // fall through
678 case itos: pop_i(rax); break;
679 case ltos: pop_l(rax, rdx); break;
680 case ftos:
681 if (UseSSE >= 1) {
682 pop_f(xmm0);
683 } else {
684 pop_f();
685 }
686 break;
687 case dtos:
688 if (UseSSE >= 2) {
689 pop_d(xmm0);
690 } else {
691 pop_d();
692 }
693 break;
694 case vtos: /* nothing to do */ break;
695 default : ShouldNotReachHere();
696 }
697 verify_oop(rax, state);
698 }
699
700
701 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
702 push(hi);
703 push(lo);
704 }
705
706 void InterpreterMacroAssembler::push_f() {
707 // Do not schedule for no AGI! Never write beyond rsp!
708 subptr(rsp, 1 * wordSize);
709 fstp_s(Address(rsp, 0));
710 }
711
712 void InterpreterMacroAssembler::push_d() {
713 // Do not schedule for no AGI! Never write beyond rsp!
714 subptr(rsp, 2 * wordSize);
715 fstp_d(Address(rsp, 0));
716 }
717
718
719 void InterpreterMacroAssembler::push(TosState state) {
720 verify_oop(rax, state);
721 switch (state) {
722 case atos: push_ptr(rax); break;
723 case btos: // fall through
724 case ztos: // fall through
725 case ctos: // fall through
726 case stos: // fall through
727 case itos: push_i(rax); break;
728 case ltos: push_l(rax, rdx); break;
729 case ftos:
730 if (UseSSE >= 1) {
731 push_f(xmm0);
732 } else {
733 push_f();
734 }
735 break;
736 case dtos:
737 if (UseSSE >= 2) {
738 push_d(xmm0);
739 } else {
740 push_d();
741 }
742 break;
743 case vtos: /* nothing to do */ break;
744 default : ShouldNotReachHere();
745 }
746 }
747 #endif // _LP64
748
749
750 // Helpers for swap and dup
751 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
752 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
753 }
754
755 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
756 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
757 }
758
759
760 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
761 // set sender sp
762 lea(_bcp_register, Address(rsp, wordSize));
763 // record last_sp
764 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
765 }
766
767
768 // Jump to from_interpreted entry of a call unless single stepping is possible
769 // in this thread in which case we must call the i2i entry
770 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
771 prepare_to_jump_from_interpreted();
772
773 if (JvmtiExport::can_post_interpreter_events()) {
774 Label run_compiled_code;
775 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
776 // compiled code in threads for which the event is enabled. Check here for
777 // interp_only_mode if these events CAN be enabled.
778 // interp_only is an int, on little endian it is sufficient to test the byte only
779 // Is a cmpl faster?
780 LP64_ONLY(temp = r15_thread;)
781 NOT_LP64(get_thread(temp);)
782 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
783 jccb(Assembler::zero, run_compiled_code);
784 jmp(Address(method, Method::interpreter_entry_offset()));
785 bind(run_compiled_code);
786 }
787
788 jmp(Address(method, Method::from_interpreted_offset()));
789 }
790
791 // The following two routines provide a hook so that an implementation
792 // can schedule the dispatch in two parts. x86 does not do this.
793 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
794 // Nothing x86 specific to be done here
795 }
796
797 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
798 dispatch_next(state, step);
799 }
800
801 void InterpreterMacroAssembler::dispatch_base(TosState state,
802 address* table,
803 bool verifyoop) {
804 verify_FPU(1, state);
805 if (VerifyActivationFrameSize) {
806 Label L;
807 mov(rcx, rbp);
808 subptr(rcx, rsp);
809 int32_t min_frame_size =
810 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
811 wordSize;
812 cmpptr(rcx, (int32_t)min_frame_size);
813 jcc(Assembler::greaterEqual, L);
814 stop("broken stack frame");
815 bind(L);
816 }
817 if (verifyoop) {
818 verify_oop(rax, state);
819 }
820 #ifdef _LP64
821 lea(rscratch1, ExternalAddress((address)table));
822 jmp(Address(rscratch1, rbx, Address::times_8));
823 #else
824 Address index(noreg, rbx, Address::times_ptr);
825 ExternalAddress tbl((address)table);
826 ArrayAddress dispatch(tbl, index);
827 jump(dispatch);
828 #endif // _LP64
829 }
830
831 void InterpreterMacroAssembler::dispatch_only(TosState state) {
832 dispatch_base(state, Interpreter::dispatch_table(state));
833 }
834
835 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
836 dispatch_base(state, Interpreter::normal_table(state));
837 }
838
839 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
840 dispatch_base(state, Interpreter::normal_table(state), false);
841 }
842
843
844 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
845 // load next bytecode (load before advancing _bcp_register to prevent AGI)
846 load_unsigned_byte(rbx, Address(_bcp_register, step));
847 // advance _bcp_register
848 increment(_bcp_register, step);
849 dispatch_base(state, Interpreter::dispatch_table(state));
850 }
851
852 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
853 // load current bytecode
854 load_unsigned_byte(rbx, Address(_bcp_register, 0));
855 dispatch_base(state, table);
856 }
857
858 void InterpreterMacroAssembler::narrow(Register result) {
859
860 // Get method->_constMethod->_result_type
861 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
862 movptr(rcx, Address(rcx, Method::const_offset()));
863 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
864
865 Label done, notBool, notByte, notChar;
866
867 // common case first
868 cmpl(rcx, T_INT);
869 jcc(Assembler::equal, done);
870
871 // mask integer result to narrower return type.
872 cmpl(rcx, T_BOOLEAN);
873 jcc(Assembler::notEqual, notBool);
874 andl(result, 0x1);
875 jmp(done);
876
877 bind(notBool);
878 cmpl(rcx, T_BYTE);
879 jcc(Assembler::notEqual, notByte);
880 LP64_ONLY(movsbl(result, result);)
881 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
882 NOT_LP64(sarl(result, 24);) // and sign-extend byte
883 jmp(done);
884
885 bind(notByte);
886 cmpl(rcx, T_CHAR);
887 jcc(Assembler::notEqual, notChar);
888 LP64_ONLY(movzwl(result, result);)
889 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
890 jmp(done);
891
892 bind(notChar);
893 // cmpl(rcx, T_SHORT); // all that's left
894 // jcc(Assembler::notEqual, done);
895 LP64_ONLY(movswl(result, result);)
896 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
897 NOT_LP64(sarl(result, 16);) // and sign-extend short
898
899 // Nothing to do for T_INT
900 bind(done);
901 }
902
903 // remove activation
904 //
905 // Unlock the receiver if this is a synchronized method.
906 // Unlock any Java monitors from syncronized blocks.
907 // Remove the activation from the stack.
908 //
909 // If there are locked Java monitors
910 // If throw_monitor_exception
911 // throws IllegalMonitorStateException
912 // Else if install_monitor_exception
913 // installs IllegalMonitorStateException
914 // Else
915 // no error processing
916 void InterpreterMacroAssembler::remove_activation(
917 TosState state,
918 Register ret_addr,
919 bool throw_monitor_exception,
920 bool install_monitor_exception,
921 bool notify_jvmdi) {
922 // Note: Registers rdx xmm0 may be in use for the
923 // result check if synchronized method
924 Label unlocked, unlock, no_unlock;
925
926 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
927 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
928 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
929 // monitor pointers need different register
930 // because rdx may have the result in it
931 NOT_LP64(get_thread(rcx);)
932
933 // get the value of _do_not_unlock_if_synchronized into rdx
934 const Address do_not_unlock_if_synchronized(rthread,
935 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
936 movbool(rbx, do_not_unlock_if_synchronized);
937 movbool(do_not_unlock_if_synchronized, false); // reset the flag
938
939 // get method access flags
940 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
941 movl(rcx, Address(rcx, Method::access_flags_offset()));
942 testl(rcx, JVM_ACC_SYNCHRONIZED);
943 jcc(Assembler::zero, unlocked);
944
945 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
946 // is set.
947 testbool(rbx);
948 jcc(Assembler::notZero, no_unlock);
949
950 // unlock monitor
951 push(state); // save result
952
953 // BasicObjectLock will be first in list, since this is a
954 // synchronized method. However, need to check that the object has
955 // not been unlocked by an explicit monitorexit bytecode.
956 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
957 wordSize - (int) sizeof(BasicObjectLock));
958 // We use c_rarg1/rdx so that if we go slow path it will be the correct
959 // register for unlock_object to pass to VM directly
960 lea(robj, monitor); // address of first monitor
961
962 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
963 testptr(rax, rax);
964 jcc(Assembler::notZero, unlock);
965
966 pop(state);
967 if (throw_monitor_exception) {
968 // Entry already unlocked, need to throw exception
969 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
970 call_VM(noreg, CAST_FROM_FN_PTR(address,
971 InterpreterRuntime::throw_illegal_monitor_state_exception));
972 should_not_reach_here();
973 } else {
974 // Monitor already unlocked during a stack unroll. If requested,
975 // install an illegal_monitor_state_exception. Continue with
976 // stack unrolling.
977 if (install_monitor_exception) {
978 NOT_LP64(empty_FPU_stack();)
979 call_VM(noreg, CAST_FROM_FN_PTR(address,
980 InterpreterRuntime::new_illegal_monitor_state_exception));
981 }
982 jmp(unlocked);
983 }
984
985 bind(unlock);
986 unlock_object(robj);
987 pop(state);
988
989 // Check that for block-structured locking (i.e., that all locked
990 // objects has been unlocked)
991 bind(unlocked);
992
993 // rax, rdx: Might contain return value
994
995 // Check that all monitors are unlocked
996 {
997 Label loop, exception, entry, restart;
998 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
999 const Address monitor_block_top(
1000 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1001 const Address monitor_block_bot(
1002 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1003
1004 bind(restart);
1005 // We use c_rarg1 so that if we go slow path it will be the correct
1006 // register for unlock_object to pass to VM directly
1007 movptr(rmon, monitor_block_top); // points to current entry, starting
1008 // with top-most entry
1009 lea(rbx, monitor_block_bot); // points to word before bottom of
1010 // monitor block
1011 jmp(entry);
1012
1013 // Entry already locked, need to throw exception
1014 bind(exception);
1015
1016 if (throw_monitor_exception) {
1017 // Throw exception
1018 NOT_LP64(empty_FPU_stack();)
1019 MacroAssembler::call_VM(noreg,
1020 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1021 throw_illegal_monitor_state_exception));
1022 should_not_reach_here();
1023 } else {
1024 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1025 // Unlock does not block, so don't have to worry about the frame.
1026 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1027
1028 push(state);
1029 mov(robj, rmon); // nop if robj and rmon are the same
1030 unlock_object(robj);
1031 pop(state);
1032
1033 if (install_monitor_exception) {
1034 NOT_LP64(empty_FPU_stack();)
1035 call_VM(noreg, CAST_FROM_FN_PTR(address,
1036 InterpreterRuntime::
1037 new_illegal_monitor_state_exception));
1038 }
1039
1040 jmp(restart);
1041 }
1042
1043 bind(loop);
1044 // check if current entry is used
1045 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
1046 jcc(Assembler::notEqual, exception);
1047
1048 addptr(rmon, entry_size); // otherwise advance to next entry
1049 bind(entry);
1050 cmpptr(rmon, rbx); // check if bottom reached
1051 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1052 }
1053
1054 bind(no_unlock);
1055
1056 // jvmti support
1057 if (notify_jvmdi) {
1058 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1059 } else {
1060 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1061 }
1062
1063 // remove activation
1064 // get sender sp
1065 movptr(rbx,
1066 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1067 if (StackReservedPages > 0) {
1068 // testing if reserved zone needs to be re-enabled
1069 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1070 Label no_reserved_zone_enabling;
1071
1072 NOT_LP64(get_thread(rthread);)
1073
1074 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_enabled);
1075 jcc(Assembler::equal, no_reserved_zone_enabling);
1076
1077 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1078 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1079
1080 call_VM_leaf(
1081 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1082 push(rthread);
1083 call_VM(noreg, CAST_FROM_FN_PTR(address,
1084 InterpreterRuntime::throw_delayed_StackOverflowError));
1085 should_not_reach_here();
1086
1087 bind(no_reserved_zone_enabling);
1088 }
1089 leave(); // remove frame anchor
1090 pop(ret_addr); // get return address
1091 mov(rsp, rbx); // set sp to sender sp
1092 }
1093
1094 void InterpreterMacroAssembler::get_method_counters(Register method,
1095 Register mcs, Label& skip) {
1096 Label has_counters;
1097 movptr(mcs, Address(method, Method::method_counters_offset()));
1098 testptr(mcs, mcs);
1099 jcc(Assembler::notZero, has_counters);
1100 call_VM(noreg, CAST_FROM_FN_PTR(address,
1101 InterpreterRuntime::build_method_counters), method);
1102 movptr(mcs, Address(method,Method::method_counters_offset()));
1103 testptr(mcs, mcs);
1104 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1105 bind(has_counters);
1106 }
1107
1108
1109 // Lock object
1110 //
1111 // Args:
1112 // rdx, c_rarg1: BasicObjectLock to be used for locking
1113 //
1114 // Kills:
1115 // rax, rbx
1116 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1117 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1118 "The argument is only for looks. It must be c_rarg1");
1119
1120 if (UseHeavyMonitors) {
1121 call_VM(noreg,
1122 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1123 lock_reg);
1124 } else {
1125 Label done;
1126
1127 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1128 const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a
1129 // problematic case where tmp_reg = no_reg.
1130 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1131
1132 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
1133 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
1134 const int mark_offset = lock_offset +
1135 BasicLock::displaced_header_offset_in_bytes();
1136
1137 Label slow_case;
1138
1139 // Load object pointer into obj_reg
1140 movptr(obj_reg, Address(lock_reg, obj_offset));
1141
1142 if (UseBiasedLocking) {
1143 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case);
1144 }
1145
1146 // Load immediate 1 into swap_reg %rax
1147 movl(swap_reg, (int32_t)1);
1148
1149 // Load (object->mark() | 1) into swap_reg %rax
1150 orptr(swap_reg, Address(obj_reg, 0));
1151
1152 // Save (object->mark() | 1) into BasicLock's displaced header
1153 movptr(Address(lock_reg, mark_offset), swap_reg);
1154
1155 assert(lock_offset == 0,
1156 "displaced header must be first word in BasicObjectLock");
1157
1158 if (os::is_MP()) lock();
1159 cmpxchgptr(lock_reg, Address(obj_reg, 0));
1160 if (PrintBiasedLockingStatistics) {
1161 cond_inc32(Assembler::zero,
1162 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1163 }
1164 jcc(Assembler::zero, done);
1165
1166 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1167
1168 // Test if the oopMark is an obvious stack pointer, i.e.,
1169 // 1) (mark & zero_bits) == 0, and
1170 // 2) rsp <= mark < mark + os::pagesize()
1171 //
1172 // These 3 tests can be done by evaluating the following
1173 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1174 // assuming both stack pointer and pagesize have their
1175 // least significant bits clear.
1176 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
1177 subptr(swap_reg, rsp);
1178 andptr(swap_reg, zero_bits - os::vm_page_size());
1179
1180 // Save the test result, for recursive case, the result is zero
1181 movptr(Address(lock_reg, mark_offset), swap_reg);
1182
1183 if (PrintBiasedLockingStatistics) {
1184 cond_inc32(Assembler::zero,
1185 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1186 }
1187 jcc(Assembler::zero, done);
1188
1189 bind(slow_case);
1190
1191 // Call the runtime routine for slow case
1192 call_VM(noreg,
1193 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1194 lock_reg);
1195
1196 bind(done);
1197 }
1198 }
1199
1200
1201 // Unlocks an object. Used in monitorexit bytecode and
1202 // remove_activation. Throws an IllegalMonitorException if object is
1203 // not locked by current thread.
1204 //
1205 // Args:
1206 // rdx, c_rarg1: BasicObjectLock for lock
1207 //
1208 // Kills:
1209 // rax
1210 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1211 // rscratch1 (scratch reg)
1212 // rax, rbx, rcx, rdx
1213 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1214 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1215 "The argument is only for looks. It must be c_rarg1");
1216
1217 if (UseHeavyMonitors) {
1218 call_VM(noreg,
1219 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1220 lock_reg);
1221 } else {
1222 Label done;
1223
1224 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1225 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1226 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1227
1228 save_bcp(); // Save in case of exception
1229
1230 // Convert from BasicObjectLock structure to object and BasicLock
1231 // structure Store the BasicLock address into %rax
1232 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
1233
1234 // Load oop into obj_reg(%c_rarg3)
1235 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
1236
1237 // Free entry
1238 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
1239
1240 if (UseBiasedLocking) {
1241 biased_locking_exit(obj_reg, header_reg, done);
1242 }
1243
1244 // Load the old header from BasicLock structure
1245 movptr(header_reg, Address(swap_reg,
1246 BasicLock::displaced_header_offset_in_bytes()));
1247
1248 // Test for recursion
1249 testptr(header_reg, header_reg);
1250
1251 // zero for recursive case
1252 jcc(Assembler::zero, done);
1253
1254 // Atomic swap back the old header
1255 if (os::is_MP()) lock();
1256 cmpxchgptr(header_reg, Address(obj_reg, 0));
1257
1258 // zero for simple unlock of a stack-lock case
1259 jcc(Assembler::zero, done);
1260
1261 // Call the runtime routine for slow case.
1262 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
1263 obj_reg); // restore obj
1264 call_VM(noreg,
1265 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1266 lock_reg);
1267
1268 bind(done);
1269
1270 restore_bcp();
1271 }
1272 }
1273
1274 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1275 Label& zero_continue) {
1276 assert(ProfileInterpreter, "must be profiling interpreter");
1277 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1278 testptr(mdp, mdp);
1279 jcc(Assembler::zero, zero_continue);
1280 }
1281
1282
1283 // Set the method data pointer for the current bcp.
1284 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1285 assert(ProfileInterpreter, "must be profiling interpreter");
1286 Label set_mdp;
1287 push(rax);
1288 push(rbx);
1289
1290 get_method(rbx);
1291 // Test MDO to avoid the call if it is NULL.
1292 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1293 testptr(rax, rax);
1294 jcc(Assembler::zero, set_mdp);
1295 // rbx: method
1296 // _bcp_register: bcp
1297 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1298 // rax: mdi
1299 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1300 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1301 addptr(rbx, in_bytes(MethodData::data_offset()));
1302 addptr(rax, rbx);
1303 bind(set_mdp);
1304 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1305 pop(rbx);
1306 pop(rax);
1307 }
1308
1309 void InterpreterMacroAssembler::verify_method_data_pointer() {
1310 assert(ProfileInterpreter, "must be profiling interpreter");
1311 #ifdef ASSERT
1312 Label verify_continue;
1313 push(rax);
1314 push(rbx);
1315 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1316 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1317 push(arg3_reg);
1318 push(arg2_reg);
1319 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1320 get_method(rbx);
1321
1322 // If the mdp is valid, it will point to a DataLayout header which is
1323 // consistent with the bcp. The converse is highly probable also.
1324 load_unsigned_short(arg2_reg,
1325 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1326 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1327 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1328 cmpptr(arg2_reg, _bcp_register);
1329 jcc(Assembler::equal, verify_continue);
1330 // rbx: method
1331 // _bcp_register: bcp
1332 // c_rarg3: mdp
1333 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1334 rbx, _bcp_register, arg3_reg);
1335 bind(verify_continue);
1336 pop(arg2_reg);
1337 pop(arg3_reg);
1338 pop(rbx);
1339 pop(rax);
1340 #endif // ASSERT
1341 }
1342
1343
1344 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1345 int constant,
1346 Register value) {
1347 assert(ProfileInterpreter, "must be profiling interpreter");
1348 Address data(mdp_in, constant);
1349 movptr(data, value);
1350 }
1351
1352
1353 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1354 int constant,
1355 bool decrement) {
1356 // Counter address
1357 Address data(mdp_in, constant);
1358
1359 increment_mdp_data_at(data, decrement);
1360 }
1361
1362 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1363 bool decrement) {
1364 assert(ProfileInterpreter, "must be profiling interpreter");
1365 // %%% this does 64bit counters at best it is wasting space
1366 // at worst it is a rare bug when counters overflow
1367
1368 if (decrement) {
1369 // Decrement the register. Set condition codes.
1370 addptr(data, (int32_t) -DataLayout::counter_increment);
1371 // If the decrement causes the counter to overflow, stay negative
1372 Label L;
1373 jcc(Assembler::negative, L);
1374 addptr(data, (int32_t) DataLayout::counter_increment);
1375 bind(L);
1376 } else {
1377 assert(DataLayout::counter_increment == 1,
1378 "flow-free idiom only works with 1");
1379 // Increment the register. Set carry flag.
1380 addptr(data, DataLayout::counter_increment);
1381 // If the increment causes the counter to overflow, pull back by 1.
1382 sbbptr(data, (int32_t)0);
1383 }
1384 }
1385
1386
1387 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1388 Register reg,
1389 int constant,
1390 bool decrement) {
1391 Address data(mdp_in, reg, Address::times_1, constant);
1392
1393 increment_mdp_data_at(data, decrement);
1394 }
1395
1396 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1397 int flag_byte_constant) {
1398 assert(ProfileInterpreter, "must be profiling interpreter");
1399 int header_offset = in_bytes(DataLayout::header_offset());
1400 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
1401 // Set the flag
1402 orl(Address(mdp_in, header_offset), header_bits);
1403 }
1404
1405
1406
1407 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1408 int offset,
1409 Register value,
1410 Register test_value_out,
1411 Label& not_equal_continue) {
1412 assert(ProfileInterpreter, "must be profiling interpreter");
1413 if (test_value_out == noreg) {
1414 cmpptr(value, Address(mdp_in, offset));
1415 } else {
1416 // Put the test value into a register, so caller can use it:
1417 movptr(test_value_out, Address(mdp_in, offset));
1418 cmpptr(test_value_out, value);
1419 }
1420 jcc(Assembler::notEqual, not_equal_continue);
1421 }
1422
1423
1424 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1425 int offset_of_disp) {
1426 assert(ProfileInterpreter, "must be profiling interpreter");
1427 Address disp_address(mdp_in, offset_of_disp);
1428 addptr(mdp_in, disp_address);
1429 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1430 }
1431
1432
1433 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1434 Register reg,
1435 int offset_of_disp) {
1436 assert(ProfileInterpreter, "must be profiling interpreter");
1437 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1438 addptr(mdp_in, disp_address);
1439 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1440 }
1441
1442
1443 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1444 int constant) {
1445 assert(ProfileInterpreter, "must be profiling interpreter");
1446 addptr(mdp_in, constant);
1447 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1448 }
1449
1450
1451 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1452 assert(ProfileInterpreter, "must be profiling interpreter");
1453 push(return_bci); // save/restore across call_VM
1454 call_VM(noreg,
1455 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1456 return_bci);
1457 pop(return_bci);
1458 }
1459
1460
1461 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1462 Register bumped_count) {
1463 if (ProfileInterpreter) {
1464 Label profile_continue;
1465
1466 // If no method data exists, go to profile_continue.
1467 // Otherwise, assign to mdp
1468 test_method_data_pointer(mdp, profile_continue);
1469
1470 // We are taking a branch. Increment the taken count.
1471 // We inline increment_mdp_data_at to return bumped_count in a register
1472 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1473 Address data(mdp, in_bytes(JumpData::taken_offset()));
1474 movptr(bumped_count, data);
1475 assert(DataLayout::counter_increment == 1,
1476 "flow-free idiom only works with 1");
1477 addptr(bumped_count, DataLayout::counter_increment);
1478 sbbptr(bumped_count, 0);
1479 movptr(data, bumped_count); // Store back out
1480
1481 // The method data pointer needs to be updated to reflect the new target.
1482 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1483 bind(profile_continue);
1484 }
1485 }
1486
1487
1488 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1489 if (ProfileInterpreter) {
1490 Label profile_continue;
1491
1492 // If no method data exists, go to profile_continue.
1493 test_method_data_pointer(mdp, profile_continue);
1494
1495 // We are taking a branch. Increment the not taken count.
1496 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1497
1498 // The method data pointer needs to be updated to correspond to
1499 // the next bytecode
1500 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1501 bind(profile_continue);
1502 }
1503 }
1504
1505 void InterpreterMacroAssembler::profile_call(Register mdp) {
1506 if (ProfileInterpreter) {
1507 Label profile_continue;
1508
1509 // If no method data exists, go to profile_continue.
1510 test_method_data_pointer(mdp, profile_continue);
1511
1512 // We are making a call. Increment the count.
1513 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1514
1515 // The method data pointer needs to be updated to reflect the new target.
1516 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1517 bind(profile_continue);
1518 }
1519 }
1520
1521
1522 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1523 if (ProfileInterpreter) {
1524 Label profile_continue;
1525
1526 // If no method data exists, go to profile_continue.
1527 test_method_data_pointer(mdp, profile_continue);
1528
1529 // We are making a call. Increment the count.
1530 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1531
1532 // The method data pointer needs to be updated to reflect the new target.
1533 update_mdp_by_constant(mdp,
1534 in_bytes(VirtualCallData::
1535 virtual_call_data_size()));
1536 bind(profile_continue);
1537 }
1538 }
1539
1540
1541 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1542 Register mdp,
1543 Register reg2,
1544 bool receiver_can_be_null) {
1545 if (ProfileInterpreter) {
1546 Label profile_continue;
1547
1548 // If no method data exists, go to profile_continue.
1549 test_method_data_pointer(mdp, profile_continue);
1550
1551 Label skip_receiver_profile;
1552 if (receiver_can_be_null) {
1553 Label not_null;
1554 testptr(receiver, receiver);
1555 jccb(Assembler::notZero, not_null);
1556 // We are making a call. Increment the count for null receiver.
1557 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1558 jmp(skip_receiver_profile);
1559 bind(not_null);
1560 }
1561
1562 // Record the receiver type.
1563 record_klass_in_profile(receiver, mdp, reg2, true);
1564 bind(skip_receiver_profile);
1565
1566 // The method data pointer needs to be updated to reflect the new target.
1567 #if INCLUDE_JVMCI
1568 if (MethodProfileWidth == 0) {
1569 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1570 }
1571 #else // INCLUDE_JVMCI
1572 update_mdp_by_constant(mdp,
1573 in_bytes(VirtualCallData::
1574 virtual_call_data_size()));
1575 #endif // INCLUDE_JVMCI
1576 bind(profile_continue);
1577 }
1578 }
1579
1580 #if INCLUDE_JVMCI
1581 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) {
1582 assert_different_registers(method, mdp, reg2);
1583 if (ProfileInterpreter && MethodProfileWidth > 0) {
1584 Label profile_continue;
1585
1586 // If no method data exists, go to profile_continue.
1587 test_method_data_pointer(mdp, profile_continue);
1588
1589 Label done;
1590 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth,
1591 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1592 bind(done);
1593
1594 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1595 bind(profile_continue);
1596 }
1597 }
1598 #endif // INCLUDE_JVMCI
1599
1600 // This routine creates a state machine for updating the multi-row
1601 // type profile at a virtual call site (or other type-sensitive bytecode).
1602 // The machine visits each row (of receiver/count) until the receiver type
1603 // is found, or until it runs out of rows. At the same time, it remembers
1604 // the location of the first empty row. (An empty row records null for its
1605 // receiver, and can be allocated for a newly-observed receiver type.)
1606 // Because there are two degrees of freedom in the state, a simple linear
1607 // search will not work; it must be a decision tree. Hence this helper
1608 // function is recursive, to generate the required tree structured code.
1609 // It's the interpreter, so we are trading off code space for speed.
1610 // See below for example code.
1611 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1612 Register receiver, Register mdp,
1613 Register reg2, int start_row,
1614 Label& done, bool is_virtual_call) {
1615 if (TypeProfileWidth == 0) {
1616 if (is_virtual_call) {
1617 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1618 }
1619 #if INCLUDE_JVMCI
1620 else if (EnableJVMCI) {
1621 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1622 }
1623 #endif // INCLUDE_JVMCI
1624 } else {
1625 int non_profiled_offset = -1;
1626 if (is_virtual_call) {
1627 non_profiled_offset = in_bytes(CounterData::count_offset());
1628 }
1629 #if INCLUDE_JVMCI
1630 else if (EnableJVMCI) {
1631 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1632 }
1633 #endif // INCLUDE_JVMCI
1634
1635 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1636 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1637 }
1638 }
1639
1640 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1641 Register reg2, int start_row, Label& done, int total_rows,
1642 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1643 int non_profiled_offset) {
1644 int last_row = total_rows - 1;
1645 assert(start_row <= last_row, "must be work left to do");
1646 // Test this row for both the item and for null.
1647 // Take any of three different outcomes:
1648 // 1. found item => increment count and goto done
1649 // 2. found null => keep looking for case 1, maybe allocate this cell
1650 // 3. found something else => keep looking for cases 1 and 2
1651 // Case 3 is handled by a recursive call.
1652 for (int row = start_row; row <= last_row; row++) {
1653 Label next_test;
1654 bool test_for_null_also = (row == start_row);
1655
1656 // See if the item is item[n].
1657 int item_offset = in_bytes(item_offset_fn(row));
1658 test_mdp_data_at(mdp, item_offset, item,
1659 (test_for_null_also ? reg2 : noreg),
1660 next_test);
1661 // (Reg2 now contains the item from the CallData.)
1662
1663 // The item is item[n]. Increment count[n].
1664 int count_offset = in_bytes(item_count_offset_fn(row));
1665 increment_mdp_data_at(mdp, count_offset);
1666 jmp(done);
1667 bind(next_test);
1668
1669 if (test_for_null_also) {
1670 Label found_null;
1671 // Failed the equality check on item[n]... Test for null.
1672 testptr(reg2, reg2);
1673 if (start_row == last_row) {
1674 // The only thing left to do is handle the null case.
1675 if (non_profiled_offset >= 0) {
1676 jccb(Assembler::zero, found_null);
1677 // Item did not match any saved item and there is no empty row for it.
1678 // Increment total counter to indicate polymorphic case.
1679 increment_mdp_data_at(mdp, non_profiled_offset);
1680 jmp(done);
1681 bind(found_null);
1682 } else {
1683 jcc(Assembler::notZero, done);
1684 }
1685 break;
1686 }
1687 // Since null is rare, make it be the branch-taken case.
1688 jcc(Assembler::zero, found_null);
1689
1690 // Put all the "Case 3" tests here.
1691 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1692 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1693
1694 // Found a null. Keep searching for a matching item,
1695 // but remember that this is an empty (unused) slot.
1696 bind(found_null);
1697 }
1698 }
1699
1700 // In the fall-through case, we found no matching item, but we
1701 // observed the item[start_row] is NULL.
1702
1703 // Fill in the item field and increment the count.
1704 int item_offset = in_bytes(item_offset_fn(start_row));
1705 set_mdp_data_at(mdp, item_offset, item);
1706 int count_offset = in_bytes(item_count_offset_fn(start_row));
1707 movl(reg2, DataLayout::counter_increment);
1708 set_mdp_data_at(mdp, count_offset, reg2);
1709 if (start_row > 0) {
1710 jmp(done);
1711 }
1712 }
1713
1714 // Example state machine code for three profile rows:
1715 // // main copy of decision tree, rooted at row[1]
1716 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1717 // if (row[0].rec != NULL) {
1718 // // inner copy of decision tree, rooted at row[1]
1719 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1720 // if (row[1].rec != NULL) {
1721 // // degenerate decision tree, rooted at row[2]
1722 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1723 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1724 // row[2].init(rec); goto done;
1725 // } else {
1726 // // remember row[1] is empty
1727 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1728 // row[1].init(rec); goto done;
1729 // }
1730 // } else {
1731 // // remember row[0] is empty
1732 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1733 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1734 // row[0].init(rec); goto done;
1735 // }
1736 // done:
1737
1738 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1739 Register mdp, Register reg2,
1740 bool is_virtual_call) {
1741 assert(ProfileInterpreter, "must be profiling");
1742 Label done;
1743
1744 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1745
1746 bind (done);
1747 }
1748
1749 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1750 Register mdp) {
1751 if (ProfileInterpreter) {
1752 Label profile_continue;
1753 uint row;
1754
1755 // If no method data exists, go to profile_continue.
1756 test_method_data_pointer(mdp, profile_continue);
1757
1758 // Update the total ret count.
1759 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1760
1761 for (row = 0; row < RetData::row_limit(); row++) {
1762 Label next_test;
1763
1764 // See if return_bci is equal to bci[n]:
1765 test_mdp_data_at(mdp,
1766 in_bytes(RetData::bci_offset(row)),
1767 return_bci, noreg,
1768 next_test);
1769
1770 // return_bci is equal to bci[n]. Increment the count.
1771 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1772
1773 // The method data pointer needs to be updated to reflect the new target.
1774 update_mdp_by_offset(mdp,
1775 in_bytes(RetData::bci_displacement_offset(row)));
1776 jmp(profile_continue);
1777 bind(next_test);
1778 }
1779
1780 update_mdp_for_ret(return_bci);
1781
1782 bind(profile_continue);
1783 }
1784 }
1785
1786
1787 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1788 if (ProfileInterpreter) {
1789 Label profile_continue;
1790
1791 // If no method data exists, go to profile_continue.
1792 test_method_data_pointer(mdp, profile_continue);
1793
1794 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1795
1796 // The method data pointer needs to be updated.
1797 int mdp_delta = in_bytes(BitData::bit_data_size());
1798 if (TypeProfileCasts) {
1799 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1800 }
1801 update_mdp_by_constant(mdp, mdp_delta);
1802
1803 bind(profile_continue);
1804 }
1805 }
1806
1807
1808 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1809 if (ProfileInterpreter && TypeProfileCasts) {
1810 Label profile_continue;
1811
1812 // If no method data exists, go to profile_continue.
1813 test_method_data_pointer(mdp, profile_continue);
1814
1815 int count_offset = in_bytes(CounterData::count_offset());
1816 // Back up the address, since we have already bumped the mdp.
1817 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1818
1819 // *Decrement* the counter. We expect to see zero or small negatives.
1820 increment_mdp_data_at(mdp, count_offset, true);
1821
1822 bind (profile_continue);
1823 }
1824 }
1825
1826
1827 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1828 if (ProfileInterpreter) {
1829 Label profile_continue;
1830
1831 // If no method data exists, go to profile_continue.
1832 test_method_data_pointer(mdp, profile_continue);
1833
1834 // The method data pointer needs to be updated.
1835 int mdp_delta = in_bytes(BitData::bit_data_size());
1836 if (TypeProfileCasts) {
1837 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1838
1839 // Record the object type.
1840 record_klass_in_profile(klass, mdp, reg2, false);
1841 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1842 NOT_LP64(restore_locals();) // Restore EDI
1843 }
1844 update_mdp_by_constant(mdp, mdp_delta);
1845
1846 bind(profile_continue);
1847 }
1848 }
1849
1850
1851 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1852 if (ProfileInterpreter) {
1853 Label profile_continue;
1854
1855 // If no method data exists, go to profile_continue.
1856 test_method_data_pointer(mdp, profile_continue);
1857
1858 // Update the default case count
1859 increment_mdp_data_at(mdp,
1860 in_bytes(MultiBranchData::default_count_offset()));
1861
1862 // The method data pointer needs to be updated.
1863 update_mdp_by_offset(mdp,
1864 in_bytes(MultiBranchData::
1865 default_displacement_offset()));
1866
1867 bind(profile_continue);
1868 }
1869 }
1870
1871
1872 void InterpreterMacroAssembler::profile_switch_case(Register index,
1873 Register mdp,
1874 Register reg2) {
1875 if (ProfileInterpreter) {
1876 Label profile_continue;
1877
1878 // If no method data exists, go to profile_continue.
1879 test_method_data_pointer(mdp, profile_continue);
1880
1881 // Build the base (index * per_case_size_in_bytes()) +
1882 // case_array_offset_in_bytes()
1883 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1884 imulptr(index, reg2); // XXX l ?
1885 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1886
1887 // Update the case count
1888 increment_mdp_data_at(mdp,
1889 index,
1890 in_bytes(MultiBranchData::relative_count_offset()));
1891
1892 // The method data pointer needs to be updated.
1893 update_mdp_by_offset(mdp,
1894 index,
1895 in_bytes(MultiBranchData::
1896 relative_displacement_offset()));
1897
1898 bind(profile_continue);
1899 }
1900 }
1901
1902
1903
1904 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1905 if (state == atos) {
1906 MacroAssembler::verify_oop(reg);
1907 }
1908 }
1909
1910 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1911 #ifndef _LP64
1912 if ((state == ftos && UseSSE < 1) ||
1913 (state == dtos && UseSSE < 2)) {
1914 MacroAssembler::verify_FPU(stack_depth);
1915 }
1916 #endif
1917 }
1918
1919 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1920 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1921 int increment, Address mask,
1922 Register scratch, bool preloaded,
1923 Condition cond, Label* where) {
1924 if (!preloaded) {
1925 movl(scratch, counter_addr);
1926 }
1927 incrementl(scratch, increment);
1928 movl(counter_addr, scratch);
1929 andl(scratch, mask);
1930 jcc(cond, *where);
1931 }
1932
1933 void InterpreterMacroAssembler::notify_method_entry() {
1934 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1935 // track stack depth. If it is possible to enter interp_only_mode we add
1936 // the code to check if the event should be sent.
1937 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1938 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1939 if (JvmtiExport::can_post_interpreter_events()) {
1940 Label L;
1941 NOT_LP64(get_thread(rthread);)
1942 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1943 testl(rdx, rdx);
1944 jcc(Assembler::zero, L);
1945 call_VM(noreg, CAST_FROM_FN_PTR(address,
1946 InterpreterRuntime::post_method_entry));
1947 bind(L);
1948 }
1949
1950 {
1951 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1952 NOT_LP64(get_thread(rthread);)
1953 get_method(rarg);
1954 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1955 rthread, rarg);
1956 }
1957
1958 // RedefineClasses() tracing support for obsolete method entry
1959 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1960 NOT_LP64(get_thread(rthread);)
1961 get_method(rarg);
1962 call_VM_leaf(
1963 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1964 rthread, rarg);
1965 }
1966 }
1967
1968
1969 void InterpreterMacroAssembler::notify_method_exit(
1970 TosState state, NotifyMethodExitMode mode) {
1971 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1972 // track stack depth. If it is possible to enter interp_only_mode we add
1973 // the code to check if the event should be sent.
1974 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1975 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1976 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1977 Label L;
1978 // Note: frame::interpreter_frame_result has a dependency on how the
1979 // method result is saved across the call to post_method_exit. If this
1980 // is changed then the interpreter_frame_result implementation will
1981 // need to be updated too.
1982
1983 // template interpreter will leave the result on the top of the stack.
1984 push(state);
1985 NOT_LP64(get_thread(rthread);)
1986 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1987 testl(rdx, rdx);
1988 jcc(Assembler::zero, L);
1989 call_VM(noreg,
1990 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1991 bind(L);
1992 pop(state);
1993 }
1994
1995 {
1996 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1997 push(state);
1998 NOT_LP64(get_thread(rthread);)
1999 get_method(rarg);
2000 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2001 rthread, rarg);
2002 pop(state);
2003 }
2004 }