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 oopDesc::bs()->interpreter_read_barrier_not_null(this, result);
516 // Add in the index
517 addptr(result, tmp);
518 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
519 }
520
521
522 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
523 // subtype of super_klass.
524 //
525 // Args:
526 // rax: superklass
527 // Rsub_klass: subklass
528 //
529 // Kills:
530 // rcx, rdi
531 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
532 Label& ok_is_subtype) {
533 assert(Rsub_klass != rax, "rax holds superklass");
534 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
535 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
536 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
537 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
538
539 // Profile the not-null value's klass.
540 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
541
542 // Do the check.
543 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
544
545 // Profile the failure of the check.
546 profile_typecheck_failed(rcx); // blows rcx
547 }
548
549
550 #ifndef _LP64
551 void InterpreterMacroAssembler::f2ieee() {
552 if (IEEEPrecision) {
553 fstp_s(Address(rsp, 0));
554 fld_s(Address(rsp, 0));
555 }
556 }
557
558
559 void InterpreterMacroAssembler::d2ieee() {
560 if (IEEEPrecision) {
561 fstp_d(Address(rsp, 0));
562 fld_d(Address(rsp, 0));
563 }
564 }
565 #endif // _LP64
566
567 // Java Expression Stack
568
569 void InterpreterMacroAssembler::pop_ptr(Register r) {
570 pop(r);
571 }
572
573 void InterpreterMacroAssembler::push_ptr(Register r) {
574 push(r);
575 }
576
577 void InterpreterMacroAssembler::push_i(Register r) {
578 push(r);
579 }
580
581 void InterpreterMacroAssembler::push_f(XMMRegister r) {
582 subptr(rsp, wordSize);
583 movflt(Address(rsp, 0), r);
584 }
585
586 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
587 movflt(r, Address(rsp, 0));
588 addptr(rsp, wordSize);
589 }
590
591 void InterpreterMacroAssembler::push_d(XMMRegister r) {
592 subptr(rsp, 2 * wordSize);
593 movdbl(Address(rsp, 0), r);
594 }
595
596 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
597 movdbl(r, Address(rsp, 0));
598 addptr(rsp, 2 * Interpreter::stackElementSize);
599 }
600
601 #ifdef _LP64
602 void InterpreterMacroAssembler::pop_i(Register r) {
603 // XXX can't use pop currently, upper half non clean
604 movl(r, Address(rsp, 0));
605 addptr(rsp, wordSize);
606 }
607
608 void InterpreterMacroAssembler::pop_l(Register r) {
609 movq(r, Address(rsp, 0));
610 addptr(rsp, 2 * Interpreter::stackElementSize);
611 }
612
613 void InterpreterMacroAssembler::push_l(Register r) {
614 subptr(rsp, 2 * wordSize);
615 movq(Address(rsp, 0), r);
616 }
617
618 void InterpreterMacroAssembler::pop(TosState state) {
619 switch (state) {
620 case atos: pop_ptr(); break;
621 case btos:
622 case ztos:
623 case ctos:
624 case stos:
625 case itos: pop_i(); break;
626 case ltos: pop_l(); break;
627 case ftos: pop_f(xmm0); break;
628 case dtos: pop_d(xmm0); break;
629 case vtos: /* nothing to do */ break;
630 default: ShouldNotReachHere();
631 }
632 verify_oop(rax, state);
633 }
634
635 void InterpreterMacroAssembler::push(TosState state) {
636 verify_oop(rax, state);
637 switch (state) {
638 case atos: push_ptr(); break;
639 case btos:
640 case ztos:
641 case ctos:
642 case stos:
643 case itos: push_i(); break;
644 case ltos: push_l(); break;
645 case ftos: push_f(xmm0); break;
646 case dtos: push_d(xmm0); break;
647 case vtos: /* nothing to do */ break;
648 default : ShouldNotReachHere();
649 }
650 }
651 #else
652 void InterpreterMacroAssembler::pop_i(Register r) {
653 pop(r);
654 }
655
656 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
657 pop(lo);
658 pop(hi);
659 }
660
661 void InterpreterMacroAssembler::pop_f() {
662 fld_s(Address(rsp, 0));
663 addptr(rsp, 1 * wordSize);
664 }
665
666 void InterpreterMacroAssembler::pop_d() {
667 fld_d(Address(rsp, 0));
668 addptr(rsp, 2 * wordSize);
669 }
670
671
672 void InterpreterMacroAssembler::pop(TosState state) {
673 switch (state) {
674 case atos: pop_ptr(rax); break;
675 case btos: // fall through
676 case ztos: // fall through
677 case ctos: // fall through
678 case stos: // fall through
679 case itos: pop_i(rax); break;
680 case ltos: pop_l(rax, rdx); break;
681 case ftos:
682 if (UseSSE >= 1) {
683 pop_f(xmm0);
684 } else {
685 pop_f();
686 }
687 break;
688 case dtos:
689 if (UseSSE >= 2) {
690 pop_d(xmm0);
691 } else {
692 pop_d();
693 }
694 break;
695 case vtos: /* nothing to do */ break;
696 default : ShouldNotReachHere();
697 }
698 verify_oop(rax, state);
699 }
700
701
702 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
703 push(hi);
704 push(lo);
705 }
706
707 void InterpreterMacroAssembler::push_f() {
708 // Do not schedule for no AGI! Never write beyond rsp!
709 subptr(rsp, 1 * wordSize);
710 fstp_s(Address(rsp, 0));
711 }
712
713 void InterpreterMacroAssembler::push_d() {
714 // Do not schedule for no AGI! Never write beyond rsp!
715 subptr(rsp, 2 * wordSize);
716 fstp_d(Address(rsp, 0));
717 }
718
719
720 void InterpreterMacroAssembler::push(TosState state) {
721 verify_oop(rax, state);
722 switch (state) {
723 case atos: push_ptr(rax); break;
724 case btos: // fall through
725 case ztos: // fall through
726 case ctos: // fall through
727 case stos: // fall through
728 case itos: push_i(rax); break;
729 case ltos: push_l(rax, rdx); break;
730 case ftos:
731 if (UseSSE >= 1) {
732 push_f(xmm0);
733 } else {
734 push_f();
735 }
736 break;
737 case dtos:
738 if (UseSSE >= 2) {
739 push_d(xmm0);
740 } else {
741 push_d();
742 }
743 break;
744 case vtos: /* nothing to do */ break;
745 default : ShouldNotReachHere();
746 }
747 }
748 #endif // _LP64
749
750
751 // Helpers for swap and dup
752 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
753 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
754 }
755
756 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
757 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
758 }
759
760
761 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
762 // set sender sp
763 lea(_bcp_register, Address(rsp, wordSize));
764 // record last_sp
765 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
766 }
767
768
769 // Jump to from_interpreted entry of a call unless single stepping is possible
770 // in this thread in which case we must call the i2i entry
771 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
772 prepare_to_jump_from_interpreted();
773
774 if (JvmtiExport::can_post_interpreter_events()) {
775 Label run_compiled_code;
776 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
777 // compiled code in threads for which the event is enabled. Check here for
778 // interp_only_mode if these events CAN be enabled.
779 // interp_only is an int, on little endian it is sufficient to test the byte only
780 // Is a cmpl faster?
781 LP64_ONLY(temp = r15_thread;)
782 NOT_LP64(get_thread(temp);)
783 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
784 jccb(Assembler::zero, run_compiled_code);
785 jmp(Address(method, Method::interpreter_entry_offset()));
786 bind(run_compiled_code);
787 }
788
789 jmp(Address(method, Method::from_interpreted_offset()));
790 }
791
792 // The following two routines provide a hook so that an implementation
793 // can schedule the dispatch in two parts. x86 does not do this.
794 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
795 // Nothing x86 specific to be done here
796 }
797
798 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
799 dispatch_next(state, step);
800 }
801
802 void InterpreterMacroAssembler::dispatch_base(TosState state,
803 address* table,
804 bool verifyoop) {
805 verify_FPU(1, state);
806 if (VerifyActivationFrameSize) {
807 Label L;
808 mov(rcx, rbp);
809 subptr(rcx, rsp);
810 int32_t min_frame_size =
811 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
812 wordSize;
813 cmpptr(rcx, (int32_t)min_frame_size);
814 jcc(Assembler::greaterEqual, L);
815 stop("broken stack frame");
816 bind(L);
817 }
818 if (verifyoop) {
819 verify_oop(rax, state);
820 }
821 #ifdef _LP64
822 lea(rscratch1, ExternalAddress((address)table));
823 jmp(Address(rscratch1, rbx, Address::times_8));
824 #else
825 Address index(noreg, rbx, Address::times_ptr);
826 ExternalAddress tbl((address)table);
827 ArrayAddress dispatch(tbl, index);
828 jump(dispatch);
829 #endif // _LP64
830 }
831
832 void InterpreterMacroAssembler::dispatch_only(TosState state) {
833 dispatch_base(state, Interpreter::dispatch_table(state));
834 }
835
836 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
837 dispatch_base(state, Interpreter::normal_table(state));
838 }
839
840 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
841 dispatch_base(state, Interpreter::normal_table(state), false);
842 }
843
844
845 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
846 // load next bytecode (load before advancing _bcp_register to prevent AGI)
847 load_unsigned_byte(rbx, Address(_bcp_register, step));
848 // advance _bcp_register
849 increment(_bcp_register, step);
850 dispatch_base(state, Interpreter::dispatch_table(state));
851 }
852
853 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
854 // load current bytecode
855 load_unsigned_byte(rbx, Address(_bcp_register, 0));
856 dispatch_base(state, table);
857 }
858
859 void InterpreterMacroAssembler::narrow(Register result) {
860
861 // Get method->_constMethod->_result_type
862 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
863 movptr(rcx, Address(rcx, Method::const_offset()));
864 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
865
866 Label done, notBool, notByte, notChar;
867
868 // common case first
869 cmpl(rcx, T_INT);
870 jcc(Assembler::equal, done);
871
872 // mask integer result to narrower return type.
873 cmpl(rcx, T_BOOLEAN);
874 jcc(Assembler::notEqual, notBool);
875 andl(result, 0x1);
876 jmp(done);
877
878 bind(notBool);
879 cmpl(rcx, T_BYTE);
880 jcc(Assembler::notEqual, notByte);
881 LP64_ONLY(movsbl(result, result);)
882 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
883 NOT_LP64(sarl(result, 24);) // and sign-extend byte
884 jmp(done);
885
886 bind(notByte);
887 cmpl(rcx, T_CHAR);
888 jcc(Assembler::notEqual, notChar);
889 LP64_ONLY(movzwl(result, result);)
890 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
891 jmp(done);
892
893 bind(notChar);
894 // cmpl(rcx, T_SHORT); // all that's left
895 // jcc(Assembler::notEqual, done);
896 LP64_ONLY(movswl(result, result);)
897 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
898 NOT_LP64(sarl(result, 16);) // and sign-extend short
899
900 // Nothing to do for T_INT
901 bind(done);
902 }
903
904 // remove activation
905 //
906 // Unlock the receiver if this is a synchronized method.
907 // Unlock any Java monitors from syncronized blocks.
908 // Remove the activation from the stack.
909 //
910 // If there are locked Java monitors
911 // If throw_monitor_exception
912 // throws IllegalMonitorStateException
913 // Else if install_monitor_exception
914 // installs IllegalMonitorStateException
915 // Else
916 // no error processing
917 void InterpreterMacroAssembler::remove_activation(
918 TosState state,
919 Register ret_addr,
920 bool throw_monitor_exception,
921 bool install_monitor_exception,
922 bool notify_jvmdi) {
923 // Note: Registers rdx xmm0 may be in use for the
924 // result check if synchronized method
925 Label unlocked, unlock, no_unlock;
926
927 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
928 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
929 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
930 // monitor pointers need different register
931 // because rdx may have the result in it
932 NOT_LP64(get_thread(rcx);)
933
934 // get the value of _do_not_unlock_if_synchronized into rdx
935 const Address do_not_unlock_if_synchronized(rthread,
936 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
937 movbool(rbx, do_not_unlock_if_synchronized);
938 movbool(do_not_unlock_if_synchronized, false); // reset the flag
939
940 // get method access flags
941 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
942 movl(rcx, Address(rcx, Method::access_flags_offset()));
943 testl(rcx, JVM_ACC_SYNCHRONIZED);
944 jcc(Assembler::zero, unlocked);
945
946 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
947 // is set.
948 testbool(rbx);
949 jcc(Assembler::notZero, no_unlock);
950
951 // unlock monitor
952 push(state); // save result
953
954 // BasicObjectLock will be first in list, since this is a
955 // synchronized method. However, need to check that the object has
956 // not been unlocked by an explicit monitorexit bytecode.
957 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
958 wordSize - (int) sizeof(BasicObjectLock));
959 // We use c_rarg1/rdx so that if we go slow path it will be the correct
960 // register for unlock_object to pass to VM directly
961 lea(robj, monitor); // address of first monitor
962
963 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
964 shenandoah_store_addr_check(rax); // Invariant
965 testptr(rax, rax);
966 jcc(Assembler::notZero, unlock);
967
968 pop(state);
969 if (throw_monitor_exception) {
970 // Entry already unlocked, need to throw exception
971 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
972 call_VM(noreg, CAST_FROM_FN_PTR(address,
973 InterpreterRuntime::throw_illegal_monitor_state_exception));
974 should_not_reach_here();
975 } else {
976 // Monitor already unlocked during a stack unroll. If requested,
977 // install an illegal_monitor_state_exception. Continue with
978 // stack unrolling.
979 if (install_monitor_exception) {
980 NOT_LP64(empty_FPU_stack();)
981 call_VM(noreg, CAST_FROM_FN_PTR(address,
982 InterpreterRuntime::new_illegal_monitor_state_exception));
983 }
984 jmp(unlocked);
985 }
986
987 bind(unlock);
988 unlock_object(robj);
989 pop(state);
990
991 // Check that for block-structured locking (i.e., that all locked
992 // objects has been unlocked)
993 bind(unlocked);
994
995 // rax, rdx: Might contain return value
996
997 // Check that all monitors are unlocked
998 {
999 Label loop, exception, entry, restart;
1000 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1001 const Address monitor_block_top(
1002 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1003 const Address monitor_block_bot(
1004 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1005
1006 bind(restart);
1007 // We use c_rarg1 so that if we go slow path it will be the correct
1008 // register for unlock_object to pass to VM directly
1009 movptr(rmon, monitor_block_top); // points to current entry, starting
1010 // with top-most entry
1011 lea(rbx, monitor_block_bot); // points to word before bottom of
1012 // monitor block
1013 jmp(entry);
1014
1015 // Entry already locked, need to throw exception
1016 bind(exception);
1017
1018 if (throw_monitor_exception) {
1019 // Throw exception
1020 NOT_LP64(empty_FPU_stack();)
1021 MacroAssembler::call_VM(noreg,
1022 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1023 throw_illegal_monitor_state_exception));
1024 should_not_reach_here();
1025 } else {
1026 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1027 // Unlock does not block, so don't have to worry about the frame.
1028 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1029
1030 push(state);
1031 mov(robj, rmon); // nop if robj and rmon are the same
1032 unlock_object(robj);
1033 pop(state);
1034
1035 if (install_monitor_exception) {
1036 NOT_LP64(empty_FPU_stack();)
1037 call_VM(noreg, CAST_FROM_FN_PTR(address,
1038 InterpreterRuntime::
1039 new_illegal_monitor_state_exception));
1040 }
1041
1042 jmp(restart);
1043 }
1044
1045 bind(loop);
1046 // check if current entry is used
1047 shenandoah_lock_check(rmon);
1048 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
1049 jcc(Assembler::notEqual, exception);
1050
1051 addptr(rmon, entry_size); // otherwise advance to next entry
1052 bind(entry);
1053 cmpptr(rmon, rbx); // check if bottom reached
1054 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1055 }
1056
1057 bind(no_unlock);
1058
1059 // jvmti support
1060 if (notify_jvmdi) {
1061 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1062 } else {
1063 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1064 }
1065
1066 // remove activation
1067 // get sender sp
1068 movptr(rbx,
1069 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1070 if (StackReservedPages > 0) {
1071 // testing if reserved zone needs to be re-enabled
1072 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1073 Label no_reserved_zone_enabling;
1074
1075 NOT_LP64(get_thread(rthread);)
1076
1077 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_enabled);
1078 jcc(Assembler::equal, no_reserved_zone_enabling);
1079
1080 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1081 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1082
1083 call_VM_leaf(
1084 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1085 push(rthread);
1086 call_VM(noreg, CAST_FROM_FN_PTR(address,
1087 InterpreterRuntime::throw_delayed_StackOverflowError));
1088 should_not_reach_here();
1089
1090 bind(no_reserved_zone_enabling);
1091 }
1092 leave(); // remove frame anchor
1093 pop(ret_addr); // get return address
1094 mov(rsp, rbx); // set sp to sender sp
1095 }
1096
1097 void InterpreterMacroAssembler::get_method_counters(Register method,
1098 Register mcs, Label& skip) {
1099 Label has_counters;
1100 movptr(mcs, Address(method, Method::method_counters_offset()));
1101 testptr(mcs, mcs);
1102 jcc(Assembler::notZero, has_counters);
1103 call_VM(noreg, CAST_FROM_FN_PTR(address,
1104 InterpreterRuntime::build_method_counters), method);
1105 movptr(mcs, Address(method,Method::method_counters_offset()));
1106 testptr(mcs, mcs);
1107 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1108 bind(has_counters);
1109 }
1110
1111
1112 // Lock object
1113 //
1114 // Args:
1115 // rdx, c_rarg1: BasicObjectLock to be used for locking
1116 //
1117 // Kills:
1118 // rax, rbx
1119 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1120 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1121 "The argument is only for looks. It must be c_rarg1");
1122
1123 if (UseHeavyMonitors) {
1124 call_VM(noreg,
1125 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1126 lock_reg);
1127 } else {
1128 Label done;
1129
1130 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1131 const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a
1132 // problematic case where tmp_reg = no_reg.
1133 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1134
1135 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
1136 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
1137 const int mark_offset = lock_offset +
1138 BasicLock::displaced_header_offset_in_bytes();
1139
1140 Label slow_case;
1141
1142 // Load object pointer into obj_reg
1143 movptr(obj_reg, Address(lock_reg, obj_offset));
1144
1145 shenandoah_store_addr_check(obj_reg);
1146
1147 if (UseBiasedLocking) {
1148 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case);
1149 }
1150
1151 // Load immediate 1 into swap_reg %rax
1152 movl(swap_reg, (int32_t)1);
1153
1154 // Load (object->mark() | 1) into swap_reg %rax
1155 orptr(swap_reg, Address(obj_reg, 0));
1156
1157 // Save (object->mark() | 1) into BasicLock's displaced header
1158 movptr(Address(lock_reg, mark_offset), swap_reg);
1159
1160 assert(lock_offset == 0,
1161 "displaced header must be first word in BasicObjectLock");
1162
1163 // obj_reg has been checked a few lines up.
1164 if (os::is_MP()) lock();
1165 cmpxchgptr(lock_reg, Address(obj_reg, 0));
1166 if (PrintBiasedLockingStatistics) {
1167 cond_inc32(Assembler::zero,
1168 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1169 }
1170 jcc(Assembler::zero, done);
1171
1172 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1173
1174 // Test if the oopMark is an obvious stack pointer, i.e.,
1175 // 1) (mark & zero_bits) == 0, and
1176 // 2) rsp <= mark < mark + os::pagesize()
1177 //
1178 // These 3 tests can be done by evaluating the following
1179 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1180 // assuming both stack pointer and pagesize have their
1181 // least significant bits clear.
1182 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
1183 subptr(swap_reg, rsp);
1184 andptr(swap_reg, zero_bits - os::vm_page_size());
1185
1186 // Save the test result, for recursive case, the result is zero
1187 movptr(Address(lock_reg, mark_offset), swap_reg);
1188
1189 if (PrintBiasedLockingStatistics) {
1190 cond_inc32(Assembler::zero,
1191 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1192 }
1193 jcc(Assembler::zero, done);
1194
1195 bind(slow_case);
1196
1197 // Call the runtime routine for slow case
1198 call_VM(noreg,
1199 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1200 lock_reg);
1201
1202 bind(done);
1203 }
1204 }
1205
1206
1207 // Unlocks an object. Used in monitorexit bytecode and
1208 // remove_activation. Throws an IllegalMonitorException if object is
1209 // not locked by current thread.
1210 //
1211 // Args:
1212 // rdx, c_rarg1: BasicObjectLock for lock
1213 //
1214 // Kills:
1215 // rax
1216 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1217 // rscratch1 (scratch reg)
1218 // rax, rbx, rcx, rdx
1219 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1220 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1221 "The argument is only for looks. It must be c_rarg1");
1222
1223 if (UseHeavyMonitors) {
1224 call_VM(noreg,
1225 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1226 lock_reg);
1227 } else {
1228 Label done;
1229
1230 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1231 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1232 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1233
1234 save_bcp(); // Save in case of exception
1235
1236 // Convert from BasicObjectLock structure to object and BasicLock
1237 // structure Store the BasicLock address into %rax
1238 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
1239
1240 // Load oop into obj_reg(%c_rarg3)
1241 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
1242 shenandoah_store_addr_check(obj_reg); // Invariant
1243
1244 // Free entry
1245 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
1246
1247 if (UseBiasedLocking) {
1248 biased_locking_exit(obj_reg, header_reg, done);
1249 }
1250
1251 // Load the old header from BasicLock structure
1252 movptr(header_reg, Address(swap_reg,
1253 BasicLock::displaced_header_offset_in_bytes()));
1254
1255 // Test for recursion
1256 testptr(header_reg, header_reg);
1257
1258 // zero for recursive case
1259 jcc(Assembler::zero, done);
1260
1261 // Atomic swap back the old header
1262 if (os::is_MP()) lock();
1263 cmpxchgptr(header_reg, Address(obj_reg, 0));
1264
1265 // zero for simple unlock of a stack-lock case
1266 jcc(Assembler::zero, done);
1267
1268 // Call the runtime routine for slow case.
1269 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
1270 obj_reg); // restore obj
1271 call_VM(noreg,
1272 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1273 lock_reg);
1274
1275 bind(done);
1276
1277 restore_bcp();
1278 }
1279 }
1280
1281 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1282 Label& zero_continue) {
1283 assert(ProfileInterpreter, "must be profiling interpreter");
1284 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1285 testptr(mdp, mdp);
1286 jcc(Assembler::zero, zero_continue);
1287 }
1288
1289
1290 // Set the method data pointer for the current bcp.
1291 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1292 assert(ProfileInterpreter, "must be profiling interpreter");
1293 Label set_mdp;
1294 push(rax);
1295 push(rbx);
1296
1297 get_method(rbx);
1298 // Test MDO to avoid the call if it is NULL.
1299 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1300 testptr(rax, rax);
1301 jcc(Assembler::zero, set_mdp);
1302 // rbx: method
1303 // _bcp_register: bcp
1304 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1305 // rax: mdi
1306 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1307 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1308 addptr(rbx, in_bytes(MethodData::data_offset()));
1309 addptr(rax, rbx);
1310 bind(set_mdp);
1311 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1312 pop(rbx);
1313 pop(rax);
1314 }
1315
1316 void InterpreterMacroAssembler::verify_method_data_pointer() {
1317 assert(ProfileInterpreter, "must be profiling interpreter");
1318 #ifdef ASSERT
1319 Label verify_continue;
1320 push(rax);
1321 push(rbx);
1322 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1323 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1324 push(arg3_reg);
1325 push(arg2_reg);
1326 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1327 get_method(rbx);
1328
1329 // If the mdp is valid, it will point to a DataLayout header which is
1330 // consistent with the bcp. The converse is highly probable also.
1331 load_unsigned_short(arg2_reg,
1332 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1333 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1334 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1335 cmpptr(arg2_reg, _bcp_register);
1336 jcc(Assembler::equal, verify_continue);
1337 // rbx: method
1338 // _bcp_register: bcp
1339 // c_rarg3: mdp
1340 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1341 rbx, _bcp_register, arg3_reg);
1342 bind(verify_continue);
1343 pop(arg2_reg);
1344 pop(arg3_reg);
1345 pop(rbx);
1346 pop(rax);
1347 #endif // ASSERT
1348 }
1349
1350
1351 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1352 int constant,
1353 Register value) {
1354 assert(ProfileInterpreter, "must be profiling interpreter");
1355 Address data(mdp_in, constant);
1356 movptr(data, value);
1357 }
1358
1359
1360 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1361 int constant,
1362 bool decrement) {
1363 // Counter address
1364 Address data(mdp_in, constant);
1365
1366 increment_mdp_data_at(data, decrement);
1367 }
1368
1369 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1370 bool decrement) {
1371 assert(ProfileInterpreter, "must be profiling interpreter");
1372 // %%% this does 64bit counters at best it is wasting space
1373 // at worst it is a rare bug when counters overflow
1374
1375 if (decrement) {
1376 // Decrement the register. Set condition codes.
1377 addptr(data, (int32_t) -DataLayout::counter_increment);
1378 // If the decrement causes the counter to overflow, stay negative
1379 Label L;
1380 jcc(Assembler::negative, L);
1381 addptr(data, (int32_t) DataLayout::counter_increment);
1382 bind(L);
1383 } else {
1384 assert(DataLayout::counter_increment == 1,
1385 "flow-free idiom only works with 1");
1386 // Increment the register. Set carry flag.
1387 addptr(data, DataLayout::counter_increment);
1388 // If the increment causes the counter to overflow, pull back by 1.
1389 sbbptr(data, (int32_t)0);
1390 }
1391 }
1392
1393
1394 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1395 Register reg,
1396 int constant,
1397 bool decrement) {
1398 Address data(mdp_in, reg, Address::times_1, constant);
1399
1400 increment_mdp_data_at(data, decrement);
1401 }
1402
1403 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1404 int flag_byte_constant) {
1405 assert(ProfileInterpreter, "must be profiling interpreter");
1406 int header_offset = in_bytes(DataLayout::header_offset());
1407 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
1408 // Set the flag
1409 orl(Address(mdp_in, header_offset), header_bits);
1410 }
1411
1412
1413
1414 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1415 int offset,
1416 Register value,
1417 Register test_value_out,
1418 Label& not_equal_continue) {
1419 assert(ProfileInterpreter, "must be profiling interpreter");
1420 if (test_value_out == noreg) {
1421 cmpptr(value, Address(mdp_in, offset));
1422 } else {
1423 // Put the test value into a register, so caller can use it:
1424 movptr(test_value_out, Address(mdp_in, offset));
1425 cmpptr(test_value_out, value);
1426 }
1427 jcc(Assembler::notEqual, not_equal_continue);
1428 }
1429
1430
1431 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1432 int offset_of_disp) {
1433 assert(ProfileInterpreter, "must be profiling interpreter");
1434 Address disp_address(mdp_in, offset_of_disp);
1435 addptr(mdp_in, disp_address);
1436 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1437 }
1438
1439
1440 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1441 Register reg,
1442 int offset_of_disp) {
1443 assert(ProfileInterpreter, "must be profiling interpreter");
1444 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1445 addptr(mdp_in, disp_address);
1446 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1447 }
1448
1449
1450 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1451 int constant) {
1452 assert(ProfileInterpreter, "must be profiling interpreter");
1453 addptr(mdp_in, constant);
1454 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1455 }
1456
1457
1458 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1459 assert(ProfileInterpreter, "must be profiling interpreter");
1460 push(return_bci); // save/restore across call_VM
1461 call_VM(noreg,
1462 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1463 return_bci);
1464 pop(return_bci);
1465 }
1466
1467
1468 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1469 Register bumped_count) {
1470 if (ProfileInterpreter) {
1471 Label profile_continue;
1472
1473 // If no method data exists, go to profile_continue.
1474 // Otherwise, assign to mdp
1475 test_method_data_pointer(mdp, profile_continue);
1476
1477 // We are taking a branch. Increment the taken count.
1478 // We inline increment_mdp_data_at to return bumped_count in a register
1479 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1480 Address data(mdp, in_bytes(JumpData::taken_offset()));
1481 movptr(bumped_count, data);
1482 assert(DataLayout::counter_increment == 1,
1483 "flow-free idiom only works with 1");
1484 addptr(bumped_count, DataLayout::counter_increment);
1485 sbbptr(bumped_count, 0);
1486 movptr(data, bumped_count); // Store back out
1487
1488 // The method data pointer needs to be updated to reflect the new target.
1489 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1490 bind(profile_continue);
1491 }
1492 }
1493
1494
1495 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1496 if (ProfileInterpreter) {
1497 Label profile_continue;
1498
1499 // If no method data exists, go to profile_continue.
1500 test_method_data_pointer(mdp, profile_continue);
1501
1502 // We are taking a branch. Increment the not taken count.
1503 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1504
1505 // The method data pointer needs to be updated to correspond to
1506 // the next bytecode
1507 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1508 bind(profile_continue);
1509 }
1510 }
1511
1512 void InterpreterMacroAssembler::profile_call(Register mdp) {
1513 if (ProfileInterpreter) {
1514 Label profile_continue;
1515
1516 // If no method data exists, go to profile_continue.
1517 test_method_data_pointer(mdp, profile_continue);
1518
1519 // We are making a call. Increment the count.
1520 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1521
1522 // The method data pointer needs to be updated to reflect the new target.
1523 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1524 bind(profile_continue);
1525 }
1526 }
1527
1528
1529 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1530 if (ProfileInterpreter) {
1531 Label profile_continue;
1532
1533 // If no method data exists, go to profile_continue.
1534 test_method_data_pointer(mdp, profile_continue);
1535
1536 // We are making a call. Increment the count.
1537 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1538
1539 // The method data pointer needs to be updated to reflect the new target.
1540 update_mdp_by_constant(mdp,
1541 in_bytes(VirtualCallData::
1542 virtual_call_data_size()));
1543 bind(profile_continue);
1544 }
1545 }
1546
1547
1548 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1549 Register mdp,
1550 Register reg2,
1551 bool receiver_can_be_null) {
1552 if (ProfileInterpreter) {
1553 Label profile_continue;
1554
1555 // If no method data exists, go to profile_continue.
1556 test_method_data_pointer(mdp, profile_continue);
1557
1558 Label skip_receiver_profile;
1559 if (receiver_can_be_null) {
1560 Label not_null;
1561 testptr(receiver, receiver);
1562 jccb(Assembler::notZero, not_null);
1563 // We are making a call. Increment the count for null receiver.
1564 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1565 jmp(skip_receiver_profile);
1566 bind(not_null);
1567 }
1568
1569 // Record the receiver type.
1570 record_klass_in_profile(receiver, mdp, reg2, true);
1571 bind(skip_receiver_profile);
1572
1573 // The method data pointer needs to be updated to reflect the new target.
1574 #if INCLUDE_JVMCI
1575 if (MethodProfileWidth == 0) {
1576 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1577 }
1578 #else // INCLUDE_JVMCI
1579 update_mdp_by_constant(mdp,
1580 in_bytes(VirtualCallData::
1581 virtual_call_data_size()));
1582 #endif // INCLUDE_JVMCI
1583 bind(profile_continue);
1584 }
1585 }
1586
1587 #if INCLUDE_JVMCI
1588 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) {
1589 assert_different_registers(method, mdp, reg2);
1590 if (ProfileInterpreter && MethodProfileWidth > 0) {
1591 Label profile_continue;
1592
1593 // If no method data exists, go to profile_continue.
1594 test_method_data_pointer(mdp, profile_continue);
1595
1596 Label done;
1597 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth,
1598 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1599 bind(done);
1600
1601 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1602 bind(profile_continue);
1603 }
1604 }
1605 #endif // INCLUDE_JVMCI
1606
1607 // This routine creates a state machine for updating the multi-row
1608 // type profile at a virtual call site (or other type-sensitive bytecode).
1609 // The machine visits each row (of receiver/count) until the receiver type
1610 // is found, or until it runs out of rows. At the same time, it remembers
1611 // the location of the first empty row. (An empty row records null for its
1612 // receiver, and can be allocated for a newly-observed receiver type.)
1613 // Because there are two degrees of freedom in the state, a simple linear
1614 // search will not work; it must be a decision tree. Hence this helper
1615 // function is recursive, to generate the required tree structured code.
1616 // It's the interpreter, so we are trading off code space for speed.
1617 // See below for example code.
1618 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1619 Register receiver, Register mdp,
1620 Register reg2, int start_row,
1621 Label& done, bool is_virtual_call) {
1622 if (TypeProfileWidth == 0) {
1623 if (is_virtual_call) {
1624 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1625 }
1626 #if INCLUDE_JVMCI
1627 else if (EnableJVMCI) {
1628 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1629 }
1630 #endif // INCLUDE_JVMCI
1631 } else {
1632 int non_profiled_offset = -1;
1633 if (is_virtual_call) {
1634 non_profiled_offset = in_bytes(CounterData::count_offset());
1635 }
1636 #if INCLUDE_JVMCI
1637 else if (EnableJVMCI) {
1638 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1639 }
1640 #endif // INCLUDE_JVMCI
1641
1642 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1643 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1644 }
1645 }
1646
1647 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1648 Register reg2, int start_row, Label& done, int total_rows,
1649 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1650 int non_profiled_offset) {
1651 int last_row = total_rows - 1;
1652 assert(start_row <= last_row, "must be work left to do");
1653 // Test this row for both the item and for null.
1654 // Take any of three different outcomes:
1655 // 1. found item => increment count and goto done
1656 // 2. found null => keep looking for case 1, maybe allocate this cell
1657 // 3. found something else => keep looking for cases 1 and 2
1658 // Case 3 is handled by a recursive call.
1659 for (int row = start_row; row <= last_row; row++) {
1660 Label next_test;
1661 bool test_for_null_also = (row == start_row);
1662
1663 // See if the item is item[n].
1664 int item_offset = in_bytes(item_offset_fn(row));
1665 test_mdp_data_at(mdp, item_offset, item,
1666 (test_for_null_also ? reg2 : noreg),
1667 next_test);
1668 // (Reg2 now contains the item from the CallData.)
1669
1670 // The item is item[n]. Increment count[n].
1671 int count_offset = in_bytes(item_count_offset_fn(row));
1672 increment_mdp_data_at(mdp, count_offset);
1673 jmp(done);
1674 bind(next_test);
1675
1676 if (test_for_null_also) {
1677 Label found_null;
1678 // Failed the equality check on item[n]... Test for null.
1679 testptr(reg2, reg2);
1680 if (start_row == last_row) {
1681 // The only thing left to do is handle the null case.
1682 if (non_profiled_offset >= 0) {
1683 jccb(Assembler::zero, found_null);
1684 // Item did not match any saved item and there is no empty row for it.
1685 // Increment total counter to indicate polymorphic case.
1686 increment_mdp_data_at(mdp, non_profiled_offset);
1687 jmp(done);
1688 bind(found_null);
1689 } else {
1690 jcc(Assembler::notZero, done);
1691 }
1692 break;
1693 }
1694 // Since null is rare, make it be the branch-taken case.
1695 jcc(Assembler::zero, found_null);
1696
1697 // Put all the "Case 3" tests here.
1698 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1699 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1700
1701 // Found a null. Keep searching for a matching item,
1702 // but remember that this is an empty (unused) slot.
1703 bind(found_null);
1704 }
1705 }
1706
1707 // In the fall-through case, we found no matching item, but we
1708 // observed the item[start_row] is NULL.
1709
1710 // Fill in the item field and increment the count.
1711 int item_offset = in_bytes(item_offset_fn(start_row));
1712 set_mdp_data_at(mdp, item_offset, item);
1713 int count_offset = in_bytes(item_count_offset_fn(start_row));
1714 movl(reg2, DataLayout::counter_increment);
1715 set_mdp_data_at(mdp, count_offset, reg2);
1716 if (start_row > 0) {
1717 jmp(done);
1718 }
1719 }
1720
1721 // Example state machine code for three profile rows:
1722 // // main copy of decision tree, rooted at row[1]
1723 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1724 // if (row[0].rec != NULL) {
1725 // // inner copy of decision tree, rooted at row[1]
1726 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1727 // if (row[1].rec != NULL) {
1728 // // degenerate decision tree, rooted at row[2]
1729 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1730 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1731 // row[2].init(rec); goto done;
1732 // } else {
1733 // // remember row[1] is empty
1734 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1735 // row[1].init(rec); goto done;
1736 // }
1737 // } else {
1738 // // remember row[0] is empty
1739 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1740 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1741 // row[0].init(rec); goto done;
1742 // }
1743 // done:
1744
1745 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1746 Register mdp, Register reg2,
1747 bool is_virtual_call) {
1748 assert(ProfileInterpreter, "must be profiling");
1749 Label done;
1750
1751 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1752
1753 bind (done);
1754 }
1755
1756 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1757 Register mdp) {
1758 if (ProfileInterpreter) {
1759 Label profile_continue;
1760 uint row;
1761
1762 // If no method data exists, go to profile_continue.
1763 test_method_data_pointer(mdp, profile_continue);
1764
1765 // Update the total ret count.
1766 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1767
1768 for (row = 0; row < RetData::row_limit(); row++) {
1769 Label next_test;
1770
1771 // See if return_bci is equal to bci[n]:
1772 test_mdp_data_at(mdp,
1773 in_bytes(RetData::bci_offset(row)),
1774 return_bci, noreg,
1775 next_test);
1776
1777 // return_bci is equal to bci[n]. Increment the count.
1778 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1779
1780 // The method data pointer needs to be updated to reflect the new target.
1781 update_mdp_by_offset(mdp,
1782 in_bytes(RetData::bci_displacement_offset(row)));
1783 jmp(profile_continue);
1784 bind(next_test);
1785 }
1786
1787 update_mdp_for_ret(return_bci);
1788
1789 bind(profile_continue);
1790 }
1791 }
1792
1793
1794 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1795 if (ProfileInterpreter) {
1796 Label profile_continue;
1797
1798 // If no method data exists, go to profile_continue.
1799 test_method_data_pointer(mdp, profile_continue);
1800
1801 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1802
1803 // The method data pointer needs to be updated.
1804 int mdp_delta = in_bytes(BitData::bit_data_size());
1805 if (TypeProfileCasts) {
1806 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1807 }
1808 update_mdp_by_constant(mdp, mdp_delta);
1809
1810 bind(profile_continue);
1811 }
1812 }
1813
1814
1815 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1816 if (ProfileInterpreter && TypeProfileCasts) {
1817 Label profile_continue;
1818
1819 // If no method data exists, go to profile_continue.
1820 test_method_data_pointer(mdp, profile_continue);
1821
1822 int count_offset = in_bytes(CounterData::count_offset());
1823 // Back up the address, since we have already bumped the mdp.
1824 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1825
1826 // *Decrement* the counter. We expect to see zero or small negatives.
1827 increment_mdp_data_at(mdp, count_offset, true);
1828
1829 bind (profile_continue);
1830 }
1831 }
1832
1833
1834 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1835 if (ProfileInterpreter) {
1836 Label profile_continue;
1837
1838 // If no method data exists, go to profile_continue.
1839 test_method_data_pointer(mdp, profile_continue);
1840
1841 // The method data pointer needs to be updated.
1842 int mdp_delta = in_bytes(BitData::bit_data_size());
1843 if (TypeProfileCasts) {
1844 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1845
1846 // Record the object type.
1847 record_klass_in_profile(klass, mdp, reg2, false);
1848 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1849 NOT_LP64(restore_locals();) // Restore EDI
1850 }
1851 update_mdp_by_constant(mdp, mdp_delta);
1852
1853 bind(profile_continue);
1854 }
1855 }
1856
1857
1858 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1859 if (ProfileInterpreter) {
1860 Label profile_continue;
1861
1862 // If no method data exists, go to profile_continue.
1863 test_method_data_pointer(mdp, profile_continue);
1864
1865 // Update the default case count
1866 increment_mdp_data_at(mdp,
1867 in_bytes(MultiBranchData::default_count_offset()));
1868
1869 // The method data pointer needs to be updated.
1870 update_mdp_by_offset(mdp,
1871 in_bytes(MultiBranchData::
1872 default_displacement_offset()));
1873
1874 bind(profile_continue);
1875 }
1876 }
1877
1878
1879 void InterpreterMacroAssembler::profile_switch_case(Register index,
1880 Register mdp,
1881 Register reg2) {
1882 if (ProfileInterpreter) {
1883 Label profile_continue;
1884
1885 // If no method data exists, go to profile_continue.
1886 test_method_data_pointer(mdp, profile_continue);
1887
1888 // Build the base (index * per_case_size_in_bytes()) +
1889 // case_array_offset_in_bytes()
1890 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1891 imulptr(index, reg2); // XXX l ?
1892 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1893
1894 // Update the case count
1895 increment_mdp_data_at(mdp,
1896 index,
1897 in_bytes(MultiBranchData::relative_count_offset()));
1898
1899 // The method data pointer needs to be updated.
1900 update_mdp_by_offset(mdp,
1901 index,
1902 in_bytes(MultiBranchData::
1903 relative_displacement_offset()));
1904
1905 bind(profile_continue);
1906 }
1907 }
1908
1909
1910
1911 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1912 if (state == atos) {
1913 MacroAssembler::verify_oop(reg);
1914 }
1915 }
1916
1917 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1918 #ifndef _LP64
1919 if ((state == ftos && UseSSE < 1) ||
1920 (state == dtos && UseSSE < 2)) {
1921 MacroAssembler::verify_FPU(stack_depth);
1922 }
1923 #endif
1924 }
1925
1926 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1927 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1928 int increment, Address mask,
1929 Register scratch, bool preloaded,
1930 Condition cond, Label* where) {
1931 if (!preloaded) {
1932 movl(scratch, counter_addr);
1933 }
1934 incrementl(scratch, increment);
1935 movl(counter_addr, scratch);
1936 andl(scratch, mask);
1937 jcc(cond, *where);
1938 }
1939
1940 void InterpreterMacroAssembler::notify_method_entry() {
1941 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1942 // track stack depth. If it is possible to enter interp_only_mode we add
1943 // the code to check if the event should be sent.
1944 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1945 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1946 if (JvmtiExport::can_post_interpreter_events()) {
1947 Label L;
1948 NOT_LP64(get_thread(rthread);)
1949 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1950 testl(rdx, rdx);
1951 jcc(Assembler::zero, L);
1952 call_VM(noreg, CAST_FROM_FN_PTR(address,
1953 InterpreterRuntime::post_method_entry));
1954 bind(L);
1955 }
1956
1957 {
1958 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1959 NOT_LP64(get_thread(rthread);)
1960 get_method(rarg);
1961 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1962 rthread, rarg);
1963 }
1964
1965 // RedefineClasses() tracing support for obsolete method entry
1966 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1967 NOT_LP64(get_thread(rthread);)
1968 get_method(rarg);
1969 call_VM_leaf(
1970 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1971 rthread, rarg);
1972 }
1973 }
1974
1975
1976 void InterpreterMacroAssembler::notify_method_exit(
1977 TosState state, NotifyMethodExitMode mode) {
1978 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1979 // track stack depth. If it is possible to enter interp_only_mode we add
1980 // the code to check if the event should be sent.
1981 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1982 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1983 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1984 Label L;
1985 // Note: frame::interpreter_frame_result has a dependency on how the
1986 // method result is saved across the call to post_method_exit. If this
1987 // is changed then the interpreter_frame_result implementation will
1988 // need to be updated too.
1989
1990 // template interpreter will leave the result on the top of the stack.
1991 push(state);
1992 NOT_LP64(get_thread(rthread);)
1993 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1994 testl(rdx, rdx);
1995 jcc(Assembler::zero, L);
1996 call_VM(noreg,
1997 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1998 bind(L);
1999 pop(state);
2000 }
2001
2002 {
2003 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2004 push(state);
2005 NOT_LP64(get_thread(rthread);)
2006 get_method(rarg);
2007 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2008 rthread, rarg);
2009 pop(state);
2010 }
2011 }