1 /*
2 * Copyright (c) 1997, 2024, 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 "compiler/compiler_globals.hpp"
27 #include "interp_masm_x86.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "logging/log.hpp"
31 #include "oops/arrayOop.hpp"
32 #include "oops/markWord.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "prims/jvmtiExport.hpp"
36 #include "prims/jvmtiThreadState.hpp"
37 #include "runtime/basicLock.hpp"
38 #include "runtime/biasedLocking.hpp"
39 #include "runtime/frame.inline.hpp"
40 #include "runtime/safepointMechanism.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/thread.inline.hpp"
43 #include "utilities/powerOfTwo.hpp"
44
45 // Implementation of InterpreterMacroAssembler
46
47 void InterpreterMacroAssembler::jump_to_entry(address entry) {
48 assert(entry, "Entry must have been generated by now");
49 jump(RuntimeAddress(entry));
50 }
51
52 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
53 Label update, next, none;
54
55 #ifdef _LP64
56 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
57 #else
58 assert_different_registers(obj, mdo_addr.base(), mdo_addr.index());
59 #endif
60
61 interp_verify_oop(obj, atos);
62
63 testptr(obj, obj);
64 jccb(Assembler::notZero, update);
65 testptr(mdo_addr, TypeEntries::null_seen);
66 jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore.
67 // atomic update to prevent overwriting Klass* with 0
68 lock();
69 orptr(mdo_addr, TypeEntries::null_seen);
70 jmp(next);
71
72 bind(update);
73 Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
74 load_klass(obj, obj, tmp_load_klass);
75 #ifdef _LP64
76 mov(rscratch1, obj);
77 #endif
78
79 xorptr(obj, mdo_addr);
80 testptr(obj, TypeEntries::type_klass_mask);
81 jccb(Assembler::zero, next); // klass seen before, nothing to
82 // do. The unknown bit may have been
83 // set already but no need to check.
84
85 testptr(obj, TypeEntries::type_unknown);
86 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
87
88 cmpptr(mdo_addr, 0);
89 jccb(Assembler::equal, none);
90 cmpptr(mdo_addr, TypeEntries::null_seen);
91 jccb(Assembler::equal, none);
92 #ifdef _LP64
93 // There is a chance that the checks above (re-reading profiling
94 // data from memory) fail if another thread has just set the
95 // profiling to this obj's klass
96 mov(obj, rscratch1);
97 xorptr(obj, mdo_addr);
98 testptr(obj, TypeEntries::type_klass_mask);
99 jccb(Assembler::zero, next);
100 #endif
101
102 // different than before. Cannot keep accurate profile.
103 orptr(mdo_addr, TypeEntries::type_unknown);
104 jmpb(next);
105
106 bind(none);
107 // first time here. Set profile type.
108 movptr(mdo_addr, obj);
109 #ifdef ASSERT
110 andptr(obj, TypeEntries::type_klass_mask);
111 verify_klass_ptr(obj);
112 #endif
113
114 bind(next);
115 }
116
117 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
118 if (!ProfileInterpreter) {
119 return;
120 }
121
122 if (MethodData::profile_arguments() || MethodData::profile_return()) {
123 Label profile_continue;
124
125 test_method_data_pointer(mdp, profile_continue);
126
127 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
128
129 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
130 jcc(Assembler::notEqual, profile_continue);
131
132 if (MethodData::profile_arguments()) {
133 Label done;
134 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
135 addptr(mdp, off_to_args);
136
137 for (int i = 0; i < TypeProfileArgsLimit; i++) {
138 if (i > 0 || MethodData::profile_return()) {
139 // If return value type is profiled we may have no argument to profile
140 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
141 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
142 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
143 jcc(Assembler::less, done);
144 }
145 movptr(tmp, Address(callee, Method::const_offset()));
146 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
147 // stack offset o (zero based) from the start of the argument
148 // list, for n arguments translates into offset n - o - 1 from
149 // the end of the argument list
150 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
151 subl(tmp, 1);
152 Address arg_addr = argument_address(tmp);
153 movptr(tmp, arg_addr);
154
155 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
156 profile_obj_type(tmp, mdo_arg_addr);
157
158 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
159 addptr(mdp, to_add);
160 off_to_args += to_add;
161 }
162
163 if (MethodData::profile_return()) {
164 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
165 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
166 }
167
168 bind(done);
169
170 if (MethodData::profile_return()) {
171 // We're right after the type profile for the last
172 // argument. tmp is the number of cells left in the
173 // CallTypeData/VirtualCallTypeData to reach its end. Non null
174 // if there's a return to profile.
175 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
176 shll(tmp, log2i_exact((int)DataLayout::cell_size));
177 addptr(mdp, tmp);
178 }
179 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
180 } else {
181 assert(MethodData::profile_return(), "either profile call args or call ret");
182 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
183 }
184
185 // mdp points right after the end of the
186 // CallTypeData/VirtualCallTypeData, right after the cells for the
187 // return value type if there's one
188
189 bind(profile_continue);
190 }
191 }
192
193 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
194 assert_different_registers(mdp, ret, tmp, _bcp_register);
195 if (ProfileInterpreter && MethodData::profile_return()) {
196 Label profile_continue;
197
198 test_method_data_pointer(mdp, profile_continue);
199
200 if (MethodData::profile_return_jsr292_only()) {
201 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
202
203 // If we don't profile all invoke bytecodes we must make sure
204 // it's a bytecode we indeed profile. We can't go back to the
205 // begining of the ProfileData we intend to update to check its
206 // type because we're right after it and we don't known its
207 // length
208 Label do_profile;
209 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
210 jcc(Assembler::equal, do_profile);
211 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
212 jcc(Assembler::equal, do_profile);
213 get_method(tmp);
214 cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
215 jcc(Assembler::notEqual, profile_continue);
216
217 bind(do_profile);
218 }
219
220 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
221 mov(tmp, ret);
222 profile_obj_type(tmp, mdo_ret_addr);
223
224 bind(profile_continue);
225 }
226 }
227
228 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
229 if (ProfileInterpreter && MethodData::profile_parameters()) {
230 Label profile_continue;
231
232 test_method_data_pointer(mdp, profile_continue);
233
234 // Load the offset of the area within the MDO used for
235 // parameters. If it's negative we're not profiling any parameters
236 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
237 testl(tmp1, tmp1);
238 jcc(Assembler::negative, profile_continue);
239
240 // Compute a pointer to the area for parameters from the offset
241 // and move the pointer to the slot for the last
242 // parameters. Collect profiling from last parameter down.
243 // mdo start + parameters offset + array length - 1
244 addptr(mdp, tmp1);
245 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
246 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
247
248 Label loop;
249 bind(loop);
250
251 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
252 int type_base = in_bytes(ParametersTypeData::type_offset(0));
253 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
254 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
255 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
256
257 // load offset on the stack from the slot for this parameter
258 movptr(tmp2, arg_off);
259 negptr(tmp2);
260 // read the parameter from the local area
261 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
262
263 // profile the parameter
264 profile_obj_type(tmp2, arg_type);
265
266 // go to next parameter
267 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
268 jcc(Assembler::positive, loop);
269
270 bind(profile_continue);
271 }
272 }
273
274 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
275 int number_of_arguments) {
276 // interpreter specific
277 //
278 // Note: No need to save/restore bcp & locals registers
279 // since these are callee saved registers and no blocking/
280 // GC can happen in leaf calls.
281 // Further Note: DO NOT save/restore bcp/locals. If a caller has
282 // already saved them so that it can use rsi/rdi as temporaries
283 // then a save/restore here will DESTROY the copy the caller
284 // saved! There used to be a save_bcp() that only happened in
285 // the ASSERT path (no restore_bcp). Which caused bizarre failures
286 // when jvm built with ASSERTs.
287 #ifdef ASSERT
288 {
289 Label L;
290 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
291 jcc(Assembler::equal, L);
292 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
293 " last_sp != NULL");
294 bind(L);
295 }
296 #endif
297 // super call
298 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
299 // interpreter specific
300 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
301 // but since they may not have been saved (and we don't want to
302 // save them here (see note above) the assert is invalid.
303 }
304
305 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
306 Register java_thread,
307 Register last_java_sp,
308 address entry_point,
309 int number_of_arguments,
310 bool check_exceptions) {
311 // interpreter specific
312 //
313 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
314 // really make a difference for these runtime calls, since they are
315 // slow anyway. Btw., bcp must be saved/restored since it may change
316 // due to GC.
317 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
318 save_bcp();
319 #ifdef ASSERT
320 {
321 Label L;
322 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
323 jcc(Assembler::equal, L);
324 stop("InterpreterMacroAssembler::call_VM_base:"
325 " last_sp != NULL");
326 bind(L);
327 }
328 #endif /* ASSERT */
329 // super call
330 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
331 entry_point, number_of_arguments,
332 check_exceptions);
333 // interpreter specific
334 restore_bcp();
335 restore_locals();
336 }
337
338 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
339 if (JvmtiExport::can_pop_frame()) {
340 Label L;
341 // Initiate popframe handling only if it is not already being
342 // processed. If the flag has the popframe_processing bit set, it
343 // means that this code is called *during* popframe handling - we
344 // don't want to reenter.
345 // This method is only called just after the call into the vm in
346 // call_VM_base, so the arg registers are available.
347 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
348 LP64_ONLY(c_rarg0);
349 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
350 testl(pop_cond, JavaThread::popframe_pending_bit);
351 jcc(Assembler::zero, L);
352 testl(pop_cond, JavaThread::popframe_processing_bit);
353 jcc(Assembler::notZero, L);
354 // Call Interpreter::remove_activation_preserving_args_entry() to get the
355 // address of the same-named entrypoint in the generated interpreter code.
356 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
357 jmp(rax);
358 bind(L);
359 NOT_LP64(get_thread(java_thread);)
360 }
361 }
362
363 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
364 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
365 NOT_LP64(get_thread(thread);)
366 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
367 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
368 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
369 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
370 #ifdef _LP64
371 switch (state) {
372 case atos: movptr(rax, oop_addr);
373 movptr(oop_addr, (int32_t)NULL_WORD);
374 interp_verify_oop(rax, state); break;
375 case ltos: movptr(rax, val_addr); break;
376 case btos: // fall through
377 case ztos: // fall through
378 case ctos: // fall through
379 case stos: // fall through
380 case itos: movl(rax, val_addr); break;
381 case ftos: load_float(val_addr); break;
382 case dtos: load_double(val_addr); break;
383 case vtos: /* nothing to do */ break;
384 default : ShouldNotReachHere();
385 }
386 // Clean up tos value in the thread object
387 movl(tos_addr, (int) ilgl);
388 movl(val_addr, (int32_t) NULL_WORD);
389 #else
390 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
391 + in_ByteSize(wordSize));
392 switch (state) {
393 case atos: movptr(rax, oop_addr);
394 movptr(oop_addr, NULL_WORD);
395 interp_verify_oop(rax, state); break;
396 case ltos:
397 movl(rdx, val_addr1); // fall through
398 case btos: // fall through
399 case ztos: // fall through
400 case ctos: // fall through
401 case stos: // fall through
402 case itos: movl(rax, val_addr); break;
403 case ftos: load_float(val_addr); break;
404 case dtos: load_double(val_addr); break;
405 case vtos: /* nothing to do */ break;
406 default : ShouldNotReachHere();
407 }
408 #endif // _LP64
409 // Clean up tos value in the thread object
410 movl(tos_addr, (int32_t) ilgl);
411 movptr(val_addr, NULL_WORD);
412 NOT_LP64(movptr(val_addr1, NULL_WORD);)
413 }
414
415
416 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
417 if (JvmtiExport::can_force_early_return()) {
418 Label L;
419 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
420 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
421
422 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
423 testptr(tmp, tmp);
424 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
425
426 // Initiate earlyret handling only if it is not already being processed.
427 // If the flag has the earlyret_processing bit set, it means that this code
428 // is called *during* earlyret handling - we don't want to reenter.
429 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
430 cmpl(tmp, JvmtiThreadState::earlyret_pending);
431 jcc(Assembler::notEqual, L);
432
433 // Call Interpreter::remove_activation_early_entry() to get the address of the
434 // same-named entrypoint in the generated interpreter code.
435 NOT_LP64(get_thread(java_thread);)
436 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
437 #ifdef _LP64
438 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
439 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
440 #else
441 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
442 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
443 #endif // _LP64
444 jmp(rax);
445 bind(L);
446 NOT_LP64(get_thread(java_thread);)
447 }
448 }
449
450 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
451 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
452 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
453 bswapl(reg);
454 shrl(reg, 16);
455 }
456
457 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
458 int bcp_offset,
459 size_t index_size) {
460 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
461 if (index_size == sizeof(u2)) {
462 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
463 } else if (index_size == sizeof(u4)) {
464 movl(index, Address(_bcp_register, bcp_offset));
465 // Check if the secondary index definition is still ~x, otherwise
466 // we have to change the following assembler code to calculate the
467 // plain index.
468 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
469 notl(index); // convert to plain index
470 } else if (index_size == sizeof(u1)) {
471 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
472 } else {
473 ShouldNotReachHere();
474 }
475 }
476
477 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
478 Register index,
479 int bcp_offset,
480 size_t index_size) {
481 assert_different_registers(cache, index);
482 get_cache_index_at_bcp(index, bcp_offset, index_size);
483 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
484 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
485 // convert from field index to ConstantPoolCacheEntry index
486 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
487 shll(index, 2);
488 }
489
490 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
491 Register index,
492 Register bytecode,
493 int byte_no,
494 int bcp_offset,
495 size_t index_size) {
496 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
497 // We use a 32-bit load here since the layout of 64-bit words on
498 // little-endian machines allow us that.
499 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
500 const int shift_count = (1 + byte_no) * BitsPerByte;
501 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
502 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
503 "correct shift count");
504 shrl(bytecode, shift_count);
505 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
506 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
507 }
508
509 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
510 Register tmp,
511 int bcp_offset,
512 size_t index_size) {
513 assert_different_registers(cache, tmp);
514
515 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
516 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
517 // convert from field index to ConstantPoolCacheEntry index
518 // and from word offset to byte offset
519 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
520 shll(tmp, 2 + LogBytesPerWord);
521 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
522 // skip past the header
523 addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
524 addptr(cache, tmp); // construct pointer to cache entry
525 }
526
527 // Load object from cpool->resolved_references(index)
528 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
529 Register index,
530 Register tmp) {
531 assert_different_registers(result, index);
532
533 get_constant_pool(result);
534 // load pointer for resolved_references[] objArray
535 movptr(result, Address(result, ConstantPool::cache_offset_in_bytes()));
536 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes()));
537 resolve_oop_handle(result, tmp);
538 load_heap_oop(result, Address(result, index,
539 UseCompressedOops ? Address::times_4 : Address::times_ptr,
540 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
541 }
542
543 // load cpool->resolved_klass_at(index)
544 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
545 Register cpool,
546 Register index) {
547 assert_different_registers(cpool, index);
548
549 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
550 Register resolved_klasses = cpool;
551 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes()));
552 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
553 }
554
555 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
556 Register method,
557 Register cache,
558 Register index) {
559 assert_different_registers(cache, index);
560
561 const int method_offset = in_bytes(
562 ConstantPoolCache::base_offset() +
563 ((byte_no == TemplateTable::f2_byte)
564 ? ConstantPoolCacheEntry::f2_offset()
565 : ConstantPoolCacheEntry::f1_offset()));
566
567 movptr(method, Address(cache, index, Address::times_ptr, method_offset)); // get f1 Method*
568 }
569
570 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
571 // subtype of super_klass.
572 //
573 // Args:
574 // rax: superklass
575 // Rsub_klass: subklass
576 //
577 // Kills:
578 // rcx, rdi
579 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
580 Label& ok_is_subtype) {
581 assert(Rsub_klass != rax, "rax holds superklass");
582 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
583 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
584 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
585 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
586
587 // Profile the not-null value's klass.
588 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
589
590 // Do the check.
591 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
592
593 // Profile the failure of the check.
594 profile_typecheck_failed(rcx); // blows rcx
595 }
596
597
598 #ifndef _LP64
599 void InterpreterMacroAssembler::f2ieee() {
600 if (IEEEPrecision) {
601 fstp_s(Address(rsp, 0));
602 fld_s(Address(rsp, 0));
603 }
604 }
605
606
607 void InterpreterMacroAssembler::d2ieee() {
608 if (IEEEPrecision) {
609 fstp_d(Address(rsp, 0));
610 fld_d(Address(rsp, 0));
611 }
612 }
613 #endif // _LP64
614
615 // Java Expression Stack
616
617 void InterpreterMacroAssembler::pop_ptr(Register r) {
618 pop(r);
619 }
620
621 void InterpreterMacroAssembler::push_ptr(Register r) {
622 push(r);
623 }
624
625 void InterpreterMacroAssembler::push_i(Register r) {
626 push(r);
627 }
628
629 void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
630 push(r);
631 }
632
633 void InterpreterMacroAssembler::push_f(XMMRegister r) {
634 subptr(rsp, wordSize);
635 movflt(Address(rsp, 0), r);
636 }
637
638 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
639 movflt(r, Address(rsp, 0));
640 addptr(rsp, wordSize);
641 }
642
643 void InterpreterMacroAssembler::push_d(XMMRegister r) {
644 subptr(rsp, 2 * wordSize);
645 movdbl(Address(rsp, 0), r);
646 }
647
648 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
649 movdbl(r, Address(rsp, 0));
650 addptr(rsp, 2 * Interpreter::stackElementSize);
651 }
652
653 #ifdef _LP64
654 void InterpreterMacroAssembler::pop_i(Register r) {
655 // XXX can't use pop currently, upper half non clean
656 movl(r, Address(rsp, 0));
657 addptr(rsp, wordSize);
658 }
659
660 void InterpreterMacroAssembler::pop_l(Register r) {
661 movq(r, Address(rsp, 0));
662 addptr(rsp, 2 * Interpreter::stackElementSize);
663 }
664
665 void InterpreterMacroAssembler::push_l(Register r) {
666 subptr(rsp, 2 * wordSize);
667 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
668 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
669 }
670
671 void InterpreterMacroAssembler::pop(TosState state) {
672 switch (state) {
673 case atos: pop_ptr(); break;
674 case btos:
675 case ztos:
676 case ctos:
677 case stos:
678 case itos: pop_i(); break;
679 case ltos: pop_l(); break;
680 case ftos: pop_f(xmm0); break;
681 case dtos: pop_d(xmm0); break;
682 case vtos: /* nothing to do */ break;
683 default: ShouldNotReachHere();
684 }
685 interp_verify_oop(rax, state);
686 }
687
688 void InterpreterMacroAssembler::push(TosState state) {
689 interp_verify_oop(rax, state);
690 switch (state) {
691 case atos: push_ptr(); break;
692 case btos:
693 case ztos:
694 case ctos:
695 case stos:
696 case itos: push_i(); break;
697 case ltos: push_l(); break;
698 case ftos: push_f(xmm0); break;
699 case dtos: push_d(xmm0); break;
700 case vtos: /* nothing to do */ break;
701 default : ShouldNotReachHere();
702 }
703 }
704 #else
705 void InterpreterMacroAssembler::pop_i(Register r) {
706 pop(r);
707 }
708
709 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
710 pop(lo);
711 pop(hi);
712 }
713
714 void InterpreterMacroAssembler::pop_f() {
715 fld_s(Address(rsp, 0));
716 addptr(rsp, 1 * wordSize);
717 }
718
719 void InterpreterMacroAssembler::pop_d() {
720 fld_d(Address(rsp, 0));
721 addptr(rsp, 2 * wordSize);
722 }
723
724
725 void InterpreterMacroAssembler::pop(TosState state) {
726 switch (state) {
727 case atos: pop_ptr(rax); break;
728 case btos: // fall through
729 case ztos: // fall through
730 case ctos: // fall through
731 case stos: // fall through
732 case itos: pop_i(rax); break;
733 case ltos: pop_l(rax, rdx); break;
734 case ftos:
735 if (UseSSE >= 1) {
736 pop_f(xmm0);
737 } else {
738 pop_f();
739 }
740 break;
741 case dtos:
742 if (UseSSE >= 2) {
743 pop_d(xmm0);
744 } else {
745 pop_d();
746 }
747 break;
748 case vtos: /* nothing to do */ break;
749 default : ShouldNotReachHere();
750 }
751 interp_verify_oop(rax, state);
752 }
753
754
755 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
756 push(hi);
757 push(lo);
758 }
759
760 void InterpreterMacroAssembler::push_f() {
761 // Do not schedule for no AGI! Never write beyond rsp!
762 subptr(rsp, 1 * wordSize);
763 fstp_s(Address(rsp, 0));
764 }
765
766 void InterpreterMacroAssembler::push_d() {
767 // Do not schedule for no AGI! Never write beyond rsp!
768 subptr(rsp, 2 * wordSize);
769 fstp_d(Address(rsp, 0));
770 }
771
772
773 void InterpreterMacroAssembler::push(TosState state) {
774 interp_verify_oop(rax, state);
775 switch (state) {
776 case atos: push_ptr(rax); break;
777 case btos: // fall through
778 case ztos: // fall through
779 case ctos: // fall through
780 case stos: // fall through
781 case itos: push_i(rax); break;
782 case ltos: push_l(rax, rdx); break;
783 case ftos:
784 if (UseSSE >= 1) {
785 push_f(xmm0);
786 } else {
787 push_f();
788 }
789 break;
790 case dtos:
791 if (UseSSE >= 2) {
792 push_d(xmm0);
793 } else {
794 push_d();
795 }
796 break;
797 case vtos: /* nothing to do */ break;
798 default : ShouldNotReachHere();
799 }
800 }
801 #endif // _LP64
802
803
804 // Helpers for swap and dup
805 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
806 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
807 }
808
809 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
810 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
811 }
812
813
814 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
815 // set sender sp
816 lea(_bcp_register, Address(rsp, wordSize));
817 // record last_sp
818 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
819 }
820
821
822 // Jump to from_interpreted entry of a call unless single stepping is possible
823 // in this thread in which case we must call the i2i entry
824 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
825 prepare_to_jump_from_interpreted();
826
827 if (JvmtiExport::can_post_interpreter_events()) {
828 Label run_compiled_code;
829 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
830 // compiled code in threads for which the event is enabled. Check here for
831 // interp_only_mode if these events CAN be enabled.
832 // interp_only is an int, on little endian it is sufficient to test the byte only
833 // Is a cmpl faster?
834 LP64_ONLY(temp = r15_thread;)
835 NOT_LP64(get_thread(temp);)
836 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
837 jccb(Assembler::zero, run_compiled_code);
838 jmp(Address(method, Method::interpreter_entry_offset()));
839 bind(run_compiled_code);
840 }
841
842 jmp(Address(method, Method::from_interpreted_offset()));
843 }
844
845 // The following two routines provide a hook so that an implementation
846 // can schedule the dispatch in two parts. x86 does not do this.
847 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
848 // Nothing x86 specific to be done here
849 }
850
851 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
852 dispatch_next(state, step);
853 }
854
855 void InterpreterMacroAssembler::dispatch_base(TosState state,
856 address* table,
857 bool verifyoop,
858 bool generate_poll) {
859 verify_FPU(1, state);
860 if (VerifyActivationFrameSize) {
861 Label L;
862 mov(rcx, rbp);
863 subptr(rcx, rsp);
864 int32_t min_frame_size =
865 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
866 wordSize;
867 cmpptr(rcx, (int32_t)min_frame_size);
868 jcc(Assembler::greaterEqual, L);
869 stop("broken stack frame");
870 bind(L);
871 }
872 if (verifyoop) {
873 interp_verify_oop(rax, state);
874 }
875
876 address* const safepoint_table = Interpreter::safept_table(state);
877 #ifdef _LP64
878 Label no_safepoint, dispatch;
879 if (table != safepoint_table && generate_poll) {
880 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
881 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
882
883 jccb(Assembler::zero, no_safepoint);
884 lea(rscratch1, ExternalAddress((address)safepoint_table));
885 jmpb(dispatch);
886 }
887
888 bind(no_safepoint);
889 lea(rscratch1, ExternalAddress((address)table));
890 bind(dispatch);
891 jmp(Address(rscratch1, rbx, Address::times_8));
892
893 #else
894 Address index(noreg, rbx, Address::times_ptr);
895 if (table != safepoint_table && generate_poll) {
896 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
897 Label no_safepoint;
898 const Register thread = rcx;
899 get_thread(thread);
900 testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
901
902 jccb(Assembler::zero, no_safepoint);
903 ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
904 jump(dispatch_addr);
905 bind(no_safepoint);
906 }
907
908 {
909 ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
910 jump(dispatch_addr);
911 }
912 #endif // _LP64
913 }
914
915 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
916 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
917 }
918
919 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
920 dispatch_base(state, Interpreter::normal_table(state));
921 }
922
923 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
924 dispatch_base(state, Interpreter::normal_table(state), false);
925 }
926
927
928 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
929 // load next bytecode (load before advancing _bcp_register to prevent AGI)
930 load_unsigned_byte(rbx, Address(_bcp_register, step));
931 // advance _bcp_register
932 increment(_bcp_register, step);
933 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
934 }
935
936 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
937 // load current bytecode
938 load_unsigned_byte(rbx, Address(_bcp_register, 0));
939 dispatch_base(state, table);
940 }
941
942 void InterpreterMacroAssembler::narrow(Register result) {
943
944 // Get method->_constMethod->_result_type
945 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
946 movptr(rcx, Address(rcx, Method::const_offset()));
947 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
948
949 Label done, notBool, notByte, notChar;
950
951 // common case first
952 cmpl(rcx, T_INT);
953 jcc(Assembler::equal, done);
954
955 // mask integer result to narrower return type.
956 cmpl(rcx, T_BOOLEAN);
957 jcc(Assembler::notEqual, notBool);
958 andl(result, 0x1);
959 jmp(done);
960
961 bind(notBool);
962 cmpl(rcx, T_BYTE);
963 jcc(Assembler::notEqual, notByte);
964 LP64_ONLY(movsbl(result, result);)
965 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
966 NOT_LP64(sarl(result, 24);) // and sign-extend byte
967 jmp(done);
968
969 bind(notByte);
970 cmpl(rcx, T_CHAR);
971 jcc(Assembler::notEqual, notChar);
972 LP64_ONLY(movzwl(result, result);)
973 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
974 jmp(done);
975
976 bind(notChar);
977 // cmpl(rcx, T_SHORT); // all that's left
978 // jcc(Assembler::notEqual, done);
979 LP64_ONLY(movswl(result, result);)
980 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
981 NOT_LP64(sarl(result, 16);) // and sign-extend short
982
983 // Nothing to do for T_INT
984 bind(done);
985 }
986
987 // remove activation
988 //
989 // Apply stack watermark barrier.
990 // Unlock the receiver if this is a synchronized method.
991 // Unlock any Java monitors from syncronized blocks.
992 // Remove the activation from the stack.
993 //
994 // If there are locked Java monitors
995 // If throw_monitor_exception
996 // throws IllegalMonitorStateException
997 // Else if install_monitor_exception
998 // installs IllegalMonitorStateException
999 // Else
1000 // no error processing
1001 void InterpreterMacroAssembler::remove_activation(
1002 TosState state,
1003 Register ret_addr,
1004 bool throw_monitor_exception,
1005 bool install_monitor_exception,
1006 bool notify_jvmdi) {
1007 // Note: Registers rdx xmm0 may be in use for the
1008 // result check if synchronized method
1009 Label unlocked, unlock, no_unlock;
1010
1011 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1012 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
1013 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
1014 // monitor pointers need different register
1015 // because rdx may have the result in it
1016 NOT_LP64(get_thread(rthread);)
1017
1018 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
1019 // that would normally not be safe to use. Such bad returns into unsafe territory of
1020 // the stack, will call InterpreterRuntime::at_unwind.
1021 Label slow_path;
1022 Label fast_path;
1023 safepoint_poll(slow_path, rthread, true /* at_return */, false /* in_nmethod */);
1024 jmp(fast_path);
1025 bind(slow_path);
1026 push(state);
1027 set_last_Java_frame(rthread, noreg, rbp, (address)pc());
1028 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
1029 NOT_LP64(get_thread(rthread);) // call_VM clobbered it, restore
1030 reset_last_Java_frame(rthread, true);
1031 pop(state);
1032 bind(fast_path);
1033
1034 // get the value of _do_not_unlock_if_synchronized into rdx
1035 const Address do_not_unlock_if_synchronized(rthread,
1036 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1037 movbool(rbx, do_not_unlock_if_synchronized);
1038 movbool(do_not_unlock_if_synchronized, false); // reset the flag
1039
1040 // get method access flags
1041 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
1042 movl(rcx, Address(rcx, Method::access_flags_offset()));
1043 testl(rcx, JVM_ACC_SYNCHRONIZED);
1044 jcc(Assembler::zero, unlocked);
1045
1046 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
1047 // is set.
1048 testbool(rbx);
1049 jcc(Assembler::notZero, no_unlock);
1050
1051 // unlock monitor
1052 push(state); // save result
1053
1054 // BasicObjectLock will be first in list, since this is a
1055 // synchronized method. However, need to check that the object has
1056 // not been unlocked by an explicit monitorexit bytecode.
1057 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
1058 wordSize - (int) sizeof(BasicObjectLock));
1059 // We use c_rarg1/rdx so that if we go slow path it will be the correct
1060 // register for unlock_object to pass to VM directly
1061 lea(robj, monitor); // address of first monitor
1062
1063 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
1064 testptr(rax, rax);
1065 jcc(Assembler::notZero, unlock);
1066
1067 pop(state);
1068 if (throw_monitor_exception) {
1069 // Entry already unlocked, need to throw exception
1070 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
1071 call_VM(noreg, CAST_FROM_FN_PTR(address,
1072 InterpreterRuntime::throw_illegal_monitor_state_exception));
1073 should_not_reach_here();
1074 } else {
1075 // Monitor already unlocked during a stack unroll. If requested,
1076 // install an illegal_monitor_state_exception. Continue with
1077 // stack unrolling.
1078 if (install_monitor_exception) {
1079 NOT_LP64(empty_FPU_stack();)
1080 call_VM(noreg, CAST_FROM_FN_PTR(address,
1081 InterpreterRuntime::new_illegal_monitor_state_exception));
1082 }
1083 jmp(unlocked);
1084 }
1085
1086 bind(unlock);
1087 unlock_object(robj);
1088 pop(state);
1089
1090 // Check that for block-structured locking (i.e., that all locked
1091 // objects has been unlocked)
1092 bind(unlocked);
1093
1094 // rax, rdx: Might contain return value
1095
1096 // Check that all monitors are unlocked
1097 {
1098 Label loop, exception, entry, restart;
1099 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1100 const Address monitor_block_top(
1101 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1102 const Address monitor_block_bot(
1103 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1104
1105 bind(restart);
1106 // We use c_rarg1 so that if we go slow path it will be the correct
1107 // register for unlock_object to pass to VM directly
1108 movptr(rmon, monitor_block_top); // points to current entry, starting
1109 // with top-most entry
1110 lea(rbx, monitor_block_bot); // points to word before bottom of
1111 // monitor block
1112 jmp(entry);
1113
1114 // Entry already locked, need to throw exception
1115 bind(exception);
1116
1117 if (throw_monitor_exception) {
1118 // Throw exception
1119 NOT_LP64(empty_FPU_stack();)
1120 MacroAssembler::call_VM(noreg,
1121 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1122 throw_illegal_monitor_state_exception));
1123 should_not_reach_here();
1124 } else {
1125 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1126 // Unlock does not block, so don't have to worry about the frame.
1127 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1128
1129 push(state);
1130 mov(robj, rmon); // nop if robj and rmon are the same
1131 unlock_object(robj);
1132 pop(state);
1133
1134 if (install_monitor_exception) {
1135 NOT_LP64(empty_FPU_stack();)
1136 call_VM(noreg, CAST_FROM_FN_PTR(address,
1137 InterpreterRuntime::
1138 new_illegal_monitor_state_exception));
1139 }
1140
1141 jmp(restart);
1142 }
1143
1144 bind(loop);
1145 // check if current entry is used
1146 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
1147 jcc(Assembler::notEqual, exception);
1148
1149 addptr(rmon, entry_size); // otherwise advance to next entry
1150 bind(entry);
1151 cmpptr(rmon, rbx); // check if bottom reached
1152 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1153 }
1154
1155 bind(no_unlock);
1156
1157 // jvmti support
1158 if (notify_jvmdi) {
1159 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1160 } else {
1161 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1162 }
1163
1164 // remove activation
1165 // get sender sp
1166 movptr(rbx,
1167 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1168 if (StackReservedPages > 0) {
1169 // testing if reserved zone needs to be re-enabled
1170 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1171 Label no_reserved_zone_enabling;
1172
1173 NOT_LP64(get_thread(rthread);)
1174
1175 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
1176 jcc(Assembler::equal, no_reserved_zone_enabling);
1177
1178 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1179 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1180
1181 call_VM_leaf(
1182 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1183 call_VM(noreg, CAST_FROM_FN_PTR(address,
1184 InterpreterRuntime::throw_delayed_StackOverflowError));
1185 should_not_reach_here();
1186
1187 bind(no_reserved_zone_enabling);
1188 }
1189 leave(); // remove frame anchor
1190 pop(ret_addr); // get return address
1191 mov(rsp, rbx); // set sp to sender sp
1192 }
1193
1194 void InterpreterMacroAssembler::get_method_counters(Register method,
1195 Register mcs, Label& skip) {
1196 Label has_counters;
1197 movptr(mcs, Address(method, Method::method_counters_offset()));
1198 testptr(mcs, mcs);
1199 jcc(Assembler::notZero, has_counters);
1200 call_VM(noreg, CAST_FROM_FN_PTR(address,
1201 InterpreterRuntime::build_method_counters), method);
1202 movptr(mcs, Address(method,Method::method_counters_offset()));
1203 testptr(mcs, mcs);
1204 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1205 bind(has_counters);
1206 }
1207
1208
1209 // Lock object
1210 //
1211 // Args:
1212 // rdx, c_rarg1: BasicObjectLock to be used for locking
1213 //
1214 // Kills:
1215 // rax, rbx
1216 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1217 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1218 "The argument is only for looks. It must be c_rarg1");
1219
1220 if (LockingMode == LM_MONITOR) {
1221 call_VM(noreg,
1222 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1223 lock_reg);
1224 } else {
1225 Label done;
1226
1227 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1228 const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a
1229 // problematic case where tmp_reg = no_reg.
1230 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1231 const Register rklass_decode_tmp = LP64_ONLY(rscratch1) NOT_LP64(noreg);
1232
1233 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
1234 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
1235 const int mark_offset = lock_offset +
1236 BasicLock::displaced_header_offset_in_bytes();
1237
1238 Label slow_case;
1239
1240 // Load object pointer into obj_reg
1241 movptr(obj_reg, Address(lock_reg, obj_offset));
1242
1243 if (DiagnoseSyncOnValueBasedClasses != 0) {
1244 load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
1245 movl(tmp_reg, Address(tmp_reg, Klass::access_flags_offset()));
1246 testl(tmp_reg, JVM_ACC_IS_VALUE_BASED_CLASS);
1247 jcc(Assembler::notZero, slow_case);
1248 }
1249
1250 if (UseBiasedLocking) {
1251 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, rklass_decode_tmp, false, done, &slow_case);
1252 }
1253
1254 if (LockingMode == LM_LIGHTWEIGHT) {
1255 #ifdef _LP64
1256 const Register thread = r15_thread;
1257 #else
1258 const Register thread = lock_reg;
1259 get_thread(thread);
1260 #endif
1261 lightweight_lock(obj_reg, swap_reg, thread, tmp_reg, slow_case);
1262 jmp(done);
1263 } else {
1264 // Load immediate 1 into swap_reg %rax
1265 movl(swap_reg, (int32_t)1);
1266
1267 // Load (object->mark() | 1) into swap_reg %rax
1268 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1269
1270 // Save (object->mark() | 1) into BasicLock's displaced header
1271 movptr(Address(lock_reg, mark_offset), swap_reg);
1272
1273 assert(lock_offset == 0,
1274 "displaced header must be first word in BasicObjectLock");
1275
1276 lock();
1277 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1278 if (PrintBiasedLockingStatistics) {
1279 cond_inc32(Assembler::zero,
1280 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1281 }
1282 jcc(Assembler::zero, done);
1283
1284 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1285
1286 // Fast check for recursive lock.
1287 //
1288 // Can apply the optimization only if this is a stack lock
1289 // allocated in this thread. For efficiency, we can focus on
1290 // recently allocated stack locks (instead of reading the stack
1291 // base and checking whether 'mark' points inside the current
1292 // thread stack):
1293 // 1) (mark & zero_bits) == 0, and
1294 // 2) rsp <= mark < mark + os::pagesize()
1295 //
1296 // Warning: rsp + os::pagesize can overflow the stack base. We must
1297 // neither apply the optimization for an inflated lock allocated
1298 // just above the thread stack (this is why condition 1 matters)
1299 // nor apply the optimization if the stack lock is inside the stack
1300 // of another thread. The latter is avoided even in case of overflow
1301 // because we have guard pages at the end of all stacks. Hence, if
1302 // we go over the stack base and hit the stack of another thread,
1303 // this should not be in a writeable area that could contain a
1304 // stack lock allocated by that thread. As a consequence, a stack
1305 // lock less than page size away from rsp is guaranteed to be
1306 // owned by the current thread.
1307 //
1308 // These 3 tests can be done by evaluating the following
1309 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1310 // assuming both stack pointer and pagesize have their
1311 // least significant bits clear.
1312 // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
1313 subptr(swap_reg, rsp);
1314 andptr(swap_reg, zero_bits - os::vm_page_size());
1315
1316 // Save the test result, for recursive case, the result is zero
1317 movptr(Address(lock_reg, mark_offset), swap_reg);
1318
1319 if (PrintBiasedLockingStatistics) {
1320 cond_inc32(Assembler::zero,
1321 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1322 }
1323 jcc(Assembler::zero, done);
1324 }
1325 bind(slow_case);
1326
1327 // Call the runtime routine for slow case
1328 if (LockingMode == LM_LIGHTWEIGHT) {
1329 call_VM(noreg,
1330 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj),
1331 obj_reg);
1332 } else {
1333 call_VM(noreg,
1334 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1335 lock_reg);
1336 }
1337 bind(done);
1338 }
1339 }
1340
1341
1342 // Unlocks an object. Used in monitorexit bytecode and
1343 // remove_activation. Throws an IllegalMonitorException if object is
1344 // not locked by current thread.
1345 //
1346 // Args:
1347 // rdx, c_rarg1: BasicObjectLock for lock
1348 //
1349 // Kills:
1350 // rax
1351 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1352 // rscratch1 (scratch reg)
1353 // rax, rbx, rcx, rdx
1354 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1355 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1356 "The argument is only for looks. It must be c_rarg1");
1357
1358 if (LockingMode == LM_MONITOR) {
1359 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1360 } else {
1361 Label done, slow_case;
1362
1363 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1364 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1365 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1366
1367 save_bcp(); // Save in case of exception
1368
1369 if (LockingMode != LM_LIGHTWEIGHT) {
1370 // Convert from BasicObjectLock structure to object and BasicLock
1371 // structure Store the BasicLock address into %rax
1372 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
1373 }
1374
1375 // Load oop into obj_reg(%c_rarg3)
1376 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
1377
1378 // Free entry
1379 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
1380
1381 if (LockingMode == LM_LIGHTWEIGHT) {
1382 #ifdef _LP64
1383 lightweight_unlock(obj_reg, swap_reg, r15_thread, header_reg, slow_case);
1384 #else
1385 // This relies on the implementation of lightweight_unlock being able to handle
1386 // that the reg_rax and thread Register parameters may alias each other.
1387 get_thread(swap_reg);
1388 lightweight_unlock(obj_reg, swap_reg, swap_reg, header_reg, slow_case);
1389 #endif
1390 jmp(done);
1391 } else {
1392 if (UseBiasedLocking) {
1393 biased_locking_exit(obj_reg, header_reg, done);
1394 }
1395
1396 // Load the old header from BasicLock structure
1397 movptr(header_reg, Address(swap_reg,
1398 BasicLock::displaced_header_offset_in_bytes()));
1399
1400 // Test for recursion
1401 testptr(header_reg, header_reg);
1402
1403 // zero for recursive case
1404 jcc(Assembler::zero, done);
1405
1406 // Atomic swap back the old header
1407 lock();
1408 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1409
1410 // zero for simple unlock of a stack-lock case
1411 jcc(Assembler::zero, done);
1412 }
1413
1414 bind(slow_case);
1415 // Call the runtime routine for slow case.
1416 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), obj_reg); // restore obj
1417 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1418
1419 bind(done);
1420
1421 restore_bcp();
1422 }
1423 }
1424
1425 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1426 Label& zero_continue) {
1427 assert(ProfileInterpreter, "must be profiling interpreter");
1428 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1429 testptr(mdp, mdp);
1430 jcc(Assembler::zero, zero_continue);
1431 }
1432
1433
1434 // Set the method data pointer for the current bcp.
1435 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1436 assert(ProfileInterpreter, "must be profiling interpreter");
1437 Label set_mdp;
1438 push(rax);
1439 push(rbx);
1440
1441 get_method(rbx);
1442 // Test MDO to avoid the call if it is NULL.
1443 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1444 testptr(rax, rax);
1445 jcc(Assembler::zero, set_mdp);
1446 // rbx: method
1447 // _bcp_register: bcp
1448 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1449 // rax: mdi
1450 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1451 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1452 addptr(rbx, in_bytes(MethodData::data_offset()));
1453 addptr(rax, rbx);
1454 bind(set_mdp);
1455 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1456 pop(rbx);
1457 pop(rax);
1458 }
1459
1460 void InterpreterMacroAssembler::verify_method_data_pointer() {
1461 assert(ProfileInterpreter, "must be profiling interpreter");
1462 #ifdef ASSERT
1463 Label verify_continue;
1464 push(rax);
1465 push(rbx);
1466 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1467 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1468 push(arg3_reg);
1469 push(arg2_reg);
1470 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1471 get_method(rbx);
1472
1473 // If the mdp is valid, it will point to a DataLayout header which is
1474 // consistent with the bcp. The converse is highly probable also.
1475 load_unsigned_short(arg2_reg,
1476 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1477 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1478 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1479 cmpptr(arg2_reg, _bcp_register);
1480 jcc(Assembler::equal, verify_continue);
1481 // rbx: method
1482 // _bcp_register: bcp
1483 // c_rarg3: mdp
1484 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1485 rbx, _bcp_register, arg3_reg);
1486 bind(verify_continue);
1487 pop(arg2_reg);
1488 pop(arg3_reg);
1489 pop(rbx);
1490 pop(rax);
1491 #endif // ASSERT
1492 }
1493
1494
1495 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1496 int constant,
1497 Register value) {
1498 assert(ProfileInterpreter, "must be profiling interpreter");
1499 Address data(mdp_in, constant);
1500 movptr(data, value);
1501 }
1502
1503
1504 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1505 int constant,
1506 bool decrement) {
1507 // Counter address
1508 Address data(mdp_in, constant);
1509
1510 increment_mdp_data_at(data, decrement);
1511 }
1512
1513 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1514 bool decrement) {
1515 assert(ProfileInterpreter, "must be profiling interpreter");
1516 // %%% this does 64bit counters at best it is wasting space
1517 // at worst it is a rare bug when counters overflow
1518
1519 if (decrement) {
1520 // Decrement the register. Set condition codes.
1521 addptr(data, (int32_t) -DataLayout::counter_increment);
1522 // If the decrement causes the counter to overflow, stay negative
1523 Label L;
1524 jcc(Assembler::negative, L);
1525 addptr(data, (int32_t) DataLayout::counter_increment);
1526 bind(L);
1527 } else {
1528 assert(DataLayout::counter_increment == 1,
1529 "flow-free idiom only works with 1");
1530 // Increment the register. Set carry flag.
1531 addptr(data, DataLayout::counter_increment);
1532 // If the increment causes the counter to overflow, pull back by 1.
1533 sbbptr(data, (int32_t)0);
1534 }
1535 }
1536
1537
1538 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1539 Register reg,
1540 int constant,
1541 bool decrement) {
1542 Address data(mdp_in, reg, Address::times_1, constant);
1543
1544 increment_mdp_data_at(data, decrement);
1545 }
1546
1547 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1548 int flag_byte_constant) {
1549 assert(ProfileInterpreter, "must be profiling interpreter");
1550 int header_offset = in_bytes(DataLayout::flags_offset());
1551 int header_bits = flag_byte_constant;
1552 // Set the flag
1553 orb(Address(mdp_in, header_offset), header_bits);
1554 }
1555
1556
1557
1558 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1559 int offset,
1560 Register value,
1561 Register test_value_out,
1562 Label& not_equal_continue) {
1563 assert(ProfileInterpreter, "must be profiling interpreter");
1564 if (test_value_out == noreg) {
1565 cmpptr(value, Address(mdp_in, offset));
1566 } else {
1567 // Put the test value into a register, so caller can use it:
1568 movptr(test_value_out, Address(mdp_in, offset));
1569 cmpptr(test_value_out, value);
1570 }
1571 jcc(Assembler::notEqual, not_equal_continue);
1572 }
1573
1574
1575 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1576 int offset_of_disp) {
1577 assert(ProfileInterpreter, "must be profiling interpreter");
1578 Address disp_address(mdp_in, offset_of_disp);
1579 addptr(mdp_in, disp_address);
1580 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1581 }
1582
1583
1584 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1585 Register reg,
1586 int offset_of_disp) {
1587 assert(ProfileInterpreter, "must be profiling interpreter");
1588 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1589 addptr(mdp_in, disp_address);
1590 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1591 }
1592
1593
1594 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1595 int constant) {
1596 assert(ProfileInterpreter, "must be profiling interpreter");
1597 addptr(mdp_in, constant);
1598 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1599 }
1600
1601
1602 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1603 assert(ProfileInterpreter, "must be profiling interpreter");
1604 push(return_bci); // save/restore across call_VM
1605 call_VM(noreg,
1606 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1607 return_bci);
1608 pop(return_bci);
1609 }
1610
1611
1612 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1613 Register bumped_count) {
1614 if (ProfileInterpreter) {
1615 Label profile_continue;
1616
1617 // If no method data exists, go to profile_continue.
1618 // Otherwise, assign to mdp
1619 test_method_data_pointer(mdp, profile_continue);
1620
1621 // We are taking a branch. Increment the taken count.
1622 // We inline increment_mdp_data_at to return bumped_count in a register
1623 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1624 Address data(mdp, in_bytes(JumpData::taken_offset()));
1625 movptr(bumped_count, data);
1626 assert(DataLayout::counter_increment == 1,
1627 "flow-free idiom only works with 1");
1628 addptr(bumped_count, DataLayout::counter_increment);
1629 sbbptr(bumped_count, 0);
1630 movptr(data, bumped_count); // Store back out
1631
1632 // The method data pointer needs to be updated to reflect the new target.
1633 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1634 bind(profile_continue);
1635 }
1636 }
1637
1638
1639 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1640 if (ProfileInterpreter) {
1641 Label profile_continue;
1642
1643 // If no method data exists, go to profile_continue.
1644 test_method_data_pointer(mdp, profile_continue);
1645
1646 // We are taking a branch. Increment the not taken count.
1647 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1648
1649 // The method data pointer needs to be updated to correspond to
1650 // the next bytecode
1651 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1652 bind(profile_continue);
1653 }
1654 }
1655
1656 void InterpreterMacroAssembler::profile_call(Register mdp) {
1657 if (ProfileInterpreter) {
1658 Label profile_continue;
1659
1660 // If no method data exists, go to profile_continue.
1661 test_method_data_pointer(mdp, profile_continue);
1662
1663 // We are making a call. Increment the count.
1664 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1665
1666 // The method data pointer needs to be updated to reflect the new target.
1667 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1668 bind(profile_continue);
1669 }
1670 }
1671
1672
1673 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1674 if (ProfileInterpreter) {
1675 Label profile_continue;
1676
1677 // If no method data exists, go to profile_continue.
1678 test_method_data_pointer(mdp, profile_continue);
1679
1680 // We are making a call. Increment the count.
1681 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1682
1683 // The method data pointer needs to be updated to reflect the new target.
1684 update_mdp_by_constant(mdp,
1685 in_bytes(VirtualCallData::
1686 virtual_call_data_size()));
1687 bind(profile_continue);
1688 }
1689 }
1690
1691
1692 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1693 Register mdp,
1694 Register reg2,
1695 bool receiver_can_be_null) {
1696 if (ProfileInterpreter) {
1697 Label profile_continue;
1698
1699 // If no method data exists, go to profile_continue.
1700 test_method_data_pointer(mdp, profile_continue);
1701
1702 Label skip_receiver_profile;
1703 if (receiver_can_be_null) {
1704 Label not_null;
1705 testptr(receiver, receiver);
1706 jccb(Assembler::notZero, not_null);
1707 // We are making a call. Increment the count for null receiver.
1708 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1709 jmp(skip_receiver_profile);
1710 bind(not_null);
1711 }
1712
1713 // Record the receiver type.
1714 record_klass_in_profile(receiver, mdp, reg2, true);
1715 bind(skip_receiver_profile);
1716
1717 // The method data pointer needs to be updated to reflect the new target.
1718 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1719 bind(profile_continue);
1720 }
1721 }
1722
1723 // This routine creates a state machine for updating the multi-row
1724 // type profile at a virtual call site (or other type-sensitive bytecode).
1725 // The machine visits each row (of receiver/count) until the receiver type
1726 // is found, or until it runs out of rows. At the same time, it remembers
1727 // the location of the first empty row. (An empty row records null for its
1728 // receiver, and can be allocated for a newly-observed receiver type.)
1729 // Because there are two degrees of freedom in the state, a simple linear
1730 // search will not work; it must be a decision tree. Hence this helper
1731 // function is recursive, to generate the required tree structured code.
1732 // It's the interpreter, so we are trading off code space for speed.
1733 // See below for example code.
1734 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1735 Register receiver, Register mdp,
1736 Register reg2, int start_row,
1737 Label& done, bool is_virtual_call) {
1738 if (TypeProfileWidth == 0) {
1739 if (is_virtual_call) {
1740 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1741 }
1742 #if INCLUDE_JVMCI
1743 else if (EnableJVMCI) {
1744 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1745 }
1746 #endif // INCLUDE_JVMCI
1747 } else {
1748 int non_profiled_offset = -1;
1749 if (is_virtual_call) {
1750 non_profiled_offset = in_bytes(CounterData::count_offset());
1751 }
1752 #if INCLUDE_JVMCI
1753 else if (EnableJVMCI) {
1754 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1755 }
1756 #endif // INCLUDE_JVMCI
1757
1758 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1759 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1760 }
1761 }
1762
1763 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1764 Register reg2, int start_row, Label& done, int total_rows,
1765 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1766 int non_profiled_offset) {
1767 int last_row = total_rows - 1;
1768 assert(start_row <= last_row, "must be work left to do");
1769 // Test this row for both the item and for null.
1770 // Take any of three different outcomes:
1771 // 1. found item => increment count and goto done
1772 // 2. found null => keep looking for case 1, maybe allocate this cell
1773 // 3. found something else => keep looking for cases 1 and 2
1774 // Case 3 is handled by a recursive call.
1775 for (int row = start_row; row <= last_row; row++) {
1776 Label next_test;
1777 bool test_for_null_also = (row == start_row);
1778
1779 // See if the item is item[n].
1780 int item_offset = in_bytes(item_offset_fn(row));
1781 test_mdp_data_at(mdp, item_offset, item,
1782 (test_for_null_also ? reg2 : noreg),
1783 next_test);
1784 // (Reg2 now contains the item from the CallData.)
1785
1786 // The item is item[n]. Increment count[n].
1787 int count_offset = in_bytes(item_count_offset_fn(row));
1788 increment_mdp_data_at(mdp, count_offset);
1789 jmp(done);
1790 bind(next_test);
1791
1792 if (test_for_null_also) {
1793 // Failed the equality check on item[n]... Test for null.
1794 testptr(reg2, reg2);
1795 if (start_row == last_row) {
1796 // The only thing left to do is handle the null case.
1797 if (non_profiled_offset >= 0) {
1798 Label found_null;
1799 jccb(Assembler::zero, found_null);
1800 // Item did not match any saved item and there is no empty row for it.
1801 // Increment total counter to indicate polymorphic case.
1802 increment_mdp_data_at(mdp, non_profiled_offset);
1803 jmp(done);
1804 bind(found_null);
1805 } else {
1806 jcc(Assembler::notZero, done);
1807 }
1808 break;
1809 }
1810 Label found_null;
1811 // Since null is rare, make it be the branch-taken case.
1812 jcc(Assembler::zero, found_null);
1813
1814 // Put all the "Case 3" tests here.
1815 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1816 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1817
1818 // Found a null. Keep searching for a matching item,
1819 // but remember that this is an empty (unused) slot.
1820 bind(found_null);
1821 }
1822 }
1823
1824 // In the fall-through case, we found no matching item, but we
1825 // observed the item[start_row] is NULL.
1826
1827 // Fill in the item field and increment the count.
1828 int item_offset = in_bytes(item_offset_fn(start_row));
1829 set_mdp_data_at(mdp, item_offset, item);
1830 int count_offset = in_bytes(item_count_offset_fn(start_row));
1831 movl(reg2, DataLayout::counter_increment);
1832 set_mdp_data_at(mdp, count_offset, reg2);
1833 if (start_row > 0) {
1834 jmp(done);
1835 }
1836 }
1837
1838 // Example state machine code for three profile rows:
1839 // // main copy of decision tree, rooted at row[1]
1840 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1841 // if (row[0].rec != NULL) {
1842 // // inner copy of decision tree, rooted at row[1]
1843 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1844 // if (row[1].rec != NULL) {
1845 // // degenerate decision tree, rooted at row[2]
1846 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1847 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1848 // row[2].init(rec); goto done;
1849 // } else {
1850 // // remember row[1] is empty
1851 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1852 // row[1].init(rec); goto done;
1853 // }
1854 // } else {
1855 // // remember row[0] is empty
1856 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1857 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1858 // row[0].init(rec); goto done;
1859 // }
1860 // done:
1861
1862 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1863 Register mdp, Register reg2,
1864 bool is_virtual_call) {
1865 assert(ProfileInterpreter, "must be profiling");
1866 Label done;
1867
1868 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1869
1870 bind (done);
1871 }
1872
1873 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1874 Register mdp) {
1875 if (ProfileInterpreter) {
1876 Label profile_continue;
1877 uint row;
1878
1879 // If no method data exists, go to profile_continue.
1880 test_method_data_pointer(mdp, profile_continue);
1881
1882 // Update the total ret count.
1883 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1884
1885 for (row = 0; row < RetData::row_limit(); row++) {
1886 Label next_test;
1887
1888 // See if return_bci is equal to bci[n]:
1889 test_mdp_data_at(mdp,
1890 in_bytes(RetData::bci_offset(row)),
1891 return_bci, noreg,
1892 next_test);
1893
1894 // return_bci is equal to bci[n]. Increment the count.
1895 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1896
1897 // The method data pointer needs to be updated to reflect the new target.
1898 update_mdp_by_offset(mdp,
1899 in_bytes(RetData::bci_displacement_offset(row)));
1900 jmp(profile_continue);
1901 bind(next_test);
1902 }
1903
1904 update_mdp_for_ret(return_bci);
1905
1906 bind(profile_continue);
1907 }
1908 }
1909
1910
1911 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1912 if (ProfileInterpreter) {
1913 Label profile_continue;
1914
1915 // If no method data exists, go to profile_continue.
1916 test_method_data_pointer(mdp, profile_continue);
1917
1918 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1919
1920 // The method data pointer needs to be updated.
1921 int mdp_delta = in_bytes(BitData::bit_data_size());
1922 if (TypeProfileCasts) {
1923 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1924 }
1925 update_mdp_by_constant(mdp, mdp_delta);
1926
1927 bind(profile_continue);
1928 }
1929 }
1930
1931
1932 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1933 if (ProfileInterpreter && TypeProfileCasts) {
1934 Label profile_continue;
1935
1936 // If no method data exists, go to profile_continue.
1937 test_method_data_pointer(mdp, profile_continue);
1938
1939 int count_offset = in_bytes(CounterData::count_offset());
1940 // Back up the address, since we have already bumped the mdp.
1941 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1942
1943 // *Decrement* the counter. We expect to see zero or small negatives.
1944 increment_mdp_data_at(mdp, count_offset, true);
1945
1946 bind (profile_continue);
1947 }
1948 }
1949
1950
1951 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1952 if (ProfileInterpreter) {
1953 Label profile_continue;
1954
1955 // If no method data exists, go to profile_continue.
1956 test_method_data_pointer(mdp, profile_continue);
1957
1958 // The method data pointer needs to be updated.
1959 int mdp_delta = in_bytes(BitData::bit_data_size());
1960 if (TypeProfileCasts) {
1961 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1962
1963 // Record the object type.
1964 record_klass_in_profile(klass, mdp, reg2, false);
1965 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1966 NOT_LP64(restore_locals();) // Restore EDI
1967 }
1968 update_mdp_by_constant(mdp, mdp_delta);
1969
1970 bind(profile_continue);
1971 }
1972 }
1973
1974
1975 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1976 if (ProfileInterpreter) {
1977 Label profile_continue;
1978
1979 // If no method data exists, go to profile_continue.
1980 test_method_data_pointer(mdp, profile_continue);
1981
1982 // Update the default case count
1983 increment_mdp_data_at(mdp,
1984 in_bytes(MultiBranchData::default_count_offset()));
1985
1986 // The method data pointer needs to be updated.
1987 update_mdp_by_offset(mdp,
1988 in_bytes(MultiBranchData::
1989 default_displacement_offset()));
1990
1991 bind(profile_continue);
1992 }
1993 }
1994
1995
1996 void InterpreterMacroAssembler::profile_switch_case(Register index,
1997 Register mdp,
1998 Register reg2) {
1999 if (ProfileInterpreter) {
2000 Label profile_continue;
2001
2002 // If no method data exists, go to profile_continue.
2003 test_method_data_pointer(mdp, profile_continue);
2004
2005 // Build the base (index * per_case_size_in_bytes()) +
2006 // case_array_offset_in_bytes()
2007 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
2008 imulptr(index, reg2); // XXX l ?
2009 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
2010
2011 // Update the case count
2012 increment_mdp_data_at(mdp,
2013 index,
2014 in_bytes(MultiBranchData::relative_count_offset()));
2015
2016 // The method data pointer needs to be updated.
2017 update_mdp_by_offset(mdp,
2018 index,
2019 in_bytes(MultiBranchData::
2020 relative_displacement_offset()));
2021
2022 bind(profile_continue);
2023 }
2024 }
2025
2026
2027
2028 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
2029 if (state == atos) {
2030 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
2031 }
2032 }
2033
2034 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2035 #ifndef _LP64
2036 if ((state == ftos && UseSSE < 1) ||
2037 (state == dtos && UseSSE < 2)) {
2038 MacroAssembler::verify_FPU(stack_depth);
2039 }
2040 #endif
2041 }
2042
2043 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
2044 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
2045 int increment, Address mask,
2046 Register scratch, bool preloaded,
2047 Condition cond, Label* where) {
2048 if (!preloaded) {
2049 movl(scratch, counter_addr);
2050 }
2051 incrementl(scratch, increment);
2052 movl(counter_addr, scratch);
2053 andl(scratch, mask);
2054 if (where != NULL) {
2055 jcc(cond, *where);
2056 }
2057 }
2058
2059 void InterpreterMacroAssembler::notify_method_entry() {
2060 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2061 // track stack depth. If it is possible to enter interp_only_mode we add
2062 // the code to check if the event should be sent.
2063 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2064 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2065 if (JvmtiExport::can_post_interpreter_events()) {
2066 Label L;
2067 NOT_LP64(get_thread(rthread);)
2068 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2069 testl(rdx, rdx);
2070 jcc(Assembler::zero, L);
2071 call_VM(noreg, CAST_FROM_FN_PTR(address,
2072 InterpreterRuntime::post_method_entry));
2073 bind(L);
2074 }
2075
2076 {
2077 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2078 NOT_LP64(get_thread(rthread);)
2079 get_method(rarg);
2080 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2081 rthread, rarg);
2082 }
2083
2084 // RedefineClasses() tracing support for obsolete method entry
2085 if (log_is_enabled(Trace, redefine, class, obsolete)) {
2086 NOT_LP64(get_thread(rthread);)
2087 get_method(rarg);
2088 call_VM_leaf(
2089 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2090 rthread, rarg);
2091 }
2092 }
2093
2094
2095 void InterpreterMacroAssembler::notify_method_exit(
2096 TosState state, NotifyMethodExitMode mode) {
2097 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2098 // track stack depth. If it is possible to enter interp_only_mode we add
2099 // the code to check if the event should be sent.
2100 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2101 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2102 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2103 Label L;
2104 // Note: frame::interpreter_frame_result has a dependency on how the
2105 // method result is saved across the call to post_method_exit. If this
2106 // is changed then the interpreter_frame_result implementation will
2107 // need to be updated too.
2108
2109 // template interpreter will leave the result on the top of the stack.
2110 push(state);
2111 NOT_LP64(get_thread(rthread);)
2112 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2113 testl(rdx, rdx);
2114 jcc(Assembler::zero, L);
2115 call_VM(noreg,
2116 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2117 bind(L);
2118 pop(state);
2119 }
2120
2121 {
2122 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2123 push(state);
2124 NOT_LP64(get_thread(rthread);)
2125 get_method(rarg);
2126 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2127 rthread, rarg);
2128 pop(state);
2129 }
2130 }