1 /*
2 * Copyright (c) 1997, 2018, 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/markWord.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/method.hpp"
34 #include "oops/valueKlass.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
44 // Implementation of InterpreterMacroAssembler
45
46 void InterpreterMacroAssembler::jump_to_entry(address entry) {
47 assert(entry, "Entry must have been generated by now");
48 jump(RuntimeAddress(entry));
49 }
50
51 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
52 Label update, next, none;
53
54 verify_oop(obj);
55
56 testptr(obj, obj);
57 jccb(Assembler::notZero, update);
58 orptr(mdo_addr, TypeEntries::null_seen);
59 jmpb(next);
60
61 bind(update);
62 load_klass(obj, obj);
63
64 xorptr(obj, mdo_addr);
65 testptr(obj, TypeEntries::type_klass_mask);
66 jccb(Assembler::zero, next); // klass seen before, nothing to
67 // do. The unknown bit may have been
68 // set already but no need to check.
69
70 testptr(obj, TypeEntries::type_unknown);
71 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
72
73 cmpptr(mdo_addr, 0);
74 jccb(Assembler::equal, none);
75 cmpptr(mdo_addr, TypeEntries::null_seen);
76 jccb(Assembler::equal, none);
77 // There is a chance that the checks above (re-reading profiling
78 // data from memory) fail if another thread has just set the
79 // profiling to this obj's klass
80 xorptr(obj, mdo_addr);
81 testptr(obj, TypeEntries::type_klass_mask);
82 jccb(Assembler::zero, next);
83
84 // different than before. Cannot keep accurate profile.
85 orptr(mdo_addr, TypeEntries::type_unknown);
86 jmpb(next);
87
88 bind(none);
89 // first time here. Set profile type.
90 movptr(mdo_addr, obj);
91
92 bind(next);
93 }
94
95 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
96 if (!ProfileInterpreter) {
97 return;
98 }
99
100 if (MethodData::profile_arguments() || MethodData::profile_return()) {
101 Label profile_continue;
102
103 test_method_data_pointer(mdp, profile_continue);
104
105 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
106
107 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
108 jcc(Assembler::notEqual, profile_continue);
109
110 if (MethodData::profile_arguments()) {
111 Label done;
112 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
113 addptr(mdp, off_to_args);
114
115 for (int i = 0; i < TypeProfileArgsLimit; i++) {
116 if (i > 0 || MethodData::profile_return()) {
117 // If return value type is profiled we may have no argument to profile
118 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
119 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
120 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
121 jcc(Assembler::less, done);
122 }
123 movptr(tmp, Address(callee, Method::const_offset()));
124 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
125 // stack offset o (zero based) from the start of the argument
126 // list, for n arguments translates into offset n - o - 1 from
127 // the end of the argument list
128 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
129 subl(tmp, 1);
130 Address arg_addr = argument_address(tmp);
131 movptr(tmp, arg_addr);
132
133 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
134 profile_obj_type(tmp, mdo_arg_addr);
135
136 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
137 addptr(mdp, to_add);
138 off_to_args += to_add;
139 }
140
141 if (MethodData::profile_return()) {
142 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
143 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
144 }
145
146 bind(done);
147
148 if (MethodData::profile_return()) {
149 // We're right after the type profile for the last
150 // argument. tmp is the number of cells left in the
151 // CallTypeData/VirtualCallTypeData to reach its end. Non null
152 // if there's a return to profile.
153 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
154 shll(tmp, exact_log2(DataLayout::cell_size));
155 addptr(mdp, tmp);
156 }
157 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
158 } else {
159 assert(MethodData::profile_return(), "either profile call args or call ret");
160 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
161 }
162
163 // mdp points right after the end of the
164 // CallTypeData/VirtualCallTypeData, right after the cells for the
165 // return value type if there's one
166
167 bind(profile_continue);
168 }
169 }
170
171 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
172 assert_different_registers(mdp, ret, tmp, _bcp_register);
173 if (ProfileInterpreter && MethodData::profile_return()) {
174 Label profile_continue;
175
176 test_method_data_pointer(mdp, profile_continue);
177
178 if (MethodData::profile_return_jsr292_only()) {
179 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
180
181 // If we don't profile all invoke bytecodes we must make sure
182 // it's a bytecode we indeed profile. We can't go back to the
183 // begining of the ProfileData we intend to update to check its
184 // type because we're right after it and we don't known its
185 // length
186 Label do_profile;
187 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
188 jcc(Assembler::equal, do_profile);
189 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
190 jcc(Assembler::equal, do_profile);
191 get_method(tmp);
192 cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm);
193 jcc(Assembler::notEqual, profile_continue);
194
195 bind(do_profile);
196 }
197
198 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
199 mov(tmp, ret);
200 profile_obj_type(tmp, mdo_ret_addr);
201
202 bind(profile_continue);
203 }
204 }
205
206 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
207 if (ProfileInterpreter && MethodData::profile_parameters()) {
208 Label profile_continue;
209
210 test_method_data_pointer(mdp, profile_continue);
211
212 // Load the offset of the area within the MDO used for
213 // parameters. If it's negative we're not profiling any parameters
214 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
215 testl(tmp1, tmp1);
216 jcc(Assembler::negative, profile_continue);
217
218 // Compute a pointer to the area for parameters from the offset
219 // and move the pointer to the slot for the last
220 // parameters. Collect profiling from last parameter down.
221 // mdo start + parameters offset + array length - 1
222 addptr(mdp, tmp1);
223 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
224 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
225
226 Label loop;
227 bind(loop);
228
229 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
230 int type_base = in_bytes(ParametersTypeData::type_offset(0));
231 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
232 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
233 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
234
235 // load offset on the stack from the slot for this parameter
236 movptr(tmp2, arg_off);
237 negptr(tmp2);
238 // read the parameter from the local area
239 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
240
241 // profile the parameter
242 profile_obj_type(tmp2, arg_type);
243
244 // go to next parameter
245 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
246 jcc(Assembler::positive, loop);
247
248 bind(profile_continue);
249 }
250 }
251
252 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
253 int number_of_arguments) {
254 // interpreter specific
255 //
256 // Note: No need to save/restore bcp & locals registers
257 // since these are callee saved registers and no blocking/
258 // GC can happen in leaf calls.
259 // Further Note: DO NOT save/restore bcp/locals. If a caller has
260 // already saved them so that it can use rsi/rdi as temporaries
261 // then a save/restore here will DESTROY the copy the caller
262 // saved! There used to be a save_bcp() that only happened in
263 // the ASSERT path (no restore_bcp). Which caused bizarre failures
264 // when jvm built with ASSERTs.
265 #ifdef ASSERT
266 {
267 Label L;
268 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
269 jcc(Assembler::equal, L);
270 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
271 " last_sp != NULL");
272 bind(L);
273 }
274 #endif
275 // super call
276 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
277 // interpreter specific
278 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
279 // but since they may not have been saved (and we don't want to
280 // save them here (see note above) the assert is invalid.
281 }
282
283 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
284 Register java_thread,
285 Register last_java_sp,
286 address entry_point,
287 int number_of_arguments,
288 bool check_exceptions) {
289 // interpreter specific
290 //
291 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
292 // really make a difference for these runtime calls, since they are
293 // slow anyway. Btw., bcp must be saved/restored since it may change
294 // due to GC.
295 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
296 save_bcp();
297 #ifdef ASSERT
298 {
299 Label L;
300 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
301 jcc(Assembler::equal, L);
302 stop("InterpreterMacroAssembler::call_VM_base:"
303 " last_sp != NULL");
304 bind(L);
305 }
306 #endif /* ASSERT */
307 // super call
308 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
309 entry_point, number_of_arguments,
310 check_exceptions);
311 // interpreter specific
312 restore_bcp();
313 restore_locals();
314 }
315
316 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
317 if (JvmtiExport::can_pop_frame()) {
318 Label L;
319 // Initiate popframe handling only if it is not already being
320 // processed. If the flag has the popframe_processing bit set, it
321 // means that this code is called *during* popframe handling - we
322 // don't want to reenter.
323 // This method is only called just after the call into the vm in
324 // call_VM_base, so the arg registers are available.
325 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
326 LP64_ONLY(c_rarg0);
327 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
328 testl(pop_cond, JavaThread::popframe_pending_bit);
329 jcc(Assembler::zero, L);
330 testl(pop_cond, JavaThread::popframe_processing_bit);
331 jcc(Assembler::notZero, L);
332 // Call Interpreter::remove_activation_preserving_args_entry() to get the
333 // address of the same-named entrypoint in the generated interpreter code.
334 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
335 jmp(rax);
336 bind(L);
337 NOT_LP64(get_thread(java_thread);)
338 }
339 }
340
341 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
342 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
343 NOT_LP64(get_thread(thread);)
344 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
345 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
346 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
347 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
348 #ifdef _LP64
349 switch (state) {
350 case atos: movptr(rax, oop_addr);
351 movptr(oop_addr, (int32_t)NULL_WORD);
352 verify_oop(rax, state); break;
353 case ltos: movptr(rax, val_addr); break;
354 case btos: // fall through
355 case ztos: // fall through
356 case ctos: // fall through
357 case stos: // fall through
358 case itos: movl(rax, val_addr); break;
359 case ftos: load_float(val_addr); break;
360 case dtos: load_double(val_addr); break;
361 case vtos: /* nothing to do */ break;
362 default : ShouldNotReachHere();
363 }
364 // Clean up tos value in the thread object
365 movl(tos_addr, (int) ilgl);
366 movl(val_addr, (int32_t) NULL_WORD);
367 #else
368 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
369 + in_ByteSize(wordSize));
370 switch (state) {
371 case atos: movptr(rax, oop_addr);
372 movptr(oop_addr, NULL_WORD);
373 verify_oop(rax, state); break;
374 case ltos:
375 movl(rdx, val_addr1); // fall through
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 #endif // _LP64
387 // Clean up tos value in the thread object
388 movl(tos_addr, (int32_t) ilgl);
389 movptr(val_addr, NULL_WORD);
390 NOT_LP64(movptr(val_addr1, NULL_WORD);)
391 }
392
393
394 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
395 if (JvmtiExport::can_force_early_return()) {
396 Label L;
397 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
398 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
399
400 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
401 testptr(tmp, tmp);
402 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
403
404 // Initiate earlyret handling only if it is not already being processed.
405 // If the flag has the earlyret_processing bit set, it means that this code
406 // is called *during* earlyret handling - we don't want to reenter.
407 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
408 cmpl(tmp, JvmtiThreadState::earlyret_pending);
409 jcc(Assembler::notEqual, L);
410
411 // Call Interpreter::remove_activation_early_entry() to get the address of the
412 // same-named entrypoint in the generated interpreter code.
413 NOT_LP64(get_thread(java_thread);)
414 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
415 #ifdef _LP64
416 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
417 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
418 #else
419 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
420 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
421 #endif // _LP64
422 jmp(rax);
423 bind(L);
424 NOT_LP64(get_thread(java_thread);)
425 }
426 }
427
428 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
429 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
430 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
431 bswapl(reg);
432 shrl(reg, 16);
433 }
434
435 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
436 int bcp_offset,
437 size_t index_size) {
438 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
439 if (index_size == sizeof(u2)) {
440 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
441 } else if (index_size == sizeof(u4)) {
442 movl(index, Address(_bcp_register, bcp_offset));
443 // Check if the secondary index definition is still ~x, otherwise
444 // we have to change the following assembler code to calculate the
445 // plain index.
446 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
447 notl(index); // convert to plain index
448 } else if (index_size == sizeof(u1)) {
449 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
450 } else {
451 ShouldNotReachHere();
452 }
453 }
454
455 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
456 Register index,
457 int bcp_offset,
458 size_t index_size) {
459 assert_different_registers(cache, index);
460 get_cache_index_at_bcp(index, bcp_offset, index_size);
461 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
462 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
463 // convert from field index to ConstantPoolCacheEntry index
464 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
465 shll(index, 2);
466 }
467
468 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
469 Register index,
470 Register bytecode,
471 int byte_no,
472 int bcp_offset,
473 size_t index_size) {
474 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
475 // We use a 32-bit load here since the layout of 64-bit words on
476 // little-endian machines allow us that.
477 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
478 const int shift_count = (1 + byte_no) * BitsPerByte;
479 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
480 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
481 "correct shift count");
482 shrl(bytecode, shift_count);
483 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
484 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
485 }
486
487 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
488 Register tmp,
489 int bcp_offset,
490 size_t index_size) {
491 assert_different_registers(cache, tmp);
492
493 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
494 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
495 // convert from field index to ConstantPoolCacheEntry index
496 // and from word offset to byte offset
497 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
498 shll(tmp, 2 + LogBytesPerWord);
499 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
500 // skip past the header
501 addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
502 addptr(cache, tmp); // construct pointer to cache entry
503 }
504
505 // Load object from cpool->resolved_references(index)
506 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
507 Register index,
508 Register tmp) {
509 assert_different_registers(result, index);
510
511 get_constant_pool(result);
512 // load pointer for resolved_references[] objArray
513 movptr(result, Address(result, ConstantPool::cache_offset_in_bytes()));
514 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes()));
515 resolve_oop_handle(result, tmp);
516 load_heap_oop(result, Address(result, index,
517 UseCompressedOops ? Address::times_4 : Address::times_ptr,
518 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
519 }
520
521 // load cpool->resolved_klass_at(index)
522 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
523 Register cpool,
524 Register index) {
525 assert_different_registers(cpool, index);
526
527 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
528 Register resolved_klasses = cpool;
529 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes()));
530 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
531 }
532
533 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
534 Register method,
535 Register cache,
536 Register index) {
537 assert_different_registers(cache, index);
538
539 const int method_offset = in_bytes(
540 ConstantPoolCache::base_offset() +
541 ((byte_no == TemplateTable::f2_byte)
542 ? ConstantPoolCacheEntry::f2_offset()
543 : ConstantPoolCacheEntry::f1_offset()));
544
545 movptr(method, Address(cache, index, Address::times_ptr, method_offset)); // get f1 Method*
546 }
547
548 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
549 // subtype of super_klass.
550 //
551 // Args:
552 // rax: superklass
553 // Rsub_klass: subklass
554 //
555 // Kills:
556 // rcx, rdi
557 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
558 Label& ok_is_subtype) {
559 assert(Rsub_klass != rax, "rax holds superklass");
560 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
561 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
562 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
563 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
564
565 // Profile the not-null value's klass.
566 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
567
568 // Do the check.
569 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
570
571 // Profile the failure of the check.
572 profile_typecheck_failed(rcx); // blows rcx
573 }
574
575
576 #ifndef _LP64
577 void InterpreterMacroAssembler::f2ieee() {
578 if (IEEEPrecision) {
579 fstp_s(Address(rsp, 0));
580 fld_s(Address(rsp, 0));
581 }
582 }
583
584
585 void InterpreterMacroAssembler::d2ieee() {
586 if (IEEEPrecision) {
587 fstp_d(Address(rsp, 0));
588 fld_d(Address(rsp, 0));
589 }
590 }
591 #endif // _LP64
592
593 // Java Expression Stack
594
595 void InterpreterMacroAssembler::pop_ptr(Register r) {
596 pop(r);
597 }
598
599 void InterpreterMacroAssembler::push_ptr(Register r) {
600 push(r);
601 }
602
603 void InterpreterMacroAssembler::push_i(Register r) {
604 push(r);
605 }
606
607 void InterpreterMacroAssembler::push_f(XMMRegister r) {
608 subptr(rsp, wordSize);
609 movflt(Address(rsp, 0), r);
610 }
611
612 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
613 movflt(r, Address(rsp, 0));
614 addptr(rsp, wordSize);
615 }
616
617 void InterpreterMacroAssembler::push_d(XMMRegister r) {
618 subptr(rsp, 2 * wordSize);
619 movdbl(Address(rsp, 0), r);
620 }
621
622 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
623 movdbl(r, Address(rsp, 0));
624 addptr(rsp, 2 * Interpreter::stackElementSize);
625 }
626
627 #ifdef _LP64
628 void InterpreterMacroAssembler::pop_i(Register r) {
629 // XXX can't use pop currently, upper half non clean
630 movl(r, Address(rsp, 0));
631 addptr(rsp, wordSize);
632 }
633
634 void InterpreterMacroAssembler::pop_l(Register r) {
635 movq(r, Address(rsp, 0));
636 addptr(rsp, 2 * Interpreter::stackElementSize);
637 }
638
639 void InterpreterMacroAssembler::push_l(Register r) {
640 subptr(rsp, 2 * wordSize);
641 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
642 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
643 }
644
645 void InterpreterMacroAssembler::pop(TosState state) {
646 switch (state) {
647 case atos: pop_ptr(); break;
648 case btos:
649 case ztos:
650 case ctos:
651 case stos:
652 case itos: pop_i(); break;
653 case ltos: pop_l(); break;
654 case ftos: pop_f(xmm0); break;
655 case dtos: pop_d(xmm0); break;
656 case vtos: /* nothing to do */ break;
657 default: ShouldNotReachHere();
658 }
659 verify_oop(rax, state);
660 }
661
662 void InterpreterMacroAssembler::push(TosState state) {
663 verify_oop(rax, state);
664 switch (state) {
665 case atos: push_ptr(); break;
666 case btos:
667 case ztos:
668 case ctos:
669 case stos:
670 case itos: push_i(); break;
671 case ltos: push_l(); break;
672 case ftos: push_f(xmm0); break;
673 case dtos: push_d(xmm0); break;
674 case vtos: /* nothing to do */ break;
675 default : ShouldNotReachHere();
676 }
677 }
678 #else
679 void InterpreterMacroAssembler::pop_i(Register r) {
680 pop(r);
681 }
682
683 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
684 pop(lo);
685 pop(hi);
686 }
687
688 void InterpreterMacroAssembler::pop_f() {
689 fld_s(Address(rsp, 0));
690 addptr(rsp, 1 * wordSize);
691 }
692
693 void InterpreterMacroAssembler::pop_d() {
694 fld_d(Address(rsp, 0));
695 addptr(rsp, 2 * wordSize);
696 }
697
698
699 void InterpreterMacroAssembler::pop(TosState state) {
700 switch (state) {
701 case atos: pop_ptr(rax); break;
702 case btos: // fall through
703 case ztos: // fall through
704 case ctos: // fall through
705 case stos: // fall through
706 case itos: pop_i(rax); break;
707 case ltos: pop_l(rax, rdx); break;
708 case ftos:
709 if (UseSSE >= 1) {
710 pop_f(xmm0);
711 } else {
712 pop_f();
713 }
714 break;
715 case dtos:
716 if (UseSSE >= 2) {
717 pop_d(xmm0);
718 } else {
719 pop_d();
720 }
721 break;
722 case vtos: /* nothing to do */ break;
723 default : ShouldNotReachHere();
724 }
725 verify_oop(rax, state);
726 }
727
728
729 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
730 push(hi);
731 push(lo);
732 }
733
734 void InterpreterMacroAssembler::push_f() {
735 // Do not schedule for no AGI! Never write beyond rsp!
736 subptr(rsp, 1 * wordSize);
737 fstp_s(Address(rsp, 0));
738 }
739
740 void InterpreterMacroAssembler::push_d() {
741 // Do not schedule for no AGI! Never write beyond rsp!
742 subptr(rsp, 2 * wordSize);
743 fstp_d(Address(rsp, 0));
744 }
745
746
747 void InterpreterMacroAssembler::push(TosState state) {
748 verify_oop(rax, state);
749 switch (state) {
750 case atos: push_ptr(rax); break;
751 case btos: // fall through
752 case ztos: // fall through
753 case ctos: // fall through
754 case stos: // fall through
755 case itos: push_i(rax); break;
756 case ltos: push_l(rax, rdx); break;
757 case ftos:
758 if (UseSSE >= 1) {
759 push_f(xmm0);
760 } else {
761 push_f();
762 }
763 break;
764 case dtos:
765 if (UseSSE >= 2) {
766 push_d(xmm0);
767 } else {
768 push_d();
769 }
770 break;
771 case vtos: /* nothing to do */ break;
772 default : ShouldNotReachHere();
773 }
774 }
775 #endif // _LP64
776
777
778 // Helpers for swap and dup
779 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
780 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
781 }
782
783 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
784 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
785 }
786
787
788 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
789 // set sender sp
790 lea(_bcp_register, Address(rsp, wordSize));
791 // record last_sp
792 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
793 }
794
795
796 // Jump to from_interpreted entry of a call unless single stepping is possible
797 // in this thread in which case we must call the i2i entry
798 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
799 prepare_to_jump_from_interpreted();
800
801 if (JvmtiExport::can_post_interpreter_events()) {
802 Label run_compiled_code;
803 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
804 // compiled code in threads for which the event is enabled. Check here for
805 // interp_only_mode if these events CAN be enabled.
806 // interp_only is an int, on little endian it is sufficient to test the byte only
807 // Is a cmpl faster?
808 LP64_ONLY(temp = r15_thread;)
809 NOT_LP64(get_thread(temp);)
810 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
811 jccb(Assembler::zero, run_compiled_code);
812 jmp(Address(method, Method::interpreter_entry_offset()));
813 bind(run_compiled_code);
814 }
815
816 jmp(Address(method, Method::from_interpreted_offset()));
817 }
818
819 // The following two routines provide a hook so that an implementation
820 // can schedule the dispatch in two parts. x86 does not do this.
821 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
822 // Nothing x86 specific to be done here
823 }
824
825 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
826 dispatch_next(state, step);
827 }
828
829 void InterpreterMacroAssembler::dispatch_base(TosState state,
830 address* table,
831 bool verifyoop,
832 bool generate_poll) {
833 verify_FPU(1, state);
834 if (VerifyActivationFrameSize) {
835 Label L;
836 mov(rcx, rbp);
837 subptr(rcx, rsp);
838 int32_t min_frame_size =
839 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
840 wordSize;
841 cmpptr(rcx, (int32_t)min_frame_size);
842 jcc(Assembler::greaterEqual, L);
843 stop("broken stack frame");
844 bind(L);
845 }
846 if (verifyoop) {
847 verify_oop(rax, state);
848 }
849
850 address* const safepoint_table = Interpreter::safept_table(state);
851 #ifdef _LP64
852 Label no_safepoint, dispatch;
853 if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) {
854 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
855 testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
856
857 jccb(Assembler::zero, no_safepoint);
858 lea(rscratch1, ExternalAddress((address)safepoint_table));
859 jmpb(dispatch);
860 }
861
862 bind(no_safepoint);
863 lea(rscratch1, ExternalAddress((address)table));
864 bind(dispatch);
865 jmp(Address(rscratch1, rbx, Address::times_8));
866
867 #else
868 Address index(noreg, rbx, Address::times_ptr);
869 if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) {
870 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
871 Label no_safepoint;
872 const Register thread = rcx;
873 get_thread(thread);
874 testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
875
876 jccb(Assembler::zero, no_safepoint);
877 ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
878 jump(dispatch_addr);
879 bind(no_safepoint);
880 }
881
882 {
883 ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
884 jump(dispatch_addr);
885 }
886 #endif // _LP64
887 }
888
889 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
890 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
891 }
892
893 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
894 dispatch_base(state, Interpreter::normal_table(state));
895 }
896
897 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
898 dispatch_base(state, Interpreter::normal_table(state), false);
899 }
900
901
902 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
903 // load next bytecode (load before advancing _bcp_register to prevent AGI)
904 load_unsigned_byte(rbx, Address(_bcp_register, step));
905 // advance _bcp_register
906 increment(_bcp_register, step);
907 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
908 }
909
910 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
911 // load current bytecode
912 load_unsigned_byte(rbx, Address(_bcp_register, 0));
913 dispatch_base(state, table);
914 }
915
916 void InterpreterMacroAssembler::narrow(Register result) {
917
918 // Get method->_constMethod->_result_type
919 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
920 movptr(rcx, Address(rcx, Method::const_offset()));
921 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
922
923 Label done, notBool, notByte, notChar;
924
925 // common case first
926 cmpl(rcx, T_INT);
927 jcc(Assembler::equal, done);
928
929 // mask integer result to narrower return type.
930 cmpl(rcx, T_BOOLEAN);
931 jcc(Assembler::notEqual, notBool);
932 andl(result, 0x1);
933 jmp(done);
934
935 bind(notBool);
936 cmpl(rcx, T_BYTE);
937 jcc(Assembler::notEqual, notByte);
938 LP64_ONLY(movsbl(result, result);)
939 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
940 NOT_LP64(sarl(result, 24);) // and sign-extend byte
941 jmp(done);
942
943 bind(notByte);
944 cmpl(rcx, T_CHAR);
945 jcc(Assembler::notEqual, notChar);
946 LP64_ONLY(movzwl(result, result);)
947 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
948 jmp(done);
949
950 bind(notChar);
951 // cmpl(rcx, T_SHORT); // all that's left
952 // jcc(Assembler::notEqual, done);
953 LP64_ONLY(movswl(result, result);)
954 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
955 NOT_LP64(sarl(result, 16);) // and sign-extend short
956
957 // Nothing to do for T_INT
958 bind(done);
959 }
960
961 // remove activation
962 //
963 // Unlock the receiver if this is a synchronized method.
964 // Unlock any Java monitors from syncronized blocks.
965 // Remove the activation from the stack.
966 //
967 // If there are locked Java monitors
968 // If throw_monitor_exception
969 // throws IllegalMonitorStateException
970 // Else if install_monitor_exception
971 // installs IllegalMonitorStateException
972 // Else
973 // no error processing
974 void InterpreterMacroAssembler::remove_activation(
975 TosState state,
976 Register ret_addr,
977 bool throw_monitor_exception,
978 bool install_monitor_exception,
979 bool notify_jvmdi) {
980 // Note: Registers rdx xmm0 may be in use for the
981 // result check if synchronized method
982 Label unlocked, unlock, no_unlock;
983
984 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
985 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
986 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
987 // monitor pointers need different register
988 // because rdx may have the result in it
989 NOT_LP64(get_thread(rcx);)
990
991 // get the value of _do_not_unlock_if_synchronized into rdx
992 const Address do_not_unlock_if_synchronized(rthread,
993 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
994 movbool(rbx, do_not_unlock_if_synchronized);
995 movbool(do_not_unlock_if_synchronized, false); // reset the flag
996
997 // get method access flags
998 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
999 movl(rcx, Address(rcx, Method::access_flags_offset()));
1000 testl(rcx, JVM_ACC_SYNCHRONIZED);
1001 jcc(Assembler::zero, unlocked);
1002
1003 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
1004 // is set.
1005 testbool(rbx);
1006 jcc(Assembler::notZero, no_unlock);
1007
1008 // unlock monitor
1009 push(state); // save result
1010
1011 // BasicObjectLock will be first in list, since this is a
1012 // synchronized method. However, need to check that the object has
1013 // not been unlocked by an explicit monitorexit bytecode.
1014 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
1015 wordSize - (int) sizeof(BasicObjectLock));
1016 // We use c_rarg1/rdx so that if we go slow path it will be the correct
1017 // register for unlock_object to pass to VM directly
1018 lea(robj, monitor); // address of first monitor
1019
1020 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
1021 testptr(rax, rax);
1022 jcc(Assembler::notZero, unlock);
1023
1024 pop(state);
1025 if (throw_monitor_exception) {
1026 // Entry already unlocked, need to throw exception
1027 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
1028 call_VM(noreg, CAST_FROM_FN_PTR(address,
1029 InterpreterRuntime::throw_illegal_monitor_state_exception));
1030 should_not_reach_here();
1031 } else {
1032 // Monitor already unlocked during a stack unroll. If requested,
1033 // install an illegal_monitor_state_exception. Continue with
1034 // stack unrolling.
1035 if (install_monitor_exception) {
1036 NOT_LP64(empty_FPU_stack();)
1037 call_VM(noreg, CAST_FROM_FN_PTR(address,
1038 InterpreterRuntime::new_illegal_monitor_state_exception));
1039 }
1040 jmp(unlocked);
1041 }
1042
1043 bind(unlock);
1044 unlock_object(robj);
1045 pop(state);
1046
1047 // Check that for block-structured locking (i.e., that all locked
1048 // objects has been unlocked)
1049 bind(unlocked);
1050
1051 // rax, rdx: Might contain return value
1052
1053 // Check that all monitors are unlocked
1054 {
1055 Label loop, exception, entry, restart;
1056 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1057 const Address monitor_block_top(
1058 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1059 const Address monitor_block_bot(
1060 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1061
1062 bind(restart);
1063 // We use c_rarg1 so that if we go slow path it will be the correct
1064 // register for unlock_object to pass to VM directly
1065 movptr(rmon, monitor_block_top); // points to current entry, starting
1066 // with top-most entry
1067 lea(rbx, monitor_block_bot); // points to word before bottom of
1068 // monitor block
1069 jmp(entry);
1070
1071 // Entry already locked, need to throw exception
1072 bind(exception);
1073
1074 if (throw_monitor_exception) {
1075 // Throw exception
1076 NOT_LP64(empty_FPU_stack();)
1077 MacroAssembler::call_VM(noreg,
1078 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1079 throw_illegal_monitor_state_exception));
1080 should_not_reach_here();
1081 } else {
1082 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1083 // Unlock does not block, so don't have to worry about the frame.
1084 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1085
1086 push(state);
1087 mov(robj, rmon); // nop if robj and rmon are the same
1088 unlock_object(robj);
1089 pop(state);
1090
1091 if (install_monitor_exception) {
1092 NOT_LP64(empty_FPU_stack();)
1093 call_VM(noreg, CAST_FROM_FN_PTR(address,
1094 InterpreterRuntime::
1095 new_illegal_monitor_state_exception));
1096 }
1097
1098 jmp(restart);
1099 }
1100
1101 bind(loop);
1102 // check if current entry is used
1103 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
1104 jcc(Assembler::notEqual, exception);
1105
1106 addptr(rmon, entry_size); // otherwise advance to next entry
1107 bind(entry);
1108 cmpptr(rmon, rbx); // check if bottom reached
1109 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1110 }
1111
1112 bind(no_unlock);
1113
1114 // jvmti support
1115 if (notify_jvmdi) {
1116 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1117 } else {
1118 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1119 }
1120
1121 if (StackReservedPages > 0) {
1122 movptr(rbx,
1123 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1124 // testing if reserved zone needs to be re-enabled
1125 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1126 Label no_reserved_zone_enabling;
1127
1128 NOT_LP64(get_thread(rthread);)
1129
1130 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_enabled);
1131 jcc(Assembler::equal, no_reserved_zone_enabling);
1132
1133 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1134 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1135
1136 call_VM_leaf(
1137 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1138 call_VM(noreg, CAST_FROM_FN_PTR(address,
1139 InterpreterRuntime::throw_delayed_StackOverflowError));
1140 should_not_reach_here();
1141
1142 bind(no_reserved_zone_enabling);
1143 }
1144
1145 // remove activation
1146 // get sender sp
1147 movptr(rbx,
1148 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1149
1150 if (state == atos && ValueTypeReturnedAsFields) {
1151 Label skip;
1152 // Test if the return type is a value type
1153 movptr(rdi, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
1154 movptr(rdi, Address(rdi, Method::const_offset()));
1155 load_unsigned_byte(rdi, Address(rdi, ConstMethod::result_type_offset()));
1156 cmpl(rdi, T_VALUETYPE);
1157 jcc(Assembler::notEqual, skip);
1158
1159 // We are returning a value type, load its fields into registers
1160 #ifndef _LP64
1161 super_call_VM_leaf(StubRoutines::load_value_type_fields_in_regs());
1162 #else
1163 // Load fields from a buffered value with a value class specific handler
1164 load_klass(rdi, rax);
1165 movptr(rdi, Address(rdi, InstanceKlass::adr_valueklass_fixed_block_offset()));
1166 movptr(rdi, Address(rdi, ValueKlass::unpack_handler_offset()));
1167
1168 testptr(rdi, rdi);
1169 jcc(Assembler::equal, skip);
1170
1171 call(rdi);
1172 #endif
1173 // call above kills the value in rbx. Reload it.
1174 movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1175 bind(skip);
1176 }
1177 leave(); // remove frame anchor
1178 pop(ret_addr); // get return address
1179 mov(rsp, rbx); // set sp to sender sp
1180 }
1181
1182 void InterpreterMacroAssembler::get_method_counters(Register method,
1183 Register mcs, Label& skip) {
1184 Label has_counters;
1185 movptr(mcs, Address(method, Method::method_counters_offset()));
1186 testptr(mcs, mcs);
1187 jcc(Assembler::notZero, has_counters);
1188 call_VM(noreg, CAST_FROM_FN_PTR(address,
1189 InterpreterRuntime::build_method_counters), method);
1190 movptr(mcs, Address(method,Method::method_counters_offset()));
1191 testptr(mcs, mcs);
1192 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1193 bind(has_counters);
1194 }
1195
1196 void InterpreterMacroAssembler::allocate_instance(Register klass, Register new_obj,
1197 Register t1, Register t2,
1198 bool clear_fields, Label& alloc_failed) {
1199 MacroAssembler::allocate_instance(klass, new_obj, t1, t2, clear_fields, alloc_failed);
1200 {
1201 SkipIfEqual skip_if(this, &DTraceAllocProbes, 0);
1202 // Trigger dtrace event for fastpath
1203 push(atos);
1204 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), new_obj);
1205 pop(atos);
1206 }
1207 }
1208
1209
1210 void InterpreterMacroAssembler::read_flattened_field(Register holder_klass,
1211 Register field_index, Register field_offset,
1212 Register obj) {
1213 Label alloc_failed, empty_value, done;
1214 const Register src = field_offset;
1215 const Register alloc_temp = LP64_ONLY(rscratch1) NOT_LP64(rsi);
1216 const Register dst_temp = LP64_ONLY(rscratch2) NOT_LP64(rdi);
1217 assert_different_registers(obj, holder_klass, field_index, field_offset, dst_temp);
1218
1219 // Grap the inline field klass
1220 push(holder_klass);
1221 const Register field_klass = holder_klass;
1222 get_value_field_klass(holder_klass, field_index, field_klass);
1223
1224 //check for empty value klass
1225 test_klass_is_empty_value(field_klass, dst_temp, empty_value);
1226
1227 // allocate buffer
1228 push(obj); // save holder
1229 allocate_instance(field_klass, obj, alloc_temp, dst_temp, false, alloc_failed);
1230
1231 // Have a oop instance buffer, copy into it
1232 data_for_oop(obj, dst_temp, field_klass);
1233 pop(alloc_temp); // restore holder
1234 lea(src, Address(alloc_temp, field_offset));
1235 // call_VM_leaf, clobbers a few regs, save restore new obj
1236 push(obj);
1237 access_value_copy(IS_DEST_UNINITIALIZED, src, dst_temp, field_klass);
1238 pop(obj);
1239 pop(holder_klass);
1240 jmp(done);
1241
1242 bind(empty_value);
1243 get_empty_value_oop(field_klass, dst_temp, obj);
1244 pop(holder_klass);
1245 jmp(done);
1246
1247 bind(alloc_failed);
1248 pop(obj);
1249 pop(holder_klass);
1250 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flattened_field),
1251 obj, field_index, holder_klass);
1252
1253 bind(done);
1254 }
1255
1256 // Lock object
1257 //
1258 // Args:
1259 // rdx, c_rarg1: BasicObjectLock to be used for locking
1260 //
1261 // Kills:
1262 // rax, rbx
1263 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1264 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1265 "The argument is only for looks. It must be c_rarg1");
1266
1267 if (UseHeavyMonitors) {
1268 call_VM(noreg,
1269 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1270 lock_reg);
1271 } else {
1272 Label done;
1273
1274 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1275 const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a
1276 // problematic case where tmp_reg = no_reg.
1277 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1278
1279 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
1280 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
1281 const int mark_offset = lock_offset +
1282 BasicLock::displaced_header_offset_in_bytes();
1283
1284 Label slow_case;
1285
1286 // Load object pointer into obj_reg
1287 movptr(obj_reg, Address(lock_reg, obj_offset));
1288
1289 if (UseBiasedLocking) {
1290 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case);
1291 }
1292
1293 // Load immediate 1 into swap_reg %rax
1294 movl(swap_reg, (int32_t)1);
1295
1296 // Load (object->mark() | 1) into swap_reg %rax
1297 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1298 if (EnableValhalla && !UseBiasedLocking) {
1299 // For slow path is_always_locked, using biased, which is never natural for !UseBiasLocking
1300 andptr(swap_reg, ~((int) markWord::biased_lock_bit_in_place));
1301 }
1302
1303 // Save (object->mark() | 1) into BasicLock's displaced header
1304 movptr(Address(lock_reg, mark_offset), swap_reg);
1305
1306 assert(lock_offset == 0,
1307 "displaced header must be first word in BasicObjectLock");
1308
1309 lock();
1310 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1311 if (PrintBiasedLockingStatistics) {
1312 cond_inc32(Assembler::zero,
1313 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1314 }
1315 jcc(Assembler::zero, done);
1316
1317 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1318
1319 // Test if the oopMark is an obvious stack pointer, i.e.,
1320 // 1) (mark & zero_bits) == 0, and
1321 // 2) rsp <= mark < mark + os::pagesize()
1322 //
1323 // These 3 tests can be done by evaluating the following
1324 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1325 // assuming both stack pointer and pagesize have their
1326 // least significant bits clear.
1327 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
1328 subptr(swap_reg, rsp);
1329 andptr(swap_reg, zero_bits - os::vm_page_size());
1330
1331 // Save the test result, for recursive case, the result is zero
1332 movptr(Address(lock_reg, mark_offset), swap_reg);
1333
1334 if (PrintBiasedLockingStatistics) {
1335 cond_inc32(Assembler::zero,
1336 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1337 }
1338 jcc(Assembler::zero, done);
1339
1340 bind(slow_case);
1341
1342 // Call the runtime routine for slow case
1343 call_VM(noreg,
1344 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1345 lock_reg);
1346
1347 bind(done);
1348 }
1349 }
1350
1351
1352 // Unlocks an object. Used in monitorexit bytecode and
1353 // remove_activation. Throws an IllegalMonitorException if object is
1354 // not locked by current thread.
1355 //
1356 // Args:
1357 // rdx, c_rarg1: BasicObjectLock for lock
1358 //
1359 // Kills:
1360 // rax
1361 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1362 // rscratch1 (scratch reg)
1363 // rax, rbx, rcx, rdx
1364 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1365 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1366 "The argument is only for looks. It must be c_rarg1");
1367
1368 if (UseHeavyMonitors) {
1369 call_VM(noreg,
1370 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1371 lock_reg);
1372 } else {
1373 Label done;
1374
1375 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1376 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1377 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1378
1379 save_bcp(); // Save in case of exception
1380
1381 // Convert from BasicObjectLock structure to object and BasicLock
1382 // structure Store the BasicLock address into %rax
1383 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
1384
1385 // Load oop into obj_reg(%c_rarg3)
1386 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
1387
1388 // Free entry
1389 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
1390
1391 if (UseBiasedLocking) {
1392 biased_locking_exit(obj_reg, header_reg, done);
1393 }
1394
1395 // Load the old header from BasicLock structure
1396 movptr(header_reg, Address(swap_reg,
1397 BasicLock::displaced_header_offset_in_bytes()));
1398
1399 // Test for recursion
1400 testptr(header_reg, header_reg);
1401
1402 // zero for recursive case
1403 jcc(Assembler::zero, done);
1404
1405 // Atomic swap back the old header
1406 lock();
1407 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1408
1409 // zero for simple unlock of a stack-lock case
1410 jcc(Assembler::zero, done);
1411
1412 // Call the runtime routine for slow case.
1413 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
1414 obj_reg); // restore obj
1415 call_VM(noreg,
1416 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1417 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 #if INCLUDE_JVMCI
1719 if (MethodProfileWidth == 0) {
1720 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1721 }
1722 #else // INCLUDE_JVMCI
1723 update_mdp_by_constant(mdp,
1724 in_bytes(VirtualCallData::
1725 virtual_call_data_size()));
1726 #endif // INCLUDE_JVMCI
1727 bind(profile_continue);
1728 }
1729 }
1730
1731 #if INCLUDE_JVMCI
1732 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) {
1733 assert_different_registers(method, mdp, reg2);
1734 if (ProfileInterpreter && MethodProfileWidth > 0) {
1735 Label profile_continue;
1736
1737 // If no method data exists, go to profile_continue.
1738 test_method_data_pointer(mdp, profile_continue);
1739
1740 Label done;
1741 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth,
1742 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1743 bind(done);
1744
1745 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1746 bind(profile_continue);
1747 }
1748 }
1749 #endif // INCLUDE_JVMCI
1750
1751 // This routine creates a state machine for updating the multi-row
1752 // type profile at a virtual call site (or other type-sensitive bytecode).
1753 // The machine visits each row (of receiver/count) until the receiver type
1754 // is found, or until it runs out of rows. At the same time, it remembers
1755 // the location of the first empty row. (An empty row records null for its
1756 // receiver, and can be allocated for a newly-observed receiver type.)
1757 // Because there are two degrees of freedom in the state, a simple linear
1758 // search will not work; it must be a decision tree. Hence this helper
1759 // function is recursive, to generate the required tree structured code.
1760 // It's the interpreter, so we are trading off code space for speed.
1761 // See below for example code.
1762 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1763 Register receiver, Register mdp,
1764 Register reg2, int start_row,
1765 Label& done, bool is_virtual_call) {
1766 if (TypeProfileWidth == 0) {
1767 if (is_virtual_call) {
1768 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1769 }
1770 #if INCLUDE_JVMCI
1771 else if (EnableJVMCI) {
1772 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1773 }
1774 #endif // INCLUDE_JVMCI
1775 } else {
1776 int non_profiled_offset = -1;
1777 if (is_virtual_call) {
1778 non_profiled_offset = in_bytes(CounterData::count_offset());
1779 }
1780 #if INCLUDE_JVMCI
1781 else if (EnableJVMCI) {
1782 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1783 }
1784 #endif // INCLUDE_JVMCI
1785
1786 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1787 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1788 }
1789 }
1790
1791 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1792 Register reg2, int start_row, Label& done, int total_rows,
1793 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1794 int non_profiled_offset) {
1795 int last_row = total_rows - 1;
1796 assert(start_row <= last_row, "must be work left to do");
1797 // Test this row for both the item and for null.
1798 // Take any of three different outcomes:
1799 // 1. found item => increment count and goto done
1800 // 2. found null => keep looking for case 1, maybe allocate this cell
1801 // 3. found something else => keep looking for cases 1 and 2
1802 // Case 3 is handled by a recursive call.
1803 for (int row = start_row; row <= last_row; row++) {
1804 Label next_test;
1805 bool test_for_null_also = (row == start_row);
1806
1807 // See if the item is item[n].
1808 int item_offset = in_bytes(item_offset_fn(row));
1809 test_mdp_data_at(mdp, item_offset, item,
1810 (test_for_null_also ? reg2 : noreg),
1811 next_test);
1812 // (Reg2 now contains the item from the CallData.)
1813
1814 // The item is item[n]. Increment count[n].
1815 int count_offset = in_bytes(item_count_offset_fn(row));
1816 increment_mdp_data_at(mdp, count_offset);
1817 jmp(done);
1818 bind(next_test);
1819
1820 if (test_for_null_also) {
1821 // Failed the equality check on item[n]... Test for null.
1822 testptr(reg2, reg2);
1823 if (start_row == last_row) {
1824 // The only thing left to do is handle the null case.
1825 if (non_profiled_offset >= 0) {
1826 Label found_null;
1827 jccb(Assembler::zero, found_null);
1828 // Item did not match any saved item and there is no empty row for it.
1829 // Increment total counter to indicate polymorphic case.
1830 increment_mdp_data_at(mdp, non_profiled_offset);
1831 jmp(done);
1832 bind(found_null);
1833 } else {
1834 jcc(Assembler::notZero, done);
1835 }
1836 break;
1837 }
1838 Label found_null;
1839 // Since null is rare, make it be the branch-taken case.
1840 jcc(Assembler::zero, found_null);
1841
1842 // Put all the "Case 3" tests here.
1843 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1844 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1845
1846 // Found a null. Keep searching for a matching item,
1847 // but remember that this is an empty (unused) slot.
1848 bind(found_null);
1849 }
1850 }
1851
1852 // In the fall-through case, we found no matching item, but we
1853 // observed the item[start_row] is NULL.
1854
1855 // Fill in the item field and increment the count.
1856 int item_offset = in_bytes(item_offset_fn(start_row));
1857 set_mdp_data_at(mdp, item_offset, item);
1858 int count_offset = in_bytes(item_count_offset_fn(start_row));
1859 movl(reg2, DataLayout::counter_increment);
1860 set_mdp_data_at(mdp, count_offset, reg2);
1861 if (start_row > 0) {
1862 jmp(done);
1863 }
1864 }
1865
1866 // Example state machine code for three profile rows:
1867 // // main copy of decision tree, rooted at row[1]
1868 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1869 // if (row[0].rec != NULL) {
1870 // // inner copy of decision tree, rooted at row[1]
1871 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1872 // if (row[1].rec != NULL) {
1873 // // degenerate decision tree, rooted at row[2]
1874 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1875 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1876 // row[2].init(rec); goto done;
1877 // } else {
1878 // // remember row[1] is empty
1879 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1880 // row[1].init(rec); goto done;
1881 // }
1882 // } else {
1883 // // remember row[0] is empty
1884 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1885 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1886 // row[0].init(rec); goto done;
1887 // }
1888 // done:
1889
1890 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1891 Register mdp, Register reg2,
1892 bool is_virtual_call) {
1893 assert(ProfileInterpreter, "must be profiling");
1894 Label done;
1895
1896 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1897
1898 bind (done);
1899 }
1900
1901 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1902 Register mdp) {
1903 if (ProfileInterpreter) {
1904 Label profile_continue;
1905 uint row;
1906
1907 // If no method data exists, go to profile_continue.
1908 test_method_data_pointer(mdp, profile_continue);
1909
1910 // Update the total ret count.
1911 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1912
1913 for (row = 0; row < RetData::row_limit(); row++) {
1914 Label next_test;
1915
1916 // See if return_bci is equal to bci[n]:
1917 test_mdp_data_at(mdp,
1918 in_bytes(RetData::bci_offset(row)),
1919 return_bci, noreg,
1920 next_test);
1921
1922 // return_bci is equal to bci[n]. Increment the count.
1923 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1924
1925 // The method data pointer needs to be updated to reflect the new target.
1926 update_mdp_by_offset(mdp,
1927 in_bytes(RetData::bci_displacement_offset(row)));
1928 jmp(profile_continue);
1929 bind(next_test);
1930 }
1931
1932 update_mdp_for_ret(return_bci);
1933
1934 bind(profile_continue);
1935 }
1936 }
1937
1938
1939 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1940 if (ProfileInterpreter) {
1941 Label profile_continue;
1942
1943 // If no method data exists, go to profile_continue.
1944 test_method_data_pointer(mdp, profile_continue);
1945
1946 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1947
1948 // The method data pointer needs to be updated.
1949 int mdp_delta = in_bytes(BitData::bit_data_size());
1950 if (TypeProfileCasts) {
1951 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1952 }
1953 update_mdp_by_constant(mdp, mdp_delta);
1954
1955 bind(profile_continue);
1956 }
1957 }
1958
1959
1960 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1961 if (ProfileInterpreter && TypeProfileCasts) {
1962 Label profile_continue;
1963
1964 // If no method data exists, go to profile_continue.
1965 test_method_data_pointer(mdp, profile_continue);
1966
1967 int count_offset = in_bytes(CounterData::count_offset());
1968 // Back up the address, since we have already bumped the mdp.
1969 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1970
1971 // *Decrement* the counter. We expect to see zero or small negatives.
1972 increment_mdp_data_at(mdp, count_offset, true);
1973
1974 bind (profile_continue);
1975 }
1976 }
1977
1978
1979 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1980 if (ProfileInterpreter) {
1981 Label profile_continue;
1982
1983 // If no method data exists, go to profile_continue.
1984 test_method_data_pointer(mdp, profile_continue);
1985
1986 // The method data pointer needs to be updated.
1987 int mdp_delta = in_bytes(BitData::bit_data_size());
1988 if (TypeProfileCasts) {
1989 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1990
1991 // Record the object type.
1992 record_klass_in_profile(klass, mdp, reg2, false);
1993 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1994 NOT_LP64(restore_locals();) // Restore EDI
1995 }
1996 update_mdp_by_constant(mdp, mdp_delta);
1997
1998 bind(profile_continue);
1999 }
2000 }
2001
2002
2003 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
2004 if (ProfileInterpreter) {
2005 Label profile_continue;
2006
2007 // If no method data exists, go to profile_continue.
2008 test_method_data_pointer(mdp, profile_continue);
2009
2010 // Update the default case count
2011 increment_mdp_data_at(mdp,
2012 in_bytes(MultiBranchData::default_count_offset()));
2013
2014 // The method data pointer needs to be updated.
2015 update_mdp_by_offset(mdp,
2016 in_bytes(MultiBranchData::
2017 default_displacement_offset()));
2018
2019 bind(profile_continue);
2020 }
2021 }
2022
2023
2024 void InterpreterMacroAssembler::profile_switch_case(Register index,
2025 Register mdp,
2026 Register reg2) {
2027 if (ProfileInterpreter) {
2028 Label profile_continue;
2029
2030 // If no method data exists, go to profile_continue.
2031 test_method_data_pointer(mdp, profile_continue);
2032
2033 // Build the base (index * per_case_size_in_bytes()) +
2034 // case_array_offset_in_bytes()
2035 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
2036 imulptr(index, reg2); // XXX l ?
2037 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
2038
2039 // Update the case count
2040 increment_mdp_data_at(mdp,
2041 index,
2042 in_bytes(MultiBranchData::relative_count_offset()));
2043
2044 // The method data pointer needs to be updated.
2045 update_mdp_by_offset(mdp,
2046 index,
2047 in_bytes(MultiBranchData::
2048 relative_displacement_offset()));
2049
2050 bind(profile_continue);
2051 }
2052 }
2053
2054
2055
2056 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2057 if (state == atos) {
2058 MacroAssembler::verify_oop(reg);
2059 }
2060 }
2061
2062 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2063 #ifndef _LP64
2064 if ((state == ftos && UseSSE < 1) ||
2065 (state == dtos && UseSSE < 2)) {
2066 MacroAssembler::verify_FPU(stack_depth);
2067 }
2068 #endif
2069 }
2070
2071 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
2072 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
2073 int increment, Address mask,
2074 Register scratch, bool preloaded,
2075 Condition cond, Label* where) {
2076 if (!preloaded) {
2077 movl(scratch, counter_addr);
2078 }
2079 incrementl(scratch, increment);
2080 movl(counter_addr, scratch);
2081 andl(scratch, mask);
2082 if (where != NULL) {
2083 jcc(cond, *where);
2084 }
2085 }
2086
2087 void InterpreterMacroAssembler::notify_method_entry() {
2088 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2089 // track stack depth. If it is possible to enter interp_only_mode we add
2090 // the code to check if the event should be sent.
2091 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2092 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2093 if (JvmtiExport::can_post_interpreter_events()) {
2094 Label L;
2095 NOT_LP64(get_thread(rthread);)
2096 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2097 testl(rdx, rdx);
2098 jcc(Assembler::zero, L);
2099 call_VM(noreg, CAST_FROM_FN_PTR(address,
2100 InterpreterRuntime::post_method_entry));
2101 bind(L);
2102 }
2103
2104 {
2105 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2106 NOT_LP64(get_thread(rthread);)
2107 get_method(rarg);
2108 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2109 rthread, rarg);
2110 }
2111
2112 // RedefineClasses() tracing support for obsolete method entry
2113 if (log_is_enabled(Trace, redefine, class, obsolete)) {
2114 NOT_LP64(get_thread(rthread);)
2115 get_method(rarg);
2116 call_VM_leaf(
2117 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2118 rthread, rarg);
2119 }
2120 }
2121
2122
2123 void InterpreterMacroAssembler::notify_method_exit(
2124 TosState state, NotifyMethodExitMode mode) {
2125 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2126 // track stack depth. If it is possible to enter interp_only_mode we add
2127 // the code to check if the event should be sent.
2128 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2129 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2130 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2131 Label L;
2132 // Note: frame::interpreter_frame_result has a dependency on how the
2133 // method result is saved across the call to post_method_exit. If this
2134 // is changed then the interpreter_frame_result implementation will
2135 // need to be updated too.
2136
2137 // template interpreter will leave the result on the top of the stack.
2138 push(state);
2139 NOT_LP64(get_thread(rthread);)
2140 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2141 testl(rdx, rdx);
2142 jcc(Assembler::zero, L);
2143 call_VM(noreg,
2144 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2145 bind(L);
2146 pop(state);
2147 }
2148
2149 {
2150 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2151 push(state);
2152 NOT_LP64(get_thread(rthread);)
2153 get_method(rarg);
2154 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2155 rthread, rarg);
2156 pop(state);
2157 }
2158 }