1 /*
2 * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "interp_masm_x86.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "logging/log.hpp"
30 #include "oops/arrayOop.hpp"
31 #include "oops/markOop.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/method.hpp"
34 #include "prims/jvmtiExport.hpp"
35 #include "prims/jvmtiThreadState.hpp"
36 #include "runtime/basicLock.hpp"
37 #include "runtime/biasedLocking.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/thread.inline.hpp"
40
41 // Implementation of InterpreterMacroAssembler
42
43 void InterpreterMacroAssembler::jump_to_entry(address entry) {
44 assert(entry, "Entry must have been generated by now");
45 jump(RuntimeAddress(entry));
46 }
47
48 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
49 Label update, next, none;
50
51 verify_oop(obj);
52
53 testptr(obj, obj);
54 jccb(Assembler::notZero, update);
55 orptr(mdo_addr, TypeEntries::null_seen);
56 jmpb(next);
57
58 bind(update);
59 load_klass(obj, obj);
60
61 xorptr(obj, mdo_addr);
62 testptr(obj, TypeEntries::type_klass_mask);
63 jccb(Assembler::zero, next); // klass seen before, nothing to
64 // do. The unknown bit may have been
65 // set already but no need to check.
66
67 testptr(obj, TypeEntries::type_unknown);
68 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
69
70 cmpptr(mdo_addr, 0);
71 jccb(Assembler::equal, none);
72 cmpptr(mdo_addr, TypeEntries::null_seen);
73 jccb(Assembler::equal, none);
74 // There is a chance that the checks above (re-reading profiling
75 // data from memory) fail if another thread has just set the
76 // profiling to this obj's klass
77 xorptr(obj, mdo_addr);
78 testptr(obj, TypeEntries::type_klass_mask);
79 jccb(Assembler::zero, next);
80
81 // different than before. Cannot keep accurate profile.
82 orptr(mdo_addr, TypeEntries::type_unknown);
83 jmpb(next);
84
85 bind(none);
86 // first time here. Set profile type.
87 movptr(mdo_addr, obj);
88
89 bind(next);
90 }
91
92 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
93 if (!ProfileInterpreter) {
94 return;
95 }
96
97 if (MethodData::profile_arguments() || MethodData::profile_return()) {
98 Label profile_continue;
99
100 test_method_data_pointer(mdp, profile_continue);
101
102 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
103
104 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
105 jcc(Assembler::notEqual, profile_continue);
106
107 if (MethodData::profile_arguments()) {
108 Label done;
109 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
110 addptr(mdp, off_to_args);
111
112 for (int i = 0; i < TypeProfileArgsLimit; i++) {
113 if (i > 0 || MethodData::profile_return()) {
114 // If return value type is profiled we may have no argument to profile
115 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
116 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
117 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
118 jcc(Assembler::less, done);
119 }
120 movptr(tmp, Address(callee, Method::const_offset()));
121 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
122 // stack offset o (zero based) from the start of the argument
123 // list, for n arguments translates into offset n - o - 1 from
124 // the end of the argument list
125 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
126 subl(tmp, 1);
127 Address arg_addr = argument_address(tmp);
128 movptr(tmp, arg_addr);
129
130 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
131 profile_obj_type(tmp, mdo_arg_addr);
132
133 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
134 addptr(mdp, to_add);
135 off_to_args += to_add;
136 }
137
138 if (MethodData::profile_return()) {
139 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
140 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
141 }
142
143 bind(done);
144
145 if (MethodData::profile_return()) {
146 // We're right after the type profile for the last
147 // argument. tmp is the number of cells left in the
148 // CallTypeData/VirtualCallTypeData to reach its end. Non null
149 // if there's a return to profile.
150 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
151 shll(tmp, exact_log2(DataLayout::cell_size));
152 addptr(mdp, tmp);
153 }
154 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
155 } else {
156 assert(MethodData::profile_return(), "either profile call args or call ret");
157 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
158 }
159
160 // mdp points right after the end of the
161 // CallTypeData/VirtualCallTypeData, right after the cells for the
162 // return value type if there's one
163
164 bind(profile_continue);
165 }
166 }
167
168 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
169 assert_different_registers(mdp, ret, tmp, _bcp_register);
170 if (ProfileInterpreter && MethodData::profile_return()) {
171 Label profile_continue, done;
172
173 test_method_data_pointer(mdp, profile_continue);
174
175 if (MethodData::profile_return_jsr292_only()) {
176 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
177
178 // If we don't profile all invoke bytecodes we must make sure
179 // it's a bytecode we indeed profile. We can't go back to the
180 // begining of the ProfileData we intend to update to check its
181 // type because we're right after it and we don't known its
182 // length
183 Label do_profile;
184 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
185 jcc(Assembler::equal, do_profile);
186 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
187 jcc(Assembler::equal, do_profile);
188 get_method(tmp);
189 cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm);
190 jcc(Assembler::notEqual, profile_continue);
191
192 bind(do_profile);
193 }
194
195 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
196 mov(tmp, ret);
197 profile_obj_type(tmp, mdo_ret_addr);
198
199 bind(profile_continue);
200 }
201 }
202
203 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
204 if (ProfileInterpreter && MethodData::profile_parameters()) {
205 Label profile_continue, done;
206
207 test_method_data_pointer(mdp, profile_continue);
208
209 // Load the offset of the area within the MDO used for
210 // parameters. If it's negative we're not profiling any parameters
211 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
212 testl(tmp1, tmp1);
213 jcc(Assembler::negative, profile_continue);
214
215 // Compute a pointer to the area for parameters from the offset
216 // and move the pointer to the slot for the last
217 // parameters. Collect profiling from last parameter down.
218 // mdo start + parameters offset + array length - 1
219 addptr(mdp, tmp1);
220 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
221 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
222
223 Label loop;
224 bind(loop);
225
226 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
227 int type_base = in_bytes(ParametersTypeData::type_offset(0));
228 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
229 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
230 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
231
232 // load offset on the stack from the slot for this parameter
233 movptr(tmp2, arg_off);
234 negptr(tmp2);
235 // read the parameter from the local area
236 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
237
238 // profile the parameter
239 profile_obj_type(tmp2, arg_type);
240
241 // go to next parameter
242 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
243 jcc(Assembler::positive, loop);
244
245 bind(profile_continue);
246 }
247 }
248
249 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
250 int number_of_arguments) {
251 // interpreter specific
252 //
253 // Note: No need to save/restore bcp & locals registers
254 // since these are callee saved registers and no blocking/
255 // GC can happen in leaf calls.
256 // Further Note: DO NOT save/restore bcp/locals. If a caller has
257 // already saved them so that it can use rsi/rdi as temporaries
258 // then a save/restore here will DESTROY the copy the caller
259 // saved! There used to be a save_bcp() that only happened in
260 // the ASSERT path (no restore_bcp). Which caused bizarre failures
261 // when jvm built with ASSERTs.
262 #ifdef ASSERT
263 {
264 Label L;
265 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
266 jcc(Assembler::equal, L);
267 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
268 " last_sp != NULL");
269 bind(L);
270 }
271 #endif
272 // super call
273 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
274 // interpreter specific
275 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
276 // but since they may not have been saved (and we don't want to
277 // save them here (see note above) the assert is invalid.
278 }
279
280 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
281 Register java_thread,
282 Register last_java_sp,
283 address entry_point,
284 int number_of_arguments,
285 bool check_exceptions) {
286 // interpreter specific
287 //
288 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
289 // really make a difference for these runtime calls, since they are
290 // slow anyway. Btw., bcp must be saved/restored since it may change
291 // due to GC.
292 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
293 save_bcp();
294 #ifdef ASSERT
295 {
296 Label L;
297 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
298 jcc(Assembler::equal, L);
299 stop("InterpreterMacroAssembler::call_VM_base:"
300 " last_sp != NULL");
301 bind(L);
302 }
303 #endif /* ASSERT */
304 // super call
305 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
306 entry_point, number_of_arguments,
307 check_exceptions);
308 // interpreter specific
309 restore_bcp();
310 restore_locals();
311 }
312
313 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
314 if (JvmtiExport::can_pop_frame()) {
315 Label L;
316 // Initiate popframe handling only if it is not already being
317 // processed. If the flag has the popframe_processing bit set, it
318 // means that this code is called *during* popframe handling - we
319 // don't want to reenter.
320 // This method is only called just after the call into the vm in
321 // call_VM_base, so the arg registers are available.
322 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
323 LP64_ONLY(c_rarg0);
324 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
325 testl(pop_cond, JavaThread::popframe_pending_bit);
326 jcc(Assembler::zero, L);
327 testl(pop_cond, JavaThread::popframe_processing_bit);
328 jcc(Assembler::notZero, L);
329 // Call Interpreter::remove_activation_preserving_args_entry() to get the
330 // address of the same-named entrypoint in the generated interpreter code.
331 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
332 jmp(rax);
333 bind(L);
334 NOT_LP64(get_thread(java_thread);)
335 }
336 }
337
338 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
339 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
340 NOT_LP64(get_thread(thread);)
341 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
342 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
343 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
344 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
345 #ifdef _LP64
346 switch (state) {
347 case atos: movptr(rax, oop_addr);
348 movptr(oop_addr, (int32_t)NULL_WORD);
349 verify_oop(rax, state); break;
350 case ltos: movptr(rax, val_addr); break;
351 case btos: // fall through
352 case ztos: // fall through
353 case ctos: // fall through
354 case stos: // fall through
355 case itos: movl(rax, val_addr); break;
356 case ftos: load_float(val_addr); break;
357 case dtos: load_double(val_addr); break;
358 case vtos: /* nothing to do */ break;
359 default : ShouldNotReachHere();
360 }
361 // Clean up tos value in the thread object
362 movl(tos_addr, (int) ilgl);
363 movl(val_addr, (int32_t) NULL_WORD);
364 #else
365 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
366 + in_ByteSize(wordSize));
367 switch (state) {
368 case atos: movptr(rax, oop_addr);
369 movptr(oop_addr, NULL_WORD);
370 verify_oop(rax, state); break;
371 case ltos:
372 movl(rdx, val_addr1); // fall through
373 case btos: // fall through
374 case ztos: // fall through
375 case ctos: // fall through
376 case stos: // fall through
377 case itos: movl(rax, val_addr); break;
378 case ftos: load_float(val_addr); break;
379 case dtos: load_double(val_addr); break;
380 case vtos: /* nothing to do */ break;
381 default : ShouldNotReachHere();
382 }
383 #endif // _LP64
384 // Clean up tos value in the thread object
385 movl(tos_addr, (int32_t) ilgl);
386 movptr(val_addr, NULL_WORD);
387 NOT_LP64(movptr(val_addr1, NULL_WORD);)
388 }
389
390
391 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
392 if (JvmtiExport::can_force_early_return()) {
393 Label L;
394 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
395 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
396
397 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
398 testptr(tmp, tmp);
399 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
400
401 // Initiate earlyret handling only if it is not already being processed.
402 // If the flag has the earlyret_processing bit set, it means that this code
403 // is called *during* earlyret handling - we don't want to reenter.
404 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
405 cmpl(tmp, JvmtiThreadState::earlyret_pending);
406 jcc(Assembler::notEqual, L);
407
408 // Call Interpreter::remove_activation_early_entry() to get the address of the
409 // same-named entrypoint in the generated interpreter code.
410 NOT_LP64(get_thread(java_thread);)
411 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
412 #ifdef _LP64
413 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
414 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
415 #else
416 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
417 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
418 #endif // _LP64
419 jmp(rax);
420 bind(L);
421 NOT_LP64(get_thread(java_thread);)
422 }
423 }
424
425 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
426 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
427 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
428 bswapl(reg);
429 shrl(reg, 16);
430 }
431
432 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
433 int bcp_offset,
434 size_t index_size) {
435 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
436 if (index_size == sizeof(u2)) {
437 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
438 } else if (index_size == sizeof(u4)) {
439 movl(index, Address(_bcp_register, bcp_offset));
440 // Check if the secondary index definition is still ~x, otherwise
441 // we have to change the following assembler code to calculate the
442 // plain index.
443 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
444 notl(index); // convert to plain index
445 } else if (index_size == sizeof(u1)) {
446 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
447 } else {
448 ShouldNotReachHere();
449 }
450 }
451
452 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
453 Register index,
454 int bcp_offset,
455 size_t index_size) {
456 assert_different_registers(cache, index);
457 get_cache_index_at_bcp(index, bcp_offset, index_size);
458 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
459 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
460 // convert from field index to ConstantPoolCacheEntry index
461 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
462 shll(index, 2);
463 }
464
465 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
466 Register index,
467 Register bytecode,
468 int byte_no,
469 int bcp_offset,
470 size_t index_size) {
471 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
472 // We use a 32-bit load here since the layout of 64-bit words on
473 // little-endian machines allow us that.
474 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
475 const int shift_count = (1 + byte_no) * BitsPerByte;
476 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
477 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
478 "correct shift count");
479 shrl(bytecode, shift_count);
480 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
481 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
482 }
483
484 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
485 Register tmp,
486 int bcp_offset,
487 size_t index_size) {
488 assert(cache != tmp, "must use different register");
489 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
490 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
491 // convert from field index to ConstantPoolCacheEntry index
492 // and from word offset to byte offset
493 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
494 shll(tmp, 2 + LogBytesPerWord);
495 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
496 // skip past the header
497 addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
498 addptr(cache, tmp); // construct pointer to cache entry
499 }
500
501 // Load object from cpool->resolved_references(index)
502 void InterpreterMacroAssembler::load_resolved_reference_at_index(
503 Register result, Register index) {
504 assert_different_registers(result, index);
505 // convert from field index to resolved_references() index and from
506 // word index to byte offset. Since this is a java object, it can be compressed
507 Register tmp = index; // reuse
508 shll(tmp, LogBytesPerHeapOop);
509
510 get_constant_pool(result);
511 // load pointer for resolved_references[] objArray
512 movptr(result, Address(result, ConstantPool::cache_offset_in_bytes()));
513 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes()));
514 resolve_oop_handle(result);
515 // Add in the index
516 addptr(result, tmp);
517 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
518 }
519
520 // load cpool->resolved_klass_at(index)
521 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register cpool,
522 Register index, Register klass) {
523 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
524 Register resolved_klasses = cpool;
525 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes()));
526 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
527 }
528
529 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
530 // subtype of super_klass.
531 //
532 // Args:
533 // rax: superklass
534 // Rsub_klass: subklass
535 //
536 // Kills:
537 // rcx, rdi
538 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
539 Label& ok_is_subtype) {
540 assert(Rsub_klass != rax, "rax holds superklass");
541 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
542 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
543 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
544 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
545
546 // Profile the not-null value's klass.
547 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
548
549 // Do the check.
550 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
551
552 // Profile the failure of the check.
553 profile_typecheck_failed(rcx); // blows rcx
554 }
555
556
557 #ifndef _LP64
558 void InterpreterMacroAssembler::f2ieee() {
559 if (IEEEPrecision) {
560 fstp_s(Address(rsp, 0));
561 fld_s(Address(rsp, 0));
562 }
563 }
564
565
566 void InterpreterMacroAssembler::d2ieee() {
567 if (IEEEPrecision) {
568 fstp_d(Address(rsp, 0));
569 fld_d(Address(rsp, 0));
570 }
571 }
572 #endif // _LP64
573
574 // Java Expression Stack
575
576 void InterpreterMacroAssembler::pop_ptr(Register r) {
577 pop(r);
578 }
579
580 void InterpreterMacroAssembler::push_ptr(Register r) {
581 push(r);
582 }
583
584 void InterpreterMacroAssembler::push_i(Register r) {
585 push(r);
586 }
587
588 void InterpreterMacroAssembler::push_f(XMMRegister r) {
589 subptr(rsp, wordSize);
590 movflt(Address(rsp, 0), r);
591 }
592
593 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
594 movflt(r, Address(rsp, 0));
595 addptr(rsp, wordSize);
596 }
597
598 void InterpreterMacroAssembler::push_d(XMMRegister r) {
599 subptr(rsp, 2 * wordSize);
600 movdbl(Address(rsp, 0), r);
601 }
602
603 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
604 movdbl(r, Address(rsp, 0));
605 addptr(rsp, 2 * Interpreter::stackElementSize);
606 }
607
608 #ifdef _LP64
609 void InterpreterMacroAssembler::pop_i(Register r) {
610 // XXX can't use pop currently, upper half non clean
611 movl(r, Address(rsp, 0));
612 addptr(rsp, wordSize);
613 }
614
615 void InterpreterMacroAssembler::pop_l(Register r) {
616 movq(r, Address(rsp, 0));
617 addptr(rsp, 2 * Interpreter::stackElementSize);
618 }
619
620 void InterpreterMacroAssembler::push_l(Register r) {
621 subptr(rsp, 2 * wordSize);
622 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
623 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
624 }
625
626 void InterpreterMacroAssembler::pop(TosState state) {
627 switch (state) {
628 case atos: pop_ptr(); break;
629 case btos:
630 case ztos:
631 case ctos:
632 case stos:
633 case itos: pop_i(); break;
634 case ltos: pop_l(); break;
635 case ftos: pop_f(xmm0); break;
636 case dtos: pop_d(xmm0); break;
637 case vtos: /* nothing to do */ break;
638 default: ShouldNotReachHere();
639 }
640 verify_oop(rax, state);
641 }
642
643 void InterpreterMacroAssembler::push(TosState state) {
644 verify_oop(rax, state);
645 switch (state) {
646 case atos: push_ptr(); break;
647 case btos:
648 case ztos:
649 case ctos:
650 case stos:
651 case itos: push_i(); break;
652 case ltos: push_l(); break;
653 case ftos: push_f(xmm0); break;
654 case dtos: push_d(xmm0); break;
655 case vtos: /* nothing to do */ break;
656 default : ShouldNotReachHere();
657 }
658 }
659 #else
660 void InterpreterMacroAssembler::pop_i(Register r) {
661 pop(r);
662 }
663
664 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
665 pop(lo);
666 pop(hi);
667 }
668
669 void InterpreterMacroAssembler::pop_f() {
670 fld_s(Address(rsp, 0));
671 addptr(rsp, 1 * wordSize);
672 }
673
674 void InterpreterMacroAssembler::pop_d() {
675 fld_d(Address(rsp, 0));
676 addptr(rsp, 2 * wordSize);
677 }
678
679
680 void InterpreterMacroAssembler::pop(TosState state) {
681 switch (state) {
682 case atos: pop_ptr(rax); break;
683 case btos: // fall through
684 case ztos: // fall through
685 case ctos: // fall through
686 case stos: // fall through
687 case itos: pop_i(rax); break;
688 case ltos: pop_l(rax, rdx); break;
689 case ftos:
690 if (UseSSE >= 1) {
691 pop_f(xmm0);
692 } else {
693 pop_f();
694 }
695 break;
696 case dtos:
697 if (UseSSE >= 2) {
698 pop_d(xmm0);
699 } else {
700 pop_d();
701 }
702 break;
703 case vtos: /* nothing to do */ break;
704 default : ShouldNotReachHere();
705 }
706 verify_oop(rax, state);
707 }
708
709
710 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
711 push(hi);
712 push(lo);
713 }
714
715 void InterpreterMacroAssembler::push_f() {
716 // Do not schedule for no AGI! Never write beyond rsp!
717 subptr(rsp, 1 * wordSize);
718 fstp_s(Address(rsp, 0));
719 }
720
721 void InterpreterMacroAssembler::push_d() {
722 // Do not schedule for no AGI! Never write beyond rsp!
723 subptr(rsp, 2 * wordSize);
724 fstp_d(Address(rsp, 0));
725 }
726
727
728 void InterpreterMacroAssembler::push(TosState state) {
729 verify_oop(rax, state);
730 switch (state) {
731 case atos: push_ptr(rax); break;
732 case btos: // fall through
733 case ztos: // fall through
734 case ctos: // fall through
735 case stos: // fall through
736 case itos: push_i(rax); break;
737 case ltos: push_l(rax, rdx); break;
738 case ftos:
739 if (UseSSE >= 1) {
740 push_f(xmm0);
741 } else {
742 push_f();
743 }
744 break;
745 case dtos:
746 if (UseSSE >= 2) {
747 push_d(xmm0);
748 } else {
749 push_d();
750 }
751 break;
752 case vtos: /* nothing to do */ break;
753 default : ShouldNotReachHere();
754 }
755 }
756 #endif // _LP64
757
758
759 // Helpers for swap and dup
760 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
761 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
762 }
763
764 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
765 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
766 }
767
768
769 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
770 // set sender sp
771 lea(_bcp_register, Address(rsp, wordSize));
772 // record last_sp
773 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
774 }
775
776
777 // Jump to from_interpreted entry of a call unless single stepping is possible
778 // in this thread in which case we must call the i2i entry
779 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
780 prepare_to_jump_from_interpreted();
781
782 if (JvmtiExport::can_post_interpreter_events()) {
783 Label run_compiled_code;
784 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
785 // compiled code in threads for which the event is enabled. Check here for
786 // interp_only_mode if these events CAN be enabled.
787 // interp_only is an int, on little endian it is sufficient to test the byte only
788 // Is a cmpl faster?
789 LP64_ONLY(temp = r15_thread;)
790 NOT_LP64(get_thread(temp);)
791 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
792 jccb(Assembler::zero, run_compiled_code);
793 jmp(Address(method, Method::interpreter_entry_offset()));
794 bind(run_compiled_code);
795 }
796
797 jmp(Address(method, Method::from_interpreted_offset()));
798 }
799
800 // The following two routines provide a hook so that an implementation
801 // can schedule the dispatch in two parts. x86 does not do this.
802 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
803 // Nothing x86 specific to be done here
804 }
805
806 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
807 dispatch_next(state, step);
808 }
809
810 void InterpreterMacroAssembler::dispatch_base(TosState state,
811 address* table,
812 bool verifyoop) {
813 verify_FPU(1, state);
814 if (VerifyActivationFrameSize) {
815 Label L;
816 mov(rcx, rbp);
817 subptr(rcx, rsp);
818 int32_t min_frame_size =
819 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
820 wordSize;
821 cmpptr(rcx, (int32_t)min_frame_size);
822 jcc(Assembler::greaterEqual, L);
823 stop("broken stack frame");
824 bind(L);
825 }
826 if (verifyoop) {
827 verify_oop(rax, state);
828 }
829 #ifdef _LP64
830 lea(rscratch1, ExternalAddress((address)table));
831 jmp(Address(rscratch1, rbx, Address::times_8));
832 #else
833 Address index(noreg, rbx, Address::times_ptr);
834 ExternalAddress tbl((address)table);
835 ArrayAddress dispatch(tbl, index);
836 jump(dispatch);
837 #endif // _LP64
838 }
839
840 void InterpreterMacroAssembler::dispatch_only(TosState state) {
841 dispatch_base(state, Interpreter::dispatch_table(state));
842 }
843
844 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
845 dispatch_base(state, Interpreter::normal_table(state));
846 }
847
848 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
849 dispatch_base(state, Interpreter::normal_table(state), false);
850 }
851
852
853 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
854 // load next bytecode (load before advancing _bcp_register to prevent AGI)
855 load_unsigned_byte(rbx, Address(_bcp_register, step));
856 // advance _bcp_register
857 increment(_bcp_register, step);
858 dispatch_base(state, Interpreter::dispatch_table(state));
859 }
860
861 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
862 // load current bytecode
863 load_unsigned_byte(rbx, Address(_bcp_register, 0));
864 dispatch_base(state, table);
865 }
866
867 void InterpreterMacroAssembler::narrow(Register result) {
868
869 // Get method->_constMethod->_result_type
870 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
871 movptr(rcx, Address(rcx, Method::const_offset()));
872 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
873
874 Label done, notBool, notByte, notChar;
875
876 // common case first
877 cmpl(rcx, T_INT);
878 jcc(Assembler::equal, done);
879
880 // mask integer result to narrower return type.
881 cmpl(rcx, T_BOOLEAN);
882 jcc(Assembler::notEqual, notBool);
883 andl(result, 0x1);
884 jmp(done);
885
886 bind(notBool);
887 cmpl(rcx, T_BYTE);
888 jcc(Assembler::notEqual, notByte);
889 LP64_ONLY(movsbl(result, result);)
890 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
891 NOT_LP64(sarl(result, 24);) // and sign-extend byte
892 jmp(done);
893
894 bind(notByte);
895 cmpl(rcx, T_CHAR);
896 jcc(Assembler::notEqual, notChar);
897 LP64_ONLY(movzwl(result, result);)
898 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
899 jmp(done);
900
901 bind(notChar);
902 // cmpl(rcx, T_SHORT); // all that's left
903 // jcc(Assembler::notEqual, done);
904 LP64_ONLY(movswl(result, result);)
905 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
906 NOT_LP64(sarl(result, 16);) // and sign-extend short
907
908 // Nothing to do for T_INT
909 bind(done);
910 }
911
912 // remove activation
913 //
914 // Unlock the receiver if this is a synchronized method.
915 // Unlock any Java monitors from syncronized blocks.
916 // Remove the activation from the stack.
917 //
918 // If there are locked Java monitors
919 // If throw_monitor_exception
920 // throws IllegalMonitorStateException
921 // Else if install_monitor_exception
922 // installs IllegalMonitorStateException
923 // Else
924 // no error processing
925 void InterpreterMacroAssembler::remove_activation(
926 TosState state,
927 Register ret_addr,
928 bool throw_monitor_exception,
929 bool install_monitor_exception,
930 bool notify_jvmdi) {
931 // Note: Registers rdx xmm0 may be in use for the
932 // result check if synchronized method
933 Label unlocked, unlock, no_unlock;
934
935 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
936 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
937 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
938 // monitor pointers need different register
939 // because rdx may have the result in it
940 NOT_LP64(get_thread(rcx);)
941
942 // get the value of _do_not_unlock_if_synchronized into rdx
943 const Address do_not_unlock_if_synchronized(rthread,
944 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
945 movbool(rbx, do_not_unlock_if_synchronized);
946 movbool(do_not_unlock_if_synchronized, false); // reset the flag
947
948 // get method access flags
949 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
950 movl(rcx, Address(rcx, Method::access_flags_offset()));
951 testl(rcx, JVM_ACC_SYNCHRONIZED);
952 jcc(Assembler::zero, unlocked);
953
954 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
955 // is set.
956 testbool(rbx);
957 jcc(Assembler::notZero, no_unlock);
958
959 // unlock monitor
960 push(state); // save result
961
962 // BasicObjectLock will be first in list, since this is a
963 // synchronized method. However, need to check that the object has
964 // not been unlocked by an explicit monitorexit bytecode.
965 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
966 wordSize - (int) sizeof(BasicObjectLock));
967 // We use c_rarg1/rdx so that if we go slow path it will be the correct
968 // register for unlock_object to pass to VM directly
969 lea(robj, monitor); // address of first monitor
970
971 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
972 testptr(rax, rax);
973 jcc(Assembler::notZero, unlock);
974
975 pop(state);
976 if (throw_monitor_exception) {
977 // Entry already unlocked, need to throw exception
978 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
979 call_VM(noreg, CAST_FROM_FN_PTR(address,
980 InterpreterRuntime::throw_illegal_monitor_state_exception));
981 should_not_reach_here();
982 } else {
983 // Monitor already unlocked during a stack unroll. If requested,
984 // install an illegal_monitor_state_exception. Continue with
985 // stack unrolling.
986 if (install_monitor_exception) {
987 NOT_LP64(empty_FPU_stack();)
988 call_VM(noreg, CAST_FROM_FN_PTR(address,
989 InterpreterRuntime::new_illegal_monitor_state_exception));
990 }
991 jmp(unlocked);
992 }
993
994 bind(unlock);
995 unlock_object(robj);
996 pop(state);
997
998 // Check that for block-structured locking (i.e., that all locked
999 // objects has been unlocked)
1000 bind(unlocked);
1001
1002 // rax, rdx: Might contain return value
1003
1004 // Check that all monitors are unlocked
1005 {
1006 Label loop, exception, entry, restart;
1007 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1008 const Address monitor_block_top(
1009 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1010 const Address monitor_block_bot(
1011 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1012
1013 bind(restart);
1014 // We use c_rarg1 so that if we go slow path it will be the correct
1015 // register for unlock_object to pass to VM directly
1016 movptr(rmon, monitor_block_top); // points to current entry, starting
1017 // with top-most entry
1018 lea(rbx, monitor_block_bot); // points to word before bottom of
1019 // monitor block
1020 jmp(entry);
1021
1022 // Entry already locked, need to throw exception
1023 bind(exception);
1024
1025 if (throw_monitor_exception) {
1026 // Throw exception
1027 NOT_LP64(empty_FPU_stack();)
1028 MacroAssembler::call_VM(noreg,
1029 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1030 throw_illegal_monitor_state_exception));
1031 should_not_reach_here();
1032 } else {
1033 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1034 // Unlock does not block, so don't have to worry about the frame.
1035 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1036
1037 push(state);
1038 mov(robj, rmon); // nop if robj and rmon are the same
1039 unlock_object(robj);
1040 pop(state);
1041
1042 if (install_monitor_exception) {
1043 NOT_LP64(empty_FPU_stack();)
1044 call_VM(noreg, CAST_FROM_FN_PTR(address,
1045 InterpreterRuntime::
1046 new_illegal_monitor_state_exception));
1047 }
1048
1049 jmp(restart);
1050 }
1051
1052 bind(loop);
1053 // check if current entry is used
1054 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
1055 jcc(Assembler::notEqual, exception);
1056
1057 addptr(rmon, entry_size); // otherwise advance to next entry
1058 bind(entry);
1059 cmpptr(rmon, rbx); // check if bottom reached
1060 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1061 }
1062
1063 bind(no_unlock);
1064
1065 // jvmti support
1066 if (notify_jvmdi) {
1067 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1068 } else {
1069 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1070 }
1071
1072 // remove activation
1073 // get sender sp
1074 movptr(rbx,
1075 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1076 if (StackReservedPages > 0) {
1077 // testing if reserved zone needs to be re-enabled
1078 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1079 Label no_reserved_zone_enabling;
1080
1081 NOT_LP64(get_thread(rthread);)
1082
1083 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_enabled);
1084 jcc(Assembler::equal, no_reserved_zone_enabling);
1085
1086 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1087 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1088
1089 call_VM_leaf(
1090 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1091 call_VM(noreg, CAST_FROM_FN_PTR(address,
1092 InterpreterRuntime::throw_delayed_StackOverflowError));
1093 should_not_reach_here();
1094
1095 bind(no_reserved_zone_enabling);
1096 }
1097 leave(); // remove frame anchor
1098 pop(ret_addr); // get return address
1099 mov(rsp, rbx); // set sp to sender sp
1100 }
1101
1102 void InterpreterMacroAssembler::get_method_counters(Register method,
1103 Register mcs, Label& skip) {
1104 Label has_counters;
1105 movptr(mcs, Address(method, Method::method_counters_offset()));
1106 testptr(mcs, mcs);
1107 jcc(Assembler::notZero, has_counters);
1108 call_VM(noreg, CAST_FROM_FN_PTR(address,
1109 InterpreterRuntime::build_method_counters), method);
1110 movptr(mcs, Address(method,Method::method_counters_offset()));
1111 testptr(mcs, mcs);
1112 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1113 bind(has_counters);
1114 }
1115
1116
1117 // Lock object
1118 //
1119 // Args:
1120 // rdx, c_rarg1: BasicObjectLock to be used for locking
1121 //
1122 // Kills:
1123 // rax, rbx
1124 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1125 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1126 "The argument is only for looks. It must be c_rarg1");
1127
1128 if (UseHeavyMonitors) {
1129 call_VM(noreg,
1130 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1131 lock_reg);
1132 } else {
1133 Label done;
1134
1135 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1136 const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a
1137 // problematic case where tmp_reg = no_reg.
1138 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1139
1140 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
1141 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
1142 const int mark_offset = lock_offset +
1143 BasicLock::displaced_header_offset_in_bytes();
1144
1145 Label slow_case;
1146
1147 // Load object pointer into obj_reg
1148 movptr(obj_reg, Address(lock_reg, obj_offset));
1149
1150 if (UseBiasedLocking) {
1151 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case);
1152 }
1153
1154 // Load immediate 1 into swap_reg %rax
1155 movl(swap_reg, (int32_t)1);
1156
1157 // Load (object->mark() | 1) into swap_reg %rax
1158 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1159
1160 // Save (object->mark() | 1) into BasicLock's displaced header
1161 movptr(Address(lock_reg, mark_offset), swap_reg);
1162
1163 assert(lock_offset == 0,
1164 "displaced header must be first word in BasicObjectLock");
1165
1166 if (os::is_MP()) lock();
1167 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1168 if (PrintBiasedLockingStatistics) {
1169 cond_inc32(Assembler::zero,
1170 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1171 }
1172 jcc(Assembler::zero, done);
1173
1174 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1175
1176 // Test if the oopMark is an obvious stack pointer, i.e.,
1177 // 1) (mark & zero_bits) == 0, and
1178 // 2) rsp <= mark < mark + os::pagesize()
1179 //
1180 // These 3 tests can be done by evaluating the following
1181 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1182 // assuming both stack pointer and pagesize have their
1183 // least significant bits clear.
1184 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
1185 subptr(swap_reg, rsp);
1186 andptr(swap_reg, zero_bits - os::vm_page_size());
1187
1188 // Save the test result, for recursive case, the result is zero
1189 movptr(Address(lock_reg, mark_offset), swap_reg);
1190
1191 if (PrintBiasedLockingStatistics) {
1192 cond_inc32(Assembler::zero,
1193 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1194 }
1195 jcc(Assembler::zero, done);
1196
1197 bind(slow_case);
1198
1199 // Call the runtime routine for slow case
1200 call_VM(noreg,
1201 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1202 lock_reg);
1203
1204 bind(done);
1205 }
1206 }
1207
1208
1209 // Unlocks an object. Used in monitorexit bytecode and
1210 // remove_activation. Throws an IllegalMonitorException if object is
1211 // not locked by current thread.
1212 //
1213 // Args:
1214 // rdx, c_rarg1: BasicObjectLock for lock
1215 //
1216 // Kills:
1217 // rax
1218 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1219 // rscratch1 (scratch reg)
1220 // rax, rbx, rcx, rdx
1221 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1222 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1223 "The argument is only for looks. It must be c_rarg1");
1224
1225 if (UseHeavyMonitors) {
1226 call_VM(noreg,
1227 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1228 lock_reg);
1229 } else {
1230 Label done;
1231
1232 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1233 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1234 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1235
1236 save_bcp(); // Save in case of exception
1237
1238 // Convert from BasicObjectLock structure to object and BasicLock
1239 // structure Store the BasicLock address into %rax
1240 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
1241
1242 // Load oop into obj_reg(%c_rarg3)
1243 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
1244
1245 // Free entry
1246 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
1247
1248 if (UseBiasedLocking) {
1249 biased_locking_exit(obj_reg, header_reg, done);
1250 }
1251
1252 // Load the old header from BasicLock structure
1253 movptr(header_reg, Address(swap_reg,
1254 BasicLock::displaced_header_offset_in_bytes()));
1255
1256 // Test for recursion
1257 testptr(header_reg, header_reg);
1258
1259 // zero for recursive case
1260 jcc(Assembler::zero, done);
1261
1262 // Atomic swap back the old header
1263 if (os::is_MP()) lock();
1264 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1265
1266 // zero for simple unlock of a stack-lock case
1267 jcc(Assembler::zero, done);
1268
1269 // Call the runtime routine for slow case.
1270 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
1271 obj_reg); // restore obj
1272 call_VM(noreg,
1273 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1274 lock_reg);
1275
1276 bind(done);
1277
1278 restore_bcp();
1279 }
1280 }
1281
1282 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1283 Label& zero_continue) {
1284 assert(ProfileInterpreter, "must be profiling interpreter");
1285 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1286 testptr(mdp, mdp);
1287 jcc(Assembler::zero, zero_continue);
1288 }
1289
1290
1291 // Set the method data pointer for the current bcp.
1292 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1293 assert(ProfileInterpreter, "must be profiling interpreter");
1294 Label set_mdp;
1295 push(rax);
1296 push(rbx);
1297
1298 get_method(rbx);
1299 // Test MDO to avoid the call if it is NULL.
1300 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1301 testptr(rax, rax);
1302 jcc(Assembler::zero, set_mdp);
1303 // rbx: method
1304 // _bcp_register: bcp
1305 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1306 // rax: mdi
1307 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1308 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1309 addptr(rbx, in_bytes(MethodData::data_offset()));
1310 addptr(rax, rbx);
1311 bind(set_mdp);
1312 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1313 pop(rbx);
1314 pop(rax);
1315 }
1316
1317 void InterpreterMacroAssembler::verify_method_data_pointer() {
1318 assert(ProfileInterpreter, "must be profiling interpreter");
1319 #ifdef ASSERT
1320 Label verify_continue;
1321 push(rax);
1322 push(rbx);
1323 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1324 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1325 push(arg3_reg);
1326 push(arg2_reg);
1327 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1328 get_method(rbx);
1329
1330 // If the mdp is valid, it will point to a DataLayout header which is
1331 // consistent with the bcp. The converse is highly probable also.
1332 load_unsigned_short(arg2_reg,
1333 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1334 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1335 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1336 cmpptr(arg2_reg, _bcp_register);
1337 jcc(Assembler::equal, verify_continue);
1338 // rbx: method
1339 // _bcp_register: bcp
1340 // c_rarg3: mdp
1341 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1342 rbx, _bcp_register, arg3_reg);
1343 bind(verify_continue);
1344 pop(arg2_reg);
1345 pop(arg3_reg);
1346 pop(rbx);
1347 pop(rax);
1348 #endif // ASSERT
1349 }
1350
1351
1352 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1353 int constant,
1354 Register value) {
1355 assert(ProfileInterpreter, "must be profiling interpreter");
1356 Address data(mdp_in, constant);
1357 movptr(data, value);
1358 }
1359
1360
1361 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1362 int constant,
1363 bool decrement) {
1364 // Counter address
1365 Address data(mdp_in, constant);
1366
1367 increment_mdp_data_at(data, decrement);
1368 }
1369
1370 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1371 bool decrement) {
1372 assert(ProfileInterpreter, "must be profiling interpreter");
1373 // %%% this does 64bit counters at best it is wasting space
1374 // at worst it is a rare bug when counters overflow
1375
1376 if (decrement) {
1377 // Decrement the register. Set condition codes.
1378 addptr(data, (int32_t) -DataLayout::counter_increment);
1379 // If the decrement causes the counter to overflow, stay negative
1380 Label L;
1381 jcc(Assembler::negative, L);
1382 addptr(data, (int32_t) DataLayout::counter_increment);
1383 bind(L);
1384 } else {
1385 assert(DataLayout::counter_increment == 1,
1386 "flow-free idiom only works with 1");
1387 // Increment the register. Set carry flag.
1388 addptr(data, DataLayout::counter_increment);
1389 // If the increment causes the counter to overflow, pull back by 1.
1390 sbbptr(data, (int32_t)0);
1391 }
1392 }
1393
1394
1395 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1396 Register reg,
1397 int constant,
1398 bool decrement) {
1399 Address data(mdp_in, reg, Address::times_1, constant);
1400
1401 increment_mdp_data_at(data, decrement);
1402 }
1403
1404 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1405 int flag_byte_constant) {
1406 assert(ProfileInterpreter, "must be profiling interpreter");
1407 int header_offset = in_bytes(DataLayout::header_offset());
1408 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
1409 // Set the flag
1410 orl(Address(mdp_in, header_offset), header_bits);
1411 }
1412
1413
1414
1415 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1416 int offset,
1417 Register value,
1418 Register test_value_out,
1419 Label& not_equal_continue) {
1420 assert(ProfileInterpreter, "must be profiling interpreter");
1421 if (test_value_out == noreg) {
1422 cmpptr(value, Address(mdp_in, offset));
1423 } else {
1424 // Put the test value into a register, so caller can use it:
1425 movptr(test_value_out, Address(mdp_in, offset));
1426 cmpptr(test_value_out, value);
1427 }
1428 jcc(Assembler::notEqual, not_equal_continue);
1429 }
1430
1431
1432 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1433 int offset_of_disp) {
1434 assert(ProfileInterpreter, "must be profiling interpreter");
1435 Address disp_address(mdp_in, offset_of_disp);
1436 addptr(mdp_in, disp_address);
1437 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1438 }
1439
1440
1441 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1442 Register reg,
1443 int offset_of_disp) {
1444 assert(ProfileInterpreter, "must be profiling interpreter");
1445 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1446 addptr(mdp_in, disp_address);
1447 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1448 }
1449
1450
1451 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1452 int constant) {
1453 assert(ProfileInterpreter, "must be profiling interpreter");
1454 addptr(mdp_in, constant);
1455 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1456 }
1457
1458
1459 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1460 assert(ProfileInterpreter, "must be profiling interpreter");
1461 push(return_bci); // save/restore across call_VM
1462 call_VM(noreg,
1463 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1464 return_bci);
1465 pop(return_bci);
1466 }
1467
1468
1469 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1470 Register bumped_count) {
1471 if (ProfileInterpreter) {
1472 Label profile_continue;
1473
1474 // If no method data exists, go to profile_continue.
1475 // Otherwise, assign to mdp
1476 test_method_data_pointer(mdp, profile_continue);
1477
1478 // We are taking a branch. Increment the taken count.
1479 // We inline increment_mdp_data_at to return bumped_count in a register
1480 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1481 Address data(mdp, in_bytes(JumpData::taken_offset()));
1482 movptr(bumped_count, data);
1483 assert(DataLayout::counter_increment == 1,
1484 "flow-free idiom only works with 1");
1485 addptr(bumped_count, DataLayout::counter_increment);
1486 sbbptr(bumped_count, 0);
1487 movptr(data, bumped_count); // Store back out
1488
1489 // The method data pointer needs to be updated to reflect the new target.
1490 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1491 bind(profile_continue);
1492 }
1493 }
1494
1495
1496 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1497 if (ProfileInterpreter) {
1498 Label profile_continue;
1499
1500 // If no method data exists, go to profile_continue.
1501 test_method_data_pointer(mdp, profile_continue);
1502
1503 // We are taking a branch. Increment the not taken count.
1504 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1505
1506 // The method data pointer needs to be updated to correspond to
1507 // the next bytecode
1508 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1509 bind(profile_continue);
1510 }
1511 }
1512
1513 void InterpreterMacroAssembler::profile_call(Register mdp) {
1514 if (ProfileInterpreter) {
1515 Label profile_continue;
1516
1517 // If no method data exists, go to profile_continue.
1518 test_method_data_pointer(mdp, profile_continue);
1519
1520 // We are making a call. Increment the count.
1521 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1522
1523 // The method data pointer needs to be updated to reflect the new target.
1524 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1525 bind(profile_continue);
1526 }
1527 }
1528
1529
1530 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1531 if (ProfileInterpreter) {
1532 Label profile_continue;
1533
1534 // If no method data exists, go to profile_continue.
1535 test_method_data_pointer(mdp, profile_continue);
1536
1537 // We are making a call. Increment the count.
1538 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1539
1540 // The method data pointer needs to be updated to reflect the new target.
1541 update_mdp_by_constant(mdp,
1542 in_bytes(VirtualCallData::
1543 virtual_call_data_size()));
1544 bind(profile_continue);
1545 }
1546 }
1547
1548
1549 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1550 Register mdp,
1551 Register reg2,
1552 bool receiver_can_be_null) {
1553 if (ProfileInterpreter) {
1554 Label profile_continue;
1555
1556 // If no method data exists, go to profile_continue.
1557 test_method_data_pointer(mdp, profile_continue);
1558
1559 Label skip_receiver_profile;
1560 if (receiver_can_be_null) {
1561 Label not_null;
1562 testptr(receiver, receiver);
1563 jccb(Assembler::notZero, not_null);
1564 // We are making a call. Increment the count for null receiver.
1565 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1566 jmp(skip_receiver_profile);
1567 bind(not_null);
1568 }
1569
1570 // Record the receiver type.
1571 record_klass_in_profile(receiver, mdp, reg2, true);
1572 bind(skip_receiver_profile);
1573
1574 // The method data pointer needs to be updated to reflect the new target.
1575 #if INCLUDE_JVMCI
1576 if (MethodProfileWidth == 0) {
1577 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1578 }
1579 #else // INCLUDE_JVMCI
1580 update_mdp_by_constant(mdp,
1581 in_bytes(VirtualCallData::
1582 virtual_call_data_size()));
1583 #endif // INCLUDE_JVMCI
1584 bind(profile_continue);
1585 }
1586 }
1587
1588 #if INCLUDE_JVMCI
1589 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) {
1590 assert_different_registers(method, mdp, reg2);
1591 if (ProfileInterpreter && MethodProfileWidth > 0) {
1592 Label profile_continue;
1593
1594 // If no method data exists, go to profile_continue.
1595 test_method_data_pointer(mdp, profile_continue);
1596
1597 Label done;
1598 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth,
1599 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1600 bind(done);
1601
1602 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1603 bind(profile_continue);
1604 }
1605 }
1606 #endif // INCLUDE_JVMCI
1607
1608 // This routine creates a state machine for updating the multi-row
1609 // type profile at a virtual call site (or other type-sensitive bytecode).
1610 // The machine visits each row (of receiver/count) until the receiver type
1611 // is found, or until it runs out of rows. At the same time, it remembers
1612 // the location of the first empty row. (An empty row records null for its
1613 // receiver, and can be allocated for a newly-observed receiver type.)
1614 // Because there are two degrees of freedom in the state, a simple linear
1615 // search will not work; it must be a decision tree. Hence this helper
1616 // function is recursive, to generate the required tree structured code.
1617 // It's the interpreter, so we are trading off code space for speed.
1618 // See below for example code.
1619 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1620 Register receiver, Register mdp,
1621 Register reg2, int start_row,
1622 Label& done, bool is_virtual_call) {
1623 if (TypeProfileWidth == 0) {
1624 if (is_virtual_call) {
1625 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1626 }
1627 #if INCLUDE_JVMCI
1628 else if (EnableJVMCI) {
1629 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1630 }
1631 #endif // INCLUDE_JVMCI
1632 } else {
1633 int non_profiled_offset = -1;
1634 if (is_virtual_call) {
1635 non_profiled_offset = in_bytes(CounterData::count_offset());
1636 }
1637 #if INCLUDE_JVMCI
1638 else if (EnableJVMCI) {
1639 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1640 }
1641 #endif // INCLUDE_JVMCI
1642
1643 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1644 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1645 }
1646 }
1647
1648 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1649 Register reg2, int start_row, Label& done, int total_rows,
1650 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1651 int non_profiled_offset) {
1652 int last_row = total_rows - 1;
1653 assert(start_row <= last_row, "must be work left to do");
1654 // Test this row for both the item and for null.
1655 // Take any of three different outcomes:
1656 // 1. found item => increment count and goto done
1657 // 2. found null => keep looking for case 1, maybe allocate this cell
1658 // 3. found something else => keep looking for cases 1 and 2
1659 // Case 3 is handled by a recursive call.
1660 for (int row = start_row; row <= last_row; row++) {
1661 Label next_test;
1662 bool test_for_null_also = (row == start_row);
1663
1664 // See if the item is item[n].
1665 int item_offset = in_bytes(item_offset_fn(row));
1666 test_mdp_data_at(mdp, item_offset, item,
1667 (test_for_null_also ? reg2 : noreg),
1668 next_test);
1669 // (Reg2 now contains the item from the CallData.)
1670
1671 // The item is item[n]. Increment count[n].
1672 int count_offset = in_bytes(item_count_offset_fn(row));
1673 increment_mdp_data_at(mdp, count_offset);
1674 jmp(done);
1675 bind(next_test);
1676
1677 if (test_for_null_also) {
1678 Label found_null;
1679 // Failed the equality check on item[n]... Test for null.
1680 testptr(reg2, reg2);
1681 if (start_row == last_row) {
1682 // The only thing left to do is handle the null case.
1683 if (non_profiled_offset >= 0) {
1684 jccb(Assembler::zero, found_null);
1685 // Item did not match any saved item and there is no empty row for it.
1686 // Increment total counter to indicate polymorphic case.
1687 increment_mdp_data_at(mdp, non_profiled_offset);
1688 jmp(done);
1689 bind(found_null);
1690 } else {
1691 jcc(Assembler::notZero, done);
1692 }
1693 break;
1694 }
1695 // Since null is rare, make it be the branch-taken case.
1696 jcc(Assembler::zero, found_null);
1697
1698 // Put all the "Case 3" tests here.
1699 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1700 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1701
1702 // Found a null. Keep searching for a matching item,
1703 // but remember that this is an empty (unused) slot.
1704 bind(found_null);
1705 }
1706 }
1707
1708 // In the fall-through case, we found no matching item, but we
1709 // observed the item[start_row] is NULL.
1710
1711 // Fill in the item field and increment the count.
1712 int item_offset = in_bytes(item_offset_fn(start_row));
1713 set_mdp_data_at(mdp, item_offset, item);
1714 int count_offset = in_bytes(item_count_offset_fn(start_row));
1715 movl(reg2, DataLayout::counter_increment);
1716 set_mdp_data_at(mdp, count_offset, reg2);
1717 if (start_row > 0) {
1718 jmp(done);
1719 }
1720 }
1721
1722 // Example state machine code for three profile rows:
1723 // // main copy of decision tree, rooted at row[1]
1724 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1725 // if (row[0].rec != NULL) {
1726 // // inner copy of decision tree, rooted at row[1]
1727 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1728 // if (row[1].rec != NULL) {
1729 // // degenerate decision tree, rooted at row[2]
1730 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1731 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1732 // row[2].init(rec); goto done;
1733 // } else {
1734 // // remember row[1] is empty
1735 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1736 // row[1].init(rec); goto done;
1737 // }
1738 // } else {
1739 // // remember row[0] is empty
1740 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1741 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1742 // row[0].init(rec); goto done;
1743 // }
1744 // done:
1745
1746 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1747 Register mdp, Register reg2,
1748 bool is_virtual_call) {
1749 assert(ProfileInterpreter, "must be profiling");
1750 Label done;
1751
1752 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1753
1754 bind (done);
1755 }
1756
1757 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1758 Register mdp) {
1759 if (ProfileInterpreter) {
1760 Label profile_continue;
1761 uint row;
1762
1763 // If no method data exists, go to profile_continue.
1764 test_method_data_pointer(mdp, profile_continue);
1765
1766 // Update the total ret count.
1767 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1768
1769 for (row = 0; row < RetData::row_limit(); row++) {
1770 Label next_test;
1771
1772 // See if return_bci is equal to bci[n]:
1773 test_mdp_data_at(mdp,
1774 in_bytes(RetData::bci_offset(row)),
1775 return_bci, noreg,
1776 next_test);
1777
1778 // return_bci is equal to bci[n]. Increment the count.
1779 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1780
1781 // The method data pointer needs to be updated to reflect the new target.
1782 update_mdp_by_offset(mdp,
1783 in_bytes(RetData::bci_displacement_offset(row)));
1784 jmp(profile_continue);
1785 bind(next_test);
1786 }
1787
1788 update_mdp_for_ret(return_bci);
1789
1790 bind(profile_continue);
1791 }
1792 }
1793
1794
1795 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1796 if (ProfileInterpreter) {
1797 Label profile_continue;
1798
1799 // If no method data exists, go to profile_continue.
1800 test_method_data_pointer(mdp, profile_continue);
1801
1802 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1803
1804 // The method data pointer needs to be updated.
1805 int mdp_delta = in_bytes(BitData::bit_data_size());
1806 if (TypeProfileCasts) {
1807 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1808 }
1809 update_mdp_by_constant(mdp, mdp_delta);
1810
1811 bind(profile_continue);
1812 }
1813 }
1814
1815
1816 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1817 if (ProfileInterpreter && TypeProfileCasts) {
1818 Label profile_continue;
1819
1820 // If no method data exists, go to profile_continue.
1821 test_method_data_pointer(mdp, profile_continue);
1822
1823 int count_offset = in_bytes(CounterData::count_offset());
1824 // Back up the address, since we have already bumped the mdp.
1825 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1826
1827 // *Decrement* the counter. We expect to see zero or small negatives.
1828 increment_mdp_data_at(mdp, count_offset, true);
1829
1830 bind (profile_continue);
1831 }
1832 }
1833
1834
1835 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1836 if (ProfileInterpreter) {
1837 Label profile_continue;
1838
1839 // If no method data exists, go to profile_continue.
1840 test_method_data_pointer(mdp, profile_continue);
1841
1842 // The method data pointer needs to be updated.
1843 int mdp_delta = in_bytes(BitData::bit_data_size());
1844 if (TypeProfileCasts) {
1845 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1846
1847 // Record the object type.
1848 record_klass_in_profile(klass, mdp, reg2, false);
1849 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1850 NOT_LP64(restore_locals();) // Restore EDI
1851 }
1852 update_mdp_by_constant(mdp, mdp_delta);
1853
1854 bind(profile_continue);
1855 }
1856 }
1857
1858
1859 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1860 if (ProfileInterpreter) {
1861 Label profile_continue;
1862
1863 // If no method data exists, go to profile_continue.
1864 test_method_data_pointer(mdp, profile_continue);
1865
1866 // Update the default case count
1867 increment_mdp_data_at(mdp,
1868 in_bytes(MultiBranchData::default_count_offset()));
1869
1870 // The method data pointer needs to be updated.
1871 update_mdp_by_offset(mdp,
1872 in_bytes(MultiBranchData::
1873 default_displacement_offset()));
1874
1875 bind(profile_continue);
1876 }
1877 }
1878
1879
1880 void InterpreterMacroAssembler::profile_switch_case(Register index,
1881 Register mdp,
1882 Register reg2) {
1883 if (ProfileInterpreter) {
1884 Label profile_continue;
1885
1886 // If no method data exists, go to profile_continue.
1887 test_method_data_pointer(mdp, profile_continue);
1888
1889 // Build the base (index * per_case_size_in_bytes()) +
1890 // case_array_offset_in_bytes()
1891 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1892 imulptr(index, reg2); // XXX l ?
1893 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1894
1895 // Update the case count
1896 increment_mdp_data_at(mdp,
1897 index,
1898 in_bytes(MultiBranchData::relative_count_offset()));
1899
1900 // The method data pointer needs to be updated.
1901 update_mdp_by_offset(mdp,
1902 index,
1903 in_bytes(MultiBranchData::
1904 relative_displacement_offset()));
1905
1906 bind(profile_continue);
1907 }
1908 }
1909
1910
1911
1912 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1913 if (state == atos) {
1914 MacroAssembler::verify_oop(reg);
1915 }
1916 }
1917
1918 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1919 #ifndef _LP64
1920 if ((state == ftos && UseSSE < 1) ||
1921 (state == dtos && UseSSE < 2)) {
1922 MacroAssembler::verify_FPU(stack_depth);
1923 }
1924 #endif
1925 }
1926
1927 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1928 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1929 int increment, Address mask,
1930 Register scratch, bool preloaded,
1931 Condition cond, Label* where) {
1932 if (!preloaded) {
1933 movl(scratch, counter_addr);
1934 }
1935 incrementl(scratch, increment);
1936 movl(counter_addr, scratch);
1937 andl(scratch, mask);
1938 jcc(cond, *where);
1939 }
1940
1941 void InterpreterMacroAssembler::notify_method_entry() {
1942 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1943 // track stack depth. If it is possible to enter interp_only_mode we add
1944 // the code to check if the event should be sent.
1945 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1946 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1947 if (JvmtiExport::can_post_interpreter_events()) {
1948 Label L;
1949 NOT_LP64(get_thread(rthread);)
1950 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1951 testl(rdx, rdx);
1952 jcc(Assembler::zero, L);
1953 call_VM(noreg, CAST_FROM_FN_PTR(address,
1954 InterpreterRuntime::post_method_entry));
1955 bind(L);
1956 }
1957
1958 {
1959 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1960 NOT_LP64(get_thread(rthread);)
1961 get_method(rarg);
1962 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1963 rthread, rarg);
1964 }
1965
1966 // RedefineClasses() tracing support for obsolete method entry
1967 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1968 NOT_LP64(get_thread(rthread);)
1969 get_method(rarg);
1970 call_VM_leaf(
1971 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1972 rthread, rarg);
1973 }
1974 }
1975
1976
1977 void InterpreterMacroAssembler::notify_method_exit(
1978 TosState state, NotifyMethodExitMode mode) {
1979 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1980 // track stack depth. If it is possible to enter interp_only_mode we add
1981 // the code to check if the event should be sent.
1982 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1983 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1984 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1985 Label L;
1986 // Note: frame::interpreter_frame_result has a dependency on how the
1987 // method result is saved across the call to post_method_exit. If this
1988 // is changed then the interpreter_frame_result implementation will
1989 // need to be updated too.
1990
1991 // template interpreter will leave the result on the top of the stack.
1992 push(state);
1993 NOT_LP64(get_thread(rthread);)
1994 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1995 testl(rdx, rdx);
1996 jcc(Assembler::zero, L);
1997 call_VM(noreg,
1998 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1999 bind(L);
2000 pop(state);
2001 }
2002
2003 {
2004 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2005 push(state);
2006 NOT_LP64(get_thread(rthread);)
2007 get_method(rarg);
2008 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2009 rthread, rarg);
2010 pop(state);
2011 }
2012 }