1 /*
2 * Copyright (c) 1997, 2015, 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 "oops/arrayOop.hpp"
30 #include "oops/markOop.hpp"
31 #include "oops/methodData.hpp"
32 #include "oops/method.hpp"
33 #include "prims/jvmtiExport.hpp"
34 #include "prims/jvmtiRedefineClassesTrace.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 #ifndef CC_INTERP
49 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
50 Label update, next, none;
51
52 verify_oop(obj);
53
54 testptr(obj, obj);
55 jccb(Assembler::notZero, update);
56 orptr(mdo_addr, TypeEntries::null_seen);
57 jmpb(next);
58
59 bind(update);
60 load_klass(obj, obj);
61
62 xorptr(obj, mdo_addr);
63 testptr(obj, TypeEntries::type_klass_mask);
64 jccb(Assembler::zero, next); // klass seen before, nothing to
65 // do. The unknown bit may have been
66 // set already but no need to check.
67
68 testptr(obj, TypeEntries::type_unknown);
69 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
70
71 cmpptr(mdo_addr, 0);
72 jccb(Assembler::equal, none);
73 cmpptr(mdo_addr, TypeEntries::null_seen);
74 jccb(Assembler::equal, none);
75 // There is a chance that the checks above (re-reading profiling
76 // data from memory) fail if another thread has just set the
77 // profiling to this obj's klass
78 xorptr(obj, mdo_addr);
79 testptr(obj, TypeEntries::type_klass_mask);
80 jccb(Assembler::zero, next);
81
82 // different than before. Cannot keep accurate profile.
83 orptr(mdo_addr, TypeEntries::type_unknown);
84 jmpb(next);
85
86 bind(none);
87 // first time here. Set profile type.
88 movptr(mdo_addr, obj);
89
90 bind(next);
91 }
92
93 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
94 if (!ProfileInterpreter) {
95 return;
96 }
97
98 if (MethodData::profile_arguments() || MethodData::profile_return()) {
99 Label profile_continue;
100
101 test_method_data_pointer(mdp, profile_continue);
102
103 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
104
105 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
106 jcc(Assembler::notEqual, profile_continue);
107
108 if (MethodData::profile_arguments()) {
109 Label done;
110 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
111 addptr(mdp, off_to_args);
112
113 for (int i = 0; i < TypeProfileArgsLimit; i++) {
114 if (i > 0 || MethodData::profile_return()) {
115 // If return value type is profiled we may have no argument to profile
116 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
117 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
118 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
119 jcc(Assembler::less, done);
120 }
121 movptr(tmp, Address(callee, Method::const_offset()));
122 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
123 // stack offset o (zero based) from the start of the argument
124 // list, for n arguments translates into offset n - o - 1 from
125 // the end of the argument list
126 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
127 subl(tmp, 1);
128 Address arg_addr = argument_address(tmp);
129 movptr(tmp, arg_addr);
130
131 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
132 profile_obj_type(tmp, mdo_arg_addr);
133
134 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
135 addptr(mdp, to_add);
136 off_to_args += to_add;
137 }
138
139 if (MethodData::profile_return()) {
140 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
141 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
142 }
143
144 bind(done);
145
146 if (MethodData::profile_return()) {
147 // We're right after the type profile for the last
148 // argument. tmp is the number of cells left in the
149 // CallTypeData/VirtualCallTypeData to reach its end. Non null
150 // if there's a return to profile.
151 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
152 shll(tmp, exact_log2(DataLayout::cell_size));
153 addptr(mdp, tmp);
154 }
155 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
156 } else {
157 assert(MethodData::profile_return(), "either profile call args or call ret");
158 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
159 }
160
161 // mdp points right after the end of the
162 // CallTypeData/VirtualCallTypeData, right after the cells for the
163 // return value type if there's one
164
165 bind(profile_continue);
166 }
167 }
168
169 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
170 assert_different_registers(mdp, ret, tmp, _bcp_register);
171 if (ProfileInterpreter && MethodData::profile_return()) {
172 Label profile_continue, done;
173
174 test_method_data_pointer(mdp, profile_continue);
175
176 if (MethodData::profile_return_jsr292_only()) {
177 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
178
179 // If we don't profile all invoke bytecodes we must make sure
180 // it's a bytecode we indeed profile. We can't go back to the
181 // begining of the ProfileData we intend to update to check its
182 // type because we're right after it and we don't known its
183 // length
184 Label do_profile;
185 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
186 jcc(Assembler::equal, do_profile);
187 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
188 jcc(Assembler::equal, do_profile);
189 get_method(tmp);
190 cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm);
191 jcc(Assembler::notEqual, profile_continue);
192
193 bind(do_profile);
194 }
195
196 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
197 mov(tmp, ret);
198 profile_obj_type(tmp, mdo_ret_addr);
199
200 bind(profile_continue);
201 }
202 }
203
204 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
205 if (ProfileInterpreter && MethodData::profile_parameters()) {
206 Label profile_continue, done;
207
208 test_method_data_pointer(mdp, profile_continue);
209
210 // Load the offset of the area within the MDO used for
211 // parameters. If it's negative we're not profiling any parameters
212 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
213 testl(tmp1, tmp1);
214 jcc(Assembler::negative, profile_continue);
215
216 // Compute a pointer to the area for parameters from the offset
217 // and move the pointer to the slot for the last
218 // parameters. Collect profiling from last parameter down.
219 // mdo start + parameters offset + array length - 1
220 addptr(mdp, tmp1);
221 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
222 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
223
224 Label loop;
225 bind(loop);
226
227 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
228 int type_base = in_bytes(ParametersTypeData::type_offset(0));
229 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
230 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
231 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
232
233 // load offset on the stack from the slot for this parameter
234 movptr(tmp2, arg_off);
235 negptr(tmp2);
236 // read the parameter from the local area
237 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
238
239 // profile the parameter
240 profile_obj_type(tmp2, arg_type);
241
242 // go to next parameter
243 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
244 jcc(Assembler::positive, loop);
245
246 bind(profile_continue);
247 }
248 }
249 #endif
250
251 #ifdef CC_INTERP
252 void InterpreterMacroAssembler::get_method(Register reg) {
253 movptr(reg, Address(rbp, -(sizeof(BytecodeInterpreter) + 2 * wordSize)));
254 movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method)));
255 }
256 #endif // CC_INTERP
257
258 #ifndef CC_INTERP
259 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
260 int number_of_arguments) {
261 // interpreter specific
262 //
263 // Note: No need to save/restore bcp & locals registers
264 // since these are callee saved registers and no blocking/
265 // GC can happen in leaf calls.
266 // Further Note: DO NOT save/restore bcp/locals. If a caller has
267 // already saved them so that it can use rsi/rdi as temporaries
268 // then a save/restore here will DESTROY the copy the caller
269 // saved! There used to be a save_bcp() that only happened in
270 // the ASSERT path (no restore_bcp). Which caused bizarre failures
271 // when jvm built with ASSERTs.
272 #ifdef ASSERT
273 {
274 Label L;
275 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
276 jcc(Assembler::equal, L);
277 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
278 " last_sp != NULL");
279 bind(L);
280 }
281 #endif
282 // super call
283 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
284 // interpreter specific
285 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
286 // but since they may not have been saved (and we don't want to
287 // save them here (see note above) the assert is invalid.
288 }
289
290 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
291 Register java_thread,
292 Register last_java_sp,
293 address entry_point,
294 int number_of_arguments,
295 bool check_exceptions) {
296 // interpreter specific
297 //
298 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
299 // really make a difference for these runtime calls, since they are
300 // slow anyway. Btw., bcp must be saved/restored since it may change
301 // due to GC.
302 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
303 save_bcp();
304 #ifdef ASSERT
305 {
306 Label L;
307 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
308 jcc(Assembler::equal, L);
309 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
310 " last_sp != NULL");
311 bind(L);
312 }
313 #endif /* ASSERT */
314 // super call
315 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
316 entry_point, number_of_arguments,
317 check_exceptions);
318 // interpreter specific
319 restore_bcp();
320 restore_locals();
321 }
322
323 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
324 if (JvmtiExport::can_pop_frame()) {
325 Label L;
326 // Initiate popframe handling only if it is not already being
327 // processed. If the flag has the popframe_processing bit set, it
328 // means that this code is called *during* popframe handling - we
329 // don't want to reenter.
330 // This method is only called just after the call into the vm in
331 // call_VM_base, so the arg registers are available.
332 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
333 LP64_ONLY(c_rarg0);
334 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
335 testl(pop_cond, JavaThread::popframe_pending_bit);
336 jcc(Assembler::zero, L);
337 testl(pop_cond, JavaThread::popframe_processing_bit);
338 jcc(Assembler::notZero, L);
339 // Call Interpreter::remove_activation_preserving_args_entry() to get the
340 // address of the same-named entrypoint in the generated interpreter code.
341 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
342 jmp(rax);
343 bind(L);
344 NOT_LP64(get_thread(java_thread);)
345 }
346 }
347
348 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
349 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
350 NOT_LP64(get_thread(thread);)
351 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
352 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
353 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
354 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
355 #ifdef _LP64
356 switch (state) {
357 case atos: movptr(rax, oop_addr);
358 movptr(oop_addr, (int32_t)NULL_WORD);
359 verify_oop(rax, state); break;
360 case ltos: movptr(rax, val_addr); break;
361 case btos: // fall through
362 case ctos: // fall through
363 case stos: // fall through
364 case itos: movl(rax, val_addr); break;
365 case ftos: load_float(val_addr); break;
366 case dtos: load_double(val_addr); break;
367 case vtos: /* nothing to do */ break;
368 default : ShouldNotReachHere();
369 }
370 // Clean up tos value in the thread object
371 movl(tos_addr, (int) ilgl);
372 movl(val_addr, (int32_t) NULL_WORD);
373 #else
374 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
375 + in_ByteSize(wordSize));
376 switch (state) {
377 case atos: movptr(rax, oop_addr);
378 movptr(oop_addr, NULL_WORD);
379 verify_oop(rax, state); break;
380 case ltos:
381 movl(rdx, val_addr1); // fall through
382 case btos: // fall through
383 case ctos: // fall through
384 case stos: // fall through
385 case itos: movl(rax, val_addr); break;
386 case ftos: load_float(val_addr); break;
387 case dtos: load_double(val_addr); break;
388 case vtos: /* nothing to do */ break;
389 default : ShouldNotReachHere();
390 }
391 #endif // _LP64
392 // Clean up tos value in the thread object
393 movl(tos_addr, (int32_t) ilgl);
394 movptr(val_addr, NULL_WORD);
395 NOT_LP64(movptr(val_addr1, NULL_WORD);)
396 }
397
398
399 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
400 if (JvmtiExport::can_force_early_return()) {
401 Label L;
402 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
403 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
404
405 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
406 testptr(tmp, tmp);
407 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
408
409 // Initiate earlyret handling only if it is not already being processed.
410 // If the flag has the earlyret_processing bit set, it means that this code
411 // is called *during* earlyret handling - we don't want to reenter.
412 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
413 cmpl(tmp, JvmtiThreadState::earlyret_pending);
414 jcc(Assembler::notEqual, L);
415
416 // Call Interpreter::remove_activation_early_entry() to get the address of the
417 // same-named entrypoint in the generated interpreter code.
418 NOT_LP64(get_thread(java_thread);)
419 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
420 #ifdef _LP64
421 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
422 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
423 #else
424 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
425 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
426 #endif // _LP64
427 jmp(rax);
428 bind(L);
429 NOT_LP64(get_thread(java_thread);)
430 }
431 }
432
433 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
434 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
435 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
436 bswapl(reg);
437 shrl(reg, 16);
438 }
439
440 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
441 int bcp_offset,
442 size_t index_size) {
443 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
444 if (index_size == sizeof(u2)) {
445 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
446 } else if (index_size == sizeof(u4)) {
447 movl(index, Address(_bcp_register, bcp_offset));
448 // Check if the secondary index definition is still ~x, otherwise
449 // we have to change the following assembler code to calculate the
450 // plain index.
451 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
452 notl(index); // convert to plain index
453 } else if (index_size == sizeof(u1)) {
454 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
455 } else {
456 ShouldNotReachHere();
457 }
458 }
459
460 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
461 Register index,
462 int bcp_offset,
463 size_t index_size) {
464 assert_different_registers(cache, index);
465 get_cache_index_at_bcp(index, bcp_offset, index_size);
466 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
467 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
468 // convert from field index to ConstantPoolCacheEntry index
469 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
470 shll(index, 2);
471 }
472
473 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
474 Register index,
475 Register bytecode,
476 int byte_no,
477 int bcp_offset,
478 size_t index_size) {
479 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
480 // We use a 32-bit load here since the layout of 64-bit words on
481 // little-endian machines allow us that.
482 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
483 const int shift_count = (1 + byte_no) * BitsPerByte;
484 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
485 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
486 "correct shift count");
487 shrl(bytecode, shift_count);
488 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
489 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
490 }
491
492 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
493 Register tmp,
494 int bcp_offset,
495 size_t index_size) {
496 assert(cache != tmp, "must use different register");
497 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
498 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
499 // convert from field index to ConstantPoolCacheEntry index
500 // and from word offset to byte offset
501 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
502 shll(tmp, 2 + LogBytesPerWord);
503 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
504 // skip past the header
505 addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
506 addptr(cache, tmp); // construct pointer to cache entry
507 }
508
509 // Load object from cpool->resolved_references(index)
510 void InterpreterMacroAssembler::load_resolved_reference_at_index(
511 Register result, Register index) {
512 assert_different_registers(result, index);
513 // convert from field index to resolved_references() index and from
514 // word index to byte offset. Since this is a java object, it can be compressed
515 Register tmp = index; // reuse
516 shll(tmp, LogBytesPerHeapOop);
517
518 get_constant_pool(result);
519 // load pointer for resolved_references[] objArray
520 movptr(result, Address(result, ConstantPool::resolved_references_offset_in_bytes()));
521 // JNIHandles::resolve(obj);
522 movptr(result, Address(result, 0));
523 // Add in the index
524 addptr(result, tmp);
525 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
526 }
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 movq(Address(rsp, 0), r);
623 }
624
625 void InterpreterMacroAssembler::pop(TosState state) {
626 switch (state) {
627 case atos: pop_ptr(); break;
628 case btos:
629 case ctos:
630 case stos:
631 case itos: pop_i(); break;
632 case ltos: pop_l(); break;
633 case ftos: pop_f(xmm0); break;
634 case dtos: pop_d(xmm0); break;
635 case vtos: /* nothing to do */ break;
636 default: ShouldNotReachHere();
637 }
638 verify_oop(rax, state);
639 }
640
641 void InterpreterMacroAssembler::push(TosState state) {
642 verify_oop(rax, state);
643 switch (state) {
644 case atos: push_ptr(); break;
645 case btos:
646 case ctos:
647 case stos:
648 case itos: push_i(); break;
649 case ltos: push_l(); break;
650 case ftos: push_f(xmm0); break;
651 case dtos: push_d(xmm0); break;
652 case vtos: /* nothing to do */ break;
653 default : ShouldNotReachHere();
654 }
655 }
656 #else
657 void InterpreterMacroAssembler::pop_i(Register r) {
658 pop(r);
659 }
660
661 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
662 pop(lo);
663 pop(hi);
664 }
665
666 void InterpreterMacroAssembler::pop_f() {
667 fld_s(Address(rsp, 0));
668 addptr(rsp, 1 * wordSize);
669 }
670
671 void InterpreterMacroAssembler::pop_d() {
672 fld_d(Address(rsp, 0));
673 addptr(rsp, 2 * wordSize);
674 }
675
676
677 void InterpreterMacroAssembler::pop(TosState state) {
678 switch (state) {
679 case atos: pop_ptr(rax); break;
680 case btos: // fall through
681 case ctos: // fall through
682 case stos: // fall through
683 case itos: pop_i(rax); break;
684 case ltos: pop_l(rax, rdx); break;
685 case ftos:
686 if (UseSSE >= 1) {
687 pop_f(xmm0);
688 } else {
689 pop_f();
690 }
691 break;
692 case dtos:
693 if (UseSSE >= 2) {
694 pop_d(xmm0);
695 } else {
696 pop_d();
697 }
698 break;
699 case vtos: /* nothing to do */ break;
700 default : ShouldNotReachHere();
701 }
702 verify_oop(rax, state);
703 }
704
705
706 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
707 push(hi);
708 push(lo);
709 }
710
711 void InterpreterMacroAssembler::push_f() {
712 // Do not schedule for no AGI! Never write beyond rsp!
713 subptr(rsp, 1 * wordSize);
714 fstp_s(Address(rsp, 0));
715 }
716
717 void InterpreterMacroAssembler::push_d() {
718 // Do not schedule for no AGI! Never write beyond rsp!
719 subptr(rsp, 2 * wordSize);
720 fstp_d(Address(rsp, 0));
721 }
722
723
724 void InterpreterMacroAssembler::push(TosState state) {
725 verify_oop(rax, state);
726 switch (state) {
727 case atos: push_ptr(rax); break;
728 case btos: // fall through
729 case ctos: // fall through
730 case stos: // fall through
731 case itos: push_i(rax); break;
732 case ltos: push_l(rax, rdx); break;
733 case ftos:
734 if (UseSSE >= 1) {
735 push_f(xmm0);
736 } else {
737 push_f();
738 }
739 break;
740 case dtos:
741 if (UseSSE >= 2) {
742 push_d(xmm0);
743 } else {
744 push_d();
745 }
746 break;
747 case vtos: /* nothing to do */ break;
748 default : ShouldNotReachHere();
749 }
750 }
751 #endif // _LP64
752
753
754 // Helpers for swap and dup
755 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
756 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
757 }
758
759 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
760 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
761 }
762
763
764 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
765 // set sender sp
766 lea(_bcp_register, Address(rsp, wordSize));
767 // record last_sp
768 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
769 }
770
771
772 // Jump to from_interpreted entry of a call unless single stepping is possible
773 // in this thread in which case we must call the i2i entry
774 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
775 prepare_to_jump_from_interpreted();
776
777 if (JvmtiExport::can_post_interpreter_events()) {
778 Label run_compiled_code;
779 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
780 // compiled code in threads for which the event is enabled. Check here for
781 // interp_only_mode if these events CAN be enabled.
782 // interp_only is an int, on little endian it is sufficient to test the byte only
783 // Is a cmpl faster?
784 LP64_ONLY(temp = r15_thread;)
785 NOT_LP64(get_thread(temp);)
786 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
787 jccb(Assembler::zero, run_compiled_code);
788 jmp(Address(method, Method::interpreter_entry_offset()));
789 bind(run_compiled_code);
790 }
791
792 jmp(Address(method, Method::from_interpreted_offset()));
793 }
794
795 // The following two routines provide a hook so that an implementation
796 // can schedule the dispatch in two parts. x86 does not do this.
797 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
798 // Nothing x86 specific to be done here
799 }
800
801 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
802 dispatch_next(state, step);
803 }
804
805 void InterpreterMacroAssembler::dispatch_base(TosState state,
806 address* table,
807 bool verifyoop) {
808 verify_FPU(1, state);
809 if (VerifyActivationFrameSize) {
810 Label L;
811 mov(rcx, rbp);
812 subptr(rcx, rsp);
813 int32_t min_frame_size =
814 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
815 wordSize;
816 cmpptr(rcx, (int32_t)min_frame_size);
817 jcc(Assembler::greaterEqual, L);
818 stop("broken stack frame");
819 bind(L);
820 }
821 if (verifyoop) {
822 verify_oop(rax, state);
823 }
824 #ifdef _LP64
825 lea(rscratch1, ExternalAddress((address)table));
826 jmp(Address(rscratch1, rbx, Address::times_8));
827 #else
828 Address index(noreg, rbx, Address::times_ptr);
829 ExternalAddress tbl((address)table);
830 ArrayAddress dispatch(tbl, index);
831 jump(dispatch);
832 #endif // _LP64
833 }
834
835 void InterpreterMacroAssembler::dispatch_only(TosState state) {
836 dispatch_base(state, Interpreter::dispatch_table(state));
837 }
838
839 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
840 dispatch_base(state, Interpreter::normal_table(state));
841 }
842
843 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
844 dispatch_base(state, Interpreter::normal_table(state), false);
845 }
846
847
848 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
849 // load next bytecode (load before advancing _bcp_register to prevent AGI)
850 load_unsigned_byte(rbx, Address(_bcp_register, step));
851 // advance _bcp_register
852 increment(_bcp_register, step);
853 dispatch_base(state, Interpreter::dispatch_table(state));
854 }
855
856 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
857 // load current bytecode
858 load_unsigned_byte(rbx, Address(_bcp_register, 0));
859 dispatch_base(state, table);
860 }
861
862 // remove activation
863 //
864 // Unlock the receiver if this is a synchronized method.
865 // Unlock any Java monitors from syncronized blocks.
866 // Remove the activation from the stack.
867 //
868 // If there are locked Java monitors
869 // If throw_monitor_exception
870 // throws IllegalMonitorStateException
871 // Else if install_monitor_exception
872 // installs IllegalMonitorStateException
873 // Else
874 // no error processing
875 void InterpreterMacroAssembler::remove_activation(
876 TosState state,
877 Register ret_addr,
878 bool throw_monitor_exception,
879 bool install_monitor_exception,
880 bool notify_jvmdi) {
881 // Note: Registers rdx xmm0 may be in use for the
882 // result check if synchronized method
883 Label unlocked, unlock, no_unlock;
884
885 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
886 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
887 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
888 // monitor pointers need different register
889 // because rdx may have the result in it
890 NOT_LP64(get_thread(rcx);)
891
892 // get the value of _do_not_unlock_if_synchronized into rdx
893 const Address do_not_unlock_if_synchronized(rthread,
894 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
895 movbool(rbx, do_not_unlock_if_synchronized);
896 movbool(do_not_unlock_if_synchronized, false); // reset the flag
897
898 // get method access flags
899 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
900 movl(rcx, Address(rcx, Method::access_flags_offset()));
901 testl(rcx, JVM_ACC_SYNCHRONIZED);
902 jcc(Assembler::zero, unlocked);
903
904 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
905 // is set.
906 testbool(rbx);
907 jcc(Assembler::notZero, no_unlock);
908
909 // unlock monitor
910 push(state); // save result
911
912 // BasicObjectLock will be first in list, since this is a
913 // synchronized method. However, need to check that the object has
914 // not been unlocked by an explicit monitorexit bytecode.
915 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
916 wordSize - (int) sizeof(BasicObjectLock));
917 // We use c_rarg1/rdx so that if we go slow path it will be the correct
918 // register for unlock_object to pass to VM directly
919 lea(robj, monitor); // address of first monitor
920
921 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
922 testptr(rax, rax);
923 jcc(Assembler::notZero, unlock);
924
925 pop(state);
926 if (throw_monitor_exception) {
927 // Entry already unlocked, need to throw exception
928 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
929 call_VM(noreg, CAST_FROM_FN_PTR(address,
930 InterpreterRuntime::throw_illegal_monitor_state_exception));
931 should_not_reach_here();
932 } else {
933 // Monitor already unlocked during a stack unroll. If requested,
934 // install an illegal_monitor_state_exception. Continue with
935 // stack unrolling.
936 if (install_monitor_exception) {
937 NOT_LP64(empty_FPU_stack();)
938 call_VM(noreg, CAST_FROM_FN_PTR(address,
939 InterpreterRuntime::new_illegal_monitor_state_exception));
940 }
941 jmp(unlocked);
942 }
943
944 bind(unlock);
945 unlock_object(robj);
946 pop(state);
947
948 // Check that for block-structured locking (i.e., that all locked
949 // objects has been unlocked)
950 bind(unlocked);
951
952 // rax, rdx: Might contain return value
953
954 // Check that all monitors are unlocked
955 {
956 Label loop, exception, entry, restart;
957 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
958 const Address monitor_block_top(
959 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
960 const Address monitor_block_bot(
961 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
962
963 bind(restart);
964 // We use c_rarg1 so that if we go slow path it will be the correct
965 // register for unlock_object to pass to VM directly
966 movptr(rmon, monitor_block_top); // points to current entry, starting
967 // with top-most entry
968 lea(rbx, monitor_block_bot); // points to word before bottom of
969 // monitor block
970 jmp(entry);
971
972 // Entry already locked, need to throw exception
973 bind(exception);
974
975 if (throw_monitor_exception) {
976 // Throw exception
977 NOT_LP64(empty_FPU_stack();)
978 MacroAssembler::call_VM(noreg,
979 CAST_FROM_FN_PTR(address, InterpreterRuntime::
980 throw_illegal_monitor_state_exception));
981 should_not_reach_here();
982 } else {
983 // Stack unrolling. Unlock object and install illegal_monitor_exception.
984 // Unlock does not block, so don't have to worry about the frame.
985 // We don't have to preserve c_rarg1 since we are going to throw an exception.
986
987 push(state);
988 mov(robj, rmon); // nop if robj and rmon are the same
989 unlock_object(robj);
990 pop(state);
991
992 if (install_monitor_exception) {
993 NOT_LP64(empty_FPU_stack();)
994 call_VM(noreg, CAST_FROM_FN_PTR(address,
995 InterpreterRuntime::
996 new_illegal_monitor_state_exception));
997 }
998
999 jmp(restart);
1000 }
1001
1002 bind(loop);
1003 // check if current entry is used
1004 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
1005 jcc(Assembler::notEqual, exception);
1006
1007 addptr(rmon, entry_size); // otherwise advance to next entry
1008 bind(entry);
1009 cmpptr(rmon, rbx); // check if bottom reached
1010 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1011 }
1012
1013 bind(no_unlock);
1014
1015 // jvmti support
1016 if (notify_jvmdi) {
1017 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1018 } else {
1019 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1020 }
1021
1022 // remove activation
1023 // get sender sp
1024 movptr(rbx,
1025 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1026 if (StackReservedPages > 0) {
1027 // testing if reserved zone needs to be re-enabled
1028 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1029 Label no_reserved_zone_enabling;
1030
1031 NOT_LP64(get_thread(rthread);)
1032
1033 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1034 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1035
1036 call_VM_leaf(
1037 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1038 push(rthread);
1039 call_VM(noreg, CAST_FROM_FN_PTR(address,
1040 InterpreterRuntime::throw_delayed_StackOverflowError));
1041 should_not_reach_here();
1042
1043 bind(no_reserved_zone_enabling);
1044 }
1045 leave(); // remove frame anchor
1046 pop(ret_addr); // get return address
1047 mov(rsp, rbx); // set sp to sender sp
1048 }
1049 #endif // !CC_INTERP
1050
1051 void InterpreterMacroAssembler::get_method_counters(Register method,
1052 Register mcs, Label& skip) {
1053 Label has_counters;
1054 movptr(mcs, Address(method, Method::method_counters_offset()));
1055 testptr(mcs, mcs);
1056 jcc(Assembler::notZero, has_counters);
1057 call_VM(noreg, CAST_FROM_FN_PTR(address,
1058 InterpreterRuntime::build_method_counters), method);
1059 movptr(mcs, Address(method,Method::method_counters_offset()));
1060 testptr(mcs, mcs);
1061 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1062 bind(has_counters);
1063 }
1064
1065
1066 // Lock object
1067 //
1068 // Args:
1069 // rdx, c_rarg1: BasicObjectLock to be used for locking
1070 //
1071 // Kills:
1072 // rax, rbx
1073 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1074 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1075 "The argument is only for looks. It must be c_rarg1");
1076
1077 if (UseHeavyMonitors) {
1078 call_VM(noreg,
1079 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1080 lock_reg);
1081 } else {
1082 Label done;
1083
1084 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1085 const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a
1086 // problematic case where tmp_reg = no_reg.
1087 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1088
1089 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
1090 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
1091 const int mark_offset = lock_offset +
1092 BasicLock::displaced_header_offset_in_bytes();
1093
1094 Label slow_case;
1095
1096 // Load object pointer into obj_reg
1097 movptr(obj_reg, Address(lock_reg, obj_offset));
1098
1099 if (UseBiasedLocking) {
1100 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case);
1101 }
1102
1103 // Load immediate 1 into swap_reg %rax
1104 movl(swap_reg, (int32_t)1);
1105
1106 // Load (object->mark() | 1) into swap_reg %rax
1107 orptr(swap_reg, Address(obj_reg, 0));
1108
1109 // Save (object->mark() | 1) into BasicLock's displaced header
1110 movptr(Address(lock_reg, mark_offset), swap_reg);
1111
1112 assert(lock_offset == 0,
1113 "displached header must be first word in BasicObjectLock");
1114
1115 if (os::is_MP()) lock();
1116 cmpxchgptr(lock_reg, Address(obj_reg, 0));
1117 if (PrintBiasedLockingStatistics) {
1118 cond_inc32(Assembler::zero,
1119 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1120 }
1121 jcc(Assembler::zero, done);
1122
1123 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1124
1125 // Test if the oopMark is an obvious stack pointer, i.e.,
1126 // 1) (mark & zero_bits) == 0, and
1127 // 2) rsp <= mark < mark + os::pagesize()
1128 //
1129 // These 3 tests can be done by evaluating the following
1130 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1131 // assuming both stack pointer and pagesize have their
1132 // least significant bits clear.
1133 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
1134 subptr(swap_reg, rsp);
1135 andptr(swap_reg, zero_bits - os::vm_page_size());
1136
1137 // Save the test result, for recursive case, the result is zero
1138 movptr(Address(lock_reg, mark_offset), swap_reg);
1139
1140 if (PrintBiasedLockingStatistics) {
1141 cond_inc32(Assembler::zero,
1142 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1143 }
1144 jcc(Assembler::zero, done);
1145
1146 bind(slow_case);
1147
1148 // Call the runtime routine for slow case
1149 call_VM(noreg,
1150 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1151 lock_reg);
1152
1153 bind(done);
1154 }
1155 }
1156
1157
1158 // Unlocks an object. Used in monitorexit bytecode and
1159 // remove_activation. Throws an IllegalMonitorException if object is
1160 // not locked by current thread.
1161 //
1162 // Args:
1163 // rdx, c_rarg1: BasicObjectLock for lock
1164 //
1165 // Kills:
1166 // rax
1167 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1168 // rscratch1, rscratch2 (scratch regs)
1169 // rax, rbx, rcx, rdx
1170 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1171 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1172 "The argument is only for looks. It must be c_rarg1");
1173
1174 if (UseHeavyMonitors) {
1175 call_VM(noreg,
1176 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1177 lock_reg);
1178 } else {
1179 Label done;
1180
1181 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1182 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1183 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1184
1185 save_bcp(); // Save in case of exception
1186
1187 // Convert from BasicObjectLock structure to object and BasicLock
1188 // structure Store the BasicLock address into %rax
1189 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
1190
1191 // Load oop into obj_reg(%c_rarg3)
1192 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
1193
1194 // Free entry
1195 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
1196
1197 if (UseBiasedLocking) {
1198 biased_locking_exit(obj_reg, header_reg, done);
1199 }
1200
1201 // Load the old header from BasicLock structure
1202 movptr(header_reg, Address(swap_reg,
1203 BasicLock::displaced_header_offset_in_bytes()));
1204
1205 // Test for recursion
1206 testptr(header_reg, header_reg);
1207
1208 // zero for recursive case
1209 jcc(Assembler::zero, done);
1210
1211 // Atomic swap back the old header
1212 if (os::is_MP()) lock();
1213 cmpxchgptr(header_reg, Address(obj_reg, 0));
1214
1215 // zero for recursive case
1216 jcc(Assembler::zero, done);
1217
1218 // Call the runtime routine for slow case.
1219 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
1220 obj_reg); // restore obj
1221 call_VM(noreg,
1222 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1223 lock_reg);
1224
1225 bind(done);
1226
1227 restore_bcp();
1228 }
1229 }
1230 #ifndef CC_INTERP
1231 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1232 Label& zero_continue) {
1233 assert(ProfileInterpreter, "must be profiling interpreter");
1234 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1235 testptr(mdp, mdp);
1236 jcc(Assembler::zero, zero_continue);
1237 }
1238
1239
1240 // Set the method data pointer for the current bcp.
1241 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1242 assert(ProfileInterpreter, "must be profiling interpreter");
1243 Label set_mdp;
1244 push(rax);
1245 push(rbx);
1246
1247 get_method(rbx);
1248 // Test MDO to avoid the call if it is NULL.
1249 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1250 testptr(rax, rax);
1251 jcc(Assembler::zero, set_mdp);
1252 // rbx: method
1253 // _bcp_register: bcp
1254 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1255 // rax: mdi
1256 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1257 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1258 addptr(rbx, in_bytes(MethodData::data_offset()));
1259 addptr(rax, rbx);
1260 bind(set_mdp);
1261 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1262 pop(rbx);
1263 pop(rax);
1264 }
1265
1266 void InterpreterMacroAssembler::verify_method_data_pointer() {
1267 assert(ProfileInterpreter, "must be profiling interpreter");
1268 #ifdef ASSERT
1269 Label verify_continue;
1270 push(rax);
1271 push(rbx);
1272 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1273 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1274 push(arg3_reg);
1275 push(arg2_reg);
1276 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1277 get_method(rbx);
1278
1279 // If the mdp is valid, it will point to a DataLayout header which is
1280 // consistent with the bcp. The converse is highly probable also.
1281 load_unsigned_short(arg2_reg,
1282 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1283 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1284 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1285 cmpptr(arg2_reg, _bcp_register);
1286 jcc(Assembler::equal, verify_continue);
1287 // rbx: method
1288 // _bcp_register: bcp
1289 // c_rarg3: mdp
1290 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1291 rbx, _bcp_register, arg3_reg);
1292 bind(verify_continue);
1293 pop(arg2_reg);
1294 pop(arg3_reg);
1295 pop(rbx);
1296 pop(rax);
1297 #endif // ASSERT
1298 }
1299
1300
1301 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1302 int constant,
1303 Register value) {
1304 assert(ProfileInterpreter, "must be profiling interpreter");
1305 Address data(mdp_in, constant);
1306 movptr(data, value);
1307 }
1308
1309
1310 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1311 int constant,
1312 bool decrement) {
1313 // Counter address
1314 Address data(mdp_in, constant);
1315
1316 increment_mdp_data_at(data, decrement);
1317 }
1318
1319 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1320 bool decrement) {
1321 assert(ProfileInterpreter, "must be profiling interpreter");
1322 // %%% this does 64bit counters at best it is wasting space
1323 // at worst it is a rare bug when counters overflow
1324
1325 if (decrement) {
1326 // Decrement the register. Set condition codes.
1327 addptr(data, (int32_t) -DataLayout::counter_increment);
1328 // If the decrement causes the counter to overflow, stay negative
1329 Label L;
1330 jcc(Assembler::negative, L);
1331 addptr(data, (int32_t) DataLayout::counter_increment);
1332 bind(L);
1333 } else {
1334 assert(DataLayout::counter_increment == 1,
1335 "flow-free idiom only works with 1");
1336 // Increment the register. Set carry flag.
1337 addptr(data, DataLayout::counter_increment);
1338 // If the increment causes the counter to overflow, pull back by 1.
1339 sbbptr(data, (int32_t)0);
1340 }
1341 }
1342
1343
1344 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1345 Register reg,
1346 int constant,
1347 bool decrement) {
1348 Address data(mdp_in, reg, Address::times_1, constant);
1349
1350 increment_mdp_data_at(data, decrement);
1351 }
1352
1353 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1354 int flag_byte_constant) {
1355 assert(ProfileInterpreter, "must be profiling interpreter");
1356 int header_offset = in_bytes(DataLayout::header_offset());
1357 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
1358 // Set the flag
1359 orl(Address(mdp_in, header_offset), header_bits);
1360 }
1361
1362
1363
1364 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1365 int offset,
1366 Register value,
1367 Register test_value_out,
1368 Label& not_equal_continue) {
1369 assert(ProfileInterpreter, "must be profiling interpreter");
1370 if (test_value_out == noreg) {
1371 cmpptr(value, Address(mdp_in, offset));
1372 } else {
1373 // Put the test value into a register, so caller can use it:
1374 movptr(test_value_out, Address(mdp_in, offset));
1375 cmpptr(test_value_out, value);
1376 }
1377 jcc(Assembler::notEqual, not_equal_continue);
1378 }
1379
1380
1381 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1382 int offset_of_disp) {
1383 assert(ProfileInterpreter, "must be profiling interpreter");
1384 Address disp_address(mdp_in, offset_of_disp);
1385 addptr(mdp_in, disp_address);
1386 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1387 }
1388
1389
1390 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1391 Register reg,
1392 int offset_of_disp) {
1393 assert(ProfileInterpreter, "must be profiling interpreter");
1394 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1395 addptr(mdp_in, disp_address);
1396 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1397 }
1398
1399
1400 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1401 int constant) {
1402 assert(ProfileInterpreter, "must be profiling interpreter");
1403 addptr(mdp_in, constant);
1404 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1405 }
1406
1407
1408 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1409 assert(ProfileInterpreter, "must be profiling interpreter");
1410 push(return_bci); // save/restore across call_VM
1411 call_VM(noreg,
1412 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1413 return_bci);
1414 pop(return_bci);
1415 }
1416
1417
1418 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1419 Register bumped_count) {
1420 if (ProfileInterpreter) {
1421 Label profile_continue;
1422
1423 // If no method data exists, go to profile_continue.
1424 // Otherwise, assign to mdp
1425 test_method_data_pointer(mdp, profile_continue);
1426
1427 // We are taking a branch. Increment the taken count.
1428 // We inline increment_mdp_data_at to return bumped_count in a register
1429 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1430 Address data(mdp, in_bytes(JumpData::taken_offset()));
1431 movptr(bumped_count, data);
1432 assert(DataLayout::counter_increment == 1,
1433 "flow-free idiom only works with 1");
1434 addptr(bumped_count, DataLayout::counter_increment);
1435 sbbptr(bumped_count, 0);
1436 movptr(data, bumped_count); // Store back out
1437
1438 // The method data pointer needs to be updated to reflect the new target.
1439 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1440 bind(profile_continue);
1441 }
1442 }
1443
1444
1445 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1446 if (ProfileInterpreter) {
1447 Label profile_continue;
1448
1449 // If no method data exists, go to profile_continue.
1450 test_method_data_pointer(mdp, profile_continue);
1451
1452 // We are taking a branch. Increment the not taken count.
1453 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1454
1455 // The method data pointer needs to be updated to correspond to
1456 // the next bytecode
1457 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1458 bind(profile_continue);
1459 }
1460 }
1461
1462 void InterpreterMacroAssembler::profile_call(Register mdp) {
1463 if (ProfileInterpreter) {
1464 Label profile_continue;
1465
1466 // If no method data exists, go to profile_continue.
1467 test_method_data_pointer(mdp, profile_continue);
1468
1469 // We are making a call. Increment the count.
1470 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1471
1472 // The method data pointer needs to be updated to reflect the new target.
1473 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1474 bind(profile_continue);
1475 }
1476 }
1477
1478
1479 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1480 if (ProfileInterpreter) {
1481 Label profile_continue;
1482
1483 // If no method data exists, go to profile_continue.
1484 test_method_data_pointer(mdp, profile_continue);
1485
1486 // We are making a call. Increment the count.
1487 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1488
1489 // The method data pointer needs to be updated to reflect the new target.
1490 update_mdp_by_constant(mdp,
1491 in_bytes(VirtualCallData::
1492 virtual_call_data_size()));
1493 bind(profile_continue);
1494 }
1495 }
1496
1497
1498 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1499 Register mdp,
1500 Register reg2,
1501 bool receiver_can_be_null) {
1502 if (ProfileInterpreter) {
1503 Label profile_continue;
1504
1505 // If no method data exists, go to profile_continue.
1506 test_method_data_pointer(mdp, profile_continue);
1507
1508 Label skip_receiver_profile;
1509 if (receiver_can_be_null) {
1510 Label not_null;
1511 testptr(receiver, receiver);
1512 jccb(Assembler::notZero, not_null);
1513 // We are making a call. Increment the count for null receiver.
1514 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1515 jmp(skip_receiver_profile);
1516 bind(not_null);
1517 }
1518
1519 // Record the receiver type.
1520 record_klass_in_profile(receiver, mdp, reg2, true);
1521 bind(skip_receiver_profile);
1522
1523 // The method data pointer needs to be updated to reflect the new target.
1524 #if INCLUDE_JVMCI
1525 if (MethodProfileWidth == 0) {
1526 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1527 }
1528 #else // INCLUDE_JVMCI
1529 update_mdp_by_constant(mdp,
1530 in_bytes(VirtualCallData::
1531 virtual_call_data_size()));
1532 #endif // INCLUDE_JVMCI
1533 bind(profile_continue);
1534 }
1535 }
1536
1537 #if INCLUDE_JVMCI
1538 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) {
1539 assert_different_registers(method, mdp, reg2);
1540 if (ProfileInterpreter && MethodProfileWidth > 0) {
1541 Label profile_continue;
1542
1543 // If no method data exists, go to profile_continue.
1544 test_method_data_pointer(mdp, profile_continue);
1545
1546 Label done;
1547 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth,
1548 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1549 bind(done);
1550
1551 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1552 bind(profile_continue);
1553 }
1554 }
1555 #endif // INCLUDE_JVMCI
1556
1557 // This routine creates a state machine for updating the multi-row
1558 // type profile at a virtual call site (or other type-sensitive bytecode).
1559 // The machine visits each row (of receiver/count) until the receiver type
1560 // is found, or until it runs out of rows. At the same time, it remembers
1561 // the location of the first empty row. (An empty row records null for its
1562 // receiver, and can be allocated for a newly-observed receiver type.)
1563 // Because there are two degrees of freedom in the state, a simple linear
1564 // search will not work; it must be a decision tree. Hence this helper
1565 // function is recursive, to generate the required tree structured code.
1566 // It's the interpreter, so we are trading off code space for speed.
1567 // See below for example code.
1568 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1569 Register receiver, Register mdp,
1570 Register reg2, int start_row,
1571 Label& done, bool is_virtual_call) {
1572 if (TypeProfileWidth == 0) {
1573 if (is_virtual_call) {
1574 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1575 }
1576 #if INCLUDE_JVMCI
1577 else if (EnableJVMCI) {
1578 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1579 }
1580 #endif // INCLUDE_JVMCI
1581 } else {
1582 int non_profiled_offset = -1;
1583 if (is_virtual_call) {
1584 non_profiled_offset = in_bytes(CounterData::count_offset());
1585 }
1586 #if INCLUDE_JVMCI
1587 else if (EnableJVMCI) {
1588 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1589 }
1590 #endif // INCLUDE_JVMCI
1591
1592 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1593 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1594 }
1595 }
1596
1597 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1598 Register reg2, int start_row, Label& done, int total_rows,
1599 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1600 int non_profiled_offset) {
1601 int last_row = total_rows - 1;
1602 assert(start_row <= last_row, "must be work left to do");
1603 // Test this row for both the item and for null.
1604 // Take any of three different outcomes:
1605 // 1. found item => increment count and goto done
1606 // 2. found null => keep looking for case 1, maybe allocate this cell
1607 // 3. found something else => keep looking for cases 1 and 2
1608 // Case 3 is handled by a recursive call.
1609 for (int row = start_row; row <= last_row; row++) {
1610 Label next_test;
1611 bool test_for_null_also = (row == start_row);
1612
1613 // See if the item is item[n].
1614 int item_offset = in_bytes(item_offset_fn(row));
1615 test_mdp_data_at(mdp, item_offset, item,
1616 (test_for_null_also ? reg2 : noreg),
1617 next_test);
1618 // (Reg2 now contains the item from the CallData.)
1619
1620 // The item is item[n]. Increment count[n].
1621 int count_offset = in_bytes(item_count_offset_fn(row));
1622 increment_mdp_data_at(mdp, count_offset);
1623 jmp(done);
1624 bind(next_test);
1625
1626 if (test_for_null_also) {
1627 Label found_null;
1628 // Failed the equality check on item[n]... Test for null.
1629 testptr(reg2, reg2);
1630 if (start_row == last_row) {
1631 // The only thing left to do is handle the null case.
1632 if (non_profiled_offset >= 0) {
1633 jccb(Assembler::zero, found_null);
1634 // Item did not match any saved item and there is no empty row for it.
1635 // Increment total counter to indicate polymorphic case.
1636 increment_mdp_data_at(mdp, non_profiled_offset);
1637 jmp(done);
1638 bind(found_null);
1639 } else {
1640 jcc(Assembler::notZero, done);
1641 }
1642 break;
1643 }
1644 // Since null is rare, make it be the branch-taken case.
1645 jcc(Assembler::zero, found_null);
1646
1647 // Put all the "Case 3" tests here.
1648 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1649 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1650
1651 // Found a null. Keep searching for a matching item,
1652 // but remember that this is an empty (unused) slot.
1653 bind(found_null);
1654 }
1655 }
1656
1657 // In the fall-through case, we found no matching item, but we
1658 // observed the item[start_row] is NULL.
1659
1660 // Fill in the item field and increment the count.
1661 int item_offset = in_bytes(item_offset_fn(start_row));
1662 set_mdp_data_at(mdp, item_offset, item);
1663 int count_offset = in_bytes(item_count_offset_fn(start_row));
1664 movl(reg2, DataLayout::counter_increment);
1665 set_mdp_data_at(mdp, count_offset, reg2);
1666 if (start_row > 0) {
1667 jmp(done);
1668 }
1669 }
1670
1671 // Example state machine code for three profile rows:
1672 // // main copy of decision tree, rooted at row[1]
1673 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1674 // if (row[0].rec != NULL) {
1675 // // inner copy of decision tree, rooted at row[1]
1676 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1677 // if (row[1].rec != NULL) {
1678 // // degenerate decision tree, rooted at row[2]
1679 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1680 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1681 // row[2].init(rec); goto done;
1682 // } else {
1683 // // remember row[1] is empty
1684 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1685 // row[1].init(rec); goto done;
1686 // }
1687 // } else {
1688 // // remember row[0] is empty
1689 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1690 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1691 // row[0].init(rec); goto done;
1692 // }
1693 // done:
1694
1695 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1696 Register mdp, Register reg2,
1697 bool is_virtual_call) {
1698 assert(ProfileInterpreter, "must be profiling");
1699 Label done;
1700
1701 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1702
1703 bind (done);
1704 }
1705
1706 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1707 Register mdp) {
1708 if (ProfileInterpreter) {
1709 Label profile_continue;
1710 uint row;
1711
1712 // If no method data exists, go to profile_continue.
1713 test_method_data_pointer(mdp, profile_continue);
1714
1715 // Update the total ret count.
1716 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1717
1718 for (row = 0; row < RetData::row_limit(); row++) {
1719 Label next_test;
1720
1721 // See if return_bci is equal to bci[n]:
1722 test_mdp_data_at(mdp,
1723 in_bytes(RetData::bci_offset(row)),
1724 return_bci, noreg,
1725 next_test);
1726
1727 // return_bci is equal to bci[n]. Increment the count.
1728 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1729
1730 // The method data pointer needs to be updated to reflect the new target.
1731 update_mdp_by_offset(mdp,
1732 in_bytes(RetData::bci_displacement_offset(row)));
1733 jmp(profile_continue);
1734 bind(next_test);
1735 }
1736
1737 update_mdp_for_ret(return_bci);
1738
1739 bind(profile_continue);
1740 }
1741 }
1742
1743
1744 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1745 if (ProfileInterpreter) {
1746 Label profile_continue;
1747
1748 // If no method data exists, go to profile_continue.
1749 test_method_data_pointer(mdp, profile_continue);
1750
1751 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1752
1753 // The method data pointer needs to be updated.
1754 int mdp_delta = in_bytes(BitData::bit_data_size());
1755 if (TypeProfileCasts) {
1756 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1757 }
1758 update_mdp_by_constant(mdp, mdp_delta);
1759
1760 bind(profile_continue);
1761 }
1762 }
1763
1764
1765 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1766 if (ProfileInterpreter && TypeProfileCasts) {
1767 Label profile_continue;
1768
1769 // If no method data exists, go to profile_continue.
1770 test_method_data_pointer(mdp, profile_continue);
1771
1772 int count_offset = in_bytes(CounterData::count_offset());
1773 // Back up the address, since we have already bumped the mdp.
1774 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1775
1776 // *Decrement* the counter. We expect to see zero or small negatives.
1777 increment_mdp_data_at(mdp, count_offset, true);
1778
1779 bind (profile_continue);
1780 }
1781 }
1782
1783
1784 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1785 if (ProfileInterpreter) {
1786 Label profile_continue;
1787
1788 // If no method data exists, go to profile_continue.
1789 test_method_data_pointer(mdp, profile_continue);
1790
1791 // The method data pointer needs to be updated.
1792 int mdp_delta = in_bytes(BitData::bit_data_size());
1793 if (TypeProfileCasts) {
1794 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1795
1796 // Record the object type.
1797 record_klass_in_profile(klass, mdp, reg2, false);
1798 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1799 NOT_LP64(restore_locals();) // Restore EDI
1800 }
1801 update_mdp_by_constant(mdp, mdp_delta);
1802
1803 bind(profile_continue);
1804 }
1805 }
1806
1807
1808 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1809 if (ProfileInterpreter) {
1810 Label profile_continue;
1811
1812 // If no method data exists, go to profile_continue.
1813 test_method_data_pointer(mdp, profile_continue);
1814
1815 // Update the default case count
1816 increment_mdp_data_at(mdp,
1817 in_bytes(MultiBranchData::default_count_offset()));
1818
1819 // The method data pointer needs to be updated.
1820 update_mdp_by_offset(mdp,
1821 in_bytes(MultiBranchData::
1822 default_displacement_offset()));
1823
1824 bind(profile_continue);
1825 }
1826 }
1827
1828
1829 void InterpreterMacroAssembler::profile_switch_case(Register index,
1830 Register mdp,
1831 Register reg2) {
1832 if (ProfileInterpreter) {
1833 Label profile_continue;
1834
1835 // If no method data exists, go to profile_continue.
1836 test_method_data_pointer(mdp, profile_continue);
1837
1838 // Build the base (index * per_case_size_in_bytes()) +
1839 // case_array_offset_in_bytes()
1840 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1841 imulptr(index, reg2); // XXX l ?
1842 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1843
1844 // Update the case count
1845 increment_mdp_data_at(mdp,
1846 index,
1847 in_bytes(MultiBranchData::relative_count_offset()));
1848
1849 // The method data pointer needs to be updated.
1850 update_mdp_by_offset(mdp,
1851 index,
1852 in_bytes(MultiBranchData::
1853 relative_displacement_offset()));
1854
1855 bind(profile_continue);
1856 }
1857 }
1858
1859
1860
1861 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1862 if (state == atos) {
1863 MacroAssembler::verify_oop(reg);
1864 }
1865 }
1866
1867 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1868 #ifndef _LP64
1869 if ((state == ftos && UseSSE < 1) ||
1870 (state == dtos && UseSSE < 2)) {
1871 MacroAssembler::verify_FPU(stack_depth);
1872 }
1873 #endif
1874 }
1875
1876 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1877 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1878 int increment, Address mask,
1879 Register scratch, bool preloaded,
1880 Condition cond, Label* where) {
1881 if (!preloaded) {
1882 movl(scratch, counter_addr);
1883 }
1884 incrementl(scratch, increment);
1885 movl(counter_addr, scratch);
1886 andl(scratch, mask);
1887 jcc(cond, *where);
1888 }
1889 #endif // CC_INTERP
1890
1891 void InterpreterMacroAssembler::notify_method_entry() {
1892 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1893 // track stack depth. If it is possible to enter interp_only_mode we add
1894 // the code to check if the event should be sent.
1895 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1896 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1897 if (JvmtiExport::can_post_interpreter_events()) {
1898 Label L;
1899 NOT_LP64(get_thread(rthread);)
1900 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1901 testl(rdx, rdx);
1902 jcc(Assembler::zero, L);
1903 call_VM(noreg, CAST_FROM_FN_PTR(address,
1904 InterpreterRuntime::post_method_entry));
1905 bind(L);
1906 }
1907
1908 {
1909 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1910 NOT_LP64(get_thread(rthread);)
1911 get_method(rarg);
1912 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1913 rthread, rarg);
1914 }
1915
1916 // RedefineClasses() tracing support for obsolete method entry
1917 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
1918 NOT_LP64(get_thread(rthread);)
1919 get_method(rarg);
1920 call_VM_leaf(
1921 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1922 rthread, rarg);
1923 }
1924 }
1925
1926
1927 void InterpreterMacroAssembler::notify_method_exit(
1928 TosState state, NotifyMethodExitMode mode) {
1929 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1930 // track stack depth. If it is possible to enter interp_only_mode we add
1931 // the code to check if the event should be sent.
1932 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1933 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1934 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1935 Label L;
1936 // Note: frame::interpreter_frame_result has a dependency on how the
1937 // method result is saved across the call to post_method_exit. If this
1938 // is changed then the interpreter_frame_result implementation will
1939 // need to be updated too.
1940
1941 // For c++ interpreter the result is always stored at a known location in the frame
1942 // template interpreter will leave it on the top of the stack.
1943 NOT_CC_INTERP(push(state);)
1944 NOT_LP64(get_thread(rthread);)
1945 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1946 testl(rdx, rdx);
1947 jcc(Assembler::zero, L);
1948 call_VM(noreg,
1949 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1950 bind(L);
1951 NOT_CC_INTERP(pop(state));
1952 }
1953
1954 {
1955 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1956 NOT_CC_INTERP(push(state));
1957 NOT_LP64(get_thread(rthread);)
1958 get_method(rarg);
1959 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1960 rthread, rarg);
1961 NOT_CC_INTERP(pop(state));
1962 }
1963 }