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 "memory/vtBuffer.hpp"
31 #include "oops/arrayOop.hpp"
32 #include "oops/markOop.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "oops/valueKlass.hpp"
36 #include "prims/jvmtiExport.hpp"
37 #include "prims/jvmtiThreadState.hpp"
38 #include "runtime/basicLock.hpp"
39 #include "runtime/biasedLocking.hpp"
40 #include "runtime/safepointMechanism.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/thread.inline.hpp"
43
44 // Implementation of InterpreterMacroAssembler
45
46 void InterpreterMacroAssembler::jump_to_entry(address entry) {
47 assert(entry, "Entry must have been generated by now");
48 jump(RuntimeAddress(entry));
49 }
50
51 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
52 Label update, next, none;
53
54 verify_oop(obj);
55
56 testptr(obj, obj);
57 jccb(Assembler::notZero, update);
58 orptr(mdo_addr, TypeEntries::null_seen);
59 jmpb(next);
60
61 bind(update);
62 load_klass(obj, obj);
63
64 xorptr(obj, mdo_addr);
65 testptr(obj, TypeEntries::type_klass_mask);
66 jccb(Assembler::zero, next); // klass seen before, nothing to
67 // do. The unknown bit may have been
68 // set already but no need to check.
69
70 testptr(obj, TypeEntries::type_unknown);
71 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
72
73 cmpptr(mdo_addr, 0);
74 jccb(Assembler::equal, none);
75 cmpptr(mdo_addr, TypeEntries::null_seen);
76 jccb(Assembler::equal, none);
77 // There is a chance that the checks above (re-reading profiling
78 // data from memory) fail if another thread has just set the
79 // profiling to this obj's klass
80 xorptr(obj, mdo_addr);
81 testptr(obj, TypeEntries::type_klass_mask);
82 jccb(Assembler::zero, next);
83
84 // different than before. Cannot keep accurate profile.
85 orptr(mdo_addr, TypeEntries::type_unknown);
86 jmpb(next);
87
88 bind(none);
89 // first time here. Set profile type.
90 movptr(mdo_addr, obj);
91
92 bind(next);
93 }
94
95 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
96 if (!ProfileInterpreter) {
97 return;
98 }
99
100 if (MethodData::profile_arguments() || MethodData::profile_return()) {
101 Label profile_continue;
102
103 test_method_data_pointer(mdp, profile_continue);
104
105 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
106
107 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
108 jcc(Assembler::notEqual, profile_continue);
109
110 if (MethodData::profile_arguments()) {
111 Label done;
112 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
113 addptr(mdp, off_to_args);
114
115 for (int i = 0; i < TypeProfileArgsLimit; i++) {
116 if (i > 0 || MethodData::profile_return()) {
117 // If return value type is profiled we may have no argument to profile
118 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
119 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
120 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
121 jcc(Assembler::less, done);
122 }
123 movptr(tmp, Address(callee, Method::const_offset()));
124 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
125 // stack offset o (zero based) from the start of the argument
126 // list, for n arguments translates into offset n - o - 1 from
127 // the end of the argument list
128 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
129 subl(tmp, 1);
130 Address arg_addr = argument_address(tmp);
131 movptr(tmp, arg_addr);
132
133 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
134 profile_obj_type(tmp, mdo_arg_addr);
135
136 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
137 addptr(mdp, to_add);
138 off_to_args += to_add;
139 }
140
141 if (MethodData::profile_return()) {
142 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
143 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
144 }
145
146 bind(done);
147
148 if (MethodData::profile_return()) {
149 // We're right after the type profile for the last
150 // argument. tmp is the number of cells left in the
151 // CallTypeData/VirtualCallTypeData to reach its end. Non null
152 // if there's a return to profile.
153 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
154 shll(tmp, exact_log2(DataLayout::cell_size));
155 addptr(mdp, tmp);
156 }
157 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
158 } else {
159 assert(MethodData::profile_return(), "either profile call args or call ret");
160 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
161 }
162
163 // mdp points right after the end of the
164 // CallTypeData/VirtualCallTypeData, right after the cells for the
165 // return value type if there's one
166
167 bind(profile_continue);
168 }
169 }
170
171 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
172 assert_different_registers(mdp, ret, tmp, _bcp_register);
173 if (ProfileInterpreter && MethodData::profile_return()) {
174 Label profile_continue, done;
175
176 test_method_data_pointer(mdp, profile_continue);
177
178 if (MethodData::profile_return_jsr292_only()) {
179 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
180
181 // If we don't profile all invoke bytecodes we must make sure
182 // it's a bytecode we indeed profile. We can't go back to the
183 // begining of the ProfileData we intend to update to check its
184 // type because we're right after it and we don't known its
185 // length
186 Label do_profile;
187 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
188 jcc(Assembler::equal, do_profile);
189 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
190 jcc(Assembler::equal, do_profile);
191 get_method(tmp);
192 cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm);
193 jcc(Assembler::notEqual, profile_continue);
194
195 bind(do_profile);
196 }
197
198 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
199 mov(tmp, ret);
200 profile_obj_type(tmp, mdo_ret_addr);
201
202 bind(profile_continue);
203 }
204 }
205
206 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
207 if (ProfileInterpreter && MethodData::profile_parameters()) {
208 Label profile_continue, done;
209
210 test_method_data_pointer(mdp, profile_continue);
211
212 // Load the offset of the area within the MDO used for
213 // parameters. If it's negative we're not profiling any parameters
214 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
215 testl(tmp1, tmp1);
216 jcc(Assembler::negative, profile_continue);
217
218 // Compute a pointer to the area for parameters from the offset
219 // and move the pointer to the slot for the last
220 // parameters. Collect profiling from last parameter down.
221 // mdo start + parameters offset + array length - 1
222 addptr(mdp, tmp1);
223 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
224 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
225
226 Label loop;
227 bind(loop);
228
229 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
230 int type_base = in_bytes(ParametersTypeData::type_offset(0));
231 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
232 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
233 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
234
235 // load offset on the stack from the slot for this parameter
236 movptr(tmp2, arg_off);
237 negptr(tmp2);
238 // read the parameter from the local area
239 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
240
241 // profile the parameter
242 profile_obj_type(tmp2, arg_type);
243
244 // go to next parameter
245 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
246 jcc(Assembler::positive, loop);
247
248 bind(profile_continue);
249 }
250 }
251
252 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
253 int number_of_arguments) {
254 // interpreter specific
255 //
256 // Note: No need to save/restore bcp & locals registers
257 // since these are callee saved registers and no blocking/
258 // GC can happen in leaf calls.
259 // Further Note: DO NOT save/restore bcp/locals. If a caller has
260 // already saved them so that it can use rsi/rdi as temporaries
261 // then a save/restore here will DESTROY the copy the caller
262 // saved! There used to be a save_bcp() that only happened in
263 // the ASSERT path (no restore_bcp). Which caused bizarre failures
264 // when jvm built with ASSERTs.
265 #ifdef ASSERT
266 {
267 Label L;
268 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
269 jcc(Assembler::equal, L);
270 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
271 " last_sp != NULL");
272 bind(L);
273 }
274 #endif
275 // super call
276 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
277 // interpreter specific
278 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
279 // but since they may not have been saved (and we don't want to
280 // save them here (see note above) the assert is invalid.
281 }
282
283 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
284 Register java_thread,
285 Register last_java_sp,
286 address entry_point,
287 int number_of_arguments,
288 bool check_exceptions) {
289 // interpreter specific
290 //
291 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
292 // really make a difference for these runtime calls, since they are
293 // slow anyway. Btw., bcp must be saved/restored since it may change
294 // due to GC.
295 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
296 save_bcp();
297 #ifdef ASSERT
298 {
299 Label L;
300 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
301 jcc(Assembler::equal, L);
302 stop("InterpreterMacroAssembler::call_VM_base:"
303 " last_sp != NULL");
304 bind(L);
305 }
306 #endif /* ASSERT */
307 // super call
308 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
309 entry_point, number_of_arguments,
310 check_exceptions);
311 // interpreter specific
312 restore_bcp();
313 restore_locals();
314 }
315
316 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
317 if (JvmtiExport::can_pop_frame()) {
318 Label L;
319 // Initiate popframe handling only if it is not already being
320 // processed. If the flag has the popframe_processing bit set, it
321 // means that this code is called *during* popframe handling - we
322 // don't want to reenter.
323 // This method is only called just after the call into the vm in
324 // call_VM_base, so the arg registers are available.
325 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
326 LP64_ONLY(c_rarg0);
327 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
328 testl(pop_cond, JavaThread::popframe_pending_bit);
329 jcc(Assembler::zero, L);
330 testl(pop_cond, JavaThread::popframe_processing_bit);
331 jcc(Assembler::notZero, L);
332 // Call Interpreter::remove_activation_preserving_args_entry() to get the
333 // address of the same-named entrypoint in the generated interpreter code.
334 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
335 jmp(rax);
336 bind(L);
337 NOT_LP64(get_thread(java_thread);)
338 }
339 }
340
341 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
342 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
343 NOT_LP64(get_thread(thread);)
344 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
345 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
346 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
347 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
348 #ifdef _LP64
349 switch (state) {
350 case qtos: // fall through
351 case atos: movptr(rax, oop_addr);
352 movptr(oop_addr, (int32_t)NULL_WORD);
353 verify_oop(rax, state); break;
354 case ltos: movptr(rax, val_addr); break;
355 case btos: // fall through
356 case ztos: // fall through
357 case ctos: // fall through
358 case stos: // fall through
359 case itos: movl(rax, val_addr); break;
360 case ftos: load_float(val_addr); break;
361 case dtos: load_double(val_addr); break;
362 case vtos: /* nothing to do */ break;
363 default : ShouldNotReachHere();
364 }
365 // Clean up tos value in the thread object
366 movl(tos_addr, (int) ilgl);
367 movl(val_addr, (int32_t) NULL_WORD);
368 #else
369 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
370 + in_ByteSize(wordSize));
371 switch (state) {
372 case qtos: // fall through
373 case atos: movptr(rax, oop_addr);
374 movptr(oop_addr, NULL_WORD);
375 verify_oop(rax, state); break;
376 case ltos:
377 movl(rdx, val_addr1); // fall through
378 case btos: // fall through
379 case ztos: // fall through
380 case ctos: // fall through
381 case stos: // fall through
382 case itos: movl(rax, val_addr); break;
383 case ftos: load_float(val_addr); break;
384 case dtos: load_double(val_addr); break;
385 case vtos: /* nothing to do */ break;
386 default : ShouldNotReachHere();
387 }
388 #endif // _LP64
389 // Clean up tos value in the thread object
390 movl(tos_addr, (int32_t) ilgl);
391 movptr(val_addr, NULL_WORD);
392 NOT_LP64(movptr(val_addr1, NULL_WORD);)
393 }
394
395
396 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
397 if (JvmtiExport::can_force_early_return()) {
398 Label L;
399 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
400 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
401
402 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
403 testptr(tmp, tmp);
404 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
405
406 // Initiate earlyret handling only if it is not already being processed.
407 // If the flag has the earlyret_processing bit set, it means that this code
408 // is called *during* earlyret handling - we don't want to reenter.
409 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
410 cmpl(tmp, JvmtiThreadState::earlyret_pending);
411 jcc(Assembler::notEqual, L);
412
413 // Call Interpreter::remove_activation_early_entry() to get the address of the
414 // same-named entrypoint in the generated interpreter code.
415 NOT_LP64(get_thread(java_thread);)
416 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
417 #ifdef _LP64
418 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
419 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
420 #else
421 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
422 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
423 #endif // _LP64
424 jmp(rax);
425 bind(L);
426 NOT_LP64(get_thread(java_thread);)
427 }
428 }
429
430 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
431 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
432 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
433 bswapl(reg);
434 shrl(reg, 16);
435 }
436
437 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
438 int bcp_offset,
439 size_t index_size) {
440 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
441 if (index_size == sizeof(u2)) {
442 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
443 } else if (index_size == sizeof(u4)) {
444 movl(index, Address(_bcp_register, bcp_offset));
445 // Check if the secondary index definition is still ~x, otherwise
446 // we have to change the following assembler code to calculate the
447 // plain index.
448 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
449 notl(index); // convert to plain index
450 } else if (index_size == sizeof(u1)) {
451 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
452 } else {
453 ShouldNotReachHere();
454 }
455 }
456
457 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
458 Register index,
459 int bcp_offset,
460 size_t index_size) {
461 assert_different_registers(cache, index);
462 get_cache_index_at_bcp(index, bcp_offset, index_size);
463 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
464 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
465 // convert from field index to ConstantPoolCacheEntry index
466 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
467 shll(index, 2);
468 }
469
470 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
471 Register index,
472 Register bytecode,
473 int byte_no,
474 int bcp_offset,
475 size_t index_size) {
476 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
477 // We use a 32-bit load here since the layout of 64-bit words on
478 // little-endian machines allow us that.
479 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
480 const int shift_count = (1 + byte_no) * BitsPerByte;
481 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
482 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
483 "correct shift count");
484 shrl(bytecode, shift_count);
485 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
486 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
487 }
488
489 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
490 Register tmp,
491 int bcp_offset,
492 size_t index_size) {
493 assert(cache != tmp, "must use different register");
494 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
495 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
496 // convert from field index to ConstantPoolCacheEntry index
497 // and from word offset to byte offset
498 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
499 shll(tmp, 2 + LogBytesPerWord);
500 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
501 // skip past the header
502 addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
503 addptr(cache, tmp); // construct pointer to cache entry
504 }
505
506 // Load object from cpool->resolved_references(index)
507 void InterpreterMacroAssembler::load_resolved_reference_at_index(
508 Register result, Register index) {
509 assert_different_registers(result, index);
510 // convert from field index to resolved_references() index and from
511 // word index to byte offset. Since this is a java object, it can be compressed
512 Register tmp = index; // reuse
513 shll(tmp, LogBytesPerHeapOop);
514
515 get_constant_pool(result);
516 // load pointer for resolved_references[] objArray
517 movptr(result, Address(result, ConstantPool::cache_offset_in_bytes()));
518 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes()));
519 resolve_oop_handle(result);
520 // Add in the index
521 addptr(result, tmp);
522 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
523 }
524
525 // load cpool->resolved_klass_at(index)
526 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register cpool,
527 Register index, Register klass) {
528 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
529 Register resolved_klasses = cpool;
530 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes()));
531 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
532 }
533
534 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
535 // subtype of super_klass.
536 //
537 // Args:
538 // rax: superklass
539 // Rsub_klass: subklass
540 //
541 // Kills:
542 // rcx, rdi
543 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
544 Label& ok_is_subtype) {
545 assert(Rsub_klass != rax, "rax holds superklass");
546 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
547 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
548 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
549 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
550
551 // Profile the not-null value's klass.
552 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
553
554 // Do the check.
555 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
556
557 // Profile the failure of the check.
558 profile_typecheck_failed(rcx); // blows rcx
559 }
560
561
562 #ifndef _LP64
563 void InterpreterMacroAssembler::f2ieee() {
564 if (IEEEPrecision) {
565 fstp_s(Address(rsp, 0));
566 fld_s(Address(rsp, 0));
567 }
568 }
569
570
571 void InterpreterMacroAssembler::d2ieee() {
572 if (IEEEPrecision) {
573 fstp_d(Address(rsp, 0));
574 fld_d(Address(rsp, 0));
575 }
576 }
577 #endif // _LP64
578
579 // Java Expression Stack
580
581 void InterpreterMacroAssembler::pop_ptr(Register r) {
582 pop(r);
583 }
584
585 void InterpreterMacroAssembler::push_ptr(Register r) {
586 push(r);
587 }
588
589 void InterpreterMacroAssembler::push_i(Register r) {
590 push(r);
591 }
592
593 void InterpreterMacroAssembler::push_f(XMMRegister r) {
594 subptr(rsp, wordSize);
595 movflt(Address(rsp, 0), r);
596 }
597
598 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
599 movflt(r, Address(rsp, 0));
600 addptr(rsp, wordSize);
601 }
602
603 void InterpreterMacroAssembler::push_d(XMMRegister r) {
604 subptr(rsp, 2 * wordSize);
605 movdbl(Address(rsp, 0), r);
606 }
607
608 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
609 movdbl(r, Address(rsp, 0));
610 addptr(rsp, 2 * Interpreter::stackElementSize);
611 }
612
613 #ifdef _LP64
614 void InterpreterMacroAssembler::pop_i(Register r) {
615 // XXX can't use pop currently, upper half non clean
616 movl(r, Address(rsp, 0));
617 addptr(rsp, wordSize);
618 }
619
620 void InterpreterMacroAssembler::pop_l(Register r) {
621 movq(r, Address(rsp, 0));
622 addptr(rsp, 2 * Interpreter::stackElementSize);
623 }
624
625 void InterpreterMacroAssembler::push_l(Register r) {
626 subptr(rsp, 2 * wordSize);
627 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
628 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
629 }
630
631 void InterpreterMacroAssembler::pop(TosState state) {
632 switch (state) {
633 case ptos: // Fall through
634 case qtos: // Fall through
635 case atos: pop_ptr(); break;
636 case btos:
637 case ztos:
638 case ctos:
639 case stos:
640 case itos: pop_i(); break;
641 case ltos: pop_l(); break;
642 case ftos: pop_f(xmm0); break;
643 case dtos: pop_d(xmm0); break;
644 case vtos: /* nothing to do */ break;
645 default: ShouldNotReachHere();
646 }
647 verify_oop(rax, state);
648 }
649
650 void InterpreterMacroAssembler::push(TosState state) {
651 verify_oop(rax, state);
652 switch (state) {
653 case qtos: // Fall through
654 case atos: push_ptr(); break;
655 case btos:
656 case ztos:
657 case ctos:
658 case stos:
659 case itos: push_i(); break;
660 case ltos: push_l(); break;
661 case ftos: push_f(xmm0); break;
662 case dtos: push_d(xmm0); break;
663 case vtos: /* nothing to do */ break;
664 default : ShouldNotReachHere();
665 }
666 }
667 #else
668 void InterpreterMacroAssembler::pop_i(Register r) {
669 pop(r);
670 }
671
672 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
673 pop(lo);
674 pop(hi);
675 }
676
677 void InterpreterMacroAssembler::pop_f() {
678 fld_s(Address(rsp, 0));
679 addptr(rsp, 1 * wordSize);
680 }
681
682 void InterpreterMacroAssembler::pop_d() {
683 fld_d(Address(rsp, 0));
684 addptr(rsp, 2 * wordSize);
685 }
686
687
688 void InterpreterMacroAssembler::pop(TosState state) {
689 switch (state) {
690 case qtos: // fall through
691 case atos: pop_ptr(rax); break;
692 case btos: // fall through
693 case ztos: // fall through
694 case ctos: // fall through
695 case stos: // fall through
696 case itos: pop_i(rax); break;
697 case ltos: pop_l(rax, rdx); break;
698 case ftos:
699 if (UseSSE >= 1) {
700 pop_f(xmm0);
701 } else {
702 pop_f();
703 }
704 break;
705 case dtos:
706 if (UseSSE >= 2) {
707 pop_d(xmm0);
708 } else {
709 pop_d();
710 }
711 break;
712 case vtos: /* nothing to do */ break;
713 default : ShouldNotReachHere();
714 }
715 verify_oop(rax, state);
716 }
717
718
719 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
720 push(hi);
721 push(lo);
722 }
723
724 void InterpreterMacroAssembler::push_f() {
725 // Do not schedule for no AGI! Never write beyond rsp!
726 subptr(rsp, 1 * wordSize);
727 fstp_s(Address(rsp, 0));
728 }
729
730 void InterpreterMacroAssembler::push_d() {
731 // Do not schedule for no AGI! Never write beyond rsp!
732 subptr(rsp, 2 * wordSize);
733 fstp_d(Address(rsp, 0));
734 }
735
736
737 void InterpreterMacroAssembler::push(TosState state) {
738 verify_oop(rax, state);
739 switch (state) {
740 case qtos: // fall through
741 case atos: push_ptr(rax); break;
742 case btos: // fall through
743 case ztos: // fall through
744 case ctos: // fall through
745 case stos: // fall through
746 case itos: push_i(rax); break;
747 case ltos: push_l(rax, rdx); break;
748 case ftos:
749 if (UseSSE >= 1) {
750 push_f(xmm0);
751 } else {
752 push_f();
753 }
754 break;
755 case dtos:
756 if (UseSSE >= 2) {
757 push_d(xmm0);
758 } else {
759 push_d();
760 }
761 break;
762 case vtos: /* nothing to do */ break;
763 default : ShouldNotReachHere();
764 }
765 }
766 #endif // _LP64
767
768
769 // Helpers for swap and dup
770 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
771 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
772 }
773
774 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
775 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
776 }
777
778
779 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
780 // set sender sp
781 lea(_bcp_register, Address(rsp, wordSize));
782 // record last_sp
783 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
784 }
785
786
787 // Jump to from_interpreted entry of a call unless single stepping is possible
788 // in this thread in which case we must call the i2i entry
789 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
790 prepare_to_jump_from_interpreted();
791
792 if (JvmtiExport::can_post_interpreter_events()) {
793 Label run_compiled_code;
794 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
795 // compiled code in threads for which the event is enabled. Check here for
796 // interp_only_mode if these events CAN be enabled.
797 // interp_only is an int, on little endian it is sufficient to test the byte only
798 // Is a cmpl faster?
799 LP64_ONLY(temp = r15_thread;)
800 NOT_LP64(get_thread(temp);)
801 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
802 jccb(Assembler::zero, run_compiled_code);
803 jmp(Address(method, Method::interpreter_entry_offset()));
804 bind(run_compiled_code);
805 }
806
807 jmp(Address(method, Method::from_interpreted_offset()));
808 }
809
810 // The following two routines provide a hook so that an implementation
811 // can schedule the dispatch in two parts. x86 does not do this.
812 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
813 // Nothing x86 specific to be done here
814 }
815
816 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
817 dispatch_next(state, step);
818 }
819
820 void InterpreterMacroAssembler::dispatch_base(TosState state,
821 address* table,
822 bool verifyoop,
823 bool generate_poll) {
824 verify_FPU(1, state);
825 if (VerifyActivationFrameSize) {
826 Label L;
827 mov(rcx, rbp);
828 subptr(rcx, rsp);
829 int32_t min_frame_size =
830 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
831 wordSize;
832 cmpptr(rcx, (int32_t)min_frame_size);
833 jcc(Assembler::greaterEqual, L);
834 stop("broken stack frame");
835 bind(L);
836 }
837 if (verifyoop) {
838 verify_oop(rax, state);
839 }
840 #ifdef _LP64
841
842 Label no_safepoint, dispatch;
843 address* const safepoint_table = Interpreter::safept_table(state);
844 if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) {
845 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
846
847 testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
848
849 jccb(Assembler::zero, no_safepoint);
850 lea(rscratch1, ExternalAddress((address)safepoint_table));
851 jmpb(dispatch);
852 }
853
854 bind(no_safepoint);
855 lea(rscratch1, ExternalAddress((address)table));
856 bind(dispatch);
857 jmp(Address(rscratch1, rbx, Address::times_8));
858
859 #else
860 Address index(noreg, rbx, Address::times_ptr);
861 ExternalAddress tbl((address)table);
862 ArrayAddress dispatch(tbl, index);
863 jump(dispatch);
864 #endif // _LP64
865 }
866
867 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
868 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
869 }
870
871 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
872 dispatch_base(state, Interpreter::normal_table(state));
873 }
874
875 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
876 dispatch_base(state, Interpreter::normal_table(state), false);
877 }
878
879
880 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
881 // load next bytecode (load before advancing _bcp_register to prevent AGI)
882 load_unsigned_byte(rbx, Address(_bcp_register, step));
883 // advance _bcp_register
884 increment(_bcp_register, step);
885 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
886 }
887
888 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
889 // load current bytecode
890 load_unsigned_byte(rbx, Address(_bcp_register, 0));
891 dispatch_base(state, table);
892 }
893
894 void InterpreterMacroAssembler::narrow(Register result) {
895
896 // Get method->_constMethod->_result_type
897 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
898 movptr(rcx, Address(rcx, Method::const_offset()));
899 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
900
901 Label done, notBool, notByte, notChar;
902
903 // common case first
904 cmpl(rcx, T_INT);
905 jcc(Assembler::equal, done);
906
907 // mask integer result to narrower return type.
908 cmpl(rcx, T_BOOLEAN);
909 jcc(Assembler::notEqual, notBool);
910 andl(result, 0x1);
911 jmp(done);
912
913 bind(notBool);
914 cmpl(rcx, T_BYTE);
915 jcc(Assembler::notEqual, notByte);
916 LP64_ONLY(movsbl(result, result);)
917 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
918 NOT_LP64(sarl(result, 24);) // and sign-extend byte
919 jmp(done);
920
921 bind(notByte);
922 cmpl(rcx, T_CHAR);
923 jcc(Assembler::notEqual, notChar);
924 LP64_ONLY(movzwl(result, result);)
925 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
926 jmp(done);
927
928 bind(notChar);
929 // cmpl(rcx, T_SHORT); // all that's left
930 // jcc(Assembler::notEqual, done);
931 LP64_ONLY(movswl(result, result);)
932 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
933 NOT_LP64(sarl(result, 16);) // and sign-extend short
934
935 // Nothing to do for T_INT
936 bind(done);
937 }
938
939 // remove activation
940 //
941 // Unlock the receiver if this is a synchronized method.
942 // Unlock any Java monitors from syncronized blocks.
943 // Remove the activation from the stack.
944 //
945 // If there are locked Java monitors
946 // If throw_monitor_exception
947 // throws IllegalMonitorStateException
948 // Else if install_monitor_exception
949 // installs IllegalMonitorStateException
950 // Else
951 // no error processing
952 void InterpreterMacroAssembler::remove_activation(
953 TosState state,
954 Register ret_addr,
955 bool throw_monitor_exception,
956 bool install_monitor_exception,
957 bool notify_jvmdi,
958 bool load_values) {
959 // Note: Registers rdx xmm0 may be in use for the
960 // result check if synchronized method
961 Label unlocked, unlock, no_unlock;
962
963 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
964 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
965 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
966 // monitor pointers need different register
967 // because rdx may have the result in it
968 NOT_LP64(get_thread(rcx);)
969
970 // get the value of _do_not_unlock_if_synchronized into rdx
971 const Address do_not_unlock_if_synchronized(rthread,
972 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
973 movbool(rbx, do_not_unlock_if_synchronized);
974 movbool(do_not_unlock_if_synchronized, false); // reset the flag
975
976 // get method access flags
977 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
978 movl(rcx, Address(rcx, Method::access_flags_offset()));
979 testl(rcx, JVM_ACC_SYNCHRONIZED);
980 jcc(Assembler::zero, unlocked);
981
982 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
983 // is set.
984 testbool(rbx);
985 jcc(Assembler::notZero, no_unlock);
986
987 // unlock monitor
988 push(state); // save result
989
990 // BasicObjectLock will be first in list, since this is a
991 // synchronized method. However, need to check that the object has
992 // not been unlocked by an explicit monitorexit bytecode.
993 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
994 wordSize - (int) sizeof(BasicObjectLock));
995 // We use c_rarg1/rdx so that if we go slow path it will be the correct
996 // register for unlock_object to pass to VM directly
997 lea(robj, monitor); // address of first monitor
998
999 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
1000 testptr(rax, rax);
1001 jcc(Assembler::notZero, unlock);
1002
1003 pop(state);
1004 if (throw_monitor_exception) {
1005 // Entry already unlocked, need to throw exception
1006 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
1007 call_VM(noreg, CAST_FROM_FN_PTR(address,
1008 InterpreterRuntime::throw_illegal_monitor_state_exception));
1009 should_not_reach_here();
1010 } else {
1011 // Monitor already unlocked during a stack unroll. If requested,
1012 // install an illegal_monitor_state_exception. Continue with
1013 // stack unrolling.
1014 if (install_monitor_exception) {
1015 NOT_LP64(empty_FPU_stack();)
1016 call_VM(noreg, CAST_FROM_FN_PTR(address,
1017 InterpreterRuntime::new_illegal_monitor_state_exception));
1018 }
1019 jmp(unlocked);
1020 }
1021
1022 bind(unlock);
1023 unlock_object(robj);
1024 pop(state);
1025
1026 // Check that for block-structured locking (i.e., that all locked
1027 // objects has been unlocked)
1028 bind(unlocked);
1029
1030 // rax, rdx: Might contain return value
1031
1032 // Check that all monitors are unlocked
1033 {
1034 Label loop, exception, entry, restart;
1035 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1036 const Address monitor_block_top(
1037 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1038 const Address monitor_block_bot(
1039 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1040
1041 bind(restart);
1042 // We use c_rarg1 so that if we go slow path it will be the correct
1043 // register for unlock_object to pass to VM directly
1044 movptr(rmon, monitor_block_top); // points to current entry, starting
1045 // with top-most entry
1046 lea(rbx, monitor_block_bot); // points to word before bottom of
1047 // monitor block
1048 jmp(entry);
1049
1050 // Entry already locked, need to throw exception
1051 bind(exception);
1052
1053 if (throw_monitor_exception) {
1054 // Throw exception
1055 NOT_LP64(empty_FPU_stack();)
1056 MacroAssembler::call_VM(noreg,
1057 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1058 throw_illegal_monitor_state_exception));
1059 should_not_reach_here();
1060 } else {
1061 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1062 // Unlock does not block, so don't have to worry about the frame.
1063 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1064
1065 push(state);
1066 mov(robj, rmon); // nop if robj and rmon are the same
1067 unlock_object(robj);
1068 pop(state);
1069
1070 if (install_monitor_exception) {
1071 NOT_LP64(empty_FPU_stack();)
1072 call_VM(noreg, CAST_FROM_FN_PTR(address,
1073 InterpreterRuntime::
1074 new_illegal_monitor_state_exception));
1075 }
1076
1077 jmp(restart);
1078 }
1079
1080 bind(loop);
1081 // check if current entry is used
1082 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
1083 jcc(Assembler::notEqual, exception);
1084
1085 addptr(rmon, entry_size); // otherwise advance to next entry
1086 bind(entry);
1087 cmpptr(rmon, rbx); // check if bottom reached
1088 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1089 }
1090
1091 bind(no_unlock);
1092
1093 // jvmti support
1094 if (notify_jvmdi) {
1095 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1096 } else {
1097 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1098 }
1099
1100 if (StackReservedPages > 0) {
1101 movptr(rbx,
1102 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1103 // testing if reserved zone needs to be re-enabled
1104 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1105 Label no_reserved_zone_enabling;
1106
1107 NOT_LP64(get_thread(rthread);)
1108
1109 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_enabled);
1110 jcc(Assembler::equal, no_reserved_zone_enabling);
1111
1112 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1113 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1114
1115 call_VM_leaf(
1116 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1117 call_VM(noreg, CAST_FROM_FN_PTR(address,
1118 InterpreterRuntime::throw_delayed_StackOverflowError));
1119 should_not_reach_here();
1120
1121 bind(no_reserved_zone_enabling);
1122 }
1123
1124 // Code below is taking care of recycling TLVB memory, no safepoint should
1125 // occur between this point and the end of the remove_activation() method
1126 const Register thread1 = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1127 const uintptr_t chunk_mask = VTBufferChunk::chunk_mask();
1128
1129 Label TLVB_cleanup_done;
1130 if (state == qtos) {
1131 Label no_buffered_value_returned;
1132 if (ReturnValuesInThreadLocalBuffer) {
1133 NOT_LP64(get_thread(thread1));
1134 cmpptr(Address(thread1, JavaThread::return_buffered_value_offset()), (intptr_t)NULL_WORD);
1135 jcc(Assembler::equal, no_buffered_value_returned);
1136 movptr(rbx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize));
1137 call_VM_leaf(CAST_FROM_FN_PTR(address,
1138 InterpreterRuntime::relocate_return_value), rax, rbx);
1139 NOT_LP64(get_thread(thread1));
1140 get_vm_result(rax, thread1);
1141 jmp(TLVB_cleanup_done); // clean up has been performed during relocation
1142 bind(no_buffered_value_returned);
1143 }
1144 }
1145 movptr(rbx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize));
1146 NOT_LP64(get_thread(thread1));
1147 movptr(rcx, Address(thread1, JavaThread::vt_alloc_ptr_offset()));
1148 cmpptr(rbx, rcx);
1149 jcc(Assembler::equal, TLVB_cleanup_done);
1150 push(state);
1151 movptr(rbx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize));
1152 call_VM_leaf(CAST_FROM_FN_PTR(address,
1153 InterpreterRuntime::recycle_vtbuffer), rbx);
1154 pop(state);
1155 bind(TLVB_cleanup_done);
1156
1157 // remove activation
1158 // get sender sp
1159 movptr(rbx,
1160 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1161
1162 if (load_values) {
1163 // We are returning a value type, load its fields into registers
1164 #ifndef _LP64
1165 super_call_VM_leaf(StubRoutines::load_value_type_fields_in_regs());
1166 #else
1167 load_klass(rdi, rax);
1168 movptr(rdi, Address(rdi, ValueKlass::unpack_handler_offset()));
1169
1170 Label skip;
1171 testptr(rdi, rdi);
1172 jcc(Assembler::equal, skip);
1173
1174 // Load fields from a buffered value with a value class specific
1175 // handler
1176 call(rdi);
1177
1178 bind(skip);
1179 #endif
1180 // call above kills the value in rbx. Reload it.
1181 movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1182 }
1183 leave(); // remove frame anchor
1184 pop(ret_addr); // get return address
1185 mov(rsp, rbx); // set sp to sender sp
1186 }
1187
1188 void InterpreterMacroAssembler::get_method_counters(Register method,
1189 Register mcs, Label& skip) {
1190 Label has_counters;
1191 movptr(mcs, Address(method, Method::method_counters_offset()));
1192 testptr(mcs, mcs);
1193 jcc(Assembler::notZero, has_counters);
1194 call_VM(noreg, CAST_FROM_FN_PTR(address,
1195 InterpreterRuntime::build_method_counters), method);
1196 movptr(mcs, Address(method,Method::method_counters_offset()));
1197 testptr(mcs, mcs);
1198 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1199 bind(has_counters);
1200 }
1201
1202
1203 // Lock object
1204 //
1205 // Args:
1206 // rdx, c_rarg1: BasicObjectLock to be used for locking
1207 //
1208 // Kills:
1209 // rax, rbx
1210 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1211 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1212 "The argument is only for looks. It must be c_rarg1");
1213
1214 if (UseHeavyMonitors) {
1215 call_VM(noreg,
1216 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1217 lock_reg);
1218 } else {
1219 Label done;
1220
1221 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1222 const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a
1223 // problematic case where tmp_reg = no_reg.
1224 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1225
1226 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
1227 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
1228 const int mark_offset = lock_offset +
1229 BasicLock::displaced_header_offset_in_bytes();
1230
1231 Label slow_case;
1232
1233 // Load object pointer into obj_reg
1234 movptr(obj_reg, Address(lock_reg, obj_offset));
1235
1236 if (UseBiasedLocking) {
1237 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case);
1238 }
1239
1240 // Load immediate 1 into swap_reg %rax
1241 movl(swap_reg, (int32_t)1);
1242
1243 // Load (object->mark() | 1) into swap_reg %rax
1244 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1245
1246 // Save (object->mark() | 1) into BasicLock's displaced header
1247 movptr(Address(lock_reg, mark_offset), swap_reg);
1248
1249 assert(lock_offset == 0,
1250 "displaced header must be first word in BasicObjectLock");
1251
1252 if (os::is_MP()) lock();
1253 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1254 if (PrintBiasedLockingStatistics) {
1255 cond_inc32(Assembler::zero,
1256 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1257 }
1258 jcc(Assembler::zero, done);
1259
1260 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1261
1262 // Test if the oopMark is an obvious stack pointer, i.e.,
1263 // 1) (mark & zero_bits) == 0, and
1264 // 2) rsp <= mark < mark + os::pagesize()
1265 //
1266 // These 3 tests can be done by evaluating the following
1267 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1268 // assuming both stack pointer and pagesize have their
1269 // least significant bits clear.
1270 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
1271 subptr(swap_reg, rsp);
1272 andptr(swap_reg, zero_bits - os::vm_page_size());
1273
1274 // Save the test result, for recursive case, the result is zero
1275 movptr(Address(lock_reg, mark_offset), swap_reg);
1276
1277 if (PrintBiasedLockingStatistics) {
1278 cond_inc32(Assembler::zero,
1279 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1280 }
1281 jcc(Assembler::zero, done);
1282
1283 bind(slow_case);
1284
1285 // Call the runtime routine for slow case
1286 call_VM(noreg,
1287 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1288 lock_reg);
1289
1290 bind(done);
1291 }
1292 }
1293
1294
1295 // Unlocks an object. Used in monitorexit bytecode and
1296 // remove_activation. Throws an IllegalMonitorException if object is
1297 // not locked by current thread.
1298 //
1299 // Args:
1300 // rdx, c_rarg1: BasicObjectLock for lock
1301 //
1302 // Kills:
1303 // rax
1304 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1305 // rscratch1 (scratch reg)
1306 // rax, rbx, rcx, rdx
1307 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1308 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1309 "The argument is only for looks. It must be c_rarg1");
1310
1311 if (UseHeavyMonitors) {
1312 call_VM(noreg,
1313 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1314 lock_reg);
1315 } else {
1316 Label done;
1317
1318 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1319 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1320 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1321
1322 save_bcp(); // Save in case of exception
1323
1324 // Convert from BasicObjectLock structure to object and BasicLock
1325 // structure Store the BasicLock address into %rax
1326 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
1327
1328 // Load oop into obj_reg(%c_rarg3)
1329 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
1330
1331 // Free entry
1332 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
1333
1334 if (UseBiasedLocking) {
1335 biased_locking_exit(obj_reg, header_reg, done);
1336 }
1337
1338 // Load the old header from BasicLock structure
1339 movptr(header_reg, Address(swap_reg,
1340 BasicLock::displaced_header_offset_in_bytes()));
1341
1342 // Test for recursion
1343 testptr(header_reg, header_reg);
1344
1345 // zero for recursive case
1346 jcc(Assembler::zero, done);
1347
1348 // Atomic swap back the old header
1349 if (os::is_MP()) lock();
1350 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1351
1352 // zero for simple unlock of a stack-lock case
1353 jcc(Assembler::zero, done);
1354
1355 // Call the runtime routine for slow case.
1356 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
1357 obj_reg); // restore obj
1358 call_VM(noreg,
1359 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1360 lock_reg);
1361
1362 bind(done);
1363
1364 restore_bcp();
1365 }
1366 }
1367
1368 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1369 Label& zero_continue) {
1370 assert(ProfileInterpreter, "must be profiling interpreter");
1371 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1372 testptr(mdp, mdp);
1373 jcc(Assembler::zero, zero_continue);
1374 }
1375
1376
1377 // Set the method data pointer for the current bcp.
1378 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1379 assert(ProfileInterpreter, "must be profiling interpreter");
1380 Label set_mdp;
1381 push(rax);
1382 push(rbx);
1383
1384 get_method(rbx);
1385 // Test MDO to avoid the call if it is NULL.
1386 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1387 testptr(rax, rax);
1388 jcc(Assembler::zero, set_mdp);
1389 // rbx: method
1390 // _bcp_register: bcp
1391 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1392 // rax: mdi
1393 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1394 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1395 addptr(rbx, in_bytes(MethodData::data_offset()));
1396 addptr(rax, rbx);
1397 bind(set_mdp);
1398 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1399 pop(rbx);
1400 pop(rax);
1401 }
1402
1403 void InterpreterMacroAssembler::verify_method_data_pointer() {
1404 assert(ProfileInterpreter, "must be profiling interpreter");
1405 #ifdef ASSERT
1406 Label verify_continue;
1407 push(rax);
1408 push(rbx);
1409 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1410 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1411 push(arg3_reg);
1412 push(arg2_reg);
1413 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1414 get_method(rbx);
1415
1416 // If the mdp is valid, it will point to a DataLayout header which is
1417 // consistent with the bcp. The converse is highly probable also.
1418 load_unsigned_short(arg2_reg,
1419 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1420 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1421 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1422 cmpptr(arg2_reg, _bcp_register);
1423 jcc(Assembler::equal, verify_continue);
1424 // rbx: method
1425 // _bcp_register: bcp
1426 // c_rarg3: mdp
1427 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1428 rbx, _bcp_register, arg3_reg);
1429 bind(verify_continue);
1430 pop(arg2_reg);
1431 pop(arg3_reg);
1432 pop(rbx);
1433 pop(rax);
1434 #endif // ASSERT
1435 }
1436
1437
1438 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1439 int constant,
1440 Register value) {
1441 assert(ProfileInterpreter, "must be profiling interpreter");
1442 Address data(mdp_in, constant);
1443 movptr(data, value);
1444 }
1445
1446
1447 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1448 int constant,
1449 bool decrement) {
1450 // Counter address
1451 Address data(mdp_in, constant);
1452
1453 increment_mdp_data_at(data, decrement);
1454 }
1455
1456 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1457 bool decrement) {
1458 assert(ProfileInterpreter, "must be profiling interpreter");
1459 // %%% this does 64bit counters at best it is wasting space
1460 // at worst it is a rare bug when counters overflow
1461
1462 if (decrement) {
1463 // Decrement the register. Set condition codes.
1464 addptr(data, (int32_t) -DataLayout::counter_increment);
1465 // If the decrement causes the counter to overflow, stay negative
1466 Label L;
1467 jcc(Assembler::negative, L);
1468 addptr(data, (int32_t) DataLayout::counter_increment);
1469 bind(L);
1470 } else {
1471 assert(DataLayout::counter_increment == 1,
1472 "flow-free idiom only works with 1");
1473 // Increment the register. Set carry flag.
1474 addptr(data, DataLayout::counter_increment);
1475 // If the increment causes the counter to overflow, pull back by 1.
1476 sbbptr(data, (int32_t)0);
1477 }
1478 }
1479
1480
1481 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1482 Register reg,
1483 int constant,
1484 bool decrement) {
1485 Address data(mdp_in, reg, Address::times_1, constant);
1486
1487 increment_mdp_data_at(data, decrement);
1488 }
1489
1490 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1491 int flag_byte_constant) {
1492 assert(ProfileInterpreter, "must be profiling interpreter");
1493 int header_offset = in_bytes(DataLayout::header_offset());
1494 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
1495 // Set the flag
1496 orl(Address(mdp_in, header_offset), header_bits);
1497 }
1498
1499
1500
1501 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1502 int offset,
1503 Register value,
1504 Register test_value_out,
1505 Label& not_equal_continue) {
1506 assert(ProfileInterpreter, "must be profiling interpreter");
1507 if (test_value_out == noreg) {
1508 cmpptr(value, Address(mdp_in, offset));
1509 } else {
1510 // Put the test value into a register, so caller can use it:
1511 movptr(test_value_out, Address(mdp_in, offset));
1512 cmpptr(test_value_out, value);
1513 }
1514 jcc(Assembler::notEqual, not_equal_continue);
1515 }
1516
1517
1518 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1519 int offset_of_disp) {
1520 assert(ProfileInterpreter, "must be profiling interpreter");
1521 Address disp_address(mdp_in, offset_of_disp);
1522 addptr(mdp_in, disp_address);
1523 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1524 }
1525
1526
1527 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1528 Register reg,
1529 int offset_of_disp) {
1530 assert(ProfileInterpreter, "must be profiling interpreter");
1531 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1532 addptr(mdp_in, disp_address);
1533 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1534 }
1535
1536
1537 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1538 int constant) {
1539 assert(ProfileInterpreter, "must be profiling interpreter");
1540 addptr(mdp_in, constant);
1541 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1542 }
1543
1544
1545 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1546 assert(ProfileInterpreter, "must be profiling interpreter");
1547 push(return_bci); // save/restore across call_VM
1548 call_VM(noreg,
1549 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1550 return_bci);
1551 pop(return_bci);
1552 }
1553
1554
1555 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1556 Register bumped_count) {
1557 if (ProfileInterpreter) {
1558 Label profile_continue;
1559
1560 // If no method data exists, go to profile_continue.
1561 // Otherwise, assign to mdp
1562 test_method_data_pointer(mdp, profile_continue);
1563
1564 // We are taking a branch. Increment the taken count.
1565 // We inline increment_mdp_data_at to return bumped_count in a register
1566 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1567 Address data(mdp, in_bytes(JumpData::taken_offset()));
1568 movptr(bumped_count, data);
1569 assert(DataLayout::counter_increment == 1,
1570 "flow-free idiom only works with 1");
1571 addptr(bumped_count, DataLayout::counter_increment);
1572 sbbptr(bumped_count, 0);
1573 movptr(data, bumped_count); // Store back out
1574
1575 // The method data pointer needs to be updated to reflect the new target.
1576 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1577 bind(profile_continue);
1578 }
1579 }
1580
1581
1582 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1583 if (ProfileInterpreter) {
1584 Label profile_continue;
1585
1586 // If no method data exists, go to profile_continue.
1587 test_method_data_pointer(mdp, profile_continue);
1588
1589 // We are taking a branch. Increment the not taken count.
1590 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1591
1592 // The method data pointer needs to be updated to correspond to
1593 // the next bytecode
1594 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1595 bind(profile_continue);
1596 }
1597 }
1598
1599 void InterpreterMacroAssembler::profile_call(Register mdp) {
1600 if (ProfileInterpreter) {
1601 Label profile_continue;
1602
1603 // If no method data exists, go to profile_continue.
1604 test_method_data_pointer(mdp, profile_continue);
1605
1606 // We are making a call. Increment the count.
1607 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1608
1609 // The method data pointer needs to be updated to reflect the new target.
1610 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1611 bind(profile_continue);
1612 }
1613 }
1614
1615
1616 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1617 if (ProfileInterpreter) {
1618 Label profile_continue;
1619
1620 // If no method data exists, go to profile_continue.
1621 test_method_data_pointer(mdp, profile_continue);
1622
1623 // We are making a call. Increment the count.
1624 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1625
1626 // The method data pointer needs to be updated to reflect the new target.
1627 update_mdp_by_constant(mdp,
1628 in_bytes(VirtualCallData::
1629 virtual_call_data_size()));
1630 bind(profile_continue);
1631 }
1632 }
1633
1634
1635 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1636 Register mdp,
1637 Register reg2,
1638 bool receiver_can_be_null) {
1639 if (ProfileInterpreter) {
1640 Label profile_continue;
1641
1642 // If no method data exists, go to profile_continue.
1643 test_method_data_pointer(mdp, profile_continue);
1644
1645 Label skip_receiver_profile;
1646 if (receiver_can_be_null) {
1647 Label not_null;
1648 testptr(receiver, receiver);
1649 jccb(Assembler::notZero, not_null);
1650 // We are making a call. Increment the count for null receiver.
1651 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1652 jmp(skip_receiver_profile);
1653 bind(not_null);
1654 }
1655
1656 // Record the receiver type.
1657 record_klass_in_profile(receiver, mdp, reg2, true);
1658 bind(skip_receiver_profile);
1659
1660 // The method data pointer needs to be updated to reflect the new target.
1661 #if INCLUDE_JVMCI
1662 if (MethodProfileWidth == 0) {
1663 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1664 }
1665 #else // INCLUDE_JVMCI
1666 update_mdp_by_constant(mdp,
1667 in_bytes(VirtualCallData::
1668 virtual_call_data_size()));
1669 #endif // INCLUDE_JVMCI
1670 bind(profile_continue);
1671 }
1672 }
1673
1674 #if INCLUDE_JVMCI
1675 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) {
1676 assert_different_registers(method, mdp, reg2);
1677 if (ProfileInterpreter && MethodProfileWidth > 0) {
1678 Label profile_continue;
1679
1680 // If no method data exists, go to profile_continue.
1681 test_method_data_pointer(mdp, profile_continue);
1682
1683 Label done;
1684 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth,
1685 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1686 bind(done);
1687
1688 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1689 bind(profile_continue);
1690 }
1691 }
1692 #endif // INCLUDE_JVMCI
1693
1694 // This routine creates a state machine for updating the multi-row
1695 // type profile at a virtual call site (or other type-sensitive bytecode).
1696 // The machine visits each row (of receiver/count) until the receiver type
1697 // is found, or until it runs out of rows. At the same time, it remembers
1698 // the location of the first empty row. (An empty row records null for its
1699 // receiver, and can be allocated for a newly-observed receiver type.)
1700 // Because there are two degrees of freedom in the state, a simple linear
1701 // search will not work; it must be a decision tree. Hence this helper
1702 // function is recursive, to generate the required tree structured code.
1703 // It's the interpreter, so we are trading off code space for speed.
1704 // See below for example code.
1705 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1706 Register receiver, Register mdp,
1707 Register reg2, int start_row,
1708 Label& done, bool is_virtual_call) {
1709 if (TypeProfileWidth == 0) {
1710 if (is_virtual_call) {
1711 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1712 }
1713 #if INCLUDE_JVMCI
1714 else if (EnableJVMCI) {
1715 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1716 }
1717 #endif // INCLUDE_JVMCI
1718 } else {
1719 int non_profiled_offset = -1;
1720 if (is_virtual_call) {
1721 non_profiled_offset = in_bytes(CounterData::count_offset());
1722 }
1723 #if INCLUDE_JVMCI
1724 else if (EnableJVMCI) {
1725 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1726 }
1727 #endif // INCLUDE_JVMCI
1728
1729 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1730 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1731 }
1732 }
1733
1734 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1735 Register reg2, int start_row, Label& done, int total_rows,
1736 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1737 int non_profiled_offset) {
1738 int last_row = total_rows - 1;
1739 assert(start_row <= last_row, "must be work left to do");
1740 // Test this row for both the item and for null.
1741 // Take any of three different outcomes:
1742 // 1. found item => increment count and goto done
1743 // 2. found null => keep looking for case 1, maybe allocate this cell
1744 // 3. found something else => keep looking for cases 1 and 2
1745 // Case 3 is handled by a recursive call.
1746 for (int row = start_row; row <= last_row; row++) {
1747 Label next_test;
1748 bool test_for_null_also = (row == start_row);
1749
1750 // See if the item is item[n].
1751 int item_offset = in_bytes(item_offset_fn(row));
1752 test_mdp_data_at(mdp, item_offset, item,
1753 (test_for_null_also ? reg2 : noreg),
1754 next_test);
1755 // (Reg2 now contains the item from the CallData.)
1756
1757 // The item is item[n]. Increment count[n].
1758 int count_offset = in_bytes(item_count_offset_fn(row));
1759 increment_mdp_data_at(mdp, count_offset);
1760 jmp(done);
1761 bind(next_test);
1762
1763 if (test_for_null_also) {
1764 Label found_null;
1765 // Failed the equality check on item[n]... Test for null.
1766 testptr(reg2, reg2);
1767 if (start_row == last_row) {
1768 // The only thing left to do is handle the null case.
1769 if (non_profiled_offset >= 0) {
1770 jccb(Assembler::zero, found_null);
1771 // Item did not match any saved item and there is no empty row for it.
1772 // Increment total counter to indicate polymorphic case.
1773 increment_mdp_data_at(mdp, non_profiled_offset);
1774 jmp(done);
1775 bind(found_null);
1776 } else {
1777 jcc(Assembler::notZero, done);
1778 }
1779 break;
1780 }
1781 // Since null is rare, make it be the branch-taken case.
1782 jcc(Assembler::zero, found_null);
1783
1784 // Put all the "Case 3" tests here.
1785 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1786 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1787
1788 // Found a null. Keep searching for a matching item,
1789 // but remember that this is an empty (unused) slot.
1790 bind(found_null);
1791 }
1792 }
1793
1794 // In the fall-through case, we found no matching item, but we
1795 // observed the item[start_row] is NULL.
1796
1797 // Fill in the item field and increment the count.
1798 int item_offset = in_bytes(item_offset_fn(start_row));
1799 set_mdp_data_at(mdp, item_offset, item);
1800 int count_offset = in_bytes(item_count_offset_fn(start_row));
1801 movl(reg2, DataLayout::counter_increment);
1802 set_mdp_data_at(mdp, count_offset, reg2);
1803 if (start_row > 0) {
1804 jmp(done);
1805 }
1806 }
1807
1808 // Example state machine code for three profile rows:
1809 // // main copy of decision tree, rooted at row[1]
1810 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1811 // if (row[0].rec != NULL) {
1812 // // inner copy of decision tree, rooted at row[1]
1813 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1814 // if (row[1].rec != NULL) {
1815 // // degenerate decision tree, rooted at row[2]
1816 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1817 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1818 // row[2].init(rec); goto done;
1819 // } else {
1820 // // remember row[1] is empty
1821 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1822 // row[1].init(rec); goto done;
1823 // }
1824 // } else {
1825 // // remember row[0] is empty
1826 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1827 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1828 // row[0].init(rec); goto done;
1829 // }
1830 // done:
1831
1832 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1833 Register mdp, Register reg2,
1834 bool is_virtual_call) {
1835 assert(ProfileInterpreter, "must be profiling");
1836 Label done;
1837
1838 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1839
1840 bind (done);
1841 }
1842
1843 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1844 Register mdp) {
1845 if (ProfileInterpreter) {
1846 Label profile_continue;
1847 uint row;
1848
1849 // If no method data exists, go to profile_continue.
1850 test_method_data_pointer(mdp, profile_continue);
1851
1852 // Update the total ret count.
1853 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1854
1855 for (row = 0; row < RetData::row_limit(); row++) {
1856 Label next_test;
1857
1858 // See if return_bci is equal to bci[n]:
1859 test_mdp_data_at(mdp,
1860 in_bytes(RetData::bci_offset(row)),
1861 return_bci, noreg,
1862 next_test);
1863
1864 // return_bci is equal to bci[n]. Increment the count.
1865 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1866
1867 // The method data pointer needs to be updated to reflect the new target.
1868 update_mdp_by_offset(mdp,
1869 in_bytes(RetData::bci_displacement_offset(row)));
1870 jmp(profile_continue);
1871 bind(next_test);
1872 }
1873
1874 update_mdp_for_ret(return_bci);
1875
1876 bind(profile_continue);
1877 }
1878 }
1879
1880
1881 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1882 if (ProfileInterpreter) {
1883 Label profile_continue;
1884
1885 // If no method data exists, go to profile_continue.
1886 test_method_data_pointer(mdp, profile_continue);
1887
1888 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1889
1890 // The method data pointer needs to be updated.
1891 int mdp_delta = in_bytes(BitData::bit_data_size());
1892 if (TypeProfileCasts) {
1893 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1894 }
1895 update_mdp_by_constant(mdp, mdp_delta);
1896
1897 bind(profile_continue);
1898 }
1899 }
1900
1901
1902 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1903 if (ProfileInterpreter && TypeProfileCasts) {
1904 Label profile_continue;
1905
1906 // If no method data exists, go to profile_continue.
1907 test_method_data_pointer(mdp, profile_continue);
1908
1909 int count_offset = in_bytes(CounterData::count_offset());
1910 // Back up the address, since we have already bumped the mdp.
1911 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1912
1913 // *Decrement* the counter. We expect to see zero or small negatives.
1914 increment_mdp_data_at(mdp, count_offset, true);
1915
1916 bind (profile_continue);
1917 }
1918 }
1919
1920
1921 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1922 if (ProfileInterpreter) {
1923 Label profile_continue;
1924
1925 // If no method data exists, go to profile_continue.
1926 test_method_data_pointer(mdp, profile_continue);
1927
1928 // The method data pointer needs to be updated.
1929 int mdp_delta = in_bytes(BitData::bit_data_size());
1930 if (TypeProfileCasts) {
1931 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1932
1933 // Record the object type.
1934 record_klass_in_profile(klass, mdp, reg2, false);
1935 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1936 NOT_LP64(restore_locals();) // Restore EDI
1937 }
1938 update_mdp_by_constant(mdp, mdp_delta);
1939
1940 bind(profile_continue);
1941 }
1942 }
1943
1944
1945 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1946 if (ProfileInterpreter) {
1947 Label profile_continue;
1948
1949 // If no method data exists, go to profile_continue.
1950 test_method_data_pointer(mdp, profile_continue);
1951
1952 // Update the default case count
1953 increment_mdp_data_at(mdp,
1954 in_bytes(MultiBranchData::default_count_offset()));
1955
1956 // The method data pointer needs to be updated.
1957 update_mdp_by_offset(mdp,
1958 in_bytes(MultiBranchData::
1959 default_displacement_offset()));
1960
1961 bind(profile_continue);
1962 }
1963 }
1964
1965
1966 void InterpreterMacroAssembler::profile_switch_case(Register index,
1967 Register mdp,
1968 Register reg2) {
1969 if (ProfileInterpreter) {
1970 Label profile_continue;
1971
1972 // If no method data exists, go to profile_continue.
1973 test_method_data_pointer(mdp, profile_continue);
1974
1975 // Build the base (index * per_case_size_in_bytes()) +
1976 // case_array_offset_in_bytes()
1977 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1978 imulptr(index, reg2); // XXX l ?
1979 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1980
1981 // Update the case count
1982 increment_mdp_data_at(mdp,
1983 index,
1984 in_bytes(MultiBranchData::relative_count_offset()));
1985
1986 // The method data pointer needs to be updated.
1987 update_mdp_by_offset(mdp,
1988 index,
1989 in_bytes(MultiBranchData::
1990 relative_displacement_offset()));
1991
1992 bind(profile_continue);
1993 }
1994 }
1995
1996
1997
1998 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1999 if (state == atos) {
2000 MacroAssembler::verify_oop(reg);
2001 }
2002 }
2003
2004 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2005 #ifndef _LP64
2006 if ((state == ftos && UseSSE < 1) ||
2007 (state == dtos && UseSSE < 2)) {
2008 MacroAssembler::verify_FPU(stack_depth);
2009 }
2010 #endif
2011 }
2012
2013 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
2014 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
2015 int increment, Address mask,
2016 Register scratch, bool preloaded,
2017 Condition cond, Label* where) {
2018 if (!preloaded) {
2019 movl(scratch, counter_addr);
2020 }
2021 incrementl(scratch, increment);
2022 movl(counter_addr, scratch);
2023 andl(scratch, mask);
2024 jcc(cond, *where);
2025 }
2026
2027 void InterpreterMacroAssembler::notify_method_entry() {
2028 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2029 // track stack depth. If it is possible to enter interp_only_mode we add
2030 // the code to check if the event should be sent.
2031 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2032 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2033 if (JvmtiExport::can_post_interpreter_events()) {
2034 Label L;
2035 NOT_LP64(get_thread(rthread);)
2036 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2037 testl(rdx, rdx);
2038 jcc(Assembler::zero, L);
2039 call_VM(noreg, CAST_FROM_FN_PTR(address,
2040 InterpreterRuntime::post_method_entry));
2041 bind(L);
2042 }
2043
2044 {
2045 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2046 NOT_LP64(get_thread(rthread);)
2047 get_method(rarg);
2048 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2049 rthread, rarg);
2050 }
2051
2052 // RedefineClasses() tracing support for obsolete method entry
2053 if (log_is_enabled(Trace, redefine, class, obsolete)) {
2054 NOT_LP64(get_thread(rthread);)
2055 get_method(rarg);
2056 call_VM_leaf(
2057 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2058 rthread, rarg);
2059 }
2060 }
2061
2062
2063 void InterpreterMacroAssembler::notify_method_exit(
2064 TosState state, NotifyMethodExitMode mode) {
2065 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2066 // track stack depth. If it is possible to enter interp_only_mode we add
2067 // the code to check if the event should be sent.
2068 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2069 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2070 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2071 Label L;
2072 // Note: frame::interpreter_frame_result has a dependency on how the
2073 // method result is saved across the call to post_method_exit. If this
2074 // is changed then the interpreter_frame_result implementation will
2075 // need to be updated too.
2076
2077 // template interpreter will leave the result on the top of the stack.
2078 push(state);
2079 NOT_LP64(get_thread(rthread);)
2080 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2081 testl(rdx, rdx);
2082 jcc(Assembler::zero, L);
2083 call_VM(noreg,
2084 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2085 bind(L);
2086 pop(state);
2087 }
2088
2089 {
2090 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2091 push(state);
2092 NOT_LP64(get_thread(rthread);)
2093 get_method(rarg);
2094 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2095 rthread, rarg);
2096 pop(state);
2097 }
2098 }