1 /*
2 * Copyright (c) 1997, 2022, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "compiler/compiler_globals.hpp"
27 #include "interp_masm_x86.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "logging/log.hpp"
31 #include "oops/arrayOop.hpp"
32 #include "oops/markWord.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "prims/jvmtiExport.hpp"
36 #include "prims/jvmtiThreadState.hpp"
37 #include "runtime/basicLock.hpp"
38 #include "runtime/frame.inline.hpp"
39 #include "runtime/safepointMechanism.hpp"
40 #include "runtime/sharedRuntime.hpp"
41 #include "runtime/thread.inline.hpp"
42 #include "utilities/powerOfTwo.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 interp_verify_oop(obj, atos);
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 Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
63 load_klass(obj, obj, tmp_load_klass);
64
65 xorptr(obj, mdo_addr);
66 testptr(obj, TypeEntries::type_klass_mask);
67 jccb(Assembler::zero, next); // klass seen before, nothing to
68 // do. The unknown bit may have been
69 // set already but no need to check.
70
71 testptr(obj, TypeEntries::type_unknown);
72 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
73
74 cmpptr(mdo_addr, 0);
75 jccb(Assembler::equal, none);
76 cmpptr(mdo_addr, TypeEntries::null_seen);
77 jccb(Assembler::equal, none);
78 // There is a chance that the checks above (re-reading profiling
79 // data from memory) fail if another thread has just set the
80 // profiling to this obj's klass
81 xorptr(obj, mdo_addr);
82 testptr(obj, TypeEntries::type_klass_mask);
83 jccb(Assembler::zero, next);
84
85 // different than before. Cannot keep accurate profile.
86 orptr(mdo_addr, TypeEntries::type_unknown);
87 jmpb(next);
88
89 bind(none);
90 // first time here. Set profile type.
91 movptr(mdo_addr, obj);
92
93 bind(next);
94 }
95
96 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
97 if (!ProfileInterpreter) {
98 return;
99 }
100
101 if (MethodData::profile_arguments() || MethodData::profile_return()) {
102 Label profile_continue;
103
104 test_method_data_pointer(mdp, profile_continue);
105
106 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
107
108 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
109 jcc(Assembler::notEqual, profile_continue);
110
111 if (MethodData::profile_arguments()) {
112 Label done;
113 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
114 addptr(mdp, off_to_args);
115
116 for (int i = 0; i < TypeProfileArgsLimit; i++) {
117 if (i > 0 || MethodData::profile_return()) {
118 // If return value type is profiled we may have no argument to profile
119 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
120 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
121 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
122 jcc(Assembler::less, done);
123 }
124 movptr(tmp, Address(callee, Method::const_offset()));
125 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
126 // stack offset o (zero based) from the start of the argument
127 // list, for n arguments translates into offset n - o - 1 from
128 // the end of the argument list
129 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
130 subl(tmp, 1);
131 Address arg_addr = argument_address(tmp);
132 movptr(tmp, arg_addr);
133
134 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
135 profile_obj_type(tmp, mdo_arg_addr);
136
137 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
138 addptr(mdp, to_add);
139 off_to_args += to_add;
140 }
141
142 if (MethodData::profile_return()) {
143 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
144 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
145 }
146
147 bind(done);
148
149 if (MethodData::profile_return()) {
150 // We're right after the type profile for the last
151 // argument. tmp is the number of cells left in the
152 // CallTypeData/VirtualCallTypeData to reach its end. Non null
153 // if there's a return to profile.
154 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
155 shll(tmp, log2i_exact((int)DataLayout::cell_size));
156 addptr(mdp, tmp);
157 }
158 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
159 } else {
160 assert(MethodData::profile_return(), "either profile call args or call ret");
161 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
162 }
163
164 // mdp points right after the end of the
165 // CallTypeData/VirtualCallTypeData, right after the cells for the
166 // return value type if there's one
167
168 bind(profile_continue);
169 }
170 }
171
172 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
173 assert_different_registers(mdp, ret, tmp, _bcp_register);
174 if (ProfileInterpreter && MethodData::profile_return()) {
175 Label profile_continue;
176
177 test_method_data_pointer(mdp, profile_continue);
178
179 if (MethodData::profile_return_jsr292_only()) {
180 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
181
182 // If we don't profile all invoke bytecodes we must make sure
183 // it's a bytecode we indeed profile. We can't go back to the
184 // beginning of the ProfileData we intend to update to check its
185 // type because we're right after it and we don't known its
186 // length
187 Label do_profile;
188 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
189 jcc(Assembler::equal, do_profile);
190 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
191 jcc(Assembler::equal, do_profile);
192 get_method(tmp);
193 cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
194 jcc(Assembler::notEqual, profile_continue);
195
196 bind(do_profile);
197 }
198
199 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
200 mov(tmp, ret);
201 profile_obj_type(tmp, mdo_ret_addr);
202
203 bind(profile_continue);
204 }
205 }
206
207 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
208 if (ProfileInterpreter && MethodData::profile_parameters()) {
209 Label profile_continue;
210
211 test_method_data_pointer(mdp, profile_continue);
212
213 // Load the offset of the area within the MDO used for
214 // parameters. If it's negative we're not profiling any parameters
215 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
216 testl(tmp1, tmp1);
217 jcc(Assembler::negative, profile_continue);
218
219 // Compute a pointer to the area for parameters from the offset
220 // and move the pointer to the slot for the last
221 // parameters. Collect profiling from last parameter down.
222 // mdo start + parameters offset + array length - 1
223 addptr(mdp, tmp1);
224 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
225 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
226
227 Label loop;
228 bind(loop);
229
230 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
231 int type_base = in_bytes(ParametersTypeData::type_offset(0));
232 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
233 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
234 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
235
236 // load offset on the stack from the slot for this parameter
237 movptr(tmp2, arg_off);
238 negptr(tmp2);
239 // read the parameter from the local area
240 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
241
242 // profile the parameter
243 profile_obj_type(tmp2, arg_type);
244
245 // go to next parameter
246 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
247 jcc(Assembler::positive, loop);
248
249 bind(profile_continue);
250 }
251 }
252
253 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
254 int number_of_arguments) {
255 // interpreter specific
256 //
257 // Note: No need to save/restore bcp & locals registers
258 // since these are callee saved registers and no blocking/
259 // GC can happen in leaf calls.
260 // Further Note: DO NOT save/restore bcp/locals. If a caller has
261 // already saved them so that it can use rsi/rdi as temporaries
262 // then a save/restore here will DESTROY the copy the caller
263 // saved! There used to be a save_bcp() that only happened in
264 // the ASSERT path (no restore_bcp). Which caused bizarre failures
265 // when jvm built with ASSERTs.
266 #ifdef ASSERT
267 {
268 Label L;
269 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
270 jcc(Assembler::equal, L);
271 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
272 " last_sp != NULL");
273 bind(L);
274 }
275 #endif
276 // super call
277 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
278 // interpreter specific
279 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
280 // but since they may not have been saved (and we don't want to
281 // save them here (see note above) the assert is invalid.
282 }
283
284 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
285 Register java_thread,
286 Register last_java_sp,
287 address entry_point,
288 int number_of_arguments,
289 bool check_exceptions) {
290 // interpreter specific
291 //
292 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
293 // really make a difference for these runtime calls, since they are
294 // slow anyway. Btw., bcp must be saved/restored since it may change
295 // due to GC.
296 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
297 save_bcp();
298 #ifdef ASSERT
299 {
300 Label L;
301 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
302 jcc(Assembler::equal, L);
303 stop("InterpreterMacroAssembler::call_VM_base:"
304 " last_sp != NULL");
305 bind(L);
306 }
307 #endif /* ASSERT */
308 // super call
309 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
310 entry_point, number_of_arguments,
311 check_exceptions);
312 // interpreter specific
313 restore_bcp();
314 restore_locals();
315 }
316
317 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
318 if (JvmtiExport::can_pop_frame()) {
319 Label L;
320 // Initiate popframe handling only if it is not already being
321 // processed. If the flag has the popframe_processing bit set, it
322 // means that this code is called *during* popframe handling - we
323 // don't want to reenter.
324 // This method is only called just after the call into the vm in
325 // call_VM_base, so the arg registers are available.
326 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
327 LP64_ONLY(c_rarg0);
328 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
329 testl(pop_cond, JavaThread::popframe_pending_bit);
330 jcc(Assembler::zero, L);
331 testl(pop_cond, JavaThread::popframe_processing_bit);
332 jcc(Assembler::notZero, L);
333 // Call Interpreter::remove_activation_preserving_args_entry() to get the
334 // address of the same-named entrypoint in the generated interpreter code.
335 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
336 jmp(rax);
337 bind(L);
338 NOT_LP64(get_thread(java_thread);)
339 }
340 }
341
342 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
343 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
344 NOT_LP64(get_thread(thread);)
345 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
346 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
347 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
348 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
349 #ifdef _LP64
350 switch (state) {
351 case atos: movptr(rax, oop_addr);
352 movptr(oop_addr, (int32_t)NULL_WORD);
353 interp_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 atos: movptr(rax, oop_addr);
373 movptr(oop_addr, NULL_WORD);
374 interp_verify_oop(rax, state); break;
375 case ltos:
376 movl(rdx, val_addr1); // fall through
377 case btos: // fall through
378 case ztos: // fall through
379 case ctos: // fall through
380 case stos: // fall through
381 case itos: movl(rax, val_addr); break;
382 case ftos: load_float(val_addr); break;
383 case dtos: load_double(val_addr); break;
384 case vtos: /* nothing to do */ break;
385 default : ShouldNotReachHere();
386 }
387 #endif // _LP64
388 // Clean up tos value in the thread object
389 movl(tos_addr, (int32_t) ilgl);
390 movptr(val_addr, NULL_WORD);
391 NOT_LP64(movptr(val_addr1, NULL_WORD);)
392 }
393
394
395 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
396 if (JvmtiExport::can_force_early_return()) {
397 Label L;
398 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
399 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
400
401 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
402 testptr(tmp, tmp);
403 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
404
405 // Initiate earlyret handling only if it is not already being processed.
406 // If the flag has the earlyret_processing bit set, it means that this code
407 // is called *during* earlyret handling - we don't want to reenter.
408 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
409 cmpl(tmp, JvmtiThreadState::earlyret_pending);
410 jcc(Assembler::notEqual, L);
411
412 // Call Interpreter::remove_activation_early_entry() to get the address of the
413 // same-named entrypoint in the generated interpreter code.
414 NOT_LP64(get_thread(java_thread);)
415 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
416 #ifdef _LP64
417 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
418 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
419 #else
420 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
421 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
422 #endif // _LP64
423 jmp(rax);
424 bind(L);
425 NOT_LP64(get_thread(java_thread);)
426 }
427 }
428
429 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
430 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
431 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
432 bswapl(reg);
433 shrl(reg, 16);
434 }
435
436 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
437 int bcp_offset,
438 size_t index_size) {
439 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
440 if (index_size == sizeof(u2)) {
441 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
442 } else if (index_size == sizeof(u4)) {
443 movl(index, Address(_bcp_register, bcp_offset));
444 // Check if the secondary index definition is still ~x, otherwise
445 // we have to change the following assembler code to calculate the
446 // plain index.
447 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
448 notl(index); // convert to plain index
449 } else if (index_size == sizeof(u1)) {
450 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
451 } else {
452 ShouldNotReachHere();
453 }
454 }
455
456 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
457 Register index,
458 int bcp_offset,
459 size_t index_size) {
460 assert_different_registers(cache, index);
461 get_cache_index_at_bcp(index, bcp_offset, index_size);
462 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
463 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
464 // convert from field index to ConstantPoolCacheEntry index
465 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
466 shll(index, 2);
467 }
468
469 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
470 Register index,
471 Register bytecode,
472 int byte_no,
473 int bcp_offset,
474 size_t index_size) {
475 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
476 // We use a 32-bit load here since the layout of 64-bit words on
477 // little-endian machines allow us that.
478 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
479 const int shift_count = (1 + byte_no) * BitsPerByte;
480 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
481 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
482 "correct shift count");
483 shrl(bytecode, shift_count);
484 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
485 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
486 }
487
488 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
489 Register tmp,
490 int bcp_offset,
491 size_t index_size) {
492 assert_different_registers(cache, tmp);
493
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(Register result,
508 Register index,
509 Register tmp) {
510 assert_different_registers(result, index);
511
512 get_constant_pool(result);
513 // load pointer for resolved_references[] objArray
514 movptr(result, Address(result, ConstantPool::cache_offset_in_bytes()));
515 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes()));
516 resolve_oop_handle(result, tmp);
517 load_heap_oop(result, Address(result, index,
518 UseCompressedOops ? Address::times_4 : Address::times_ptr,
519 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
520 }
521
522 // load cpool->resolved_klass_at(index)
523 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
524 Register cpool,
525 Register index) {
526 assert_different_registers(cpool, index);
527
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 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
535 Register method,
536 Register cache,
537 Register index) {
538 assert_different_registers(cache, index);
539
540 const int method_offset = in_bytes(
541 ConstantPoolCache::base_offset() +
542 ((byte_no == TemplateTable::f2_byte)
543 ? ConstantPoolCacheEntry::f2_offset()
544 : ConstantPoolCacheEntry::f1_offset()));
545
546 movptr(method, Address(cache, index, Address::times_ptr, method_offset)); // get f1 Method*
547 }
548
549 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
550 // subtype of super_klass.
551 //
552 // Args:
553 // rax: superklass
554 // Rsub_klass: subklass
555 //
556 // Kills:
557 // rcx, rdi
558 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
559 Label& ok_is_subtype) {
560 assert(Rsub_klass != rax, "rax holds superklass");
561 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
562 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
563 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
564 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
565
566 // Profile the not-null value's klass.
567 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
568
569 // Do the check.
570 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
571
572 // Profile the failure of the check.
573 profile_typecheck_failed(rcx); // blows rcx
574 }
575
576
577 #ifndef _LP64
578 void InterpreterMacroAssembler::f2ieee() {
579 if (IEEEPrecision) {
580 fstp_s(Address(rsp, 0));
581 fld_s(Address(rsp, 0));
582 }
583 }
584
585
586 void InterpreterMacroAssembler::d2ieee() {
587 if (IEEEPrecision) {
588 fstp_d(Address(rsp, 0));
589 fld_d(Address(rsp, 0));
590 }
591 }
592 #endif // _LP64
593
594 // Java Expression Stack
595
596 void InterpreterMacroAssembler::pop_ptr(Register r) {
597 pop(r);
598 }
599
600 void InterpreterMacroAssembler::push_ptr(Register r) {
601 push(r);
602 }
603
604 void InterpreterMacroAssembler::push_i(Register r) {
605 push(r);
606 }
607
608 void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
609 push(r);
610 }
611
612 void InterpreterMacroAssembler::push_f(XMMRegister r) {
613 subptr(rsp, wordSize);
614 movflt(Address(rsp, 0), r);
615 }
616
617 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
618 movflt(r, Address(rsp, 0));
619 addptr(rsp, wordSize);
620 }
621
622 void InterpreterMacroAssembler::push_d(XMMRegister r) {
623 subptr(rsp, 2 * wordSize);
624 movdbl(Address(rsp, 0), r);
625 }
626
627 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
628 movdbl(r, Address(rsp, 0));
629 addptr(rsp, 2 * Interpreter::stackElementSize);
630 }
631
632 #ifdef _LP64
633 void InterpreterMacroAssembler::pop_i(Register r) {
634 // XXX can't use pop currently, upper half non clean
635 movl(r, Address(rsp, 0));
636 addptr(rsp, wordSize);
637 }
638
639 void InterpreterMacroAssembler::pop_l(Register r) {
640 movq(r, Address(rsp, 0));
641 addptr(rsp, 2 * Interpreter::stackElementSize);
642 }
643
644 void InterpreterMacroAssembler::push_l(Register r) {
645 subptr(rsp, 2 * wordSize);
646 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
647 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
648 }
649
650 void InterpreterMacroAssembler::pop(TosState state) {
651 switch (state) {
652 case atos: pop_ptr(); break;
653 case btos:
654 case ztos:
655 case ctos:
656 case stos:
657 case itos: pop_i(); break;
658 case ltos: pop_l(); break;
659 case ftos: pop_f(xmm0); break;
660 case dtos: pop_d(xmm0); break;
661 case vtos: /* nothing to do */ break;
662 default: ShouldNotReachHere();
663 }
664 interp_verify_oop(rax, state);
665 }
666
667 void InterpreterMacroAssembler::push(TosState state) {
668 interp_verify_oop(rax, state);
669 switch (state) {
670 case atos: push_ptr(); break;
671 case btos:
672 case ztos:
673 case ctos:
674 case stos:
675 case itos: push_i(); break;
676 case ltos: push_l(); break;
677 case ftos: push_f(xmm0); break;
678 case dtos: push_d(xmm0); break;
679 case vtos: /* nothing to do */ break;
680 default : ShouldNotReachHere();
681 }
682 }
683 #else
684 void InterpreterMacroAssembler::pop_i(Register r) {
685 pop(r);
686 }
687
688 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
689 pop(lo);
690 pop(hi);
691 }
692
693 void InterpreterMacroAssembler::pop_f() {
694 fld_s(Address(rsp, 0));
695 addptr(rsp, 1 * wordSize);
696 }
697
698 void InterpreterMacroAssembler::pop_d() {
699 fld_d(Address(rsp, 0));
700 addptr(rsp, 2 * wordSize);
701 }
702
703
704 void InterpreterMacroAssembler::pop(TosState state) {
705 switch (state) {
706 case atos: pop_ptr(rax); break;
707 case btos: // fall through
708 case ztos: // fall through
709 case ctos: // fall through
710 case stos: // fall through
711 case itos: pop_i(rax); break;
712 case ltos: pop_l(rax, rdx); break;
713 case ftos:
714 if (UseSSE >= 1) {
715 pop_f(xmm0);
716 } else {
717 pop_f();
718 }
719 break;
720 case dtos:
721 if (UseSSE >= 2) {
722 pop_d(xmm0);
723 } else {
724 pop_d();
725 }
726 break;
727 case vtos: /* nothing to do */ break;
728 default : ShouldNotReachHere();
729 }
730 interp_verify_oop(rax, state);
731 }
732
733
734 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
735 push(hi);
736 push(lo);
737 }
738
739 void InterpreterMacroAssembler::push_f() {
740 // Do not schedule for no AGI! Never write beyond rsp!
741 subptr(rsp, 1 * wordSize);
742 fstp_s(Address(rsp, 0));
743 }
744
745 void InterpreterMacroAssembler::push_d() {
746 // Do not schedule for no AGI! Never write beyond rsp!
747 subptr(rsp, 2 * wordSize);
748 fstp_d(Address(rsp, 0));
749 }
750
751
752 void InterpreterMacroAssembler::push(TosState state) {
753 interp_verify_oop(rax, state);
754 switch (state) {
755 case atos: push_ptr(rax); break;
756 case btos: // fall through
757 case ztos: // fall through
758 case ctos: // fall through
759 case stos: // fall through
760 case itos: push_i(rax); break;
761 case ltos: push_l(rax, rdx); break;
762 case ftos:
763 if (UseSSE >= 1) {
764 push_f(xmm0);
765 } else {
766 push_f();
767 }
768 break;
769 case dtos:
770 if (UseSSE >= 2) {
771 push_d(xmm0);
772 } else {
773 push_d();
774 }
775 break;
776 case vtos: /* nothing to do */ break;
777 default : ShouldNotReachHere();
778 }
779 }
780 #endif // _LP64
781
782
783 // Helpers for swap and dup
784 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
785 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
786 }
787
788 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
789 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
790 }
791
792
793 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
794 // set sender sp
795 lea(_bcp_register, Address(rsp, wordSize));
796 // record last_sp
797 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
798 }
799
800
801 // Jump to from_interpreted entry of a call unless single stepping is possible
802 // in this thread in which case we must call the i2i entry
803 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
804 prepare_to_jump_from_interpreted();
805
806 if (JvmtiExport::can_post_interpreter_events()) {
807 Label run_compiled_code;
808 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
809 // compiled code in threads for which the event is enabled. Check here for
810 // interp_only_mode if these events CAN be enabled.
811 // interp_only is an int, on little endian it is sufficient to test the byte only
812 // Is a cmpl faster?
813 LP64_ONLY(temp = r15_thread;)
814 NOT_LP64(get_thread(temp);)
815 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
816 jccb(Assembler::zero, run_compiled_code);
817 jmp(Address(method, Method::interpreter_entry_offset()));
818 bind(run_compiled_code);
819 }
820
821 jmp(Address(method, Method::from_interpreted_offset()));
822 }
823
824 // The following two routines provide a hook so that an implementation
825 // can schedule the dispatch in two parts. x86 does not do this.
826 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
827 // Nothing x86 specific to be done here
828 }
829
830 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
831 dispatch_next(state, step);
832 }
833
834 void InterpreterMacroAssembler::dispatch_base(TosState state,
835 address* table,
836 bool verifyoop,
837 bool generate_poll) {
838 verify_FPU(1, state);
839 if (VerifyActivationFrameSize) {
840 Label L;
841 mov(rcx, rbp);
842 subptr(rcx, rsp);
843 int32_t min_frame_size =
844 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
845 wordSize;
846 cmpptr(rcx, (int32_t)min_frame_size);
847 jcc(Assembler::greaterEqual, L);
848 stop("broken stack frame");
849 bind(L);
850 }
851 if (verifyoop) {
852 interp_verify_oop(rax, state);
853 }
854
855 address* const safepoint_table = Interpreter::safept_table(state);
856 #ifdef _LP64
857 Label no_safepoint, dispatch;
858 if (table != safepoint_table && generate_poll) {
859 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
860 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
861
862 jccb(Assembler::zero, no_safepoint);
863 lea(rscratch1, ExternalAddress((address)safepoint_table));
864 jmpb(dispatch);
865 }
866
867 bind(no_safepoint);
868 lea(rscratch1, ExternalAddress((address)table));
869 bind(dispatch);
870 jmp(Address(rscratch1, rbx, Address::times_8));
871
872 #else
873 Address index(noreg, rbx, Address::times_ptr);
874 if (table != safepoint_table && generate_poll) {
875 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
876 Label no_safepoint;
877 const Register thread = rcx;
878 get_thread(thread);
879 testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
880
881 jccb(Assembler::zero, no_safepoint);
882 ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
883 jump(dispatch_addr);
884 bind(no_safepoint);
885 }
886
887 {
888 ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
889 jump(dispatch_addr);
890 }
891 #endif // _LP64
892 }
893
894 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
895 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
896 }
897
898 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
899 dispatch_base(state, Interpreter::normal_table(state));
900 }
901
902 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
903 dispatch_base(state, Interpreter::normal_table(state), false);
904 }
905
906
907 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
908 // load next bytecode (load before advancing _bcp_register to prevent AGI)
909 load_unsigned_byte(rbx, Address(_bcp_register, step));
910 // advance _bcp_register
911 increment(_bcp_register, step);
912 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
913 }
914
915 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
916 // load current bytecode
917 load_unsigned_byte(rbx, Address(_bcp_register, 0));
918 dispatch_base(state, table);
919 }
920
921 void InterpreterMacroAssembler::narrow(Register result) {
922
923 // Get method->_constMethod->_result_type
924 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
925 movptr(rcx, Address(rcx, Method::const_offset()));
926 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
927
928 Label done, notBool, notByte, notChar;
929
930 // common case first
931 cmpl(rcx, T_INT);
932 jcc(Assembler::equal, done);
933
934 // mask integer result to narrower return type.
935 cmpl(rcx, T_BOOLEAN);
936 jcc(Assembler::notEqual, notBool);
937 andl(result, 0x1);
938 jmp(done);
939
940 bind(notBool);
941 cmpl(rcx, T_BYTE);
942 jcc(Assembler::notEqual, notByte);
943 LP64_ONLY(movsbl(result, result);)
944 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
945 NOT_LP64(sarl(result, 24);) // and sign-extend byte
946 jmp(done);
947
948 bind(notByte);
949 cmpl(rcx, T_CHAR);
950 jcc(Assembler::notEqual, notChar);
951 LP64_ONLY(movzwl(result, result);)
952 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
953 jmp(done);
954
955 bind(notChar);
956 // cmpl(rcx, T_SHORT); // all that's left
957 // jcc(Assembler::notEqual, done);
958 LP64_ONLY(movswl(result, result);)
959 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
960 NOT_LP64(sarl(result, 16);) // and sign-extend short
961
962 // Nothing to do for T_INT
963 bind(done);
964 }
965
966 // remove activation
967 //
968 // Apply stack watermark barrier.
969 // Unlock the receiver if this is a synchronized method.
970 // Unlock any Java monitors from synchronized blocks.
971 // Remove the activation from the stack.
972 //
973 // If there are locked Java monitors
974 // If throw_monitor_exception
975 // throws IllegalMonitorStateException
976 // Else if install_monitor_exception
977 // installs IllegalMonitorStateException
978 // Else
979 // no error processing
980 void InterpreterMacroAssembler::remove_activation(
981 TosState state,
982 Register ret_addr,
983 bool throw_monitor_exception,
984 bool install_monitor_exception,
985 bool notify_jvmdi) {
986 // Note: Registers rdx xmm0 may be in use for the
987 // result check if synchronized method
988 Label unlocked, unlock, no_unlock;
989
990 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
991 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
992 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
993 // monitor pointers need different register
994 // because rdx may have the result in it
995 NOT_LP64(get_thread(rthread);)
996
997 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
998 // that would normally not be safe to use. Such bad returns into unsafe territory of
999 // the stack, will call InterpreterRuntime::at_unwind.
1000 Label slow_path;
1001 Label fast_path;
1002 safepoint_poll(slow_path, rthread, true /* at_return */, false /* in_nmethod */);
1003 jmp(fast_path);
1004 bind(slow_path);
1005 push(state);
1006 set_last_Java_frame(rthread, noreg, rbp, (address)pc());
1007 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
1008 NOT_LP64(get_thread(rthread);) // call_VM clobbered it, restore
1009 reset_last_Java_frame(rthread, true);
1010 pop(state);
1011 bind(fast_path);
1012
1013 // get the value of _do_not_unlock_if_synchronized into rdx
1014 const Address do_not_unlock_if_synchronized(rthread,
1015 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1016 movbool(rbx, do_not_unlock_if_synchronized);
1017 movbool(do_not_unlock_if_synchronized, false); // reset the flag
1018
1019 // get method access flags
1020 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
1021 movl(rcx, Address(rcx, Method::access_flags_offset()));
1022 testl(rcx, JVM_ACC_SYNCHRONIZED);
1023 jcc(Assembler::zero, unlocked);
1024
1025 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
1026 // is set.
1027 testbool(rbx);
1028 jcc(Assembler::notZero, no_unlock);
1029
1030 // unlock monitor
1031 push(state); // save result
1032
1033 // BasicObjectLock will be first in list, since this is a
1034 // synchronized method. However, need to check that the object has
1035 // not been unlocked by an explicit monitorexit bytecode.
1036 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
1037 wordSize - (int) sizeof(BasicObjectLock));
1038 // We use c_rarg1/rdx so that if we go slow path it will be the correct
1039 // register for unlock_object to pass to VM directly
1040 lea(robj, monitor); // address of first monitor
1041
1042 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
1043 testptr(rax, rax);
1044 jcc(Assembler::notZero, unlock);
1045
1046 pop(state);
1047 if (throw_monitor_exception) {
1048 // Entry already unlocked, need to throw exception
1049 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
1050 call_VM(noreg, CAST_FROM_FN_PTR(address,
1051 InterpreterRuntime::throw_illegal_monitor_state_exception));
1052 should_not_reach_here();
1053 } else {
1054 // Monitor already unlocked during a stack unroll. If requested,
1055 // install an illegal_monitor_state_exception. Continue with
1056 // stack unrolling.
1057 if (install_monitor_exception) {
1058 NOT_LP64(empty_FPU_stack();)
1059 call_VM(noreg, CAST_FROM_FN_PTR(address,
1060 InterpreterRuntime::new_illegal_monitor_state_exception));
1061 }
1062 jmp(unlocked);
1063 }
1064
1065 bind(unlock);
1066 unlock_object(robj);
1067 pop(state);
1068
1069 // Check that for block-structured locking (i.e., that all locked
1070 // objects has been unlocked)
1071 bind(unlocked);
1072
1073 // rax, rdx: Might contain return value
1074
1075 // Check that all monitors are unlocked
1076 {
1077 Label loop, exception, entry, restart;
1078 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1079 const Address monitor_block_top(
1080 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1081 const Address monitor_block_bot(
1082 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1083
1084 bind(restart);
1085 // We use c_rarg1 so that if we go slow path it will be the correct
1086 // register for unlock_object to pass to VM directly
1087 movptr(rmon, monitor_block_top); // points to current entry, starting
1088 // with top-most entry
1089 lea(rbx, monitor_block_bot); // points to word before bottom of
1090 // monitor block
1091 jmp(entry);
1092
1093 // Entry already locked, need to throw exception
1094 bind(exception);
1095
1096 if (throw_monitor_exception) {
1097 // Throw exception
1098 NOT_LP64(empty_FPU_stack();)
1099 MacroAssembler::call_VM(noreg,
1100 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1101 throw_illegal_monitor_state_exception));
1102 should_not_reach_here();
1103 } else {
1104 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1105 // Unlock does not block, so don't have to worry about the frame.
1106 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1107
1108 push(state);
1109 mov(robj, rmon); // nop if robj and rmon are the same
1110 unlock_object(robj);
1111 pop(state);
1112
1113 if (install_monitor_exception) {
1114 NOT_LP64(empty_FPU_stack();)
1115 call_VM(noreg, CAST_FROM_FN_PTR(address,
1116 InterpreterRuntime::
1117 new_illegal_monitor_state_exception));
1118 }
1119
1120 jmp(restart);
1121 }
1122
1123 bind(loop);
1124 // check if current entry is used
1125 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
1126 jcc(Assembler::notEqual, exception);
1127
1128 addptr(rmon, entry_size); // otherwise advance to next entry
1129 bind(entry);
1130 cmpptr(rmon, rbx); // check if bottom reached
1131 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1132 }
1133
1134 bind(no_unlock);
1135
1136 // jvmti support
1137 if (notify_jvmdi) {
1138 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1139 } else {
1140 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1141 }
1142
1143 // remove activation
1144 // get sender sp
1145 movptr(rbx,
1146 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1147 if (StackReservedPages > 0) {
1148 // testing if reserved zone needs to be re-enabled
1149 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1150 Label no_reserved_zone_enabling;
1151
1152 NOT_LP64(get_thread(rthread);)
1153
1154 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
1155 jcc(Assembler::equal, no_reserved_zone_enabling);
1156
1157 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1158 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1159
1160 call_VM_leaf(
1161 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1162 call_VM(noreg, CAST_FROM_FN_PTR(address,
1163 InterpreterRuntime::throw_delayed_StackOverflowError));
1164 should_not_reach_here();
1165
1166 bind(no_reserved_zone_enabling);
1167 }
1168 leave(); // remove frame anchor
1169 pop(ret_addr); // get return address
1170 mov(rsp, rbx); // set sp to sender sp
1171 }
1172
1173 void InterpreterMacroAssembler::get_method_counters(Register method,
1174 Register mcs, Label& skip) {
1175 Label has_counters;
1176 movptr(mcs, Address(method, Method::method_counters_offset()));
1177 testptr(mcs, mcs);
1178 jcc(Assembler::notZero, has_counters);
1179 call_VM(noreg, CAST_FROM_FN_PTR(address,
1180 InterpreterRuntime::build_method_counters), method);
1181 movptr(mcs, Address(method,Method::method_counters_offset()));
1182 testptr(mcs, mcs);
1183 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1184 bind(has_counters);
1185 }
1186
1187
1188 // Lock object
1189 //
1190 // Args:
1191 // rdx, c_rarg1: BasicObjectLock to be used for locking
1192 //
1193 // Kills:
1194 // rax, rbx
1195 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1196 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1197 "The argument is only for looks. It must be c_rarg1");
1198
1199 if (UseHeavyMonitors) {
1200 call_VM(noreg,
1201 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1202 lock_reg);
1203 } else {
1204 Label done;
1205
1206 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1207 const Register tmp_reg = rbx;
1208 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1209 const Register rklass_decode_tmp = LP64_ONLY(rscratch1) NOT_LP64(noreg);
1210
1211 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
1212 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
1213 const int mark_offset = lock_offset +
1214 BasicLock::displaced_header_offset_in_bytes();
1215
1216 Label slow_case;
1217
1218 // Load object pointer into obj_reg
1219 movptr(obj_reg, Address(lock_reg, obj_offset));
1220
1221 if (DiagnoseSyncOnValueBasedClasses != 0) {
1222 load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
1223 movl(tmp_reg, Address(tmp_reg, Klass::access_flags_offset()));
1224 testl(tmp_reg, JVM_ACC_IS_VALUE_BASED_CLASS);
1225 jcc(Assembler::notZero, slow_case);
1226 }
1227
1228 // Load immediate 1 into swap_reg %rax
1229 movl(swap_reg, (int32_t)1);
1230
1231 // Load (object->mark() | 1) into swap_reg %rax
1232 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1233
1234 // Save (object->mark() | 1) into BasicLock's displaced header
1235 movptr(Address(lock_reg, mark_offset), swap_reg);
1236
1237 assert(lock_offset == 0,
1238 "displaced header must be first word in BasicObjectLock");
1239
1240 lock();
1241 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1242 jcc(Assembler::zero, done);
1243
1244 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1245
1246 // Fast check for recursive lock.
1247 //
1248 // Can apply the optimization only if this is a stack lock
1249 // allocated in this thread. For efficiency, we can focus on
1250 // recently allocated stack locks (instead of reading the stack
1251 // base and checking whether 'mark' points inside the current
1252 // thread stack):
1253 // 1) (mark & zero_bits) == 0, and
1254 // 2) rsp <= mark < mark + os::pagesize()
1255 //
1256 // Warning: rsp + os::pagesize can overflow the stack base. We must
1257 // neither apply the optimization for an inflated lock allocated
1258 // just above the thread stack (this is why condition 1 matters)
1259 // nor apply the optimization if the stack lock is inside the stack
1260 // of another thread. The latter is avoided even in case of overflow
1261 // because we have guard pages at the end of all stacks. Hence, if
1262 // we go over the stack base and hit the stack of another thread,
1263 // this should not be in a writeable area that could contain a
1264 // stack lock allocated by that thread. As a consequence, a stack
1265 // lock less than page size away from rsp is guaranteed to be
1266 // owned by the current thread.
1267 //
1268 // These 3 tests can be done by evaluating the following
1269 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1270 // assuming both stack pointer and pagesize have their
1271 // least significant bits clear.
1272 // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
1273 subptr(swap_reg, rsp);
1274 andptr(swap_reg, zero_bits - os::vm_page_size());
1275
1276 // Save the test result, for recursive case, the result is zero
1277 movptr(Address(lock_reg, mark_offset), swap_reg);
1278 jcc(Assembler::zero, done);
1279
1280 bind(slow_case);
1281
1282 // Call the runtime routine for slow case
1283 call_VM(noreg,
1284 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1285 lock_reg);
1286
1287 bind(done);
1288 }
1289 }
1290
1291
1292 // Unlocks an object. Used in monitorexit bytecode and
1293 // remove_activation. Throws an IllegalMonitorException if object is
1294 // not locked by current thread.
1295 //
1296 // Args:
1297 // rdx, c_rarg1: BasicObjectLock for lock
1298 //
1299 // Kills:
1300 // rax
1301 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1302 // rscratch1 (scratch reg)
1303 // rax, rbx, rcx, rdx
1304 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1305 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1306 "The argument is only for looks. It must be c_rarg1");
1307
1308 if (UseHeavyMonitors) {
1309 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1310 } else {
1311 Label done;
1312
1313 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1314 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1315 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1316
1317 save_bcp(); // Save in case of exception
1318
1319 // Convert from BasicObjectLock structure to object and BasicLock
1320 // structure Store the BasicLock address into %rax
1321 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
1322
1323 // Load oop into obj_reg(%c_rarg3)
1324 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
1325
1326 // Free entry
1327 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
1328
1329 // Load the old header from BasicLock structure
1330 movptr(header_reg, Address(swap_reg,
1331 BasicLock::displaced_header_offset_in_bytes()));
1332
1333 // Test for recursion
1334 testptr(header_reg, header_reg);
1335
1336 // zero for recursive case
1337 jcc(Assembler::zero, done);
1338
1339 // Atomic swap back the old header
1340 lock();
1341 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1342
1343 // zero for simple unlock of a stack-lock case
1344 jcc(Assembler::zero, done);
1345
1346
1347 // Call the runtime routine for slow case.
1348 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), obj_reg); // restore obj
1349 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1350
1351 bind(done);
1352
1353 restore_bcp();
1354 }
1355 }
1356
1357 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1358 Label& zero_continue) {
1359 assert(ProfileInterpreter, "must be profiling interpreter");
1360 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1361 testptr(mdp, mdp);
1362 jcc(Assembler::zero, zero_continue);
1363 }
1364
1365
1366 // Set the method data pointer for the current bcp.
1367 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1368 assert(ProfileInterpreter, "must be profiling interpreter");
1369 Label set_mdp;
1370 push(rax);
1371 push(rbx);
1372
1373 get_method(rbx);
1374 // Test MDO to avoid the call if it is NULL.
1375 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1376 testptr(rax, rax);
1377 jcc(Assembler::zero, set_mdp);
1378 // rbx: method
1379 // _bcp_register: bcp
1380 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1381 // rax: mdi
1382 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1383 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1384 addptr(rbx, in_bytes(MethodData::data_offset()));
1385 addptr(rax, rbx);
1386 bind(set_mdp);
1387 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1388 pop(rbx);
1389 pop(rax);
1390 }
1391
1392 void InterpreterMacroAssembler::verify_method_data_pointer() {
1393 assert(ProfileInterpreter, "must be profiling interpreter");
1394 #ifdef ASSERT
1395 Label verify_continue;
1396 push(rax);
1397 push(rbx);
1398 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1399 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1400 push(arg3_reg);
1401 push(arg2_reg);
1402 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1403 get_method(rbx);
1404
1405 // If the mdp is valid, it will point to a DataLayout header which is
1406 // consistent with the bcp. The converse is highly probable also.
1407 load_unsigned_short(arg2_reg,
1408 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1409 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1410 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1411 cmpptr(arg2_reg, _bcp_register);
1412 jcc(Assembler::equal, verify_continue);
1413 // rbx: method
1414 // _bcp_register: bcp
1415 // c_rarg3: mdp
1416 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1417 rbx, _bcp_register, arg3_reg);
1418 bind(verify_continue);
1419 pop(arg2_reg);
1420 pop(arg3_reg);
1421 pop(rbx);
1422 pop(rax);
1423 #endif // ASSERT
1424 }
1425
1426
1427 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1428 int constant,
1429 Register value) {
1430 assert(ProfileInterpreter, "must be profiling interpreter");
1431 Address data(mdp_in, constant);
1432 movptr(data, value);
1433 }
1434
1435
1436 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1437 int constant,
1438 bool decrement) {
1439 // Counter address
1440 Address data(mdp_in, constant);
1441
1442 increment_mdp_data_at(data, decrement);
1443 }
1444
1445 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1446 bool decrement) {
1447 assert(ProfileInterpreter, "must be profiling interpreter");
1448 // %%% this does 64bit counters at best it is wasting space
1449 // at worst it is a rare bug when counters overflow
1450
1451 if (decrement) {
1452 // Decrement the register. Set condition codes.
1453 addptr(data, (int32_t) -DataLayout::counter_increment);
1454 // If the decrement causes the counter to overflow, stay negative
1455 Label L;
1456 jcc(Assembler::negative, L);
1457 addptr(data, (int32_t) DataLayout::counter_increment);
1458 bind(L);
1459 } else {
1460 assert(DataLayout::counter_increment == 1,
1461 "flow-free idiom only works with 1");
1462 // Increment the register. Set carry flag.
1463 addptr(data, DataLayout::counter_increment);
1464 // If the increment causes the counter to overflow, pull back by 1.
1465 sbbptr(data, (int32_t)0);
1466 }
1467 }
1468
1469
1470 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1471 Register reg,
1472 int constant,
1473 bool decrement) {
1474 Address data(mdp_in, reg, Address::times_1, constant);
1475
1476 increment_mdp_data_at(data, decrement);
1477 }
1478
1479 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1480 int flag_byte_constant) {
1481 assert(ProfileInterpreter, "must be profiling interpreter");
1482 int header_offset = in_bytes(DataLayout::flags_offset());
1483 int header_bits = flag_byte_constant;
1484 // Set the flag
1485 orb(Address(mdp_in, header_offset), header_bits);
1486 }
1487
1488
1489
1490 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1491 int offset,
1492 Register value,
1493 Register test_value_out,
1494 Label& not_equal_continue) {
1495 assert(ProfileInterpreter, "must be profiling interpreter");
1496 if (test_value_out == noreg) {
1497 cmpptr(value, Address(mdp_in, offset));
1498 } else {
1499 // Put the test value into a register, so caller can use it:
1500 movptr(test_value_out, Address(mdp_in, offset));
1501 cmpptr(test_value_out, value);
1502 }
1503 jcc(Assembler::notEqual, not_equal_continue);
1504 }
1505
1506
1507 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1508 int offset_of_disp) {
1509 assert(ProfileInterpreter, "must be profiling interpreter");
1510 Address disp_address(mdp_in, offset_of_disp);
1511 addptr(mdp_in, disp_address);
1512 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1513 }
1514
1515
1516 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1517 Register reg,
1518 int offset_of_disp) {
1519 assert(ProfileInterpreter, "must be profiling interpreter");
1520 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1521 addptr(mdp_in, disp_address);
1522 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1523 }
1524
1525
1526 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1527 int constant) {
1528 assert(ProfileInterpreter, "must be profiling interpreter");
1529 addptr(mdp_in, constant);
1530 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1531 }
1532
1533
1534 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1535 assert(ProfileInterpreter, "must be profiling interpreter");
1536 push(return_bci); // save/restore across call_VM
1537 call_VM(noreg,
1538 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1539 return_bci);
1540 pop(return_bci);
1541 }
1542
1543
1544 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1545 Register bumped_count) {
1546 if (ProfileInterpreter) {
1547 Label profile_continue;
1548
1549 // If no method data exists, go to profile_continue.
1550 // Otherwise, assign to mdp
1551 test_method_data_pointer(mdp, profile_continue);
1552
1553 // We are taking a branch. Increment the taken count.
1554 // We inline increment_mdp_data_at to return bumped_count in a register
1555 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1556 Address data(mdp, in_bytes(JumpData::taken_offset()));
1557 movptr(bumped_count, data);
1558 assert(DataLayout::counter_increment == 1,
1559 "flow-free idiom only works with 1");
1560 addptr(bumped_count, DataLayout::counter_increment);
1561 sbbptr(bumped_count, 0);
1562 movptr(data, bumped_count); // Store back out
1563
1564 // The method data pointer needs to be updated to reflect the new target.
1565 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1566 bind(profile_continue);
1567 }
1568 }
1569
1570
1571 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1572 if (ProfileInterpreter) {
1573 Label profile_continue;
1574
1575 // If no method data exists, go to profile_continue.
1576 test_method_data_pointer(mdp, profile_continue);
1577
1578 // We are taking a branch. Increment the not taken count.
1579 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1580
1581 // The method data pointer needs to be updated to correspond to
1582 // the next bytecode
1583 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1584 bind(profile_continue);
1585 }
1586 }
1587
1588 void InterpreterMacroAssembler::profile_call(Register mdp) {
1589 if (ProfileInterpreter) {
1590 Label profile_continue;
1591
1592 // If no method data exists, go to profile_continue.
1593 test_method_data_pointer(mdp, profile_continue);
1594
1595 // We are making a call. Increment the count.
1596 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1597
1598 // The method data pointer needs to be updated to reflect the new target.
1599 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1600 bind(profile_continue);
1601 }
1602 }
1603
1604
1605 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1606 if (ProfileInterpreter) {
1607 Label profile_continue;
1608
1609 // If no method data exists, go to profile_continue.
1610 test_method_data_pointer(mdp, profile_continue);
1611
1612 // We are making a call. Increment the count.
1613 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1614
1615 // The method data pointer needs to be updated to reflect the new target.
1616 update_mdp_by_constant(mdp,
1617 in_bytes(VirtualCallData::
1618 virtual_call_data_size()));
1619 bind(profile_continue);
1620 }
1621 }
1622
1623
1624 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1625 Register mdp,
1626 Register reg2,
1627 bool receiver_can_be_null) {
1628 if (ProfileInterpreter) {
1629 Label profile_continue;
1630
1631 // If no method data exists, go to profile_continue.
1632 test_method_data_pointer(mdp, profile_continue);
1633
1634 Label skip_receiver_profile;
1635 if (receiver_can_be_null) {
1636 Label not_null;
1637 testptr(receiver, receiver);
1638 jccb(Assembler::notZero, not_null);
1639 // We are making a call. Increment the count for null receiver.
1640 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1641 jmp(skip_receiver_profile);
1642 bind(not_null);
1643 }
1644
1645 // Record the receiver type.
1646 record_klass_in_profile(receiver, mdp, reg2, true);
1647 bind(skip_receiver_profile);
1648
1649 // The method data pointer needs to be updated to reflect the new target.
1650 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1651 bind(profile_continue);
1652 }
1653 }
1654
1655 // This routine creates a state machine for updating the multi-row
1656 // type profile at a virtual call site (or other type-sensitive bytecode).
1657 // The machine visits each row (of receiver/count) until the receiver type
1658 // is found, or until it runs out of rows. At the same time, it remembers
1659 // the location of the first empty row. (An empty row records null for its
1660 // receiver, and can be allocated for a newly-observed receiver type.)
1661 // Because there are two degrees of freedom in the state, a simple linear
1662 // search will not work; it must be a decision tree. Hence this helper
1663 // function is recursive, to generate the required tree structured code.
1664 // It's the interpreter, so we are trading off code space for speed.
1665 // See below for example code.
1666 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1667 Register receiver, Register mdp,
1668 Register reg2, int start_row,
1669 Label& done, bool is_virtual_call) {
1670 if (TypeProfileWidth == 0) {
1671 if (is_virtual_call) {
1672 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1673 }
1674 #if INCLUDE_JVMCI
1675 else if (EnableJVMCI) {
1676 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1677 }
1678 #endif // INCLUDE_JVMCI
1679 } else {
1680 int non_profiled_offset = -1;
1681 if (is_virtual_call) {
1682 non_profiled_offset = in_bytes(CounterData::count_offset());
1683 }
1684 #if INCLUDE_JVMCI
1685 else if (EnableJVMCI) {
1686 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1687 }
1688 #endif // INCLUDE_JVMCI
1689
1690 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1691 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1692 }
1693 }
1694
1695 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1696 Register reg2, int start_row, Label& done, int total_rows,
1697 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1698 int non_profiled_offset) {
1699 int last_row = total_rows - 1;
1700 assert(start_row <= last_row, "must be work left to do");
1701 // Test this row for both the item and for null.
1702 // Take any of three different outcomes:
1703 // 1. found item => increment count and goto done
1704 // 2. found null => keep looking for case 1, maybe allocate this cell
1705 // 3. found something else => keep looking for cases 1 and 2
1706 // Case 3 is handled by a recursive call.
1707 for (int row = start_row; row <= last_row; row++) {
1708 Label next_test;
1709 bool test_for_null_also = (row == start_row);
1710
1711 // See if the item is item[n].
1712 int item_offset = in_bytes(item_offset_fn(row));
1713 test_mdp_data_at(mdp, item_offset, item,
1714 (test_for_null_also ? reg2 : noreg),
1715 next_test);
1716 // (Reg2 now contains the item from the CallData.)
1717
1718 // The item is item[n]. Increment count[n].
1719 int count_offset = in_bytes(item_count_offset_fn(row));
1720 increment_mdp_data_at(mdp, count_offset);
1721 jmp(done);
1722 bind(next_test);
1723
1724 if (test_for_null_also) {
1725 // Failed the equality check on item[n]... Test for null.
1726 testptr(reg2, reg2);
1727 if (start_row == last_row) {
1728 // The only thing left to do is handle the null case.
1729 if (non_profiled_offset >= 0) {
1730 Label found_null;
1731 jccb(Assembler::zero, found_null);
1732 // Item did not match any saved item and there is no empty row for it.
1733 // Increment total counter to indicate polymorphic case.
1734 increment_mdp_data_at(mdp, non_profiled_offset);
1735 jmp(done);
1736 bind(found_null);
1737 } else {
1738 jcc(Assembler::notZero, done);
1739 }
1740 break;
1741 }
1742 Label found_null;
1743 // Since null is rare, make it be the branch-taken case.
1744 jcc(Assembler::zero, found_null);
1745
1746 // Put all the "Case 3" tests here.
1747 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1748 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1749
1750 // Found a null. Keep searching for a matching item,
1751 // but remember that this is an empty (unused) slot.
1752 bind(found_null);
1753 }
1754 }
1755
1756 // In the fall-through case, we found no matching item, but we
1757 // observed the item[start_row] is NULL.
1758
1759 // Fill in the item field and increment the count.
1760 int item_offset = in_bytes(item_offset_fn(start_row));
1761 set_mdp_data_at(mdp, item_offset, item);
1762 int count_offset = in_bytes(item_count_offset_fn(start_row));
1763 movl(reg2, DataLayout::counter_increment);
1764 set_mdp_data_at(mdp, count_offset, reg2);
1765 if (start_row > 0) {
1766 jmp(done);
1767 }
1768 }
1769
1770 // Example state machine code for three profile rows:
1771 // // main copy of decision tree, rooted at row[1]
1772 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1773 // if (row[0].rec != NULL) {
1774 // // inner copy of decision tree, rooted at row[1]
1775 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1776 // if (row[1].rec != NULL) {
1777 // // degenerate decision tree, rooted at row[2]
1778 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1779 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1780 // row[2].init(rec); goto done;
1781 // } else {
1782 // // remember row[1] is empty
1783 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1784 // row[1].init(rec); goto done;
1785 // }
1786 // } else {
1787 // // remember row[0] is empty
1788 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1789 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1790 // row[0].init(rec); goto done;
1791 // }
1792 // done:
1793
1794 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1795 Register mdp, Register reg2,
1796 bool is_virtual_call) {
1797 assert(ProfileInterpreter, "must be profiling");
1798 Label done;
1799
1800 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1801
1802 bind (done);
1803 }
1804
1805 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1806 Register mdp) {
1807 if (ProfileInterpreter) {
1808 Label profile_continue;
1809 uint row;
1810
1811 // If no method data exists, go to profile_continue.
1812 test_method_data_pointer(mdp, profile_continue);
1813
1814 // Update the total ret count.
1815 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1816
1817 for (row = 0; row < RetData::row_limit(); row++) {
1818 Label next_test;
1819
1820 // See if return_bci is equal to bci[n]:
1821 test_mdp_data_at(mdp,
1822 in_bytes(RetData::bci_offset(row)),
1823 return_bci, noreg,
1824 next_test);
1825
1826 // return_bci is equal to bci[n]. Increment the count.
1827 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1828
1829 // The method data pointer needs to be updated to reflect the new target.
1830 update_mdp_by_offset(mdp,
1831 in_bytes(RetData::bci_displacement_offset(row)));
1832 jmp(profile_continue);
1833 bind(next_test);
1834 }
1835
1836 update_mdp_for_ret(return_bci);
1837
1838 bind(profile_continue);
1839 }
1840 }
1841
1842
1843 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1844 if (ProfileInterpreter) {
1845 Label profile_continue;
1846
1847 // If no method data exists, go to profile_continue.
1848 test_method_data_pointer(mdp, profile_continue);
1849
1850 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1851
1852 // The method data pointer needs to be updated.
1853 int mdp_delta = in_bytes(BitData::bit_data_size());
1854 if (TypeProfileCasts) {
1855 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1856 }
1857 update_mdp_by_constant(mdp, mdp_delta);
1858
1859 bind(profile_continue);
1860 }
1861 }
1862
1863
1864 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1865 if (ProfileInterpreter && TypeProfileCasts) {
1866 Label profile_continue;
1867
1868 // If no method data exists, go to profile_continue.
1869 test_method_data_pointer(mdp, profile_continue);
1870
1871 int count_offset = in_bytes(CounterData::count_offset());
1872 // Back up the address, since we have already bumped the mdp.
1873 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1874
1875 // *Decrement* the counter. We expect to see zero or small negatives.
1876 increment_mdp_data_at(mdp, count_offset, true);
1877
1878 bind (profile_continue);
1879 }
1880 }
1881
1882
1883 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1884 if (ProfileInterpreter) {
1885 Label profile_continue;
1886
1887 // If no method data exists, go to profile_continue.
1888 test_method_data_pointer(mdp, profile_continue);
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 // Record the object type.
1896 record_klass_in_profile(klass, mdp, reg2, false);
1897 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1898 NOT_LP64(restore_locals();) // Restore EDI
1899 }
1900 update_mdp_by_constant(mdp, mdp_delta);
1901
1902 bind(profile_continue);
1903 }
1904 }
1905
1906
1907 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1908 if (ProfileInterpreter) {
1909 Label profile_continue;
1910
1911 // If no method data exists, go to profile_continue.
1912 test_method_data_pointer(mdp, profile_continue);
1913
1914 // Update the default case count
1915 increment_mdp_data_at(mdp,
1916 in_bytes(MultiBranchData::default_count_offset()));
1917
1918 // The method data pointer needs to be updated.
1919 update_mdp_by_offset(mdp,
1920 in_bytes(MultiBranchData::
1921 default_displacement_offset()));
1922
1923 bind(profile_continue);
1924 }
1925 }
1926
1927
1928 void InterpreterMacroAssembler::profile_switch_case(Register index,
1929 Register mdp,
1930 Register reg2) {
1931 if (ProfileInterpreter) {
1932 Label profile_continue;
1933
1934 // If no method data exists, go to profile_continue.
1935 test_method_data_pointer(mdp, profile_continue);
1936
1937 // Build the base (index * per_case_size_in_bytes()) +
1938 // case_array_offset_in_bytes()
1939 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1940 imulptr(index, reg2); // XXX l ?
1941 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1942
1943 // Update the case count
1944 increment_mdp_data_at(mdp,
1945 index,
1946 in_bytes(MultiBranchData::relative_count_offset()));
1947
1948 // The method data pointer needs to be updated.
1949 update_mdp_by_offset(mdp,
1950 index,
1951 in_bytes(MultiBranchData::
1952 relative_displacement_offset()));
1953
1954 bind(profile_continue);
1955 }
1956 }
1957
1958
1959
1960 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1961 if (state == atos) {
1962 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1963 }
1964 }
1965
1966 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1967 #ifndef _LP64
1968 if ((state == ftos && UseSSE < 1) ||
1969 (state == dtos && UseSSE < 2)) {
1970 MacroAssembler::verify_FPU(stack_depth);
1971 }
1972 #endif
1973 }
1974
1975 // Jump if ((*counter_addr += increment) & mask) == 0
1976 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
1977 Register scratch, Label* where) {
1978 // This update is actually not atomic and can lose a number of updates
1979 // under heavy contention, but the alternative of using the (contended)
1980 // atomic update here penalizes profiling paths too much.
1981 movl(scratch, counter_addr);
1982 incrementl(scratch, InvocationCounter::count_increment);
1983 movl(counter_addr, scratch);
1984 andl(scratch, mask);
1985 if (where != NULL) {
1986 jcc(Assembler::zero, *where);
1987 }
1988 }
1989
1990 void InterpreterMacroAssembler::notify_method_entry() {
1991 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1992 // track stack depth. If it is possible to enter interp_only_mode we add
1993 // the code to check if the event should be sent.
1994 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1995 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1996 if (JvmtiExport::can_post_interpreter_events()) {
1997 Label L;
1998 NOT_LP64(get_thread(rthread);)
1999 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2000 testl(rdx, rdx);
2001 jcc(Assembler::zero, L);
2002 call_VM(noreg, CAST_FROM_FN_PTR(address,
2003 InterpreterRuntime::post_method_entry));
2004 bind(L);
2005 }
2006
2007 {
2008 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2009 NOT_LP64(get_thread(rthread);)
2010 get_method(rarg);
2011 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2012 rthread, rarg);
2013 }
2014
2015 // RedefineClasses() tracing support for obsolete method entry
2016 if (log_is_enabled(Trace, redefine, class, obsolete)) {
2017 NOT_LP64(get_thread(rthread);)
2018 get_method(rarg);
2019 call_VM_leaf(
2020 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2021 rthread, rarg);
2022 }
2023 }
2024
2025
2026 void InterpreterMacroAssembler::notify_method_exit(
2027 TosState state, NotifyMethodExitMode mode) {
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 (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2034 Label L;
2035 // Note: frame::interpreter_frame_result has a dependency on how the
2036 // method result is saved across the call to post_method_exit. If this
2037 // is changed then the interpreter_frame_result implementation will
2038 // need to be updated too.
2039
2040 // template interpreter will leave the result on the top of the stack.
2041 push(state);
2042 NOT_LP64(get_thread(rthread);)
2043 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2044 testl(rdx, rdx);
2045 jcc(Assembler::zero, L);
2046 call_VM(noreg,
2047 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2048 bind(L);
2049 pop(state);
2050 }
2051
2052 {
2053 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2054 push(state);
2055 NOT_LP64(get_thread(rthread);)
2056 get_method(rarg);
2057 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2058 rthread, rarg);
2059 pop(state);
2060 }
2061 }