1 /*
2 * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "interp_masm_x86.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "logging/log.hpp"
30 #include "oops/arrayOop.hpp"
31 #include "oops/markOop.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/method.hpp"
34 #include "prims/jvmtiExport.hpp"
35 #include "prims/jvmtiThreadState.hpp"
36 #include "runtime/basicLock.hpp"
37 #include "runtime/biasedLocking.hpp"
38 #include "runtime/frame.inline.hpp"
39 #include "runtime/safepointMechanism.hpp"
40 #include "runtime/sharedRuntime.hpp"
41 #include "runtime/thread.inline.hpp"
42
43 // Implementation of InterpreterMacroAssembler
44
45 void InterpreterMacroAssembler::jump_to_entry(address entry) {
46 assert(entry, "Entry must have been generated by now");
47 jump(RuntimeAddress(entry));
48 }
49
50 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
51 Label update, next, none;
52
53 verify_oop(obj);
54
55 testptr(obj, obj);
56 jccb(Assembler::notZero, update);
57 orptr(mdo_addr, TypeEntries::null_seen);
58 jmpb(next);
59
60 bind(update);
61 load_klass(obj, obj);
62
63 xorptr(obj, mdo_addr);
64 testptr(obj, TypeEntries::type_klass_mask);
65 jccb(Assembler::zero, next); // klass seen before, nothing to
66 // do. The unknown bit may have been
67 // set already but no need to check.
68
69 testptr(obj, TypeEntries::type_unknown);
70 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
71
72 cmpptr(mdo_addr, 0);
73 jccb(Assembler::equal, none);
74 cmpptr(mdo_addr, TypeEntries::null_seen);
75 jccb(Assembler::equal, none);
76 // There is a chance that the checks above (re-reading profiling
77 // data from memory) fail if another thread has just set the
78 // profiling to this obj's klass
79 xorptr(obj, mdo_addr);
80 testptr(obj, TypeEntries::type_klass_mask);
81 jccb(Assembler::zero, next);
82
83 // different than before. Cannot keep accurate profile.
84 orptr(mdo_addr, TypeEntries::type_unknown);
85 jmpb(next);
86
87 bind(none);
88 // first time here. Set profile type.
89 movptr(mdo_addr, obj);
90
91 bind(next);
92 }
93
94 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
95 if (!ProfileInterpreter) {
96 return;
97 }
98
99 if (MethodData::profile_arguments() || MethodData::profile_return()) {
100 Label profile_continue;
101
102 test_method_data_pointer(mdp, profile_continue);
103
104 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
105
106 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
107 jcc(Assembler::notEqual, profile_continue);
108
109 if (MethodData::profile_arguments()) {
110 Label done;
111 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
112 addptr(mdp, off_to_args);
113
114 for (int i = 0; i < TypeProfileArgsLimit; i++) {
115 if (i > 0 || MethodData::profile_return()) {
116 // If return value type is profiled we may have no argument to profile
117 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
118 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
119 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
120 jcc(Assembler::less, done);
121 }
122 movptr(tmp, Address(callee, Method::const_offset()));
123 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
124 // stack offset o (zero based) from the start of the argument
125 // list, for n arguments translates into offset n - o - 1 from
126 // the end of the argument list
127 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
128 subl(tmp, 1);
129 Address arg_addr = argument_address(tmp);
130 movptr(tmp, arg_addr);
131
132 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
133 profile_obj_type(tmp, mdo_arg_addr);
134
135 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
136 addptr(mdp, to_add);
137 off_to_args += to_add;
138 }
139
140 if (MethodData::profile_return()) {
141 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
142 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
143 }
144
145 bind(done);
146
147 if (MethodData::profile_return()) {
148 // We're right after the type profile for the last
149 // argument. tmp is the number of cells left in the
150 // CallTypeData/VirtualCallTypeData to reach its end. Non null
151 // if there's a return to profile.
152 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
153 shll(tmp, exact_log2(DataLayout::cell_size));
154 addptr(mdp, tmp);
155 }
156 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
157 } else {
158 assert(MethodData::profile_return(), "either profile call args or call ret");
159 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
160 }
161
162 // mdp points right after the end of the
163 // CallTypeData/VirtualCallTypeData, right after the cells for the
164 // return value type if there's one
165
166 bind(profile_continue);
167 }
168 }
169
170 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
171 assert_different_registers(mdp, ret, tmp, _bcp_register);
172 if (ProfileInterpreter && MethodData::profile_return()) {
173 Label profile_continue, done;
174
175 test_method_data_pointer(mdp, profile_continue);
176
177 if (MethodData::profile_return_jsr292_only()) {
178 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
179
180 // If we don't profile all invoke bytecodes we must make sure
181 // it's a bytecode we indeed profile. We can't go back to the
182 // begining of the ProfileData we intend to update to check its
183 // type because we're right after it and we don't known its
184 // length
185 Label do_profile;
186 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
187 jcc(Assembler::equal, do_profile);
188 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
189 jcc(Assembler::equal, do_profile);
190 get_method(tmp);
191 cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm);
192 jcc(Assembler::notEqual, profile_continue);
193
194 bind(do_profile);
195 }
196
197 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
198 mov(tmp, ret);
199 profile_obj_type(tmp, mdo_ret_addr);
200
201 bind(profile_continue);
202 }
203 }
204
205 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
206 if (ProfileInterpreter && MethodData::profile_parameters()) {
207 Label profile_continue, done;
208
209 test_method_data_pointer(mdp, profile_continue);
210
211 // Load the offset of the area within the MDO used for
212 // parameters. If it's negative we're not profiling any parameters
213 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
214 testl(tmp1, tmp1);
215 jcc(Assembler::negative, profile_continue);
216
217 // Compute a pointer to the area for parameters from the offset
218 // and move the pointer to the slot for the last
219 // parameters. Collect profiling from last parameter down.
220 // mdo start + parameters offset + array length - 1
221 addptr(mdp, tmp1);
222 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
223 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
224
225 Label loop;
226 bind(loop);
227
228 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
229 int type_base = in_bytes(ParametersTypeData::type_offset(0));
230 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
231 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
232 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
233
234 // load offset on the stack from the slot for this parameter
235 movptr(tmp2, arg_off);
236 negptr(tmp2);
237 // read the parameter from the local area
238 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
239
240 // profile the parameter
241 profile_obj_type(tmp2, arg_type);
242
243 // go to next parameter
244 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
245 jcc(Assembler::positive, loop);
246
247 bind(profile_continue);
248 }
249 }
250
251 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
252 int number_of_arguments) {
253 // interpreter specific
254 //
255 // Note: No need to save/restore bcp & locals registers
256 // since these are callee saved registers and no blocking/
257 // GC can happen in leaf calls.
258 // Further Note: DO NOT save/restore bcp/locals. If a caller has
259 // already saved them so that it can use rsi/rdi as temporaries
260 // then a save/restore here will DESTROY the copy the caller
261 // saved! There used to be a save_bcp() that only happened in
262 // the ASSERT path (no restore_bcp). Which caused bizarre failures
263 // when jvm built with ASSERTs.
264 #ifdef ASSERT
265 {
266 Label L;
267 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
268 jcc(Assembler::equal, L);
269 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
270 " last_sp != NULL");
271 bind(L);
272 }
273 #endif
274 // super call
275 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
276 // interpreter specific
277 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
278 // but since they may not have been saved (and we don't want to
279 // save them here (see note above) the assert is invalid.
280 }
281
282 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
283 Register java_thread,
284 Register last_java_sp,
285 address entry_point,
286 int number_of_arguments,
287 bool check_exceptions) {
288 // interpreter specific
289 //
290 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
291 // really make a difference for these runtime calls, since they are
292 // slow anyway. Btw., bcp must be saved/restored since it may change
293 // due to GC.
294 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
295 save_bcp();
296 #ifdef ASSERT
297 {
298 Label L;
299 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
300 jcc(Assembler::equal, L);
301 stop("InterpreterMacroAssembler::call_VM_base:"
302 " last_sp != NULL");
303 bind(L);
304 }
305 #endif /* ASSERT */
306 // super call
307 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
308 entry_point, number_of_arguments,
309 check_exceptions);
310 // interpreter specific
311 restore_bcp();
312 restore_locals();
313 }
314
315 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
316 if (JvmtiExport::can_pop_frame()) {
317 Label L;
318 // Initiate popframe handling only if it is not already being
319 // processed. If the flag has the popframe_processing bit set, it
320 // means that this code is called *during* popframe handling - we
321 // don't want to reenter.
322 // This method is only called just after the call into the vm in
323 // call_VM_base, so the arg registers are available.
324 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
325 LP64_ONLY(c_rarg0);
326 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
327 testl(pop_cond, JavaThread::popframe_pending_bit);
328 jcc(Assembler::zero, L);
329 testl(pop_cond, JavaThread::popframe_processing_bit);
330 jcc(Assembler::notZero, L);
331 // Call Interpreter::remove_activation_preserving_args_entry() to get the
332 // address of the same-named entrypoint in the generated interpreter code.
333 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
334 jmp(rax);
335 bind(L);
336 NOT_LP64(get_thread(java_thread);)
337 }
338 }
339
340 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
341 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
342 NOT_LP64(get_thread(thread);)
343 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
344 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
345 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
346 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
347 #ifdef _LP64
348 switch (state) {
349 case atos: movptr(rax, oop_addr);
350 movptr(oop_addr, (int32_t)NULL_WORD);
351 verify_oop(rax, state); break;
352 case ltos: movptr(rax, val_addr); break;
353 case btos: // fall through
354 case ztos: // fall through
355 case ctos: // fall through
356 case stos: // fall through
357 case itos: movl(rax, val_addr); break;
358 case ftos: load_float(val_addr); break;
359 case dtos: load_double(val_addr); break;
360 case vtos: /* nothing to do */ break;
361 default : ShouldNotReachHere();
362 }
363 // Clean up tos value in the thread object
364 movl(tos_addr, (int) ilgl);
365 movl(val_addr, (int32_t) NULL_WORD);
366 #else
367 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
368 + in_ByteSize(wordSize));
369 switch (state) {
370 case atos: movptr(rax, oop_addr);
371 movptr(oop_addr, NULL_WORD);
372 verify_oop(rax, state); break;
373 case ltos:
374 movl(rdx, val_addr1); // fall through
375 case btos: // fall through
376 case ztos: // fall through
377 case ctos: // fall through
378 case stos: // fall through
379 case itos: movl(rax, val_addr); break;
380 case ftos: load_float(val_addr); break;
381 case dtos: load_double(val_addr); break;
382 case vtos: /* nothing to do */ break;
383 default : ShouldNotReachHere();
384 }
385 #endif // _LP64
386 // Clean up tos value in the thread object
387 movl(tos_addr, (int32_t) ilgl);
388 movptr(val_addr, NULL_WORD);
389 NOT_LP64(movptr(val_addr1, NULL_WORD);)
390 }
391
392
393 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
394 if (JvmtiExport::can_force_early_return()) {
395 Label L;
396 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
397 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
398
399 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
400 testptr(tmp, tmp);
401 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
402
403 // Initiate earlyret handling only if it is not already being processed.
404 // If the flag has the earlyret_processing bit set, it means that this code
405 // is called *during* earlyret handling - we don't want to reenter.
406 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
407 cmpl(tmp, JvmtiThreadState::earlyret_pending);
408 jcc(Assembler::notEqual, L);
409
410 // Call Interpreter::remove_activation_early_entry() to get the address of the
411 // same-named entrypoint in the generated interpreter code.
412 NOT_LP64(get_thread(java_thread);)
413 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
414 #ifdef _LP64
415 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
416 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
417 #else
418 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
419 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
420 #endif // _LP64
421 jmp(rax);
422 bind(L);
423 NOT_LP64(get_thread(java_thread);)
424 }
425 }
426
427 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
428 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
429 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
430 bswapl(reg);
431 shrl(reg, 16);
432 }
433
434 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
435 int bcp_offset,
436 size_t index_size) {
437 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
438 if (index_size == sizeof(u2)) {
439 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
440 } else if (index_size == sizeof(u4)) {
441 movl(index, Address(_bcp_register, bcp_offset));
442 // Check if the secondary index definition is still ~x, otherwise
443 // we have to change the following assembler code to calculate the
444 // plain index.
445 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
446 notl(index); // convert to plain index
447 } else if (index_size == sizeof(u1)) {
448 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
449 } else {
450 ShouldNotReachHere();
451 }
452 }
453
454 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
455 Register index,
456 int bcp_offset,
457 size_t index_size) {
458 assert_different_registers(cache, index);
459 get_cache_index_at_bcp(index, bcp_offset, index_size);
460 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
461 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
462 // convert from field index to ConstantPoolCacheEntry index
463 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
464 shll(index, 2);
465 }
466
467 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
468 Register index,
469 Register bytecode,
470 int byte_no,
471 int bcp_offset,
472 size_t index_size) {
473 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
474 // We use a 32-bit load here since the layout of 64-bit words on
475 // little-endian machines allow us that.
476 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
477 const int shift_count = (1 + byte_no) * BitsPerByte;
478 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
479 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
480 "correct shift count");
481 shrl(bytecode, shift_count);
482 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
483 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
484 }
485
486 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
487 Register tmp,
488 int bcp_offset,
489 size_t index_size) {
490 assert(cache != tmp, "must use different register");
491 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
492 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
493 // convert from field index to ConstantPoolCacheEntry index
494 // and from word offset to byte offset
495 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
496 shll(tmp, 2 + LogBytesPerWord);
497 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
498 // skip past the header
499 addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
500 addptr(cache, tmp); // construct pointer to cache entry
501 }
502
503 // Load object from cpool->resolved_references(index)
504 void InterpreterMacroAssembler::load_resolved_reference_at_index(
505 Register result, Register index) {
506 assert_different_registers(result, index);
507 // convert from field index to resolved_references() index and from
508 // word index to byte offset. Since this is a java object, it can be compressed
509 Register tmp = index; // reuse
510 shll(tmp, LogBytesPerHeapOop);
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);
517 // Add in the index
518 addptr(result, tmp);
519 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
520 // The resulting oop is null if the reference is not yet resolved.
521 // It is Universe::the_null_sentinel() if the reference resolved to NULL via condy.
522 }
523
524 // load cpool->resolved_klass_at(index)
525 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register cpool,
526 Register index, Register klass) {
527 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
528 Register resolved_klasses = cpool;
529 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes()));
530 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
531 }
532
533 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
534 // subtype of super_klass.
535 //
536 // Args:
537 // rax: superklass
538 // Rsub_klass: subklass
539 //
540 // Kills:
541 // rcx, rdi
542 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
543 Label& ok_is_subtype) {
544 assert(Rsub_klass != rax, "rax holds superklass");
545 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
546 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
547 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
548 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
549
550 // Profile the not-null value's klass.
551 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
552
553 // Do the check.
554 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
555
556 // Profile the failure of the check.
557 profile_typecheck_failed(rcx); // blows rcx
558 }
559
560
561 #ifndef _LP64
562 void InterpreterMacroAssembler::f2ieee() {
563 if (IEEEPrecision) {
564 fstp_s(Address(rsp, 0));
565 fld_s(Address(rsp, 0));
566 }
567 }
568
569
570 void InterpreterMacroAssembler::d2ieee() {
571 if (IEEEPrecision) {
572 fstp_d(Address(rsp, 0));
573 fld_d(Address(rsp, 0));
574 }
575 }
576 #endif // _LP64
577
578 // Java Expression Stack
579
580 void InterpreterMacroAssembler::pop_ptr(Register r) {
581 pop(r);
582 }
583
584 void InterpreterMacroAssembler::push_ptr(Register r) {
585 push(r);
586 }
587
588 void InterpreterMacroAssembler::push_i(Register r) {
589 push(r);
590 }
591
592 void InterpreterMacroAssembler::push_f(XMMRegister r) {
593 subptr(rsp, wordSize);
594 movflt(Address(rsp, 0), r);
595 }
596
597 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
598 movflt(r, Address(rsp, 0));
599 addptr(rsp, wordSize);
600 }
601
602 void InterpreterMacroAssembler::push_d(XMMRegister r) {
603 subptr(rsp, 2 * wordSize);
604 movdbl(Address(rsp, 0), r);
605 }
606
607 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
608 movdbl(r, Address(rsp, 0));
609 addptr(rsp, 2 * Interpreter::stackElementSize);
610 }
611
612 #ifdef _LP64
613 void InterpreterMacroAssembler::pop_i(Register r) {
614 // XXX can't use pop currently, upper half non clean
615 movl(r, Address(rsp, 0));
616 addptr(rsp, wordSize);
617 }
618
619 void InterpreterMacroAssembler::pop_l(Register r) {
620 movq(r, Address(rsp, 0));
621 addptr(rsp, 2 * Interpreter::stackElementSize);
622 }
623
624 void InterpreterMacroAssembler::push_l(Register r) {
625 subptr(rsp, 2 * wordSize);
626 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
627 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
628 }
629
630 void InterpreterMacroAssembler::pop(TosState state) {
631 switch (state) {
632 case atos: pop_ptr(); break;
633 case btos:
634 case ztos:
635 case ctos:
636 case stos:
637 case itos: pop_i(); break;
638 case ltos: pop_l(); break;
639 case ftos: pop_f(xmm0); break;
640 case dtos: pop_d(xmm0); break;
641 case vtos: /* nothing to do */ break;
642 default: ShouldNotReachHere();
643 }
644 verify_oop(rax, state);
645 }
646
647 void InterpreterMacroAssembler::push(TosState state) {
648 verify_oop(rax, state);
649 switch (state) {
650 case atos: push_ptr(); break;
651 case btos:
652 case ztos:
653 case ctos:
654 case stos:
655 case itos: push_i(); break;
656 case ltos: push_l(); break;
657 case ftos: push_f(xmm0); break;
658 case dtos: push_d(xmm0); break;
659 case vtos: /* nothing to do */ break;
660 default : ShouldNotReachHere();
661 }
662 }
663 #else
664 void InterpreterMacroAssembler::pop_i(Register r) {
665 pop(r);
666 }
667
668 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
669 pop(lo);
670 pop(hi);
671 }
672
673 void InterpreterMacroAssembler::pop_f() {
674 fld_s(Address(rsp, 0));
675 addptr(rsp, 1 * wordSize);
676 }
677
678 void InterpreterMacroAssembler::pop_d() {
679 fld_d(Address(rsp, 0));
680 addptr(rsp, 2 * wordSize);
681 }
682
683
684 void InterpreterMacroAssembler::pop(TosState state) {
685 switch (state) {
686 case atos: pop_ptr(rax); break;
687 case btos: // fall through
688 case ztos: // fall through
689 case ctos: // fall through
690 case stos: // fall through
691 case itos: pop_i(rax); break;
692 case ltos: pop_l(rax, rdx); break;
693 case ftos:
694 if (UseSSE >= 1) {
695 pop_f(xmm0);
696 } else {
697 pop_f();
698 }
699 break;
700 case dtos:
701 if (UseSSE >= 2) {
702 pop_d(xmm0);
703 } else {
704 pop_d();
705 }
706 break;
707 case vtos: /* nothing to do */ break;
708 default : ShouldNotReachHere();
709 }
710 verify_oop(rax, state);
711 }
712
713
714 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
715 push(hi);
716 push(lo);
717 }
718
719 void InterpreterMacroAssembler::push_f() {
720 // Do not schedule for no AGI! Never write beyond rsp!
721 subptr(rsp, 1 * wordSize);
722 fstp_s(Address(rsp, 0));
723 }
724
725 void InterpreterMacroAssembler::push_d() {
726 // Do not schedule for no AGI! Never write beyond rsp!
727 subptr(rsp, 2 * wordSize);
728 fstp_d(Address(rsp, 0));
729 }
730
731
732 void InterpreterMacroAssembler::push(TosState state) {
733 verify_oop(rax, state);
734 switch (state) {
735 case atos: push_ptr(rax); break;
736 case btos: // fall through
737 case ztos: // fall through
738 case ctos: // fall through
739 case stos: // fall through
740 case itos: push_i(rax); break;
741 case ltos: push_l(rax, rdx); break;
742 case ftos:
743 if (UseSSE >= 1) {
744 push_f(xmm0);
745 } else {
746 push_f();
747 }
748 break;
749 case dtos:
750 if (UseSSE >= 2) {
751 push_d(xmm0);
752 } else {
753 push_d();
754 }
755 break;
756 case vtos: /* nothing to do */ break;
757 default : ShouldNotReachHere();
758 }
759 }
760 #endif // _LP64
761
762
763 // Helpers for swap and dup
764 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
765 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
766 }
767
768 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
769 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
770 }
771
772
773 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
774 // set sender sp
775 lea(_bcp_register, Address(rsp, wordSize));
776 // record last_sp
777 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
778 }
779
780
781 // Jump to from_interpreted entry of a call unless single stepping is possible
782 // in this thread in which case we must call the i2i entry
783 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
784 prepare_to_jump_from_interpreted();
785
786 if (JvmtiExport::can_post_interpreter_events()) {
787 Label run_compiled_code;
788 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
789 // compiled code in threads for which the event is enabled. Check here for
790 // interp_only_mode if these events CAN be enabled.
791 // interp_only is an int, on little endian it is sufficient to test the byte only
792 // Is a cmpl faster?
793 LP64_ONLY(temp = r15_thread;)
794 NOT_LP64(get_thread(temp);)
795 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
796 jccb(Assembler::zero, run_compiled_code);
797 jmp(Address(method, Method::interpreter_entry_offset()));
798 bind(run_compiled_code);
799 }
800
801 jmp(Address(method, Method::from_interpreted_offset()));
802 }
803
804 // The following two routines provide a hook so that an implementation
805 // can schedule the dispatch in two parts. x86 does not do this.
806 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
807 // Nothing x86 specific to be done here
808 }
809
810 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
811 dispatch_next(state, step);
812 }
813
814 void InterpreterMacroAssembler::dispatch_base(TosState state,
815 address* table,
816 bool verifyoop,
817 bool generate_poll) {
818 verify_FPU(1, state);
819 if (VerifyActivationFrameSize) {
820 Label L;
821 mov(rcx, rbp);
822 subptr(rcx, rsp);
823 int32_t min_frame_size =
824 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
825 wordSize;
826 cmpptr(rcx, (int32_t)min_frame_size);
827 jcc(Assembler::greaterEqual, L);
828 stop("broken stack frame");
829 bind(L);
830 }
831 if (verifyoop) {
832 verify_oop(rax, state);
833 }
834
835 address* const safepoint_table = Interpreter::safept_table(state);
836 #ifdef _LP64
837 Label no_safepoint, dispatch;
838 if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) {
839 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
840 testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
841
842 jccb(Assembler::zero, no_safepoint);
843 lea(rscratch1, ExternalAddress((address)safepoint_table));
844 jmpb(dispatch);
845 }
846
847 bind(no_safepoint);
848 lea(rscratch1, ExternalAddress((address)table));
849 bind(dispatch);
850 jmp(Address(rscratch1, rbx, Address::times_8));
851
852 #else
853 Address index(noreg, rbx, Address::times_ptr);
854 if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) {
855 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
856 Label no_safepoint;
857 const Register thread = rcx;
858 get_thread(thread);
859 testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
860
861 jccb(Assembler::zero, no_safepoint);
862 ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
863 jump(dispatch_addr);
864 bind(no_safepoint);
865 }
866
867 {
868 ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
869 jump(dispatch_addr);
870 }
871 #endif // _LP64
872 }
873
874 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
875 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
876 }
877
878 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
879 dispatch_base(state, Interpreter::normal_table(state));
880 }
881
882 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
883 dispatch_base(state, Interpreter::normal_table(state), false);
884 }
885
886
887 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
888 // load next bytecode (load before advancing _bcp_register to prevent AGI)
889 load_unsigned_byte(rbx, Address(_bcp_register, step));
890 // advance _bcp_register
891 increment(_bcp_register, step);
892 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
893 }
894
895 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
896 // load current bytecode
897 load_unsigned_byte(rbx, Address(_bcp_register, 0));
898 dispatch_base(state, table);
899 }
900
901 void InterpreterMacroAssembler::narrow(Register result) {
902
903 // Get method->_constMethod->_result_type
904 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
905 movptr(rcx, Address(rcx, Method::const_offset()));
906 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
907
908 Label done, notBool, notByte, notChar;
909
910 // common case first
911 cmpl(rcx, T_INT);
912 jcc(Assembler::equal, done);
913
914 // mask integer result to narrower return type.
915 cmpl(rcx, T_BOOLEAN);
916 jcc(Assembler::notEqual, notBool);
917 andl(result, 0x1);
918 jmp(done);
919
920 bind(notBool);
921 cmpl(rcx, T_BYTE);
922 jcc(Assembler::notEqual, notByte);
923 LP64_ONLY(movsbl(result, result);)
924 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
925 NOT_LP64(sarl(result, 24);) // and sign-extend byte
926 jmp(done);
927
928 bind(notByte);
929 cmpl(rcx, T_CHAR);
930 jcc(Assembler::notEqual, notChar);
931 LP64_ONLY(movzwl(result, result);)
932 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
933 jmp(done);
934
935 bind(notChar);
936 // cmpl(rcx, T_SHORT); // all that's left
937 // jcc(Assembler::notEqual, done);
938 LP64_ONLY(movswl(result, result);)
939 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
940 NOT_LP64(sarl(result, 16);) // and sign-extend short
941
942 // Nothing to do for T_INT
943 bind(done);
944 }
945
946 // remove activation
947 //
948 // Unlock the receiver if this is a synchronized method.
949 // Unlock any Java monitors from syncronized blocks.
950 // Remove the activation from the stack.
951 //
952 // If there are locked Java monitors
953 // If throw_monitor_exception
954 // throws IllegalMonitorStateException
955 // Else if install_monitor_exception
956 // installs IllegalMonitorStateException
957 // Else
958 // no error processing
959 void InterpreterMacroAssembler::remove_activation(
960 TosState state,
961 Register ret_addr,
962 bool throw_monitor_exception,
963 bool install_monitor_exception,
964 bool notify_jvmdi) {
965 // Note: Registers rdx xmm0 may be in use for the
966 // result check if synchronized method
967 Label unlocked, unlock, no_unlock;
968
969 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
970 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
971 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
972 // monitor pointers need different register
973 // because rdx may have the result in it
974 NOT_LP64(get_thread(rcx);)
975
976 // get the value of _do_not_unlock_if_synchronized into rdx
977 const Address do_not_unlock_if_synchronized(rthread,
978 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
979 movbool(rbx, do_not_unlock_if_synchronized);
980 movbool(do_not_unlock_if_synchronized, false); // reset the flag
981
982 // get method access flags
983 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
984 movl(rcx, Address(rcx, Method::access_flags_offset()));
985 testl(rcx, JVM_ACC_SYNCHRONIZED);
986 jcc(Assembler::zero, unlocked);
987
988 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
989 // is set.
990 testbool(rbx);
991 jcc(Assembler::notZero, no_unlock);
992
993 // unlock monitor
994 push(state); // save result
995
996 // BasicObjectLock will be first in list, since this is a
997 // synchronized method. However, need to check that the object has
998 // not been unlocked by an explicit monitorexit bytecode.
999 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
1000 wordSize - (int) sizeof(BasicObjectLock));
1001 // We use c_rarg1/rdx so that if we go slow path it will be the correct
1002 // register for unlock_object to pass to VM directly
1003 lea(robj, monitor); // address of first monitor
1004
1005 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
1006 testptr(rax, rax);
1007 jcc(Assembler::notZero, unlock);
1008
1009 pop(state);
1010 if (throw_monitor_exception) {
1011 // Entry already unlocked, need to throw exception
1012 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
1013 call_VM(noreg, CAST_FROM_FN_PTR(address,
1014 InterpreterRuntime::throw_illegal_monitor_state_exception));
1015 should_not_reach_here();
1016 } else {
1017 // Monitor already unlocked during a stack unroll. If requested,
1018 // install an illegal_monitor_state_exception. Continue with
1019 // stack unrolling.
1020 if (install_monitor_exception) {
1021 NOT_LP64(empty_FPU_stack();)
1022 call_VM(noreg, CAST_FROM_FN_PTR(address,
1023 InterpreterRuntime::new_illegal_monitor_state_exception));
1024 }
1025 jmp(unlocked);
1026 }
1027
1028 bind(unlock);
1029 unlock_object(robj);
1030 pop(state);
1031
1032 // Check that for block-structured locking (i.e., that all locked
1033 // objects has been unlocked)
1034 bind(unlocked);
1035
1036 // rax, rdx: Might contain return value
1037
1038 // Check that all monitors are unlocked
1039 {
1040 Label loop, exception, entry, restart;
1041 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1042 const Address monitor_block_top(
1043 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1044 const Address monitor_block_bot(
1045 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1046
1047 bind(restart);
1048 // We use c_rarg1 so that if we go slow path it will be the correct
1049 // register for unlock_object to pass to VM directly
1050 movptr(rmon, monitor_block_top); // points to current entry, starting
1051 // with top-most entry
1052 lea(rbx, monitor_block_bot); // points to word before bottom of
1053 // monitor block
1054 jmp(entry);
1055
1056 // Entry already locked, need to throw exception
1057 bind(exception);
1058
1059 if (throw_monitor_exception) {
1060 // Throw exception
1061 NOT_LP64(empty_FPU_stack();)
1062 MacroAssembler::call_VM(noreg,
1063 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1064 throw_illegal_monitor_state_exception));
1065 should_not_reach_here();
1066 } else {
1067 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1068 // Unlock does not block, so don't have to worry about the frame.
1069 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1070
1071 push(state);
1072 mov(robj, rmon); // nop if robj and rmon are the same
1073 unlock_object(robj);
1074 pop(state);
1075
1076 if (install_monitor_exception) {
1077 NOT_LP64(empty_FPU_stack();)
1078 call_VM(noreg, CAST_FROM_FN_PTR(address,
1079 InterpreterRuntime::
1080 new_illegal_monitor_state_exception));
1081 }
1082
1083 jmp(restart);
1084 }
1085
1086 bind(loop);
1087 // check if current entry is used
1088 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
1089 jcc(Assembler::notEqual, exception);
1090
1091 addptr(rmon, entry_size); // otherwise advance to next entry
1092 bind(entry);
1093 cmpptr(rmon, rbx); // check if bottom reached
1094 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1095 }
1096
1097 bind(no_unlock);
1098
1099 // jvmti support
1100 if (notify_jvmdi) {
1101 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1102 } else {
1103 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1104 }
1105
1106 // remove activation
1107 // get sender sp
1108 movptr(rbx,
1109 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1110 if (StackReservedPages > 0) {
1111 // testing if reserved zone needs to be re-enabled
1112 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1113 Label no_reserved_zone_enabling;
1114
1115 NOT_LP64(get_thread(rthread);)
1116
1117 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_enabled);
1118 jcc(Assembler::equal, no_reserved_zone_enabling);
1119
1120 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1121 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1122
1123 call_VM_leaf(
1124 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1125 call_VM(noreg, CAST_FROM_FN_PTR(address,
1126 InterpreterRuntime::throw_delayed_StackOverflowError));
1127 should_not_reach_here();
1128
1129 bind(no_reserved_zone_enabling);
1130 }
1131 leave(); // remove frame anchor
1132 pop(ret_addr); // get return address
1133 mov(rsp, rbx); // set sp to sender sp
1134 }
1135
1136 void InterpreterMacroAssembler::get_method_counters(Register method,
1137 Register mcs, Label& skip) {
1138 Label has_counters;
1139 movptr(mcs, Address(method, Method::method_counters_offset()));
1140 testptr(mcs, mcs);
1141 jcc(Assembler::notZero, has_counters);
1142 call_VM(noreg, CAST_FROM_FN_PTR(address,
1143 InterpreterRuntime::build_method_counters), method);
1144 movptr(mcs, Address(method,Method::method_counters_offset()));
1145 testptr(mcs, mcs);
1146 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1147 bind(has_counters);
1148 }
1149
1150
1151 // Lock object
1152 //
1153 // Args:
1154 // rdx, c_rarg1: BasicObjectLock to be used for locking
1155 //
1156 // Kills:
1157 // rax, rbx
1158 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1159 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1160 "The argument is only for looks. It must be c_rarg1");
1161
1162 if (UseHeavyMonitors) {
1163 call_VM(noreg,
1164 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1165 lock_reg);
1166 } else {
1167 Label done;
1168
1169 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1170 const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a
1171 // problematic case where tmp_reg = no_reg.
1172 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1173
1174 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
1175 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
1176 const int mark_offset = lock_offset +
1177 BasicLock::displaced_header_offset_in_bytes();
1178
1179 Label slow_case;
1180
1181 // Load object pointer into obj_reg
1182 movptr(obj_reg, Address(lock_reg, obj_offset));
1183
1184 if (UseBiasedLocking) {
1185 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case);
1186 }
1187
1188 // Load immediate 1 into swap_reg %rax
1189 movl(swap_reg, (int32_t)1);
1190
1191 // Load (object->mark() | 1) into swap_reg %rax
1192 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1193
1194 // Save (object->mark() | 1) into BasicLock's displaced header
1195 movptr(Address(lock_reg, mark_offset), swap_reg);
1196
1197 assert(lock_offset == 0,
1198 "displaced header must be first word in BasicObjectLock");
1199
1200 if (os::is_MP()) lock();
1201 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1202 if (PrintBiasedLockingStatistics) {
1203 cond_inc32(Assembler::zero,
1204 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1205 }
1206 jcc(Assembler::zero, done);
1207
1208 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1209
1210 // Test if the oopMark is an obvious stack pointer, i.e.,
1211 // 1) (mark & zero_bits) == 0, and
1212 // 2) rsp <= mark < mark + os::pagesize()
1213 //
1214 // These 3 tests can be done by evaluating the following
1215 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1216 // assuming both stack pointer and pagesize have their
1217 // least significant bits clear.
1218 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
1219 subptr(swap_reg, rsp);
1220 andptr(swap_reg, zero_bits - os::vm_page_size());
1221
1222 // Save the test result, for recursive case, the result is zero
1223 movptr(Address(lock_reg, mark_offset), swap_reg);
1224
1225 if (PrintBiasedLockingStatistics) {
1226 cond_inc32(Assembler::zero,
1227 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1228 }
1229 jcc(Assembler::zero, done);
1230
1231 bind(slow_case);
1232
1233 // Call the runtime routine for slow case
1234 call_VM(noreg,
1235 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1236 lock_reg);
1237
1238 bind(done);
1239 }
1240 }
1241
1242
1243 // Unlocks an object. Used in monitorexit bytecode and
1244 // remove_activation. Throws an IllegalMonitorException if object is
1245 // not locked by current thread.
1246 //
1247 // Args:
1248 // rdx, c_rarg1: BasicObjectLock for lock
1249 //
1250 // Kills:
1251 // rax
1252 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1253 // rscratch1 (scratch reg)
1254 // rax, rbx, rcx, rdx
1255 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1256 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1257 "The argument is only for looks. It must be c_rarg1");
1258
1259 if (UseHeavyMonitors) {
1260 call_VM(noreg,
1261 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1262 lock_reg);
1263 } else {
1264 Label done;
1265
1266 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1267 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1268 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1269
1270 save_bcp(); // Save in case of exception
1271
1272 // Convert from BasicObjectLock structure to object and BasicLock
1273 // structure Store the BasicLock address into %rax
1274 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
1275
1276 // Load oop into obj_reg(%c_rarg3)
1277 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
1278
1279 // Free entry
1280 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
1281
1282 if (UseBiasedLocking) {
1283 biased_locking_exit(obj_reg, header_reg, done);
1284 }
1285
1286 // Load the old header from BasicLock structure
1287 movptr(header_reg, Address(swap_reg,
1288 BasicLock::displaced_header_offset_in_bytes()));
1289
1290 // Test for recursion
1291 testptr(header_reg, header_reg);
1292
1293 // zero for recursive case
1294 jcc(Assembler::zero, done);
1295
1296 // Atomic swap back the old header
1297 if (os::is_MP()) lock();
1298 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1299
1300 // zero for simple unlock of a stack-lock case
1301 jcc(Assembler::zero, done);
1302
1303 // Call the runtime routine for slow case.
1304 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
1305 obj_reg); // restore obj
1306 call_VM(noreg,
1307 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1308 lock_reg);
1309
1310 bind(done);
1311
1312 restore_bcp();
1313 }
1314 }
1315
1316 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1317 Label& zero_continue) {
1318 assert(ProfileInterpreter, "must be profiling interpreter");
1319 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1320 testptr(mdp, mdp);
1321 jcc(Assembler::zero, zero_continue);
1322 }
1323
1324
1325 // Set the method data pointer for the current bcp.
1326 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1327 assert(ProfileInterpreter, "must be profiling interpreter");
1328 Label set_mdp;
1329 push(rax);
1330 push(rbx);
1331
1332 get_method(rbx);
1333 // Test MDO to avoid the call if it is NULL.
1334 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1335 testptr(rax, rax);
1336 jcc(Assembler::zero, set_mdp);
1337 // rbx: method
1338 // _bcp_register: bcp
1339 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1340 // rax: mdi
1341 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1342 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1343 addptr(rbx, in_bytes(MethodData::data_offset()));
1344 addptr(rax, rbx);
1345 bind(set_mdp);
1346 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1347 pop(rbx);
1348 pop(rax);
1349 }
1350
1351 void InterpreterMacroAssembler::verify_method_data_pointer() {
1352 assert(ProfileInterpreter, "must be profiling interpreter");
1353 #ifdef ASSERT
1354 Label verify_continue;
1355 push(rax);
1356 push(rbx);
1357 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1358 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1359 push(arg3_reg);
1360 push(arg2_reg);
1361 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1362 get_method(rbx);
1363
1364 // If the mdp is valid, it will point to a DataLayout header which is
1365 // consistent with the bcp. The converse is highly probable also.
1366 load_unsigned_short(arg2_reg,
1367 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1368 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1369 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1370 cmpptr(arg2_reg, _bcp_register);
1371 jcc(Assembler::equal, verify_continue);
1372 // rbx: method
1373 // _bcp_register: bcp
1374 // c_rarg3: mdp
1375 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1376 rbx, _bcp_register, arg3_reg);
1377 bind(verify_continue);
1378 pop(arg2_reg);
1379 pop(arg3_reg);
1380 pop(rbx);
1381 pop(rax);
1382 #endif // ASSERT
1383 }
1384
1385
1386 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1387 int constant,
1388 Register value) {
1389 assert(ProfileInterpreter, "must be profiling interpreter");
1390 Address data(mdp_in, constant);
1391 movptr(data, value);
1392 }
1393
1394
1395 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1396 int constant,
1397 bool decrement) {
1398 // Counter address
1399 Address data(mdp_in, constant);
1400
1401 increment_mdp_data_at(data, decrement);
1402 }
1403
1404 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1405 bool decrement) {
1406 assert(ProfileInterpreter, "must be profiling interpreter");
1407 // %%% this does 64bit counters at best it is wasting space
1408 // at worst it is a rare bug when counters overflow
1409
1410 if (decrement) {
1411 // Decrement the register. Set condition codes.
1412 addptr(data, (int32_t) -DataLayout::counter_increment);
1413 // If the decrement causes the counter to overflow, stay negative
1414 Label L;
1415 jcc(Assembler::negative, L);
1416 addptr(data, (int32_t) DataLayout::counter_increment);
1417 bind(L);
1418 } else {
1419 assert(DataLayout::counter_increment == 1,
1420 "flow-free idiom only works with 1");
1421 // Increment the register. Set carry flag.
1422 addptr(data, DataLayout::counter_increment);
1423 // If the increment causes the counter to overflow, pull back by 1.
1424 sbbptr(data, (int32_t)0);
1425 }
1426 }
1427
1428
1429 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1430 Register reg,
1431 int constant,
1432 bool decrement) {
1433 Address data(mdp_in, reg, Address::times_1, constant);
1434
1435 increment_mdp_data_at(data, decrement);
1436 }
1437
1438 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1439 int flag_byte_constant) {
1440 assert(ProfileInterpreter, "must be profiling interpreter");
1441 int header_offset = in_bytes(DataLayout::header_offset());
1442 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
1443 // Set the flag
1444 orl(Address(mdp_in, header_offset), header_bits);
1445 }
1446
1447
1448
1449 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1450 int offset,
1451 Register value,
1452 Register test_value_out,
1453 Label& not_equal_continue) {
1454 assert(ProfileInterpreter, "must be profiling interpreter");
1455 if (test_value_out == noreg) {
1456 cmpptr(value, Address(mdp_in, offset));
1457 } else {
1458 // Put the test value into a register, so caller can use it:
1459 movptr(test_value_out, Address(mdp_in, offset));
1460 cmpptr(test_value_out, value);
1461 }
1462 jcc(Assembler::notEqual, not_equal_continue);
1463 }
1464
1465
1466 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1467 int offset_of_disp) {
1468 assert(ProfileInterpreter, "must be profiling interpreter");
1469 Address disp_address(mdp_in, offset_of_disp);
1470 addptr(mdp_in, disp_address);
1471 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1472 }
1473
1474
1475 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1476 Register reg,
1477 int offset_of_disp) {
1478 assert(ProfileInterpreter, "must be profiling interpreter");
1479 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1480 addptr(mdp_in, disp_address);
1481 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1482 }
1483
1484
1485 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1486 int constant) {
1487 assert(ProfileInterpreter, "must be profiling interpreter");
1488 addptr(mdp_in, constant);
1489 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1490 }
1491
1492
1493 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1494 assert(ProfileInterpreter, "must be profiling interpreter");
1495 push(return_bci); // save/restore across call_VM
1496 call_VM(noreg,
1497 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1498 return_bci);
1499 pop(return_bci);
1500 }
1501
1502
1503 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1504 Register bumped_count) {
1505 if (ProfileInterpreter) {
1506 Label profile_continue;
1507
1508 // If no method data exists, go to profile_continue.
1509 // Otherwise, assign to mdp
1510 test_method_data_pointer(mdp, profile_continue);
1511
1512 // We are taking a branch. Increment the taken count.
1513 // We inline increment_mdp_data_at to return bumped_count in a register
1514 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1515 Address data(mdp, in_bytes(JumpData::taken_offset()));
1516 movptr(bumped_count, data);
1517 assert(DataLayout::counter_increment == 1,
1518 "flow-free idiom only works with 1");
1519 addptr(bumped_count, DataLayout::counter_increment);
1520 sbbptr(bumped_count, 0);
1521 movptr(data, bumped_count); // Store back out
1522
1523 // The method data pointer needs to be updated to reflect the new target.
1524 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1525 bind(profile_continue);
1526 }
1527 }
1528
1529
1530 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1531 if (ProfileInterpreter) {
1532 Label profile_continue;
1533
1534 // If no method data exists, go to profile_continue.
1535 test_method_data_pointer(mdp, profile_continue);
1536
1537 // We are taking a branch. Increment the not taken count.
1538 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1539
1540 // The method data pointer needs to be updated to correspond to
1541 // the next bytecode
1542 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1543 bind(profile_continue);
1544 }
1545 }
1546
1547 void InterpreterMacroAssembler::profile_call(Register mdp) {
1548 if (ProfileInterpreter) {
1549 Label profile_continue;
1550
1551 // If no method data exists, go to profile_continue.
1552 test_method_data_pointer(mdp, profile_continue);
1553
1554 // We are making a call. Increment the count.
1555 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1556
1557 // The method data pointer needs to be updated to reflect the new target.
1558 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1559 bind(profile_continue);
1560 }
1561 }
1562
1563
1564 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1565 if (ProfileInterpreter) {
1566 Label profile_continue;
1567
1568 // If no method data exists, go to profile_continue.
1569 test_method_data_pointer(mdp, profile_continue);
1570
1571 // We are making a call. Increment the count.
1572 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1573
1574 // The method data pointer needs to be updated to reflect the new target.
1575 update_mdp_by_constant(mdp,
1576 in_bytes(VirtualCallData::
1577 virtual_call_data_size()));
1578 bind(profile_continue);
1579 }
1580 }
1581
1582
1583 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1584 Register mdp,
1585 Register reg2,
1586 bool receiver_can_be_null) {
1587 if (ProfileInterpreter) {
1588 Label profile_continue;
1589
1590 // If no method data exists, go to profile_continue.
1591 test_method_data_pointer(mdp, profile_continue);
1592
1593 Label skip_receiver_profile;
1594 if (receiver_can_be_null) {
1595 Label not_null;
1596 testptr(receiver, receiver);
1597 jccb(Assembler::notZero, not_null);
1598 // We are making a call. Increment the count for null receiver.
1599 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1600 jmp(skip_receiver_profile);
1601 bind(not_null);
1602 }
1603
1604 // Record the receiver type.
1605 record_klass_in_profile(receiver, mdp, reg2, true);
1606 bind(skip_receiver_profile);
1607
1608 // The method data pointer needs to be updated to reflect the new target.
1609 #if INCLUDE_JVMCI
1610 if (MethodProfileWidth == 0) {
1611 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1612 }
1613 #else // INCLUDE_JVMCI
1614 update_mdp_by_constant(mdp,
1615 in_bytes(VirtualCallData::
1616 virtual_call_data_size()));
1617 #endif // INCLUDE_JVMCI
1618 bind(profile_continue);
1619 }
1620 }
1621
1622 #if INCLUDE_JVMCI
1623 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) {
1624 assert_different_registers(method, mdp, reg2);
1625 if (ProfileInterpreter && MethodProfileWidth > 0) {
1626 Label profile_continue;
1627
1628 // If no method data exists, go to profile_continue.
1629 test_method_data_pointer(mdp, profile_continue);
1630
1631 Label done;
1632 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth,
1633 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1634 bind(done);
1635
1636 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1637 bind(profile_continue);
1638 }
1639 }
1640 #endif // INCLUDE_JVMCI
1641
1642 // This routine creates a state machine for updating the multi-row
1643 // type profile at a virtual call site (or other type-sensitive bytecode).
1644 // The machine visits each row (of receiver/count) until the receiver type
1645 // is found, or until it runs out of rows. At the same time, it remembers
1646 // the location of the first empty row. (An empty row records null for its
1647 // receiver, and can be allocated for a newly-observed receiver type.)
1648 // Because there are two degrees of freedom in the state, a simple linear
1649 // search will not work; it must be a decision tree. Hence this helper
1650 // function is recursive, to generate the required tree structured code.
1651 // It's the interpreter, so we are trading off code space for speed.
1652 // See below for example code.
1653 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1654 Register receiver, Register mdp,
1655 Register reg2, int start_row,
1656 Label& done, bool is_virtual_call) {
1657 if (TypeProfileWidth == 0) {
1658 if (is_virtual_call) {
1659 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1660 }
1661 #if INCLUDE_JVMCI
1662 else if (EnableJVMCI) {
1663 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1664 }
1665 #endif // INCLUDE_JVMCI
1666 } else {
1667 int non_profiled_offset = -1;
1668 if (is_virtual_call) {
1669 non_profiled_offset = in_bytes(CounterData::count_offset());
1670 }
1671 #if INCLUDE_JVMCI
1672 else if (EnableJVMCI) {
1673 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1674 }
1675 #endif // INCLUDE_JVMCI
1676
1677 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1678 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1679 }
1680 }
1681
1682 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1683 Register reg2, int start_row, Label& done, int total_rows,
1684 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1685 int non_profiled_offset) {
1686 int last_row = total_rows - 1;
1687 assert(start_row <= last_row, "must be work left to do");
1688 // Test this row for both the item and for null.
1689 // Take any of three different outcomes:
1690 // 1. found item => increment count and goto done
1691 // 2. found null => keep looking for case 1, maybe allocate this cell
1692 // 3. found something else => keep looking for cases 1 and 2
1693 // Case 3 is handled by a recursive call.
1694 for (int row = start_row; row <= last_row; row++) {
1695 Label next_test;
1696 bool test_for_null_also = (row == start_row);
1697
1698 // See if the item is item[n].
1699 int item_offset = in_bytes(item_offset_fn(row));
1700 test_mdp_data_at(mdp, item_offset, item,
1701 (test_for_null_also ? reg2 : noreg),
1702 next_test);
1703 // (Reg2 now contains the item from the CallData.)
1704
1705 // The item is item[n]. Increment count[n].
1706 int count_offset = in_bytes(item_count_offset_fn(row));
1707 increment_mdp_data_at(mdp, count_offset);
1708 jmp(done);
1709 bind(next_test);
1710
1711 if (test_for_null_also) {
1712 Label found_null;
1713 // Failed the equality check on item[n]... Test for null.
1714 testptr(reg2, reg2);
1715 if (start_row == last_row) {
1716 // The only thing left to do is handle the null case.
1717 if (non_profiled_offset >= 0) {
1718 jccb(Assembler::zero, found_null);
1719 // Item did not match any saved item and there is no empty row for it.
1720 // Increment total counter to indicate polymorphic case.
1721 increment_mdp_data_at(mdp, non_profiled_offset);
1722 jmp(done);
1723 bind(found_null);
1724 } else {
1725 jcc(Assembler::notZero, done);
1726 }
1727 break;
1728 }
1729 // Since null is rare, make it be the branch-taken case.
1730 jcc(Assembler::zero, found_null);
1731
1732 // Put all the "Case 3" tests here.
1733 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1734 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1735
1736 // Found a null. Keep searching for a matching item,
1737 // but remember that this is an empty (unused) slot.
1738 bind(found_null);
1739 }
1740 }
1741
1742 // In the fall-through case, we found no matching item, but we
1743 // observed the item[start_row] is NULL.
1744
1745 // Fill in the item field and increment the count.
1746 int item_offset = in_bytes(item_offset_fn(start_row));
1747 set_mdp_data_at(mdp, item_offset, item);
1748 int count_offset = in_bytes(item_count_offset_fn(start_row));
1749 movl(reg2, DataLayout::counter_increment);
1750 set_mdp_data_at(mdp, count_offset, reg2);
1751 if (start_row > 0) {
1752 jmp(done);
1753 }
1754 }
1755
1756 // Example state machine code for three profile rows:
1757 // // main copy of decision tree, rooted at row[1]
1758 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1759 // if (row[0].rec != NULL) {
1760 // // inner copy of decision tree, rooted at row[1]
1761 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1762 // if (row[1].rec != NULL) {
1763 // // degenerate decision tree, rooted at row[2]
1764 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1765 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1766 // row[2].init(rec); goto done;
1767 // } else {
1768 // // remember row[1] is empty
1769 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1770 // row[1].init(rec); goto done;
1771 // }
1772 // } else {
1773 // // remember row[0] is empty
1774 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1775 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1776 // row[0].init(rec); goto done;
1777 // }
1778 // done:
1779
1780 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1781 Register mdp, Register reg2,
1782 bool is_virtual_call) {
1783 assert(ProfileInterpreter, "must be profiling");
1784 Label done;
1785
1786 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1787
1788 bind (done);
1789 }
1790
1791 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1792 Register mdp) {
1793 if (ProfileInterpreter) {
1794 Label profile_continue;
1795 uint row;
1796
1797 // If no method data exists, go to profile_continue.
1798 test_method_data_pointer(mdp, profile_continue);
1799
1800 // Update the total ret count.
1801 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1802
1803 for (row = 0; row < RetData::row_limit(); row++) {
1804 Label next_test;
1805
1806 // See if return_bci is equal to bci[n]:
1807 test_mdp_data_at(mdp,
1808 in_bytes(RetData::bci_offset(row)),
1809 return_bci, noreg,
1810 next_test);
1811
1812 // return_bci is equal to bci[n]. Increment the count.
1813 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1814
1815 // The method data pointer needs to be updated to reflect the new target.
1816 update_mdp_by_offset(mdp,
1817 in_bytes(RetData::bci_displacement_offset(row)));
1818 jmp(profile_continue);
1819 bind(next_test);
1820 }
1821
1822 update_mdp_for_ret(return_bci);
1823
1824 bind(profile_continue);
1825 }
1826 }
1827
1828
1829 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1830 if (ProfileInterpreter) {
1831 Label profile_continue;
1832
1833 // If no method data exists, go to profile_continue.
1834 test_method_data_pointer(mdp, profile_continue);
1835
1836 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1837
1838 // The method data pointer needs to be updated.
1839 int mdp_delta = in_bytes(BitData::bit_data_size());
1840 if (TypeProfileCasts) {
1841 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1842 }
1843 update_mdp_by_constant(mdp, mdp_delta);
1844
1845 bind(profile_continue);
1846 }
1847 }
1848
1849
1850 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1851 if (ProfileInterpreter && TypeProfileCasts) {
1852 Label profile_continue;
1853
1854 // If no method data exists, go to profile_continue.
1855 test_method_data_pointer(mdp, profile_continue);
1856
1857 int count_offset = in_bytes(CounterData::count_offset());
1858 // Back up the address, since we have already bumped the mdp.
1859 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1860
1861 // *Decrement* the counter. We expect to see zero or small negatives.
1862 increment_mdp_data_at(mdp, count_offset, true);
1863
1864 bind (profile_continue);
1865 }
1866 }
1867
1868
1869 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1870 if (ProfileInterpreter) {
1871 Label profile_continue;
1872
1873 // If no method data exists, go to profile_continue.
1874 test_method_data_pointer(mdp, profile_continue);
1875
1876 // The method data pointer needs to be updated.
1877 int mdp_delta = in_bytes(BitData::bit_data_size());
1878 if (TypeProfileCasts) {
1879 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1880
1881 // Record the object type.
1882 record_klass_in_profile(klass, mdp, reg2, false);
1883 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1884 NOT_LP64(restore_locals();) // Restore EDI
1885 }
1886 update_mdp_by_constant(mdp, mdp_delta);
1887
1888 bind(profile_continue);
1889 }
1890 }
1891
1892
1893 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1894 if (ProfileInterpreter) {
1895 Label profile_continue;
1896
1897 // If no method data exists, go to profile_continue.
1898 test_method_data_pointer(mdp, profile_continue);
1899
1900 // Update the default case count
1901 increment_mdp_data_at(mdp,
1902 in_bytes(MultiBranchData::default_count_offset()));
1903
1904 // The method data pointer needs to be updated.
1905 update_mdp_by_offset(mdp,
1906 in_bytes(MultiBranchData::
1907 default_displacement_offset()));
1908
1909 bind(profile_continue);
1910 }
1911 }
1912
1913
1914 void InterpreterMacroAssembler::profile_switch_case(Register index,
1915 Register mdp,
1916 Register reg2) {
1917 if (ProfileInterpreter) {
1918 Label profile_continue;
1919
1920 // If no method data exists, go to profile_continue.
1921 test_method_data_pointer(mdp, profile_continue);
1922
1923 // Build the base (index * per_case_size_in_bytes()) +
1924 // case_array_offset_in_bytes()
1925 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1926 imulptr(index, reg2); // XXX l ?
1927 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1928
1929 // Update the case count
1930 increment_mdp_data_at(mdp,
1931 index,
1932 in_bytes(MultiBranchData::relative_count_offset()));
1933
1934 // The method data pointer needs to be updated.
1935 update_mdp_by_offset(mdp,
1936 index,
1937 in_bytes(MultiBranchData::
1938 relative_displacement_offset()));
1939
1940 bind(profile_continue);
1941 }
1942 }
1943
1944
1945
1946 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1947 if (state == atos) {
1948 MacroAssembler::verify_oop(reg);
1949 }
1950 }
1951
1952 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1953 #ifndef _LP64
1954 if ((state == ftos && UseSSE < 1) ||
1955 (state == dtos && UseSSE < 2)) {
1956 MacroAssembler::verify_FPU(stack_depth);
1957 }
1958 #endif
1959 }
1960
1961 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1962 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1963 int increment, Address mask,
1964 Register scratch, bool preloaded,
1965 Condition cond, Label* where) {
1966 if (!preloaded) {
1967 movl(scratch, counter_addr);
1968 }
1969 incrementl(scratch, increment);
1970 movl(counter_addr, scratch);
1971 andl(scratch, mask);
1972 jcc(cond, *where);
1973 }
1974
1975 void InterpreterMacroAssembler::notify_method_entry() {
1976 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1977 // track stack depth. If it is possible to enter interp_only_mode we add
1978 // the code to check if the event should be sent.
1979 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1980 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1981 if (JvmtiExport::can_post_interpreter_events()) {
1982 Label L;
1983 NOT_LP64(get_thread(rthread);)
1984 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1985 testl(rdx, rdx);
1986 jcc(Assembler::zero, L);
1987 call_VM(noreg, CAST_FROM_FN_PTR(address,
1988 InterpreterRuntime::post_method_entry));
1989 bind(L);
1990 }
1991
1992 {
1993 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1994 NOT_LP64(get_thread(rthread);)
1995 get_method(rarg);
1996 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1997 rthread, rarg);
1998 }
1999
2000 // RedefineClasses() tracing support for obsolete method entry
2001 if (log_is_enabled(Trace, redefine, class, obsolete)) {
2002 NOT_LP64(get_thread(rthread);)
2003 get_method(rarg);
2004 call_VM_leaf(
2005 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2006 rthread, rarg);
2007 }
2008 }
2009
2010
2011 void InterpreterMacroAssembler::notify_method_exit(
2012 TosState state, NotifyMethodExitMode mode) {
2013 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2014 // track stack depth. If it is possible to enter interp_only_mode we add
2015 // the code to check if the event should be sent.
2016 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2017 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2018 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2019 Label L;
2020 // Note: frame::interpreter_frame_result has a dependency on how the
2021 // method result is saved across the call to post_method_exit. If this
2022 // is changed then the interpreter_frame_result implementation will
2023 // need to be updated too.
2024
2025 // template interpreter will leave the result on the top of the stack.
2026 push(state);
2027 NOT_LP64(get_thread(rthread);)
2028 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2029 testl(rdx, rdx);
2030 jcc(Assembler::zero, L);
2031 call_VM(noreg,
2032 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2033 bind(L);
2034 pop(state);
2035 }
2036
2037 {
2038 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2039 push(state);
2040 NOT_LP64(get_thread(rthread);)
2041 get_method(rarg);
2042 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2043 rthread, rarg);
2044 pop(state);
2045 }
2046 }