1 /*
2 * Copyright (c) 1997, 2024, 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 "oops/resolvedFieldEntry.hpp"
36 #include "oops/resolvedIndyEntry.hpp"
37 #include "oops/resolvedMethodEntry.hpp"
38 #include "prims/jvmtiExport.hpp"
39 #include "prims/jvmtiThreadState.hpp"
40 #include "runtime/basicLock.hpp"
41 #include "runtime/frame.inline.hpp"
42 #include "runtime/javaThread.hpp"
43 #include "runtime/safepointMechanism.hpp"
44 #include "runtime/sharedRuntime.hpp"
45 #include "utilities/powerOfTwo.hpp"
46
47 // Implementation of InterpreterMacroAssembler
48
49 void InterpreterMacroAssembler::jump_to_entry(address entry) {
50 assert(entry, "Entry must have been generated by now");
51 jump(RuntimeAddress(entry));
52 }
53
54 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
55 Label update, next, none;
56
57 #ifdef _LP64
58 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
59 #else
60 assert_different_registers(obj, mdo_addr.base(), mdo_addr.index());
61 #endif
62
63 interp_verify_oop(obj, atos);
64
65 testptr(obj, obj);
66 jccb(Assembler::notZero, update);
67 testptr(mdo_addr, TypeEntries::null_seen);
68 jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore.
69 // atomic update to prevent overwriting Klass* with 0
70 lock();
71 orptr(mdo_addr, TypeEntries::null_seen);
72 jmpb(next);
73
74 bind(update);
75 load_klass(obj, obj, rscratch1);
76 #ifdef _LP64
77 mov(rscratch1, obj);
78 #endif
79
80 xorptr(obj, mdo_addr);
81 testptr(obj, TypeEntries::type_klass_mask);
82 jccb(Assembler::zero, next); // klass seen before, nothing to
83 // do. The unknown bit may have been
84 // set already but no need to check.
85
86 testptr(obj, TypeEntries::type_unknown);
87 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
88
89 cmpptr(mdo_addr, 0);
90 jccb(Assembler::equal, none);
91 cmpptr(mdo_addr, TypeEntries::null_seen);
92 jccb(Assembler::equal, none);
93 #ifdef _LP64
94 // There is a chance that the checks above (re-reading profiling
95 // data from memory) fail if another thread has just set the
96 // profiling to this obj's klass
97 mov(obj, rscratch1);
98 xorptr(obj, mdo_addr);
99 testptr(obj, TypeEntries::type_klass_mask);
100 jccb(Assembler::zero, next);
101 #endif
102
103 // different than before. Cannot keep accurate profile.
104 orptr(mdo_addr, TypeEntries::type_unknown);
105 jmpb(next);
106
107 bind(none);
108 // first time here. Set profile type.
109 movptr(mdo_addr, obj);
110 #ifdef ASSERT
111 andptr(obj, TypeEntries::type_klass_mask);
112 verify_klass_ptr(obj);
113 #endif
114
115 bind(next);
116 }
117
118 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
119 if (!ProfileInterpreter) {
120 return;
121 }
122
123 if (MethodData::profile_arguments() || MethodData::profile_return()) {
124 Label profile_continue;
125
126 test_method_data_pointer(mdp, profile_continue);
127
128 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
129
130 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
131 jcc(Assembler::notEqual, profile_continue);
132
133 if (MethodData::profile_arguments()) {
134 Label done;
135 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
136 addptr(mdp, off_to_args);
137
138 for (int i = 0; i < TypeProfileArgsLimit; i++) {
139 if (i > 0 || MethodData::profile_return()) {
140 // If return value type is profiled we may have no argument to profile
141 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
142 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
143 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
144 jcc(Assembler::less, done);
145 }
146 movptr(tmp, Address(callee, Method::const_offset()));
147 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
148 // stack offset o (zero based) from the start of the argument
149 // list, for n arguments translates into offset n - o - 1 from
150 // the end of the argument list
151 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
152 subl(tmp, 1);
153 Address arg_addr = argument_address(tmp);
154 movptr(tmp, arg_addr);
155
156 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
157 profile_obj_type(tmp, mdo_arg_addr);
158
159 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
160 addptr(mdp, to_add);
161 off_to_args += to_add;
162 }
163
164 if (MethodData::profile_return()) {
165 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
166 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
167 }
168
169 bind(done);
170
171 if (MethodData::profile_return()) {
172 // We're right after the type profile for the last
173 // argument. tmp is the number of cells left in the
174 // CallTypeData/VirtualCallTypeData to reach its end. Non null
175 // if there's a return to profile.
176 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
177 shll(tmp, log2i_exact((int)DataLayout::cell_size));
178 addptr(mdp, tmp);
179 }
180 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
181 } else {
182 assert(MethodData::profile_return(), "either profile call args or call ret");
183 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
184 }
185
186 // mdp points right after the end of the
187 // CallTypeData/VirtualCallTypeData, right after the cells for the
188 // return value type if there's one
189
190 bind(profile_continue);
191 }
192 }
193
194 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
195 assert_different_registers(mdp, ret, tmp, _bcp_register);
196 if (ProfileInterpreter && MethodData::profile_return()) {
197 Label profile_continue;
198
199 test_method_data_pointer(mdp, profile_continue);
200
201 if (MethodData::profile_return_jsr292_only()) {
202 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
203
204 // If we don't profile all invoke bytecodes we must make sure
205 // it's a bytecode we indeed profile. We can't go back to the
206 // beginning of the ProfileData we intend to update to check its
207 // type because we're right after it and we don't known its
208 // length
209 Label do_profile;
210 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
211 jcc(Assembler::equal, do_profile);
212 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
213 jcc(Assembler::equal, do_profile);
214 get_method(tmp);
215 cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
216 jcc(Assembler::notEqual, profile_continue);
217
218 bind(do_profile);
219 }
220
221 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
222 mov(tmp, ret);
223 profile_obj_type(tmp, mdo_ret_addr);
224
225 bind(profile_continue);
226 }
227 }
228
229 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
230 if (ProfileInterpreter && MethodData::profile_parameters()) {
231 Label profile_continue;
232
233 test_method_data_pointer(mdp, profile_continue);
234
235 // Load the offset of the area within the MDO used for
236 // parameters. If it's negative we're not profiling any parameters
237 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
238 testl(tmp1, tmp1);
239 jcc(Assembler::negative, profile_continue);
240
241 // Compute a pointer to the area for parameters from the offset
242 // and move the pointer to the slot for the last
243 // parameters. Collect profiling from last parameter down.
244 // mdo start + parameters offset + array length - 1
245 addptr(mdp, tmp1);
246 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
247 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
248
249 Label loop;
250 bind(loop);
251
252 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
253 int type_base = in_bytes(ParametersTypeData::type_offset(0));
254 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
255 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
256 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
257
258 // load offset on the stack from the slot for this parameter
259 movptr(tmp2, arg_off);
260 negptr(tmp2);
261 // read the parameter from the local area
262 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
263
264 // profile the parameter
265 profile_obj_type(tmp2, arg_type);
266
267 // go to next parameter
268 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
269 jcc(Assembler::positive, loop);
270
271 bind(profile_continue);
272 }
273 }
274
275 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
276 int number_of_arguments) {
277 // interpreter specific
278 //
279 // Note: No need to save/restore bcp & locals registers
280 // since these are callee saved registers and no blocking/
281 // GC can happen in leaf calls.
282 // Further Note: DO NOT save/restore bcp/locals. If a caller has
283 // already saved them so that it can use rsi/rdi as temporaries
284 // then a save/restore here will DESTROY the copy the caller
285 // saved! There used to be a save_bcp() that only happened in
286 // the ASSERT path (no restore_bcp). Which caused bizarre failures
287 // when jvm built with ASSERTs.
288 #ifdef ASSERT
289 {
290 Label L;
291 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
292 jcc(Assembler::equal, L);
293 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
294 " last_sp != null");
295 bind(L);
296 }
297 #endif
298 // super call
299 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
300 // interpreter specific
301 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
302 // but since they may not have been saved (and we don't want to
303 // save them here (see note above) the assert is invalid.
304 }
305
306 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
307 Register java_thread,
308 Register last_java_sp,
309 address entry_point,
310 int number_of_arguments,
311 bool check_exceptions) {
312 // interpreter specific
313 //
314 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
315 // really make a difference for these runtime calls, since they are
316 // slow anyway. Btw., bcp must be saved/restored since it may change
317 // due to GC.
318 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
319 save_bcp();
320 #ifdef ASSERT
321 {
322 Label L;
323 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
324 jcc(Assembler::equal, L);
325 stop("InterpreterMacroAssembler::call_VM_base:"
326 " last_sp isn't null");
327 bind(L);
328 }
329 #endif /* ASSERT */
330 // super call
331 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
332 entry_point, number_of_arguments,
333 check_exceptions);
334 // interpreter specific
335 restore_bcp();
336 restore_locals();
337 }
338
339 #ifdef _LP64
340 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
341 address entry_point,
342 Register arg_1) {
343 assert(arg_1 == c_rarg1, "");
344 Label resume_pc, not_preempted;
345
346 #ifdef ASSERT
347 {
348 Label L;
349 cmpptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
350 jcc(Assembler::equal, L);
351 stop("Should not have alternate return address set");
352 bind(L);
353 }
354 #endif /* ASSERT */
355
356 // Force freeze slow path.
357 push_cont_fastpath();
358
359 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
360 // Note: call_VM_helper requires last_Java_pc for anchor to be at the top of the stack.
361 lea(rscratch1, resume_pc);
362 push(rscratch1);
363 MacroAssembler::call_VM_helper(oop_result, entry_point, 1, false /*check_exceptions*/);
364 pop(rscratch1);
365
366 pop_cont_fastpath();
367
368 // Check if preempted.
369 movptr(rscratch1, Address(r15_thread, JavaThread::preempt_alternate_return_offset()));
370 cmpptr(rscratch1, NULL_WORD);
371 jccb(Assembler::zero, not_preempted);
372 movptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
373 jmp(rscratch1);
374
375 // In case of preemption, this is where we will resume once we finally acquire the monitor.
376 bind(resume_pc);
377 restore_after_resume(false /* is_native */);
378
379 bind(not_preempted);
380 }
381
382 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
383 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
384 call(rscratch1);
385 if (is_native) {
386 // On resume we need to set up stack as expected.
387 push(dtos);
388 push(ltos);
389 }
390 }
391 #else
392 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
393 address entry_point,
394 Register arg_1) {
395 MacroAssembler::call_VM(oop_result, entry_point, arg_1);
396 }
397 #endif // _LP64
398
399 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
400 if (JvmtiExport::can_pop_frame()) {
401 Label L;
402 // Initiate popframe handling only if it is not already being
403 // processed. If the flag has the popframe_processing bit set, it
404 // means that this code is called *during* popframe handling - we
405 // don't want to reenter.
406 // This method is only called just after the call into the vm in
407 // call_VM_base, so the arg registers are available.
408 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
409 LP64_ONLY(c_rarg0);
410 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
411 testl(pop_cond, JavaThread::popframe_pending_bit);
412 jcc(Assembler::zero, L);
413 testl(pop_cond, JavaThread::popframe_processing_bit);
414 jcc(Assembler::notZero, L);
415 // Call Interpreter::remove_activation_preserving_args_entry() to get the
416 // address of the same-named entrypoint in the generated interpreter code.
417 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
418 jmp(rax);
419 bind(L);
420 NOT_LP64(get_thread(java_thread);)
421 }
422 }
423
424 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
425 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
426 NOT_LP64(get_thread(thread);)
427 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
428 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
429 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
430 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
431 #ifdef _LP64
432 switch (state) {
433 case atos: movptr(rax, oop_addr);
434 movptr(oop_addr, NULL_WORD);
435 interp_verify_oop(rax, state); break;
436 case ltos: movptr(rax, val_addr); break;
437 case btos: // fall through
438 case ztos: // fall through
439 case ctos: // fall through
440 case stos: // fall through
441 case itos: movl(rax, val_addr); break;
442 case ftos: load_float(val_addr); break;
443 case dtos: load_double(val_addr); break;
444 case vtos: /* nothing to do */ break;
445 default : ShouldNotReachHere();
446 }
447 // Clean up tos value in the thread object
448 movl(tos_addr, ilgl);
449 movl(val_addr, NULL_WORD);
450 #else
451 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
452 + in_ByteSize(wordSize));
453 switch (state) {
454 case atos: movptr(rax, oop_addr);
455 movptr(oop_addr, NULL_WORD);
456 interp_verify_oop(rax, state); break;
457 case ltos:
458 movl(rdx, val_addr1); // fall through
459 case btos: // fall through
460 case ztos: // fall through
461 case ctos: // fall through
462 case stos: // fall through
463 case itos: movl(rax, val_addr); break;
464 case ftos: load_float(val_addr); break;
465 case dtos: load_double(val_addr); break;
466 case vtos: /* nothing to do */ break;
467 default : ShouldNotReachHere();
468 }
469 #endif // _LP64
470 // Clean up tos value in the thread object
471 movl(tos_addr, ilgl);
472 movptr(val_addr, NULL_WORD);
473 NOT_LP64(movptr(val_addr1, NULL_WORD);)
474 }
475
476
477 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
478 if (JvmtiExport::can_force_early_return()) {
479 Label L;
480 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
481 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
482
483 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
484 testptr(tmp, tmp);
485 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit;
486
487 // Initiate earlyret handling only if it is not already being processed.
488 // If the flag has the earlyret_processing bit set, it means that this code
489 // is called *during* earlyret handling - we don't want to reenter.
490 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
491 cmpl(tmp, JvmtiThreadState::earlyret_pending);
492 jcc(Assembler::notEqual, L);
493
494 // Call Interpreter::remove_activation_early_entry() to get the address of the
495 // same-named entrypoint in the generated interpreter code.
496 NOT_LP64(get_thread(java_thread);)
497 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
498 #ifdef _LP64
499 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
500 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
501 #else
502 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
503 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
504 #endif // _LP64
505 jmp(rax);
506 bind(L);
507 NOT_LP64(get_thread(java_thread);)
508 }
509 }
510
511 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
512 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
513 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
514 bswapl(reg);
515 shrl(reg, 16);
516 }
517
518 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
519 int bcp_offset,
520 size_t index_size) {
521 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
522 if (index_size == sizeof(u2)) {
523 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
524 } else if (index_size == sizeof(u4)) {
525 movl(index, Address(_bcp_register, bcp_offset));
526 } else if (index_size == sizeof(u1)) {
527 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
528 } else {
529 ShouldNotReachHere();
530 }
531 }
532
533 // Load object from cpool->resolved_references(index)
534 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
535 Register index,
536 Register tmp) {
537 assert_different_registers(result, index);
538
539 get_constant_pool(result);
540 // load pointer for resolved_references[] objArray
541 movptr(result, Address(result, ConstantPool::cache_offset()));
542 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
543 resolve_oop_handle(result, tmp);
544 load_heap_oop(result, Address(result, index,
545 UseCompressedOops ? Address::times_4 : Address::times_ptr,
546 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
547 }
548
549 // load cpool->resolved_klass_at(index)
550 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
551 Register cpool,
552 Register index) {
553 assert_different_registers(cpool, index);
554
555 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
556 Register resolved_klasses = cpool;
557 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset()));
558 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
559 }
560
561 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
562 // subtype of super_klass.
563 //
564 // Args:
565 // rax: superklass
566 // Rsub_klass: subklass
567 //
568 // Kills:
569 // rcx, rdi
570 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
571 Label& ok_is_subtype) {
572 assert(Rsub_klass != rax, "rax holds superklass");
573 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
574 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
575 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
576 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
577
578 // Profile the not-null value's klass.
579 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
580
581 // Do the check.
582 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
583 }
584
585
586 #ifndef _LP64
587 void InterpreterMacroAssembler::f2ieee() {
588 if (IEEEPrecision) {
589 fstp_s(Address(rsp, 0));
590 fld_s(Address(rsp, 0));
591 }
592 }
593
594
595 void InterpreterMacroAssembler::d2ieee() {
596 if (IEEEPrecision) {
597 fstp_d(Address(rsp, 0));
598 fld_d(Address(rsp, 0));
599 }
600 }
601 #endif // _LP64
602
603 // Java Expression Stack
604
605 void InterpreterMacroAssembler::pop_ptr(Register r) {
606 pop(r);
607 }
608
609 void InterpreterMacroAssembler::push_ptr(Register r) {
610 push(r);
611 }
612
613 void InterpreterMacroAssembler::push_i(Register r) {
614 push(r);
615 }
616
617 void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
618 push(r);
619 }
620
621 void InterpreterMacroAssembler::push_f(XMMRegister r) {
622 subptr(rsp, wordSize);
623 movflt(Address(rsp, 0), r);
624 }
625
626 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
627 movflt(r, Address(rsp, 0));
628 addptr(rsp, wordSize);
629 }
630
631 void InterpreterMacroAssembler::push_d(XMMRegister r) {
632 subptr(rsp, 2 * wordSize);
633 movdbl(Address(rsp, 0), r);
634 }
635
636 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
637 movdbl(r, Address(rsp, 0));
638 addptr(rsp, 2 * Interpreter::stackElementSize);
639 }
640
641 #ifdef _LP64
642 void InterpreterMacroAssembler::pop_i(Register r) {
643 // XXX can't use pop currently, upper half non clean
644 movl(r, Address(rsp, 0));
645 addptr(rsp, wordSize);
646 }
647
648 void InterpreterMacroAssembler::pop_l(Register r) {
649 movq(r, Address(rsp, 0));
650 addptr(rsp, 2 * Interpreter::stackElementSize);
651 }
652
653 void InterpreterMacroAssembler::push_l(Register r) {
654 subptr(rsp, 2 * wordSize);
655 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
656 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
657 }
658
659 void InterpreterMacroAssembler::pop(TosState state) {
660 switch (state) {
661 case atos: pop_ptr(); break;
662 case btos:
663 case ztos:
664 case ctos:
665 case stos:
666 case itos: pop_i(); break;
667 case ltos: pop_l(); break;
668 case ftos: pop_f(xmm0); break;
669 case dtos: pop_d(xmm0); break;
670 case vtos: /* nothing to do */ break;
671 default: ShouldNotReachHere();
672 }
673 interp_verify_oop(rax, state);
674 }
675
676 void InterpreterMacroAssembler::push(TosState state) {
677 interp_verify_oop(rax, state);
678 switch (state) {
679 case atos: push_ptr(); break;
680 case btos:
681 case ztos:
682 case ctos:
683 case stos:
684 case itos: push_i(); break;
685 case ltos: push_l(); break;
686 case ftos: push_f(xmm0); break;
687 case dtos: push_d(xmm0); break;
688 case vtos: /* nothing to do */ break;
689 default : ShouldNotReachHere();
690 }
691 }
692 #else
693 void InterpreterMacroAssembler::pop_i(Register r) {
694 pop(r);
695 }
696
697 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
698 pop(lo);
699 pop(hi);
700 }
701
702 void InterpreterMacroAssembler::pop_f() {
703 fld_s(Address(rsp, 0));
704 addptr(rsp, 1 * wordSize);
705 }
706
707 void InterpreterMacroAssembler::pop_d() {
708 fld_d(Address(rsp, 0));
709 addptr(rsp, 2 * wordSize);
710 }
711
712
713 void InterpreterMacroAssembler::pop(TosState state) {
714 switch (state) {
715 case atos: pop_ptr(rax); break;
716 case btos: // fall through
717 case ztos: // fall through
718 case ctos: // fall through
719 case stos: // fall through
720 case itos: pop_i(rax); break;
721 case ltos: pop_l(rax, rdx); break;
722 case ftos:
723 if (UseSSE >= 1) {
724 pop_f(xmm0);
725 } else {
726 pop_f();
727 }
728 break;
729 case dtos:
730 if (UseSSE >= 2) {
731 pop_d(xmm0);
732 } else {
733 pop_d();
734 }
735 break;
736 case vtos: /* nothing to do */ break;
737 default : ShouldNotReachHere();
738 }
739 interp_verify_oop(rax, state);
740 }
741
742
743 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
744 push(hi);
745 push(lo);
746 }
747
748 void InterpreterMacroAssembler::push_f() {
749 // Do not schedule for no AGI! Never write beyond rsp!
750 subptr(rsp, 1 * wordSize);
751 fstp_s(Address(rsp, 0));
752 }
753
754 void InterpreterMacroAssembler::push_d() {
755 // Do not schedule for no AGI! Never write beyond rsp!
756 subptr(rsp, 2 * wordSize);
757 fstp_d(Address(rsp, 0));
758 }
759
760
761 void InterpreterMacroAssembler::push(TosState state) {
762 interp_verify_oop(rax, state);
763 switch (state) {
764 case atos: push_ptr(rax); break;
765 case btos: // fall through
766 case ztos: // fall through
767 case ctos: // fall through
768 case stos: // fall through
769 case itos: push_i(rax); break;
770 case ltos: push_l(rax, rdx); break;
771 case ftos:
772 if (UseSSE >= 1) {
773 push_f(xmm0);
774 } else {
775 push_f();
776 }
777 break;
778 case dtos:
779 if (UseSSE >= 2) {
780 push_d(xmm0);
781 } else {
782 push_d();
783 }
784 break;
785 case vtos: /* nothing to do */ break;
786 default : ShouldNotReachHere();
787 }
788 }
789 #endif // _LP64
790
791
792 // Helpers for swap and dup
793 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
794 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
795 }
796
797 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
798 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
799 }
800
801
802 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
803 // set sender sp
804 lea(_bcp_register, Address(rsp, wordSize));
805 // record last_sp
806 mov(rcx, _bcp_register);
807 subptr(rcx, rbp);
808 sarptr(rcx, LogBytesPerWord);
809 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx);
810 }
811
812
813 // Jump to from_interpreted entry of a call unless single stepping is possible
814 // in this thread in which case we must call the i2i entry
815 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
816 prepare_to_jump_from_interpreted();
817
818 if (JvmtiExport::can_post_interpreter_events()) {
819 Label run_compiled_code;
820 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
821 // compiled code in threads for which the event is enabled. Check here for
822 // interp_only_mode if these events CAN be enabled.
823 // interp_only is an int, on little endian it is sufficient to test the byte only
824 // Is a cmpl faster?
825 LP64_ONLY(temp = r15_thread;)
826 NOT_LP64(get_thread(temp);)
827 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
828 jccb(Assembler::zero, run_compiled_code);
829 jmp(Address(method, Method::interpreter_entry_offset()));
830 bind(run_compiled_code);
831 }
832
833 jmp(Address(method, Method::from_interpreted_offset()));
834 }
835
836 // The following two routines provide a hook so that an implementation
837 // can schedule the dispatch in two parts. x86 does not do this.
838 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
839 // Nothing x86 specific to be done here
840 }
841
842 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
843 dispatch_next(state, step);
844 }
845
846 void InterpreterMacroAssembler::dispatch_base(TosState state,
847 address* table,
848 bool verifyoop,
849 bool generate_poll) {
850 verify_FPU(1, state);
851 if (VerifyActivationFrameSize) {
852 Label L;
853 mov(rcx, rbp);
854 subptr(rcx, rsp);
855 int32_t min_frame_size =
856 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
857 wordSize;
858 cmpptr(rcx, min_frame_size);
859 jcc(Assembler::greaterEqual, L);
860 stop("broken stack frame");
861 bind(L);
862 }
863 if (verifyoop) {
864 interp_verify_oop(rax, state);
865 }
866
867 address* const safepoint_table = Interpreter::safept_table(state);
868 #ifdef _LP64
869 Label no_safepoint, dispatch;
870 if (table != safepoint_table && generate_poll) {
871 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
872 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
873
874 jccb(Assembler::zero, no_safepoint);
875 lea(rscratch1, ExternalAddress((address)safepoint_table));
876 jmpb(dispatch);
877 }
878
879 bind(no_safepoint);
880 lea(rscratch1, ExternalAddress((address)table));
881 bind(dispatch);
882 jmp(Address(rscratch1, rbx, Address::times_8));
883
884 #else
885 Address index(noreg, rbx, Address::times_ptr);
886 if (table != safepoint_table && generate_poll) {
887 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
888 Label no_safepoint;
889 const Register thread = rcx;
890 get_thread(thread);
891 testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
892
893 jccb(Assembler::zero, no_safepoint);
894 ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
895 jump(dispatch_addr, noreg);
896 bind(no_safepoint);
897 }
898
899 {
900 ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
901 jump(dispatch_addr, noreg);
902 }
903 #endif // _LP64
904 }
905
906 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
907 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
908 }
909
910 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
911 dispatch_base(state, Interpreter::normal_table(state));
912 }
913
914 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
915 dispatch_base(state, Interpreter::normal_table(state), false);
916 }
917
918
919 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
920 // load next bytecode (load before advancing _bcp_register to prevent AGI)
921 load_unsigned_byte(rbx, Address(_bcp_register, step));
922 // advance _bcp_register
923 increment(_bcp_register, step);
924 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
925 }
926
927 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
928 // load current bytecode
929 load_unsigned_byte(rbx, Address(_bcp_register, 0));
930 dispatch_base(state, table);
931 }
932
933 void InterpreterMacroAssembler::narrow(Register result) {
934
935 // Get method->_constMethod->_result_type
936 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
937 movptr(rcx, Address(rcx, Method::const_offset()));
938 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
939
940 Label done, notBool, notByte, notChar;
941
942 // common case first
943 cmpl(rcx, T_INT);
944 jcc(Assembler::equal, done);
945
946 // mask integer result to narrower return type.
947 cmpl(rcx, T_BOOLEAN);
948 jcc(Assembler::notEqual, notBool);
949 andl(result, 0x1);
950 jmp(done);
951
952 bind(notBool);
953 cmpl(rcx, T_BYTE);
954 jcc(Assembler::notEqual, notByte);
955 LP64_ONLY(movsbl(result, result);)
956 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
957 NOT_LP64(sarl(result, 24);) // and sign-extend byte
958 jmp(done);
959
960 bind(notByte);
961 cmpl(rcx, T_CHAR);
962 jcc(Assembler::notEqual, notChar);
963 LP64_ONLY(movzwl(result, result);)
964 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
965 jmp(done);
966
967 bind(notChar);
968 // cmpl(rcx, T_SHORT); // all that's left
969 // jcc(Assembler::notEqual, done);
970 LP64_ONLY(movswl(result, result);)
971 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
972 NOT_LP64(sarl(result, 16);) // and sign-extend short
973
974 // Nothing to do for T_INT
975 bind(done);
976 }
977
978 // remove activation
979 //
980 // Apply stack watermark barrier.
981 // Unlock the receiver if this is a synchronized method.
982 // Unlock any Java monitors from synchronized blocks.
983 // Remove the activation from the stack.
984 //
985 // If there are locked Java monitors
986 // If throw_monitor_exception
987 // throws IllegalMonitorStateException
988 // Else if install_monitor_exception
989 // installs IllegalMonitorStateException
990 // Else
991 // no error processing
992 void InterpreterMacroAssembler::remove_activation(
993 TosState state,
994 Register ret_addr,
995 bool throw_monitor_exception,
996 bool install_monitor_exception,
997 bool notify_jvmdi) {
998 // Note: Registers rdx xmm0 may be in use for the
999 // result check if synchronized method
1000 Label unlocked, unlock, no_unlock;
1001
1002 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1003 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
1004 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
1005 // monitor pointers need different register
1006 // because rdx may have the result in it
1007 NOT_LP64(get_thread(rthread);)
1008
1009 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
1010 // that would normally not be safe to use. Such bad returns into unsafe territory of
1011 // the stack, will call InterpreterRuntime::at_unwind.
1012 Label slow_path;
1013 Label fast_path;
1014 safepoint_poll(slow_path, rthread, true /* at_return */, false /* in_nmethod */);
1015 jmp(fast_path);
1016 bind(slow_path);
1017 push(state);
1018 set_last_Java_frame(rthread, noreg, rbp, (address)pc(), rscratch1);
1019 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
1020 NOT_LP64(get_thread(rthread);) // call_VM clobbered it, restore
1021 reset_last_Java_frame(rthread, true);
1022 pop(state);
1023 bind(fast_path);
1024
1025 // get the value of _do_not_unlock_if_synchronized into rdx
1026 const Address do_not_unlock_if_synchronized(rthread,
1027 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1028 movbool(rbx, do_not_unlock_if_synchronized);
1029 movbool(do_not_unlock_if_synchronized, false); // reset the flag
1030
1031 // get method access flags
1032 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
1033 movl(rcx, Address(rcx, Method::access_flags_offset()));
1034 testl(rcx, JVM_ACC_SYNCHRONIZED);
1035 jcc(Assembler::zero, unlocked);
1036
1037 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
1038 // is set.
1039 testbool(rbx);
1040 jcc(Assembler::notZero, no_unlock);
1041
1042 // unlock monitor
1043 push(state); // save result
1044
1045 // BasicObjectLock will be first in list, since this is a
1046 // synchronized method. However, need to check that the object has
1047 // not been unlocked by an explicit monitorexit bytecode.
1048 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
1049 wordSize - (int) sizeof(BasicObjectLock));
1050 // We use c_rarg1/rdx so that if we go slow path it will be the correct
1051 // register for unlock_object to pass to VM directly
1052 lea(robj, monitor); // address of first monitor
1053
1054 movptr(rax, Address(robj, BasicObjectLock::obj_offset()));
1055 testptr(rax, rax);
1056 jcc(Assembler::notZero, unlock);
1057
1058 pop(state);
1059 if (throw_monitor_exception) {
1060 // Entry already unlocked, need to throw exception
1061 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
1062 call_VM(noreg, CAST_FROM_FN_PTR(address,
1063 InterpreterRuntime::throw_illegal_monitor_state_exception));
1064 should_not_reach_here();
1065 } else {
1066 // Monitor already unlocked during a stack unroll. If requested,
1067 // install an illegal_monitor_state_exception. Continue with
1068 // stack unrolling.
1069 if (install_monitor_exception) {
1070 NOT_LP64(empty_FPU_stack();)
1071 call_VM(noreg, CAST_FROM_FN_PTR(address,
1072 InterpreterRuntime::new_illegal_monitor_state_exception));
1073 }
1074 jmp(unlocked);
1075 }
1076
1077 bind(unlock);
1078 unlock_object(robj);
1079 pop(state);
1080
1081 // Check that for block-structured locking (i.e., that all locked
1082 // objects has been unlocked)
1083 bind(unlocked);
1084
1085 // rax, rdx: Might contain return value
1086
1087 // Check that all monitors are unlocked
1088 {
1089 Label loop, exception, entry, restart;
1090 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
1091 const Address monitor_block_top(
1092 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1093 const Address monitor_block_bot(
1094 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1095
1096 bind(restart);
1097 // We use c_rarg1 so that if we go slow path it will be the correct
1098 // register for unlock_object to pass to VM directly
1099 movptr(rmon, monitor_block_top); // derelativize pointer
1100 lea(rmon, Address(rbp, rmon, Address::times_ptr));
1101 // c_rarg1 points to current entry, starting with top-most entry
1102
1103 lea(rbx, monitor_block_bot); // points to word before bottom of
1104 // monitor block
1105 jmp(entry);
1106
1107 // Entry already locked, need to throw exception
1108 bind(exception);
1109
1110 if (throw_monitor_exception) {
1111 // Throw exception
1112 NOT_LP64(empty_FPU_stack();)
1113 MacroAssembler::call_VM(noreg,
1114 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1115 throw_illegal_monitor_state_exception));
1116 should_not_reach_here();
1117 } else {
1118 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1119 // Unlock does not block, so don't have to worry about the frame.
1120 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1121
1122 push(state);
1123 mov(robj, rmon); // nop if robj and rmon are the same
1124 unlock_object(robj);
1125 pop(state);
1126
1127 if (install_monitor_exception) {
1128 NOT_LP64(empty_FPU_stack();)
1129 call_VM(noreg, CAST_FROM_FN_PTR(address,
1130 InterpreterRuntime::
1131 new_illegal_monitor_state_exception));
1132 }
1133
1134 jmp(restart);
1135 }
1136
1137 bind(loop);
1138 // check if current entry is used
1139 cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD);
1140 jcc(Assembler::notEqual, exception);
1141
1142 addptr(rmon, entry_size); // otherwise advance to next entry
1143 bind(entry);
1144 cmpptr(rmon, rbx); // check if bottom reached
1145 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1146 }
1147
1148 bind(no_unlock);
1149
1150 // jvmti support
1151 if (notify_jvmdi) {
1152 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1153 } else {
1154 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1155 }
1156
1157 // remove activation
1158 // get sender sp
1159 movptr(rbx,
1160 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1161 if (StackReservedPages > 0) {
1162 // testing if reserved zone needs to be re-enabled
1163 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1164 Label no_reserved_zone_enabling;
1165
1166 NOT_LP64(get_thread(rthread);)
1167
1168 // check if already enabled - if so no re-enabling needed
1169 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
1170 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
1171 jcc(Assembler::equal, no_reserved_zone_enabling);
1172
1173 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1174 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1175
1176 call_VM_leaf(
1177 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1178 call_VM(noreg, CAST_FROM_FN_PTR(address,
1179 InterpreterRuntime::throw_delayed_StackOverflowError));
1180 should_not_reach_here();
1181
1182 bind(no_reserved_zone_enabling);
1183 }
1184 leave(); // remove frame anchor
1185 pop(ret_addr); // get return address
1186 mov(rsp, rbx); // set sp to sender sp
1187 pop_cont_fastpath();
1188 }
1189
1190 void InterpreterMacroAssembler::get_method_counters(Register method,
1191 Register mcs, Label& skip) {
1192 Label has_counters;
1193 movptr(mcs, Address(method, Method::method_counters_offset()));
1194 testptr(mcs, mcs);
1195 jcc(Assembler::notZero, has_counters);
1196 call_VM(noreg, CAST_FROM_FN_PTR(address,
1197 InterpreterRuntime::build_method_counters), method);
1198 movptr(mcs, Address(method,Method::method_counters_offset()));
1199 testptr(mcs, mcs);
1200 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1201 bind(has_counters);
1202 }
1203
1204
1205 // Lock object
1206 //
1207 // Args:
1208 // rdx, c_rarg1: BasicObjectLock to be used for locking
1209 //
1210 // Kills:
1211 // rax, rbx
1212 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1213 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1214 "The argument is only for looks. It must be c_rarg1");
1215
1216 if (LockingMode == LM_MONITOR) {
1217 call_VM_preemptable(noreg,
1218 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1219 lock_reg);
1220 } else {
1221 Label count_locking, done, slow_case;
1222
1223 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1224 const Register tmp_reg = rbx;
1225 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1226 const Register rklass_decode_tmp = rscratch1;
1227
1228 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
1229 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1230 const int mark_offset = lock_offset +
1231 BasicLock::displaced_header_offset_in_bytes();
1232
1233 // Load object pointer into obj_reg
1234 movptr(obj_reg, Address(lock_reg, obj_offset));
1235
1236 if (DiagnoseSyncOnValueBasedClasses != 0) {
1237 load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
1238 testb(Address(tmp_reg, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
1239 jcc(Assembler::notZero, slow_case);
1240 }
1241
1242 if (LockingMode == LM_LIGHTWEIGHT) {
1243 #ifdef _LP64
1244 const Register thread = r15_thread;
1245 lightweight_lock(lock_reg, obj_reg, swap_reg, thread, tmp_reg, slow_case);
1246 #else
1247 // Lacking registers and thread on x86_32. Always take slow path.
1248 jmp(slow_case);
1249 #endif
1250 } else if (LockingMode == LM_LEGACY) {
1251 // Load immediate 1 into swap_reg %rax
1252 movl(swap_reg, 1);
1253
1254 // Load (object->mark() | 1) into swap_reg %rax
1255 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1256
1257 // Save (object->mark() | 1) into BasicLock's displaced header
1258 movptr(Address(lock_reg, mark_offset), swap_reg);
1259
1260 assert(lock_offset == 0,
1261 "displaced header must be first word in BasicObjectLock");
1262
1263 lock();
1264 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1265 jcc(Assembler::zero, count_locking);
1266
1267 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1268
1269 // Fast check for recursive lock.
1270 //
1271 // Can apply the optimization only if this is a stack lock
1272 // allocated in this thread. For efficiency, we can focus on
1273 // recently allocated stack locks (instead of reading the stack
1274 // base and checking whether 'mark' points inside the current
1275 // thread stack):
1276 // 1) (mark & zero_bits) == 0, and
1277 // 2) rsp <= mark < mark + os::pagesize()
1278 //
1279 // Warning: rsp + os::pagesize can overflow the stack base. We must
1280 // neither apply the optimization for an inflated lock allocated
1281 // just above the thread stack (this is why condition 1 matters)
1282 // nor apply the optimization if the stack lock is inside the stack
1283 // of another thread. The latter is avoided even in case of overflow
1284 // because we have guard pages at the end of all stacks. Hence, if
1285 // we go over the stack base and hit the stack of another thread,
1286 // this should not be in a writeable area that could contain a
1287 // stack lock allocated by that thread. As a consequence, a stack
1288 // lock less than page size away from rsp is guaranteed to be
1289 // owned by the current thread.
1290 //
1291 // These 3 tests can be done by evaluating the following
1292 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1293 // assuming both stack pointer and pagesize have their
1294 // least significant bits clear.
1295 // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
1296 subptr(swap_reg, rsp);
1297 andptr(swap_reg, zero_bits - (int)os::vm_page_size());
1298
1299 // Save the test result, for recursive case, the result is zero
1300 movptr(Address(lock_reg, mark_offset), swap_reg);
1301 jcc(Assembler::notZero, slow_case);
1302
1303 bind(count_locking);
1304 inc_held_monitor_count();
1305 }
1306 jmp(done);
1307
1308 bind(slow_case);
1309
1310 // Call the runtime routine for slow case
1311 call_VM_preemptable(noreg,
1312 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1313 lock_reg);
1314 bind(done);
1315 }
1316 }
1317
1318
1319 // Unlocks an object. Used in monitorexit bytecode and
1320 // remove_activation. Throws an IllegalMonitorException if object is
1321 // not locked by current thread.
1322 //
1323 // Args:
1324 // rdx, c_rarg1: BasicObjectLock for lock
1325 //
1326 // Kills:
1327 // rax
1328 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1329 // rscratch1 (scratch reg)
1330 // rax, rbx, rcx, rdx
1331 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1332 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1333 "The argument is only for looks. It must be c_rarg1");
1334
1335 if (LockingMode == LM_MONITOR) {
1336 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1337 } else {
1338 Label count_locking, done, slow_case;
1339
1340 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1341 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1342 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1343
1344 save_bcp(); // Save in case of exception
1345
1346 if (LockingMode != LM_LIGHTWEIGHT) {
1347 // Convert from BasicObjectLock structure to object and BasicLock
1348 // structure Store the BasicLock address into %rax
1349 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
1350 }
1351
1352 // Load oop into obj_reg(%c_rarg3)
1353 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
1354
1355 // Free entry
1356 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD);
1357
1358 if (LockingMode == LM_LIGHTWEIGHT) {
1359 #ifdef _LP64
1360 lightweight_unlock(obj_reg, swap_reg, r15_thread, header_reg, slow_case);
1361 #else
1362 // Lacking registers and thread on x86_32. Always take slow path.
1363 jmp(slow_case);
1364 #endif
1365 } else if (LockingMode == LM_LEGACY) {
1366 // Load the old header from BasicLock structure
1367 movptr(header_reg, Address(swap_reg,
1368 BasicLock::displaced_header_offset_in_bytes()));
1369
1370 // Test for recursion
1371 testptr(header_reg, header_reg);
1372
1373 // zero for recursive case
1374 jcc(Assembler::zero, count_locking);
1375
1376 // Atomic swap back the old header
1377 lock();
1378 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1379
1380 // zero for simple unlock of a stack-lock case
1381 jcc(Assembler::notZero, slow_case);
1382
1383 bind(count_locking);
1384 dec_held_monitor_count();
1385 }
1386 jmp(done);
1387
1388 bind(slow_case);
1389 // Call the runtime routine for slow case.
1390 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj
1391 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1392
1393 bind(done);
1394
1395 restore_bcp();
1396 }
1397 }
1398
1399 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1400 Label& zero_continue) {
1401 assert(ProfileInterpreter, "must be profiling interpreter");
1402 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1403 testptr(mdp, mdp);
1404 jcc(Assembler::zero, zero_continue);
1405 }
1406
1407
1408 // Set the method data pointer for the current bcp.
1409 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1410 assert(ProfileInterpreter, "must be profiling interpreter");
1411 Label set_mdp;
1412 push(rax);
1413 push(rbx);
1414
1415 get_method(rbx);
1416 // Test MDO to avoid the call if it is null.
1417 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1418 testptr(rax, rax);
1419 jcc(Assembler::zero, set_mdp);
1420 // rbx: method
1421 // _bcp_register: bcp
1422 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1423 // rax: mdi
1424 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1425 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1426 addptr(rbx, in_bytes(MethodData::data_offset()));
1427 addptr(rax, rbx);
1428 bind(set_mdp);
1429 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1430 pop(rbx);
1431 pop(rax);
1432 }
1433
1434 void InterpreterMacroAssembler::verify_method_data_pointer() {
1435 assert(ProfileInterpreter, "must be profiling interpreter");
1436 #ifdef ASSERT
1437 Label verify_continue;
1438 push(rax);
1439 push(rbx);
1440 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1441 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1442 push(arg3_reg);
1443 push(arg2_reg);
1444 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1445 get_method(rbx);
1446
1447 // If the mdp is valid, it will point to a DataLayout header which is
1448 // consistent with the bcp. The converse is highly probable also.
1449 load_unsigned_short(arg2_reg,
1450 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1451 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1452 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1453 cmpptr(arg2_reg, _bcp_register);
1454 jcc(Assembler::equal, verify_continue);
1455 // rbx: method
1456 // _bcp_register: bcp
1457 // c_rarg3: mdp
1458 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1459 rbx, _bcp_register, arg3_reg);
1460 bind(verify_continue);
1461 pop(arg2_reg);
1462 pop(arg3_reg);
1463 pop(rbx);
1464 pop(rax);
1465 #endif // ASSERT
1466 }
1467
1468
1469 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1470 int constant,
1471 Register value) {
1472 assert(ProfileInterpreter, "must be profiling interpreter");
1473 Address data(mdp_in, constant);
1474 movptr(data, value);
1475 }
1476
1477
1478 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1479 int constant,
1480 bool decrement) {
1481 // Counter address
1482 Address data(mdp_in, constant);
1483
1484 increment_mdp_data_at(data, decrement);
1485 }
1486
1487 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1488 bool decrement) {
1489 assert(ProfileInterpreter, "must be profiling interpreter");
1490 // %%% this does 64bit counters at best it is wasting space
1491 // at worst it is a rare bug when counters overflow
1492
1493 if (decrement) {
1494 // Decrement the register. Set condition codes.
1495 addptr(data, -DataLayout::counter_increment);
1496 // If the decrement causes the counter to overflow, stay negative
1497 Label L;
1498 jcc(Assembler::negative, L);
1499 addptr(data, DataLayout::counter_increment);
1500 bind(L);
1501 } else {
1502 assert(DataLayout::counter_increment == 1,
1503 "flow-free idiom only works with 1");
1504 // Increment the register. Set carry flag.
1505 addptr(data, DataLayout::counter_increment);
1506 // If the increment causes the counter to overflow, pull back by 1.
1507 sbbptr(data, 0);
1508 }
1509 }
1510
1511
1512 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1513 Register reg,
1514 int constant,
1515 bool decrement) {
1516 Address data(mdp_in, reg, Address::times_1, constant);
1517
1518 increment_mdp_data_at(data, decrement);
1519 }
1520
1521 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1522 int flag_byte_constant) {
1523 assert(ProfileInterpreter, "must be profiling interpreter");
1524 int header_offset = in_bytes(DataLayout::flags_offset());
1525 int header_bits = flag_byte_constant;
1526 // Set the flag
1527 orb(Address(mdp_in, header_offset), header_bits);
1528 }
1529
1530
1531
1532 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1533 int offset,
1534 Register value,
1535 Register test_value_out,
1536 Label& not_equal_continue) {
1537 assert(ProfileInterpreter, "must be profiling interpreter");
1538 if (test_value_out == noreg) {
1539 cmpptr(value, Address(mdp_in, offset));
1540 } else {
1541 // Put the test value into a register, so caller can use it:
1542 movptr(test_value_out, Address(mdp_in, offset));
1543 cmpptr(test_value_out, value);
1544 }
1545 jcc(Assembler::notEqual, not_equal_continue);
1546 }
1547
1548
1549 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1550 int offset_of_disp) {
1551 assert(ProfileInterpreter, "must be profiling interpreter");
1552 Address disp_address(mdp_in, offset_of_disp);
1553 addptr(mdp_in, disp_address);
1554 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1555 }
1556
1557
1558 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1559 Register reg,
1560 int offset_of_disp) {
1561 assert(ProfileInterpreter, "must be profiling interpreter");
1562 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1563 addptr(mdp_in, disp_address);
1564 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1565 }
1566
1567
1568 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1569 int constant) {
1570 assert(ProfileInterpreter, "must be profiling interpreter");
1571 addptr(mdp_in, constant);
1572 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1573 }
1574
1575
1576 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1577 assert(ProfileInterpreter, "must be profiling interpreter");
1578 push(return_bci); // save/restore across call_VM
1579 call_VM(noreg,
1580 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1581 return_bci);
1582 pop(return_bci);
1583 }
1584
1585
1586 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1587 Register bumped_count) {
1588 if (ProfileInterpreter) {
1589 Label profile_continue;
1590
1591 // If no method data exists, go to profile_continue.
1592 // Otherwise, assign to mdp
1593 test_method_data_pointer(mdp, profile_continue);
1594
1595 // We are taking a branch. Increment the taken count.
1596 // We inline increment_mdp_data_at to return bumped_count in a register
1597 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1598 Address data(mdp, in_bytes(JumpData::taken_offset()));
1599 movptr(bumped_count, data);
1600 assert(DataLayout::counter_increment == 1,
1601 "flow-free idiom only works with 1");
1602 addptr(bumped_count, DataLayout::counter_increment);
1603 sbbptr(bumped_count, 0);
1604 movptr(data, bumped_count); // Store back out
1605
1606 // The method data pointer needs to be updated to reflect the new target.
1607 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1608 bind(profile_continue);
1609 }
1610 }
1611
1612
1613 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1614 if (ProfileInterpreter) {
1615 Label profile_continue;
1616
1617 // If no method data exists, go to profile_continue.
1618 test_method_data_pointer(mdp, profile_continue);
1619
1620 // We are taking a branch. Increment the not taken count.
1621 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1622
1623 // The method data pointer needs to be updated to correspond to
1624 // the next bytecode
1625 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1626 bind(profile_continue);
1627 }
1628 }
1629
1630 void InterpreterMacroAssembler::profile_call(Register mdp) {
1631 if (ProfileInterpreter) {
1632 Label profile_continue;
1633
1634 // If no method data exists, go to profile_continue.
1635 test_method_data_pointer(mdp, profile_continue);
1636
1637 // We are making a call. Increment the count.
1638 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1639
1640 // The method data pointer needs to be updated to reflect the new target.
1641 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1642 bind(profile_continue);
1643 }
1644 }
1645
1646
1647 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1648 if (ProfileInterpreter) {
1649 Label profile_continue;
1650
1651 // If no method data exists, go to profile_continue.
1652 test_method_data_pointer(mdp, profile_continue);
1653
1654 // We are making a call. Increment the count.
1655 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1656
1657 // The method data pointer needs to be updated to reflect the new target.
1658 update_mdp_by_constant(mdp,
1659 in_bytes(VirtualCallData::
1660 virtual_call_data_size()));
1661 bind(profile_continue);
1662 }
1663 }
1664
1665
1666 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1667 Register mdp,
1668 Register reg2,
1669 bool receiver_can_be_null) {
1670 if (ProfileInterpreter) {
1671 Label profile_continue;
1672
1673 // If no method data exists, go to profile_continue.
1674 test_method_data_pointer(mdp, profile_continue);
1675
1676 Label skip_receiver_profile;
1677 if (receiver_can_be_null) {
1678 Label not_null;
1679 testptr(receiver, receiver);
1680 jccb(Assembler::notZero, not_null);
1681 // We are making a call. Increment the count for null receiver.
1682 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1683 jmp(skip_receiver_profile);
1684 bind(not_null);
1685 }
1686
1687 // Record the receiver type.
1688 record_klass_in_profile(receiver, mdp, reg2, true);
1689 bind(skip_receiver_profile);
1690
1691 // The method data pointer needs to be updated to reflect the new target.
1692 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1693 bind(profile_continue);
1694 }
1695 }
1696
1697 // This routine creates a state machine for updating the multi-row
1698 // type profile at a virtual call site (or other type-sensitive bytecode).
1699 // The machine visits each row (of receiver/count) until the receiver type
1700 // is found, or until it runs out of rows. At the same time, it remembers
1701 // the location of the first empty row. (An empty row records null for its
1702 // receiver, and can be allocated for a newly-observed receiver type.)
1703 // Because there are two degrees of freedom in the state, a simple linear
1704 // search will not work; it must be a decision tree. Hence this helper
1705 // function is recursive, to generate the required tree structured code.
1706 // It's the interpreter, so we are trading off code space for speed.
1707 // See below for example code.
1708 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1709 Register receiver, Register mdp,
1710 Register reg2, int start_row,
1711 Label& done, bool is_virtual_call) {
1712 if (TypeProfileWidth == 0) {
1713 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1714 } else {
1715 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1716 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1717 }
1718 }
1719
1720 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row,
1721 Label& done, int total_rows,
1722 OffsetFunction item_offset_fn,
1723 OffsetFunction item_count_offset_fn) {
1724 int last_row = total_rows - 1;
1725 assert(start_row <= last_row, "must be work left to do");
1726 // Test this row for both the item and for null.
1727 // Take any of three different outcomes:
1728 // 1. found item => increment count and goto done
1729 // 2. found null => keep looking for case 1, maybe allocate this cell
1730 // 3. found something else => keep looking for cases 1 and 2
1731 // Case 3 is handled by a recursive call.
1732 for (int row = start_row; row <= last_row; row++) {
1733 Label next_test;
1734 bool test_for_null_also = (row == start_row);
1735
1736 // See if the item is item[n].
1737 int item_offset = in_bytes(item_offset_fn(row));
1738 test_mdp_data_at(mdp, item_offset, item,
1739 (test_for_null_also ? reg2 : noreg),
1740 next_test);
1741 // (Reg2 now contains the item from the CallData.)
1742
1743 // The item is item[n]. Increment count[n].
1744 int count_offset = in_bytes(item_count_offset_fn(row));
1745 increment_mdp_data_at(mdp, count_offset);
1746 jmp(done);
1747 bind(next_test);
1748
1749 if (test_for_null_also) {
1750 // Failed the equality check on item[n]... Test for null.
1751 testptr(reg2, reg2);
1752 if (start_row == last_row) {
1753 // The only thing left to do is handle the null case.
1754 Label found_null;
1755 jccb(Assembler::zero, found_null);
1756 // Item did not match any saved item and there is no empty row for it.
1757 // Increment total counter to indicate polymorphic case.
1758 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1759 jmp(done);
1760 bind(found_null);
1761 break;
1762 }
1763 Label found_null;
1764 // Since null is rare, make it be the branch-taken case.
1765 jcc(Assembler::zero, found_null);
1766
1767 // Put all the "Case 3" tests here.
1768 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1769 item_offset_fn, item_count_offset_fn);
1770
1771 // Found a null. Keep searching for a matching item,
1772 // but remember that this is an empty (unused) slot.
1773 bind(found_null);
1774 }
1775 }
1776
1777 // In the fall-through case, we found no matching item, but we
1778 // observed the item[start_row] is null.
1779
1780 // Fill in the item field and increment the count.
1781 int item_offset = in_bytes(item_offset_fn(start_row));
1782 set_mdp_data_at(mdp, item_offset, item);
1783 int count_offset = in_bytes(item_count_offset_fn(start_row));
1784 movl(reg2, DataLayout::counter_increment);
1785 set_mdp_data_at(mdp, count_offset, reg2);
1786 if (start_row > 0) {
1787 jmp(done);
1788 }
1789 }
1790
1791 // Example state machine code for three profile rows:
1792 // // main copy of decision tree, rooted at row[1]
1793 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1794 // if (row[0].rec != nullptr) {
1795 // // inner copy of decision tree, rooted at row[1]
1796 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1797 // if (row[1].rec != nullptr) {
1798 // // degenerate decision tree, rooted at row[2]
1799 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1800 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1801 // row[2].init(rec); goto done;
1802 // } else {
1803 // // remember row[1] is empty
1804 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1805 // row[1].init(rec); goto done;
1806 // }
1807 // } else {
1808 // // remember row[0] is empty
1809 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1810 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1811 // row[0].init(rec); goto done;
1812 // }
1813 // done:
1814
1815 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1816 Register mdp, Register reg2,
1817 bool is_virtual_call) {
1818 assert(ProfileInterpreter, "must be profiling");
1819 Label done;
1820
1821 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1822
1823 bind (done);
1824 }
1825
1826 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1827 Register mdp) {
1828 if (ProfileInterpreter) {
1829 Label profile_continue;
1830 uint row;
1831
1832 // If no method data exists, go to profile_continue.
1833 test_method_data_pointer(mdp, profile_continue);
1834
1835 // Update the total ret count.
1836 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1837
1838 for (row = 0; row < RetData::row_limit(); row++) {
1839 Label next_test;
1840
1841 // See if return_bci is equal to bci[n]:
1842 test_mdp_data_at(mdp,
1843 in_bytes(RetData::bci_offset(row)),
1844 return_bci, noreg,
1845 next_test);
1846
1847 // return_bci is equal to bci[n]. Increment the count.
1848 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1849
1850 // The method data pointer needs to be updated to reflect the new target.
1851 update_mdp_by_offset(mdp,
1852 in_bytes(RetData::bci_displacement_offset(row)));
1853 jmp(profile_continue);
1854 bind(next_test);
1855 }
1856
1857 update_mdp_for_ret(return_bci);
1858
1859 bind(profile_continue);
1860 }
1861 }
1862
1863
1864 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1865 if (ProfileInterpreter) {
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 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1872
1873 // The method data pointer needs to be updated.
1874 int mdp_delta = in_bytes(BitData::bit_data_size());
1875 if (TypeProfileCasts) {
1876 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1877 }
1878 update_mdp_by_constant(mdp, mdp_delta);
1879
1880 bind(profile_continue);
1881 }
1882 }
1883
1884
1885 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1886 if (ProfileInterpreter) {
1887 Label profile_continue;
1888
1889 // If no method data exists, go to profile_continue.
1890 test_method_data_pointer(mdp, profile_continue);
1891
1892 // The method data pointer needs to be updated.
1893 int mdp_delta = in_bytes(BitData::bit_data_size());
1894 if (TypeProfileCasts) {
1895 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1896
1897 // Record the object type.
1898 record_klass_in_profile(klass, mdp, reg2, false);
1899 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1900 NOT_LP64(restore_locals();) // Restore EDI
1901 }
1902 update_mdp_by_constant(mdp, mdp_delta);
1903
1904 bind(profile_continue);
1905 }
1906 }
1907
1908
1909 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1910 if (ProfileInterpreter) {
1911 Label profile_continue;
1912
1913 // If no method data exists, go to profile_continue.
1914 test_method_data_pointer(mdp, profile_continue);
1915
1916 // Update the default case count
1917 increment_mdp_data_at(mdp,
1918 in_bytes(MultiBranchData::default_count_offset()));
1919
1920 // The method data pointer needs to be updated.
1921 update_mdp_by_offset(mdp,
1922 in_bytes(MultiBranchData::
1923 default_displacement_offset()));
1924
1925 bind(profile_continue);
1926 }
1927 }
1928
1929
1930 void InterpreterMacroAssembler::profile_switch_case(Register index,
1931 Register mdp,
1932 Register reg2) {
1933 if (ProfileInterpreter) {
1934 Label profile_continue;
1935
1936 // If no method data exists, go to profile_continue.
1937 test_method_data_pointer(mdp, profile_continue);
1938
1939 // Build the base (index * per_case_size_in_bytes()) +
1940 // case_array_offset_in_bytes()
1941 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1942 imulptr(index, reg2); // XXX l ?
1943 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1944
1945 // Update the case count
1946 increment_mdp_data_at(mdp,
1947 index,
1948 in_bytes(MultiBranchData::relative_count_offset()));
1949
1950 // The method data pointer needs to be updated.
1951 update_mdp_by_offset(mdp,
1952 index,
1953 in_bytes(MultiBranchData::
1954 relative_displacement_offset()));
1955
1956 bind(profile_continue);
1957 }
1958 }
1959
1960
1961
1962 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1963 if (state == atos) {
1964 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1965 }
1966 }
1967
1968 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1969 #ifndef _LP64
1970 if ((state == ftos && UseSSE < 1) ||
1971 (state == dtos && UseSSE < 2)) {
1972 MacroAssembler::verify_FPU(stack_depth);
1973 }
1974 #endif
1975 }
1976
1977 // Jump if ((*counter_addr += increment) & mask) == 0
1978 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
1979 Register scratch, Label* where) {
1980 // This update is actually not atomic and can lose a number of updates
1981 // under heavy contention, but the alternative of using the (contended)
1982 // atomic update here penalizes profiling paths too much.
1983 movl(scratch, counter_addr);
1984 incrementl(scratch, InvocationCounter::count_increment);
1985 movl(counter_addr, scratch);
1986 andl(scratch, mask);
1987 if (where != nullptr) {
1988 jcc(Assembler::zero, *where);
1989 }
1990 }
1991
1992 void InterpreterMacroAssembler::notify_method_entry() {
1993 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1994 // track stack depth. If it is possible to enter interp_only_mode we add
1995 // the code to check if the event should be sent.
1996 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1997 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1998 if (JvmtiExport::can_post_interpreter_events()) {
1999 Label L;
2000 NOT_LP64(get_thread(rthread);)
2001 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2002 testl(rdx, rdx);
2003 jcc(Assembler::zero, L);
2004 call_VM(noreg, CAST_FROM_FN_PTR(address,
2005 InterpreterRuntime::post_method_entry));
2006 bind(L);
2007 }
2008
2009 if (DTraceMethodProbes) {
2010 NOT_LP64(get_thread(rthread);)
2011 get_method(rarg);
2012 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2013 rthread, rarg);
2014 }
2015
2016 // RedefineClasses() tracing support for obsolete method entry
2017 if (log_is_enabled(Trace, redefine, class, obsolete)) {
2018 NOT_LP64(get_thread(rthread);)
2019 get_method(rarg);
2020 call_VM_leaf(
2021 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2022 rthread, rarg);
2023 }
2024 }
2025
2026
2027 void InterpreterMacroAssembler::notify_method_exit(
2028 TosState state, NotifyMethodExitMode mode) {
2029 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2030 // track stack depth. If it is possible to enter interp_only_mode we add
2031 // the code to check if the event should be sent.
2032 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2033 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2034 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2035 Label L;
2036 // Note: frame::interpreter_frame_result has a dependency on how the
2037 // method result is saved across the call to post_method_exit. If this
2038 // is changed then the interpreter_frame_result implementation will
2039 // need to be updated too.
2040
2041 // template interpreter will leave the result on the top of the stack.
2042 push(state);
2043 NOT_LP64(get_thread(rthread);)
2044 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2045 testl(rdx, rdx);
2046 jcc(Assembler::zero, L);
2047 call_VM(noreg,
2048 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2049 bind(L);
2050 pop(state);
2051 }
2052
2053 if (DTraceMethodProbes) {
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 }
2062
2063 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
2064 // Get index out of bytecode pointer
2065 get_cache_index_at_bcp(index, 1, sizeof(u4));
2066 // Get address of invokedynamic array
2067 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2068 movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
2069 if (is_power_of_2(sizeof(ResolvedIndyEntry))) {
2070 shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2
2071 } else {
2072 imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
2073 }
2074 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes()));
2075 }
2076
2077 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
2078 // Get index out of bytecode pointer
2079 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2080 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2081
2082 movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset()));
2083 // Take shortcut if the size is a power of 2
2084 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
2085 shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
2086 } else {
2087 imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
2088 }
2089 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes()));
2090 }
2091
2092 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
2093 // Get index out of bytecode pointer
2094 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2095 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2096
2097 movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset()));
2098 imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
2099 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes()));
2100 }