1 /*
2 * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "interp_masm_x86.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "oops/arrayOop.hpp"
30 #include "oops/markOop.hpp"
31 #include "oops/methodData.hpp"
32 #include "oops/method.hpp"
33 #include "prims/jvmtiExport.hpp"
34 #include "prims/jvmtiRedefineClassesTrace.hpp"
35 #include "prims/jvmtiThreadState.hpp"
36 #include "runtime/basicLock.hpp"
37 #include "runtime/biasedLocking.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/thread.inline.hpp"
40
41 // Implementation of InterpreterMacroAssembler
42
43 void InterpreterMacroAssembler::jump_to_entry(address entry) {
44 assert(entry, "Entry must have been generated by now");
45 jump(RuntimeAddress(entry));
46 }
47
48 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
49 Label update, next, none;
50
51 verify_oop(obj);
52
53 testptr(obj, obj);
54 jccb(Assembler::notZero, update);
55 orptr(mdo_addr, TypeEntries::null_seen);
56 jmpb(next);
57
58 bind(update);
59 load_klass(obj, obj);
60
61 xorptr(obj, mdo_addr);
62 testptr(obj, TypeEntries::type_klass_mask);
63 jccb(Assembler::zero, next); // klass seen before, nothing to
64 // do. The unknown bit may have been
65 // set already but no need to check.
66
67 testptr(obj, TypeEntries::type_unknown);
68 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
69
70 cmpptr(mdo_addr, 0);
71 jccb(Assembler::equal, none);
72 cmpptr(mdo_addr, TypeEntries::null_seen);
73 jccb(Assembler::equal, none);
74 // There is a chance that the checks above (re-reading profiling
75 // data from memory) fail if another thread has just set the
76 // profiling to this obj's klass
77 xorptr(obj, mdo_addr);
78 testptr(obj, TypeEntries::type_klass_mask);
79 jccb(Assembler::zero, next);
80
81 // different than before. Cannot keep accurate profile.
82 orptr(mdo_addr, TypeEntries::type_unknown);
83 jmpb(next);
84
85 bind(none);
86 // first time here. Set profile type.
87 movptr(mdo_addr, obj);
88
89 bind(next);
90 }
91
92 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
93 if (!ProfileInterpreter) {
94 return;
95 }
96
97 if (MethodData::profile_arguments() || MethodData::profile_return()) {
98 Label profile_continue;
99
100 test_method_data_pointer(mdp, profile_continue);
101
102 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
103
104 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
105 jcc(Assembler::notEqual, profile_continue);
106
107 if (MethodData::profile_arguments()) {
108 Label done;
109 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
110 addptr(mdp, off_to_args);
111
112 for (int i = 0; i < TypeProfileArgsLimit; i++) {
113 if (i > 0 || MethodData::profile_return()) {
114 // If return value type is profiled we may have no argument to profile
115 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
116 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
117 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
118 jcc(Assembler::less, done);
119 }
120 movptr(tmp, Address(callee, Method::const_offset()));
121 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
122 // stack offset o (zero based) from the start of the argument
123 // list, for n arguments translates into offset n - o - 1 from
124 // the end of the argument list
125 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
126 subl(tmp, 1);
127 Address arg_addr = argument_address(tmp);
128 movptr(tmp, arg_addr);
129
130 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
131 profile_obj_type(tmp, mdo_arg_addr);
132
133 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
134 addptr(mdp, to_add);
135 off_to_args += to_add;
136 }
137
138 if (MethodData::profile_return()) {
139 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
140 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
141 }
142
143 bind(done);
144
145 if (MethodData::profile_return()) {
146 // We're right after the type profile for the last
147 // argument. tmp is the number of cells left in the
148 // CallTypeData/VirtualCallTypeData to reach its end. Non null
149 // if there's a return to profile.
150 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
151 shll(tmp, exact_log2(DataLayout::cell_size));
152 addptr(mdp, tmp);
153 }
154 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
155 } else {
156 assert(MethodData::profile_return(), "either profile call args or call ret");
157 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
158 }
159
160 // mdp points right after the end of the
161 // CallTypeData/VirtualCallTypeData, right after the cells for the
162 // return value type if there's one
163
164 bind(profile_continue);
165 }
166 }
167
168 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
169 assert_different_registers(mdp, ret, tmp, _bcp_register);
170 if (ProfileInterpreter && MethodData::profile_return()) {
171 Label profile_continue, done;
172
173 test_method_data_pointer(mdp, profile_continue);
174
175 if (MethodData::profile_return_jsr292_only()) {
176 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
177
178 // If we don't profile all invoke bytecodes we must make sure
179 // it's a bytecode we indeed profile. We can't go back to the
180 // begining of the ProfileData we intend to update to check its
181 // type because we're right after it and we don't known its
182 // length
183 Label do_profile;
184 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
185 jcc(Assembler::equal, do_profile);
186 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
187 jcc(Assembler::equal, do_profile);
188 get_method(tmp);
189 cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm);
190 jcc(Assembler::notEqual, profile_continue);
191
192 bind(do_profile);
193 }
194
195 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
196 mov(tmp, ret);
197 profile_obj_type(tmp, mdo_ret_addr);
198
199 bind(profile_continue);
200 }
201 }
202
203 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
204 if (ProfileInterpreter && MethodData::profile_parameters()) {
205 Label profile_continue, done;
206
207 test_method_data_pointer(mdp, profile_continue);
208
209 // Load the offset of the area within the MDO used for
210 // parameters. If it's negative we're not profiling any parameters
211 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
212 testl(tmp1, tmp1);
213 jcc(Assembler::negative, profile_continue);
214
215 // Compute a pointer to the area for parameters from the offset
216 // and move the pointer to the slot for the last
217 // parameters. Collect profiling from last parameter down.
218 // mdo start + parameters offset + array length - 1
219 addptr(mdp, tmp1);
220 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
221 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
222
223 Label loop;
224 bind(loop);
225
226 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
227 int type_base = in_bytes(ParametersTypeData::type_offset(0));
228 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
229 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
230 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
231
232 // load offset on the stack from the slot for this parameter
233 movptr(tmp2, arg_off);
234 negptr(tmp2);
235 // read the parameter from the local area
236 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
237
238 // profile the parameter
239 profile_obj_type(tmp2, arg_type);
240
241 // go to next parameter
242 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
243 jcc(Assembler::positive, loop);
244
245 bind(profile_continue);
246 }
247 }
248
249 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
250 int number_of_arguments) {
251 // interpreter specific
252 //
253 // Note: No need to save/restore bcp & locals registers
254 // since these are callee saved registers and no blocking/
255 // GC can happen in leaf calls.
256 // Further Note: DO NOT save/restore bcp/locals. If a caller has
257 // already saved them so that it can use rsi/rdi as temporaries
258 // then a save/restore here will DESTROY the copy the caller
259 // saved! There used to be a save_bcp() that only happened in
260 // the ASSERT path (no restore_bcp). Which caused bizarre failures
261 // when jvm built with ASSERTs.
262 #ifdef ASSERT
263 {
264 Label L;
265 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
266 jcc(Assembler::equal, L);
267 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
268 " last_sp != NULL");
269 bind(L);
270 }
271 #endif
272 // super call
273 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
274 // interpreter specific
275 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
276 // but since they may not have been saved (and we don't want to
277 // save them here (see note above) the assert is invalid.
278 }
279
280 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
281 Register java_thread,
282 Register last_java_sp,
283 address entry_point,
284 int number_of_arguments,
285 bool check_exceptions) {
286 // interpreter specific
287 //
288 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
289 // really make a difference for these runtime calls, since they are
290 // slow anyway. Btw., bcp must be saved/restored since it may change
291 // due to GC.
292 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
293 save_bcp();
294 #ifdef ASSERT
295 {
296 Label L;
297 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
298 jcc(Assembler::equal, L);
299 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
300 " last_sp != NULL");
301 bind(L);
302 }
303 #endif /* ASSERT */
304 // super call
305 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
306 entry_point, number_of_arguments,
307 check_exceptions);
308 // interpreter specific
309 restore_bcp();
310 restore_locals();
311 }
312
313 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
314 if (JvmtiExport::can_pop_frame()) {
315 Label L;
316 // Initiate popframe handling only if it is not already being
317 // processed. If the flag has the popframe_processing bit set, it
318 // means that this code is called *during* popframe handling - we
319 // don't want to reenter.
320 // This method is only called just after the call into the vm in
321 // call_VM_base, so the arg registers are available.
322 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
323 LP64_ONLY(c_rarg0);
324 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
325 testl(pop_cond, JavaThread::popframe_pending_bit);
326 jcc(Assembler::zero, L);
327 testl(pop_cond, JavaThread::popframe_processing_bit);
328 jcc(Assembler::notZero, L);
329 // Call Interpreter::remove_activation_preserving_args_entry() to get the
330 // address of the same-named entrypoint in the generated interpreter code.
331 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
332 jmp(rax);
333 bind(L);
334 NOT_LP64(get_thread(java_thread);)
335 }
336 }
337
338 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
339 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
340 NOT_LP64(get_thread(thread);)
341 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
342 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
343 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
344 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
345 #ifdef _LP64
346 switch (state) {
347 case atos: movptr(rax, oop_addr);
348 movptr(oop_addr, (int32_t)NULL_WORD);
349 verify_oop(rax, state); break;
350 case ltos: movptr(rax, val_addr); break;
351 case btos: // fall through
352 case ctos: // fall through
353 case stos: // fall through
354 case itos: movl(rax, val_addr); break;
355 case ftos: load_float(val_addr); break;
356 case dtos: load_double(val_addr); break;
357 case vtos: /* nothing to do */ break;
358 default : ShouldNotReachHere();
359 }
360 // Clean up tos value in the thread object
361 movl(tos_addr, (int) ilgl);
362 movl(val_addr, (int32_t) NULL_WORD);
363 #else
364 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
365 + in_ByteSize(wordSize));
366 switch (state) {
367 case atos: movptr(rax, oop_addr);
368 movptr(oop_addr, NULL_WORD);
369 verify_oop(rax, state); break;
370 case ltos:
371 movl(rdx, val_addr1); // fall through
372 case btos: // fall through
373 case ctos: // fall through
374 case stos: // fall through
375 case itos: movl(rax, val_addr); break;
376 case ftos: load_float(val_addr); break;
377 case dtos: load_double(val_addr); break;
378 case vtos: /* nothing to do */ break;
379 default : ShouldNotReachHere();
380 }
381 #endif // _LP64
382 // Clean up tos value in the thread object
383 movl(tos_addr, (int32_t) ilgl);
384 movptr(val_addr, NULL_WORD);
385 NOT_LP64(movptr(val_addr1, NULL_WORD);)
386 }
387
388
389 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
390 if (JvmtiExport::can_force_early_return()) {
391 Label L;
392 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
393 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
394
395 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
396 testptr(tmp, tmp);
397 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
398
399 // Initiate earlyret handling only if it is not already being processed.
400 // If the flag has the earlyret_processing bit set, it means that this code
401 // is called *during* earlyret handling - we don't want to reenter.
402 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
403 cmpl(tmp, JvmtiThreadState::earlyret_pending);
404 jcc(Assembler::notEqual, L);
405
406 // Call Interpreter::remove_activation_early_entry() to get the address of the
407 // same-named entrypoint in the generated interpreter code.
408 NOT_LP64(get_thread(java_thread);)
409 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
410 #ifdef _LP64
411 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
412 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
413 #else
414 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
415 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
416 #endif // _LP64
417 jmp(rax);
418 bind(L);
419 NOT_LP64(get_thread(java_thread);)
420 }
421 }
422
423 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
424 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
425 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
426 bswapl(reg);
427 shrl(reg, 16);
428 }
429
430 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
431 int bcp_offset,
432 size_t index_size) {
433 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
434 if (index_size == sizeof(u2)) {
435 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
436 } else if (index_size == sizeof(u4)) {
437 movl(index, Address(_bcp_register, bcp_offset));
438 // Check if the secondary index definition is still ~x, otherwise
439 // we have to change the following assembler code to calculate the
440 // plain index.
441 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
442 notl(index); // convert to plain index
443 } else if (index_size == sizeof(u1)) {
444 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
445 } else {
446 ShouldNotReachHere();
447 }
448 }
449
450 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
451 Register index,
452 int bcp_offset,
453 size_t index_size) {
454 assert_different_registers(cache, index);
455 get_cache_index_at_bcp(index, bcp_offset, index_size);
456 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
457 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
458 // convert from field index to ConstantPoolCacheEntry index
459 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
460 shll(index, 2);
461 }
462
463 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
464 Register index,
465 Register bytecode,
466 int byte_no,
467 int bcp_offset,
468 size_t index_size) {
469 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
470 // We use a 32-bit load here since the layout of 64-bit words on
471 // little-endian machines allow us that.
472 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
473 const int shift_count = (1 + byte_no) * BitsPerByte;
474 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
475 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
476 "correct shift count");
477 shrl(bytecode, shift_count);
478 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
479 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
480 }
481
482 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
483 Register tmp,
484 int bcp_offset,
485 size_t index_size) {
486 assert(cache != tmp, "must use different register");
487 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
488 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
489 // convert from field index to ConstantPoolCacheEntry index
490 // and from word offset to byte offset
491 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
492 shll(tmp, 2 + LogBytesPerWord);
493 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
494 // skip past the header
495 addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
496 addptr(cache, tmp); // construct pointer to cache entry
497 }
498
499 // Load object from cpool->resolved_references(index)
500 void InterpreterMacroAssembler::load_resolved_reference_at_index(
501 Register result, Register index) {
502 assert_different_registers(result, index);
503 // convert from field index to resolved_references() index and from
504 // word index to byte offset. Since this is a java object, it can be compressed
505 Register tmp = index; // reuse
506 shll(tmp, LogBytesPerHeapOop);
507
508 get_constant_pool(result);
509 // load pointer for resolved_references[] objArray
510 movptr(result, Address(result, ConstantPool::resolved_references_offset_in_bytes()));
511 // JNIHandles::resolve(obj);
512 movptr(result, Address(result, 0));
513 // Add in the index
514 addptr(result, tmp);
515 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
516 }
517
518
519 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
520 // subtype of super_klass.
521 //
522 // Args:
523 // rax: superklass
524 // Rsub_klass: subklass
525 //
526 // Kills:
527 // rcx, rdi
528 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
529 Label& ok_is_subtype) {
530 assert(Rsub_klass != rax, "rax holds superklass");
531 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
532 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
533 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
534 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
535
536 // Profile the not-null value's klass.
537 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
538
539 // Do the check.
540 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
541
542 // Profile the failure of the check.
543 profile_typecheck_failed(rcx); // blows rcx
544 }
545
546
547 #ifndef _LP64
548 void InterpreterMacroAssembler::f2ieee() {
549 if (IEEEPrecision) {
550 fstp_s(Address(rsp, 0));
551 fld_s(Address(rsp, 0));
552 }
553 }
554
555
556 void InterpreterMacroAssembler::d2ieee() {
557 if (IEEEPrecision) {
558 fstp_d(Address(rsp, 0));
559 fld_d(Address(rsp, 0));
560 }
561 }
562 #endif // _LP64
563
564 // Java Expression Stack
565
566 void InterpreterMacroAssembler::pop_ptr(Register r) {
567 pop(r);
568 }
569
570 void InterpreterMacroAssembler::push_ptr(Register r) {
571 push(r);
572 }
573
574 void InterpreterMacroAssembler::push_i(Register r) {
575 push(r);
576 }
577
578 void InterpreterMacroAssembler::push_f(XMMRegister r) {
579 subptr(rsp, wordSize);
580 movflt(Address(rsp, 0), r);
581 }
582
583 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
584 movflt(r, Address(rsp, 0));
585 addptr(rsp, wordSize);
586 }
587
588 void InterpreterMacroAssembler::push_d(XMMRegister r) {
589 subptr(rsp, 2 * wordSize);
590 movdbl(Address(rsp, 0), r);
591 }
592
593 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
594 movdbl(r, Address(rsp, 0));
595 addptr(rsp, 2 * Interpreter::stackElementSize);
596 }
597
598 #ifdef _LP64
599 void InterpreterMacroAssembler::pop_i(Register r) {
600 // XXX can't use pop currently, upper half non clean
601 movl(r, Address(rsp, 0));
602 addptr(rsp, wordSize);
603 }
604
605 void InterpreterMacroAssembler::pop_l(Register r) {
606 movq(r, Address(rsp, 0));
607 addptr(rsp, 2 * Interpreter::stackElementSize);
608 }
609
610 void InterpreterMacroAssembler::push_l(Register r) {
611 subptr(rsp, 2 * wordSize);
612 movq(Address(rsp, 0), r);
613 }
614
615 void InterpreterMacroAssembler::pop(TosState state) {
616 switch (state) {
617 case atos: pop_ptr(); break;
618 case btos:
619 case ctos:
620 case stos:
621 case itos: pop_i(); break;
622 case ltos: pop_l(); break;
623 case ftos: pop_f(xmm0); break;
624 case dtos: pop_d(xmm0); break;
625 case vtos: /* nothing to do */ break;
626 default: ShouldNotReachHere();
627 }
628 verify_oop(rax, state);
629 }
630
631 void InterpreterMacroAssembler::push(TosState state) {
632 verify_oop(rax, state);
633 switch (state) {
634 case atos: push_ptr(); break;
635 case btos:
636 case ctos:
637 case stos:
638 case itos: push_i(); break;
639 case ltos: push_l(); break;
640 case ftos: push_f(xmm0); break;
641 case dtos: push_d(xmm0); break;
642 case vtos: /* nothing to do */ break;
643 default : ShouldNotReachHere();
644 }
645 }
646 #else
647 void InterpreterMacroAssembler::pop_i(Register r) {
648 pop(r);
649 }
650
651 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
652 pop(lo);
653 pop(hi);
654 }
655
656 void InterpreterMacroAssembler::pop_f() {
657 fld_s(Address(rsp, 0));
658 addptr(rsp, 1 * wordSize);
659 }
660
661 void InterpreterMacroAssembler::pop_d() {
662 fld_d(Address(rsp, 0));
663 addptr(rsp, 2 * wordSize);
664 }
665
666
667 void InterpreterMacroAssembler::pop(TosState state) {
668 switch (state) {
669 case atos: pop_ptr(rax); break;
670 case btos: // fall through
671 case ctos: // fall through
672 case stos: // fall through
673 case itos: pop_i(rax); break;
674 case ltos: pop_l(rax, rdx); break;
675 case ftos:
676 if (UseSSE >= 1) {
677 pop_f(xmm0);
678 } else {
679 pop_f();
680 }
681 break;
682 case dtos:
683 if (UseSSE >= 2) {
684 pop_d(xmm0);
685 } else {
686 pop_d();
687 }
688 break;
689 case vtos: /* nothing to do */ break;
690 default : ShouldNotReachHere();
691 }
692 verify_oop(rax, state);
693 }
694
695
696 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
697 push(hi);
698 push(lo);
699 }
700
701 void InterpreterMacroAssembler::push_f() {
702 // Do not schedule for no AGI! Never write beyond rsp!
703 subptr(rsp, 1 * wordSize);
704 fstp_s(Address(rsp, 0));
705 }
706
707 void InterpreterMacroAssembler::push_d() {
708 // Do not schedule for no AGI! Never write beyond rsp!
709 subptr(rsp, 2 * wordSize);
710 fstp_d(Address(rsp, 0));
711 }
712
713
714 void InterpreterMacroAssembler::push(TosState state) {
715 verify_oop(rax, state);
716 switch (state) {
717 case atos: push_ptr(rax); break;
718 case btos: // fall through
719 case ctos: // fall through
720 case stos: // fall through
721 case itos: push_i(rax); break;
722 case ltos: push_l(rax, rdx); break;
723 case ftos:
724 if (UseSSE >= 1) {
725 push_f(xmm0);
726 } else {
727 push_f();
728 }
729 break;
730 case dtos:
731 if (UseSSE >= 2) {
732 push_d(xmm0);
733 } else {
734 push_d();
735 }
736 break;
737 case vtos: /* nothing to do */ break;
738 default : ShouldNotReachHere();
739 }
740 }
741 #endif // _LP64
742
743
744 // Helpers for swap and dup
745 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
746 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
747 }
748
749 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
750 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
751 }
752
753
754 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
755 // set sender sp
756 lea(_bcp_register, Address(rsp, wordSize));
757 // record last_sp
758 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
759 }
760
761
762 // Jump to from_interpreted entry of a call unless single stepping is possible
763 // in this thread in which case we must call the i2i entry
764 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
765 prepare_to_jump_from_interpreted();
766
767 if (JvmtiExport::can_post_interpreter_events()) {
768 Label run_compiled_code;
769 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
770 // compiled code in threads for which the event is enabled. Check here for
771 // interp_only_mode if these events CAN be enabled.
772 // interp_only is an int, on little endian it is sufficient to test the byte only
773 // Is a cmpl faster?
774 LP64_ONLY(temp = r15_thread;)
775 NOT_LP64(get_thread(temp);)
776 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
777 jccb(Assembler::zero, run_compiled_code);
778 jmp(Address(method, Method::interpreter_entry_offset()));
779 bind(run_compiled_code);
780 }
781
782 jmp(Address(method, Method::from_interpreted_offset()));
783 }
784
785 // The following two routines provide a hook so that an implementation
786 // can schedule the dispatch in two parts. x86 does not do this.
787 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
788 // Nothing x86 specific to be done here
789 }
790
791 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
792 dispatch_next(state, step);
793 }
794
795 void InterpreterMacroAssembler::dispatch_base(TosState state,
796 address* table,
797 bool verifyoop) {
798 verify_FPU(1, state);
799 if (VerifyActivationFrameSize) {
800 Label L;
801 mov(rcx, rbp);
802 subptr(rcx, rsp);
803 int32_t min_frame_size =
804 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
805 wordSize;
806 cmpptr(rcx, (int32_t)min_frame_size);
807 jcc(Assembler::greaterEqual, L);
808 stop("broken stack frame");
809 bind(L);
810 }
811 if (verifyoop) {
812 verify_oop(rax, state);
813 }
814 #ifdef _LP64
815 lea(rscratch1, ExternalAddress((address)table));
816 jmp(Address(rscratch1, rbx, Address::times_8));
817 #else
818 Address index(noreg, rbx, Address::times_ptr);
819 ExternalAddress tbl((address)table);
820 ArrayAddress dispatch(tbl, index);
821 jump(dispatch);
822 #endif // _LP64
823 }
824
825 void InterpreterMacroAssembler::dispatch_only(TosState state) {
826 dispatch_base(state, Interpreter::dispatch_table(state));
827 }
828
829 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
830 dispatch_base(state, Interpreter::normal_table(state));
831 }
832
833 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
834 dispatch_base(state, Interpreter::normal_table(state), false);
835 }
836
837
838 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
839 // load next bytecode (load before advancing _bcp_register to prevent AGI)
840 load_unsigned_byte(rbx, Address(_bcp_register, step));
841 // advance _bcp_register
842 increment(_bcp_register, step);
843 dispatch_base(state, Interpreter::dispatch_table(state));
844 }
845
846 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
847 // load current bytecode
848 load_unsigned_byte(rbx, Address(_bcp_register, 0));
849 dispatch_base(state, table);
850 }
851
852 // remove activation
853 //
854 // Unlock the receiver if this is a synchronized method.
855 // Unlock any Java monitors from syncronized blocks.
856 // Remove the activation from the stack.
857 //
858 // If there are locked Java monitors
859 // If throw_monitor_exception
860 // throws IllegalMonitorStateException
861 // Else if install_monitor_exception
862 // installs IllegalMonitorStateException
863 // Else
864 // no error processing
865 void InterpreterMacroAssembler::remove_activation(
866 TosState state,
867 Register ret_addr,
868 bool throw_monitor_exception,
869 bool install_monitor_exception,
870 bool notify_jvmdi) {
871 // Note: Registers rdx xmm0 may be in use for the
872 // result check if synchronized method
873 Label unlocked, unlock, no_unlock;
874
875 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
876 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
877 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
878 // monitor pointers need different register
879 // because rdx may have the result in it
880 NOT_LP64(get_thread(rcx);)
881
882 // get the value of _do_not_unlock_if_synchronized into rdx
883 const Address do_not_unlock_if_synchronized(rthread,
884 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
885 movbool(rbx, do_not_unlock_if_synchronized);
886 movbool(do_not_unlock_if_synchronized, false); // reset the flag
887
888 // get method access flags
889 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
890 movl(rcx, Address(rcx, Method::access_flags_offset()));
891 testl(rcx, JVM_ACC_SYNCHRONIZED);
892 jcc(Assembler::zero, unlocked);
893
894 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
895 // is set.
896 testbool(rbx);
897 jcc(Assembler::notZero, no_unlock);
898
899 // unlock monitor
900 push(state); // save result
901
902 // BasicObjectLock will be first in list, since this is a
903 // synchronized method. However, need to check that the object has
904 // not been unlocked by an explicit monitorexit bytecode.
905 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
906 wordSize - (int) sizeof(BasicObjectLock));
907 // We use c_rarg1/rdx so that if we go slow path it will be the correct
908 // register for unlock_object to pass to VM directly
909 lea(robj, monitor); // address of first monitor
910
911 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
912 testptr(rax, rax);
913 jcc(Assembler::notZero, unlock);
914
915 pop(state);
916 if (throw_monitor_exception) {
917 // Entry already unlocked, need to throw exception
918 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
919 call_VM(noreg, CAST_FROM_FN_PTR(address,
920 InterpreterRuntime::throw_illegal_monitor_state_exception));
921 should_not_reach_here();
922 } else {
923 // Monitor already unlocked during a stack unroll. If requested,
924 // install an illegal_monitor_state_exception. Continue with
925 // stack unrolling.
926 if (install_monitor_exception) {
927 NOT_LP64(empty_FPU_stack();)
928 call_VM(noreg, CAST_FROM_FN_PTR(address,
929 InterpreterRuntime::new_illegal_monitor_state_exception));
930 }
931 jmp(unlocked);
932 }
933
934 bind(unlock);
935 unlock_object(robj);
936 pop(state);
937
938 // Check that for block-structured locking (i.e., that all locked
939 // objects has been unlocked)
940 bind(unlocked);
941
942 // rax, rdx: Might contain return value
943
944 // Check that all monitors are unlocked
945 {
946 Label loop, exception, entry, restart;
947 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
948 const Address monitor_block_top(
949 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
950 const Address monitor_block_bot(
951 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
952
953 bind(restart);
954 // We use c_rarg1 so that if we go slow path it will be the correct
955 // register for unlock_object to pass to VM directly
956 movptr(rmon, monitor_block_top); // points to current entry, starting
957 // with top-most entry
958 lea(rbx, monitor_block_bot); // points to word before bottom of
959 // monitor block
960 jmp(entry);
961
962 // Entry already locked, need to throw exception
963 bind(exception);
964
965 if (throw_monitor_exception) {
966 // Throw exception
967 NOT_LP64(empty_FPU_stack();)
968 MacroAssembler::call_VM(noreg,
969 CAST_FROM_FN_PTR(address, InterpreterRuntime::
970 throw_illegal_monitor_state_exception));
971 should_not_reach_here();
972 } else {
973 // Stack unrolling. Unlock object and install illegal_monitor_exception.
974 // Unlock does not block, so don't have to worry about the frame.
975 // We don't have to preserve c_rarg1 since we are going to throw an exception.
976
977 push(state);
978 mov(robj, rmon); // nop if robj and rmon are the same
979 unlock_object(robj);
980 pop(state);
981
982 if (install_monitor_exception) {
983 NOT_LP64(empty_FPU_stack();)
984 call_VM(noreg, CAST_FROM_FN_PTR(address,
985 InterpreterRuntime::
986 new_illegal_monitor_state_exception));
987 }
988
989 jmp(restart);
990 }
991
992 bind(loop);
993 // check if current entry is used
994 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
995 jcc(Assembler::notEqual, exception);
996
997 addptr(rmon, entry_size); // otherwise advance to next entry
998 bind(entry);
999 cmpptr(rmon, rbx); // check if bottom reached
1000 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1001 }
1002
1003 bind(no_unlock);
1004
1005 // jvmti support
1006 if (notify_jvmdi) {
1007 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1008 } else {
1009 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1010 }
1011
1012 // remove activation
1013 // get sender sp
1014 movptr(rbx,
1015 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1016 if (StackReservedPages > 0) {
1017 // testing if reserved zone needs to be re-enabled
1018 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1019 Label no_reserved_zone_enabling;
1020
1021 NOT_LP64(get_thread(rthread);)
1022
1023 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1024 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1025
1026 call_VM_leaf(
1027 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1028 push(rthread);
1029 call_VM(noreg, CAST_FROM_FN_PTR(address,
1030 InterpreterRuntime::throw_delayed_StackOverflowError));
1031 should_not_reach_here();
1032
1033 bind(no_reserved_zone_enabling);
1034 }
1035 leave(); // remove frame anchor
1036 pop(ret_addr); // get return address
1037 mov(rsp, rbx); // set sp to sender sp
1038 }
1039
1040 void InterpreterMacroAssembler::get_method_counters(Register method,
1041 Register mcs, Label& skip) {
1042 Label has_counters;
1043 movptr(mcs, Address(method, Method::method_counters_offset()));
1044 testptr(mcs, mcs);
1045 jcc(Assembler::notZero, has_counters);
1046 call_VM(noreg, CAST_FROM_FN_PTR(address,
1047 InterpreterRuntime::build_method_counters), method);
1048 movptr(mcs, Address(method,Method::method_counters_offset()));
1049 testptr(mcs, mcs);
1050 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1051 bind(has_counters);
1052 }
1053
1054
1055 // Lock object
1056 //
1057 // Args:
1058 // rdx, c_rarg1: BasicObjectLock to be used for locking
1059 //
1060 // Kills:
1061 // rax, rbx
1062 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1063 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1064 "The argument is only for looks. It must be c_rarg1");
1065
1066 if (UseHeavyMonitors) {
1067 call_VM(noreg,
1068 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1069 lock_reg);
1070 } else {
1071 Label done;
1072
1073 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1074 const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a
1075 // problematic case where tmp_reg = no_reg.
1076 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1077
1078 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
1079 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
1080 const int mark_offset = lock_offset +
1081 BasicLock::displaced_header_offset_in_bytes();
1082
1083 Label slow_case;
1084
1085 // Load object pointer into obj_reg
1086 movptr(obj_reg, Address(lock_reg, obj_offset));
1087
1088 if (UseBiasedLocking) {
1089 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case);
1090 }
1091
1092 // Load immediate 1 into swap_reg %rax
1093 movl(swap_reg, (int32_t)1);
1094
1095 // Load (object->mark() | 1) into swap_reg %rax
1096 orptr(swap_reg, Address(obj_reg, 0));
1097
1098 // Save (object->mark() | 1) into BasicLock's displaced header
1099 movptr(Address(lock_reg, mark_offset), swap_reg);
1100
1101 assert(lock_offset == 0,
1102 "displached header must be first word in BasicObjectLock");
1103
1104 if (os::is_MP()) lock();
1105 cmpxchgptr(lock_reg, Address(obj_reg, 0));
1106 if (PrintBiasedLockingStatistics) {
1107 cond_inc32(Assembler::zero,
1108 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1109 }
1110 jcc(Assembler::zero, done);
1111
1112 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1113
1114 // Test if the oopMark is an obvious stack pointer, i.e.,
1115 // 1) (mark & zero_bits) == 0, and
1116 // 2) rsp <= mark < mark + os::pagesize()
1117 //
1118 // These 3 tests can be done by evaluating the following
1119 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1120 // assuming both stack pointer and pagesize have their
1121 // least significant bits clear.
1122 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
1123 subptr(swap_reg, rsp);
1124 andptr(swap_reg, zero_bits - os::vm_page_size());
1125
1126 // Save the test result, for recursive case, the result is zero
1127 movptr(Address(lock_reg, mark_offset), swap_reg);
1128
1129 if (PrintBiasedLockingStatistics) {
1130 cond_inc32(Assembler::zero,
1131 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1132 }
1133 jcc(Assembler::zero, done);
1134
1135 bind(slow_case);
1136
1137 // Call the runtime routine for slow case
1138 call_VM(noreg,
1139 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1140 lock_reg);
1141
1142 bind(done);
1143 }
1144 }
1145
1146
1147 // Unlocks an object. Used in monitorexit bytecode and
1148 // remove_activation. Throws an IllegalMonitorException if object is
1149 // not locked by current thread.
1150 //
1151 // Args:
1152 // rdx, c_rarg1: BasicObjectLock for lock
1153 //
1154 // Kills:
1155 // rax
1156 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1157 // rscratch1, rscratch2 (scratch regs)
1158 // rax, rbx, rcx, rdx
1159 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1160 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1161 "The argument is only for looks. It must be c_rarg1");
1162
1163 if (UseHeavyMonitors) {
1164 call_VM(noreg,
1165 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1166 lock_reg);
1167 } else {
1168 Label done;
1169
1170 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1171 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1172 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1173
1174 save_bcp(); // Save in case of exception
1175
1176 // Convert from BasicObjectLock structure to object and BasicLock
1177 // structure Store the BasicLock address into %rax
1178 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
1179
1180 // Load oop into obj_reg(%c_rarg3)
1181 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
1182
1183 // Free entry
1184 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
1185
1186 if (UseBiasedLocking) {
1187 biased_locking_exit(obj_reg, header_reg, done);
1188 }
1189
1190 // Load the old header from BasicLock structure
1191 movptr(header_reg, Address(swap_reg,
1192 BasicLock::displaced_header_offset_in_bytes()));
1193
1194 // Test for recursion
1195 testptr(header_reg, header_reg);
1196
1197 // zero for recursive case
1198 jcc(Assembler::zero, done);
1199
1200 // Atomic swap back the old header
1201 if (os::is_MP()) lock();
1202 cmpxchgptr(header_reg, Address(obj_reg, 0));
1203
1204 // zero for recursive case
1205 jcc(Assembler::zero, done);
1206
1207 // Call the runtime routine for slow case.
1208 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
1209 obj_reg); // restore obj
1210 call_VM(noreg,
1211 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1212 lock_reg);
1213
1214 bind(done);
1215
1216 restore_bcp();
1217 }
1218 }
1219
1220 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1221 Label& zero_continue) {
1222 assert(ProfileInterpreter, "must be profiling interpreter");
1223 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1224 testptr(mdp, mdp);
1225 jcc(Assembler::zero, zero_continue);
1226 }
1227
1228
1229 // Set the method data pointer for the current bcp.
1230 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1231 assert(ProfileInterpreter, "must be profiling interpreter");
1232 Label set_mdp;
1233 push(rax);
1234 push(rbx);
1235
1236 get_method(rbx);
1237 // Test MDO to avoid the call if it is NULL.
1238 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1239 testptr(rax, rax);
1240 jcc(Assembler::zero, set_mdp);
1241 // rbx: method
1242 // _bcp_register: bcp
1243 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1244 // rax: mdi
1245 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1246 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1247 addptr(rbx, in_bytes(MethodData::data_offset()));
1248 addptr(rax, rbx);
1249 bind(set_mdp);
1250 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1251 pop(rbx);
1252 pop(rax);
1253 }
1254
1255 void InterpreterMacroAssembler::verify_method_data_pointer() {
1256 assert(ProfileInterpreter, "must be profiling interpreter");
1257 #ifdef ASSERT
1258 Label verify_continue;
1259 push(rax);
1260 push(rbx);
1261 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1262 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1263 push(arg3_reg);
1264 push(arg2_reg);
1265 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1266 get_method(rbx);
1267
1268 // If the mdp is valid, it will point to a DataLayout header which is
1269 // consistent with the bcp. The converse is highly probable also.
1270 load_unsigned_short(arg2_reg,
1271 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1272 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1273 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1274 cmpptr(arg2_reg, _bcp_register);
1275 jcc(Assembler::equal, verify_continue);
1276 // rbx: method
1277 // _bcp_register: bcp
1278 // c_rarg3: mdp
1279 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1280 rbx, _bcp_register, arg3_reg);
1281 bind(verify_continue);
1282 pop(arg2_reg);
1283 pop(arg3_reg);
1284 pop(rbx);
1285 pop(rax);
1286 #endif // ASSERT
1287 }
1288
1289
1290 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1291 int constant,
1292 Register value) {
1293 assert(ProfileInterpreter, "must be profiling interpreter");
1294 Address data(mdp_in, constant);
1295 movptr(data, value);
1296 }
1297
1298
1299 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1300 int constant,
1301 bool decrement) {
1302 // Counter address
1303 Address data(mdp_in, constant);
1304
1305 increment_mdp_data_at(data, decrement);
1306 }
1307
1308 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1309 bool decrement) {
1310 assert(ProfileInterpreter, "must be profiling interpreter");
1311 // %%% this does 64bit counters at best it is wasting space
1312 // at worst it is a rare bug when counters overflow
1313
1314 if (decrement) {
1315 // Decrement the register. Set condition codes.
1316 addptr(data, (int32_t) -DataLayout::counter_increment);
1317 // If the decrement causes the counter to overflow, stay negative
1318 Label L;
1319 jcc(Assembler::negative, L);
1320 addptr(data, (int32_t) DataLayout::counter_increment);
1321 bind(L);
1322 } else {
1323 assert(DataLayout::counter_increment == 1,
1324 "flow-free idiom only works with 1");
1325 // Increment the register. Set carry flag.
1326 addptr(data, DataLayout::counter_increment);
1327 // If the increment causes the counter to overflow, pull back by 1.
1328 sbbptr(data, (int32_t)0);
1329 }
1330 }
1331
1332
1333 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1334 Register reg,
1335 int constant,
1336 bool decrement) {
1337 Address data(mdp_in, reg, Address::times_1, constant);
1338
1339 increment_mdp_data_at(data, decrement);
1340 }
1341
1342 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1343 int flag_byte_constant) {
1344 assert(ProfileInterpreter, "must be profiling interpreter");
1345 int header_offset = in_bytes(DataLayout::header_offset());
1346 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
1347 // Set the flag
1348 orl(Address(mdp_in, header_offset), header_bits);
1349 }
1350
1351
1352
1353 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1354 int offset,
1355 Register value,
1356 Register test_value_out,
1357 Label& not_equal_continue) {
1358 assert(ProfileInterpreter, "must be profiling interpreter");
1359 if (test_value_out == noreg) {
1360 cmpptr(value, Address(mdp_in, offset));
1361 } else {
1362 // Put the test value into a register, so caller can use it:
1363 movptr(test_value_out, Address(mdp_in, offset));
1364 cmpptr(test_value_out, value);
1365 }
1366 jcc(Assembler::notEqual, not_equal_continue);
1367 }
1368
1369
1370 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1371 int offset_of_disp) {
1372 assert(ProfileInterpreter, "must be profiling interpreter");
1373 Address disp_address(mdp_in, offset_of_disp);
1374 addptr(mdp_in, disp_address);
1375 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1376 }
1377
1378
1379 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1380 Register reg,
1381 int offset_of_disp) {
1382 assert(ProfileInterpreter, "must be profiling interpreter");
1383 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1384 addptr(mdp_in, disp_address);
1385 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1386 }
1387
1388
1389 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1390 int constant) {
1391 assert(ProfileInterpreter, "must be profiling interpreter");
1392 addptr(mdp_in, constant);
1393 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1394 }
1395
1396
1397 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1398 assert(ProfileInterpreter, "must be profiling interpreter");
1399 push(return_bci); // save/restore across call_VM
1400 call_VM(noreg,
1401 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1402 return_bci);
1403 pop(return_bci);
1404 }
1405
1406
1407 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1408 Register bumped_count) {
1409 if (ProfileInterpreter) {
1410 Label profile_continue;
1411
1412 // If no method data exists, go to profile_continue.
1413 // Otherwise, assign to mdp
1414 test_method_data_pointer(mdp, profile_continue);
1415
1416 // We are taking a branch. Increment the taken count.
1417 // We inline increment_mdp_data_at to return bumped_count in a register
1418 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1419 Address data(mdp, in_bytes(JumpData::taken_offset()));
1420 movptr(bumped_count, data);
1421 assert(DataLayout::counter_increment == 1,
1422 "flow-free idiom only works with 1");
1423 addptr(bumped_count, DataLayout::counter_increment);
1424 sbbptr(bumped_count, 0);
1425 movptr(data, bumped_count); // Store back out
1426
1427 // The method data pointer needs to be updated to reflect the new target.
1428 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1429 bind(profile_continue);
1430 }
1431 }
1432
1433
1434 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1435 if (ProfileInterpreter) {
1436 Label profile_continue;
1437
1438 // If no method data exists, go to profile_continue.
1439 test_method_data_pointer(mdp, profile_continue);
1440
1441 // We are taking a branch. Increment the not taken count.
1442 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1443
1444 // The method data pointer needs to be updated to correspond to
1445 // the next bytecode
1446 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1447 bind(profile_continue);
1448 }
1449 }
1450
1451 void InterpreterMacroAssembler::profile_call(Register mdp) {
1452 if (ProfileInterpreter) {
1453 Label profile_continue;
1454
1455 // If no method data exists, go to profile_continue.
1456 test_method_data_pointer(mdp, profile_continue);
1457
1458 // We are making a call. Increment the count.
1459 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1460
1461 // The method data pointer needs to be updated to reflect the new target.
1462 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1463 bind(profile_continue);
1464 }
1465 }
1466
1467
1468 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1469 if (ProfileInterpreter) {
1470 Label profile_continue;
1471
1472 // If no method data exists, go to profile_continue.
1473 test_method_data_pointer(mdp, profile_continue);
1474
1475 // We are making a call. Increment the count.
1476 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1477
1478 // The method data pointer needs to be updated to reflect the new target.
1479 update_mdp_by_constant(mdp,
1480 in_bytes(VirtualCallData::
1481 virtual_call_data_size()));
1482 bind(profile_continue);
1483 }
1484 }
1485
1486
1487 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1488 Register mdp,
1489 Register reg2,
1490 bool receiver_can_be_null) {
1491 if (ProfileInterpreter) {
1492 Label profile_continue;
1493
1494 // If no method data exists, go to profile_continue.
1495 test_method_data_pointer(mdp, profile_continue);
1496
1497 Label skip_receiver_profile;
1498 if (receiver_can_be_null) {
1499 Label not_null;
1500 testptr(receiver, receiver);
1501 jccb(Assembler::notZero, not_null);
1502 // We are making a call. Increment the count for null receiver.
1503 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1504 jmp(skip_receiver_profile);
1505 bind(not_null);
1506 }
1507
1508 // Record the receiver type.
1509 record_klass_in_profile(receiver, mdp, reg2, true);
1510 bind(skip_receiver_profile);
1511
1512 // The method data pointer needs to be updated to reflect the new target.
1513 #if INCLUDE_JVMCI
1514 if (MethodProfileWidth == 0) {
1515 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1516 }
1517 #else // INCLUDE_JVMCI
1518 update_mdp_by_constant(mdp,
1519 in_bytes(VirtualCallData::
1520 virtual_call_data_size()));
1521 #endif // INCLUDE_JVMCI
1522 bind(profile_continue);
1523 }
1524 }
1525
1526 #if INCLUDE_JVMCI
1527 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) {
1528 assert_different_registers(method, mdp, reg2);
1529 if (ProfileInterpreter && MethodProfileWidth > 0) {
1530 Label profile_continue;
1531
1532 // If no method data exists, go to profile_continue.
1533 test_method_data_pointer(mdp, profile_continue);
1534
1535 Label done;
1536 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth,
1537 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1538 bind(done);
1539
1540 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1541 bind(profile_continue);
1542 }
1543 }
1544 #endif // INCLUDE_JVMCI
1545
1546 // This routine creates a state machine for updating the multi-row
1547 // type profile at a virtual call site (or other type-sensitive bytecode).
1548 // The machine visits each row (of receiver/count) until the receiver type
1549 // is found, or until it runs out of rows. At the same time, it remembers
1550 // the location of the first empty row. (An empty row records null for its
1551 // receiver, and can be allocated for a newly-observed receiver type.)
1552 // Because there are two degrees of freedom in the state, a simple linear
1553 // search will not work; it must be a decision tree. Hence this helper
1554 // function is recursive, to generate the required tree structured code.
1555 // It's the interpreter, so we are trading off code space for speed.
1556 // See below for example code.
1557 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1558 Register receiver, Register mdp,
1559 Register reg2, int start_row,
1560 Label& done, bool is_virtual_call) {
1561 if (TypeProfileWidth == 0) {
1562 if (is_virtual_call) {
1563 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1564 }
1565 #if INCLUDE_JVMCI
1566 else if (EnableJVMCI) {
1567 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1568 }
1569 #endif // INCLUDE_JVMCI
1570 } else {
1571 int non_profiled_offset = -1;
1572 if (is_virtual_call) {
1573 non_profiled_offset = in_bytes(CounterData::count_offset());
1574 }
1575 #if INCLUDE_JVMCI
1576 else if (EnableJVMCI) {
1577 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1578 }
1579 #endif // INCLUDE_JVMCI
1580
1581 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1582 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1583 }
1584 }
1585
1586 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1587 Register reg2, int start_row, Label& done, int total_rows,
1588 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1589 int non_profiled_offset) {
1590 int last_row = total_rows - 1;
1591 assert(start_row <= last_row, "must be work left to do");
1592 // Test this row for both the item and for null.
1593 // Take any of three different outcomes:
1594 // 1. found item => increment count and goto done
1595 // 2. found null => keep looking for case 1, maybe allocate this cell
1596 // 3. found something else => keep looking for cases 1 and 2
1597 // Case 3 is handled by a recursive call.
1598 for (int row = start_row; row <= last_row; row++) {
1599 Label next_test;
1600 bool test_for_null_also = (row == start_row);
1601
1602 // See if the item is item[n].
1603 int item_offset = in_bytes(item_offset_fn(row));
1604 test_mdp_data_at(mdp, item_offset, item,
1605 (test_for_null_also ? reg2 : noreg),
1606 next_test);
1607 // (Reg2 now contains the item from the CallData.)
1608
1609 // The item is item[n]. Increment count[n].
1610 int count_offset = in_bytes(item_count_offset_fn(row));
1611 increment_mdp_data_at(mdp, count_offset);
1612 jmp(done);
1613 bind(next_test);
1614
1615 if (test_for_null_also) {
1616 Label found_null;
1617 // Failed the equality check on item[n]... Test for null.
1618 testptr(reg2, reg2);
1619 if (start_row == last_row) {
1620 // The only thing left to do is handle the null case.
1621 if (non_profiled_offset >= 0) {
1622 jccb(Assembler::zero, found_null);
1623 // Item did not match any saved item and there is no empty row for it.
1624 // Increment total counter to indicate polymorphic case.
1625 increment_mdp_data_at(mdp, non_profiled_offset);
1626 jmp(done);
1627 bind(found_null);
1628 } else {
1629 jcc(Assembler::notZero, done);
1630 }
1631 break;
1632 }
1633 // Since null is rare, make it be the branch-taken case.
1634 jcc(Assembler::zero, found_null);
1635
1636 // Put all the "Case 3" tests here.
1637 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1638 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1639
1640 // Found a null. Keep searching for a matching item,
1641 // but remember that this is an empty (unused) slot.
1642 bind(found_null);
1643 }
1644 }
1645
1646 // In the fall-through case, we found no matching item, but we
1647 // observed the item[start_row] is NULL.
1648
1649 // Fill in the item field and increment the count.
1650 int item_offset = in_bytes(item_offset_fn(start_row));
1651 set_mdp_data_at(mdp, item_offset, item);
1652 int count_offset = in_bytes(item_count_offset_fn(start_row));
1653 movl(reg2, DataLayout::counter_increment);
1654 set_mdp_data_at(mdp, count_offset, reg2);
1655 if (start_row > 0) {
1656 jmp(done);
1657 }
1658 }
1659
1660 // Example state machine code for three profile rows:
1661 // // main copy of decision tree, rooted at row[1]
1662 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1663 // if (row[0].rec != NULL) {
1664 // // inner copy of decision tree, rooted at row[1]
1665 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1666 // if (row[1].rec != NULL) {
1667 // // degenerate decision tree, rooted at row[2]
1668 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1669 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1670 // row[2].init(rec); goto done;
1671 // } else {
1672 // // remember row[1] is empty
1673 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1674 // row[1].init(rec); goto done;
1675 // }
1676 // } else {
1677 // // remember row[0] is empty
1678 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1679 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1680 // row[0].init(rec); goto done;
1681 // }
1682 // done:
1683
1684 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1685 Register mdp, Register reg2,
1686 bool is_virtual_call) {
1687 assert(ProfileInterpreter, "must be profiling");
1688 Label done;
1689
1690 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1691
1692 bind (done);
1693 }
1694
1695 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1696 Register mdp) {
1697 if (ProfileInterpreter) {
1698 Label profile_continue;
1699 uint row;
1700
1701 // If no method data exists, go to profile_continue.
1702 test_method_data_pointer(mdp, profile_continue);
1703
1704 // Update the total ret count.
1705 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1706
1707 for (row = 0; row < RetData::row_limit(); row++) {
1708 Label next_test;
1709
1710 // See if return_bci is equal to bci[n]:
1711 test_mdp_data_at(mdp,
1712 in_bytes(RetData::bci_offset(row)),
1713 return_bci, noreg,
1714 next_test);
1715
1716 // return_bci is equal to bci[n]. Increment the count.
1717 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1718
1719 // The method data pointer needs to be updated to reflect the new target.
1720 update_mdp_by_offset(mdp,
1721 in_bytes(RetData::bci_displacement_offset(row)));
1722 jmp(profile_continue);
1723 bind(next_test);
1724 }
1725
1726 update_mdp_for_ret(return_bci);
1727
1728 bind(profile_continue);
1729 }
1730 }
1731
1732
1733 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1734 if (ProfileInterpreter) {
1735 Label profile_continue;
1736
1737 // If no method data exists, go to profile_continue.
1738 test_method_data_pointer(mdp, profile_continue);
1739
1740 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1741
1742 // The method data pointer needs to be updated.
1743 int mdp_delta = in_bytes(BitData::bit_data_size());
1744 if (TypeProfileCasts) {
1745 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1746 }
1747 update_mdp_by_constant(mdp, mdp_delta);
1748
1749 bind(profile_continue);
1750 }
1751 }
1752
1753
1754 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1755 if (ProfileInterpreter && TypeProfileCasts) {
1756 Label profile_continue;
1757
1758 // If no method data exists, go to profile_continue.
1759 test_method_data_pointer(mdp, profile_continue);
1760
1761 int count_offset = in_bytes(CounterData::count_offset());
1762 // Back up the address, since we have already bumped the mdp.
1763 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1764
1765 // *Decrement* the counter. We expect to see zero or small negatives.
1766 increment_mdp_data_at(mdp, count_offset, true);
1767
1768 bind (profile_continue);
1769 }
1770 }
1771
1772
1773 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1774 if (ProfileInterpreter) {
1775 Label profile_continue;
1776
1777 // If no method data exists, go to profile_continue.
1778 test_method_data_pointer(mdp, profile_continue);
1779
1780 // The method data pointer needs to be updated.
1781 int mdp_delta = in_bytes(BitData::bit_data_size());
1782 if (TypeProfileCasts) {
1783 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1784
1785 // Record the object type.
1786 record_klass_in_profile(klass, mdp, reg2, false);
1787 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1788 NOT_LP64(restore_locals();) // Restore EDI
1789 }
1790 update_mdp_by_constant(mdp, mdp_delta);
1791
1792 bind(profile_continue);
1793 }
1794 }
1795
1796
1797 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1798 if (ProfileInterpreter) {
1799 Label profile_continue;
1800
1801 // If no method data exists, go to profile_continue.
1802 test_method_data_pointer(mdp, profile_continue);
1803
1804 // Update the default case count
1805 increment_mdp_data_at(mdp,
1806 in_bytes(MultiBranchData::default_count_offset()));
1807
1808 // The method data pointer needs to be updated.
1809 update_mdp_by_offset(mdp,
1810 in_bytes(MultiBranchData::
1811 default_displacement_offset()));
1812
1813 bind(profile_continue);
1814 }
1815 }
1816
1817
1818 void InterpreterMacroAssembler::profile_switch_case(Register index,
1819 Register mdp,
1820 Register reg2) {
1821 if (ProfileInterpreter) {
1822 Label profile_continue;
1823
1824 // If no method data exists, go to profile_continue.
1825 test_method_data_pointer(mdp, profile_continue);
1826
1827 // Build the base (index * per_case_size_in_bytes()) +
1828 // case_array_offset_in_bytes()
1829 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1830 imulptr(index, reg2); // XXX l ?
1831 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1832
1833 // Update the case count
1834 increment_mdp_data_at(mdp,
1835 index,
1836 in_bytes(MultiBranchData::relative_count_offset()));
1837
1838 // The method data pointer needs to be updated.
1839 update_mdp_by_offset(mdp,
1840 index,
1841 in_bytes(MultiBranchData::
1842 relative_displacement_offset()));
1843
1844 bind(profile_continue);
1845 }
1846 }
1847
1848
1849
1850 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1851 if (state == atos) {
1852 MacroAssembler::verify_oop(reg);
1853 }
1854 }
1855
1856 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1857 #ifndef _LP64
1858 if ((state == ftos && UseSSE < 1) ||
1859 (state == dtos && UseSSE < 2)) {
1860 MacroAssembler::verify_FPU(stack_depth);
1861 }
1862 #endif
1863 }
1864
1865 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1866 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1867 int increment, Address mask,
1868 Register scratch, bool preloaded,
1869 Condition cond, Label* where) {
1870 if (!preloaded) {
1871 movl(scratch, counter_addr);
1872 }
1873 incrementl(scratch, increment);
1874 movl(counter_addr, scratch);
1875 andl(scratch, mask);
1876 jcc(cond, *where);
1877 }
1878
1879 void InterpreterMacroAssembler::notify_method_entry() {
1880 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1881 // track stack depth. If it is possible to enter interp_only_mode we add
1882 // the code to check if the event should be sent.
1883 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1884 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1885 if (JvmtiExport::can_post_interpreter_events()) {
1886 Label L;
1887 NOT_LP64(get_thread(rthread);)
1888 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1889 testl(rdx, rdx);
1890 jcc(Assembler::zero, L);
1891 call_VM(noreg, CAST_FROM_FN_PTR(address,
1892 InterpreterRuntime::post_method_entry));
1893 bind(L);
1894 }
1895
1896 {
1897 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1898 NOT_LP64(get_thread(rthread);)
1899 get_method(rarg);
1900 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1901 rthread, rarg);
1902 }
1903
1904 // RedefineClasses() tracing support for obsolete method entry
1905 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
1906 NOT_LP64(get_thread(rthread);)
1907 get_method(rarg);
1908 call_VM_leaf(
1909 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1910 rthread, rarg);
1911 }
1912 }
1913
1914
1915 void InterpreterMacroAssembler::notify_method_exit(
1916 TosState state, NotifyMethodExitMode mode) {
1917 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1918 // track stack depth. If it is possible to enter interp_only_mode we add
1919 // the code to check if the event should be sent.
1920 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1921 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1922 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1923 Label L;
1924 // Note: frame::interpreter_frame_result has a dependency on how the
1925 // method result is saved across the call to post_method_exit. If this
1926 // is changed then the interpreter_frame_result implementation will
1927 // need to be updated too.
1928
1929 // template interpreter will leave the result on the top of the stack.
1930 push(state);
1931 NOT_LP64(get_thread(rthread);)
1932 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1933 testl(rdx, rdx);
1934 jcc(Assembler::zero, L);
1935 call_VM(noreg,
1936 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1937 bind(L);
1938 pop(state);
1939 }
1940
1941 {
1942 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1943 push(state);
1944 NOT_LP64(get_thread(rthread);)
1945 get_method(rarg);
1946 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1947 rthread, rarg);
1948 pop(state);
1949 }
1950 }