1 | /* |
2 | * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved. |
3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
4 | * |
5 | * This code is free software; you can redistribute it and/or modify it |
6 | * under the terms of the GNU General Public License version 2 only, as |
7 | * published by the Free Software Foundation. |
8 | * |
9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
12 | * version 2 for more details (a copy is included in the LICENSE file that |
13 | * accompanied this code). |
14 | * |
15 | * You should have received a copy of the GNU General Public License version |
16 | * 2 along with this work; if not, write to the Free Software Foundation, |
17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
18 | * |
19 | * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
20 | * or visit www.oracle.com if you need additional information or have any |
21 | * questions. |
22 | * |
23 | */ |
24 | |
25 | #include "precompiled.hpp" |
26 | #include "interp_masm_x86.hpp" |
27 | #include "interpreter/interpreter.hpp" |
28 | #include "interpreter/interpreterRuntime.hpp" |
29 | #include "logging/log.hpp" |
30 | #include "oops/arrayOop.hpp" |
31 | #include "oops/markOop.hpp" |
32 | #include "oops/methodData.hpp" |
33 | #include "oops/method.hpp" |
34 | #include "prims/jvmtiExport.hpp" |
35 | #include "prims/jvmtiThreadState.hpp" |
36 | #include "runtime/basicLock.hpp" |
37 | #include "runtime/biasedLocking.hpp" |
38 | #include "runtime/frame.inline.hpp" |
39 | #include "runtime/safepointMechanism.hpp" |
40 | #include "runtime/sharedRuntime.hpp" |
41 | #include "runtime/thread.inline.hpp" |
42 | |
43 | // Implementation of InterpreterMacroAssembler |
44 | |
45 | void InterpreterMacroAssembler::jump_to_entry(address entry) { |
46 | assert(entry, "Entry must have been generated by now" ); |
47 | jump(RuntimeAddress(entry)); |
48 | } |
49 | |
50 | void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) { |
51 | Label update, next, none; |
52 | |
53 | verify_oop(obj); |
54 | |
55 | testptr(obj, obj); |
56 | jccb(Assembler::notZero, update); |
57 | orptr(mdo_addr, TypeEntries::null_seen); |
58 | jmpb(next); |
59 | |
60 | bind(update); |
61 | load_klass(obj, obj); |
62 | |
63 | xorptr(obj, mdo_addr); |
64 | testptr(obj, TypeEntries::type_klass_mask); |
65 | jccb(Assembler::zero, next); // klass seen before, nothing to |
66 | // do. The unknown bit may have been |
67 | // set already but no need to check. |
68 | |
69 | testptr(obj, TypeEntries::type_unknown); |
70 | jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore. |
71 | |
72 | cmpptr(mdo_addr, 0); |
73 | jccb(Assembler::equal, none); |
74 | cmpptr(mdo_addr, TypeEntries::null_seen); |
75 | jccb(Assembler::equal, none); |
76 | // There is a chance that the checks above (re-reading profiling |
77 | // data from memory) fail if another thread has just set the |
78 | // profiling to this obj's klass |
79 | xorptr(obj, mdo_addr); |
80 | testptr(obj, TypeEntries::type_klass_mask); |
81 | jccb(Assembler::zero, next); |
82 | |
83 | // different than before. Cannot keep accurate profile. |
84 | orptr(mdo_addr, TypeEntries::type_unknown); |
85 | jmpb(next); |
86 | |
87 | bind(none); |
88 | // first time here. Set profile type. |
89 | movptr(mdo_addr, obj); |
90 | |
91 | bind(next); |
92 | } |
93 | |
94 | void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) { |
95 | if (!ProfileInterpreter) { |
96 | return; |
97 | } |
98 | |
99 | if (MethodData::profile_arguments() || MethodData::profile_return()) { |
100 | Label profile_continue; |
101 | |
102 | test_method_data_pointer(mdp, profile_continue); |
103 | |
104 | int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); |
105 | |
106 | cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); |
107 | jcc(Assembler::notEqual, profile_continue); |
108 | |
109 | if (MethodData::profile_arguments()) { |
110 | Label done; |
111 | int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); |
112 | addptr(mdp, off_to_args); |
113 | |
114 | for (int i = 0; i < TypeProfileArgsLimit; i++) { |
115 | if (i > 0 || MethodData::profile_return()) { |
116 | // If return value type is profiled we may have no argument to profile |
117 | movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args)); |
118 | subl(tmp, i*TypeStackSlotEntries::per_arg_count()); |
119 | cmpl(tmp, TypeStackSlotEntries::per_arg_count()); |
120 | jcc(Assembler::less, done); |
121 | } |
122 | movptr(tmp, Address(callee, Method::const_offset())); |
123 | load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset())); |
124 | // stack offset o (zero based) from the start of the argument |
125 | // list, for n arguments translates into offset n - o - 1 from |
126 | // the end of the argument list |
127 | subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args)); |
128 | subl(tmp, 1); |
129 | Address arg_addr = argument_address(tmp); |
130 | movptr(tmp, arg_addr); |
131 | |
132 | Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args); |
133 | profile_obj_type(tmp, mdo_arg_addr); |
134 | |
135 | int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); |
136 | addptr(mdp, to_add); |
137 | off_to_args += to_add; |
138 | } |
139 | |
140 | if (MethodData::profile_return()) { |
141 | movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args)); |
142 | subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); |
143 | } |
144 | |
145 | bind(done); |
146 | |
147 | if (MethodData::profile_return()) { |
148 | // We're right after the type profile for the last |
149 | // argument. tmp is the number of cells left in the |
150 | // CallTypeData/VirtualCallTypeData to reach its end. Non null |
151 | // if there's a return to profile. |
152 | assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type" ); |
153 | shll(tmp, exact_log2(DataLayout::cell_size)); |
154 | addptr(mdp, tmp); |
155 | } |
156 | movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp); |
157 | } else { |
158 | assert(MethodData::profile_return(), "either profile call args or call ret" ); |
159 | update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size())); |
160 | } |
161 | |
162 | // mdp points right after the end of the |
163 | // CallTypeData/VirtualCallTypeData, right after the cells for the |
164 | // return value type if there's one |
165 | |
166 | bind(profile_continue); |
167 | } |
168 | } |
169 | |
170 | void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) { |
171 | assert_different_registers(mdp, ret, tmp, _bcp_register); |
172 | if (ProfileInterpreter && MethodData::profile_return()) { |
173 | Label profile_continue; |
174 | |
175 | test_method_data_pointer(mdp, profile_continue); |
176 | |
177 | if (MethodData::profile_return_jsr292_only()) { |
178 | assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2" ); |
179 | |
180 | // If we don't profile all invoke bytecodes we must make sure |
181 | // it's a bytecode we indeed profile. We can't go back to the |
182 | // begining of the ProfileData we intend to update to check its |
183 | // type because we're right after it and we don't known its |
184 | // length |
185 | Label do_profile; |
186 | cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic); |
187 | jcc(Assembler::equal, do_profile); |
188 | cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle); |
189 | jcc(Assembler::equal, do_profile); |
190 | get_method(tmp); |
191 | cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm); |
192 | jcc(Assembler::notEqual, profile_continue); |
193 | |
194 | bind(do_profile); |
195 | } |
196 | |
197 | Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size())); |
198 | mov(tmp, ret); |
199 | profile_obj_type(tmp, mdo_ret_addr); |
200 | |
201 | bind(profile_continue); |
202 | } |
203 | } |
204 | |
205 | void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) { |
206 | if (ProfileInterpreter && MethodData::profile_parameters()) { |
207 | Label profile_continue; |
208 | |
209 | test_method_data_pointer(mdp, profile_continue); |
210 | |
211 | // Load the offset of the area within the MDO used for |
212 | // parameters. If it's negative we're not profiling any parameters |
213 | movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()))); |
214 | testl(tmp1, tmp1); |
215 | jcc(Assembler::negative, profile_continue); |
216 | |
217 | // Compute a pointer to the area for parameters from the offset |
218 | // and move the pointer to the slot for the last |
219 | // parameters. Collect profiling from last parameter down. |
220 | // mdo start + parameters offset + array length - 1 |
221 | addptr(mdp, tmp1); |
222 | movptr(tmp1, Address(mdp, ArrayData::array_len_offset())); |
223 | decrement(tmp1, TypeStackSlotEntries::per_arg_count()); |
224 | |
225 | Label loop; |
226 | bind(loop); |
227 | |
228 | int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); |
229 | int type_base = in_bytes(ParametersTypeData::type_offset(0)); |
230 | Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size); |
231 | Address arg_off(mdp, tmp1, per_arg_scale, off_base); |
232 | Address arg_type(mdp, tmp1, per_arg_scale, type_base); |
233 | |
234 | // load offset on the stack from the slot for this parameter |
235 | movptr(tmp2, arg_off); |
236 | negptr(tmp2); |
237 | // read the parameter from the local area |
238 | movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale())); |
239 | |
240 | // profile the parameter |
241 | profile_obj_type(tmp2, arg_type); |
242 | |
243 | // go to next parameter |
244 | decrement(tmp1, TypeStackSlotEntries::per_arg_count()); |
245 | jcc(Assembler::positive, loop); |
246 | |
247 | bind(profile_continue); |
248 | } |
249 | } |
250 | |
251 | void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, |
252 | int number_of_arguments) { |
253 | // interpreter specific |
254 | // |
255 | // Note: No need to save/restore bcp & locals registers |
256 | // since these are callee saved registers and no blocking/ |
257 | // GC can happen in leaf calls. |
258 | // Further Note: DO NOT save/restore bcp/locals. If a caller has |
259 | // already saved them so that it can use rsi/rdi as temporaries |
260 | // then a save/restore here will DESTROY the copy the caller |
261 | // saved! There used to be a save_bcp() that only happened in |
262 | // the ASSERT path (no restore_bcp). Which caused bizarre failures |
263 | // when jvm built with ASSERTs. |
264 | #ifdef ASSERT |
265 | { |
266 | Label L; |
267 | cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); |
268 | jcc(Assembler::equal, L); |
269 | stop("InterpreterMacroAssembler::call_VM_leaf_base:" |
270 | " last_sp != NULL" ); |
271 | bind(L); |
272 | } |
273 | #endif |
274 | // super call |
275 | MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments); |
276 | // interpreter specific |
277 | // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals |
278 | // but since they may not have been saved (and we don't want to |
279 | // save them here (see note above) the assert is invalid. |
280 | } |
281 | |
282 | void InterpreterMacroAssembler::call_VM_base(Register oop_result, |
283 | Register java_thread, |
284 | Register last_java_sp, |
285 | address entry_point, |
286 | int number_of_arguments, |
287 | bool check_exceptions) { |
288 | // interpreter specific |
289 | // |
290 | // Note: Could avoid restoring locals ptr (callee saved) - however doesn't |
291 | // really make a difference for these runtime calls, since they are |
292 | // slow anyway. Btw., bcp must be saved/restored since it may change |
293 | // due to GC. |
294 | NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread" );) |
295 | save_bcp(); |
296 | #ifdef ASSERT |
297 | { |
298 | Label L; |
299 | cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); |
300 | jcc(Assembler::equal, L); |
301 | stop("InterpreterMacroAssembler::call_VM_base:" |
302 | " last_sp != NULL" ); |
303 | bind(L); |
304 | } |
305 | #endif /* ASSERT */ |
306 | // super call |
307 | MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp, |
308 | entry_point, number_of_arguments, |
309 | check_exceptions); |
310 | // interpreter specific |
311 | restore_bcp(); |
312 | restore_locals(); |
313 | } |
314 | |
315 | void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) { |
316 | if (JvmtiExport::can_pop_frame()) { |
317 | Label L; |
318 | // Initiate popframe handling only if it is not already being |
319 | // processed. If the flag has the popframe_processing bit set, it |
320 | // means that this code is called *during* popframe handling - we |
321 | // don't want to reenter. |
322 | // This method is only called just after the call into the vm in |
323 | // call_VM_base, so the arg registers are available. |
324 | Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit |
325 | LP64_ONLY(c_rarg0); |
326 | movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset())); |
327 | testl(pop_cond, JavaThread::popframe_pending_bit); |
328 | jcc(Assembler::zero, L); |
329 | testl(pop_cond, JavaThread::popframe_processing_bit); |
330 | jcc(Assembler::notZero, L); |
331 | // Call Interpreter::remove_activation_preserving_args_entry() to get the |
332 | // address of the same-named entrypoint in the generated interpreter code. |
333 | call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); |
334 | jmp(rax); |
335 | bind(L); |
336 | NOT_LP64(get_thread(java_thread);) |
337 | } |
338 | } |
339 | |
340 | void InterpreterMacroAssembler::load_earlyret_value(TosState state) { |
341 | Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx); |
342 | NOT_LP64(get_thread(thread);) |
343 | movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset())); |
344 | const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset()); |
345 | const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset()); |
346 | const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset()); |
347 | #ifdef _LP64 |
348 | switch (state) { |
349 | case atos: movptr(rax, oop_addr); |
350 | movptr(oop_addr, (int32_t)NULL_WORD); |
351 | verify_oop(rax, state); break; |
352 | case ltos: movptr(rax, val_addr); break; |
353 | case btos: // fall through |
354 | case ztos: // fall through |
355 | case ctos: // fall through |
356 | case stos: // fall through |
357 | case itos: movl(rax, val_addr); break; |
358 | case ftos: load_float(val_addr); break; |
359 | case dtos: load_double(val_addr); break; |
360 | case vtos: /* nothing to do */ break; |
361 | default : ShouldNotReachHere(); |
362 | } |
363 | // Clean up tos value in the thread object |
364 | movl(tos_addr, (int) ilgl); |
365 | movl(val_addr, (int32_t) NULL_WORD); |
366 | #else |
367 | const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset() |
368 | + in_ByteSize(wordSize)); |
369 | switch (state) { |
370 | case atos: movptr(rax, oop_addr); |
371 | movptr(oop_addr, NULL_WORD); |
372 | verify_oop(rax, state); break; |
373 | case ltos: |
374 | movl(rdx, val_addr1); // fall through |
375 | case btos: // fall through |
376 | case ztos: // fall through |
377 | case ctos: // fall through |
378 | case stos: // fall through |
379 | case itos: movl(rax, val_addr); break; |
380 | case ftos: load_float(val_addr); break; |
381 | case dtos: load_double(val_addr); break; |
382 | case vtos: /* nothing to do */ break; |
383 | default : ShouldNotReachHere(); |
384 | } |
385 | #endif // _LP64 |
386 | // Clean up tos value in the thread object |
387 | movl(tos_addr, (int32_t) ilgl); |
388 | movptr(val_addr, NULL_WORD); |
389 | NOT_LP64(movptr(val_addr1, NULL_WORD);) |
390 | } |
391 | |
392 | |
393 | void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) { |
394 | if (JvmtiExport::can_force_early_return()) { |
395 | Label L; |
396 | Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread); |
397 | Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread); |
398 | |
399 | movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset())); |
400 | testptr(tmp, tmp); |
401 | jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit; |
402 | |
403 | // Initiate earlyret handling only if it is not already being processed. |
404 | // If the flag has the earlyret_processing bit set, it means that this code |
405 | // is called *during* earlyret handling - we don't want to reenter. |
406 | movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset())); |
407 | cmpl(tmp, JvmtiThreadState::earlyret_pending); |
408 | jcc(Assembler::notEqual, L); |
409 | |
410 | // Call Interpreter::remove_activation_early_entry() to get the address of the |
411 | // same-named entrypoint in the generated interpreter code. |
412 | NOT_LP64(get_thread(java_thread);) |
413 | movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset())); |
414 | #ifdef _LP64 |
415 | movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset())); |
416 | call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp); |
417 | #else |
418 | pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset())); |
419 | call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1); |
420 | #endif // _LP64 |
421 | jmp(rax); |
422 | bind(L); |
423 | NOT_LP64(get_thread(java_thread);) |
424 | } |
425 | } |
426 | |
427 | void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) { |
428 | assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode" ); |
429 | load_unsigned_short(reg, Address(_bcp_register, bcp_offset)); |
430 | bswapl(reg); |
431 | shrl(reg, 16); |
432 | } |
433 | |
434 | void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, |
435 | int bcp_offset, |
436 | size_t index_size) { |
437 | assert(bcp_offset > 0, "bcp is still pointing to start of bytecode" ); |
438 | if (index_size == sizeof(u2)) { |
439 | load_unsigned_short(index, Address(_bcp_register, bcp_offset)); |
440 | } else if (index_size == sizeof(u4)) { |
441 | movl(index, Address(_bcp_register, bcp_offset)); |
442 | // Check if the secondary index definition is still ~x, otherwise |
443 | // we have to change the following assembler code to calculate the |
444 | // plain index. |
445 | assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line" ); |
446 | notl(index); // convert to plain index |
447 | } else if (index_size == sizeof(u1)) { |
448 | load_unsigned_byte(index, Address(_bcp_register, bcp_offset)); |
449 | } else { |
450 | ShouldNotReachHere(); |
451 | } |
452 | } |
453 | |
454 | void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, |
455 | Register index, |
456 | int bcp_offset, |
457 | size_t index_size) { |
458 | assert_different_registers(cache, index); |
459 | get_cache_index_at_bcp(index, bcp_offset, index_size); |
460 | movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); |
461 | assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below" ); |
462 | // convert from field index to ConstantPoolCacheEntry index |
463 | assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line" ); |
464 | shll(index, 2); |
465 | } |
466 | |
467 | void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache, |
468 | Register index, |
469 | Register bytecode, |
470 | int byte_no, |
471 | int bcp_offset, |
472 | size_t index_size) { |
473 | get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size); |
474 | // We use a 32-bit load here since the layout of 64-bit words on |
475 | // little-endian machines allow us that. |
476 | movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset())); |
477 | const int shift_count = (1 + byte_no) * BitsPerByte; |
478 | assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) || |
479 | (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift), |
480 | "correct shift count" ); |
481 | shrl(bytecode, shift_count); |
482 | assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask" ); |
483 | andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask); |
484 | } |
485 | |
486 | void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, |
487 | Register tmp, |
488 | int bcp_offset, |
489 | size_t index_size) { |
490 | assert_different_registers(cache, tmp); |
491 | |
492 | get_cache_index_at_bcp(tmp, bcp_offset, index_size); |
493 | assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below" ); |
494 | // convert from field index to ConstantPoolCacheEntry index |
495 | // and from word offset to byte offset |
496 | assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line" ); |
497 | shll(tmp, 2 + LogBytesPerWord); |
498 | movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); |
499 | // skip past the header |
500 | addptr(cache, in_bytes(ConstantPoolCache::base_offset())); |
501 | addptr(cache, tmp); // construct pointer to cache entry |
502 | } |
503 | |
504 | // Load object from cpool->resolved_references(index) |
505 | void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, |
506 | Register index, |
507 | Register tmp) { |
508 | assert_different_registers(result, index); |
509 | |
510 | get_constant_pool(result); |
511 | // load pointer for resolved_references[] objArray |
512 | movptr(result, Address(result, ConstantPool::cache_offset_in_bytes())); |
513 | movptr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes())); |
514 | resolve_oop_handle(result, tmp); |
515 | load_heap_oop(result, Address(result, index, |
516 | UseCompressedOops ? Address::times_4 : Address::times_ptr, |
517 | arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp); |
518 | } |
519 | |
520 | // load cpool->resolved_klass_at(index) |
521 | void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass, |
522 | Register cpool, |
523 | Register index) { |
524 | assert_different_registers(cpool, index); |
525 | |
526 | movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool))); |
527 | Register resolved_klasses = cpool; |
528 | movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes())); |
529 | movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes())); |
530 | } |
531 | |
532 | void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no, |
533 | Register method, |
534 | Register cache, |
535 | Register index) { |
536 | assert_different_registers(cache, index); |
537 | |
538 | const int method_offset = in_bytes( |
539 | ConstantPoolCache::base_offset() + |
540 | ((byte_no == TemplateTable::f2_byte) |
541 | ? ConstantPoolCacheEntry::f2_offset() |
542 | : ConstantPoolCacheEntry::f1_offset())); |
543 | |
544 | movptr(method, Address(cache, index, Address::times_ptr, method_offset)); // get f1 Method* |
545 | } |
546 | |
547 | // Generate a subtype check: branch to ok_is_subtype if sub_klass is a |
548 | // subtype of super_klass. |
549 | // |
550 | // Args: |
551 | // rax: superklass |
552 | // Rsub_klass: subklass |
553 | // |
554 | // Kills: |
555 | // rcx, rdi |
556 | void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, |
557 | Label& ok_is_subtype) { |
558 | assert(Rsub_klass != rax, "rax holds superklass" ); |
559 | LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals" );) |
560 | LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp" );) |
561 | assert(Rsub_klass != rcx, "rcx holds 2ndary super array length" ); |
562 | assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr" ); |
563 | |
564 | // Profile the not-null value's klass. |
565 | profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi |
566 | |
567 | // Do the check. |
568 | check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx |
569 | |
570 | // Profile the failure of the check. |
571 | profile_typecheck_failed(rcx); // blows rcx |
572 | } |
573 | |
574 | |
575 | #ifndef _LP64 |
576 | void InterpreterMacroAssembler::f2ieee() { |
577 | if (IEEEPrecision) { |
578 | fstp_s(Address(rsp, 0)); |
579 | fld_s(Address(rsp, 0)); |
580 | } |
581 | } |
582 | |
583 | |
584 | void InterpreterMacroAssembler::d2ieee() { |
585 | if (IEEEPrecision) { |
586 | fstp_d(Address(rsp, 0)); |
587 | fld_d(Address(rsp, 0)); |
588 | } |
589 | } |
590 | #endif // _LP64 |
591 | |
592 | // Java Expression Stack |
593 | |
594 | void InterpreterMacroAssembler::pop_ptr(Register r) { |
595 | pop(r); |
596 | } |
597 | |
598 | void InterpreterMacroAssembler::push_ptr(Register r) { |
599 | push(r); |
600 | } |
601 | |
602 | void InterpreterMacroAssembler::push_i(Register r) { |
603 | push(r); |
604 | } |
605 | |
606 | void InterpreterMacroAssembler::push_f(XMMRegister r) { |
607 | subptr(rsp, wordSize); |
608 | movflt(Address(rsp, 0), r); |
609 | } |
610 | |
611 | void InterpreterMacroAssembler::pop_f(XMMRegister r) { |
612 | movflt(r, Address(rsp, 0)); |
613 | addptr(rsp, wordSize); |
614 | } |
615 | |
616 | void InterpreterMacroAssembler::push_d(XMMRegister r) { |
617 | subptr(rsp, 2 * wordSize); |
618 | movdbl(Address(rsp, 0), r); |
619 | } |
620 | |
621 | void InterpreterMacroAssembler::pop_d(XMMRegister r) { |
622 | movdbl(r, Address(rsp, 0)); |
623 | addptr(rsp, 2 * Interpreter::stackElementSize); |
624 | } |
625 | |
626 | #ifdef _LP64 |
627 | void InterpreterMacroAssembler::pop_i(Register r) { |
628 | // XXX can't use pop currently, upper half non clean |
629 | movl(r, Address(rsp, 0)); |
630 | addptr(rsp, wordSize); |
631 | } |
632 | |
633 | void InterpreterMacroAssembler::pop_l(Register r) { |
634 | movq(r, Address(rsp, 0)); |
635 | addptr(rsp, 2 * Interpreter::stackElementSize); |
636 | } |
637 | |
638 | void InterpreterMacroAssembler::push_l(Register r) { |
639 | subptr(rsp, 2 * wordSize); |
640 | movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r ); |
641 | movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD ); |
642 | } |
643 | |
644 | void InterpreterMacroAssembler::pop(TosState state) { |
645 | switch (state) { |
646 | case atos: pop_ptr(); break; |
647 | case btos: |
648 | case ztos: |
649 | case ctos: |
650 | case stos: |
651 | case itos: pop_i(); break; |
652 | case ltos: pop_l(); break; |
653 | case ftos: pop_f(xmm0); break; |
654 | case dtos: pop_d(xmm0); break; |
655 | case vtos: /* nothing to do */ break; |
656 | default: ShouldNotReachHere(); |
657 | } |
658 | verify_oop(rax, state); |
659 | } |
660 | |
661 | void InterpreterMacroAssembler::push(TosState state) { |
662 | verify_oop(rax, state); |
663 | switch (state) { |
664 | case atos: push_ptr(); break; |
665 | case btos: |
666 | case ztos: |
667 | case ctos: |
668 | case stos: |
669 | case itos: push_i(); break; |
670 | case ltos: push_l(); break; |
671 | case ftos: push_f(xmm0); break; |
672 | case dtos: push_d(xmm0); break; |
673 | case vtos: /* nothing to do */ break; |
674 | default : ShouldNotReachHere(); |
675 | } |
676 | } |
677 | #else |
678 | void InterpreterMacroAssembler::pop_i(Register r) { |
679 | pop(r); |
680 | } |
681 | |
682 | void InterpreterMacroAssembler::pop_l(Register lo, Register hi) { |
683 | pop(lo); |
684 | pop(hi); |
685 | } |
686 | |
687 | void InterpreterMacroAssembler::pop_f() { |
688 | fld_s(Address(rsp, 0)); |
689 | addptr(rsp, 1 * wordSize); |
690 | } |
691 | |
692 | void InterpreterMacroAssembler::pop_d() { |
693 | fld_d(Address(rsp, 0)); |
694 | addptr(rsp, 2 * wordSize); |
695 | } |
696 | |
697 | |
698 | void InterpreterMacroAssembler::pop(TosState state) { |
699 | switch (state) { |
700 | case atos: pop_ptr(rax); break; |
701 | case btos: // fall through |
702 | case ztos: // fall through |
703 | case ctos: // fall through |
704 | case stos: // fall through |
705 | case itos: pop_i(rax); break; |
706 | case ltos: pop_l(rax, rdx); break; |
707 | case ftos: |
708 | if (UseSSE >= 1) { |
709 | pop_f(xmm0); |
710 | } else { |
711 | pop_f(); |
712 | } |
713 | break; |
714 | case dtos: |
715 | if (UseSSE >= 2) { |
716 | pop_d(xmm0); |
717 | } else { |
718 | pop_d(); |
719 | } |
720 | break; |
721 | case vtos: /* nothing to do */ break; |
722 | default : ShouldNotReachHere(); |
723 | } |
724 | verify_oop(rax, state); |
725 | } |
726 | |
727 | |
728 | void InterpreterMacroAssembler::push_l(Register lo, Register hi) { |
729 | push(hi); |
730 | push(lo); |
731 | } |
732 | |
733 | void InterpreterMacroAssembler::push_f() { |
734 | // Do not schedule for no AGI! Never write beyond rsp! |
735 | subptr(rsp, 1 * wordSize); |
736 | fstp_s(Address(rsp, 0)); |
737 | } |
738 | |
739 | void InterpreterMacroAssembler::push_d() { |
740 | // Do not schedule for no AGI! Never write beyond rsp! |
741 | subptr(rsp, 2 * wordSize); |
742 | fstp_d(Address(rsp, 0)); |
743 | } |
744 | |
745 | |
746 | void InterpreterMacroAssembler::push(TosState state) { |
747 | verify_oop(rax, state); |
748 | switch (state) { |
749 | case atos: push_ptr(rax); break; |
750 | case btos: // fall through |
751 | case ztos: // fall through |
752 | case ctos: // fall through |
753 | case stos: // fall through |
754 | case itos: push_i(rax); break; |
755 | case ltos: push_l(rax, rdx); break; |
756 | case ftos: |
757 | if (UseSSE >= 1) { |
758 | push_f(xmm0); |
759 | } else { |
760 | push_f(); |
761 | } |
762 | break; |
763 | case dtos: |
764 | if (UseSSE >= 2) { |
765 | push_d(xmm0); |
766 | } else { |
767 | push_d(); |
768 | } |
769 | break; |
770 | case vtos: /* nothing to do */ break; |
771 | default : ShouldNotReachHere(); |
772 | } |
773 | } |
774 | #endif // _LP64 |
775 | |
776 | |
777 | // Helpers for swap and dup |
778 | void InterpreterMacroAssembler::load_ptr(int n, Register val) { |
779 | movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n))); |
780 | } |
781 | |
782 | void InterpreterMacroAssembler::store_ptr(int n, Register val) { |
783 | movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val); |
784 | } |
785 | |
786 | |
787 | void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() { |
788 | // set sender sp |
789 | lea(_bcp_register, Address(rsp, wordSize)); |
790 | // record last_sp |
791 | movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register); |
792 | } |
793 | |
794 | |
795 | // Jump to from_interpreted entry of a call unless single stepping is possible |
796 | // in this thread in which case we must call the i2i entry |
797 | void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) { |
798 | prepare_to_jump_from_interpreted(); |
799 | |
800 | if (JvmtiExport::can_post_interpreter_events()) { |
801 | Label run_compiled_code; |
802 | // JVMTI events, such as single-stepping, are implemented partly by avoiding running |
803 | // compiled code in threads for which the event is enabled. Check here for |
804 | // interp_only_mode if these events CAN be enabled. |
805 | // interp_only is an int, on little endian it is sufficient to test the byte only |
806 | // Is a cmpl faster? |
807 | LP64_ONLY(temp = r15_thread;) |
808 | NOT_LP64(get_thread(temp);) |
809 | cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0); |
810 | jccb(Assembler::zero, run_compiled_code); |
811 | jmp(Address(method, Method::interpreter_entry_offset())); |
812 | bind(run_compiled_code); |
813 | } |
814 | |
815 | jmp(Address(method, Method::from_interpreted_offset())); |
816 | } |
817 | |
818 | // The following two routines provide a hook so that an implementation |
819 | // can schedule the dispatch in two parts. x86 does not do this. |
820 | void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) { |
821 | // Nothing x86 specific to be done here |
822 | } |
823 | |
824 | void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) { |
825 | dispatch_next(state, step); |
826 | } |
827 | |
828 | void InterpreterMacroAssembler::dispatch_base(TosState state, |
829 | address* table, |
830 | bool verifyoop, |
831 | bool generate_poll) { |
832 | verify_FPU(1, state); |
833 | if (VerifyActivationFrameSize) { |
834 | Label L; |
835 | mov(rcx, rbp); |
836 | subptr(rcx, rsp); |
837 | int32_t min_frame_size = |
838 | (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * |
839 | wordSize; |
840 | cmpptr(rcx, (int32_t)min_frame_size); |
841 | jcc(Assembler::greaterEqual, L); |
842 | stop("broken stack frame" ); |
843 | bind(L); |
844 | } |
845 | if (verifyoop) { |
846 | verify_oop(rax, state); |
847 | } |
848 | |
849 | address* const safepoint_table = Interpreter::safept_table(state); |
850 | #ifdef _LP64 |
851 | Label no_safepoint, dispatch; |
852 | if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) { |
853 | NOT_PRODUCT(block_comment("Thread-local Safepoint poll" )); |
854 | testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit()); |
855 | |
856 | jccb(Assembler::zero, no_safepoint); |
857 | lea(rscratch1, ExternalAddress((address)safepoint_table)); |
858 | jmpb(dispatch); |
859 | } |
860 | |
861 | bind(no_safepoint); |
862 | lea(rscratch1, ExternalAddress((address)table)); |
863 | bind(dispatch); |
864 | jmp(Address(rscratch1, rbx, Address::times_8)); |
865 | |
866 | #else |
867 | Address index(noreg, rbx, Address::times_ptr); |
868 | if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) { |
869 | NOT_PRODUCT(block_comment("Thread-local Safepoint poll" )); |
870 | Label no_safepoint; |
871 | const Register thread = rcx; |
872 | get_thread(thread); |
873 | testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit()); |
874 | |
875 | jccb(Assembler::zero, no_safepoint); |
876 | ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index); |
877 | jump(dispatch_addr); |
878 | bind(no_safepoint); |
879 | } |
880 | |
881 | { |
882 | ArrayAddress dispatch_addr(ExternalAddress((address)table), index); |
883 | jump(dispatch_addr); |
884 | } |
885 | #endif // _LP64 |
886 | } |
887 | |
888 | void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) { |
889 | dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); |
890 | } |
891 | |
892 | void InterpreterMacroAssembler::dispatch_only_normal(TosState state) { |
893 | dispatch_base(state, Interpreter::normal_table(state)); |
894 | } |
895 | |
896 | void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) { |
897 | dispatch_base(state, Interpreter::normal_table(state), false); |
898 | } |
899 | |
900 | |
901 | void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) { |
902 | // load next bytecode (load before advancing _bcp_register to prevent AGI) |
903 | load_unsigned_byte(rbx, Address(_bcp_register, step)); |
904 | // advance _bcp_register |
905 | increment(_bcp_register, step); |
906 | dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); |
907 | } |
908 | |
909 | void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { |
910 | // load current bytecode |
911 | load_unsigned_byte(rbx, Address(_bcp_register, 0)); |
912 | dispatch_base(state, table); |
913 | } |
914 | |
915 | void InterpreterMacroAssembler::narrow(Register result) { |
916 | |
917 | // Get method->_constMethod->_result_type |
918 | movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); |
919 | movptr(rcx, Address(rcx, Method::const_offset())); |
920 | load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset())); |
921 | |
922 | Label done, notBool, notByte, notChar; |
923 | |
924 | // common case first |
925 | cmpl(rcx, T_INT); |
926 | jcc(Assembler::equal, done); |
927 | |
928 | // mask integer result to narrower return type. |
929 | cmpl(rcx, T_BOOLEAN); |
930 | jcc(Assembler::notEqual, notBool); |
931 | andl(result, 0x1); |
932 | jmp(done); |
933 | |
934 | bind(notBool); |
935 | cmpl(rcx, T_BYTE); |
936 | jcc(Assembler::notEqual, notByte); |
937 | LP64_ONLY(movsbl(result, result);) |
938 | NOT_LP64(shll(result, 24);) // truncate upper 24 bits |
939 | NOT_LP64(sarl(result, 24);) // and sign-extend byte |
940 | jmp(done); |
941 | |
942 | bind(notByte); |
943 | cmpl(rcx, T_CHAR); |
944 | jcc(Assembler::notEqual, notChar); |
945 | LP64_ONLY(movzwl(result, result);) |
946 | NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits |
947 | jmp(done); |
948 | |
949 | bind(notChar); |
950 | // cmpl(rcx, T_SHORT); // all that's left |
951 | // jcc(Assembler::notEqual, done); |
952 | LP64_ONLY(movswl(result, result);) |
953 | NOT_LP64(shll(result, 16);) // truncate upper 16 bits |
954 | NOT_LP64(sarl(result, 16);) // and sign-extend short |
955 | |
956 | // Nothing to do for T_INT |
957 | bind(done); |
958 | } |
959 | |
960 | // remove activation |
961 | // |
962 | // Unlock the receiver if this is a synchronized method. |
963 | // Unlock any Java monitors from syncronized blocks. |
964 | // Remove the activation from the stack. |
965 | // |
966 | // If there are locked Java monitors |
967 | // If throw_monitor_exception |
968 | // throws IllegalMonitorStateException |
969 | // Else if install_monitor_exception |
970 | // installs IllegalMonitorStateException |
971 | // Else |
972 | // no error processing |
973 | void InterpreterMacroAssembler::remove_activation( |
974 | TosState state, |
975 | Register ret_addr, |
976 | bool throw_monitor_exception, |
977 | bool install_monitor_exception, |
978 | bool notify_jvmdi) { |
979 | // Note: Registers rdx xmm0 may be in use for the |
980 | // result check if synchronized method |
981 | Label unlocked, unlock, no_unlock; |
982 | |
983 | const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); |
984 | const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx); |
985 | const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx); |
986 | // monitor pointers need different register |
987 | // because rdx may have the result in it |
988 | NOT_LP64(get_thread(rcx);) |
989 | |
990 | // get the value of _do_not_unlock_if_synchronized into rdx |
991 | const Address do_not_unlock_if_synchronized(rthread, |
992 | in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); |
993 | movbool(rbx, do_not_unlock_if_synchronized); |
994 | movbool(do_not_unlock_if_synchronized, false); // reset the flag |
995 | |
996 | // get method access flags |
997 | movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); |
998 | movl(rcx, Address(rcx, Method::access_flags_offset())); |
999 | testl(rcx, JVM_ACC_SYNCHRONIZED); |
1000 | jcc(Assembler::zero, unlocked); |
1001 | |
1002 | // Don't unlock anything if the _do_not_unlock_if_synchronized flag |
1003 | // is set. |
1004 | testbool(rbx); |
1005 | jcc(Assembler::notZero, no_unlock); |
1006 | |
1007 | // unlock monitor |
1008 | push(state); // save result |
1009 | |
1010 | // BasicObjectLock will be first in list, since this is a |
1011 | // synchronized method. However, need to check that the object has |
1012 | // not been unlocked by an explicit monitorexit bytecode. |
1013 | const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * |
1014 | wordSize - (int) sizeof(BasicObjectLock)); |
1015 | // We use c_rarg1/rdx so that if we go slow path it will be the correct |
1016 | // register for unlock_object to pass to VM directly |
1017 | lea(robj, monitor); // address of first monitor |
1018 | |
1019 | movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes())); |
1020 | testptr(rax, rax); |
1021 | jcc(Assembler::notZero, unlock); |
1022 | |
1023 | pop(state); |
1024 | if (throw_monitor_exception) { |
1025 | // Entry already unlocked, need to throw exception |
1026 | NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow |
1027 | call_VM(noreg, CAST_FROM_FN_PTR(address, |
1028 | InterpreterRuntime::throw_illegal_monitor_state_exception)); |
1029 | should_not_reach_here(); |
1030 | } else { |
1031 | // Monitor already unlocked during a stack unroll. If requested, |
1032 | // install an illegal_monitor_state_exception. Continue with |
1033 | // stack unrolling. |
1034 | if (install_monitor_exception) { |
1035 | NOT_LP64(empty_FPU_stack();) |
1036 | call_VM(noreg, CAST_FROM_FN_PTR(address, |
1037 | InterpreterRuntime::new_illegal_monitor_state_exception)); |
1038 | } |
1039 | jmp(unlocked); |
1040 | } |
1041 | |
1042 | bind(unlock); |
1043 | unlock_object(robj); |
1044 | pop(state); |
1045 | |
1046 | // Check that for block-structured locking (i.e., that all locked |
1047 | // objects has been unlocked) |
1048 | bind(unlocked); |
1049 | |
1050 | // rax, rdx: Might contain return value |
1051 | |
1052 | // Check that all monitors are unlocked |
1053 | { |
1054 | Label loop, exception, entry, restart; |
1055 | const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; |
1056 | const Address monitor_block_top( |
1057 | rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); |
1058 | const Address monitor_block_bot( |
1059 | rbp, frame::interpreter_frame_initial_sp_offset * wordSize); |
1060 | |
1061 | bind(restart); |
1062 | // We use c_rarg1 so that if we go slow path it will be the correct |
1063 | // register for unlock_object to pass to VM directly |
1064 | movptr(rmon, monitor_block_top); // points to current entry, starting |
1065 | // with top-most entry |
1066 | lea(rbx, monitor_block_bot); // points to word before bottom of |
1067 | // monitor block |
1068 | jmp(entry); |
1069 | |
1070 | // Entry already locked, need to throw exception |
1071 | bind(exception); |
1072 | |
1073 | if (throw_monitor_exception) { |
1074 | // Throw exception |
1075 | NOT_LP64(empty_FPU_stack();) |
1076 | MacroAssembler::call_VM(noreg, |
1077 | CAST_FROM_FN_PTR(address, InterpreterRuntime:: |
1078 | throw_illegal_monitor_state_exception)); |
1079 | should_not_reach_here(); |
1080 | } else { |
1081 | // Stack unrolling. Unlock object and install illegal_monitor_exception. |
1082 | // Unlock does not block, so don't have to worry about the frame. |
1083 | // We don't have to preserve c_rarg1 since we are going to throw an exception. |
1084 | |
1085 | push(state); |
1086 | mov(robj, rmon); // nop if robj and rmon are the same |
1087 | unlock_object(robj); |
1088 | pop(state); |
1089 | |
1090 | if (install_monitor_exception) { |
1091 | NOT_LP64(empty_FPU_stack();) |
1092 | call_VM(noreg, CAST_FROM_FN_PTR(address, |
1093 | InterpreterRuntime:: |
1094 | new_illegal_monitor_state_exception)); |
1095 | } |
1096 | |
1097 | jmp(restart); |
1098 | } |
1099 | |
1100 | bind(loop); |
1101 | // check if current entry is used |
1102 | cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL); |
1103 | jcc(Assembler::notEqual, exception); |
1104 | |
1105 | addptr(rmon, entry_size); // otherwise advance to next entry |
1106 | bind(entry); |
1107 | cmpptr(rmon, rbx); // check if bottom reached |
1108 | jcc(Assembler::notEqual, loop); // if not at bottom then check this entry |
1109 | } |
1110 | |
1111 | bind(no_unlock); |
1112 | |
1113 | // jvmti support |
1114 | if (notify_jvmdi) { |
1115 | notify_method_exit(state, NotifyJVMTI); // preserve TOSCA |
1116 | } else { |
1117 | notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA |
1118 | } |
1119 | |
1120 | // remove activation |
1121 | // get sender sp |
1122 | movptr(rbx, |
1123 | Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); |
1124 | if (StackReservedPages > 0) { |
1125 | // testing if reserved zone needs to be re-enabled |
1126 | Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); |
1127 | Label no_reserved_zone_enabling; |
1128 | |
1129 | NOT_LP64(get_thread(rthread);) |
1130 | |
1131 | cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_enabled); |
1132 | jcc(Assembler::equal, no_reserved_zone_enabling); |
1133 | |
1134 | cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset())); |
1135 | jcc(Assembler::lessEqual, no_reserved_zone_enabling); |
1136 | |
1137 | call_VM_leaf( |
1138 | CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread); |
1139 | call_VM(noreg, CAST_FROM_FN_PTR(address, |
1140 | InterpreterRuntime::throw_delayed_StackOverflowError)); |
1141 | should_not_reach_here(); |
1142 | |
1143 | bind(no_reserved_zone_enabling); |
1144 | } |
1145 | leave(); // remove frame anchor |
1146 | pop(ret_addr); // get return address |
1147 | mov(rsp, rbx); // set sp to sender sp |
1148 | } |
1149 | |
1150 | void InterpreterMacroAssembler::get_method_counters(Register method, |
1151 | Register mcs, Label& skip) { |
1152 | Label has_counters; |
1153 | movptr(mcs, Address(method, Method::method_counters_offset())); |
1154 | testptr(mcs, mcs); |
1155 | jcc(Assembler::notZero, has_counters); |
1156 | call_VM(noreg, CAST_FROM_FN_PTR(address, |
1157 | InterpreterRuntime::build_method_counters), method); |
1158 | movptr(mcs, Address(method,Method::method_counters_offset())); |
1159 | testptr(mcs, mcs); |
1160 | jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory |
1161 | bind(has_counters); |
1162 | } |
1163 | |
1164 | |
1165 | // Lock object |
1166 | // |
1167 | // Args: |
1168 | // rdx, c_rarg1: BasicObjectLock to be used for locking |
1169 | // |
1170 | // Kills: |
1171 | // rax, rbx |
1172 | void InterpreterMacroAssembler::lock_object(Register lock_reg) { |
1173 | assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx), |
1174 | "The argument is only for looks. It must be c_rarg1" ); |
1175 | |
1176 | if (UseHeavyMonitors) { |
1177 | call_VM(noreg, |
1178 | CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), |
1179 | lock_reg); |
1180 | } else { |
1181 | Label done; |
1182 | |
1183 | const Register swap_reg = rax; // Must use rax for cmpxchg instruction |
1184 | const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a |
1185 | // problematic case where tmp_reg = no_reg. |
1186 | const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop |
1187 | |
1188 | const int obj_offset = BasicObjectLock::obj_offset_in_bytes(); |
1189 | const int lock_offset = BasicObjectLock::lock_offset_in_bytes (); |
1190 | const int mark_offset = lock_offset + |
1191 | BasicLock::displaced_header_offset_in_bytes(); |
1192 | |
1193 | Label slow_case; |
1194 | |
1195 | // Load object pointer into obj_reg |
1196 | movptr(obj_reg, Address(lock_reg, obj_offset)); |
1197 | |
1198 | if (UseBiasedLocking) { |
1199 | biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case); |
1200 | } |
1201 | |
1202 | // Load immediate 1 into swap_reg %rax |
1203 | movl(swap_reg, (int32_t)1); |
1204 | |
1205 | // Load (object->mark() | 1) into swap_reg %rax |
1206 | orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); |
1207 | |
1208 | // Save (object->mark() | 1) into BasicLock's displaced header |
1209 | movptr(Address(lock_reg, mark_offset), swap_reg); |
1210 | |
1211 | assert(lock_offset == 0, |
1212 | "displaced header must be first word in BasicObjectLock" ); |
1213 | |
1214 | lock(); |
1215 | cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); |
1216 | if (PrintBiasedLockingStatistics) { |
1217 | cond_inc32(Assembler::zero, |
1218 | ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr())); |
1219 | } |
1220 | jcc(Assembler::zero, done); |
1221 | |
1222 | const int zero_bits = LP64_ONLY(7) NOT_LP64(3); |
1223 | |
1224 | // Test if the oopMark is an obvious stack pointer, i.e., |
1225 | // 1) (mark & zero_bits) == 0, and |
1226 | // 2) rsp <= mark < mark + os::pagesize() |
1227 | // |
1228 | // These 3 tests can be done by evaluating the following |
1229 | // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())), |
1230 | // assuming both stack pointer and pagesize have their |
1231 | // least significant bits clear. |
1232 | // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg |
1233 | subptr(swap_reg, rsp); |
1234 | andptr(swap_reg, zero_bits - os::vm_page_size()); |
1235 | |
1236 | // Save the test result, for recursive case, the result is zero |
1237 | movptr(Address(lock_reg, mark_offset), swap_reg); |
1238 | |
1239 | if (PrintBiasedLockingStatistics) { |
1240 | cond_inc32(Assembler::zero, |
1241 | ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr())); |
1242 | } |
1243 | jcc(Assembler::zero, done); |
1244 | |
1245 | bind(slow_case); |
1246 | |
1247 | // Call the runtime routine for slow case |
1248 | call_VM(noreg, |
1249 | CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), |
1250 | lock_reg); |
1251 | |
1252 | bind(done); |
1253 | } |
1254 | } |
1255 | |
1256 | |
1257 | // Unlocks an object. Used in monitorexit bytecode and |
1258 | // remove_activation. Throws an IllegalMonitorException if object is |
1259 | // not locked by current thread. |
1260 | // |
1261 | // Args: |
1262 | // rdx, c_rarg1: BasicObjectLock for lock |
1263 | // |
1264 | // Kills: |
1265 | // rax |
1266 | // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs) |
1267 | // rscratch1 (scratch reg) |
1268 | // rax, rbx, rcx, rdx |
1269 | void InterpreterMacroAssembler::unlock_object(Register lock_reg) { |
1270 | assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx), |
1271 | "The argument is only for looks. It must be c_rarg1" ); |
1272 | |
1273 | if (UseHeavyMonitors) { |
1274 | call_VM(noreg, |
1275 | CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), |
1276 | lock_reg); |
1277 | } else { |
1278 | Label done; |
1279 | |
1280 | const Register swap_reg = rax; // Must use rax for cmpxchg instruction |
1281 | const Register = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark |
1282 | const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop |
1283 | |
1284 | save_bcp(); // Save in case of exception |
1285 | |
1286 | // Convert from BasicObjectLock structure to object and BasicLock |
1287 | // structure Store the BasicLock address into %rax |
1288 | lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes())); |
1289 | |
1290 | // Load oop into obj_reg(%c_rarg3) |
1291 | movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); |
1292 | |
1293 | // Free entry |
1294 | movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); |
1295 | |
1296 | if (UseBiasedLocking) { |
1297 | biased_locking_exit(obj_reg, header_reg, done); |
1298 | } |
1299 | |
1300 | // Load the old header from BasicLock structure |
1301 | movptr(header_reg, Address(swap_reg, |
1302 | BasicLock::displaced_header_offset_in_bytes())); |
1303 | |
1304 | // Test for recursion |
1305 | testptr(header_reg, header_reg); |
1306 | |
1307 | // zero for recursive case |
1308 | jcc(Assembler::zero, done); |
1309 | |
1310 | // Atomic swap back the old header |
1311 | lock(); |
1312 | cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); |
1313 | |
1314 | // zero for simple unlock of a stack-lock case |
1315 | jcc(Assembler::zero, done); |
1316 | |
1317 | // Call the runtime routine for slow case. |
1318 | movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), |
1319 | obj_reg); // restore obj |
1320 | call_VM(noreg, |
1321 | CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), |
1322 | lock_reg); |
1323 | |
1324 | bind(done); |
1325 | |
1326 | restore_bcp(); |
1327 | } |
1328 | } |
1329 | |
1330 | void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, |
1331 | Label& zero_continue) { |
1332 | assert(ProfileInterpreter, "must be profiling interpreter" ); |
1333 | movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize)); |
1334 | testptr(mdp, mdp); |
1335 | jcc(Assembler::zero, zero_continue); |
1336 | } |
1337 | |
1338 | |
1339 | // Set the method data pointer for the current bcp. |
1340 | void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { |
1341 | assert(ProfileInterpreter, "must be profiling interpreter" ); |
1342 | Label set_mdp; |
1343 | push(rax); |
1344 | push(rbx); |
1345 | |
1346 | get_method(rbx); |
1347 | // Test MDO to avoid the call if it is NULL. |
1348 | movptr(rax, Address(rbx, in_bytes(Method::method_data_offset()))); |
1349 | testptr(rax, rax); |
1350 | jcc(Assembler::zero, set_mdp); |
1351 | // rbx: method |
1352 | // _bcp_register: bcp |
1353 | call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register); |
1354 | // rax: mdi |
1355 | // mdo is guaranteed to be non-zero here, we checked for it before the call. |
1356 | movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset()))); |
1357 | addptr(rbx, in_bytes(MethodData::data_offset())); |
1358 | addptr(rax, rbx); |
1359 | bind(set_mdp); |
1360 | movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax); |
1361 | pop(rbx); |
1362 | pop(rax); |
1363 | } |
1364 | |
1365 | void InterpreterMacroAssembler::verify_method_data_pointer() { |
1366 | assert(ProfileInterpreter, "must be profiling interpreter" ); |
1367 | #ifdef ASSERT |
1368 | Label verify_continue; |
1369 | push(rax); |
1370 | push(rbx); |
1371 | Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); |
1372 | Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx); |
1373 | push(arg3_reg); |
1374 | push(arg2_reg); |
1375 | test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue |
1376 | get_method(rbx); |
1377 | |
1378 | // If the mdp is valid, it will point to a DataLayout header which is |
1379 | // consistent with the bcp. The converse is highly probable also. |
1380 | load_unsigned_short(arg2_reg, |
1381 | Address(arg3_reg, in_bytes(DataLayout::bci_offset()))); |
1382 | addptr(arg2_reg, Address(rbx, Method::const_offset())); |
1383 | lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset())); |
1384 | cmpptr(arg2_reg, _bcp_register); |
1385 | jcc(Assembler::equal, verify_continue); |
1386 | // rbx: method |
1387 | // _bcp_register: bcp |
1388 | // c_rarg3: mdp |
1389 | call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), |
1390 | rbx, _bcp_register, arg3_reg); |
1391 | bind(verify_continue); |
1392 | pop(arg2_reg); |
1393 | pop(arg3_reg); |
1394 | pop(rbx); |
1395 | pop(rax); |
1396 | #endif // ASSERT |
1397 | } |
1398 | |
1399 | |
1400 | void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, |
1401 | int constant, |
1402 | Register value) { |
1403 | assert(ProfileInterpreter, "must be profiling interpreter" ); |
1404 | Address data(mdp_in, constant); |
1405 | movptr(data, value); |
1406 | } |
1407 | |
1408 | |
1409 | void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, |
1410 | int constant, |
1411 | bool decrement) { |
1412 | // Counter address |
1413 | Address data(mdp_in, constant); |
1414 | |
1415 | increment_mdp_data_at(data, decrement); |
1416 | } |
1417 | |
1418 | void InterpreterMacroAssembler::increment_mdp_data_at(Address data, |
1419 | bool decrement) { |
1420 | assert(ProfileInterpreter, "must be profiling interpreter" ); |
1421 | // %%% this does 64bit counters at best it is wasting space |
1422 | // at worst it is a rare bug when counters overflow |
1423 | |
1424 | if (decrement) { |
1425 | // Decrement the register. Set condition codes. |
1426 | addptr(data, (int32_t) -DataLayout::counter_increment); |
1427 | // If the decrement causes the counter to overflow, stay negative |
1428 | Label L; |
1429 | jcc(Assembler::negative, L); |
1430 | addptr(data, (int32_t) DataLayout::counter_increment); |
1431 | bind(L); |
1432 | } else { |
1433 | assert(DataLayout::counter_increment == 1, |
1434 | "flow-free idiom only works with 1" ); |
1435 | // Increment the register. Set carry flag. |
1436 | addptr(data, DataLayout::counter_increment); |
1437 | // If the increment causes the counter to overflow, pull back by 1. |
1438 | sbbptr(data, (int32_t)0); |
1439 | } |
1440 | } |
1441 | |
1442 | |
1443 | void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, |
1444 | Register reg, |
1445 | int constant, |
1446 | bool decrement) { |
1447 | Address data(mdp_in, reg, Address::times_1, constant); |
1448 | |
1449 | increment_mdp_data_at(data, decrement); |
1450 | } |
1451 | |
1452 | void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, |
1453 | int flag_byte_constant) { |
1454 | assert(ProfileInterpreter, "must be profiling interpreter" ); |
1455 | int = in_bytes(DataLayout::flags_offset()); |
1456 | int = flag_byte_constant; |
1457 | // Set the flag |
1458 | orb(Address(mdp_in, header_offset), header_bits); |
1459 | } |
1460 | |
1461 | |
1462 | |
1463 | void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, |
1464 | int offset, |
1465 | Register value, |
1466 | Register test_value_out, |
1467 | Label& not_equal_continue) { |
1468 | assert(ProfileInterpreter, "must be profiling interpreter" ); |
1469 | if (test_value_out == noreg) { |
1470 | cmpptr(value, Address(mdp_in, offset)); |
1471 | } else { |
1472 | // Put the test value into a register, so caller can use it: |
1473 | movptr(test_value_out, Address(mdp_in, offset)); |
1474 | cmpptr(test_value_out, value); |
1475 | } |
1476 | jcc(Assembler::notEqual, not_equal_continue); |
1477 | } |
1478 | |
1479 | |
1480 | void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, |
1481 | int offset_of_disp) { |
1482 | assert(ProfileInterpreter, "must be profiling interpreter" ); |
1483 | Address disp_address(mdp_in, offset_of_disp); |
1484 | addptr(mdp_in, disp_address); |
1485 | movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); |
1486 | } |
1487 | |
1488 | |
1489 | void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, |
1490 | Register reg, |
1491 | int offset_of_disp) { |
1492 | assert(ProfileInterpreter, "must be profiling interpreter" ); |
1493 | Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp); |
1494 | addptr(mdp_in, disp_address); |
1495 | movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); |
1496 | } |
1497 | |
1498 | |
1499 | void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, |
1500 | int constant) { |
1501 | assert(ProfileInterpreter, "must be profiling interpreter" ); |
1502 | addptr(mdp_in, constant); |
1503 | movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); |
1504 | } |
1505 | |
1506 | |
1507 | void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { |
1508 | assert(ProfileInterpreter, "must be profiling interpreter" ); |
1509 | push(return_bci); // save/restore across call_VM |
1510 | call_VM(noreg, |
1511 | CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), |
1512 | return_bci); |
1513 | pop(return_bci); |
1514 | } |
1515 | |
1516 | |
1517 | void InterpreterMacroAssembler::profile_taken_branch(Register mdp, |
1518 | Register bumped_count) { |
1519 | if (ProfileInterpreter) { |
1520 | Label profile_continue; |
1521 | |
1522 | // If no method data exists, go to profile_continue. |
1523 | // Otherwise, assign to mdp |
1524 | test_method_data_pointer(mdp, profile_continue); |
1525 | |
1526 | // We are taking a branch. Increment the taken count. |
1527 | // We inline increment_mdp_data_at to return bumped_count in a register |
1528 | //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); |
1529 | Address data(mdp, in_bytes(JumpData::taken_offset())); |
1530 | movptr(bumped_count, data); |
1531 | assert(DataLayout::counter_increment == 1, |
1532 | "flow-free idiom only works with 1" ); |
1533 | addptr(bumped_count, DataLayout::counter_increment); |
1534 | sbbptr(bumped_count, 0); |
1535 | movptr(data, bumped_count); // Store back out |
1536 | |
1537 | // The method data pointer needs to be updated to reflect the new target. |
1538 | update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); |
1539 | bind(profile_continue); |
1540 | } |
1541 | } |
1542 | |
1543 | |
1544 | void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { |
1545 | if (ProfileInterpreter) { |
1546 | Label profile_continue; |
1547 | |
1548 | // If no method data exists, go to profile_continue. |
1549 | test_method_data_pointer(mdp, profile_continue); |
1550 | |
1551 | // We are taking a branch. Increment the not taken count. |
1552 | increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset())); |
1553 | |
1554 | // The method data pointer needs to be updated to correspond to |
1555 | // the next bytecode |
1556 | update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); |
1557 | bind(profile_continue); |
1558 | } |
1559 | } |
1560 | |
1561 | void InterpreterMacroAssembler::profile_call(Register mdp) { |
1562 | if (ProfileInterpreter) { |
1563 | Label profile_continue; |
1564 | |
1565 | // If no method data exists, go to profile_continue. |
1566 | test_method_data_pointer(mdp, profile_continue); |
1567 | |
1568 | // We are making a call. Increment the count. |
1569 | increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); |
1570 | |
1571 | // The method data pointer needs to be updated to reflect the new target. |
1572 | update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); |
1573 | bind(profile_continue); |
1574 | } |
1575 | } |
1576 | |
1577 | |
1578 | void InterpreterMacroAssembler::profile_final_call(Register mdp) { |
1579 | if (ProfileInterpreter) { |
1580 | Label profile_continue; |
1581 | |
1582 | // If no method data exists, go to profile_continue. |
1583 | test_method_data_pointer(mdp, profile_continue); |
1584 | |
1585 | // We are making a call. Increment the count. |
1586 | increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); |
1587 | |
1588 | // The method data pointer needs to be updated to reflect the new target. |
1589 | update_mdp_by_constant(mdp, |
1590 | in_bytes(VirtualCallData:: |
1591 | virtual_call_data_size())); |
1592 | bind(profile_continue); |
1593 | } |
1594 | } |
1595 | |
1596 | |
1597 | void InterpreterMacroAssembler::profile_virtual_call(Register receiver, |
1598 | Register mdp, |
1599 | Register reg2, |
1600 | bool receiver_can_be_null) { |
1601 | if (ProfileInterpreter) { |
1602 | Label profile_continue; |
1603 | |
1604 | // If no method data exists, go to profile_continue. |
1605 | test_method_data_pointer(mdp, profile_continue); |
1606 | |
1607 | Label skip_receiver_profile; |
1608 | if (receiver_can_be_null) { |
1609 | Label not_null; |
1610 | testptr(receiver, receiver); |
1611 | jccb(Assembler::notZero, not_null); |
1612 | // We are making a call. Increment the count for null receiver. |
1613 | increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); |
1614 | jmp(skip_receiver_profile); |
1615 | bind(not_null); |
1616 | } |
1617 | |
1618 | // Record the receiver type. |
1619 | record_klass_in_profile(receiver, mdp, reg2, true); |
1620 | bind(skip_receiver_profile); |
1621 | |
1622 | // The method data pointer needs to be updated to reflect the new target. |
1623 | #if INCLUDE_JVMCI |
1624 | if (MethodProfileWidth == 0) { |
1625 | update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); |
1626 | } |
1627 | #else // INCLUDE_JVMCI |
1628 | update_mdp_by_constant(mdp, |
1629 | in_bytes(VirtualCallData:: |
1630 | virtual_call_data_size())); |
1631 | #endif // INCLUDE_JVMCI |
1632 | bind(profile_continue); |
1633 | } |
1634 | } |
1635 | |
1636 | #if INCLUDE_JVMCI |
1637 | void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) { |
1638 | assert_different_registers(method, mdp, reg2); |
1639 | if (ProfileInterpreter && MethodProfileWidth > 0) { |
1640 | Label profile_continue; |
1641 | |
1642 | // If no method data exists, go to profile_continue. |
1643 | test_method_data_pointer(mdp, profile_continue); |
1644 | |
1645 | Label done; |
1646 | record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth, |
1647 | &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset())); |
1648 | bind(done); |
1649 | |
1650 | update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); |
1651 | bind(profile_continue); |
1652 | } |
1653 | } |
1654 | #endif // INCLUDE_JVMCI |
1655 | |
1656 | // This routine creates a state machine for updating the multi-row |
1657 | // type profile at a virtual call site (or other type-sensitive bytecode). |
1658 | // The machine visits each row (of receiver/count) until the receiver type |
1659 | // is found, or until it runs out of rows. At the same time, it remembers |
1660 | // the location of the first empty row. (An empty row records null for its |
1661 | // receiver, and can be allocated for a newly-observed receiver type.) |
1662 | // Because there are two degrees of freedom in the state, a simple linear |
1663 | // search will not work; it must be a decision tree. Hence this helper |
1664 | // function is recursive, to generate the required tree structured code. |
1665 | // It's the interpreter, so we are trading off code space for speed. |
1666 | // See below for example code. |
1667 | void InterpreterMacroAssembler::record_klass_in_profile_helper( |
1668 | Register receiver, Register mdp, |
1669 | Register reg2, int start_row, |
1670 | Label& done, bool is_virtual_call) { |
1671 | if (TypeProfileWidth == 0) { |
1672 | if (is_virtual_call) { |
1673 | increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); |
1674 | } |
1675 | #if INCLUDE_JVMCI |
1676 | else if (EnableJVMCI) { |
1677 | increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset())); |
1678 | } |
1679 | #endif // INCLUDE_JVMCI |
1680 | } else { |
1681 | int non_profiled_offset = -1; |
1682 | if (is_virtual_call) { |
1683 | non_profiled_offset = in_bytes(CounterData::count_offset()); |
1684 | } |
1685 | #if INCLUDE_JVMCI |
1686 | else if (EnableJVMCI) { |
1687 | non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()); |
1688 | } |
1689 | #endif // INCLUDE_JVMCI |
1690 | |
1691 | record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth, |
1692 | &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset); |
1693 | } |
1694 | } |
1695 | |
1696 | void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, |
1697 | Register reg2, int start_row, Label& done, int total_rows, |
1698 | OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, |
1699 | int non_profiled_offset) { |
1700 | int last_row = total_rows - 1; |
1701 | assert(start_row <= last_row, "must be work left to do" ); |
1702 | // Test this row for both the item and for null. |
1703 | // Take any of three different outcomes: |
1704 | // 1. found item => increment count and goto done |
1705 | // 2. found null => keep looking for case 1, maybe allocate this cell |
1706 | // 3. found something else => keep looking for cases 1 and 2 |
1707 | // Case 3 is handled by a recursive call. |
1708 | for (int row = start_row; row <= last_row; row++) { |
1709 | Label next_test; |
1710 | bool test_for_null_also = (row == start_row); |
1711 | |
1712 | // See if the item is item[n]. |
1713 | int item_offset = in_bytes(item_offset_fn(row)); |
1714 | test_mdp_data_at(mdp, item_offset, item, |
1715 | (test_for_null_also ? reg2 : noreg), |
1716 | next_test); |
1717 | // (Reg2 now contains the item from the CallData.) |
1718 | |
1719 | // The item is item[n]. Increment count[n]. |
1720 | int count_offset = in_bytes(item_count_offset_fn(row)); |
1721 | increment_mdp_data_at(mdp, count_offset); |
1722 | jmp(done); |
1723 | bind(next_test); |
1724 | |
1725 | if (test_for_null_also) { |
1726 | // Failed the equality check on item[n]... Test for null. |
1727 | testptr(reg2, reg2); |
1728 | if (start_row == last_row) { |
1729 | // The only thing left to do is handle the null case. |
1730 | if (non_profiled_offset >= 0) { |
1731 | Label found_null; |
1732 | jccb(Assembler::zero, found_null); |
1733 | // Item did not match any saved item and there is no empty row for it. |
1734 | // Increment total counter to indicate polymorphic case. |
1735 | increment_mdp_data_at(mdp, non_profiled_offset); |
1736 | jmp(done); |
1737 | bind(found_null); |
1738 | } else { |
1739 | jcc(Assembler::notZero, done); |
1740 | } |
1741 | break; |
1742 | } |
1743 | Label found_null; |
1744 | // Since null is rare, make it be the branch-taken case. |
1745 | jcc(Assembler::zero, found_null); |
1746 | |
1747 | // Put all the "Case 3" tests here. |
1748 | record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows, |
1749 | item_offset_fn, item_count_offset_fn, non_profiled_offset); |
1750 | |
1751 | // Found a null. Keep searching for a matching item, |
1752 | // but remember that this is an empty (unused) slot. |
1753 | bind(found_null); |
1754 | } |
1755 | } |
1756 | |
1757 | // In the fall-through case, we found no matching item, but we |
1758 | // observed the item[start_row] is NULL. |
1759 | |
1760 | // Fill in the item field and increment the count. |
1761 | int item_offset = in_bytes(item_offset_fn(start_row)); |
1762 | set_mdp_data_at(mdp, item_offset, item); |
1763 | int count_offset = in_bytes(item_count_offset_fn(start_row)); |
1764 | movl(reg2, DataLayout::counter_increment); |
1765 | set_mdp_data_at(mdp, count_offset, reg2); |
1766 | if (start_row > 0) { |
1767 | jmp(done); |
1768 | } |
1769 | } |
1770 | |
1771 | // Example state machine code for three profile rows: |
1772 | // // main copy of decision tree, rooted at row[1] |
1773 | // if (row[0].rec == rec) { row[0].incr(); goto done; } |
1774 | // if (row[0].rec != NULL) { |
1775 | // // inner copy of decision tree, rooted at row[1] |
1776 | // if (row[1].rec == rec) { row[1].incr(); goto done; } |
1777 | // if (row[1].rec != NULL) { |
1778 | // // degenerate decision tree, rooted at row[2] |
1779 | // if (row[2].rec == rec) { row[2].incr(); goto done; } |
1780 | // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow |
1781 | // row[2].init(rec); goto done; |
1782 | // } else { |
1783 | // // remember row[1] is empty |
1784 | // if (row[2].rec == rec) { row[2].incr(); goto done; } |
1785 | // row[1].init(rec); goto done; |
1786 | // } |
1787 | // } else { |
1788 | // // remember row[0] is empty |
1789 | // if (row[1].rec == rec) { row[1].incr(); goto done; } |
1790 | // if (row[2].rec == rec) { row[2].incr(); goto done; } |
1791 | // row[0].init(rec); goto done; |
1792 | // } |
1793 | // done: |
1794 | |
1795 | void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, |
1796 | Register mdp, Register reg2, |
1797 | bool is_virtual_call) { |
1798 | assert(ProfileInterpreter, "must be profiling" ); |
1799 | Label done; |
1800 | |
1801 | record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); |
1802 | |
1803 | bind (done); |
1804 | } |
1805 | |
1806 | void InterpreterMacroAssembler::profile_ret(Register return_bci, |
1807 | Register mdp) { |
1808 | if (ProfileInterpreter) { |
1809 | Label profile_continue; |
1810 | uint row; |
1811 | |
1812 | // If no method data exists, go to profile_continue. |
1813 | test_method_data_pointer(mdp, profile_continue); |
1814 | |
1815 | // Update the total ret count. |
1816 | increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); |
1817 | |
1818 | for (row = 0; row < RetData::row_limit(); row++) { |
1819 | Label next_test; |
1820 | |
1821 | // See if return_bci is equal to bci[n]: |
1822 | test_mdp_data_at(mdp, |
1823 | in_bytes(RetData::bci_offset(row)), |
1824 | return_bci, noreg, |
1825 | next_test); |
1826 | |
1827 | // return_bci is equal to bci[n]. Increment the count. |
1828 | increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); |
1829 | |
1830 | // The method data pointer needs to be updated to reflect the new target. |
1831 | update_mdp_by_offset(mdp, |
1832 | in_bytes(RetData::bci_displacement_offset(row))); |
1833 | jmp(profile_continue); |
1834 | bind(next_test); |
1835 | } |
1836 | |
1837 | update_mdp_for_ret(return_bci); |
1838 | |
1839 | bind(profile_continue); |
1840 | } |
1841 | } |
1842 | |
1843 | |
1844 | void InterpreterMacroAssembler::profile_null_seen(Register mdp) { |
1845 | if (ProfileInterpreter) { |
1846 | Label profile_continue; |
1847 | |
1848 | // If no method data exists, go to profile_continue. |
1849 | test_method_data_pointer(mdp, profile_continue); |
1850 | |
1851 | set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); |
1852 | |
1853 | // The method data pointer needs to be updated. |
1854 | int mdp_delta = in_bytes(BitData::bit_data_size()); |
1855 | if (TypeProfileCasts) { |
1856 | mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); |
1857 | } |
1858 | update_mdp_by_constant(mdp, mdp_delta); |
1859 | |
1860 | bind(profile_continue); |
1861 | } |
1862 | } |
1863 | |
1864 | |
1865 | void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) { |
1866 | if (ProfileInterpreter && TypeProfileCasts) { |
1867 | Label profile_continue; |
1868 | |
1869 | // If no method data exists, go to profile_continue. |
1870 | test_method_data_pointer(mdp, profile_continue); |
1871 | |
1872 | int count_offset = in_bytes(CounterData::count_offset()); |
1873 | // Back up the address, since we have already bumped the mdp. |
1874 | count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); |
1875 | |
1876 | // *Decrement* the counter. We expect to see zero or small negatives. |
1877 | increment_mdp_data_at(mdp, count_offset, true); |
1878 | |
1879 | bind (profile_continue); |
1880 | } |
1881 | } |
1882 | |
1883 | |
1884 | void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) { |
1885 | if (ProfileInterpreter) { |
1886 | Label profile_continue; |
1887 | |
1888 | // If no method data exists, go to profile_continue. |
1889 | test_method_data_pointer(mdp, profile_continue); |
1890 | |
1891 | // The method data pointer needs to be updated. |
1892 | int mdp_delta = in_bytes(BitData::bit_data_size()); |
1893 | if (TypeProfileCasts) { |
1894 | mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); |
1895 | |
1896 | // Record the object type. |
1897 | record_klass_in_profile(klass, mdp, reg2, false); |
1898 | NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg" );) |
1899 | NOT_LP64(restore_locals();) // Restore EDI |
1900 | } |
1901 | update_mdp_by_constant(mdp, mdp_delta); |
1902 | |
1903 | bind(profile_continue); |
1904 | } |
1905 | } |
1906 | |
1907 | |
1908 | void InterpreterMacroAssembler::profile_switch_default(Register mdp) { |
1909 | if (ProfileInterpreter) { |
1910 | Label profile_continue; |
1911 | |
1912 | // If no method data exists, go to profile_continue. |
1913 | test_method_data_pointer(mdp, profile_continue); |
1914 | |
1915 | // Update the default case count |
1916 | increment_mdp_data_at(mdp, |
1917 | in_bytes(MultiBranchData::default_count_offset())); |
1918 | |
1919 | // The method data pointer needs to be updated. |
1920 | update_mdp_by_offset(mdp, |
1921 | in_bytes(MultiBranchData:: |
1922 | default_displacement_offset())); |
1923 | |
1924 | bind(profile_continue); |
1925 | } |
1926 | } |
1927 | |
1928 | |
1929 | void InterpreterMacroAssembler::profile_switch_case(Register index, |
1930 | Register mdp, |
1931 | Register reg2) { |
1932 | if (ProfileInterpreter) { |
1933 | Label profile_continue; |
1934 | |
1935 | // If no method data exists, go to profile_continue. |
1936 | test_method_data_pointer(mdp, profile_continue); |
1937 | |
1938 | // Build the base (index * per_case_size_in_bytes()) + |
1939 | // case_array_offset_in_bytes() |
1940 | movl(reg2, in_bytes(MultiBranchData::per_case_size())); |
1941 | imulptr(index, reg2); // XXX l ? |
1942 | addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ? |
1943 | |
1944 | // Update the case count |
1945 | increment_mdp_data_at(mdp, |
1946 | index, |
1947 | in_bytes(MultiBranchData::relative_count_offset())); |
1948 | |
1949 | // The method data pointer needs to be updated. |
1950 | update_mdp_by_offset(mdp, |
1951 | index, |
1952 | in_bytes(MultiBranchData:: |
1953 | relative_displacement_offset())); |
1954 | |
1955 | bind(profile_continue); |
1956 | } |
1957 | } |
1958 | |
1959 | |
1960 | |
1961 | void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { |
1962 | if (state == atos) { |
1963 | MacroAssembler::verify_oop(reg); |
1964 | } |
1965 | } |
1966 | |
1967 | void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { |
1968 | #ifndef _LP64 |
1969 | if ((state == ftos && UseSSE < 1) || |
1970 | (state == dtos && UseSSE < 2)) { |
1971 | MacroAssembler::verify_FPU(stack_depth); |
1972 | } |
1973 | #endif |
1974 | } |
1975 | |
1976 | // Jump if ((*counter_addr += increment) & mask) satisfies the condition. |
1977 | void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, |
1978 | int increment, Address mask, |
1979 | Register scratch, bool preloaded, |
1980 | Condition cond, Label* where) { |
1981 | if (!preloaded) { |
1982 | movl(scratch, counter_addr); |
1983 | } |
1984 | incrementl(scratch, increment); |
1985 | movl(counter_addr, scratch); |
1986 | andl(scratch, mask); |
1987 | if (where != NULL) { |
1988 | jcc(cond, *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 | { |
2010 | SkipIfEqual skip(this, &DTraceMethodProbes, false); |
2011 | NOT_LP64(get_thread(rthread);) |
2012 | get_method(rarg); |
2013 | call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), |
2014 | rthread, rarg); |
2015 | } |
2016 | |
2017 | // RedefineClasses() tracing support for obsolete method entry |
2018 | if (log_is_enabled(Trace, redefine, class, obsolete)) { |
2019 | NOT_LP64(get_thread(rthread);) |
2020 | get_method(rarg); |
2021 | call_VM_leaf( |
2022 | CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), |
2023 | rthread, rarg); |
2024 | } |
2025 | } |
2026 | |
2027 | |
2028 | void InterpreterMacroAssembler::notify_method_exit( |
2029 | TosState state, NotifyMethodExitMode mode) { |
2030 | // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to |
2031 | // track stack depth. If it is possible to enter interp_only_mode we add |
2032 | // the code to check if the event should be sent. |
2033 | Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); |
2034 | Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx); |
2035 | if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { |
2036 | Label L; |
2037 | // Note: frame::interpreter_frame_result has a dependency on how the |
2038 | // method result is saved across the call to post_method_exit. If this |
2039 | // is changed then the interpreter_frame_result implementation will |
2040 | // need to be updated too. |
2041 | |
2042 | // template interpreter will leave the result on the top of the stack. |
2043 | push(state); |
2044 | NOT_LP64(get_thread(rthread);) |
2045 | movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset())); |
2046 | testl(rdx, rdx); |
2047 | jcc(Assembler::zero, L); |
2048 | call_VM(noreg, |
2049 | CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); |
2050 | bind(L); |
2051 | pop(state); |
2052 | } |
2053 | |
2054 | { |
2055 | SkipIfEqual skip(this, &DTraceMethodProbes, false); |
2056 | push(state); |
2057 | NOT_LP64(get_thread(rthread);) |
2058 | get_method(rarg); |
2059 | call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), |
2060 | rthread, rarg); |
2061 | pop(state); |
2062 | } |
2063 | } |
2064 | |