Print this page
| Split |
Close |
| Expand all |
| Collapse all |
--- old/src/cpu/x86/vm/interp_masm_x86_64.cpp
+++ new/src/cpu/x86/vm/interp_masm_x86_64.cpp
1 1 /*
2 2 * Copyright (c) 2003, 2011, Oracle and/or its affiliates. All rights reserved.
3 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 4 *
5 5 * This code is free software; you can redistribute it and/or modify it
6 6 * under the terms of the GNU General Public License version 2 only, as
7 7 * published by the Free Software Foundation.
8 8 *
9 9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 12 * version 2 for more details (a copy is included in the LICENSE file that
13 13 * accompanied this code).
14 14 *
15 15 * You should have received a copy of the GNU General Public License version
16 16 * 2 along with this work; if not, write to the Free Software Foundation,
17 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 18 *
19 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 20 * or visit www.oracle.com if you need additional information or have any
21 21 * questions.
22 22 *
23 23 */
24 24
25 25 #include "precompiled.hpp"
26 26 #include "interp_masm_x86_64.hpp"
27 27 #include "interpreter/interpreter.hpp"
28 28 #include "interpreter/interpreterRuntime.hpp"
29 29 #include "oops/arrayOop.hpp"
30 30 #include "oops/markOop.hpp"
31 31 #include "oops/methodDataOop.hpp"
32 32 #include "oops/methodOop.hpp"
33 33 #include "prims/jvmtiExport.hpp"
34 34 #include "prims/jvmtiRedefineClassesTrace.hpp"
35 35 #include "prims/jvmtiThreadState.hpp"
36 36 #include "runtime/basicLock.hpp"
37 37 #include "runtime/biasedLocking.hpp"
38 38 #include "runtime/sharedRuntime.hpp"
39 39 #ifdef TARGET_OS_FAMILY_linux
40 40 # include "thread_linux.inline.hpp"
41 41 #endif
42 42 #ifdef TARGET_OS_FAMILY_solaris
43 43 # include "thread_solaris.inline.hpp"
44 44 #endif
45 45 #ifdef TARGET_OS_FAMILY_windows
46 46 # include "thread_windows.inline.hpp"
47 47 #endif
48 48
49 49
50 50 // Implementation of InterpreterMacroAssembler
51 51
52 52 #ifdef CC_INTERP
53 53 void InterpreterMacroAssembler::get_method(Register reg) {
54 54 movptr(reg, Address(rbp, -((int)sizeof(BytecodeInterpreter) + 2 * wordSize)));
55 55 movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method)));
56 56 }
57 57 #endif // CC_INTERP
58 58
59 59 #ifndef CC_INTERP
60 60
61 61 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
62 62 int number_of_arguments) {
63 63 // interpreter specific
64 64 //
65 65 // Note: No need to save/restore bcp & locals (r13 & r14) pointer
66 66 // since these are callee saved registers and no blocking/
67 67 // GC can happen in leaf calls.
68 68 // Further Note: DO NOT save/restore bcp/locals. If a caller has
69 69 // already saved them so that it can use esi/edi as temporaries
70 70 // then a save/restore here will DESTROY the copy the caller
71 71 // saved! There used to be a save_bcp() that only happened in
72 72 // the ASSERT path (no restore_bcp). Which caused bizarre failures
73 73 // when jvm built with ASSERTs.
74 74 #ifdef ASSERT
75 75 {
76 76 Label L;
77 77 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
78 78 jcc(Assembler::equal, L);
79 79 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
80 80 " last_sp != NULL");
81 81 bind(L);
82 82 }
83 83 #endif
84 84 // super call
85 85 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
86 86 // interpreter specific
87 87 // Used to ASSERT that r13/r14 were equal to frame's bcp/locals
88 88 // but since they may not have been saved (and we don't want to
89 89 // save thme here (see note above) the assert is invalid.
90 90 }
91 91
92 92 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
93 93 Register java_thread,
94 94 Register last_java_sp,
95 95 address entry_point,
96 96 int number_of_arguments,
97 97 bool check_exceptions) {
98 98 // interpreter specific
99 99 //
100 100 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
101 101 // really make a difference for these runtime calls, since they are
102 102 // slow anyway. Btw., bcp must be saved/restored since it may change
103 103 // due to GC.
104 104 // assert(java_thread == noreg , "not expecting a precomputed java thread");
105 105 save_bcp();
106 106 #ifdef ASSERT
107 107 {
108 108 Label L;
109 109 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
110 110 jcc(Assembler::equal, L);
111 111 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
112 112 " last_sp != NULL");
113 113 bind(L);
114 114 }
115 115 #endif /* ASSERT */
116 116 // super call
117 117 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
118 118 entry_point, number_of_arguments,
119 119 check_exceptions);
120 120 // interpreter specific
121 121 restore_bcp();
122 122 restore_locals();
123 123 }
124 124
125 125
126 126 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
127 127 if (JvmtiExport::can_pop_frame()) {
128 128 Label L;
129 129 // Initiate popframe handling only if it is not already being
130 130 // processed. If the flag has the popframe_processing bit set, it
131 131 // means that this code is called *during* popframe handling - we
132 132 // don't want to reenter.
133 133 // This method is only called just after the call into the vm in
134 134 // call_VM_base, so the arg registers are available.
135 135 movl(c_rarg0, Address(r15_thread, JavaThread::popframe_condition_offset()));
136 136 testl(c_rarg0, JavaThread::popframe_pending_bit);
137 137 jcc(Assembler::zero, L);
138 138 testl(c_rarg0, JavaThread::popframe_processing_bit);
139 139 jcc(Assembler::notZero, L);
140 140 // Call Interpreter::remove_activation_preserving_args_entry() to get the
141 141 // address of the same-named entrypoint in the generated interpreter code.
142 142 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
143 143 jmp(rax);
144 144 bind(L);
145 145 }
146 146 }
147 147
148 148
149 149 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
150 150 movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
151 151 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
152 152 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
153 153 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
154 154 switch (state) {
155 155 case atos: movptr(rax, oop_addr);
156 156 movptr(oop_addr, (int32_t)NULL_WORD);
157 157 verify_oop(rax, state); break;
158 158 case ltos: movptr(rax, val_addr); break;
159 159 case btos: // fall through
160 160 case ctos: // fall through
161 161 case stos: // fall through
162 162 case itos: movl(rax, val_addr); break;
163 163 case ftos: movflt(xmm0, val_addr); break;
164 164 case dtos: movdbl(xmm0, val_addr); break;
165 165 case vtos: /* nothing to do */ break;
166 166 default : ShouldNotReachHere();
167 167 }
168 168 // Clean up tos value in the thread object
169 169 movl(tos_addr, (int) ilgl);
170 170 movl(val_addr, (int32_t) NULL_WORD);
171 171 }
172 172
173 173
174 174 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
175 175 if (JvmtiExport::can_force_early_return()) {
176 176 Label L;
177 177 movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
178 178 testptr(c_rarg0, c_rarg0);
179 179 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
180 180
181 181 // Initiate earlyret handling only if it is not already being processed.
182 182 // If the flag has the earlyret_processing bit set, it means that this code
183 183 // is called *during* earlyret handling - we don't want to reenter.
184 184 movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_state_offset()));
185 185 cmpl(c_rarg0, JvmtiThreadState::earlyret_pending);
186 186 jcc(Assembler::notEqual, L);
187 187
188 188 // Call Interpreter::remove_activation_early_entry() to get the address of the
189 189 // same-named entrypoint in the generated interpreter code.
190 190 movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
191 191 movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_tos_offset()));
192 192 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), c_rarg0);
193 193 jmp(rax);
194 194 bind(L);
195 195 }
196 196 }
197 197
198 198
199 199 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
200 200 Register reg,
201 201 int bcp_offset) {
202 202 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
203 203 movl(reg, Address(r13, bcp_offset));
204 204 bswapl(reg);
205 205 shrl(reg, 16);
|
↓ open down ↓ |
205 lines elided |
↑ open up ↑ |
206 206 }
207 207
208 208
209 209 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
210 210 int bcp_offset,
211 211 size_t index_size) {
212 212 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
213 213 if (index_size == sizeof(u2)) {
214 214 load_unsigned_short(index, Address(r13, bcp_offset));
215 215 } else if (index_size == sizeof(u4)) {
216 - assert(EnableInvokeDynamic, "giant index used only for EnableInvokeDynamic");
216 + assert(EnableInvokeDynamic, "giant index used only for JSR 292");
217 217 movl(index, Address(r13, bcp_offset));
218 218 // Check if the secondary index definition is still ~x, otherwise
219 219 // we have to change the following assembler code to calculate the
220 220 // plain index.
221 221 assert(constantPoolCacheOopDesc::decode_secondary_index(~123) == 123, "else change next line");
222 222 notl(index); // convert to plain index
223 223 } else if (index_size == sizeof(u1)) {
224 - assert(EnableMethodHandles, "tiny index used only for EnableMethodHandles");
224 + assert(EnableInvokeDynamic, "tiny index used only for JSR 292");
225 225 load_unsigned_byte(index, Address(r13, bcp_offset));
226 226 } else {
227 227 ShouldNotReachHere();
228 228 }
229 229 }
230 230
231 231
232 232 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
233 233 Register index,
234 234 int bcp_offset,
235 235 size_t index_size) {
236 236 assert(cache != index, "must use different registers");
237 237 get_cache_index_at_bcp(index, bcp_offset, index_size);
238 238 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
239 239 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
240 240 // convert from field index to ConstantPoolCacheEntry index
241 241 shll(index, 2);
242 242 }
243 243
244 244
245 245 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
246 246 Register tmp,
247 247 int bcp_offset,
248 248 size_t index_size) {
249 249 assert(cache != tmp, "must use different register");
250 250 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
251 251 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
252 252 // convert from field index to ConstantPoolCacheEntry index
253 253 // and from word offset to byte offset
254 254 shll(tmp, 2 + LogBytesPerWord);
255 255 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
256 256 // skip past the header
257 257 addptr(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
258 258 addptr(cache, tmp); // construct pointer to cache entry
259 259 }
260 260
261 261
262 262 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
263 263 // subtype of super_klass.
264 264 //
265 265 // Args:
266 266 // rax: superklass
267 267 // Rsub_klass: subklass
268 268 //
269 269 // Kills:
270 270 // rcx, rdi
271 271 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
272 272 Label& ok_is_subtype) {
273 273 assert(Rsub_klass != rax, "rax holds superklass");
274 274 assert(Rsub_klass != r14, "r14 holds locals");
275 275 assert(Rsub_klass != r13, "r13 holds bcp");
276 276 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
277 277 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
278 278
279 279 // Profile the not-null value's klass.
280 280 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
281 281
282 282 // Do the check.
283 283 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
284 284
285 285 // Profile the failure of the check.
286 286 profile_typecheck_failed(rcx); // blows rcx
287 287 }
288 288
289 289
290 290
291 291 // Java Expression Stack
292 292
293 293 void InterpreterMacroAssembler::pop_ptr(Register r) {
294 294 pop(r);
295 295 }
296 296
297 297 void InterpreterMacroAssembler::pop_i(Register r) {
298 298 // XXX can't use pop currently, upper half non clean
299 299 movl(r, Address(rsp, 0));
300 300 addptr(rsp, wordSize);
301 301 }
302 302
303 303 void InterpreterMacroAssembler::pop_l(Register r) {
304 304 movq(r, Address(rsp, 0));
305 305 addptr(rsp, 2 * Interpreter::stackElementSize);
306 306 }
307 307
308 308 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
309 309 movflt(r, Address(rsp, 0));
310 310 addptr(rsp, wordSize);
311 311 }
312 312
313 313 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
314 314 movdbl(r, Address(rsp, 0));
315 315 addptr(rsp, 2 * Interpreter::stackElementSize);
316 316 }
317 317
318 318 void InterpreterMacroAssembler::push_ptr(Register r) {
319 319 push(r);
320 320 }
321 321
322 322 void InterpreterMacroAssembler::push_i(Register r) {
323 323 push(r);
324 324 }
325 325
326 326 void InterpreterMacroAssembler::push_l(Register r) {
327 327 subptr(rsp, 2 * wordSize);
328 328 movq(Address(rsp, 0), r);
329 329 }
330 330
331 331 void InterpreterMacroAssembler::push_f(XMMRegister r) {
332 332 subptr(rsp, wordSize);
333 333 movflt(Address(rsp, 0), r);
334 334 }
335 335
336 336 void InterpreterMacroAssembler::push_d(XMMRegister r) {
337 337 subptr(rsp, 2 * wordSize);
338 338 movdbl(Address(rsp, 0), r);
339 339 }
340 340
341 341 void InterpreterMacroAssembler::pop(TosState state) {
342 342 switch (state) {
343 343 case atos: pop_ptr(); break;
344 344 case btos:
345 345 case ctos:
346 346 case stos:
347 347 case itos: pop_i(); break;
348 348 case ltos: pop_l(); break;
349 349 case ftos: pop_f(); break;
350 350 case dtos: pop_d(); break;
351 351 case vtos: /* nothing to do */ break;
352 352 default: ShouldNotReachHere();
353 353 }
354 354 verify_oop(rax, state);
355 355 }
356 356
357 357 void InterpreterMacroAssembler::push(TosState state) {
358 358 verify_oop(rax, state);
359 359 switch (state) {
360 360 case atos: push_ptr(); break;
361 361 case btos:
362 362 case ctos:
363 363 case stos:
364 364 case itos: push_i(); break;
365 365 case ltos: push_l(); break;
366 366 case ftos: push_f(); break;
367 367 case dtos: push_d(); break;
368 368 case vtos: /* nothing to do */ break;
369 369 default : ShouldNotReachHere();
370 370 }
371 371 }
372 372
373 373
374 374 // Helpers for swap and dup
375 375 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
376 376 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
377 377 }
378 378
379 379 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
380 380 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
381 381 }
382 382
383 383
384 384 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point) {
385 385 MacroAssembler::call_VM_leaf_base(entry_point, 0);
386 386 }
387 387
388 388
389 389 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
390 390 Register arg_1) {
391 391 if (c_rarg0 != arg_1) {
392 392 mov(c_rarg0, arg_1);
393 393 }
394 394 MacroAssembler::call_VM_leaf_base(entry_point, 1);
395 395 }
396 396
397 397
398 398 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
399 399 Register arg_1,
400 400 Register arg_2) {
401 401 assert(c_rarg0 != arg_2, "smashed argument");
402 402 assert(c_rarg1 != arg_1, "smashed argument");
403 403 if (c_rarg0 != arg_1) {
404 404 mov(c_rarg0, arg_1);
405 405 }
406 406 if (c_rarg1 != arg_2) {
407 407 mov(c_rarg1, arg_2);
408 408 }
409 409 MacroAssembler::call_VM_leaf_base(entry_point, 2);
410 410 }
411 411
412 412 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
413 413 Register arg_1,
414 414 Register arg_2,
415 415 Register arg_3) {
416 416 assert(c_rarg0 != arg_2, "smashed argument");
417 417 assert(c_rarg0 != arg_3, "smashed argument");
418 418 assert(c_rarg1 != arg_1, "smashed argument");
419 419 assert(c_rarg1 != arg_3, "smashed argument");
420 420 assert(c_rarg2 != arg_1, "smashed argument");
421 421 assert(c_rarg2 != arg_2, "smashed argument");
422 422 if (c_rarg0 != arg_1) {
423 423 mov(c_rarg0, arg_1);
424 424 }
425 425 if (c_rarg1 != arg_2) {
426 426 mov(c_rarg1, arg_2);
427 427 }
428 428 if (c_rarg2 != arg_3) {
429 429 mov(c_rarg2, arg_3);
430 430 }
431 431 MacroAssembler::call_VM_leaf_base(entry_point, 3);
432 432 }
433 433
434 434 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
435 435 // set sender sp
436 436 lea(r13, Address(rsp, wordSize));
437 437 // record last_sp
438 438 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), r13);
439 439 }
440 440
441 441
442 442 // Jump to from_interpreted entry of a call unless single stepping is possible
443 443 // in this thread in which case we must call the i2i entry
444 444 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
445 445 prepare_to_jump_from_interpreted();
446 446
447 447 if (JvmtiExport::can_post_interpreter_events()) {
448 448 Label run_compiled_code;
449 449 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
450 450 // compiled code in threads for which the event is enabled. Check here for
451 451 // interp_only_mode if these events CAN be enabled.
452 452 // interp_only is an int, on little endian it is sufficient to test the byte only
453 453 // Is a cmpl faster?
454 454 cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0);
455 455 jcc(Assembler::zero, run_compiled_code);
456 456 jmp(Address(method, methodOopDesc::interpreter_entry_offset()));
457 457 bind(run_compiled_code);
458 458 }
459 459
460 460 jmp(Address(method, methodOopDesc::from_interpreted_offset()));
461 461
462 462 }
463 463
464 464
465 465 // The following two routines provide a hook so that an implementation
466 466 // can schedule the dispatch in two parts. amd64 does not do this.
467 467 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
468 468 // Nothing amd64 specific to be done here
469 469 }
470 470
471 471 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
472 472 dispatch_next(state, step);
473 473 }
474 474
475 475 void InterpreterMacroAssembler::dispatch_base(TosState state,
476 476 address* table,
477 477 bool verifyoop) {
478 478 verify_FPU(1, state);
479 479 if (VerifyActivationFrameSize) {
480 480 Label L;
481 481 mov(rcx, rbp);
482 482 subptr(rcx, rsp);
483 483 int32_t min_frame_size =
484 484 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
485 485 wordSize;
486 486 cmpptr(rcx, (int32_t)min_frame_size);
487 487 jcc(Assembler::greaterEqual, L);
488 488 stop("broken stack frame");
489 489 bind(L);
490 490 }
491 491 if (verifyoop) {
492 492 verify_oop(rax, state);
493 493 }
494 494 lea(rscratch1, ExternalAddress((address)table));
495 495 jmp(Address(rscratch1, rbx, Address::times_8));
496 496 }
497 497
498 498 void InterpreterMacroAssembler::dispatch_only(TosState state) {
499 499 dispatch_base(state, Interpreter::dispatch_table(state));
500 500 }
501 501
502 502 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
503 503 dispatch_base(state, Interpreter::normal_table(state));
504 504 }
505 505
506 506 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
507 507 dispatch_base(state, Interpreter::normal_table(state), false);
508 508 }
509 509
510 510
511 511 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
512 512 // load next bytecode (load before advancing r13 to prevent AGI)
513 513 load_unsigned_byte(rbx, Address(r13, step));
514 514 // advance r13
515 515 increment(r13, step);
516 516 dispatch_base(state, Interpreter::dispatch_table(state));
517 517 }
518 518
519 519 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
520 520 // load current bytecode
521 521 load_unsigned_byte(rbx, Address(r13, 0));
522 522 dispatch_base(state, table);
523 523 }
524 524
525 525 // remove activation
526 526 //
527 527 // Unlock the receiver if this is a synchronized method.
528 528 // Unlock any Java monitors from syncronized blocks.
529 529 // Remove the activation from the stack.
530 530 //
531 531 // If there are locked Java monitors
532 532 // If throw_monitor_exception
533 533 // throws IllegalMonitorStateException
534 534 // Else if install_monitor_exception
535 535 // installs IllegalMonitorStateException
536 536 // Else
537 537 // no error processing
538 538 void InterpreterMacroAssembler::remove_activation(
539 539 TosState state,
540 540 Register ret_addr,
541 541 bool throw_monitor_exception,
542 542 bool install_monitor_exception,
543 543 bool notify_jvmdi) {
544 544 // Note: Registers rdx xmm0 may be in use for the
545 545 // result check if synchronized method
546 546 Label unlocked, unlock, no_unlock;
547 547
548 548 // get the value of _do_not_unlock_if_synchronized into rdx
549 549 const Address do_not_unlock_if_synchronized(r15_thread,
550 550 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
551 551 movbool(rdx, do_not_unlock_if_synchronized);
552 552 movbool(do_not_unlock_if_synchronized, false); // reset the flag
553 553
554 554 // get method access flags
555 555 movptr(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
556 556 movl(rcx, Address(rbx, methodOopDesc::access_flags_offset()));
557 557 testl(rcx, JVM_ACC_SYNCHRONIZED);
558 558 jcc(Assembler::zero, unlocked);
559 559
560 560 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
561 561 // is set.
562 562 testbool(rdx);
563 563 jcc(Assembler::notZero, no_unlock);
564 564
565 565 // unlock monitor
566 566 push(state); // save result
567 567
568 568 // BasicObjectLock will be first in list, since this is a
569 569 // synchronized method. However, need to check that the object has
570 570 // not been unlocked by an explicit monitorexit bytecode.
571 571 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
572 572 wordSize - (int) sizeof(BasicObjectLock));
573 573 // We use c_rarg1 so that if we go slow path it will be the correct
574 574 // register for unlock_object to pass to VM directly
575 575 lea(c_rarg1, monitor); // address of first monitor
576 576
577 577 movptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
578 578 testptr(rax, rax);
579 579 jcc(Assembler::notZero, unlock);
580 580
581 581 pop(state);
582 582 if (throw_monitor_exception) {
583 583 // Entry already unlocked, need to throw exception
584 584 call_VM(noreg, CAST_FROM_FN_PTR(address,
585 585 InterpreterRuntime::throw_illegal_monitor_state_exception));
586 586 should_not_reach_here();
587 587 } else {
588 588 // Monitor already unlocked during a stack unroll. If requested,
589 589 // install an illegal_monitor_state_exception. Continue with
590 590 // stack unrolling.
591 591 if (install_monitor_exception) {
592 592 call_VM(noreg, CAST_FROM_FN_PTR(address,
593 593 InterpreterRuntime::new_illegal_monitor_state_exception));
594 594 }
595 595 jmp(unlocked);
596 596 }
597 597
598 598 bind(unlock);
599 599 unlock_object(c_rarg1);
600 600 pop(state);
601 601
602 602 // Check that for block-structured locking (i.e., that all locked
603 603 // objects has been unlocked)
604 604 bind(unlocked);
605 605
606 606 // rax: Might contain return value
607 607
608 608 // Check that all monitors are unlocked
609 609 {
610 610 Label loop, exception, entry, restart;
611 611 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
612 612 const Address monitor_block_top(
613 613 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
614 614 const Address monitor_block_bot(
615 615 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
616 616
617 617 bind(restart);
618 618 // We use c_rarg1 so that if we go slow path it will be the correct
619 619 // register for unlock_object to pass to VM directly
620 620 movptr(c_rarg1, monitor_block_top); // points to current entry, starting
621 621 // with top-most entry
622 622 lea(rbx, monitor_block_bot); // points to word before bottom of
623 623 // monitor block
624 624 jmp(entry);
625 625
626 626 // Entry already locked, need to throw exception
627 627 bind(exception);
628 628
629 629 if (throw_monitor_exception) {
630 630 // Throw exception
631 631 MacroAssembler::call_VM(noreg,
632 632 CAST_FROM_FN_PTR(address, InterpreterRuntime::
633 633 throw_illegal_monitor_state_exception));
634 634 should_not_reach_here();
635 635 } else {
636 636 // Stack unrolling. Unlock object and install illegal_monitor_exception.
637 637 // Unlock does not block, so don't have to worry about the frame.
638 638 // We don't have to preserve c_rarg1 since we are going to throw an exception.
639 639
640 640 push(state);
641 641 unlock_object(c_rarg1);
642 642 pop(state);
643 643
644 644 if (install_monitor_exception) {
645 645 call_VM(noreg, CAST_FROM_FN_PTR(address,
646 646 InterpreterRuntime::
647 647 new_illegal_monitor_state_exception));
648 648 }
649 649
650 650 jmp(restart);
651 651 }
652 652
653 653 bind(loop);
654 654 // check if current entry is used
655 655 cmpptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
656 656 jcc(Assembler::notEqual, exception);
657 657
658 658 addptr(c_rarg1, entry_size); // otherwise advance to next entry
659 659 bind(entry);
660 660 cmpptr(c_rarg1, rbx); // check if bottom reached
661 661 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
662 662 }
663 663
664 664 bind(no_unlock);
665 665
666 666 // jvmti support
667 667 if (notify_jvmdi) {
668 668 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
669 669 } else {
670 670 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
671 671 }
672 672
673 673 // remove activation
674 674 // get sender sp
675 675 movptr(rbx,
676 676 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
677 677 leave(); // remove frame anchor
678 678 pop(ret_addr); // get return address
679 679 mov(rsp, rbx); // set sp to sender sp
680 680 }
681 681
682 682 #endif // C_INTERP
683 683
684 684 // Lock object
685 685 //
686 686 // Args:
687 687 // c_rarg1: BasicObjectLock to be used for locking
688 688 //
689 689 // Kills:
690 690 // rax
691 691 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
692 692 // rscratch1, rscratch2 (scratch regs)
693 693 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
694 694 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
695 695
696 696 if (UseHeavyMonitors) {
697 697 call_VM(noreg,
698 698 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
699 699 lock_reg);
700 700 } else {
701 701 Label done;
702 702
703 703 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
704 704 const Register obj_reg = c_rarg3; // Will contain the oop
705 705
706 706 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
707 707 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
708 708 const int mark_offset = lock_offset +
709 709 BasicLock::displaced_header_offset_in_bytes();
710 710
711 711 Label slow_case;
712 712
713 713 // Load object pointer into obj_reg %c_rarg3
714 714 movptr(obj_reg, Address(lock_reg, obj_offset));
715 715
716 716 if (UseBiasedLocking) {
717 717 biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, done, &slow_case);
718 718 }
719 719
720 720 // Load immediate 1 into swap_reg %rax
721 721 movl(swap_reg, 1);
722 722
723 723 // Load (object->mark() | 1) into swap_reg %rax
724 724 orptr(swap_reg, Address(obj_reg, 0));
725 725
726 726 // Save (object->mark() | 1) into BasicLock's displaced header
727 727 movptr(Address(lock_reg, mark_offset), swap_reg);
728 728
729 729 assert(lock_offset == 0,
730 730 "displached header must be first word in BasicObjectLock");
731 731
732 732 if (os::is_MP()) lock();
733 733 cmpxchgptr(lock_reg, Address(obj_reg, 0));
734 734 if (PrintBiasedLockingStatistics) {
735 735 cond_inc32(Assembler::zero,
736 736 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
737 737 }
738 738 jcc(Assembler::zero, done);
739 739
740 740 // Test if the oopMark is an obvious stack pointer, i.e.,
741 741 // 1) (mark & 7) == 0, and
742 742 // 2) rsp <= mark < mark + os::pagesize()
743 743 //
744 744 // These 3 tests can be done by evaluating the following
745 745 // expression: ((mark - rsp) & (7 - os::vm_page_size())),
746 746 // assuming both stack pointer and pagesize have their
747 747 // least significant 3 bits clear.
748 748 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
749 749 subptr(swap_reg, rsp);
750 750 andptr(swap_reg, 7 - os::vm_page_size());
751 751
752 752 // Save the test result, for recursive case, the result is zero
753 753 movptr(Address(lock_reg, mark_offset), swap_reg);
754 754
755 755 if (PrintBiasedLockingStatistics) {
756 756 cond_inc32(Assembler::zero,
757 757 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
758 758 }
759 759 jcc(Assembler::zero, done);
760 760
761 761 bind(slow_case);
762 762
763 763 // Call the runtime routine for slow case
764 764 call_VM(noreg,
765 765 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
766 766 lock_reg);
767 767
768 768 bind(done);
769 769 }
770 770 }
771 771
772 772
773 773 // Unlocks an object. Used in monitorexit bytecode and
774 774 // remove_activation. Throws an IllegalMonitorException if object is
775 775 // not locked by current thread.
776 776 //
777 777 // Args:
778 778 // c_rarg1: BasicObjectLock for lock
779 779 //
780 780 // Kills:
781 781 // rax
782 782 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
783 783 // rscratch1, rscratch2 (scratch regs)
784 784 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
785 785 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
786 786
787 787 if (UseHeavyMonitors) {
788 788 call_VM(noreg,
789 789 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
790 790 lock_reg);
791 791 } else {
792 792 Label done;
793 793
794 794 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
795 795 const Register header_reg = c_rarg2; // Will contain the old oopMark
796 796 const Register obj_reg = c_rarg3; // Will contain the oop
797 797
798 798 save_bcp(); // Save in case of exception
799 799
800 800 // Convert from BasicObjectLock structure to object and BasicLock
801 801 // structure Store the BasicLock address into %rax
802 802 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
803 803
804 804 // Load oop into obj_reg(%c_rarg3)
805 805 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
806 806
807 807 // Free entry
808 808 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
809 809
810 810 if (UseBiasedLocking) {
811 811 biased_locking_exit(obj_reg, header_reg, done);
812 812 }
813 813
814 814 // Load the old header from BasicLock structure
815 815 movptr(header_reg, Address(swap_reg,
816 816 BasicLock::displaced_header_offset_in_bytes()));
817 817
818 818 // Test for recursion
819 819 testptr(header_reg, header_reg);
820 820
821 821 // zero for recursive case
822 822 jcc(Assembler::zero, done);
823 823
824 824 // Atomic swap back the old header
825 825 if (os::is_MP()) lock();
826 826 cmpxchgptr(header_reg, Address(obj_reg, 0));
827 827
828 828 // zero for recursive case
829 829 jcc(Assembler::zero, done);
830 830
831 831 // Call the runtime routine for slow case.
832 832 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
833 833 obj_reg); // restore obj
834 834 call_VM(noreg,
835 835 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
836 836 lock_reg);
837 837
838 838 bind(done);
839 839
840 840 restore_bcp();
841 841 }
842 842 }
843 843
844 844 #ifndef CC_INTERP
845 845
846 846 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
847 847 Label& zero_continue) {
848 848 assert(ProfileInterpreter, "must be profiling interpreter");
849 849 movptr(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
850 850 testptr(mdp, mdp);
851 851 jcc(Assembler::zero, zero_continue);
852 852 }
853 853
854 854
855 855 // Set the method data pointer for the current bcp.
856 856 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
857 857 assert(ProfileInterpreter, "must be profiling interpreter");
858 858 Label set_mdp;
859 859 push(rax);
860 860 push(rbx);
861 861
862 862 get_method(rbx);
863 863 // Test MDO to avoid the call if it is NULL.
864 864 movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
865 865 testptr(rax, rax);
866 866 jcc(Assembler::zero, set_mdp);
867 867 // rbx: method
868 868 // r13: bcp
869 869 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, r13);
870 870 // rax: mdi
871 871 // mdo is guaranteed to be non-zero here, we checked for it before the call.
872 872 movptr(rbx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
873 873 addptr(rbx, in_bytes(methodDataOopDesc::data_offset()));
874 874 addptr(rax, rbx);
875 875 bind(set_mdp);
876 876 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
877 877 pop(rbx);
878 878 pop(rax);
879 879 }
880 880
881 881 void InterpreterMacroAssembler::verify_method_data_pointer() {
882 882 assert(ProfileInterpreter, "must be profiling interpreter");
883 883 #ifdef ASSERT
884 884 Label verify_continue;
885 885 push(rax);
886 886 push(rbx);
887 887 push(c_rarg3);
888 888 push(c_rarg2);
889 889 test_method_data_pointer(c_rarg3, verify_continue); // If mdp is zero, continue
890 890 get_method(rbx);
891 891
892 892 // If the mdp is valid, it will point to a DataLayout header which is
893 893 // consistent with the bcp. The converse is highly probable also.
894 894 load_unsigned_short(c_rarg2,
895 895 Address(c_rarg3, in_bytes(DataLayout::bci_offset())));
896 896 addptr(c_rarg2, Address(rbx, methodOopDesc::const_offset()));
897 897 lea(c_rarg2, Address(c_rarg2, constMethodOopDesc::codes_offset()));
898 898 cmpptr(c_rarg2, r13);
899 899 jcc(Assembler::equal, verify_continue);
900 900 // rbx: method
901 901 // r13: bcp
902 902 // c_rarg3: mdp
903 903 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
904 904 rbx, r13, c_rarg3);
905 905 bind(verify_continue);
906 906 pop(c_rarg2);
907 907 pop(c_rarg3);
908 908 pop(rbx);
909 909 pop(rax);
910 910 #endif // ASSERT
911 911 }
912 912
913 913
914 914 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
915 915 int constant,
916 916 Register value) {
917 917 assert(ProfileInterpreter, "must be profiling interpreter");
918 918 Address data(mdp_in, constant);
919 919 movptr(data, value);
920 920 }
921 921
922 922
923 923 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
924 924 int constant,
925 925 bool decrement) {
926 926 // Counter address
927 927 Address data(mdp_in, constant);
928 928
929 929 increment_mdp_data_at(data, decrement);
930 930 }
931 931
932 932 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
933 933 bool decrement) {
934 934 assert(ProfileInterpreter, "must be profiling interpreter");
935 935 // %%% this does 64bit counters at best it is wasting space
936 936 // at worst it is a rare bug when counters overflow
937 937
938 938 if (decrement) {
939 939 // Decrement the register. Set condition codes.
940 940 addptr(data, (int32_t) -DataLayout::counter_increment);
941 941 // If the decrement causes the counter to overflow, stay negative
942 942 Label L;
943 943 jcc(Assembler::negative, L);
944 944 addptr(data, (int32_t) DataLayout::counter_increment);
945 945 bind(L);
946 946 } else {
947 947 assert(DataLayout::counter_increment == 1,
948 948 "flow-free idiom only works with 1");
949 949 // Increment the register. Set carry flag.
950 950 addptr(data, DataLayout::counter_increment);
951 951 // If the increment causes the counter to overflow, pull back by 1.
952 952 sbbptr(data, (int32_t)0);
953 953 }
954 954 }
955 955
956 956
957 957 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
958 958 Register reg,
959 959 int constant,
960 960 bool decrement) {
961 961 Address data(mdp_in, reg, Address::times_1, constant);
962 962
963 963 increment_mdp_data_at(data, decrement);
964 964 }
965 965
966 966 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
967 967 int flag_byte_constant) {
968 968 assert(ProfileInterpreter, "must be profiling interpreter");
969 969 int header_offset = in_bytes(DataLayout::header_offset());
970 970 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
971 971 // Set the flag
972 972 orl(Address(mdp_in, header_offset), header_bits);
973 973 }
974 974
975 975
976 976
977 977 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
978 978 int offset,
979 979 Register value,
980 980 Register test_value_out,
981 981 Label& not_equal_continue) {
982 982 assert(ProfileInterpreter, "must be profiling interpreter");
983 983 if (test_value_out == noreg) {
984 984 cmpptr(value, Address(mdp_in, offset));
985 985 } else {
986 986 // Put the test value into a register, so caller can use it:
987 987 movptr(test_value_out, Address(mdp_in, offset));
988 988 cmpptr(test_value_out, value);
989 989 }
990 990 jcc(Assembler::notEqual, not_equal_continue);
991 991 }
992 992
993 993
994 994 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
995 995 int offset_of_disp) {
996 996 assert(ProfileInterpreter, "must be profiling interpreter");
997 997 Address disp_address(mdp_in, offset_of_disp);
998 998 addptr(mdp_in, disp_address);
999 999 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1000 1000 }
1001 1001
1002 1002
1003 1003 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1004 1004 Register reg,
1005 1005 int offset_of_disp) {
1006 1006 assert(ProfileInterpreter, "must be profiling interpreter");
1007 1007 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1008 1008 addptr(mdp_in, disp_address);
1009 1009 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1010 1010 }
1011 1011
1012 1012
1013 1013 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1014 1014 int constant) {
1015 1015 assert(ProfileInterpreter, "must be profiling interpreter");
1016 1016 addptr(mdp_in, constant);
1017 1017 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1018 1018 }
1019 1019
1020 1020
1021 1021 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1022 1022 assert(ProfileInterpreter, "must be profiling interpreter");
1023 1023 push(return_bci); // save/restore across call_VM
1024 1024 call_VM(noreg,
1025 1025 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1026 1026 return_bci);
1027 1027 pop(return_bci);
1028 1028 }
1029 1029
1030 1030
1031 1031 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1032 1032 Register bumped_count) {
1033 1033 if (ProfileInterpreter) {
1034 1034 Label profile_continue;
1035 1035
1036 1036 // If no method data exists, go to profile_continue.
1037 1037 // Otherwise, assign to mdp
1038 1038 test_method_data_pointer(mdp, profile_continue);
1039 1039
1040 1040 // We are taking a branch. Increment the taken count.
1041 1041 // We inline increment_mdp_data_at to return bumped_count in a register
1042 1042 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1043 1043 Address data(mdp, in_bytes(JumpData::taken_offset()));
1044 1044 movptr(bumped_count, data);
1045 1045 assert(DataLayout::counter_increment == 1,
1046 1046 "flow-free idiom only works with 1");
1047 1047 addptr(bumped_count, DataLayout::counter_increment);
1048 1048 sbbptr(bumped_count, 0);
1049 1049 movptr(data, bumped_count); // Store back out
1050 1050
1051 1051 // The method data pointer needs to be updated to reflect the new target.
1052 1052 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1053 1053 bind(profile_continue);
1054 1054 }
1055 1055 }
1056 1056
1057 1057
1058 1058 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1059 1059 if (ProfileInterpreter) {
1060 1060 Label profile_continue;
1061 1061
1062 1062 // If no method data exists, go to profile_continue.
1063 1063 test_method_data_pointer(mdp, profile_continue);
1064 1064
1065 1065 // We are taking a branch. Increment the not taken count.
1066 1066 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1067 1067
1068 1068 // The method data pointer needs to be updated to correspond to
1069 1069 // the next bytecode
1070 1070 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1071 1071 bind(profile_continue);
1072 1072 }
1073 1073 }
1074 1074
1075 1075
1076 1076 void InterpreterMacroAssembler::profile_call(Register mdp) {
1077 1077 if (ProfileInterpreter) {
1078 1078 Label profile_continue;
1079 1079
1080 1080 // If no method data exists, go to profile_continue.
1081 1081 test_method_data_pointer(mdp, profile_continue);
1082 1082
1083 1083 // We are making a call. Increment the count.
1084 1084 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1085 1085
1086 1086 // The method data pointer needs to be updated to reflect the new target.
1087 1087 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1088 1088 bind(profile_continue);
1089 1089 }
1090 1090 }
1091 1091
1092 1092
1093 1093 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1094 1094 if (ProfileInterpreter) {
1095 1095 Label profile_continue;
1096 1096
1097 1097 // If no method data exists, go to profile_continue.
1098 1098 test_method_data_pointer(mdp, profile_continue);
1099 1099
1100 1100 // We are making a call. Increment the count.
1101 1101 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1102 1102
1103 1103 // The method data pointer needs to be updated to reflect the new target.
1104 1104 update_mdp_by_constant(mdp,
1105 1105 in_bytes(VirtualCallData::
1106 1106 virtual_call_data_size()));
1107 1107 bind(profile_continue);
1108 1108 }
1109 1109 }
1110 1110
1111 1111
1112 1112 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1113 1113 Register mdp,
1114 1114 Register reg2,
1115 1115 bool receiver_can_be_null) {
1116 1116 if (ProfileInterpreter) {
1117 1117 Label profile_continue;
1118 1118
1119 1119 // If no method data exists, go to profile_continue.
1120 1120 test_method_data_pointer(mdp, profile_continue);
1121 1121
1122 1122 Label skip_receiver_profile;
1123 1123 if (receiver_can_be_null) {
1124 1124 Label not_null;
1125 1125 testptr(receiver, receiver);
1126 1126 jccb(Assembler::notZero, not_null);
1127 1127 // We are making a call. Increment the count for null receiver.
1128 1128 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1129 1129 jmp(skip_receiver_profile);
1130 1130 bind(not_null);
1131 1131 }
1132 1132
1133 1133 // Record the receiver type.
1134 1134 record_klass_in_profile(receiver, mdp, reg2, true);
1135 1135 bind(skip_receiver_profile);
1136 1136
1137 1137 // The method data pointer needs to be updated to reflect the new target.
1138 1138 update_mdp_by_constant(mdp,
1139 1139 in_bytes(VirtualCallData::
1140 1140 virtual_call_data_size()));
1141 1141 bind(profile_continue);
1142 1142 }
1143 1143 }
1144 1144
1145 1145 // This routine creates a state machine for updating the multi-row
1146 1146 // type profile at a virtual call site (or other type-sensitive bytecode).
1147 1147 // The machine visits each row (of receiver/count) until the receiver type
1148 1148 // is found, or until it runs out of rows. At the same time, it remembers
1149 1149 // the location of the first empty row. (An empty row records null for its
1150 1150 // receiver, and can be allocated for a newly-observed receiver type.)
1151 1151 // Because there are two degrees of freedom in the state, a simple linear
1152 1152 // search will not work; it must be a decision tree. Hence this helper
1153 1153 // function is recursive, to generate the required tree structured code.
1154 1154 // It's the interpreter, so we are trading off code space for speed.
1155 1155 // See below for example code.
1156 1156 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1157 1157 Register receiver, Register mdp,
1158 1158 Register reg2, int start_row,
1159 1159 Label& done, bool is_virtual_call) {
1160 1160 if (TypeProfileWidth == 0) {
1161 1161 if (is_virtual_call) {
1162 1162 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1163 1163 }
1164 1164 return;
1165 1165 }
1166 1166
1167 1167 int last_row = VirtualCallData::row_limit() - 1;
1168 1168 assert(start_row <= last_row, "must be work left to do");
1169 1169 // Test this row for both the receiver and for null.
1170 1170 // Take any of three different outcomes:
1171 1171 // 1. found receiver => increment count and goto done
1172 1172 // 2. found null => keep looking for case 1, maybe allocate this cell
1173 1173 // 3. found something else => keep looking for cases 1 and 2
1174 1174 // Case 3 is handled by a recursive call.
1175 1175 for (int row = start_row; row <= last_row; row++) {
1176 1176 Label next_test;
1177 1177 bool test_for_null_also = (row == start_row);
1178 1178
1179 1179 // See if the receiver is receiver[n].
1180 1180 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1181 1181 test_mdp_data_at(mdp, recvr_offset, receiver,
1182 1182 (test_for_null_also ? reg2 : noreg),
1183 1183 next_test);
1184 1184 // (Reg2 now contains the receiver from the CallData.)
1185 1185
1186 1186 // The receiver is receiver[n]. Increment count[n].
1187 1187 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1188 1188 increment_mdp_data_at(mdp, count_offset);
1189 1189 jmp(done);
1190 1190 bind(next_test);
1191 1191
1192 1192 if (test_for_null_also) {
1193 1193 Label found_null;
1194 1194 // Failed the equality check on receiver[n]... Test for null.
1195 1195 testptr(reg2, reg2);
1196 1196 if (start_row == last_row) {
1197 1197 // The only thing left to do is handle the null case.
1198 1198 if (is_virtual_call) {
1199 1199 jccb(Assembler::zero, found_null);
1200 1200 // Receiver did not match any saved receiver and there is no empty row for it.
1201 1201 // Increment total counter to indicate polymorphic case.
1202 1202 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1203 1203 jmp(done);
1204 1204 bind(found_null);
1205 1205 } else {
1206 1206 jcc(Assembler::notZero, done);
1207 1207 }
1208 1208 break;
1209 1209 }
1210 1210 // Since null is rare, make it be the branch-taken case.
1211 1211 jcc(Assembler::zero, found_null);
1212 1212
1213 1213 // Put all the "Case 3" tests here.
1214 1214 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
1215 1215
1216 1216 // Found a null. Keep searching for a matching receiver,
1217 1217 // but remember that this is an empty (unused) slot.
1218 1218 bind(found_null);
1219 1219 }
1220 1220 }
1221 1221
1222 1222 // In the fall-through case, we found no matching receiver, but we
1223 1223 // observed the receiver[start_row] is NULL.
1224 1224
1225 1225 // Fill in the receiver field and increment the count.
1226 1226 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1227 1227 set_mdp_data_at(mdp, recvr_offset, receiver);
1228 1228 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1229 1229 movl(reg2, DataLayout::counter_increment);
1230 1230 set_mdp_data_at(mdp, count_offset, reg2);
1231 1231 if (start_row > 0) {
1232 1232 jmp(done);
1233 1233 }
1234 1234 }
1235 1235
1236 1236 // Example state machine code for three profile rows:
1237 1237 // // main copy of decision tree, rooted at row[1]
1238 1238 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1239 1239 // if (row[0].rec != NULL) {
1240 1240 // // inner copy of decision tree, rooted at row[1]
1241 1241 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1242 1242 // if (row[1].rec != NULL) {
1243 1243 // // degenerate decision tree, rooted at row[2]
1244 1244 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1245 1245 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1246 1246 // row[2].init(rec); goto done;
1247 1247 // } else {
1248 1248 // // remember row[1] is empty
1249 1249 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1250 1250 // row[1].init(rec); goto done;
1251 1251 // }
1252 1252 // } else {
1253 1253 // // remember row[0] is empty
1254 1254 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1255 1255 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1256 1256 // row[0].init(rec); goto done;
1257 1257 // }
1258 1258 // done:
1259 1259
1260 1260 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1261 1261 Register mdp, Register reg2,
1262 1262 bool is_virtual_call) {
1263 1263 assert(ProfileInterpreter, "must be profiling");
1264 1264 Label done;
1265 1265
1266 1266 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1267 1267
1268 1268 bind (done);
1269 1269 }
1270 1270
1271 1271 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1272 1272 Register mdp) {
1273 1273 if (ProfileInterpreter) {
1274 1274 Label profile_continue;
1275 1275 uint row;
1276 1276
1277 1277 // If no method data exists, go to profile_continue.
1278 1278 test_method_data_pointer(mdp, profile_continue);
1279 1279
1280 1280 // Update the total ret count.
1281 1281 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1282 1282
1283 1283 for (row = 0; row < RetData::row_limit(); row++) {
1284 1284 Label next_test;
1285 1285
1286 1286 // See if return_bci is equal to bci[n]:
1287 1287 test_mdp_data_at(mdp,
1288 1288 in_bytes(RetData::bci_offset(row)),
1289 1289 return_bci, noreg,
1290 1290 next_test);
1291 1291
1292 1292 // return_bci is equal to bci[n]. Increment the count.
1293 1293 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1294 1294
1295 1295 // The method data pointer needs to be updated to reflect the new target.
1296 1296 update_mdp_by_offset(mdp,
1297 1297 in_bytes(RetData::bci_displacement_offset(row)));
1298 1298 jmp(profile_continue);
1299 1299 bind(next_test);
1300 1300 }
1301 1301
1302 1302 update_mdp_for_ret(return_bci);
1303 1303
1304 1304 bind(profile_continue);
1305 1305 }
1306 1306 }
1307 1307
1308 1308
1309 1309 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1310 1310 if (ProfileInterpreter) {
1311 1311 Label profile_continue;
1312 1312
1313 1313 // If no method data exists, go to profile_continue.
1314 1314 test_method_data_pointer(mdp, profile_continue);
1315 1315
1316 1316 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1317 1317
1318 1318 // The method data pointer needs to be updated.
1319 1319 int mdp_delta = in_bytes(BitData::bit_data_size());
1320 1320 if (TypeProfileCasts) {
1321 1321 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1322 1322 }
1323 1323 update_mdp_by_constant(mdp, mdp_delta);
1324 1324
1325 1325 bind(profile_continue);
1326 1326 }
1327 1327 }
1328 1328
1329 1329
1330 1330 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1331 1331 if (ProfileInterpreter && TypeProfileCasts) {
1332 1332 Label profile_continue;
1333 1333
1334 1334 // If no method data exists, go to profile_continue.
1335 1335 test_method_data_pointer(mdp, profile_continue);
1336 1336
1337 1337 int count_offset = in_bytes(CounterData::count_offset());
1338 1338 // Back up the address, since we have already bumped the mdp.
1339 1339 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1340 1340
1341 1341 // *Decrement* the counter. We expect to see zero or small negatives.
1342 1342 increment_mdp_data_at(mdp, count_offset, true);
1343 1343
1344 1344 bind (profile_continue);
1345 1345 }
1346 1346 }
1347 1347
1348 1348
1349 1349 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1350 1350 if (ProfileInterpreter) {
1351 1351 Label profile_continue;
1352 1352
1353 1353 // If no method data exists, go to profile_continue.
1354 1354 test_method_data_pointer(mdp, profile_continue);
1355 1355
1356 1356 // The method data pointer needs to be updated.
1357 1357 int mdp_delta = in_bytes(BitData::bit_data_size());
1358 1358 if (TypeProfileCasts) {
1359 1359 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1360 1360
1361 1361 // Record the object type.
1362 1362 record_klass_in_profile(klass, mdp, reg2, false);
1363 1363 }
1364 1364 update_mdp_by_constant(mdp, mdp_delta);
1365 1365
1366 1366 bind(profile_continue);
1367 1367 }
1368 1368 }
1369 1369
1370 1370
1371 1371 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1372 1372 if (ProfileInterpreter) {
1373 1373 Label profile_continue;
1374 1374
1375 1375 // If no method data exists, go to profile_continue.
1376 1376 test_method_data_pointer(mdp, profile_continue);
1377 1377
1378 1378 // Update the default case count
1379 1379 increment_mdp_data_at(mdp,
1380 1380 in_bytes(MultiBranchData::default_count_offset()));
1381 1381
1382 1382 // The method data pointer needs to be updated.
1383 1383 update_mdp_by_offset(mdp,
1384 1384 in_bytes(MultiBranchData::
1385 1385 default_displacement_offset()));
1386 1386
1387 1387 bind(profile_continue);
1388 1388 }
1389 1389 }
1390 1390
1391 1391
1392 1392 void InterpreterMacroAssembler::profile_switch_case(Register index,
1393 1393 Register mdp,
1394 1394 Register reg2) {
1395 1395 if (ProfileInterpreter) {
1396 1396 Label profile_continue;
1397 1397
1398 1398 // If no method data exists, go to profile_continue.
1399 1399 test_method_data_pointer(mdp, profile_continue);
1400 1400
1401 1401 // Build the base (index * per_case_size_in_bytes()) +
1402 1402 // case_array_offset_in_bytes()
1403 1403 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1404 1404 imulptr(index, reg2); // XXX l ?
1405 1405 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1406 1406
1407 1407 // Update the case count
1408 1408 increment_mdp_data_at(mdp,
1409 1409 index,
1410 1410 in_bytes(MultiBranchData::relative_count_offset()));
1411 1411
1412 1412 // The method data pointer needs to be updated.
1413 1413 update_mdp_by_offset(mdp,
1414 1414 index,
1415 1415 in_bytes(MultiBranchData::
1416 1416 relative_displacement_offset()));
1417 1417
1418 1418 bind(profile_continue);
1419 1419 }
1420 1420 }
1421 1421
1422 1422
1423 1423
1424 1424 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1425 1425 if (state == atos) {
1426 1426 MacroAssembler::verify_oop(reg);
1427 1427 }
1428 1428 }
1429 1429
1430 1430 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1431 1431 }
1432 1432 #endif // !CC_INTERP
1433 1433
1434 1434
1435 1435 void InterpreterMacroAssembler::notify_method_entry() {
1436 1436 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1437 1437 // track stack depth. If it is possible to enter interp_only_mode we add
1438 1438 // the code to check if the event should be sent.
1439 1439 if (JvmtiExport::can_post_interpreter_events()) {
1440 1440 Label L;
1441 1441 movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1442 1442 testl(rdx, rdx);
1443 1443 jcc(Assembler::zero, L);
1444 1444 call_VM(noreg, CAST_FROM_FN_PTR(address,
1445 1445 InterpreterRuntime::post_method_entry));
1446 1446 bind(L);
1447 1447 }
1448 1448
1449 1449 {
1450 1450 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1451 1451 get_method(c_rarg1);
1452 1452 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1453 1453 r15_thread, c_rarg1);
1454 1454 }
1455 1455
1456 1456 // RedefineClasses() tracing support for obsolete method entry
1457 1457 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
1458 1458 get_method(c_rarg1);
1459 1459 call_VM_leaf(
1460 1460 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1461 1461 r15_thread, c_rarg1);
1462 1462 }
1463 1463 }
1464 1464
1465 1465
1466 1466 void InterpreterMacroAssembler::notify_method_exit(
1467 1467 TosState state, NotifyMethodExitMode mode) {
1468 1468 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1469 1469 // track stack depth. If it is possible to enter interp_only_mode we add
1470 1470 // the code to check if the event should be sent.
1471 1471 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1472 1472 Label L;
1473 1473 // Note: frame::interpreter_frame_result has a dependency on how the
1474 1474 // method result is saved across the call to post_method_exit. If this
1475 1475 // is changed then the interpreter_frame_result implementation will
1476 1476 // need to be updated too.
1477 1477
1478 1478 // For c++ interpreter the result is always stored at a known location in the frame
1479 1479 // template interpreter will leave it on the top of the stack.
1480 1480 NOT_CC_INTERP(push(state);)
1481 1481 movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1482 1482 testl(rdx, rdx);
1483 1483 jcc(Assembler::zero, L);
1484 1484 call_VM(noreg,
1485 1485 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1486 1486 bind(L);
1487 1487 NOT_CC_INTERP(pop(state));
1488 1488 }
1489 1489
1490 1490 {
1491 1491 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1492 1492 NOT_CC_INTERP(push(state));
1493 1493 get_method(c_rarg1);
1494 1494 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1495 1495 r15_thread, c_rarg1);
1496 1496 NOT_CC_INTERP(pop(state));
1497 1497 }
1498 1498 }
1499 1499
1500 1500 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1501 1501 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1502 1502 int increment, int mask,
1503 1503 Register scratch, bool preloaded,
1504 1504 Condition cond, Label* where) {
1505 1505 if (!preloaded) {
1506 1506 movl(scratch, counter_addr);
1507 1507 }
1508 1508 incrementl(scratch, increment);
1509 1509 movl(counter_addr, scratch);
1510 1510 andl(scratch, mask);
1511 1511 jcc(cond, *where);
1512 1512 }
|
↓ open down ↓ |
1278 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX