Print this page
rev 2237 : [mq]: initial-intrinsification-changes
rev 2238 : [mq]: code-review-comments-vladimir
Split |
Close |
Expand all |
Collapse all |
--- old/src/cpu/x86/vm/templateInterpreter_x86_64.cpp
+++ new/src/cpu/x86/vm/templateInterpreter_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 "asm/assembler.hpp"
27 27 #include "interpreter/bytecodeHistogram.hpp"
28 28 #include "interpreter/interpreter.hpp"
29 29 #include "interpreter/interpreterGenerator.hpp"
30 30 #include "interpreter/interpreterRuntime.hpp"
31 31 #include "interpreter/templateTable.hpp"
32 32 #include "oops/arrayOop.hpp"
33 33 #include "oops/methodDataOop.hpp"
34 34 #include "oops/methodOop.hpp"
35 35 #include "oops/oop.inline.hpp"
36 36 #include "prims/jvmtiExport.hpp"
37 37 #include "prims/jvmtiThreadState.hpp"
38 38 #include "runtime/arguments.hpp"
39 39 #include "runtime/deoptimization.hpp"
40 40 #include "runtime/frame.inline.hpp"
41 41 #include "runtime/sharedRuntime.hpp"
42 42 #include "runtime/stubRoutines.hpp"
43 43 #include "runtime/synchronizer.hpp"
44 44 #include "runtime/timer.hpp"
45 45 #include "runtime/vframeArray.hpp"
46 46 #include "utilities/debug.hpp"
47 47
48 48 #define __ _masm->
49 49
50 50 #ifndef CC_INTERP
51 51
52 52 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
53 53 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
54 54 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
55 55
56 56 //-----------------------------------------------------------------------------
57 57
58 58 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
59 59 address entry = __ pc();
60 60
61 61 #ifdef ASSERT
62 62 {
63 63 Label L;
64 64 __ lea(rax, Address(rbp,
65 65 frame::interpreter_frame_monitor_block_top_offset *
66 66 wordSize));
67 67 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
68 68 // grows negative)
69 69 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
70 70 __ stop ("interpreter frame not set up");
71 71 __ bind(L);
72 72 }
73 73 #endif // ASSERT
74 74 // Restore bcp under the assumption that the current frame is still
75 75 // interpreted
76 76 __ restore_bcp();
77 77
78 78 // expression stack must be empty before entering the VM if an
79 79 // exception happened
80 80 __ empty_expression_stack();
81 81 // throw exception
82 82 __ call_VM(noreg,
83 83 CAST_FROM_FN_PTR(address,
84 84 InterpreterRuntime::throw_StackOverflowError));
85 85 return entry;
86 86 }
87 87
88 88 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
89 89 const char* name) {
90 90 address entry = __ pc();
91 91 // expression stack must be empty before entering the VM if an
92 92 // exception happened
93 93 __ empty_expression_stack();
94 94 // setup parameters
95 95 // ??? convention: expect aberrant index in register ebx
96 96 __ lea(c_rarg1, ExternalAddress((address)name));
97 97 __ call_VM(noreg,
98 98 CAST_FROM_FN_PTR(address,
99 99 InterpreterRuntime::
100 100 throw_ArrayIndexOutOfBoundsException),
101 101 c_rarg1, rbx);
102 102 return entry;
103 103 }
104 104
105 105 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
106 106 address entry = __ pc();
107 107
108 108 // object is at TOS
109 109 __ pop(c_rarg1);
110 110
111 111 // expression stack must be empty before entering the VM if an
112 112 // exception happened
113 113 __ empty_expression_stack();
114 114
115 115 __ call_VM(noreg,
116 116 CAST_FROM_FN_PTR(address,
117 117 InterpreterRuntime::
118 118 throw_ClassCastException),
119 119 c_rarg1);
120 120 return entry;
121 121 }
122 122
123 123 // Arguments are: required type at TOS+8, failing object (or NULL) at TOS+4.
124 124 address TemplateInterpreterGenerator::generate_WrongMethodType_handler() {
125 125 address entry = __ pc();
126 126
127 127 __ pop(c_rarg2); // failing object is at TOS
128 128 __ pop(c_rarg1); // required type is at TOS+8
129 129
130 130 __ verify_oop(c_rarg1);
131 131 __ verify_oop(c_rarg2);
132 132
133 133 // Various method handle types use interpreter registers as temps.
134 134 __ restore_bcp();
135 135 __ restore_locals();
136 136
137 137 // Expression stack must be empty before entering the VM for an exception.
138 138 __ empty_expression_stack();
139 139
140 140 __ call_VM(noreg,
141 141 CAST_FROM_FN_PTR(address,
142 142 InterpreterRuntime::throw_WrongMethodTypeException),
143 143 // pass required type, failing object (or NULL)
144 144 c_rarg1, c_rarg2);
145 145 return entry;
146 146 }
147 147
148 148 address TemplateInterpreterGenerator::generate_exception_handler_common(
149 149 const char* name, const char* message, bool pass_oop) {
150 150 assert(!pass_oop || message == NULL, "either oop or message but not both");
151 151 address entry = __ pc();
152 152 if (pass_oop) {
153 153 // object is at TOS
154 154 __ pop(c_rarg2);
155 155 }
156 156 // expression stack must be empty before entering the VM if an
157 157 // exception happened
158 158 __ empty_expression_stack();
159 159 // setup parameters
160 160 __ lea(c_rarg1, ExternalAddress((address)name));
161 161 if (pass_oop) {
162 162 __ call_VM(rax, CAST_FROM_FN_PTR(address,
163 163 InterpreterRuntime::
164 164 create_klass_exception),
165 165 c_rarg1, c_rarg2);
166 166 } else {
167 167 // kind of lame ExternalAddress can't take NULL because
168 168 // external_word_Relocation will assert.
169 169 if (message != NULL) {
170 170 __ lea(c_rarg2, ExternalAddress((address)message));
171 171 } else {
172 172 __ movptr(c_rarg2, NULL_WORD);
173 173 }
174 174 __ call_VM(rax,
175 175 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
176 176 c_rarg1, c_rarg2);
177 177 }
178 178 // throw exception
179 179 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
180 180 return entry;
181 181 }
182 182
183 183
184 184 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
185 185 address entry = __ pc();
186 186 // NULL last_sp until next java call
187 187 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
188 188 __ dispatch_next(state);
189 189 return entry;
190 190 }
191 191
192 192
193 193 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
194 194 address entry = __ pc();
195 195
196 196 // Restore stack bottom in case i2c adjusted stack
197 197 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
198 198 // and NULL it as marker that esp is now tos until next java call
199 199 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
200 200
201 201 __ restore_bcp();
202 202 __ restore_locals();
203 203
204 204 Label L_got_cache, L_giant_index;
205 205 if (EnableInvokeDynamic) {
206 206 __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic);
207 207 __ jcc(Assembler::equal, L_giant_index);
208 208 }
209 209 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
210 210 __ bind(L_got_cache);
211 211 __ movl(rbx, Address(rbx, rcx,
212 212 Address::times_ptr,
213 213 in_bytes(constantPoolCacheOopDesc::base_offset()) +
214 214 3 * wordSize));
215 215 __ andl(rbx, 0xFF);
216 216 __ lea(rsp, Address(rsp, rbx, Address::times_8));
217 217 __ dispatch_next(state, step);
218 218
219 219 // out of the main line of code...
220 220 if (EnableInvokeDynamic) {
221 221 __ bind(L_giant_index);
222 222 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
223 223 __ jmp(L_got_cache);
224 224 }
225 225
226 226 return entry;
227 227 }
228 228
229 229
230 230 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
231 231 int step) {
232 232 address entry = __ pc();
233 233 // NULL last_sp until next java call
234 234 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
235 235 __ restore_bcp();
236 236 __ restore_locals();
237 237 // handle exceptions
238 238 {
239 239 Label L;
240 240 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
241 241 __ jcc(Assembler::zero, L);
242 242 __ call_VM(noreg,
243 243 CAST_FROM_FN_PTR(address,
244 244 InterpreterRuntime::throw_pending_exception));
245 245 __ should_not_reach_here();
246 246 __ bind(L);
247 247 }
248 248 __ dispatch_next(state, step);
249 249 return entry;
250 250 }
251 251
252 252 int AbstractInterpreter::BasicType_as_index(BasicType type) {
253 253 int i = 0;
254 254 switch (type) {
255 255 case T_BOOLEAN: i = 0; break;
256 256 case T_CHAR : i = 1; break;
257 257 case T_BYTE : i = 2; break;
258 258 case T_SHORT : i = 3; break;
259 259 case T_INT : i = 4; break;
260 260 case T_LONG : i = 5; break;
261 261 case T_VOID : i = 6; break;
262 262 case T_FLOAT : i = 7; break;
263 263 case T_DOUBLE : i = 8; break;
264 264 case T_OBJECT : i = 9; break;
265 265 case T_ARRAY : i = 9; break;
266 266 default : ShouldNotReachHere();
267 267 }
268 268 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
269 269 "index out of bounds");
270 270 return i;
271 271 }
272 272
273 273
274 274 address TemplateInterpreterGenerator::generate_result_handler_for(
275 275 BasicType type) {
276 276 address entry = __ pc();
277 277 switch (type) {
278 278 case T_BOOLEAN: __ c2bool(rax); break;
279 279 case T_CHAR : __ movzwl(rax, rax); break;
280 280 case T_BYTE : __ sign_extend_byte(rax); break;
281 281 case T_SHORT : __ sign_extend_short(rax); break;
282 282 case T_INT : /* nothing to do */ break;
283 283 case T_LONG : /* nothing to do */ break;
284 284 case T_VOID : /* nothing to do */ break;
285 285 case T_FLOAT : /* nothing to do */ break;
286 286 case T_DOUBLE : /* nothing to do */ break;
287 287 case T_OBJECT :
288 288 // retrieve result from frame
289 289 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
290 290 // and verify it
291 291 __ verify_oop(rax);
292 292 break;
293 293 default : ShouldNotReachHere();
294 294 }
295 295 __ ret(0); // return from result handler
296 296 return entry;
297 297 }
298 298
299 299 address TemplateInterpreterGenerator::generate_safept_entry_for(
300 300 TosState state,
301 301 address runtime_entry) {
302 302 address entry = __ pc();
303 303 __ push(state);
304 304 __ call_VM(noreg, runtime_entry);
305 305 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
306 306 return entry;
307 307 }
308 308
309 309
310 310
311 311 // Helpers for commoning out cases in the various type of method entries.
312 312 //
313 313
314 314
315 315 // increment invocation count & check for overflow
316 316 //
317 317 // Note: checking for negative value instead of overflow
318 318 // so we have a 'sticky' overflow test
319 319 //
320 320 // rbx: method
321 321 // ecx: invocation counter
322 322 //
323 323 void InterpreterGenerator::generate_counter_incr(
324 324 Label* overflow,
325 325 Label* profile_method,
326 326 Label* profile_method_continue) {
327 327 const Address invocation_counter(rbx, in_bytes(methodOopDesc::invocation_counter_offset()) +
328 328 in_bytes(InvocationCounter::counter_offset()));
329 329 // Note: In tiered we increment either counters in methodOop or in MDO depending if we're profiling or not.
330 330 if (TieredCompilation) {
331 331 int increment = InvocationCounter::count_increment;
332 332 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
333 333 Label no_mdo, done;
334 334 if (ProfileInterpreter) {
335 335 // Are we profiling?
336 336 __ movptr(rax, Address(rbx, methodOopDesc::method_data_offset()));
337 337 __ testptr(rax, rax);
338 338 __ jccb(Assembler::zero, no_mdo);
339 339 // Increment counter in the MDO
340 340 const Address mdo_invocation_counter(rax, in_bytes(methodDataOopDesc::invocation_counter_offset()) +
341 341 in_bytes(InvocationCounter::counter_offset()));
342 342 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
343 343 __ jmpb(done);
344 344 }
345 345 __ bind(no_mdo);
346 346 // Increment counter in methodOop (we don't need to load it, it's in ecx).
347 347 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
348 348 __ bind(done);
349 349 } else {
350 350 const Address backedge_counter(rbx,
351 351 methodOopDesc::backedge_counter_offset() +
352 352 InvocationCounter::counter_offset());
353 353
354 354 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
355 355 __ incrementl(Address(rbx,
356 356 methodOopDesc::interpreter_invocation_counter_offset()));
357 357 }
358 358 // Update standard invocation counters
359 359 __ movl(rax, backedge_counter); // load backedge counter
360 360
361 361 __ incrementl(rcx, InvocationCounter::count_increment);
362 362 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
363 363
364 364 __ movl(invocation_counter, rcx); // save invocation count
365 365 __ addl(rcx, rax); // add both counters
366 366
367 367 // profile_method is non-null only for interpreted method so
368 368 // profile_method != NULL == !native_call
369 369
370 370 if (ProfileInterpreter && profile_method != NULL) {
371 371 // Test to see if we should create a method data oop
372 372 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
373 373 __ jcc(Assembler::less, *profile_method_continue);
374 374
375 375 // if no method data exists, go to profile_method
376 376 __ test_method_data_pointer(rax, *profile_method);
377 377 }
378 378
379 379 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
380 380 __ jcc(Assembler::aboveEqual, *overflow);
381 381 }
382 382 }
383 383
384 384 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
385 385
386 386 // Asm interpreter on entry
387 387 // r14 - locals
388 388 // r13 - bcp
389 389 // rbx - method
390 390 // edx - cpool --- DOES NOT APPEAR TO BE TRUE
391 391 // rbp - interpreter frame
392 392
393 393 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
394 394 // Everything as it was on entry
395 395 // rdx is not restored. Doesn't appear to really be set.
396 396
397 397 const Address size_of_parameters(rbx,
398 398 methodOopDesc::size_of_parameters_offset());
399 399
400 400 // InterpreterRuntime::frequency_counter_overflow takes two
401 401 // arguments, the first (thread) is passed by call_VM, the second
402 402 // indicates if the counter overflow occurs at a backwards branch
403 403 // (NULL bcp). We pass zero for it. The call returns the address
404 404 // of the verified entry point for the method or NULL if the
405 405 // compilation did not complete (either went background or bailed
406 406 // out).
407 407 __ movl(c_rarg1, 0);
408 408 __ call_VM(noreg,
409 409 CAST_FROM_FN_PTR(address,
410 410 InterpreterRuntime::frequency_counter_overflow),
411 411 c_rarg1);
412 412
413 413 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
414 414 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
415 415 // and jump to the interpreted entry.
416 416 __ jmp(*do_continue, relocInfo::none);
417 417 }
418 418
419 419 // See if we've got enough room on the stack for locals plus overhead.
420 420 // The expression stack grows down incrementally, so the normal guard
421 421 // page mechanism will work for that.
422 422 //
423 423 // NOTE: Since the additional locals are also always pushed (wasn't
424 424 // obvious in generate_method_entry) so the guard should work for them
425 425 // too.
426 426 //
427 427 // Args:
428 428 // rdx: number of additional locals this frame needs (what we must check)
429 429 // rbx: methodOop
430 430 //
431 431 // Kills:
432 432 // rax
433 433 void InterpreterGenerator::generate_stack_overflow_check(void) {
434 434
435 435 // monitor entry size: see picture of stack set
436 436 // (generate_method_entry) and frame_amd64.hpp
437 437 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
438 438
439 439 // total overhead size: entry_size + (saved rbp through expr stack
440 440 // bottom). be sure to change this if you add/subtract anything
441 441 // to/from the overhead area
442 442 const int overhead_size =
443 443 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
444 444
445 445 const int page_size = os::vm_page_size();
446 446
447 447 Label after_frame_check;
448 448
449 449 // see if the frame is greater than one page in size. If so,
450 450 // then we need to verify there is enough stack space remaining
451 451 // for the additional locals.
452 452 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
453 453 __ jcc(Assembler::belowEqual, after_frame_check);
454 454
455 455 // compute rsp as if this were going to be the last frame on
456 456 // the stack before the red zone
457 457
458 458 const Address stack_base(r15_thread, Thread::stack_base_offset());
459 459 const Address stack_size(r15_thread, Thread::stack_size_offset());
460 460
461 461 // locals + overhead, in bytes
462 462 __ mov(rax, rdx);
463 463 __ shlptr(rax, Interpreter::logStackElementSize); // 2 slots per parameter.
464 464 __ addptr(rax, overhead_size);
465 465
466 466 #ifdef ASSERT
467 467 Label stack_base_okay, stack_size_okay;
468 468 // verify that thread stack base is non-zero
469 469 __ cmpptr(stack_base, (int32_t)NULL_WORD);
470 470 __ jcc(Assembler::notEqual, stack_base_okay);
471 471 __ stop("stack base is zero");
472 472 __ bind(stack_base_okay);
473 473 // verify that thread stack size is non-zero
474 474 __ cmpptr(stack_size, 0);
475 475 __ jcc(Assembler::notEqual, stack_size_okay);
476 476 __ stop("stack size is zero");
477 477 __ bind(stack_size_okay);
478 478 #endif
479 479
480 480 // Add stack base to locals and subtract stack size
481 481 __ addptr(rax, stack_base);
482 482 __ subptr(rax, stack_size);
483 483
484 484 // Use the maximum number of pages we might bang.
485 485 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
486 486 (StackRedPages+StackYellowPages);
487 487
488 488 // add in the red and yellow zone sizes
489 489 __ addptr(rax, max_pages * page_size);
490 490
491 491 // check against the current stack bottom
492 492 __ cmpptr(rsp, rax);
493 493 __ jcc(Assembler::above, after_frame_check);
494 494
495 495 __ pop(rax); // get return address
496 496 __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
497 497
498 498 // all done with frame size check
499 499 __ bind(after_frame_check);
500 500 }
501 501
502 502 // Allocate monitor and lock method (asm interpreter)
503 503 //
504 504 // Args:
505 505 // rbx: methodOop
506 506 // r14: locals
507 507 //
508 508 // Kills:
509 509 // rax
510 510 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
511 511 // rscratch1, rscratch2 (scratch regs)
512 512 void InterpreterGenerator::lock_method(void) {
513 513 // synchronize method
514 514 const Address access_flags(rbx, methodOopDesc::access_flags_offset());
515 515 const Address monitor_block_top(
516 516 rbp,
517 517 frame::interpreter_frame_monitor_block_top_offset * wordSize);
518 518 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
519 519
520 520 #ifdef ASSERT
521 521 {
522 522 Label L;
523 523 __ movl(rax, access_flags);
524 524 __ testl(rax, JVM_ACC_SYNCHRONIZED);
525 525 __ jcc(Assembler::notZero, L);
526 526 __ stop("method doesn't need synchronization");
527 527 __ bind(L);
528 528 }
529 529 #endif // ASSERT
530 530
531 531 // get synchronization object
532 532 {
533 533 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() +
534 534 Klass::java_mirror_offset_in_bytes();
535 535 Label done;
536 536 __ movl(rax, access_flags);
537 537 __ testl(rax, JVM_ACC_STATIC);
538 538 // get receiver (assume this is frequent case)
539 539 __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
540 540 __ jcc(Assembler::zero, done);
541 541 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
542 542 __ movptr(rax, Address(rax,
543 543 constantPoolOopDesc::pool_holder_offset_in_bytes()));
544 544 __ movptr(rax, Address(rax, mirror_offset));
545 545
546 546 #ifdef ASSERT
547 547 {
548 548 Label L;
549 549 __ testptr(rax, rax);
550 550 __ jcc(Assembler::notZero, L);
551 551 __ stop("synchronization object is NULL");
552 552 __ bind(L);
553 553 }
554 554 #endif // ASSERT
555 555
556 556 __ bind(done);
557 557 }
558 558
559 559 // add space for monitor & lock
560 560 __ subptr(rsp, entry_size); // add space for a monitor entry
561 561 __ movptr(monitor_block_top, rsp); // set new monitor block top
562 562 // store object
563 563 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
564 564 __ movptr(c_rarg1, rsp); // object address
565 565 __ lock_object(c_rarg1);
566 566 }
567 567
568 568 // Generate a fixed interpreter frame. This is identical setup for
569 569 // interpreted methods and for native methods hence the shared code.
570 570 //
571 571 // Args:
572 572 // rax: return address
573 573 // rbx: methodOop
574 574 // r14: pointer to locals
575 575 // r13: sender sp
576 576 // rdx: cp cache
577 577 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
578 578 // initialize fixed part of activation frame
579 579 __ push(rax); // save return address
580 580 __ enter(); // save old & set new rbp
581 581 __ push(r13); // set sender sp
582 582 __ push((int)NULL_WORD); // leave last_sp as null
583 583 __ movptr(r13, Address(rbx, methodOopDesc::const_offset())); // get constMethodOop
584 584 __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase
585 585 __ push(rbx); // save methodOop
586 586 if (ProfileInterpreter) {
587 587 Label method_data_continue;
588 588 __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
589 589 __ testptr(rdx, rdx);
590 590 __ jcc(Assembler::zero, method_data_continue);
591 591 __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
592 592 __ bind(method_data_continue);
593 593 __ push(rdx); // set the mdp (method data pointer)
594 594 } else {
595 595 __ push(0);
596 596 }
597 597
598 598 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
599 599 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
600 600 __ push(rdx); // set constant pool cache
601 601 __ push(r14); // set locals pointer
602 602 if (native_call) {
603 603 __ push(0); // no bcp
604 604 } else {
605 605 __ push(r13); // set bcp
606 606 }
607 607 __ push(0); // reserve word for pointer to expression stack bottom
608 608 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
609 609 }
610 610
611 611 // End of helpers
612 612
613 613 // Various method entries
614 614 //------------------------------------------------------------------------------------------------------------------------
615 615 //
616 616 //
617 617
618 618 // Call an accessor method (assuming it is resolved, otherwise drop
619 619 // into vanilla (slow path) entry
620 620 address InterpreterGenerator::generate_accessor_entry(void) {
621 621 // rbx: methodOop
622 622
623 623 // r13: senderSP must preserver for slow path, set SP to it on fast path
624 624
625 625 address entry_point = __ pc();
626 626 Label xreturn_path;
627 627
628 628 // do fastpath for resolved accessor methods
629 629 if (UseFastAccessorMethods) {
630 630 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites
631 631 // thereof; parameter size = 1
632 632 // Note: We can only use this code if the getfield has been resolved
633 633 // and if we don't have a null-pointer exception => check for
634 634 // these conditions first and use slow path if necessary.
635 635 Label slow_path;
636 636 // If we need a safepoint check, generate full interpreter entry.
637 637 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
638 638 SafepointSynchronize::_not_synchronized);
639 639
640 640 __ jcc(Assembler::notEqual, slow_path);
641 641 // rbx: method
642 642 __ movptr(rax, Address(rsp, wordSize));
643 643
644 644 // check if local 0 != NULL and read field
645 645 __ testptr(rax, rax);
646 646 __ jcc(Assembler::zero, slow_path);
647 647
648 648 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
649 649 // read first instruction word and extract bytecode @ 1 and index @ 2
650 650 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
651 651 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
652 652 // Shift codes right to get the index on the right.
653 653 // The bytecode fetched looks like <index><0xb4><0x2a>
654 654 __ shrl(rdx, 2 * BitsPerByte);
655 655 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
656 656 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
657 657
658 658 // rax: local 0
659 659 // rbx: method
660 660 // rdx: constant pool cache index
661 661 // rdi: constant pool cache
662 662
663 663 // check if getfield has been resolved and read constant pool cache entry
664 664 // check the validity of the cache entry by testing whether _indices field
665 665 // contains Bytecode::_getfield in b1 byte.
666 666 assert(in_words(ConstantPoolCacheEntry::size()) == 4,
667 667 "adjust shift below");
668 668 __ movl(rcx,
669 669 Address(rdi,
670 670 rdx,
671 671 Address::times_8,
672 672 constantPoolCacheOopDesc::base_offset() +
673 673 ConstantPoolCacheEntry::indices_offset()));
674 674 __ shrl(rcx, 2 * BitsPerByte);
675 675 __ andl(rcx, 0xFF);
676 676 __ cmpl(rcx, Bytecodes::_getfield);
677 677 __ jcc(Assembler::notEqual, slow_path);
678 678
679 679 // Note: constant pool entry is not valid before bytecode is resolved
680 680 __ movptr(rcx,
681 681 Address(rdi,
682 682 rdx,
683 683 Address::times_8,
684 684 constantPoolCacheOopDesc::base_offset() +
685 685 ConstantPoolCacheEntry::f2_offset()));
686 686 // edx: flags
687 687 __ movl(rdx,
688 688 Address(rdi,
689 689 rdx,
690 690 Address::times_8,
691 691 constantPoolCacheOopDesc::base_offset() +
692 692 ConstantPoolCacheEntry::flags_offset()));
693 693
694 694 Label notObj, notInt, notByte, notShort;
695 695 const Address field_address(rax, rcx, Address::times_1);
696 696
697 697 // Need to differentiate between igetfield, agetfield, bgetfield etc.
698 698 // because they are different sizes.
699 699 // Use the type from the constant pool cache
700 700 __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
701 701 // Make sure we don't need to mask edx for tosBits after the above shift
702 702 ConstantPoolCacheEntry::verify_tosBits();
703 703
704 704 __ cmpl(rdx, atos);
705 705 __ jcc(Assembler::notEqual, notObj);
706 706 // atos
707 707 __ load_heap_oop(rax, field_address);
708 708 __ jmp(xreturn_path);
709 709
710 710 __ bind(notObj);
711 711 __ cmpl(rdx, itos);
712 712 __ jcc(Assembler::notEqual, notInt);
713 713 // itos
714 714 __ movl(rax, field_address);
715 715 __ jmp(xreturn_path);
716 716
717 717 __ bind(notInt);
718 718 __ cmpl(rdx, btos);
719 719 __ jcc(Assembler::notEqual, notByte);
720 720 // btos
721 721 __ load_signed_byte(rax, field_address);
722 722 __ jmp(xreturn_path);
723 723
724 724 __ bind(notByte);
725 725 __ cmpl(rdx, stos);
726 726 __ jcc(Assembler::notEqual, notShort);
727 727 // stos
728 728 __ load_signed_short(rax, field_address);
729 729 __ jmp(xreturn_path);
730 730
731 731 __ bind(notShort);
732 732 #ifdef ASSERT
733 733 Label okay;
734 734 __ cmpl(rdx, ctos);
735 735 __ jcc(Assembler::equal, okay);
736 736 __ stop("what type is this?");
737 737 __ bind(okay);
738 738 #endif
739 739 // ctos
740 740 __ load_unsigned_short(rax, field_address);
741 741
742 742 __ bind(xreturn_path);
743 743
744 744 // _ireturn/_areturn
745 745 __ pop(rdi);
746 746 __ mov(rsp, r13);
747 747 __ jmp(rdi);
748 748 __ ret(0);
749 749
↓ open down ↓ |
749 lines elided |
↑ open up ↑ |
750 750 // generate a vanilla interpreter entry as the slow path
751 751 __ bind(slow_path);
752 752 (void) generate_normal_entry(false);
753 753 } else {
754 754 (void) generate_normal_entry(false);
755 755 }
756 756
757 757 return entry_point;
758 758 }
759 759
760 +// Method entry for java.lang.ref.Reference.get.
761 +address InterpreterGenerator::generate_Reference_get_entry(void) {
762 +#ifndef SERIALGC
763 + // Code: _aload_0, _getfield, _areturn
764 + // parameter size = 1
765 + //
766 + // The code that gets generated by this routine is split into 2 parts:
767 + // 1. The "intrinsified" code for G1 (or any SATB based GC),
768 + // 2. The slow path - which is an expansion of the regular method entry.
769 + //
770 + // Notes:-
771 + // * In the G1 code we do not check whether we need to block for
772 + // a safepoint. If G1 is enabled then we must execute the specialized
773 + // code for Reference.get (except when the Reference object is null)
774 + // so that we can log the value in the referent field with an SATB
775 + // update buffer.
776 + // If the code for the getfield template is modified so that the
777 + // G1 pre-barrier code is executed when the current method is
778 + // Reference.get() then going through the normal method entry
779 + // will be fine.
780 + // * The G1 code can, however, check the receiver object (the instance
781 + // of java.lang.Reference) and jump to the slow path if null. If the
782 + // Reference object is null then we obviously cannot fetch the referent
783 + // and so we don't need to call the G1 pre-barrier. Thus we can use the
784 + // regular method entry code to generate the NPE.
785 + //
786 + // This code is based on generate_accessor_enty.
787 + //
788 + // rbx: methodOop
789 +
790 + // r13: senderSP must preserve for slow path, set SP to it on fast path
791 +
792 + address entry = __ pc();
793 +
794 + const int referent_offset = java_lang_ref_Reference::referent_offset;
795 + guarantee(referent_offset > 0, "referent offset not initialized");
796 +
797 + if (UseG1GC) {
798 + Label slow_path;
799 + // rbx: method
800 +
801 + // Check if local 0 != NULL
802 + // If the receiver is null then it is OK to jump to the slow path.
803 + __ movptr(rax, Address(rsp, wordSize));
804 +
805 + __ testptr(rax, rax);
806 + __ jcc(Assembler::zero, slow_path);
807 +
808 + // rax: local 0
809 + // rbx: method (but can be used as scratch now)
810 + // rdx: scratch
811 + // rdi: scratch
812 +
813 + // Generate the G1 pre-barrier code to log the value of
814 + // the referent field in an SATB buffer.
815 +
816 + // Load the value of the referent field.
817 + const Address field_address(rax, referent_offset);
818 + __ load_heap_oop(rax, field_address);
819 +
820 + // Generate the G1 pre-barrier code to log the value of
821 + // the referent field in an SATB buffer.
822 + __ g1_write_barrier_pre(noreg /* obj */,
823 + rax /* pre_val */,
824 + r15_thread /* thread */,
825 + rbx /* tmp */,
826 + true /* tosca_live */,
827 + true /* expand_call */);
828 +
829 + // _areturn
830 + __ pop(rdi); // get return address
831 + __ mov(rsp, r13); // set sp to sender sp
832 + __ jmp(rdi);
833 + __ ret(0);
834 +
835 + // generate a vanilla interpreter entry as the slow path
836 + __ bind(slow_path);
837 + (void) generate_normal_entry(false);
838 +
839 + return entry;
840 + }
841 +#endif // SERIALGC
842 +
843 + // If G1 is not enabled then attempt to go through the accessor entry point
844 + // Reference.get is an accessor
845 + return generate_accessor_entry();
846 +}
847 +
848 +
760 849 // Interpreter stub for calling a native method. (asm interpreter)
761 850 // This sets up a somewhat different looking stack for calling the
762 851 // native method than the typical interpreter frame setup.
763 852 address InterpreterGenerator::generate_native_entry(bool synchronized) {
764 853 // determine code generation flags
765 854 bool inc_counter = UseCompiler || CountCompiledCalls;
766 855
767 856 // rbx: methodOop
768 857 // r13: sender sp
769 858
770 859 address entry_point = __ pc();
771 860
772 861 const Address size_of_parameters(rbx, methodOopDesc::
773 862 size_of_parameters_offset());
774 863 const Address invocation_counter(rbx, methodOopDesc::
775 864 invocation_counter_offset() +
776 865 InvocationCounter::counter_offset());
777 866 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
778 867
779 868 // get parameter size (always needed)
780 869 __ load_unsigned_short(rcx, size_of_parameters);
781 870
782 871 // native calls don't need the stack size check since they have no
783 872 // expression stack and the arguments are already on the stack and
784 873 // we only add a handful of words to the stack
785 874
786 875 // rbx: methodOop
787 876 // rcx: size of parameters
788 877 // r13: sender sp
789 878 __ pop(rax); // get return address
790 879
791 880 // for natives the size of locals is zero
792 881
793 882 // compute beginning of parameters (r14)
794 883 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
795 884
796 885 // add 2 zero-initialized slots for native calls
797 886 // initialize result_handler slot
798 887 __ push((int) NULL_WORD);
799 888 // slot for oop temp
800 889 // (static native method holder mirror/jni oop result)
801 890 __ push((int) NULL_WORD);
802 891
803 892 if (inc_counter) {
804 893 __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
805 894 }
806 895
807 896 // initialize fixed part of activation frame
808 897 generate_fixed_frame(true);
809 898
810 899 // make sure method is native & not abstract
811 900 #ifdef ASSERT
812 901 __ movl(rax, access_flags);
813 902 {
814 903 Label L;
815 904 __ testl(rax, JVM_ACC_NATIVE);
816 905 __ jcc(Assembler::notZero, L);
817 906 __ stop("tried to execute non-native method as native");
818 907 __ bind(L);
819 908 }
820 909 {
821 910 Label L;
822 911 __ testl(rax, JVM_ACC_ABSTRACT);
823 912 __ jcc(Assembler::zero, L);
824 913 __ stop("tried to execute abstract method in interpreter");
825 914 __ bind(L);
826 915 }
827 916 #endif
828 917
829 918 // Since at this point in the method invocation the exception handler
830 919 // would try to exit the monitor of synchronized methods which hasn't
831 920 // been entered yet, we set the thread local variable
832 921 // _do_not_unlock_if_synchronized to true. The remove_activation will
833 922 // check this flag.
834 923
835 924 const Address do_not_unlock_if_synchronized(r15_thread,
836 925 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
837 926 __ movbool(do_not_unlock_if_synchronized, true);
838 927
839 928 // increment invocation count & check for overflow
840 929 Label invocation_counter_overflow;
841 930 if (inc_counter) {
842 931 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
843 932 }
844 933
845 934 Label continue_after_compile;
846 935 __ bind(continue_after_compile);
847 936
848 937 bang_stack_shadow_pages(true);
849 938
850 939 // reset the _do_not_unlock_if_synchronized flag
851 940 __ movbool(do_not_unlock_if_synchronized, false);
852 941
853 942 // check for synchronized methods
854 943 // Must happen AFTER invocation_counter check and stack overflow check,
855 944 // so method is not locked if overflows.
856 945 if (synchronized) {
857 946 lock_method();
858 947 } else {
859 948 // no synchronization necessary
860 949 #ifdef ASSERT
861 950 {
862 951 Label L;
863 952 __ movl(rax, access_flags);
864 953 __ testl(rax, JVM_ACC_SYNCHRONIZED);
865 954 __ jcc(Assembler::zero, L);
866 955 __ stop("method needs synchronization");
867 956 __ bind(L);
868 957 }
869 958 #endif
870 959 }
871 960
872 961 // start execution
873 962 #ifdef ASSERT
874 963 {
875 964 Label L;
876 965 const Address monitor_block_top(rbp,
877 966 frame::interpreter_frame_monitor_block_top_offset * wordSize);
878 967 __ movptr(rax, monitor_block_top);
879 968 __ cmpptr(rax, rsp);
880 969 __ jcc(Assembler::equal, L);
881 970 __ stop("broken stack frame setup in interpreter");
882 971 __ bind(L);
883 972 }
884 973 #endif
885 974
886 975 // jvmti support
887 976 __ notify_method_entry();
888 977
889 978 // work registers
890 979 const Register method = rbx;
891 980 const Register t = r11;
892 981
893 982 // allocate space for parameters
894 983 __ get_method(method);
895 984 __ verify_oop(method);
896 985 __ load_unsigned_short(t,
897 986 Address(method,
898 987 methodOopDesc::size_of_parameters_offset()));
899 988 __ shll(t, Interpreter::logStackElementSize);
900 989
901 990 __ subptr(rsp, t);
902 991 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
903 992 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
904 993
905 994 // get signature handler
906 995 {
907 996 Label L;
908 997 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
909 998 __ testptr(t, t);
910 999 __ jcc(Assembler::notZero, L);
911 1000 __ call_VM(noreg,
912 1001 CAST_FROM_FN_PTR(address,
913 1002 InterpreterRuntime::prepare_native_call),
914 1003 method);
915 1004 __ get_method(method);
916 1005 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
917 1006 __ bind(L);
918 1007 }
919 1008
920 1009 // call signature handler
921 1010 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
922 1011 "adjust this code");
923 1012 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
924 1013 "adjust this code");
925 1014 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
926 1015 "adjust this code");
927 1016
928 1017 // The generated handlers do not touch RBX (the method oop).
929 1018 // However, large signatures cannot be cached and are generated
930 1019 // each time here. The slow-path generator can do a GC on return,
931 1020 // so we must reload it after the call.
932 1021 __ call(t);
933 1022 __ get_method(method); // slow path can do a GC, reload RBX
934 1023
935 1024
936 1025 // result handler is in rax
937 1026 // set result handler
938 1027 __ movptr(Address(rbp,
939 1028 (frame::interpreter_frame_result_handler_offset) * wordSize),
940 1029 rax);
941 1030
942 1031 // pass mirror handle if static call
943 1032 {
944 1033 Label L;
945 1034 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() +
946 1035 Klass::java_mirror_offset_in_bytes();
947 1036 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
948 1037 __ testl(t, JVM_ACC_STATIC);
949 1038 __ jcc(Assembler::zero, L);
950 1039 // get mirror
951 1040 __ movptr(t, Address(method, methodOopDesc::constants_offset()));
952 1041 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
953 1042 __ movptr(t, Address(t, mirror_offset));
954 1043 // copy mirror into activation frame
955 1044 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
956 1045 t);
957 1046 // pass handle to mirror
958 1047 __ lea(c_rarg1,
959 1048 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
960 1049 __ bind(L);
961 1050 }
962 1051
963 1052 // get native function entry point
964 1053 {
965 1054 Label L;
966 1055 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
967 1056 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
968 1057 __ movptr(rscratch2, unsatisfied.addr());
969 1058 __ cmpptr(rax, rscratch2);
970 1059 __ jcc(Assembler::notEqual, L);
971 1060 __ call_VM(noreg,
972 1061 CAST_FROM_FN_PTR(address,
973 1062 InterpreterRuntime::prepare_native_call),
974 1063 method);
975 1064 __ get_method(method);
976 1065 __ verify_oop(method);
977 1066 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
978 1067 __ bind(L);
979 1068 }
980 1069
981 1070 // pass JNIEnv
982 1071 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
983 1072
984 1073 // It is enough that the pc() points into the right code
985 1074 // segment. It does not have to be the correct return pc.
986 1075 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
987 1076
988 1077 // change thread state
989 1078 #ifdef ASSERT
990 1079 {
991 1080 Label L;
992 1081 __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
993 1082 __ cmpl(t, _thread_in_Java);
994 1083 __ jcc(Assembler::equal, L);
995 1084 __ stop("Wrong thread state in native stub");
996 1085 __ bind(L);
997 1086 }
998 1087 #endif
999 1088
1000 1089 // Change state to native
1001 1090
1002 1091 __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1003 1092 _thread_in_native);
1004 1093
1005 1094 // Call the native method.
1006 1095 __ call(rax);
1007 1096 // result potentially in rax or xmm0
1008 1097
1009 1098 // Depending on runtime options, either restore the MXCSR
1010 1099 // register after returning from the JNI Call or verify that
1011 1100 // it wasn't changed during -Xcheck:jni.
1012 1101 if (RestoreMXCSROnJNICalls) {
1013 1102 __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
1014 1103 }
1015 1104 else if (CheckJNICalls) {
1016 1105 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry())));
1017 1106 }
1018 1107
1019 1108 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1020 1109 // in order to extract the result of a method call. If the order of these
1021 1110 // pushes change or anything else is added to the stack then the code in
1022 1111 // interpreter_frame_result must also change.
1023 1112
1024 1113 __ push(dtos);
1025 1114 __ push(ltos);
1026 1115
1027 1116 // change thread state
1028 1117 __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1029 1118 _thread_in_native_trans);
1030 1119
1031 1120 if (os::is_MP()) {
1032 1121 if (UseMembar) {
1033 1122 // Force this write out before the read below
1034 1123 __ membar(Assembler::Membar_mask_bits(
1035 1124 Assembler::LoadLoad | Assembler::LoadStore |
1036 1125 Assembler::StoreLoad | Assembler::StoreStore));
1037 1126 } else {
1038 1127 // Write serialization page so VM thread can do a pseudo remote membar.
1039 1128 // We use the current thread pointer to calculate a thread specific
1040 1129 // offset to write to within the page. This minimizes bus traffic
1041 1130 // due to cache line collision.
1042 1131 __ serialize_memory(r15_thread, rscratch2);
1043 1132 }
1044 1133 }
1045 1134
1046 1135 // check for safepoint operation in progress and/or pending suspend requests
1047 1136 {
1048 1137 Label Continue;
1049 1138 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1050 1139 SafepointSynchronize::_not_synchronized);
1051 1140
1052 1141 Label L;
1053 1142 __ jcc(Assembler::notEqual, L);
1054 1143 __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1055 1144 __ jcc(Assembler::equal, Continue);
1056 1145 __ bind(L);
1057 1146
1058 1147 // Don't use call_VM as it will see a possible pending exception
1059 1148 // and forward it and never return here preventing us from
1060 1149 // clearing _last_native_pc down below. Also can't use
1061 1150 // call_VM_leaf either as it will check to see if r13 & r14 are
1062 1151 // preserved and correspond to the bcp/locals pointers. So we do a
1063 1152 // runtime call by hand.
1064 1153 //
1065 1154 __ mov(c_rarg0, r15_thread);
1066 1155 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1067 1156 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1068 1157 __ andptr(rsp, -16); // align stack as required by ABI
1069 1158 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1070 1159 __ mov(rsp, r12); // restore sp
1071 1160 __ reinit_heapbase();
1072 1161 __ bind(Continue);
1073 1162 }
1074 1163
1075 1164 // change thread state
1076 1165 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1077 1166
1078 1167 // reset_last_Java_frame
1079 1168 __ reset_last_Java_frame(true, true);
1080 1169
1081 1170 // reset handle block
1082 1171 __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1083 1172 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1084 1173
1085 1174 // If result is an oop unbox and store it in frame where gc will see it
1086 1175 // and result handler will pick it up
1087 1176
1088 1177 {
1089 1178 Label no_oop, store_result;
1090 1179 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1091 1180 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1092 1181 __ jcc(Assembler::notEqual, no_oop);
1093 1182 // retrieve result
1094 1183 __ pop(ltos);
1095 1184 __ testptr(rax, rax);
1096 1185 __ jcc(Assembler::zero, store_result);
1097 1186 __ movptr(rax, Address(rax, 0));
1098 1187 __ bind(store_result);
1099 1188 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1100 1189 // keep stack depth as expected by pushing oop which will eventually be discarde
1101 1190 __ push(ltos);
1102 1191 __ bind(no_oop);
1103 1192 }
1104 1193
1105 1194
1106 1195 {
1107 1196 Label no_reguard;
1108 1197 __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1109 1198 JavaThread::stack_guard_yellow_disabled);
1110 1199 __ jcc(Assembler::notEqual, no_reguard);
1111 1200
1112 1201 __ pusha(); // XXX only save smashed registers
1113 1202 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1114 1203 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1115 1204 __ andptr(rsp, -16); // align stack as required by ABI
1116 1205 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1117 1206 __ mov(rsp, r12); // restore sp
1118 1207 __ popa(); // XXX only restore smashed registers
1119 1208 __ reinit_heapbase();
1120 1209
1121 1210 __ bind(no_reguard);
1122 1211 }
1123 1212
1124 1213
1125 1214 // The method register is junk from after the thread_in_native transition
1126 1215 // until here. Also can't call_VM until the bcp has been
1127 1216 // restored. Need bcp for throwing exception below so get it now.
1128 1217 __ get_method(method);
1129 1218 __ verify_oop(method);
1130 1219
1131 1220 // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1132 1221 // r13 == code_base()
1133 1222 __ movptr(r13, Address(method, methodOopDesc::const_offset())); // get constMethodOop
1134 1223 __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase
1135 1224 // handle exceptions (exception handling will handle unlocking!)
1136 1225 {
1137 1226 Label L;
1138 1227 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1139 1228 __ jcc(Assembler::zero, L);
1140 1229 // Note: At some point we may want to unify this with the code
1141 1230 // used in call_VM_base(); i.e., we should use the
1142 1231 // StubRoutines::forward_exception code. For now this doesn't work
1143 1232 // here because the rsp is not correctly set at this point.
1144 1233 __ MacroAssembler::call_VM(noreg,
1145 1234 CAST_FROM_FN_PTR(address,
1146 1235 InterpreterRuntime::throw_pending_exception));
1147 1236 __ should_not_reach_here();
1148 1237 __ bind(L);
1149 1238 }
1150 1239
1151 1240 // do unlocking if necessary
1152 1241 {
1153 1242 Label L;
1154 1243 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1155 1244 __ testl(t, JVM_ACC_SYNCHRONIZED);
1156 1245 __ jcc(Assembler::zero, L);
1157 1246 // the code below should be shared with interpreter macro
1158 1247 // assembler implementation
1159 1248 {
1160 1249 Label unlock;
1161 1250 // BasicObjectLock will be first in list, since this is a
1162 1251 // synchronized method. However, need to check that the object
1163 1252 // has not been unlocked by an explicit monitorexit bytecode.
1164 1253 const Address monitor(rbp,
1165 1254 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1166 1255 wordSize - sizeof(BasicObjectLock)));
1167 1256
1168 1257 // monitor expect in c_rarg1 for slow unlock path
1169 1258 __ lea(c_rarg1, monitor); // address of first monitor
1170 1259
1171 1260 __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1172 1261 __ testptr(t, t);
1173 1262 __ jcc(Assembler::notZero, unlock);
1174 1263
1175 1264 // Entry already unlocked, need to throw exception
1176 1265 __ MacroAssembler::call_VM(noreg,
1177 1266 CAST_FROM_FN_PTR(address,
1178 1267 InterpreterRuntime::throw_illegal_monitor_state_exception));
1179 1268 __ should_not_reach_here();
1180 1269
1181 1270 __ bind(unlock);
1182 1271 __ unlock_object(c_rarg1);
1183 1272 }
1184 1273 __ bind(L);
1185 1274 }
1186 1275
1187 1276 // jvmti support
1188 1277 // Note: This must happen _after_ handling/throwing any exceptions since
1189 1278 // the exception handler code notifies the runtime of method exits
1190 1279 // too. If this happens before, method entry/exit notifications are
1191 1280 // not properly paired (was bug - gri 11/22/99).
1192 1281 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1193 1282
1194 1283 // restore potential result in edx:eax, call result handler to
1195 1284 // restore potential result in ST0 & handle result
1196 1285
1197 1286 __ pop(ltos);
1198 1287 __ pop(dtos);
1199 1288
1200 1289 __ movptr(t, Address(rbp,
1201 1290 (frame::interpreter_frame_result_handler_offset) * wordSize));
1202 1291 __ call(t);
1203 1292
1204 1293 // remove activation
1205 1294 __ movptr(t, Address(rbp,
1206 1295 frame::interpreter_frame_sender_sp_offset *
1207 1296 wordSize)); // get sender sp
1208 1297 __ leave(); // remove frame anchor
1209 1298 __ pop(rdi); // get return address
1210 1299 __ mov(rsp, t); // set sp to sender sp
1211 1300 __ jmp(rdi);
1212 1301
1213 1302 if (inc_counter) {
1214 1303 // Handle overflow of counter and compile method
1215 1304 __ bind(invocation_counter_overflow);
1216 1305 generate_counter_overflow(&continue_after_compile);
1217 1306 }
1218 1307
1219 1308 return entry_point;
1220 1309 }
1221 1310
1222 1311 //
1223 1312 // Generic interpreted method entry to (asm) interpreter
1224 1313 //
1225 1314 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1226 1315 // determine code generation flags
1227 1316 bool inc_counter = UseCompiler || CountCompiledCalls;
1228 1317
1229 1318 // ebx: methodOop
1230 1319 // r13: sender sp
1231 1320 address entry_point = __ pc();
1232 1321
1233 1322 const Address size_of_parameters(rbx,
1234 1323 methodOopDesc::size_of_parameters_offset());
1235 1324 const Address size_of_locals(rbx, methodOopDesc::size_of_locals_offset());
1236 1325 const Address invocation_counter(rbx,
1237 1326 methodOopDesc::invocation_counter_offset() +
1238 1327 InvocationCounter::counter_offset());
1239 1328 const Address access_flags(rbx, methodOopDesc::access_flags_offset());
1240 1329
1241 1330 // get parameter size (always needed)
1242 1331 __ load_unsigned_short(rcx, size_of_parameters);
1243 1332
1244 1333 // rbx: methodOop
1245 1334 // rcx: size of parameters
1246 1335 // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1247 1336
1248 1337 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1249 1338 __ subl(rdx, rcx); // rdx = no. of additional locals
1250 1339
1251 1340 // YYY
1252 1341 // __ incrementl(rdx);
1253 1342 // __ andl(rdx, -2);
1254 1343
1255 1344 // see if we've got enough room on the stack for locals plus overhead.
1256 1345 generate_stack_overflow_check();
1257 1346
1258 1347 // get return address
1259 1348 __ pop(rax);
1260 1349
1261 1350 // compute beginning of parameters (r14)
1262 1351 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1263 1352
1264 1353 // rdx - # of additional locals
1265 1354 // allocate space for locals
1266 1355 // explicitly initialize locals
1267 1356 {
1268 1357 Label exit, loop;
1269 1358 __ testl(rdx, rdx);
1270 1359 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1271 1360 __ bind(loop);
1272 1361 __ push((int) NULL_WORD); // initialize local variables
1273 1362 __ decrementl(rdx); // until everything initialized
1274 1363 __ jcc(Assembler::greater, loop);
1275 1364 __ bind(exit);
1276 1365 }
1277 1366
1278 1367 // (pre-)fetch invocation count
1279 1368 if (inc_counter) {
1280 1369 __ movl(rcx, invocation_counter);
1281 1370 }
1282 1371 // initialize fixed part of activation frame
1283 1372 generate_fixed_frame(false);
1284 1373
1285 1374 // make sure method is not native & not abstract
1286 1375 #ifdef ASSERT
1287 1376 __ movl(rax, access_flags);
1288 1377 {
1289 1378 Label L;
1290 1379 __ testl(rax, JVM_ACC_NATIVE);
1291 1380 __ jcc(Assembler::zero, L);
1292 1381 __ stop("tried to execute native method as non-native");
1293 1382 __ bind(L);
1294 1383 }
1295 1384 {
1296 1385 Label L;
1297 1386 __ testl(rax, JVM_ACC_ABSTRACT);
1298 1387 __ jcc(Assembler::zero, L);
1299 1388 __ stop("tried to execute abstract method in interpreter");
1300 1389 __ bind(L);
1301 1390 }
1302 1391 #endif
1303 1392
1304 1393 // Since at this point in the method invocation the exception
1305 1394 // handler would try to exit the monitor of synchronized methods
1306 1395 // which hasn't been entered yet, we set the thread local variable
1307 1396 // _do_not_unlock_if_synchronized to true. The remove_activation
1308 1397 // will check this flag.
1309 1398
1310 1399 const Address do_not_unlock_if_synchronized(r15_thread,
1311 1400 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1312 1401 __ movbool(do_not_unlock_if_synchronized, true);
1313 1402
1314 1403 // increment invocation count & check for overflow
1315 1404 Label invocation_counter_overflow;
1316 1405 Label profile_method;
1317 1406 Label profile_method_continue;
1318 1407 if (inc_counter) {
1319 1408 generate_counter_incr(&invocation_counter_overflow,
1320 1409 &profile_method,
1321 1410 &profile_method_continue);
1322 1411 if (ProfileInterpreter) {
1323 1412 __ bind(profile_method_continue);
1324 1413 }
1325 1414 }
1326 1415
1327 1416 Label continue_after_compile;
1328 1417 __ bind(continue_after_compile);
1329 1418
1330 1419 // check for synchronized interpreted methods
1331 1420 bang_stack_shadow_pages(false);
1332 1421
1333 1422 // reset the _do_not_unlock_if_synchronized flag
1334 1423 __ movbool(do_not_unlock_if_synchronized, false);
1335 1424
1336 1425 // check for synchronized methods
1337 1426 // Must happen AFTER invocation_counter check and stack overflow check,
1338 1427 // so method is not locked if overflows.
1339 1428 if (synchronized) {
1340 1429 // Allocate monitor and lock method
1341 1430 lock_method();
1342 1431 } else {
1343 1432 // no synchronization necessary
1344 1433 #ifdef ASSERT
1345 1434 {
1346 1435 Label L;
1347 1436 __ movl(rax, access_flags);
1348 1437 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1349 1438 __ jcc(Assembler::zero, L);
1350 1439 __ stop("method needs synchronization");
1351 1440 __ bind(L);
1352 1441 }
1353 1442 #endif
1354 1443 }
1355 1444
1356 1445 // start execution
1357 1446 #ifdef ASSERT
1358 1447 {
1359 1448 Label L;
1360 1449 const Address monitor_block_top (rbp,
1361 1450 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1362 1451 __ movptr(rax, monitor_block_top);
1363 1452 __ cmpptr(rax, rsp);
1364 1453 __ jcc(Assembler::equal, L);
1365 1454 __ stop("broken stack frame setup in interpreter");
1366 1455 __ bind(L);
1367 1456 }
1368 1457 #endif
1369 1458
1370 1459 // jvmti support
1371 1460 __ notify_method_entry();
1372 1461
1373 1462 __ dispatch_next(vtos);
1374 1463
1375 1464 // invocation counter overflow
1376 1465 if (inc_counter) {
1377 1466 if (ProfileInterpreter) {
1378 1467 // We have decided to profile this method in the interpreter
1379 1468 __ bind(profile_method);
1380 1469 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1381 1470 __ set_method_data_pointer_for_bcp();
1382 1471 __ get_method(rbx);
1383 1472 __ jmp(profile_method_continue);
1384 1473 }
1385 1474 // Handle overflow of counter and compile method
1386 1475 __ bind(invocation_counter_overflow);
1387 1476 generate_counter_overflow(&continue_after_compile);
1388 1477 }
1389 1478
1390 1479 return entry_point;
1391 1480 }
1392 1481
1393 1482 // Entry points
1394 1483 //
1395 1484 // Here we generate the various kind of entries into the interpreter.
1396 1485 // The two main entry type are generic bytecode methods and native
1397 1486 // call method. These both come in synchronized and non-synchronized
1398 1487 // versions but the frame layout they create is very similar. The
1399 1488 // other method entry types are really just special purpose entries
1400 1489 // that are really entry and interpretation all in one. These are for
1401 1490 // trivial methods like accessor, empty, or special math methods.
1402 1491 //
1403 1492 // When control flow reaches any of the entry types for the interpreter
1404 1493 // the following holds ->
1405 1494 //
1406 1495 // Arguments:
1407 1496 //
1408 1497 // rbx: methodOop
1409 1498 //
1410 1499 // Stack layout immediately at entry
1411 1500 //
1412 1501 // [ return address ] <--- rsp
1413 1502 // [ parameter n ]
1414 1503 // ...
1415 1504 // [ parameter 1 ]
1416 1505 // [ expression stack ] (caller's java expression stack)
1417 1506
1418 1507 // Assuming that we don't go to one of the trivial specialized entries
1419 1508 // the stack will look like below when we are ready to execute the
1420 1509 // first bytecode (or call the native routine). The register usage
1421 1510 // will be as the template based interpreter expects (see
1422 1511 // interpreter_amd64.hpp).
1423 1512 //
1424 1513 // local variables follow incoming parameters immediately; i.e.
1425 1514 // the return address is moved to the end of the locals).
1426 1515 //
1427 1516 // [ monitor entry ] <--- rsp
1428 1517 // ...
1429 1518 // [ monitor entry ]
1430 1519 // [ expr. stack bottom ]
1431 1520 // [ saved r13 ]
1432 1521 // [ current r14 ]
1433 1522 // [ methodOop ]
1434 1523 // [ saved ebp ] <--- rbp
1435 1524 // [ return address ]
1436 1525 // [ local variable m ]
1437 1526 // ...
1438 1527 // [ local variable 1 ]
1439 1528 // [ parameter n ]
1440 1529 // ...
1441 1530 // [ parameter 1 ] <--- r14
1442 1531
1443 1532 address AbstractInterpreterGenerator::generate_method_entry(
1444 1533 AbstractInterpreter::MethodKind kind) {
1445 1534 // determine code generation flags
1446 1535 bool synchronized = false;
1447 1536 address entry_point = NULL;
1448 1537
1449 1538 switch (kind) {
1450 1539 case Interpreter::zerolocals : break;
1451 1540 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1452 1541 case Interpreter::native : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break;
1453 1542 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true); break;
1454 1543 case Interpreter::empty : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry(); break;
1455 1544 case Interpreter::accessor : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry(); break;
↓ open down ↓ |
686 lines elided |
↑ open up ↑ |
1456 1545 case Interpreter::abstract : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry(); break;
1457 1546 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*) this)->generate_method_handle_entry();break;
1458 1547
1459 1548 case Interpreter::java_lang_math_sin : // fall thru
1460 1549 case Interpreter::java_lang_math_cos : // fall thru
1461 1550 case Interpreter::java_lang_math_tan : // fall thru
1462 1551 case Interpreter::java_lang_math_abs : // fall thru
1463 1552 case Interpreter::java_lang_math_log : // fall thru
1464 1553 case Interpreter::java_lang_math_log10 : // fall thru
1465 1554 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind); break;
1555 + case Interpreter::java_lang_ref_reference_get
1556 + : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1466 1557 default : ShouldNotReachHere(); break;
1467 1558 }
1468 1559
1469 1560 if (entry_point) {
1470 1561 return entry_point;
1471 1562 }
1472 1563
1473 1564 return ((InterpreterGenerator*) this)->
1474 1565 generate_normal_entry(synchronized);
1475 1566 }
1476 1567
1477 1568 // These should never be compiled since the interpreter will prefer
1478 1569 // the compiled version to the intrinsic version.
1479 1570 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1480 1571 switch (method_kind(m)) {
1481 1572 case Interpreter::java_lang_math_sin : // fall thru
1482 1573 case Interpreter::java_lang_math_cos : // fall thru
1483 1574 case Interpreter::java_lang_math_tan : // fall thru
1484 1575 case Interpreter::java_lang_math_abs : // fall thru
1485 1576 case Interpreter::java_lang_math_log : // fall thru
1486 1577 case Interpreter::java_lang_math_log10 : // fall thru
1487 1578 case Interpreter::java_lang_math_sqrt :
1488 1579 return false;
1489 1580 default:
1490 1581 return true;
1491 1582 }
1492 1583 }
1493 1584
1494 1585 // How much stack a method activation needs in words.
1495 1586 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1496 1587 const int entry_size = frame::interpreter_frame_monitor_size();
1497 1588
1498 1589 // total overhead size: entry_size + (saved rbp thru expr stack
1499 1590 // bottom). be sure to change this if you add/subtract anything
1500 1591 // to/from the overhead area
1501 1592 const int overhead_size =
1502 1593 -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1503 1594
1504 1595 const int stub_code = frame::entry_frame_after_call_words;
1505 1596 const int extra_stack = methodOopDesc::extra_stack_entries();
1506 1597 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1507 1598 Interpreter::stackElementWords;
1508 1599 return (overhead_size + method_stack + stub_code);
1509 1600 }
1510 1601
1511 1602 int AbstractInterpreter::layout_activation(methodOop method,
1512 1603 int tempcount,
1513 1604 int popframe_extra_args,
1514 1605 int moncount,
1515 1606 int callee_param_count,
1516 1607 int callee_locals,
1517 1608 frame* caller,
1518 1609 frame* interpreter_frame,
1519 1610 bool is_top_frame) {
1520 1611 // Note: This calculation must exactly parallel the frame setup
1521 1612 // in AbstractInterpreterGenerator::generate_method_entry.
1522 1613 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1523 1614 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1524 1615 // right size, as determined by a previous call to this method.
1525 1616 // It is also guaranteed to be walkable even though it is in a skeletal state
1526 1617
1527 1618 // fixed size of an interpreter frame:
1528 1619 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1529 1620 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1530 1621 Interpreter::stackElementWords;
1531 1622
1532 1623 int overhead = frame::sender_sp_offset -
1533 1624 frame::interpreter_frame_initial_sp_offset;
1534 1625 // Our locals were accounted for by the caller (or last_frame_adjust
1535 1626 // on the transistion) Since the callee parameters already account
1536 1627 // for the callee's params we only need to account for the extra
1537 1628 // locals.
1538 1629 int size = overhead +
1539 1630 (callee_locals - callee_param_count)*Interpreter::stackElementWords +
1540 1631 moncount * frame::interpreter_frame_monitor_size() +
1541 1632 tempcount* Interpreter::stackElementWords + popframe_extra_args;
1542 1633 if (interpreter_frame != NULL) {
1543 1634 #ifdef ASSERT
1544 1635 if (!EnableMethodHandles)
1545 1636 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1546 1637 // Probably, since deoptimization doesn't work yet.
1547 1638 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1548 1639 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1549 1640 #endif
1550 1641
1551 1642 interpreter_frame->interpreter_frame_set_method(method);
1552 1643 // NOTE the difference in using sender_sp and
1553 1644 // interpreter_frame_sender_sp interpreter_frame_sender_sp is
1554 1645 // the original sp of the caller (the unextended_sp) and
1555 1646 // sender_sp is fp+16 XXX
1556 1647 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1557 1648
1558 1649 interpreter_frame->interpreter_frame_set_locals(locals);
1559 1650 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1560 1651 BasicObjectLock* monbot = montop - moncount;
1561 1652 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1562 1653
1563 1654 // Set last_sp
1564 1655 intptr_t* esp = (intptr_t*) monbot -
1565 1656 tempcount*Interpreter::stackElementWords -
1566 1657 popframe_extra_args;
1567 1658 interpreter_frame->interpreter_frame_set_last_sp(esp);
1568 1659
1569 1660 // All frames but the initial (oldest) interpreter frame we fill in have
1570 1661 // a value for sender_sp that allows walking the stack but isn't
1571 1662 // truly correct. Correct the value here.
1572 1663 if (extra_locals != 0 &&
1573 1664 interpreter_frame->sender_sp() ==
1574 1665 interpreter_frame->interpreter_frame_sender_sp()) {
1575 1666 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() +
1576 1667 extra_locals);
1577 1668 }
1578 1669 *interpreter_frame->interpreter_frame_cache_addr() =
1579 1670 method->constants()->cache();
1580 1671 }
1581 1672 return size;
1582 1673 }
1583 1674
1584 1675 //-----------------------------------------------------------------------------
1585 1676 // Exceptions
1586 1677
1587 1678 void TemplateInterpreterGenerator::generate_throw_exception() {
1588 1679 // Entry point in previous activation (i.e., if the caller was
1589 1680 // interpreted)
1590 1681 Interpreter::_rethrow_exception_entry = __ pc();
1591 1682 // Restore sp to interpreter_frame_last_sp even though we are going
1592 1683 // to empty the expression stack for the exception processing.
1593 1684 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1594 1685 // rax: exception
1595 1686 // rdx: return address/pc that threw exception
1596 1687 __ restore_bcp(); // r13 points to call/send
1597 1688 __ restore_locals();
1598 1689 __ reinit_heapbase(); // restore r12 as heapbase.
1599 1690 // Entry point for exceptions thrown within interpreter code
1600 1691 Interpreter::_throw_exception_entry = __ pc();
1601 1692 // expression stack is undefined here
1602 1693 // rax: exception
1603 1694 // r13: exception bcp
1604 1695 __ verify_oop(rax);
1605 1696 __ mov(c_rarg1, rax);
1606 1697
1607 1698 // expression stack must be empty before entering the VM in case of
1608 1699 // an exception
1609 1700 __ empty_expression_stack();
1610 1701 // find exception handler address and preserve exception oop
1611 1702 __ call_VM(rdx,
1612 1703 CAST_FROM_FN_PTR(address,
1613 1704 InterpreterRuntime::exception_handler_for_exception),
1614 1705 c_rarg1);
1615 1706 // rax: exception handler entry point
1616 1707 // rdx: preserved exception oop
1617 1708 // r13: bcp for exception handler
1618 1709 __ push_ptr(rdx); // push exception which is now the only value on the stack
1619 1710 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1620 1711
1621 1712 // If the exception is not handled in the current frame the frame is
1622 1713 // removed and the exception is rethrown (i.e. exception
1623 1714 // continuation is _rethrow_exception).
1624 1715 //
1625 1716 // Note: At this point the bci is still the bxi for the instruction
1626 1717 // which caused the exception and the expression stack is
1627 1718 // empty. Thus, for any VM calls at this point, GC will find a legal
1628 1719 // oop map (with empty expression stack).
1629 1720
1630 1721 // In current activation
1631 1722 // tos: exception
1632 1723 // esi: exception bcp
1633 1724
1634 1725 //
1635 1726 // JVMTI PopFrame support
1636 1727 //
1637 1728
1638 1729 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1639 1730 __ empty_expression_stack();
1640 1731 // Set the popframe_processing bit in pending_popframe_condition
1641 1732 // indicating that we are currently handling popframe, so that
1642 1733 // call_VMs that may happen later do not trigger new popframe
1643 1734 // handling cycles.
1644 1735 __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1645 1736 __ orl(rdx, JavaThread::popframe_processing_bit);
1646 1737 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1647 1738
1648 1739 {
1649 1740 // Check to see whether we are returning to a deoptimized frame.
1650 1741 // (The PopFrame call ensures that the caller of the popped frame is
1651 1742 // either interpreted or compiled and deoptimizes it if compiled.)
1652 1743 // In this case, we can't call dispatch_next() after the frame is
1653 1744 // popped, but instead must save the incoming arguments and restore
1654 1745 // them after deoptimization has occurred.
1655 1746 //
1656 1747 // Note that we don't compare the return PC against the
1657 1748 // deoptimization blob's unpack entry because of the presence of
1658 1749 // adapter frames in C2.
1659 1750 Label caller_not_deoptimized;
1660 1751 __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1661 1752 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1662 1753 InterpreterRuntime::interpreter_contains), c_rarg1);
1663 1754 __ testl(rax, rax);
1664 1755 __ jcc(Assembler::notZero, caller_not_deoptimized);
1665 1756
1666 1757 // Compute size of arguments for saving when returning to
1667 1758 // deoptimized caller
1668 1759 __ get_method(rax);
1669 1760 __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::
1670 1761 size_of_parameters_offset())));
1671 1762 __ shll(rax, Interpreter::logStackElementSize);
1672 1763 __ restore_locals(); // XXX do we need this?
1673 1764 __ subptr(r14, rax);
1674 1765 __ addptr(r14, wordSize);
1675 1766 // Save these arguments
1676 1767 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1677 1768 Deoptimization::
1678 1769 popframe_preserve_args),
1679 1770 r15_thread, rax, r14);
1680 1771
1681 1772 __ remove_activation(vtos, rdx,
1682 1773 /* throw_monitor_exception */ false,
1683 1774 /* install_monitor_exception */ false,
1684 1775 /* notify_jvmdi */ false);
1685 1776
1686 1777 // Inform deoptimization that it is responsible for restoring
1687 1778 // these arguments
1688 1779 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1689 1780 JavaThread::popframe_force_deopt_reexecution_bit);
1690 1781
1691 1782 // Continue in deoptimization handler
1692 1783 __ jmp(rdx);
1693 1784
1694 1785 __ bind(caller_not_deoptimized);
1695 1786 }
1696 1787
1697 1788 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1698 1789 /* throw_monitor_exception */ false,
1699 1790 /* install_monitor_exception */ false,
1700 1791 /* notify_jvmdi */ false);
1701 1792
1702 1793 // Finish with popframe handling
1703 1794 // A previous I2C followed by a deoptimization might have moved the
1704 1795 // outgoing arguments further up the stack. PopFrame expects the
1705 1796 // mutations to those outgoing arguments to be preserved and other
1706 1797 // constraints basically require this frame to look exactly as
1707 1798 // though it had previously invoked an interpreted activation with
1708 1799 // no space between the top of the expression stack (current
1709 1800 // last_sp) and the top of stack. Rather than force deopt to
1710 1801 // maintain this kind of invariant all the time we call a small
1711 1802 // fixup routine to move the mutated arguments onto the top of our
1712 1803 // expression stack if necessary.
1713 1804 __ mov(c_rarg1, rsp);
1714 1805 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1715 1806 // PC must point into interpreter here
1716 1807 __ set_last_Java_frame(noreg, rbp, __ pc());
1717 1808 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1718 1809 __ reset_last_Java_frame(true, true);
1719 1810 // Restore the last_sp and null it out
1720 1811 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1721 1812 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1722 1813
1723 1814 __ restore_bcp(); // XXX do we need this?
1724 1815 __ restore_locals(); // XXX do we need this?
1725 1816 // The method data pointer was incremented already during
1726 1817 // call profiling. We have to restore the mdp for the current bcp.
1727 1818 if (ProfileInterpreter) {
1728 1819 __ set_method_data_pointer_for_bcp();
1729 1820 }
1730 1821
1731 1822 // Clear the popframe condition flag
1732 1823 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1733 1824 JavaThread::popframe_inactive);
1734 1825
1735 1826 __ dispatch_next(vtos);
1736 1827 // end of PopFrame support
1737 1828
1738 1829 Interpreter::_remove_activation_entry = __ pc();
1739 1830
1740 1831 // preserve exception over this code sequence
1741 1832 __ pop_ptr(rax);
1742 1833 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1743 1834 // remove the activation (without doing throws on illegalMonitorExceptions)
1744 1835 __ remove_activation(vtos, rdx, false, true, false);
1745 1836 // restore exception
1746 1837 __ movptr(rax, Address(r15_thread, JavaThread::vm_result_offset()));
1747 1838 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD);
1748 1839 __ verify_oop(rax);
1749 1840
1750 1841 // In between activations - previous activation type unknown yet
1751 1842 // compute continuation point - the continuation point expects the
1752 1843 // following registers set up:
1753 1844 //
1754 1845 // rax: exception
1755 1846 // rdx: return address/pc that threw exception
1756 1847 // rsp: expression stack of caller
1757 1848 // rbp: ebp of caller
1758 1849 __ push(rax); // save exception
1759 1850 __ push(rdx); // save return address
1760 1851 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1761 1852 SharedRuntime::exception_handler_for_return_address),
1762 1853 r15_thread, rdx);
1763 1854 __ mov(rbx, rax); // save exception handler
1764 1855 __ pop(rdx); // restore return address
1765 1856 __ pop(rax); // restore exception
1766 1857 // Note that an "issuing PC" is actually the next PC after the call
1767 1858 __ jmp(rbx); // jump to exception
1768 1859 // handler of caller
1769 1860 }
1770 1861
1771 1862
1772 1863 //
1773 1864 // JVMTI ForceEarlyReturn support
1774 1865 //
1775 1866 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1776 1867 address entry = __ pc();
1777 1868
1778 1869 __ restore_bcp();
1779 1870 __ restore_locals();
1780 1871 __ empty_expression_stack();
1781 1872 __ load_earlyret_value(state);
1782 1873
1783 1874 __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1784 1875 Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1785 1876
1786 1877 // Clear the earlyret state
1787 1878 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1788 1879
1789 1880 __ remove_activation(state, rsi,
1790 1881 false, /* throw_monitor_exception */
1791 1882 false, /* install_monitor_exception */
1792 1883 true); /* notify_jvmdi */
1793 1884 __ jmp(rsi);
1794 1885
1795 1886 return entry;
1796 1887 } // end of ForceEarlyReturn support
1797 1888
1798 1889
1799 1890 //-----------------------------------------------------------------------------
1800 1891 // Helper for vtos entry point generation
1801 1892
1802 1893 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1803 1894 address& bep,
1804 1895 address& cep,
1805 1896 address& sep,
1806 1897 address& aep,
1807 1898 address& iep,
1808 1899 address& lep,
1809 1900 address& fep,
1810 1901 address& dep,
1811 1902 address& vep) {
1812 1903 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1813 1904 Label L;
1814 1905 aep = __ pc(); __ push_ptr(); __ jmp(L);
1815 1906 fep = __ pc(); __ push_f(); __ jmp(L);
1816 1907 dep = __ pc(); __ push_d(); __ jmp(L);
1817 1908 lep = __ pc(); __ push_l(); __ jmp(L);
1818 1909 bep = cep = sep =
1819 1910 iep = __ pc(); __ push_i();
1820 1911 vep = __ pc();
1821 1912 __ bind(L);
1822 1913 generate_and_dispatch(t);
1823 1914 }
1824 1915
1825 1916
1826 1917 //-----------------------------------------------------------------------------
1827 1918 // Generation of individual instructions
1828 1919
1829 1920 // helpers for generate_and_dispatch
1830 1921
1831 1922
1832 1923 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1833 1924 : TemplateInterpreterGenerator(code) {
1834 1925 generate_all(); // down here so it can be "virtual"
1835 1926 }
1836 1927
1837 1928 //-----------------------------------------------------------------------------
1838 1929
1839 1930 // Non-product code
1840 1931 #ifndef PRODUCT
1841 1932 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1842 1933 address entry = __ pc();
1843 1934
1844 1935 __ push(state);
1845 1936 __ push(c_rarg0);
1846 1937 __ push(c_rarg1);
1847 1938 __ push(c_rarg2);
1848 1939 __ push(c_rarg3);
1849 1940 __ mov(c_rarg2, rax); // Pass itos
1850 1941 #ifdef _WIN64
1851 1942 __ movflt(xmm3, xmm0); // Pass ftos
1852 1943 #endif
1853 1944 __ call_VM(noreg,
1854 1945 CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1855 1946 c_rarg1, c_rarg2, c_rarg3);
1856 1947 __ pop(c_rarg3);
1857 1948 __ pop(c_rarg2);
1858 1949 __ pop(c_rarg1);
1859 1950 __ pop(c_rarg0);
1860 1951 __ pop(state);
1861 1952 __ ret(0); // return from result handler
1862 1953
1863 1954 return entry;
1864 1955 }
1865 1956
1866 1957 void TemplateInterpreterGenerator::count_bytecode() {
1867 1958 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1868 1959 }
1869 1960
1870 1961 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1871 1962 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1872 1963 }
1873 1964
1874 1965 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1875 1966 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1876 1967 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1877 1968 __ orl(rbx,
1878 1969 ((int) t->bytecode()) <<
1879 1970 BytecodePairHistogram::log2_number_of_codes);
1880 1971 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1881 1972 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1882 1973 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1883 1974 }
1884 1975
1885 1976
1886 1977 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1887 1978 // Call a little run-time stub to avoid blow-up for each bytecode.
1888 1979 // The run-time runtime saves the right registers, depending on
1889 1980 // the tosca in-state for the given template.
1890 1981
1891 1982 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1892 1983 "entry must have been generated");
1893 1984 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1894 1985 __ andptr(rsp, -16); // align stack as required by ABI
1895 1986 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1896 1987 __ mov(rsp, r12); // restore sp
1897 1988 __ reinit_heapbase();
1898 1989 }
1899 1990
1900 1991
1901 1992 void TemplateInterpreterGenerator::stop_interpreter_at() {
1902 1993 Label L;
1903 1994 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1904 1995 StopInterpreterAt);
1905 1996 __ jcc(Assembler::notEqual, L);
1906 1997 __ int3();
1907 1998 __ bind(L);
1908 1999 }
1909 2000 #endif // !PRODUCT
1910 2001 #endif // ! CC_INTERP
↓ open down ↓ |
435 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX