Mercurial > hg > truffle
annotate src/cpu/x86/vm/templateInterpreter_x86_64.cpp @ 622:56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
Summary: minor assembler enhancements preparing for method handles
Reviewed-by: kvn
author | jrose |
---|---|
date | Wed, 04 Mar 2009 09:58:39 -0800 |
parents | 98cb887364d3 |
children | 819880572f09 |
rev | line source |
---|---|
0 | 1 /* |
196 | 2 * Copyright 2003-2008 Sun Microsystems, Inc. All Rights Reserved. |
0 | 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
4 * | |
5 * This code is free software; you can redistribute it and/or modify it | |
6 * under the terms of the GNU General Public License version 2 only, as | |
7 * published by the Free Software Foundation. | |
8 * | |
9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
12 * version 2 for more details (a copy is included in the LICENSE file that | |
13 * accompanied this code). | |
14 * | |
15 * You should have received a copy of the GNU General Public License version | |
16 * 2 along with this work; if not, write to the Free Software Foundation, | |
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
18 * | |
19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
21 * have any questions. | |
22 * | |
23 */ | |
24 | |
25 #include "incls/_precompiled.incl" | |
26 #include "incls/_interpreter_x86_64.cpp.incl" | |
27 | |
28 #define __ _masm-> | |
29 | |
304 | 30 #ifndef CC_INTERP |
31 | |
0 | 32 const int method_offset = frame::interpreter_frame_method_offset * wordSize; |
33 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize; | |
34 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize; | |
35 | |
36 //----------------------------------------------------------------------------- | |
37 | |
38 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { | |
39 address entry = __ pc(); | |
40 | |
41 #ifdef ASSERT | |
42 { | |
43 Label L; | |
304 | 44 __ lea(rax, Address(rbp, |
45 frame::interpreter_frame_monitor_block_top_offset * | |
46 wordSize)); | |
47 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack | |
48 // grows negative) | |
0 | 49 __ jcc(Assembler::aboveEqual, L); // check if frame is complete |
50 __ stop ("interpreter frame not set up"); | |
51 __ bind(L); | |
52 } | |
53 #endif // ASSERT | |
54 // Restore bcp under the assumption that the current frame is still | |
55 // interpreted | |
56 __ restore_bcp(); | |
57 | |
58 // expression stack must be empty before entering the VM if an | |
59 // exception happened | |
60 __ empty_expression_stack(); | |
61 // throw exception | |
62 __ call_VM(noreg, | |
63 CAST_FROM_FN_PTR(address, | |
64 InterpreterRuntime::throw_StackOverflowError)); | |
65 return entry; | |
66 } | |
67 | |
68 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler( | |
69 const char* name) { | |
70 address entry = __ pc(); | |
71 // expression stack must be empty before entering the VM if an | |
72 // exception happened | |
73 __ empty_expression_stack(); | |
74 // setup parameters | |
75 // ??? convention: expect aberrant index in register ebx | |
76 __ lea(c_rarg1, ExternalAddress((address)name)); | |
77 __ call_VM(noreg, | |
78 CAST_FROM_FN_PTR(address, | |
79 InterpreterRuntime:: | |
80 throw_ArrayIndexOutOfBoundsException), | |
81 c_rarg1, rbx); | |
82 return entry; | |
83 } | |
84 | |
85 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { | |
86 address entry = __ pc(); | |
87 | |
88 // object is at TOS | |
304 | 89 __ pop(c_rarg1); |
0 | 90 |
91 // expression stack must be empty before entering the VM if an | |
92 // exception happened | |
93 __ empty_expression_stack(); | |
94 | |
95 __ call_VM(noreg, | |
96 CAST_FROM_FN_PTR(address, | |
97 InterpreterRuntime:: | |
98 throw_ClassCastException), | |
99 c_rarg1); | |
100 return entry; | |
101 } | |
102 | |
103 address TemplateInterpreterGenerator::generate_exception_handler_common( | |
104 const char* name, const char* message, bool pass_oop) { | |
105 assert(!pass_oop || message == NULL, "either oop or message but not both"); | |
106 address entry = __ pc(); | |
107 if (pass_oop) { | |
108 // object is at TOS | |
304 | 109 __ pop(c_rarg2); |
0 | 110 } |
111 // expression stack must be empty before entering the VM if an | |
112 // exception happened | |
113 __ empty_expression_stack(); | |
114 // setup parameters | |
115 __ lea(c_rarg1, ExternalAddress((address)name)); | |
116 if (pass_oop) { | |
117 __ call_VM(rax, CAST_FROM_FN_PTR(address, | |
118 InterpreterRuntime:: | |
119 create_klass_exception), | |
120 c_rarg1, c_rarg2); | |
121 } else { | |
122 // kind of lame ExternalAddress can't take NULL because | |
123 // external_word_Relocation will assert. | |
124 if (message != NULL) { | |
125 __ lea(c_rarg2, ExternalAddress((address)message)); | |
126 } else { | |
127 __ movptr(c_rarg2, NULL_WORD); | |
128 } | |
129 __ call_VM(rax, | |
130 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), | |
131 c_rarg1, c_rarg2); | |
132 } | |
133 // throw exception | |
134 __ jump(ExternalAddress(Interpreter::throw_exception_entry())); | |
135 return entry; | |
136 } | |
137 | |
138 | |
139 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) { | |
140 address entry = __ pc(); | |
141 // NULL last_sp until next java call | |
304 | 142 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); |
0 | 143 __ dispatch_next(state); |
144 return entry; | |
145 } | |
146 | |
147 | |
148 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, | |
149 int step) { | |
150 | |
151 // amd64 doesn't need to do anything special about compiled returns | |
152 // to the interpreter so the code that exists on x86 to place a sentinel | |
153 // here and the specialized cleanup code is not needed here. | |
154 | |
155 address entry = __ pc(); | |
156 | |
157 // Restore stack bottom in case i2c adjusted stack | |
304 | 158 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); |
0 | 159 // and NULL it as marker that esp is now tos until next java call |
304 | 160 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); |
0 | 161 |
162 __ restore_bcp(); | |
163 __ restore_locals(); | |
304 | 164 |
0 | 165 __ get_cache_and_index_at_bcp(rbx, rcx, 1); |
166 __ movl(rbx, Address(rbx, rcx, | |
167 Address::times_8, | |
168 in_bytes(constantPoolCacheOopDesc::base_offset()) + | |
169 3 * wordSize)); | |
170 __ andl(rbx, 0xFF); | |
171 if (TaggedStackInterpreter) __ shll(rbx, 1); // 2 slots per parameter. | |
304 | 172 __ lea(rsp, Address(rsp, rbx, Address::times_8)); |
0 | 173 __ dispatch_next(state, step); |
174 return entry; | |
175 } | |
176 | |
177 | |
178 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, | |
179 int step) { | |
180 address entry = __ pc(); | |
181 // NULL last_sp until next java call | |
304 | 182 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); |
0 | 183 __ restore_bcp(); |
184 __ restore_locals(); | |
185 // handle exceptions | |
186 { | |
187 Label L; | |
304 | 188 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD); |
0 | 189 __ jcc(Assembler::zero, L); |
190 __ call_VM(noreg, | |
191 CAST_FROM_FN_PTR(address, | |
192 InterpreterRuntime::throw_pending_exception)); | |
193 __ should_not_reach_here(); | |
194 __ bind(L); | |
195 } | |
196 __ dispatch_next(state, step); | |
197 return entry; | |
198 } | |
199 | |
200 int AbstractInterpreter::BasicType_as_index(BasicType type) { | |
201 int i = 0; | |
202 switch (type) { | |
203 case T_BOOLEAN: i = 0; break; | |
204 case T_CHAR : i = 1; break; | |
205 case T_BYTE : i = 2; break; | |
206 case T_SHORT : i = 3; break; | |
207 case T_INT : i = 4; break; | |
208 case T_LONG : i = 5; break; | |
209 case T_VOID : i = 6; break; | |
210 case T_FLOAT : i = 7; break; | |
211 case T_DOUBLE : i = 8; break; | |
212 case T_OBJECT : i = 9; break; | |
213 case T_ARRAY : i = 9; break; | |
214 default : ShouldNotReachHere(); | |
215 } | |
216 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, | |
217 "index out of bounds"); | |
218 return i; | |
219 } | |
220 | |
221 | |
222 address TemplateInterpreterGenerator::generate_result_handler_for( | |
223 BasicType type) { | |
224 address entry = __ pc(); | |
225 switch (type) { | |
226 case T_BOOLEAN: __ c2bool(rax); break; | |
227 case T_CHAR : __ movzwl(rax, rax); break; | |
228 case T_BYTE : __ sign_extend_byte(rax); break; | |
229 case T_SHORT : __ sign_extend_short(rax); break; | |
230 case T_INT : /* nothing to do */ break; | |
231 case T_LONG : /* nothing to do */ break; | |
232 case T_VOID : /* nothing to do */ break; | |
233 case T_FLOAT : /* nothing to do */ break; | |
234 case T_DOUBLE : /* nothing to do */ break; | |
235 case T_OBJECT : | |
236 // retrieve result from frame | |
304 | 237 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize)); |
0 | 238 // and verify it |
239 __ verify_oop(rax); | |
240 break; | |
241 default : ShouldNotReachHere(); | |
242 } | |
243 __ ret(0); // return from result handler | |
244 return entry; | |
245 } | |
246 | |
247 address TemplateInterpreterGenerator::generate_safept_entry_for( | |
248 TosState state, | |
249 address runtime_entry) { | |
250 address entry = __ pc(); | |
251 __ push(state); | |
252 __ call_VM(noreg, runtime_entry); | |
253 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); | |
254 return entry; | |
255 } | |
256 | |
257 | |
258 | |
259 // Helpers for commoning out cases in the various type of method entries. | |
260 // | |
261 | |
262 | |
263 // increment invocation count & check for overflow | |
264 // | |
265 // Note: checking for negative value instead of overflow | |
266 // so we have a 'sticky' overflow test | |
267 // | |
268 // rbx: method | |
269 // ecx: invocation counter | |
270 // | |
271 void InterpreterGenerator::generate_counter_incr( | |
272 Label* overflow, | |
273 Label* profile_method, | |
274 Label* profile_method_continue) { | |
275 | |
276 const Address invocation_counter(rbx, | |
277 methodOopDesc::invocation_counter_offset() + | |
278 InvocationCounter::counter_offset()); | |
279 const Address backedge_counter(rbx, | |
280 methodOopDesc::backedge_counter_offset() + | |
281 InvocationCounter::counter_offset()); | |
282 | |
283 if (ProfileInterpreter) { // %%% Merge this into methodDataOop | |
284 __ incrementl(Address(rbx, | |
285 methodOopDesc::interpreter_invocation_counter_offset())); | |
286 } | |
287 // Update standard invocation counters | |
288 __ movl(rax, backedge_counter); // load backedge counter | |
289 | |
290 __ incrementl(rcx, InvocationCounter::count_increment); | |
291 __ andl(rax, InvocationCounter::count_mask_value); // mask out the | |
292 // status bits | |
293 | |
294 __ movl(invocation_counter, rcx); // save invocation count | |
295 __ addl(rcx, rax); // add both counters | |
296 | |
297 // profile_method is non-null only for interpreted method so | |
298 // profile_method != NULL == !native_call | |
299 | |
300 if (ProfileInterpreter && profile_method != NULL) { | |
301 // Test to see if we should create a method data oop | |
302 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit)); | |
303 __ jcc(Assembler::less, *profile_method_continue); | |
304 | |
305 // if no method data exists, go to profile_method | |
306 __ test_method_data_pointer(rax, *profile_method); | |
307 } | |
308 | |
309 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit)); | |
310 __ jcc(Assembler::aboveEqual, *overflow); | |
311 } | |
312 | |
313 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) { | |
314 | |
315 // Asm interpreter on entry | |
316 // r14 - locals | |
317 // r13 - bcp | |
318 // rbx - method | |
319 // edx - cpool --- DOES NOT APPEAR TO BE TRUE | |
320 // rbp - interpreter frame | |
321 | |
322 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] | |
323 // Everything as it was on entry | |
324 // rdx is not restored. Doesn't appear to really be set. | |
325 | |
326 const Address size_of_parameters(rbx, | |
327 methodOopDesc::size_of_parameters_offset()); | |
328 | |
329 // InterpreterRuntime::frequency_counter_overflow takes two | |
330 // arguments, the first (thread) is passed by call_VM, the second | |
331 // indicates if the counter overflow occurs at a backwards branch | |
332 // (NULL bcp). We pass zero for it. The call returns the address | |
333 // of the verified entry point for the method or NULL if the | |
334 // compilation did not complete (either went background or bailed | |
335 // out). | |
336 __ movl(c_rarg1, 0); | |
337 __ call_VM(noreg, | |
338 CAST_FROM_FN_PTR(address, | |
339 InterpreterRuntime::frequency_counter_overflow), | |
340 c_rarg1); | |
341 | |
304 | 342 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop |
0 | 343 // Preserve invariant that r13/r14 contain bcp/locals of sender frame |
344 // and jump to the interpreted entry. | |
345 __ jmp(*do_continue, relocInfo::none); | |
346 } | |
347 | |
348 // See if we've got enough room on the stack for locals plus overhead. | |
349 // The expression stack grows down incrementally, so the normal guard | |
350 // page mechanism will work for that. | |
351 // | |
352 // NOTE: Since the additional locals are also always pushed (wasn't | |
353 // obvious in generate_method_entry) so the guard should work for them | |
354 // too. | |
355 // | |
356 // Args: | |
357 // rdx: number of additional locals this frame needs (what we must check) | |
358 // rbx: methodOop | |
359 // | |
360 // Kills: | |
361 // rax | |
362 void InterpreterGenerator::generate_stack_overflow_check(void) { | |
363 | |
364 // monitor entry size: see picture of stack set | |
365 // (generate_method_entry) and frame_amd64.hpp | |
366 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; | |
367 | |
368 // total overhead size: entry_size + (saved rbp through expr stack | |
369 // bottom). be sure to change this if you add/subtract anything | |
370 // to/from the overhead area | |
371 const int overhead_size = | |
372 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size; | |
373 | |
374 const int page_size = os::vm_page_size(); | |
375 | |
376 Label after_frame_check; | |
377 | |
378 // see if the frame is greater than one page in size. If so, | |
379 // then we need to verify there is enough stack space remaining | |
380 // for the additional locals. | |
381 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize()); | |
382 __ jcc(Assembler::belowEqual, after_frame_check); | |
383 | |
384 // compute rsp as if this were going to be the last frame on | |
385 // the stack before the red zone | |
386 | |
387 const Address stack_base(r15_thread, Thread::stack_base_offset()); | |
388 const Address stack_size(r15_thread, Thread::stack_size_offset()); | |
389 | |
390 // locals + overhead, in bytes | |
304 | 391 __ mov(rax, rdx); |
392 __ shlptr(rax, Interpreter::logStackElementSize()); // 2 slots per parameter. | |
393 __ addptr(rax, overhead_size); | |
0 | 394 |
395 #ifdef ASSERT | |
396 Label stack_base_okay, stack_size_okay; | |
397 // verify that thread stack base is non-zero | |
304 | 398 __ cmpptr(stack_base, (int32_t)NULL_WORD); |
0 | 399 __ jcc(Assembler::notEqual, stack_base_okay); |
400 __ stop("stack base is zero"); | |
401 __ bind(stack_base_okay); | |
402 // verify that thread stack size is non-zero | |
304 | 403 __ cmpptr(stack_size, 0); |
0 | 404 __ jcc(Assembler::notEqual, stack_size_okay); |
405 __ stop("stack size is zero"); | |
406 __ bind(stack_size_okay); | |
407 #endif | |
408 | |
409 // Add stack base to locals and subtract stack size | |
304 | 410 __ addptr(rax, stack_base); |
411 __ subptr(rax, stack_size); | |
0 | 412 |
413 // add in the red and yellow zone sizes | |
304 | 414 __ addptr(rax, (StackRedPages + StackYellowPages) * page_size); |
0 | 415 |
416 // check against the current stack bottom | |
304 | 417 __ cmpptr(rsp, rax); |
0 | 418 __ jcc(Assembler::above, after_frame_check); |
419 | |
304 | 420 __ pop(rax); // get return address |
0 | 421 __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry())); |
422 | |
423 // all done with frame size check | |
424 __ bind(after_frame_check); | |
425 } | |
426 | |
427 // Allocate monitor and lock method (asm interpreter) | |
428 // | |
429 // Args: | |
430 // rbx: methodOop | |
431 // r14: locals | |
432 // | |
433 // Kills: | |
434 // rax | |
435 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs) | |
436 // rscratch1, rscratch2 (scratch regs) | |
437 void InterpreterGenerator::lock_method(void) { | |
438 // synchronize method | |
439 const Address access_flags(rbx, methodOopDesc::access_flags_offset()); | |
440 const Address monitor_block_top( | |
441 rbp, | |
442 frame::interpreter_frame_monitor_block_top_offset * wordSize); | |
443 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; | |
444 | |
445 #ifdef ASSERT | |
446 { | |
447 Label L; | |
448 __ movl(rax, access_flags); | |
449 __ testl(rax, JVM_ACC_SYNCHRONIZED); | |
450 __ jcc(Assembler::notZero, L); | |
451 __ stop("method doesn't need synchronization"); | |
452 __ bind(L); | |
453 } | |
454 #endif // ASSERT | |
455 | |
456 // get synchronization object | |
457 { | |
458 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + | |
459 Klass::java_mirror_offset_in_bytes(); | |
460 Label done; | |
461 __ movl(rax, access_flags); | |
462 __ testl(rax, JVM_ACC_STATIC); | |
463 // get receiver (assume this is frequent case) | |
304 | 464 __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0))); |
0 | 465 __ jcc(Assembler::zero, done); |
304 | 466 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset())); |
467 __ movptr(rax, Address(rax, | |
468 constantPoolOopDesc::pool_holder_offset_in_bytes())); | |
469 __ movptr(rax, Address(rax, mirror_offset)); | |
0 | 470 |
471 #ifdef ASSERT | |
472 { | |
473 Label L; | |
304 | 474 __ testptr(rax, rax); |
0 | 475 __ jcc(Assembler::notZero, L); |
476 __ stop("synchronization object is NULL"); | |
477 __ bind(L); | |
478 } | |
479 #endif // ASSERT | |
480 | |
481 __ bind(done); | |
482 } | |
483 | |
484 // add space for monitor & lock | |
304 | 485 __ subptr(rsp, entry_size); // add space for a monitor entry |
486 __ movptr(monitor_block_top, rsp); // set new monitor block top | |
0 | 487 // store object |
304 | 488 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); |
489 __ movptr(c_rarg1, rsp); // object address | |
0 | 490 __ lock_object(c_rarg1); |
491 } | |
492 | |
493 // Generate a fixed interpreter frame. This is identical setup for | |
494 // interpreted methods and for native methods hence the shared code. | |
495 // | |
496 // Args: | |
497 // rax: return address | |
498 // rbx: methodOop | |
499 // r14: pointer to locals | |
500 // r13: sender sp | |
501 // rdx: cp cache | |
502 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { | |
503 // initialize fixed part of activation frame | |
304 | 504 __ push(rax); // save return address |
0 | 505 __ enter(); // save old & set new rbp |
304 | 506 __ push(r13); // set sender sp |
507 __ push((int)NULL_WORD); // leave last_sp as null | |
508 __ movptr(r13, Address(rbx, methodOopDesc::const_offset())); // get constMethodOop | |
509 __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase | |
510 __ push(rbx); // save methodOop | |
0 | 511 if (ProfileInterpreter) { |
512 Label method_data_continue; | |
304 | 513 __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset()))); |
514 __ testptr(rdx, rdx); | |
0 | 515 __ jcc(Assembler::zero, method_data_continue); |
304 | 516 __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset())); |
0 | 517 __ bind(method_data_continue); |
304 | 518 __ push(rdx); // set the mdp (method data pointer) |
0 | 519 } else { |
304 | 520 __ push(0); |
0 | 521 } |
522 | |
304 | 523 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset())); |
524 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes())); | |
525 __ push(rdx); // set constant pool cache | |
526 __ push(r14); // set locals pointer | |
0 | 527 if (native_call) { |
304 | 528 __ push(0); // no bcp |
0 | 529 } else { |
304 | 530 __ push(r13); // set bcp |
0 | 531 } |
304 | 532 __ push(0); // reserve word for pointer to expression stack bottom |
533 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom | |
0 | 534 } |
535 | |
536 // End of helpers | |
537 | |
304 | 538 // Various method entries |
539 //------------------------------------------------------------------------------------------------------------------------ | |
540 // | |
541 // | |
542 | |
543 // Call an accessor method (assuming it is resolved, otherwise drop | |
544 // into vanilla (slow path) entry | |
545 address InterpreterGenerator::generate_accessor_entry(void) { | |
546 // rbx: methodOop | |
547 | |
548 // r13: senderSP must preserver for slow path, set SP to it on fast path | |
549 | |
550 address entry_point = __ pc(); | |
551 Label xreturn_path; | |
552 | |
553 // do fastpath for resolved accessor methods | |
554 if (UseFastAccessorMethods) { | |
555 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites | |
556 // thereof; parameter size = 1 | |
557 // Note: We can only use this code if the getfield has been resolved | |
558 // and if we don't have a null-pointer exception => check for | |
559 // these conditions first and use slow path if necessary. | |
560 Label slow_path; | |
561 // If we need a safepoint check, generate full interpreter entry. | |
562 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), | |
563 SafepointSynchronize::_not_synchronized); | |
564 | |
565 __ jcc(Assembler::notEqual, slow_path); | |
566 // rbx: method | |
567 __ movptr(rax, Address(rsp, wordSize)); | |
568 | |
569 // check if local 0 != NULL and read field | |
570 __ testptr(rax, rax); | |
571 __ jcc(Assembler::zero, slow_path); | |
572 | |
573 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset())); | |
574 // read first instruction word and extract bytecode @ 1 and index @ 2 | |
575 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset())); | |
576 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset())); | |
577 // Shift codes right to get the index on the right. | |
578 // The bytecode fetched looks like <index><0xb4><0x2a> | |
579 __ shrl(rdx, 2 * BitsPerByte); | |
580 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); | |
581 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes())); | |
582 | |
583 // rax: local 0 | |
584 // rbx: method | |
585 // rdx: constant pool cache index | |
586 // rdi: constant pool cache | |
587 | |
588 // check if getfield has been resolved and read constant pool cache entry | |
589 // check the validity of the cache entry by testing whether _indices field | |
590 // contains Bytecode::_getfield in b1 byte. | |
591 assert(in_words(ConstantPoolCacheEntry::size()) == 4, | |
592 "adjust shift below"); | |
593 __ movl(rcx, | |
594 Address(rdi, | |
595 rdx, | |
596 Address::times_8, | |
597 constantPoolCacheOopDesc::base_offset() + | |
598 ConstantPoolCacheEntry::indices_offset())); | |
599 __ shrl(rcx, 2 * BitsPerByte); | |
600 __ andl(rcx, 0xFF); | |
601 __ cmpl(rcx, Bytecodes::_getfield); | |
602 __ jcc(Assembler::notEqual, slow_path); | |
603 | |
604 // Note: constant pool entry is not valid before bytecode is resolved | |
605 __ movptr(rcx, | |
606 Address(rdi, | |
607 rdx, | |
608 Address::times_8, | |
609 constantPoolCacheOopDesc::base_offset() + | |
610 ConstantPoolCacheEntry::f2_offset())); | |
611 // edx: flags | |
612 __ movl(rdx, | |
613 Address(rdi, | |
614 rdx, | |
615 Address::times_8, | |
616 constantPoolCacheOopDesc::base_offset() + | |
617 ConstantPoolCacheEntry::flags_offset())); | |
618 | |
619 Label notObj, notInt, notByte, notShort; | |
620 const Address field_address(rax, rcx, Address::times_1); | |
621 | |
622 // Need to differentiate between igetfield, agetfield, bgetfield etc. | |
623 // because they are different sizes. | |
624 // Use the type from the constant pool cache | |
625 __ shrl(rdx, ConstantPoolCacheEntry::tosBits); | |
626 // Make sure we don't need to mask edx for tosBits after the above shift | |
627 ConstantPoolCacheEntry::verify_tosBits(); | |
628 | |
629 __ cmpl(rdx, atos); | |
630 __ jcc(Assembler::notEqual, notObj); | |
631 // atos | |
632 __ load_heap_oop(rax, field_address); | |
633 __ jmp(xreturn_path); | |
634 | |
635 __ bind(notObj); | |
636 __ cmpl(rdx, itos); | |
637 __ jcc(Assembler::notEqual, notInt); | |
638 // itos | |
639 __ movl(rax, field_address); | |
640 __ jmp(xreturn_path); | |
641 | |
642 __ bind(notInt); | |
643 __ cmpl(rdx, btos); | |
644 __ jcc(Assembler::notEqual, notByte); | |
645 // btos | |
646 __ load_signed_byte(rax, field_address); | |
647 __ jmp(xreturn_path); | |
648 | |
649 __ bind(notByte); | |
650 __ cmpl(rdx, stos); | |
651 __ jcc(Assembler::notEqual, notShort); | |
652 // stos | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
653 __ load_signed_short(rax, field_address); |
304 | 654 __ jmp(xreturn_path); |
655 | |
656 __ bind(notShort); | |
657 #ifdef ASSERT | |
658 Label okay; | |
659 __ cmpl(rdx, ctos); | |
660 __ jcc(Assembler::equal, okay); | |
661 __ stop("what type is this?"); | |
662 __ bind(okay); | |
663 #endif | |
664 // ctos | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
665 __ load_unsigned_short(rax, field_address); |
304 | 666 |
667 __ bind(xreturn_path); | |
668 | |
669 // _ireturn/_areturn | |
670 __ pop(rdi); | |
671 __ mov(rsp, r13); | |
672 __ jmp(rdi); | |
673 __ ret(0); | |
674 | |
675 // generate a vanilla interpreter entry as the slow path | |
676 __ bind(slow_path); | |
677 (void) generate_normal_entry(false); | |
678 } else { | |
679 (void) generate_normal_entry(false); | |
680 } | |
681 | |
682 return entry_point; | |
683 } | |
684 | |
0 | 685 // Interpreter stub for calling a native method. (asm interpreter) |
686 // This sets up a somewhat different looking stack for calling the | |
687 // native method than the typical interpreter frame setup. | |
688 address InterpreterGenerator::generate_native_entry(bool synchronized) { | |
689 // determine code generation flags | |
690 bool inc_counter = UseCompiler || CountCompiledCalls; | |
691 | |
692 // rbx: methodOop | |
693 // r13: sender sp | |
694 | |
695 address entry_point = __ pc(); | |
696 | |
697 const Address size_of_parameters(rbx, methodOopDesc:: | |
698 size_of_parameters_offset()); | |
699 const Address invocation_counter(rbx, methodOopDesc:: | |
700 invocation_counter_offset() + | |
701 InvocationCounter::counter_offset()); | |
702 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); | |
703 | |
704 // get parameter size (always needed) | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
705 __ load_unsigned_short(rcx, size_of_parameters); |
0 | 706 |
707 // native calls don't need the stack size check since they have no | |
708 // expression stack and the arguments are already on the stack and | |
709 // we only add a handful of words to the stack | |
710 | |
711 // rbx: methodOop | |
712 // rcx: size of parameters | |
713 // r13: sender sp | |
304 | 714 __ pop(rax); // get return address |
0 | 715 |
716 // for natives the size of locals is zero | |
717 | |
718 // compute beginning of parameters (r14) | |
719 if (TaggedStackInterpreter) __ shll(rcx, 1); // 2 slots per parameter. | |
304 | 720 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize)); |
0 | 721 |
722 // add 2 zero-initialized slots for native calls | |
723 // initialize result_handler slot | |
304 | 724 __ push((int) NULL_WORD); |
0 | 725 // slot for oop temp |
726 // (static native method holder mirror/jni oop result) | |
304 | 727 __ push((int) NULL_WORD); |
0 | 728 |
729 if (inc_counter) { | |
730 __ movl(rcx, invocation_counter); // (pre-)fetch invocation count | |
731 } | |
732 | |
733 // initialize fixed part of activation frame | |
734 generate_fixed_frame(true); | |
735 | |
736 // make sure method is native & not abstract | |
737 #ifdef ASSERT | |
738 __ movl(rax, access_flags); | |
739 { | |
740 Label L; | |
741 __ testl(rax, JVM_ACC_NATIVE); | |
742 __ jcc(Assembler::notZero, L); | |
743 __ stop("tried to execute non-native method as native"); | |
744 __ bind(L); | |
745 } | |
746 { | |
747 Label L; | |
748 __ testl(rax, JVM_ACC_ABSTRACT); | |
749 __ jcc(Assembler::zero, L); | |
750 __ stop("tried to execute abstract method in interpreter"); | |
751 __ bind(L); | |
752 } | |
753 #endif | |
754 | |
755 // Since at this point in the method invocation the exception handler | |
756 // would try to exit the monitor of synchronized methods which hasn't | |
757 // been entered yet, we set the thread local variable | |
758 // _do_not_unlock_if_synchronized to true. The remove_activation will | |
759 // check this flag. | |
760 | |
761 const Address do_not_unlock_if_synchronized(r15_thread, | |
762 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); | |
763 __ movbool(do_not_unlock_if_synchronized, true); | |
764 | |
765 // increment invocation count & check for overflow | |
766 Label invocation_counter_overflow; | |
767 if (inc_counter) { | |
768 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); | |
769 } | |
770 | |
771 Label continue_after_compile; | |
772 __ bind(continue_after_compile); | |
773 | |
774 bang_stack_shadow_pages(true); | |
775 | |
776 // reset the _do_not_unlock_if_synchronized flag | |
777 __ movbool(do_not_unlock_if_synchronized, false); | |
778 | |
779 // check for synchronized methods | |
780 // Must happen AFTER invocation_counter check and stack overflow check, | |
781 // so method is not locked if overflows. | |
782 if (synchronized) { | |
783 lock_method(); | |
784 } else { | |
785 // no synchronization necessary | |
786 #ifdef ASSERT | |
787 { | |
788 Label L; | |
789 __ movl(rax, access_flags); | |
790 __ testl(rax, JVM_ACC_SYNCHRONIZED); | |
791 __ jcc(Assembler::zero, L); | |
792 __ stop("method needs synchronization"); | |
793 __ bind(L); | |
794 } | |
795 #endif | |
796 } | |
797 | |
798 // start execution | |
799 #ifdef ASSERT | |
800 { | |
801 Label L; | |
802 const Address monitor_block_top(rbp, | |
803 frame::interpreter_frame_monitor_block_top_offset * wordSize); | |
304 | 804 __ movptr(rax, monitor_block_top); |
805 __ cmpptr(rax, rsp); | |
0 | 806 __ jcc(Assembler::equal, L); |
807 __ stop("broken stack frame setup in interpreter"); | |
808 __ bind(L); | |
809 } | |
810 #endif | |
811 | |
812 // jvmti support | |
813 __ notify_method_entry(); | |
814 | |
815 // work registers | |
816 const Register method = rbx; | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
817 const Register t = r11; |
0 | 818 |
819 // allocate space for parameters | |
820 __ get_method(method); | |
821 __ verify_oop(method); | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
822 __ load_unsigned_short(t, |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
823 Address(method, |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
824 methodOopDesc::size_of_parameters_offset())); |
0 | 825 __ shll(t, Interpreter::logStackElementSize()); |
826 | |
304 | 827 __ subptr(rsp, t); |
828 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows | |
605 | 829 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI) |
0 | 830 |
831 // get signature handler | |
832 { | |
833 Label L; | |
304 | 834 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset())); |
835 __ testptr(t, t); | |
0 | 836 __ jcc(Assembler::notZero, L); |
837 __ call_VM(noreg, | |
838 CAST_FROM_FN_PTR(address, | |
839 InterpreterRuntime::prepare_native_call), | |
840 method); | |
841 __ get_method(method); | |
304 | 842 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset())); |
0 | 843 __ bind(L); |
844 } | |
845 | |
846 // call signature handler | |
847 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14, | |
848 "adjust this code"); | |
849 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp, | |
850 "adjust this code"); | |
851 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1, | |
852 "adjust this code"); | |
853 | |
854 // The generated handlers do not touch RBX (the method oop). | |
855 // However, large signatures cannot be cached and are generated | |
856 // each time here. The slow-path generator can do a GC on return, | |
857 // so we must reload it after the call. | |
858 __ call(t); | |
859 __ get_method(method); // slow path can do a GC, reload RBX | |
860 | |
861 | |
862 // result handler is in rax | |
863 // set result handler | |
304 | 864 __ movptr(Address(rbp, |
865 (frame::interpreter_frame_result_handler_offset) * wordSize), | |
866 rax); | |
0 | 867 |
868 // pass mirror handle if static call | |
869 { | |
870 Label L; | |
871 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + | |
872 Klass::java_mirror_offset_in_bytes(); | |
873 __ movl(t, Address(method, methodOopDesc::access_flags_offset())); | |
874 __ testl(t, JVM_ACC_STATIC); | |
875 __ jcc(Assembler::zero, L); | |
876 // get mirror | |
304 | 877 __ movptr(t, Address(method, methodOopDesc::constants_offset())); |
878 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes())); | |
879 __ movptr(t, Address(t, mirror_offset)); | |
0 | 880 // copy mirror into activation frame |
304 | 881 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), |
0 | 882 t); |
883 // pass handle to mirror | |
304 | 884 __ lea(c_rarg1, |
885 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize)); | |
0 | 886 __ bind(L); |
887 } | |
888 | |
889 // get native function entry point | |
890 { | |
891 Label L; | |
304 | 892 __ movptr(rax, Address(method, methodOopDesc::native_function_offset())); |
0 | 893 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry()); |
894 __ movptr(rscratch2, unsatisfied.addr()); | |
304 | 895 __ cmpptr(rax, rscratch2); |
0 | 896 __ jcc(Assembler::notEqual, L); |
897 __ call_VM(noreg, | |
898 CAST_FROM_FN_PTR(address, | |
899 InterpreterRuntime::prepare_native_call), | |
900 method); | |
901 __ get_method(method); | |
902 __ verify_oop(method); | |
304 | 903 __ movptr(rax, Address(method, methodOopDesc::native_function_offset())); |
0 | 904 __ bind(L); |
905 } | |
906 | |
907 // pass JNIEnv | |
304 | 908 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset())); |
0 | 909 |
910 // It is enough that the pc() points into the right code | |
911 // segment. It does not have to be the correct return pc. | |
912 __ set_last_Java_frame(rsp, rbp, (address) __ pc()); | |
913 | |
914 // change thread state | |
915 #ifdef ASSERT | |
916 { | |
917 Label L; | |
918 __ movl(t, Address(r15_thread, JavaThread::thread_state_offset())); | |
919 __ cmpl(t, _thread_in_Java); | |
920 __ jcc(Assembler::equal, L); | |
921 __ stop("Wrong thread state in native stub"); | |
922 __ bind(L); | |
923 } | |
924 #endif | |
925 | |
926 // Change state to native | |
927 | |
928 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), | |
929 _thread_in_native); | |
930 | |
931 // Call the native method. | |
932 __ call(rax); | |
933 // result potentially in rax or xmm0 | |
934 | |
935 // Depending on runtime options, either restore the MXCSR | |
936 // register after returning from the JNI Call or verify that | |
937 // it wasn't changed during -Xcheck:jni. | |
938 if (RestoreMXCSROnJNICalls) { | |
304 | 939 __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std())); |
0 | 940 } |
941 else if (CheckJNICalls) { | |
304 | 942 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry()))); |
0 | 943 } |
944 | |
945 // NOTE: The order of these pushes is known to frame::interpreter_frame_result | |
946 // in order to extract the result of a method call. If the order of these | |
947 // pushes change or anything else is added to the stack then the code in | |
948 // interpreter_frame_result must also change. | |
949 | |
950 __ push(dtos); | |
951 __ push(ltos); | |
952 | |
953 // change thread state | |
954 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), | |
955 _thread_in_native_trans); | |
956 | |
957 if (os::is_MP()) { | |
958 if (UseMembar) { | |
959 // Force this write out before the read below | |
960 __ membar(Assembler::Membar_mask_bits( | |
961 Assembler::LoadLoad | Assembler::LoadStore | | |
962 Assembler::StoreLoad | Assembler::StoreStore)); | |
963 } else { | |
964 // Write serialization page so VM thread can do a pseudo remote membar. | |
965 // We use the current thread pointer to calculate a thread specific | |
966 // offset to write to within the page. This minimizes bus traffic | |
967 // due to cache line collision. | |
968 __ serialize_memory(r15_thread, rscratch2); | |
969 } | |
970 } | |
971 | |
972 // check for safepoint operation in progress and/or pending suspend requests | |
973 { | |
974 Label Continue; | |
975 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), | |
976 SafepointSynchronize::_not_synchronized); | |
977 | |
978 Label L; | |
979 __ jcc(Assembler::notEqual, L); | |
980 __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0); | |
981 __ jcc(Assembler::equal, Continue); | |
982 __ bind(L); | |
983 | |
984 // Don't use call_VM as it will see a possible pending exception | |
985 // and forward it and never return here preventing us from | |
986 // clearing _last_native_pc down below. Also can't use | |
987 // call_VM_leaf either as it will check to see if r13 & r14 are | |
988 // preserved and correspond to the bcp/locals pointers. So we do a | |
989 // runtime call by hand. | |
990 // | |
304 | 991 __ mov(c_rarg0, r15_thread); |
992 __ mov(r12, rsp); // remember sp | |
993 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows | |
994 __ andptr(rsp, -16); // align stack as required by ABI | |
0 | 995 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans))); |
304 | 996 __ mov(rsp, r12); // restore sp |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
997 __ reinit_heapbase(); |
0 | 998 __ bind(Continue); |
999 } | |
1000 | |
1001 // change thread state | |
1002 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java); | |
1003 | |
1004 // reset_last_Java_frame | |
1005 __ reset_last_Java_frame(true, true); | |
1006 | |
1007 // reset handle block | |
304 | 1008 __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset())); |
1009 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD); | |
0 | 1010 |
1011 // If result is an oop unbox and store it in frame where gc will see it | |
1012 // and result handler will pick it up | |
1013 | |
1014 { | |
1015 Label no_oop, store_result; | |
1016 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT))); | |
304 | 1017 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize)); |
0 | 1018 __ jcc(Assembler::notEqual, no_oop); |
1019 // retrieve result | |
1020 __ pop(ltos); | |
304 | 1021 __ testptr(rax, rax); |
0 | 1022 __ jcc(Assembler::zero, store_result); |
304 | 1023 __ movptr(rax, Address(rax, 0)); |
0 | 1024 __ bind(store_result); |
304 | 1025 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax); |
0 | 1026 // keep stack depth as expected by pushing oop which will eventually be discarde |
1027 __ push(ltos); | |
1028 __ bind(no_oop); | |
1029 } | |
1030 | |
1031 | |
1032 { | |
1033 Label no_reguard; | |
1034 __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()), | |
1035 JavaThread::stack_guard_yellow_disabled); | |
1036 __ jcc(Assembler::notEqual, no_reguard); | |
1037 | |
304 | 1038 __ pusha(); // XXX only save smashed registers |
1039 __ mov(r12, rsp); // remember sp | |
1040 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows | |
1041 __ andptr(rsp, -16); // align stack as required by ABI | |
0 | 1042 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); |
304 | 1043 __ mov(rsp, r12); // restore sp |
1044 __ popa(); // XXX only restore smashed registers | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1045 __ reinit_heapbase(); |
0 | 1046 |
1047 __ bind(no_reguard); | |
1048 } | |
1049 | |
1050 | |
1051 // The method register is junk from after the thread_in_native transition | |
1052 // until here. Also can't call_VM until the bcp has been | |
1053 // restored. Need bcp for throwing exception below so get it now. | |
1054 __ get_method(method); | |
1055 __ verify_oop(method); | |
1056 | |
1057 // restore r13 to have legal interpreter frame, i.e., bci == 0 <=> | |
1058 // r13 == code_base() | |
304 | 1059 __ movptr(r13, Address(method, methodOopDesc::const_offset())); // get constMethodOop |
1060 __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase | |
0 | 1061 // handle exceptions (exception handling will handle unlocking!) |
1062 { | |
1063 Label L; | |
304 | 1064 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD); |
0 | 1065 __ jcc(Assembler::zero, L); |
1066 // Note: At some point we may want to unify this with the code | |
1067 // used in call_VM_base(); i.e., we should use the | |
1068 // StubRoutines::forward_exception code. For now this doesn't work | |
1069 // here because the rsp is not correctly set at this point. | |
1070 __ MacroAssembler::call_VM(noreg, | |
1071 CAST_FROM_FN_PTR(address, | |
1072 InterpreterRuntime::throw_pending_exception)); | |
1073 __ should_not_reach_here(); | |
1074 __ bind(L); | |
1075 } | |
1076 | |
1077 // do unlocking if necessary | |
1078 { | |
1079 Label L; | |
1080 __ movl(t, Address(method, methodOopDesc::access_flags_offset())); | |
1081 __ testl(t, JVM_ACC_SYNCHRONIZED); | |
1082 __ jcc(Assembler::zero, L); | |
1083 // the code below should be shared with interpreter macro | |
1084 // assembler implementation | |
1085 { | |
1086 Label unlock; | |
1087 // BasicObjectLock will be first in list, since this is a | |
1088 // synchronized method. However, need to check that the object | |
1089 // has not been unlocked by an explicit monitorexit bytecode. | |
1090 const Address monitor(rbp, | |
1091 (intptr_t)(frame::interpreter_frame_initial_sp_offset * | |
1092 wordSize - sizeof(BasicObjectLock))); | |
1093 | |
1094 // monitor expect in c_rarg1 for slow unlock path | |
304 | 1095 __ lea(c_rarg1, monitor); // address of first monitor |
0 | 1096 |
304 | 1097 __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); |
1098 __ testptr(t, t); | |
0 | 1099 __ jcc(Assembler::notZero, unlock); |
1100 | |
1101 // Entry already unlocked, need to throw exception | |
1102 __ MacroAssembler::call_VM(noreg, | |
1103 CAST_FROM_FN_PTR(address, | |
1104 InterpreterRuntime::throw_illegal_monitor_state_exception)); | |
1105 __ should_not_reach_here(); | |
1106 | |
1107 __ bind(unlock); | |
1108 __ unlock_object(c_rarg1); | |
1109 } | |
1110 __ bind(L); | |
1111 } | |
1112 | |
1113 // jvmti support | |
1114 // Note: This must happen _after_ handling/throwing any exceptions since | |
1115 // the exception handler code notifies the runtime of method exits | |
1116 // too. If this happens before, method entry/exit notifications are | |
1117 // not properly paired (was bug - gri 11/22/99). | |
1118 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI); | |
1119 | |
1120 // restore potential result in edx:eax, call result handler to | |
1121 // restore potential result in ST0 & handle result | |
1122 | |
1123 __ pop(ltos); | |
1124 __ pop(dtos); | |
1125 | |
304 | 1126 __ movptr(t, Address(rbp, |
1127 (frame::interpreter_frame_result_handler_offset) * wordSize)); | |
0 | 1128 __ call(t); |
1129 | |
1130 // remove activation | |
304 | 1131 __ movptr(t, Address(rbp, |
1132 frame::interpreter_frame_sender_sp_offset * | |
1133 wordSize)); // get sender sp | |
0 | 1134 __ leave(); // remove frame anchor |
304 | 1135 __ pop(rdi); // get return address |
1136 __ mov(rsp, t); // set sp to sender sp | |
0 | 1137 __ jmp(rdi); |
1138 | |
1139 if (inc_counter) { | |
1140 // Handle overflow of counter and compile method | |
1141 __ bind(invocation_counter_overflow); | |
1142 generate_counter_overflow(&continue_after_compile); | |
1143 } | |
1144 | |
1145 return entry_point; | |
1146 } | |
1147 | |
1148 // | |
1149 // Generic interpreted method entry to (asm) interpreter | |
1150 // | |
1151 address InterpreterGenerator::generate_normal_entry(bool synchronized) { | |
1152 // determine code generation flags | |
1153 bool inc_counter = UseCompiler || CountCompiledCalls; | |
1154 | |
1155 // ebx: methodOop | |
1156 // r13: sender sp | |
1157 address entry_point = __ pc(); | |
1158 | |
1159 const Address size_of_parameters(rbx, | |
1160 methodOopDesc::size_of_parameters_offset()); | |
1161 const Address size_of_locals(rbx, methodOopDesc::size_of_locals_offset()); | |
1162 const Address invocation_counter(rbx, | |
1163 methodOopDesc::invocation_counter_offset() + | |
1164 InvocationCounter::counter_offset()); | |
1165 const Address access_flags(rbx, methodOopDesc::access_flags_offset()); | |
1166 | |
1167 // get parameter size (always needed) | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
1168 __ load_unsigned_short(rcx, size_of_parameters); |
0 | 1169 |
1170 // rbx: methodOop | |
1171 // rcx: size of parameters | |
1172 // r13: sender_sp (could differ from sp+wordSize if we were called via c2i ) | |
1173 | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
1174 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words |
0 | 1175 __ subl(rdx, rcx); // rdx = no. of additional locals |
1176 | |
1177 // YYY | |
1178 // __ incrementl(rdx); | |
1179 // __ andl(rdx, -2); | |
1180 | |
1181 // see if we've got enough room on the stack for locals plus overhead. | |
1182 generate_stack_overflow_check(); | |
1183 | |
1184 // get return address | |
304 | 1185 __ pop(rax); |
0 | 1186 |
1187 // compute beginning of parameters (r14) | |
1188 if (TaggedStackInterpreter) __ shll(rcx, 1); // 2 slots per parameter. | |
304 | 1189 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize)); |
0 | 1190 |
1191 // rdx - # of additional locals | |
1192 // allocate space for locals | |
1193 // explicitly initialize locals | |
1194 { | |
1195 Label exit, loop; | |
1196 __ testl(rdx, rdx); | |
1197 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0 | |
1198 __ bind(loop); | |
304 | 1199 if (TaggedStackInterpreter) __ push((int) NULL_WORD); // push tag |
1200 __ push((int) NULL_WORD); // initialize local variables | |
0 | 1201 __ decrementl(rdx); // until everything initialized |
1202 __ jcc(Assembler::greater, loop); | |
1203 __ bind(exit); | |
1204 } | |
1205 | |
1206 // (pre-)fetch invocation count | |
1207 if (inc_counter) { | |
1208 __ movl(rcx, invocation_counter); | |
1209 } | |
1210 // initialize fixed part of activation frame | |
1211 generate_fixed_frame(false); | |
1212 | |
1213 // make sure method is not native & not abstract | |
1214 #ifdef ASSERT | |
1215 __ movl(rax, access_flags); | |
1216 { | |
1217 Label L; | |
1218 __ testl(rax, JVM_ACC_NATIVE); | |
1219 __ jcc(Assembler::zero, L); | |
1220 __ stop("tried to execute native method as non-native"); | |
1221 __ bind(L); | |
1222 } | |
1223 { | |
1224 Label L; | |
1225 __ testl(rax, JVM_ACC_ABSTRACT); | |
1226 __ jcc(Assembler::zero, L); | |
1227 __ stop("tried to execute abstract method in interpreter"); | |
1228 __ bind(L); | |
1229 } | |
1230 #endif | |
1231 | |
1232 // Since at this point in the method invocation the exception | |
1233 // handler would try to exit the monitor of synchronized methods | |
1234 // which hasn't been entered yet, we set the thread local variable | |
1235 // _do_not_unlock_if_synchronized to true. The remove_activation | |
1236 // will check this flag. | |
1237 | |
1238 const Address do_not_unlock_if_synchronized(r15_thread, | |
1239 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); | |
1240 __ movbool(do_not_unlock_if_synchronized, true); | |
1241 | |
1242 // increment invocation count & check for overflow | |
1243 Label invocation_counter_overflow; | |
1244 Label profile_method; | |
1245 Label profile_method_continue; | |
1246 if (inc_counter) { | |
1247 generate_counter_incr(&invocation_counter_overflow, | |
1248 &profile_method, | |
1249 &profile_method_continue); | |
1250 if (ProfileInterpreter) { | |
1251 __ bind(profile_method_continue); | |
1252 } | |
1253 } | |
1254 | |
1255 Label continue_after_compile; | |
1256 __ bind(continue_after_compile); | |
1257 | |
1258 // check for synchronized interpreted methods | |
1259 bang_stack_shadow_pages(false); | |
1260 | |
1261 // reset the _do_not_unlock_if_synchronized flag | |
1262 __ movbool(do_not_unlock_if_synchronized, false); | |
1263 | |
1264 // check for synchronized methods | |
1265 // Must happen AFTER invocation_counter check and stack overflow check, | |
1266 // so method is not locked if overflows. | |
1267 if (synchronized) { | |
1268 // Allocate monitor and lock method | |
1269 lock_method(); | |
1270 } else { | |
1271 // no synchronization necessary | |
1272 #ifdef ASSERT | |
1273 { | |
1274 Label L; | |
1275 __ movl(rax, access_flags); | |
1276 __ testl(rax, JVM_ACC_SYNCHRONIZED); | |
1277 __ jcc(Assembler::zero, L); | |
1278 __ stop("method needs synchronization"); | |
1279 __ bind(L); | |
1280 } | |
1281 #endif | |
1282 } | |
1283 | |
1284 // start execution | |
1285 #ifdef ASSERT | |
1286 { | |
1287 Label L; | |
1288 const Address monitor_block_top (rbp, | |
1289 frame::interpreter_frame_monitor_block_top_offset * wordSize); | |
304 | 1290 __ movptr(rax, monitor_block_top); |
1291 __ cmpptr(rax, rsp); | |
0 | 1292 __ jcc(Assembler::equal, L); |
1293 __ stop("broken stack frame setup in interpreter"); | |
1294 __ bind(L); | |
1295 } | |
1296 #endif | |
1297 | |
1298 // jvmti support | |
1299 __ notify_method_entry(); | |
1300 | |
1301 __ dispatch_next(vtos); | |
1302 | |
1303 // invocation counter overflow | |
1304 if (inc_counter) { | |
1305 if (ProfileInterpreter) { | |
1306 // We have decided to profile this method in the interpreter | |
1307 __ bind(profile_method); | |
1308 | |
1309 __ call_VM(noreg, | |
1310 CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), | |
1311 r13, true); | |
1312 | |
304 | 1313 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop |
1314 __ movptr(rax, Address(rbx, | |
1315 in_bytes(methodOopDesc::method_data_offset()))); | |
1316 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), | |
1317 rax); | |
0 | 1318 __ test_method_data_pointer(rax, profile_method_continue); |
304 | 1319 __ addptr(rax, in_bytes(methodDataOopDesc::data_offset())); |
1320 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), | |
0 | 1321 rax); |
1322 __ jmp(profile_method_continue); | |
1323 } | |
1324 // Handle overflow of counter and compile method | |
1325 __ bind(invocation_counter_overflow); | |
1326 generate_counter_overflow(&continue_after_compile); | |
1327 } | |
1328 | |
1329 return entry_point; | |
1330 } | |
1331 | |
1332 // Entry points | |
1333 // | |
1334 // Here we generate the various kind of entries into the interpreter. | |
1335 // The two main entry type are generic bytecode methods and native | |
1336 // call method. These both come in synchronized and non-synchronized | |
1337 // versions but the frame layout they create is very similar. The | |
1338 // other method entry types are really just special purpose entries | |
1339 // that are really entry and interpretation all in one. These are for | |
1340 // trivial methods like accessor, empty, or special math methods. | |
1341 // | |
1342 // When control flow reaches any of the entry types for the interpreter | |
1343 // the following holds -> | |
1344 // | |
1345 // Arguments: | |
1346 // | |
1347 // rbx: methodOop | |
1348 // | |
1349 // Stack layout immediately at entry | |
1350 // | |
1351 // [ return address ] <--- rsp | |
1352 // [ parameter n ] | |
1353 // ... | |
1354 // [ parameter 1 ] | |
1355 // [ expression stack ] (caller's java expression stack) | |
1356 | |
1357 // Assuming that we don't go to one of the trivial specialized entries | |
1358 // the stack will look like below when we are ready to execute the | |
1359 // first bytecode (or call the native routine). The register usage | |
1360 // will be as the template based interpreter expects (see | |
1361 // interpreter_amd64.hpp). | |
1362 // | |
1363 // local variables follow incoming parameters immediately; i.e. | |
1364 // the return address is moved to the end of the locals). | |
1365 // | |
1366 // [ monitor entry ] <--- rsp | |
1367 // ... | |
1368 // [ monitor entry ] | |
1369 // [ expr. stack bottom ] | |
1370 // [ saved r13 ] | |
1371 // [ current r14 ] | |
1372 // [ methodOop ] | |
1373 // [ saved ebp ] <--- rbp | |
1374 // [ return address ] | |
1375 // [ local variable m ] | |
1376 // ... | |
1377 // [ local variable 1 ] | |
1378 // [ parameter n ] | |
1379 // ... | |
1380 // [ parameter 1 ] <--- r14 | |
1381 | |
1382 address AbstractInterpreterGenerator::generate_method_entry( | |
1383 AbstractInterpreter::MethodKind kind) { | |
1384 // determine code generation flags | |
1385 bool synchronized = false; | |
1386 address entry_point = NULL; | |
1387 | |
1388 switch (kind) { | |
1389 case Interpreter::zerolocals : break; | |
1390 case Interpreter::zerolocals_synchronized: synchronized = true; break; | |
1391 case Interpreter::native : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break; | |
1392 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true); break; | |
1393 case Interpreter::empty : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry(); break; | |
1394 case Interpreter::accessor : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry(); break; | |
1395 case Interpreter::abstract : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry(); break; | |
1396 case Interpreter::java_lang_math_sin : break; | |
1397 case Interpreter::java_lang_math_cos : break; | |
1398 case Interpreter::java_lang_math_tan : break; | |
1399 case Interpreter::java_lang_math_abs : break; | |
1400 case Interpreter::java_lang_math_log : break; | |
1401 case Interpreter::java_lang_math_log10 : break; | |
1402 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind); break; | |
1403 default : ShouldNotReachHere(); break; | |
1404 } | |
1405 | |
1406 if (entry_point) { | |
1407 return entry_point; | |
1408 } | |
1409 | |
1410 return ((InterpreterGenerator*) this)-> | |
1411 generate_normal_entry(synchronized); | |
1412 } | |
1413 | |
1414 // How much stack a method activation needs in words. | |
1415 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) { | |
1416 const int entry_size = frame::interpreter_frame_monitor_size(); | |
1417 | |
1418 // total overhead size: entry_size + (saved rbp thru expr stack | |
1419 // bottom). be sure to change this if you add/subtract anything | |
1420 // to/from the overhead area | |
1421 const int overhead_size = | |
1422 -(frame::interpreter_frame_initial_sp_offset) + entry_size; | |
1423 | |
1424 const int stub_code = frame::entry_frame_after_call_words; | |
1425 const int method_stack = (method->max_locals() + method->max_stack()) * | |
1426 Interpreter::stackElementWords(); | |
1427 return (overhead_size + method_stack + stub_code); | |
1428 } | |
1429 | |
1430 int AbstractInterpreter::layout_activation(methodOop method, | |
1431 int tempcount, | |
1432 int popframe_extra_args, | |
1433 int moncount, | |
1434 int callee_param_count, | |
1435 int callee_locals, | |
1436 frame* caller, | |
1437 frame* interpreter_frame, | |
1438 bool is_top_frame) { | |
1439 // Note: This calculation must exactly parallel the frame setup | |
1440 // in AbstractInterpreterGenerator::generate_method_entry. | |
1441 // If interpreter_frame!=NULL, set up the method, locals, and monitors. | |
1442 // The frame interpreter_frame, if not NULL, is guaranteed to be the | |
1443 // right size, as determined by a previous call to this method. | |
1444 // It is also guaranteed to be walkable even though it is in a skeletal state | |
1445 | |
1446 // fixed size of an interpreter frame: | |
1447 int max_locals = method->max_locals() * Interpreter::stackElementWords(); | |
1448 int extra_locals = (method->max_locals() - method->size_of_parameters()) * | |
1449 Interpreter::stackElementWords(); | |
1450 | |
1451 int overhead = frame::sender_sp_offset - | |
1452 frame::interpreter_frame_initial_sp_offset; | |
1453 // Our locals were accounted for by the caller (or last_frame_adjust | |
1454 // on the transistion) Since the callee parameters already account | |
1455 // for the callee's params we only need to account for the extra | |
1456 // locals. | |
1457 int size = overhead + | |
1458 (callee_locals - callee_param_count)*Interpreter::stackElementWords() + | |
1459 moncount * frame::interpreter_frame_monitor_size() + | |
1460 tempcount* Interpreter::stackElementWords() + popframe_extra_args; | |
1461 if (interpreter_frame != NULL) { | |
1462 #ifdef ASSERT | |
1463 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), | |
1464 "Frame not properly walkable"); | |
1465 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)"); | |
1466 #endif | |
1467 | |
1468 interpreter_frame->interpreter_frame_set_method(method); | |
1469 // NOTE the difference in using sender_sp and | |
1470 // interpreter_frame_sender_sp interpreter_frame_sender_sp is | |
1471 // the original sp of the caller (the unextended_sp) and | |
1472 // sender_sp is fp+16 XXX | |
1473 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1; | |
1474 | |
1475 interpreter_frame->interpreter_frame_set_locals(locals); | |
1476 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin(); | |
1477 BasicObjectLock* monbot = montop - moncount; | |
1478 interpreter_frame->interpreter_frame_set_monitor_end(monbot); | |
1479 | |
1480 // Set last_sp | |
1481 intptr_t* esp = (intptr_t*) monbot - | |
1482 tempcount*Interpreter::stackElementWords() - | |
1483 popframe_extra_args; | |
1484 interpreter_frame->interpreter_frame_set_last_sp(esp); | |
1485 | |
1486 // All frames but the initial (oldest) interpreter frame we fill in have | |
1487 // a value for sender_sp that allows walking the stack but isn't | |
1488 // truly correct. Correct the value here. | |
1489 if (extra_locals != 0 && | |
1490 interpreter_frame->sender_sp() == | |
1491 interpreter_frame->interpreter_frame_sender_sp()) { | |
1492 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + | |
1493 extra_locals); | |
1494 } | |
1495 *interpreter_frame->interpreter_frame_cache_addr() = | |
1496 method->constants()->cache(); | |
1497 } | |
1498 return size; | |
1499 } | |
1500 | |
1501 //----------------------------------------------------------------------------- | |
1502 // Exceptions | |
1503 | |
1504 void TemplateInterpreterGenerator::generate_throw_exception() { | |
1505 // Entry point in previous activation (i.e., if the caller was | |
1506 // interpreted) | |
1507 Interpreter::_rethrow_exception_entry = __ pc(); | |
1508 // Restore sp to interpreter_frame_last_sp even though we are going | |
1509 // to empty the expression stack for the exception processing. | |
304 | 1510 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); |
0 | 1511 // rax: exception |
1512 // rdx: return address/pc that threw exception | |
1513 __ restore_bcp(); // r13 points to call/send | |
1514 __ restore_locals(); | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1515 __ reinit_heapbase(); // restore r12 as heapbase. |
0 | 1516 // Entry point for exceptions thrown within interpreter code |
1517 Interpreter::_throw_exception_entry = __ pc(); | |
1518 // expression stack is undefined here | |
1519 // rax: exception | |
1520 // r13: exception bcp | |
1521 __ verify_oop(rax); | |
304 | 1522 __ mov(c_rarg1, rax); |
0 | 1523 |
1524 // expression stack must be empty before entering the VM in case of | |
1525 // an exception | |
1526 __ empty_expression_stack(); | |
1527 // find exception handler address and preserve exception oop | |
1528 __ call_VM(rdx, | |
1529 CAST_FROM_FN_PTR(address, | |
1530 InterpreterRuntime::exception_handler_for_exception), | |
1531 c_rarg1); | |
1532 // rax: exception handler entry point | |
1533 // rdx: preserved exception oop | |
1534 // r13: bcp for exception handler | |
1535 __ push_ptr(rdx); // push exception which is now the only value on the stack | |
1536 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!) | |
1537 | |
1538 // If the exception is not handled in the current frame the frame is | |
1539 // removed and the exception is rethrown (i.e. exception | |
1540 // continuation is _rethrow_exception). | |
1541 // | |
1542 // Note: At this point the bci is still the bxi for the instruction | |
1543 // which caused the exception and the expression stack is | |
1544 // empty. Thus, for any VM calls at this point, GC will find a legal | |
1545 // oop map (with empty expression stack). | |
1546 | |
1547 // In current activation | |
1548 // tos: exception | |
1549 // esi: exception bcp | |
1550 | |
1551 // | |
1552 // JVMTI PopFrame support | |
1553 // | |
1554 | |
1555 Interpreter::_remove_activation_preserving_args_entry = __ pc(); | |
1556 __ empty_expression_stack(); | |
1557 // Set the popframe_processing bit in pending_popframe_condition | |
1558 // indicating that we are currently handling popframe, so that | |
1559 // call_VMs that may happen later do not trigger new popframe | |
1560 // handling cycles. | |
1561 __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset())); | |
1562 __ orl(rdx, JavaThread::popframe_processing_bit); | |
1563 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx); | |
1564 | |
1565 { | |
1566 // Check to see whether we are returning to a deoptimized frame. | |
1567 // (The PopFrame call ensures that the caller of the popped frame is | |
1568 // either interpreted or compiled and deoptimizes it if compiled.) | |
1569 // In this case, we can't call dispatch_next() after the frame is | |
1570 // popped, but instead must save the incoming arguments and restore | |
1571 // them after deoptimization has occurred. | |
1572 // | |
1573 // Note that we don't compare the return PC against the | |
1574 // deoptimization blob's unpack entry because of the presence of | |
1575 // adapter frames in C2. | |
1576 Label caller_not_deoptimized; | |
304 | 1577 __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize)); |
0 | 1578 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, |
1579 InterpreterRuntime::interpreter_contains), c_rarg1); | |
1580 __ testl(rax, rax); | |
1581 __ jcc(Assembler::notZero, caller_not_deoptimized); | |
1582 | |
1583 // Compute size of arguments for saving when returning to | |
1584 // deoptimized caller | |
1585 __ get_method(rax); | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
1586 __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc:: |
0 | 1587 size_of_parameters_offset()))); |
1588 __ shll(rax, Interpreter::logStackElementSize()); | |
1589 __ restore_locals(); // XXX do we need this? | |
304 | 1590 __ subptr(r14, rax); |
1591 __ addptr(r14, wordSize); | |
0 | 1592 // Save these arguments |
1593 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, | |
1594 Deoptimization:: | |
1595 popframe_preserve_args), | |
1596 r15_thread, rax, r14); | |
1597 | |
1598 __ remove_activation(vtos, rdx, | |
1599 /* throw_monitor_exception */ false, | |
1600 /* install_monitor_exception */ false, | |
1601 /* notify_jvmdi */ false); | |
1602 | |
1603 // Inform deoptimization that it is responsible for restoring | |
1604 // these arguments | |
1605 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), | |
1606 JavaThread::popframe_force_deopt_reexecution_bit); | |
1607 | |
1608 // Continue in deoptimization handler | |
1609 __ jmp(rdx); | |
1610 | |
1611 __ bind(caller_not_deoptimized); | |
1612 } | |
1613 | |
1614 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */ | |
1615 /* throw_monitor_exception */ false, | |
1616 /* install_monitor_exception */ false, | |
1617 /* notify_jvmdi */ false); | |
1618 | |
1619 // Finish with popframe handling | |
1620 // A previous I2C followed by a deoptimization might have moved the | |
1621 // outgoing arguments further up the stack. PopFrame expects the | |
1622 // mutations to those outgoing arguments to be preserved and other | |
1623 // constraints basically require this frame to look exactly as | |
1624 // though it had previously invoked an interpreted activation with | |
1625 // no space between the top of the expression stack (current | |
1626 // last_sp) and the top of stack. Rather than force deopt to | |
1627 // maintain this kind of invariant all the time we call a small | |
1628 // fixup routine to move the mutated arguments onto the top of our | |
1629 // expression stack if necessary. | |
304 | 1630 __ mov(c_rarg1, rsp); |
1631 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); | |
0 | 1632 // PC must point into interpreter here |
1633 __ set_last_Java_frame(noreg, rbp, __ pc()); | |
1634 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2); | |
1635 __ reset_last_Java_frame(true, true); | |
1636 // Restore the last_sp and null it out | |
304 | 1637 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); |
1638 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); | |
0 | 1639 |
1640 __ restore_bcp(); // XXX do we need this? | |
1641 __ restore_locals(); // XXX do we need this? | |
1642 // The method data pointer was incremented already during | |
1643 // call profiling. We have to restore the mdp for the current bcp. | |
1644 if (ProfileInterpreter) { | |
1645 __ set_method_data_pointer_for_bcp(); | |
1646 } | |
1647 | |
1648 // Clear the popframe condition flag | |
1649 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), | |
1650 JavaThread::popframe_inactive); | |
1651 | |
1652 __ dispatch_next(vtos); | |
1653 // end of PopFrame support | |
1654 | |
1655 Interpreter::_remove_activation_entry = __ pc(); | |
1656 | |
1657 // preserve exception over this code sequence | |
1658 __ pop_ptr(rax); | |
304 | 1659 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax); |
0 | 1660 // remove the activation (without doing throws on illegalMonitorExceptions) |
1661 __ remove_activation(vtos, rdx, false, true, false); | |
1662 // restore exception | |
304 | 1663 __ movptr(rax, Address(r15_thread, JavaThread::vm_result_offset())); |
1664 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD); | |
0 | 1665 __ verify_oop(rax); |
1666 | |
1667 // In between activations - previous activation type unknown yet | |
1668 // compute continuation point - the continuation point expects the | |
1669 // following registers set up: | |
1670 // | |
1671 // rax: exception | |
1672 // rdx: return address/pc that threw exception | |
1673 // rsp: expression stack of caller | |
1674 // rbp: ebp of caller | |
304 | 1675 __ push(rax); // save exception |
1676 __ push(rdx); // save return address | |
0 | 1677 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, |
1678 SharedRuntime::exception_handler_for_return_address), | |
1679 rdx); | |
304 | 1680 __ mov(rbx, rax); // save exception handler |
1681 __ pop(rdx); // restore return address | |
1682 __ pop(rax); // restore exception | |
0 | 1683 // Note that an "issuing PC" is actually the next PC after the call |
1684 __ jmp(rbx); // jump to exception | |
1685 // handler of caller | |
1686 } | |
1687 | |
1688 | |
1689 // | |
1690 // JVMTI ForceEarlyReturn support | |
1691 // | |
1692 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { | |
1693 address entry = __ pc(); | |
1694 | |
1695 __ restore_bcp(); | |
1696 __ restore_locals(); | |
1697 __ empty_expression_stack(); | |
1698 __ load_earlyret_value(state); | |
1699 | |
304 | 1700 __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset())); |
0 | 1701 Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset()); |
1702 | |
1703 // Clear the earlyret state | |
1704 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive); | |
1705 | |
1706 __ remove_activation(state, rsi, | |
1707 false, /* throw_monitor_exception */ | |
1708 false, /* install_monitor_exception */ | |
1709 true); /* notify_jvmdi */ | |
1710 __ jmp(rsi); | |
1711 | |
1712 return entry; | |
1713 } // end of ForceEarlyReturn support | |
1714 | |
1715 | |
1716 //----------------------------------------------------------------------------- | |
1717 // Helper for vtos entry point generation | |
1718 | |
1719 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, | |
1720 address& bep, | |
1721 address& cep, | |
1722 address& sep, | |
1723 address& aep, | |
1724 address& iep, | |
1725 address& lep, | |
1726 address& fep, | |
1727 address& dep, | |
1728 address& vep) { | |
1729 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); | |
1730 Label L; | |
1731 aep = __ pc(); __ push_ptr(); __ jmp(L); | |
1732 fep = __ pc(); __ push_f(); __ jmp(L); | |
1733 dep = __ pc(); __ push_d(); __ jmp(L); | |
1734 lep = __ pc(); __ push_l(); __ jmp(L); | |
1735 bep = cep = sep = | |
1736 iep = __ pc(); __ push_i(); | |
1737 vep = __ pc(); | |
1738 __ bind(L); | |
1739 generate_and_dispatch(t); | |
1740 } | |
1741 | |
1742 | |
1743 //----------------------------------------------------------------------------- | |
1744 // Generation of individual instructions | |
1745 | |
1746 // helpers for generate_and_dispatch | |
1747 | |
1748 | |
1749 InterpreterGenerator::InterpreterGenerator(StubQueue* code) | |
1750 : TemplateInterpreterGenerator(code) { | |
1751 generate_all(); // down here so it can be "virtual" | |
1752 } | |
1753 | |
1754 //----------------------------------------------------------------------------- | |
1755 | |
1756 // Non-product code | |
1757 #ifndef PRODUCT | |
1758 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { | |
1759 address entry = __ pc(); | |
1760 | |
1761 __ push(state); | |
304 | 1762 __ push(c_rarg0); |
1763 __ push(c_rarg1); | |
1764 __ push(c_rarg2); | |
1765 __ push(c_rarg3); | |
1766 __ mov(c_rarg2, rax); // Pass itos | |
0 | 1767 #ifdef _WIN64 |
1768 __ movflt(xmm3, xmm0); // Pass ftos | |
1769 #endif | |
1770 __ call_VM(noreg, | |
1771 CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), | |
1772 c_rarg1, c_rarg2, c_rarg3); | |
304 | 1773 __ pop(c_rarg3); |
1774 __ pop(c_rarg2); | |
1775 __ pop(c_rarg1); | |
1776 __ pop(c_rarg0); | |
0 | 1777 __ pop(state); |
1778 __ ret(0); // return from result handler | |
1779 | |
1780 return entry; | |
1781 } | |
1782 | |
1783 void TemplateInterpreterGenerator::count_bytecode() { | |
1784 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value)); | |
1785 } | |
1786 | |
1787 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { | |
1788 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()])); | |
1789 } | |
1790 | |
1791 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { | |
1792 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index)); | |
1793 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes); | |
1794 __ orl(rbx, | |
1795 ((int) t->bytecode()) << | |
1796 BytecodePairHistogram::log2_number_of_codes); | |
1797 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx); | |
1798 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters)); | |
1799 __ incrementl(Address(rscratch1, rbx, Address::times_4)); | |
1800 } | |
1801 | |
1802 | |
1803 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { | |
1804 // Call a little run-time stub to avoid blow-up for each bytecode. | |
1805 // The run-time runtime saves the right registers, depending on | |
1806 // the tosca in-state for the given template. | |
1807 | |
1808 assert(Interpreter::trace_code(t->tos_in()) != NULL, | |
1809 "entry must have been generated"); | |
304 | 1810 __ mov(r12, rsp); // remember sp |
1811 __ andptr(rsp, -16); // align stack as required by ABI | |
0 | 1812 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in()))); |
304 | 1813 __ mov(rsp, r12); // restore sp |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1814 __ reinit_heapbase(); |
0 | 1815 } |
1816 | |
1817 | |
1818 void TemplateInterpreterGenerator::stop_interpreter_at() { | |
1819 Label L; | |
1820 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value), | |
1821 StopInterpreterAt); | |
1822 __ jcc(Assembler::notEqual, L); | |
1823 __ int3(); | |
1824 __ bind(L); | |
1825 } | |
1826 #endif // !PRODUCT | |
304 | 1827 #endif // ! CC_INTERP |