1 /*
2 * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "gc/shared/barrierSetAssembler.hpp"
28 #include "interpreter/bytecodeHistogram.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "interpreter/interpreterRuntime.hpp"
32 #include "interpreter/templateInterpreterGenerator.hpp"
33 #include "interpreter/templateTable.hpp"
34 #include "oops/arrayOop.hpp"
35 #include "oops/methodData.hpp"
36 #include "oops/method.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "prims/jvmtiExport.hpp"
39 #include "prims/jvmtiThreadState.hpp"
40 #include "runtime/arguments.hpp"
41 #include "runtime/deoptimization.hpp"
42 #include "runtime/frame.inline.hpp"
43 #include "runtime/sharedRuntime.hpp"
44 #include "runtime/stubRoutines.hpp"
45 #include "runtime/synchronizer.hpp"
46 #include "runtime/timer.hpp"
47 #include "runtime/vframeArray.hpp"
48 #include "utilities/debug.hpp"
49 #include "utilities/macros.hpp"
50
51 #define __ _masm->
52
53 // Size of interpreter code. Increase if too small. Interpreter will
54 // fail with a guarantee ("not enough space for interpreter generation");
55 // if too small.
56 // Run with +PrintInterpreter to get the VM to print out the size.
57 // Max size with JVMTI
58 #ifdef AMD64
59 int TemplateInterpreter::InterpreterCodeSize = JVMCI_ONLY(268) NOT_JVMCI(256) * 1024;
60 #else
61 int TemplateInterpreter::InterpreterCodeSize = 224 * 1024;
62 #endif // AMD64
63
64 // Global Register Names
65 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
66 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
67
68 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
69 const int bcp_offset = frame::interpreter_frame_bcp_offset * wordSize;
70 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
71
72
73 //-----------------------------------------------------------------------------
74
75 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
76 address entry = __ pc();
77
78 #ifdef ASSERT
79 {
80 Label L;
81 __ lea(rax, Address(rbp,
82 frame::interpreter_frame_monitor_block_top_offset *
83 wordSize));
84 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
85 // grows negative)
86 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
87 __ stop ("interpreter frame not set up");
88 __ bind(L);
89 }
90 #endif // ASSERT
91 // Restore bcp under the assumption that the current frame is still
92 // interpreted
93 __ restore_bcp();
94
95 // expression stack must be empty before entering the VM if an
96 // exception happened
97 __ empty_expression_stack();
98 // throw exception
99 __ call_VM(noreg,
100 CAST_FROM_FN_PTR(address,
101 InterpreterRuntime::throw_StackOverflowError));
102 return entry;
103 }
104
105 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
106 address entry = __ pc();
107 // expression stack must be empty before entering the VM if an
108 // exception happened
109 __ empty_expression_stack();
110 // setup parameters
111 // ??? convention: expect aberrant index in register ebx
112 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
113 // Pass array to create more detailed exceptions.
114 __ call_VM(noreg,
115 CAST_FROM_FN_PTR(address,
116 InterpreterRuntime::
117 throw_ArrayIndexOutOfBoundsException),
118 rarg, rbx);
119 return entry;
120 }
121
122 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
123 address entry = __ pc();
124
125 // object is at TOS
126 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
127 __ pop(rarg);
128
129 // expression stack must be empty before entering the VM if an
130 // exception happened
131 __ empty_expression_stack();
132
133 __ call_VM(noreg,
134 CAST_FROM_FN_PTR(address,
135 InterpreterRuntime::
136 throw_ClassCastException),
137 rarg);
138 return entry;
139 }
140
141 address TemplateInterpreterGenerator::generate_exception_handler_common(
142 const char* name, const char* message, bool pass_oop) {
143 assert(!pass_oop || message == NULL, "either oop or message but not both");
144 address entry = __ pc();
145
146 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
147 Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
148
149 if (pass_oop) {
150 // object is at TOS
151 __ pop(rarg2);
152 }
153 // expression stack must be empty before entering the VM if an
154 // exception happened
155 __ empty_expression_stack();
156 // setup parameters
157 __ lea(rarg, ExternalAddress((address)name));
158 if (pass_oop) {
159 __ call_VM(rax, CAST_FROM_FN_PTR(address,
160 InterpreterRuntime::
161 create_klass_exception),
162 rarg, rarg2);
163 } else {
164 __ lea(rarg2, ExternalAddress((address)message));
165 __ call_VM(rax,
166 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
167 rarg, rarg2);
168 }
169 // throw exception
170 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
171 return entry;
172 }
173
174 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
175 address entry = __ pc();
176
177 #ifndef _LP64
178 #ifdef COMPILER2
179 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
180 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
181 for (int i = 1; i < 8; i++) {
182 __ ffree(i);
183 }
184 } else if (UseSSE < 2) {
185 __ empty_FPU_stack();
186 }
187 #endif // COMPILER2
188 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
189 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
190 } else {
191 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
192 }
193
194 if (state == ftos) {
195 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
196 } else if (state == dtos) {
197 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
198 }
199 #endif // _LP64
200
201 // Restore stack bottom in case i2c adjusted stack
202 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
203 // and NULL it as marker that esp is now tos until next java call
204 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
205
206 __ restore_bcp();
207 __ restore_locals();
208
209 if (state == atos) {
210 Register mdp = rbx;
211 Register tmp = rcx;
212 __ profile_return_type(mdp, rax, tmp);
213 }
214
215 const Register cache = rbx;
216 const Register index = rcx;
217 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
218
219 const Register flags = cache;
220 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
221 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
222 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
223
224 const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
225 if (JvmtiExport::can_pop_frame()) {
226 NOT_LP64(__ get_thread(java_thread));
227 __ check_and_handle_popframe(java_thread);
228 }
229 if (JvmtiExport::can_force_early_return()) {
230 NOT_LP64(__ get_thread(java_thread));
231 __ check_and_handle_earlyret(java_thread);
232 }
233
234 __ dispatch_next(state, step);
235
236 return entry;
237 }
238
239
240 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
241 address entry = __ pc();
242
243 #ifndef _LP64
244 if (state == ftos) {
245 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
246 } else if (state == dtos) {
247 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
248 }
249 #endif // _LP64
250
251 // NULL last_sp until next java call
252 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
253 __ restore_bcp();
254 __ restore_locals();
255 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
256 NOT_LP64(__ get_thread(thread));
257 #if INCLUDE_JVMCI
258 // Check if we need to take lock at entry of synchronized method. This can
259 // only occur on method entry so emit it only for vtos with step 0.
260 if ((EnableJVMCI || UseAOT) && state == vtos && step == 0) {
261 Label L;
262 __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
263 __ jcc(Assembler::zero, L);
264 // Clear flag.
265 __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
266 // Satisfy calling convention for lock_method().
267 __ get_method(rbx);
268 // Take lock.
269 lock_method();
270 __ bind(L);
271 } else {
272 #ifdef ASSERT
273 if (EnableJVMCI) {
274 Label L;
275 __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
276 __ jccb(Assembler::zero, L);
277 __ stop("unexpected pending monitor in deopt entry");
278 __ bind(L);
279 }
280 #endif
281 }
282 #endif
283 // handle exceptions
284 {
285 Label L;
286 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
287 __ jcc(Assembler::zero, L);
288 __ call_VM(noreg,
289 CAST_FROM_FN_PTR(address,
290 InterpreterRuntime::throw_pending_exception));
291 __ should_not_reach_here();
292 __ bind(L);
293 }
294 if (continuation == NULL) {
295 __ dispatch_next(state, step);
296 } else {
297 __ jump_to_entry(continuation);
298 }
299 return entry;
300 }
301
302 address TemplateInterpreterGenerator::generate_result_handler_for(
303 BasicType type) {
304 address entry = __ pc();
305 switch (type) {
306 case T_BOOLEAN: __ c2bool(rax); break;
307 #ifndef _LP64
308 case T_CHAR : __ andptr(rax, 0xFFFF); break;
309 #else
310 case T_CHAR : __ movzwl(rax, rax); break;
311 #endif // _LP64
312 case T_BYTE : __ sign_extend_byte(rax); break;
313 case T_SHORT : __ sign_extend_short(rax); break;
314 case T_INT : /* nothing to do */ break;
315 case T_LONG : /* nothing to do */ break;
316 case T_VOID : /* nothing to do */ break;
317 #ifndef _LP64
318 case T_DOUBLE :
319 case T_FLOAT :
320 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
321 __ pop(t); // remove return address first
322 // Must return a result for interpreter or compiler. In SSE
323 // mode, results are returned in xmm0 and the FPU stack must
324 // be empty.
325 if (type == T_FLOAT && UseSSE >= 1) {
326 // Load ST0
327 __ fld_d(Address(rsp, 0));
328 // Store as float and empty fpu stack
329 __ fstp_s(Address(rsp, 0));
330 // and reload
331 __ movflt(xmm0, Address(rsp, 0));
332 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
333 __ movdbl(xmm0, Address(rsp, 0));
334 } else {
335 // restore ST0
336 __ fld_d(Address(rsp, 0));
337 }
338 // and pop the temp
339 __ addptr(rsp, 2 * wordSize);
340 __ push(t); // restore return address
341 }
342 break;
343 #else
344 case T_FLOAT : /* nothing to do */ break;
345 case T_DOUBLE : /* nothing to do */ break;
346 #endif // _LP64
347
348 case T_OBJECT :
349 // retrieve result from frame
350 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
351 // and verify it
352 __ verify_oop(rax);
353 break;
354 default : ShouldNotReachHere();
355 }
356 __ ret(0); // return from result handler
357 return entry;
358 }
359
360 address TemplateInterpreterGenerator::generate_safept_entry_for(
361 TosState state,
362 address runtime_entry) {
363 address entry = __ pc();
364 __ push(state);
365 __ call_VM(noreg, runtime_entry);
366 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
367 return entry;
368 }
369
370
371
372 // Helpers for commoning out cases in the various type of method entries.
373 //
374
375
376 // increment invocation count & check for overflow
377 //
378 // Note: checking for negative value instead of overflow
379 // so we have a 'sticky' overflow test
380 //
381 // rbx: method
382 // rcx: invocation counter
383 //
384 void TemplateInterpreterGenerator::generate_counter_incr(
385 Label* overflow,
386 Label* profile_method,
387 Label* profile_method_continue) {
388 Label done;
389 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
390 if (TieredCompilation) {
391 int increment = InvocationCounter::count_increment;
392 Label no_mdo;
393 if (ProfileInterpreter) {
394 // Are we profiling?
395 __ movptr(rax, Address(rbx, Method::method_data_offset()));
396 __ testptr(rax, rax);
397 __ jccb(Assembler::zero, no_mdo);
398 // Increment counter in the MDO
399 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
400 in_bytes(InvocationCounter::counter_offset()));
401 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
402 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
403 __ jmp(done);
404 }
405 __ bind(no_mdo);
406 // Increment counter in MethodCounters
407 const Address invocation_counter(rax,
408 MethodCounters::invocation_counter_offset() +
409 InvocationCounter::counter_offset());
410 __ get_method_counters(rbx, rax, done);
411 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
412 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
413 false, Assembler::zero, overflow);
414 __ bind(done);
415 } else { // not TieredCompilation
416 const Address backedge_counter(rax,
417 MethodCounters::backedge_counter_offset() +
418 InvocationCounter::counter_offset());
419 const Address invocation_counter(rax,
420 MethodCounters::invocation_counter_offset() +
421 InvocationCounter::counter_offset());
422
423 __ get_method_counters(rbx, rax, done);
424
425 if (ProfileInterpreter) {
426 __ incrementl(Address(rax,
427 MethodCounters::interpreter_invocation_counter_offset()));
428 }
429 // Update standard invocation counters
430 __ movl(rcx, invocation_counter);
431 __ incrementl(rcx, InvocationCounter::count_increment);
432 __ movl(invocation_counter, rcx); // save invocation count
433
434 __ movl(rax, backedge_counter); // load backedge counter
435 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
436
437 __ addl(rcx, rax); // add both counters
438
439 // profile_method is non-null only for interpreted method so
440 // profile_method != NULL == !native_call
441
442 if (ProfileInterpreter && profile_method != NULL) {
443 // Test to see if we should create a method data oop
444 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
445 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
446 __ jcc(Assembler::less, *profile_method_continue);
447
448 // if no method data exists, go to profile_method
449 __ test_method_data_pointer(rax, *profile_method);
450 }
451
452 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
453 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
454 __ jcc(Assembler::aboveEqual, *overflow);
455 __ bind(done);
456 }
457 }
458
459 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
460
461 // Asm interpreter on entry
462 // r14/rdi - locals
463 // r13/rsi - bcp
464 // rbx - method
465 // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
466 // rbp - interpreter frame
467
468 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
469 // Everything as it was on entry
470 // rdx is not restored. Doesn't appear to really be set.
471
472 // InterpreterRuntime::frequency_counter_overflow takes two
473 // arguments, the first (thread) is passed by call_VM, the second
474 // indicates if the counter overflow occurs at a backwards branch
475 // (NULL bcp). We pass zero for it. The call returns the address
476 // of the verified entry point for the method or NULL if the
477 // compilation did not complete (either went background or bailed
478 // out).
479 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
480 __ movl(rarg, 0);
481 __ call_VM(noreg,
482 CAST_FROM_FN_PTR(address,
483 InterpreterRuntime::frequency_counter_overflow),
484 rarg);
485
486 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
487 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
488 // and jump to the interpreted entry.
489 __ jmp(do_continue, relocInfo::none);
490 }
491
492 // See if we've got enough room on the stack for locals plus overhead below
493 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
494 // without going through the signal handler, i.e., reserved and yellow zones
495 // will not be made usable. The shadow zone must suffice to handle the
496 // overflow.
497 // The expression stack grows down incrementally, so the normal guard
498 // page mechanism will work for that.
499 //
500 // NOTE: Since the additional locals are also always pushed (wasn't
501 // obvious in generate_fixed_frame) so the guard should work for them
502 // too.
503 //
504 // Args:
505 // rdx: number of additional locals this frame needs (what we must check)
506 // rbx: Method*
507 //
508 // Kills:
509 // rax
510 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
511
512 // monitor entry size: see picture of stack in frame_x86.hpp
513 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
514
515 // total overhead size: entry_size + (saved rbp through expr stack
516 // bottom). be sure to change this if you add/subtract anything
517 // to/from the overhead area
518 const int overhead_size =
519 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
520
521 const int page_size = os::vm_page_size();
522
523 Label after_frame_check;
524
525 // see if the frame is greater than one page in size. If so,
526 // then we need to verify there is enough stack space remaining
527 // for the additional locals.
528 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
529 __ jcc(Assembler::belowEqual, after_frame_check);
530
531 // compute rsp as if this were going to be the last frame on
532 // the stack before the red zone
533
534 Label after_frame_check_pop;
535 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
536 #ifndef _LP64
537 __ push(thread);
538 __ get_thread(thread);
539 #endif
540
541 const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
542
543 // locals + overhead, in bytes
544 __ mov(rax, rdx);
545 __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
546 __ addptr(rax, overhead_size);
547
548 #ifdef ASSERT
549 Label limit_okay;
550 // Verify that thread stack overflow limit is non-zero.
551 __ cmpptr(stack_limit, (int32_t)NULL_WORD);
552 __ jcc(Assembler::notEqual, limit_okay);
553 __ stop("stack overflow limit is zero");
554 __ bind(limit_okay);
555 #endif
556
557 // Add locals/frame size to stack limit.
558 __ addptr(rax, stack_limit);
559
560 // Check against the current stack bottom.
561 __ cmpptr(rsp, rax);
562
563 __ jcc(Assembler::above, after_frame_check_pop);
564 NOT_LP64(__ pop(rsi)); // get saved bcp
565
566 // Restore sender's sp as SP. This is necessary if the sender's
567 // frame is an extended compiled frame (see gen_c2i_adapter())
568 // and safer anyway in case of JSR292 adaptations.
569
570 __ pop(rax); // return address must be moved if SP is changed
571 __ mov(rsp, rbcp);
572 __ push(rax);
573
574 // Note: the restored frame is not necessarily interpreted.
575 // Use the shared runtime version of the StackOverflowError.
576 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
577 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
578 // all done with frame size check
579 __ bind(after_frame_check_pop);
580 NOT_LP64(__ pop(rsi));
581
582 // all done with frame size check
583 __ bind(after_frame_check);
584 }
585
586 // Allocate monitor and lock method (asm interpreter)
587 //
588 // Args:
589 // rbx: Method*
590 // r14/rdi: locals
591 //
592 // Kills:
593 // rax
594 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
595 // rscratch1, rscratch2 (scratch regs)
596 void TemplateInterpreterGenerator::lock_method() {
597 // synchronize method
598 const Address access_flags(rbx, Method::access_flags_offset());
599 const Address monitor_block_top(
600 rbp,
601 frame::interpreter_frame_monitor_block_top_offset * wordSize);
602 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
603
604 #ifdef ASSERT
605 {
606 Label L;
607 __ movl(rax, access_flags);
608 __ testl(rax, JVM_ACC_SYNCHRONIZED);
609 __ jcc(Assembler::notZero, L);
610 __ stop("method doesn't need synchronization");
611 __ bind(L);
612 }
613 #endif // ASSERT
614
615 // get synchronization object
616 {
617 Label done;
618 __ movl(rax, access_flags);
619 __ testl(rax, JVM_ACC_STATIC);
620 // get receiver (assume this is frequent case)
621 __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
622 __ jcc(Assembler::zero, done);
623 __ load_mirror(rax, rbx);
624
625 #ifdef ASSERT
626 {
627 Label L;
628 __ testptr(rax, rax);
629 __ jcc(Assembler::notZero, L);
630 __ stop("synchronization object is NULL");
631 __ bind(L);
632 }
633 #endif // ASSERT
634
635 __ bind(done);
636 }
637
638 // add space for monitor & lock
639 __ subptr(rsp, entry_size); // add space for a monitor entry
640 __ movptr(monitor_block_top, rsp); // set new monitor block top
641 // store object
642 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
643 const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
644 __ movptr(lockreg, rsp); // object address
645 __ lock_object(lockreg);
646 }
647
648 // Generate a fixed interpreter frame. This is identical setup for
649 // interpreted methods and for native methods hence the shared code.
650 //
651 // Args:
652 // rax: return address
653 // rbx: Method*
654 // r14/rdi: pointer to locals
655 // r13/rsi: sender sp
656 // rdx: cp cache
657 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
658 // initialize fixed part of activation frame
659 __ push(rax); // save return address
660 __ enter(); // save old & set new rbp
661 __ push(rbcp); // set sender sp
662 __ push((int)NULL_WORD); // leave last_sp as null
663 __ movptr(rbcp, Address(rbx, Method::const_offset())); // get ConstMethod*
664 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
665 __ push(rbx); // save Method*
666 // Get mirror and store it in the frame as GC root for this Method*
667 __ load_mirror(rdx, rbx);
668 __ push(rdx);
669 if (ProfileInterpreter) {
670 Label method_data_continue;
671 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
672 __ testptr(rdx, rdx);
673 __ jcc(Assembler::zero, method_data_continue);
674 __ addptr(rdx, in_bytes(MethodData::data_offset()));
675 __ bind(method_data_continue);
676 __ push(rdx); // set the mdp (method data pointer)
677 } else {
678 __ push(0);
679 }
680
681 __ movptr(rdx, Address(rbx, Method::const_offset()));
682 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
683 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
684 __ push(rdx); // set constant pool cache
685 __ push(rlocals); // set locals pointer
686 if (native_call) {
687 __ push(0); // no bcp
688 } else {
689 __ push(rbcp); // set bcp
690 }
691 __ push(0); // reserve word for pointer to expression stack bottom
692 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
693 }
694
695 // End of helpers
696
697 // Method entry for java.lang.ref.Reference.get.
698 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
699 // Code: _aload_0, _getfield, _areturn
700 // parameter size = 1
701 //
702 // The code that gets generated by this routine is split into 2 parts:
703 // 1. The "intrinsified" code performing an ON_WEAK_OOP_REF load,
704 // 2. The slow path - which is an expansion of the regular method entry.
705 //
706 // Notes:-
707 // * An intrinsic is always executed, where an ON_WEAK_OOP_REF load is performed.
708 // * We may jump to the slow path iff the receiver is null. If the
709 // Reference object is null then we no longer perform an ON_WEAK_OOP_REF load
710 // Thus we can use the regular method entry code to generate the NPE.
711 //
712 // rbx: Method*
713
714 // r13: senderSP must preserve for slow path, set SP to it on fast path
715
716 address entry = __ pc();
717
718 const int referent_offset = java_lang_ref_Reference::referent_offset;
719 guarantee(referent_offset > 0, "referent offset not initialized");
720
721 Label slow_path;
722 // rbx: method
723
724 // Check if local 0 != NULL
725 // If the receiver is null then it is OK to jump to the slow path.
726 __ movptr(rax, Address(rsp, wordSize));
727
728 __ testptr(rax, rax);
729 __ jcc(Assembler::zero, slow_path);
730
731 // rax: local 0
732 // rbx: method (but can be used as scratch now)
733 // rdx: scratch
734 // rdi: scratch
735
736 // Preserve the sender sp in case the load barrier
737 // calls the runtime
738 NOT_LP64(__ push(rsi));
739
740 // Load the value of the referent field.
741 const Address field_address(rax, referent_offset);
742 __ load_heap_oop(rax, field_address, /*tmp1*/ rbx, /*tmp_thread*/ rdx, ON_WEAK_OOP_REF);
743
744 // _areturn
745 const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
746 NOT_LP64(__ pop(rsi)); // get sender sp
747 __ pop(rdi); // get return address
748 __ mov(rsp, sender_sp); // set sp to sender sp
749 __ jmp(rdi);
750 __ ret(0);
751
752 // generate a vanilla interpreter entry as the slow path
753 __ bind(slow_path);
754 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
755 return entry;
756 }
757
758 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
759 // Quick & dirty stack overflow checking: bang the stack & handle trap.
760 // Note that we do the banging after the frame is setup, since the exception
761 // handling code expects to find a valid interpreter frame on the stack.
762 // Doing the banging earlier fails if the caller frame is not an interpreter
763 // frame.
764 // (Also, the exception throwing code expects to unlock any synchronized
765 // method receiever, so do the banging after locking the receiver.)
766
767 // Bang each page in the shadow zone. We can't assume it's been done for
768 // an interpreter frame with greater than a page of locals, so each page
769 // needs to be checked. Only true for non-native.
770 if (UseStackBanging) {
771 const int page_size = os::vm_page_size();
772 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
773 const int start_page = native_call ? n_shadow_pages : 1;
774 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
775 __ bang_stack_with_offset(pages*page_size);
776 }
777 }
778 }
779
780 // Interpreter stub for calling a native method. (asm interpreter)
781 // This sets up a somewhat different looking stack for calling the
782 // native method than the typical interpreter frame setup.
783 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
784 // determine code generation flags
785 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
786
787 // rbx: Method*
788 // rbcp: sender sp
789
790 address entry_point = __ pc();
791
792 const Address constMethod (rbx, Method::const_offset());
793 const Address access_flags (rbx, Method::access_flags_offset());
794 const Address size_of_parameters(rcx, ConstMethod::
795 size_of_parameters_offset());
796
797
798 // get parameter size (always needed)
799 __ movptr(rcx, constMethod);
800 __ load_unsigned_short(rcx, size_of_parameters);
801
802 // native calls don't need the stack size check since they have no
803 // expression stack and the arguments are already on the stack and
804 // we only add a handful of words to the stack
805
806 // rbx: Method*
807 // rcx: size of parameters
808 // rbcp: sender sp
809 __ pop(rax); // get return address
810
811 // for natives the size of locals is zero
812
813 // compute beginning of parameters
814 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
815
816 // add 2 zero-initialized slots for native calls
817 // initialize result_handler slot
818 __ push((int) NULL_WORD);
819 // slot for oop temp
820 // (static native method holder mirror/jni oop result)
821 __ push((int) NULL_WORD);
822
823 // initialize fixed part of activation frame
824 generate_fixed_frame(true);
825
826 // make sure method is native & not abstract
827 #ifdef ASSERT
828 __ movl(rax, access_flags);
829 {
830 Label L;
831 __ testl(rax, JVM_ACC_NATIVE);
832 __ jcc(Assembler::notZero, L);
833 __ stop("tried to execute non-native method as native");
834 __ bind(L);
835 }
836 {
837 Label L;
838 __ testl(rax, JVM_ACC_ABSTRACT);
839 __ jcc(Assembler::zero, L);
840 __ stop("tried to execute abstract method in interpreter");
841 __ bind(L);
842 }
843 #endif
844
845 // Since at this point in the method invocation the exception handler
846 // would try to exit the monitor of synchronized methods which hasn't
847 // been entered yet, we set the thread local variable
848 // _do_not_unlock_if_synchronized to true. The remove_activation will
849 // check this flag.
850
851 const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
852 NOT_LP64(__ get_thread(thread1));
853 const Address do_not_unlock_if_synchronized(thread1,
854 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
855 __ movbool(do_not_unlock_if_synchronized, true);
856
857 // increment invocation count & check for overflow
858 Label invocation_counter_overflow;
859 if (inc_counter) {
860 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
861 }
862
863 Label continue_after_compile;
864 __ bind(continue_after_compile);
865
866 bang_stack_shadow_pages(true);
867
868 // reset the _do_not_unlock_if_synchronized flag
869 NOT_LP64(__ get_thread(thread1));
870 __ movbool(do_not_unlock_if_synchronized, false);
871
872 // check for synchronized methods
873 // Must happen AFTER invocation_counter check and stack overflow check,
874 // so method is not locked if overflows.
875 if (synchronized) {
876 lock_method();
877 } else {
878 // no synchronization necessary
879 #ifdef ASSERT
880 {
881 Label L;
882 __ movl(rax, access_flags);
883 __ testl(rax, JVM_ACC_SYNCHRONIZED);
884 __ jcc(Assembler::zero, L);
885 __ stop("method needs synchronization");
886 __ bind(L);
887 }
888 #endif
889 }
890
891 // start execution
892 #ifdef ASSERT
893 {
894 Label L;
895 const Address monitor_block_top(rbp,
896 frame::interpreter_frame_monitor_block_top_offset * wordSize);
897 __ movptr(rax, monitor_block_top);
898 __ cmpptr(rax, rsp);
899 __ jcc(Assembler::equal, L);
900 __ stop("broken stack frame setup in interpreter");
901 __ bind(L);
902 }
903 #endif
904
905 // jvmti support
906 __ notify_method_entry();
907
908 // work registers
909 const Register method = rbx;
910 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
911 const Register t = NOT_LP64(rcx) LP64_ONLY(r11);
912
913 // allocate space for parameters
914 __ get_method(method);
915 __ movptr(t, Address(method, Method::const_offset()));
916 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
917
918 #ifndef _LP64
919 __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
920 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
921 __ subptr(rsp, t);
922 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
923 #else
924 __ shll(t, Interpreter::logStackElementSize);
925
926 __ subptr(rsp, t);
927 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
928 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
929 #endif // _LP64
930
931 // get signature handler
932 {
933 Label L;
934 __ movptr(t, Address(method, Method::signature_handler_offset()));
935 __ testptr(t, t);
936 __ jcc(Assembler::notZero, L);
937 __ call_VM(noreg,
938 CAST_FROM_FN_PTR(address,
939 InterpreterRuntime::prepare_native_call),
940 method);
941 __ get_method(method);
942 __ movptr(t, Address(method, Method::signature_handler_offset()));
943 __ bind(L);
944 }
945
946 // call signature handler
947 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
948 "adjust this code");
949 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
950 "adjust this code");
951 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
952 "adjust this code");
953
954 // The generated handlers do not touch RBX (the method oop).
955 // However, large signatures cannot be cached and are generated
956 // each time here. The slow-path generator can do a GC on return,
957 // so we must reload it after the call.
958 __ call(t);
959 __ get_method(method); // slow path can do a GC, reload RBX
960
961
962 // result handler is in rax
963 // set result handler
964 __ movptr(Address(rbp,
965 (frame::interpreter_frame_result_handler_offset) * wordSize),
966 rax);
967
968 // pass mirror handle if static call
969 {
970 Label L;
971 __ movl(t, Address(method, Method::access_flags_offset()));
972 __ testl(t, JVM_ACC_STATIC);
973 __ jcc(Assembler::zero, L);
974 // get mirror
975 __ load_mirror(t, method);
976 // copy mirror into activation frame
977 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
978 t);
979 // pass handle to mirror
980 #ifndef _LP64
981 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
982 __ movptr(Address(rsp, wordSize), t);
983 #else
984 __ lea(c_rarg1,
985 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
986 #endif // _LP64
987 __ bind(L);
988 }
989
990 // get native function entry point
991 {
992 Label L;
993 __ movptr(rax, Address(method, Method::native_function_offset()));
994 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
995 __ cmpptr(rax, unsatisfied.addr());
996 __ jcc(Assembler::notEqual, L);
997 __ call_VM(noreg,
998 CAST_FROM_FN_PTR(address,
999 InterpreterRuntime::prepare_native_call),
1000 method);
1001 __ get_method(method);
1002 __ movptr(rax, Address(method, Method::native_function_offset()));
1003 __ bind(L);
1004 }
1005
1006 // pass JNIEnv
1007 #ifndef _LP64
1008 __ get_thread(thread);
1009 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1010 __ movptr(Address(rsp, 0), t);
1011
1012 // set_last_Java_frame_before_call
1013 // It is enough that the pc()
1014 // points into the right code segment. It does not have to be the correct return pc.
1015 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1016 #else
1017 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1018
1019 // It is enough that the pc() points into the right code
1020 // segment. It does not have to be the correct return pc.
1021 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1022 #endif // _LP64
1023
1024 // change thread state
1025 #ifdef ASSERT
1026 {
1027 Label L;
1028 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1029 __ cmpl(t, _thread_in_Java);
1030 __ jcc(Assembler::equal, L);
1031 __ stop("Wrong thread state in native stub");
1032 __ bind(L);
1033 }
1034 #endif
1035
1036 // Change state to native
1037
1038 __ movl(Address(thread, JavaThread::thread_state_offset()),
1039 _thread_in_native);
1040
1041 // Call the native method.
1042 __ call(rax);
1043 // 32: result potentially in rdx:rax or ST0
1044 // 64: result potentially in rax or xmm0
1045
1046 // Verify or restore cpu control state after JNI call
1047 __ restore_cpu_control_state_after_jni();
1048
1049 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1050 // in order to extract the result of a method call. If the order of these
1051 // pushes change or anything else is added to the stack then the code in
1052 // interpreter_frame_result must also change.
1053
1054 #ifndef _LP64
1055 // save potential result in ST(0) & rdx:rax
1056 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1057 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1058 // It is safe to do this push because state is _thread_in_native and return address will be found
1059 // via _last_native_pc and not via _last_jave_sp
1060
1061 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1062 // If the order changes or anything else is added to the stack the code in
1063 // interpreter_frame_result will have to be changed.
1064
1065 { Label L;
1066 Label push_double;
1067 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1068 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1069 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1070 float_handler.addr());
1071 __ jcc(Assembler::equal, push_double);
1072 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1073 double_handler.addr());
1074 __ jcc(Assembler::notEqual, L);
1075 __ bind(push_double);
1076 __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1077 __ bind(L);
1078 }
1079 #else
1080 __ push(dtos);
1081 #endif // _LP64
1082
1083 __ push(ltos);
1084
1085 // change thread state
1086 NOT_LP64(__ get_thread(thread));
1087 __ movl(Address(thread, JavaThread::thread_state_offset()),
1088 _thread_in_native_trans);
1089
1090 if (os::is_MP()) {
1091 if (UseMembar) {
1092 // Force this write out before the read below
1093 __ membar(Assembler::Membar_mask_bits(
1094 Assembler::LoadLoad | Assembler::LoadStore |
1095 Assembler::StoreLoad | Assembler::StoreStore));
1096 } else {
1097 // Write serialization page so VM thread can do a pseudo remote membar.
1098 // We use the current thread pointer to calculate a thread specific
1099 // offset to write to within the page. This minimizes bus traffic
1100 // due to cache line collision.
1101 __ serialize_memory(thread, rcx);
1102 }
1103 }
1104
1105 #ifndef _LP64
1106 if (AlwaysRestoreFPU) {
1107 // Make sure the control word is correct.
1108 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1109 }
1110 #endif // _LP64
1111
1112 // check for safepoint operation in progress and/or pending suspend requests
1113 {
1114 Label Continue;
1115 Label slow_path;
1116
1117 #ifndef _LP64
1118 __ safepoint_poll(slow_path, thread, noreg);
1119 #else
1120 __ safepoint_poll(slow_path, r15_thread, rscratch1);
1121 #endif
1122
1123 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1124 __ jcc(Assembler::equal, Continue);
1125 __ bind(slow_path);
1126
1127 // Don't use call_VM as it will see a possible pending exception
1128 // and forward it and never return here preventing us from
1129 // clearing _last_native_pc down below. Also can't use
1130 // call_VM_leaf either as it will check to see if r13 & r14 are
1131 // preserved and correspond to the bcp/locals pointers. So we do a
1132 // runtime call by hand.
1133 //
1134 #ifndef _LP64
1135 __ push(thread);
1136 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1137 JavaThread::check_special_condition_for_native_trans)));
1138 __ increment(rsp, wordSize);
1139 __ get_thread(thread);
1140 #else
1141 __ mov(c_rarg0, r15_thread);
1142 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1143 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1144 __ andptr(rsp, -16); // align stack as required by ABI
1145 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1146 __ mov(rsp, r12); // restore sp
1147 __ reinit_heapbase();
1148 #endif // _LP64
1149 __ bind(Continue);
1150 }
1151
1152 // change thread state
1153 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1154
1155 // reset_last_Java_frame
1156 __ reset_last_Java_frame(thread, true);
1157
1158 if (CheckJNICalls) {
1159 // clear_pending_jni_exception_check
1160 __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1161 }
1162
1163 // reset handle block
1164 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1165 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1166
1167 // If result is an oop unbox and store it in frame where gc will see it
1168 // and result handler will pick it up
1169
1170 {
1171 Label no_oop, not_weak, store_result;
1172 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1173 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1174 __ jcc(Assembler::notEqual, no_oop);
1175 // retrieve result
1176 __ pop(ltos);
1177 // Unbox oop result, e.g. JNIHandles::resolve value.
1178 __ resolve_jobject(rax /* value */,
1179 thread /* thread */,
1180 t /* tmp */);
1181 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1182 // keep stack depth as expected by pushing oop which will eventually be discarded
1183 __ push(ltos);
1184 __ bind(no_oop);
1185 }
1186
1187
1188 {
1189 Label no_reguard;
1190 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1191 JavaThread::stack_guard_yellow_reserved_disabled);
1192 __ jcc(Assembler::notEqual, no_reguard);
1193
1194 __ pusha(); // XXX only save smashed registers
1195 #ifndef _LP64
1196 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1197 __ popa();
1198 #else
1199 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1200 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1201 __ andptr(rsp, -16); // align stack as required by ABI
1202 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1203 __ mov(rsp, r12); // restore sp
1204 __ popa(); // XXX only restore smashed registers
1205 __ reinit_heapbase();
1206 #endif // _LP64
1207
1208 __ bind(no_reguard);
1209 }
1210
1211
1212 // The method register is junk from after the thread_in_native transition
1213 // until here. Also can't call_VM until the bcp has been
1214 // restored. Need bcp for throwing exception below so get it now.
1215 __ get_method(method);
1216
1217 // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1218 __ movptr(rbcp, Address(method, Method::const_offset())); // get ConstMethod*
1219 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
1220
1221 // handle exceptions (exception handling will handle unlocking!)
1222 {
1223 Label L;
1224 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1225 __ jcc(Assembler::zero, L);
1226 // Note: At some point we may want to unify this with the code
1227 // used in call_VM_base(); i.e., we should use the
1228 // StubRoutines::forward_exception code. For now this doesn't work
1229 // here because the rsp is not correctly set at this point.
1230 __ MacroAssembler::call_VM(noreg,
1231 CAST_FROM_FN_PTR(address,
1232 InterpreterRuntime::throw_pending_exception));
1233 __ should_not_reach_here();
1234 __ bind(L);
1235 }
1236
1237 // do unlocking if necessary
1238 {
1239 Label L;
1240 __ movl(t, Address(method, Method::access_flags_offset()));
1241 __ testl(t, JVM_ACC_SYNCHRONIZED);
1242 __ jcc(Assembler::zero, L);
1243 // the code below should be shared with interpreter macro
1244 // assembler implementation
1245 {
1246 Label unlock;
1247 // BasicObjectLock will be first in list, since this is a
1248 // synchronized method. However, need to check that the object
1249 // has not been unlocked by an explicit monitorexit bytecode.
1250 const Address monitor(rbp,
1251 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1252 wordSize - (int)sizeof(BasicObjectLock)));
1253
1254 const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1255
1256 // monitor expect in c_rarg1 for slow unlock path
1257 __ lea(regmon, monitor); // address of first monitor
1258
1259 __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1260 __ testptr(t, t);
1261 __ jcc(Assembler::notZero, unlock);
1262
1263 // Entry already unlocked, need to throw exception
1264 __ MacroAssembler::call_VM(noreg,
1265 CAST_FROM_FN_PTR(address,
1266 InterpreterRuntime::throw_illegal_monitor_state_exception));
1267 __ should_not_reach_here();
1268
1269 __ bind(unlock);
1270 __ unlock_object(regmon);
1271 }
1272 __ bind(L);
1273 }
1274
1275 // jvmti support
1276 // Note: This must happen _after_ handling/throwing any exceptions since
1277 // the exception handler code notifies the runtime of method exits
1278 // too. If this happens before, method entry/exit notifications are
1279 // not properly paired (was bug - gri 11/22/99).
1280 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1281
1282 // restore potential result in edx:eax, call result handler to
1283 // restore potential result in ST0 & handle result
1284
1285 __ pop(ltos);
1286 LP64_ONLY( __ pop(dtos));
1287
1288 __ movptr(t, Address(rbp,
1289 (frame::interpreter_frame_result_handler_offset) * wordSize));
1290 __ call(t);
1291
1292 // remove activation
1293 __ movptr(t, Address(rbp,
1294 frame::interpreter_frame_sender_sp_offset *
1295 wordSize)); // get sender sp
1296 __ leave(); // remove frame anchor
1297 __ pop(rdi); // get return address
1298 __ mov(rsp, t); // set sp to sender sp
1299 __ jmp(rdi);
1300
1301 if (inc_counter) {
1302 // Handle overflow of counter and compile method
1303 __ bind(invocation_counter_overflow);
1304 generate_counter_overflow(continue_after_compile);
1305 }
1306
1307 return entry_point;
1308 }
1309
1310 // Abstract method entry
1311 // Attempt to execute abstract method. Throw exception
1312 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1313
1314 address entry_point = __ pc();
1315
1316 // abstract method entry
1317
1318 // pop return address, reset last_sp to NULL
1319 __ empty_expression_stack();
1320 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1321 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1322
1323 // throw exception
1324 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod), rbx);
1325 // the call_VM checks for exception, so we should never return here.
1326 __ should_not_reach_here();
1327
1328 return entry_point;
1329 }
1330
1331 //
1332 // Generic interpreted method entry to (asm) interpreter
1333 //
1334 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1335 // determine code generation flags
1336 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1337
1338 // ebx: Method*
1339 // rbcp: sender sp
1340 address entry_point = __ pc();
1341
1342 const Address constMethod(rbx, Method::const_offset());
1343 const Address access_flags(rbx, Method::access_flags_offset());
1344 const Address size_of_parameters(rdx,
1345 ConstMethod::size_of_parameters_offset());
1346 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1347
1348
1349 // get parameter size (always needed)
1350 __ movptr(rdx, constMethod);
1351 __ load_unsigned_short(rcx, size_of_parameters);
1352
1353 // rbx: Method*
1354 // rcx: size of parameters
1355 // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1356
1357 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1358 __ subl(rdx, rcx); // rdx = no. of additional locals
1359
1360 // YYY
1361 // __ incrementl(rdx);
1362 // __ andl(rdx, -2);
1363
1364 // see if we've got enough room on the stack for locals plus overhead.
1365 generate_stack_overflow_check();
1366
1367 // get return address
1368 __ pop(rax);
1369
1370 // compute beginning of parameters
1371 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1372
1373 // rdx - # of additional locals
1374 // allocate space for locals
1375 // explicitly initialize locals
1376 {
1377 Label exit, loop;
1378 __ testl(rdx, rdx);
1379 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1380 __ bind(loop);
1381 __ push((int) NULL_WORD); // initialize local variables
1382 __ decrementl(rdx); // until everything initialized
1383 __ jcc(Assembler::greater, loop);
1384 __ bind(exit);
1385 }
1386
1387 // initialize fixed part of activation frame
1388 generate_fixed_frame(false);
1389
1390 // make sure method is not native & not abstract
1391 #ifdef ASSERT
1392 __ movl(rax, access_flags);
1393 {
1394 Label L;
1395 __ testl(rax, JVM_ACC_NATIVE);
1396 __ jcc(Assembler::zero, L);
1397 __ stop("tried to execute native method as non-native");
1398 __ bind(L);
1399 }
1400 {
1401 Label L;
1402 __ testl(rax, JVM_ACC_ABSTRACT);
1403 __ jcc(Assembler::zero, L);
1404 __ stop("tried to execute abstract method in interpreter");
1405 __ bind(L);
1406 }
1407 #endif
1408
1409 // Since at this point in the method invocation the exception
1410 // handler would try to exit the monitor of synchronized methods
1411 // which hasn't been entered yet, we set the thread local variable
1412 // _do_not_unlock_if_synchronized to true. The remove_activation
1413 // will check this flag.
1414
1415 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1416 NOT_LP64(__ get_thread(thread));
1417 const Address do_not_unlock_if_synchronized(thread,
1418 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1419 __ movbool(do_not_unlock_if_synchronized, true);
1420
1421 __ profile_parameters_type(rax, rcx, rdx);
1422 // increment invocation count & check for overflow
1423 Label invocation_counter_overflow;
1424 Label profile_method;
1425 Label profile_method_continue;
1426 if (inc_counter) {
1427 generate_counter_incr(&invocation_counter_overflow,
1428 &profile_method,
1429 &profile_method_continue);
1430 if (ProfileInterpreter) {
1431 __ bind(profile_method_continue);
1432 }
1433 }
1434
1435 Label continue_after_compile;
1436 __ bind(continue_after_compile);
1437
1438 // check for synchronized interpreted methods
1439 bang_stack_shadow_pages(false);
1440
1441 // reset the _do_not_unlock_if_synchronized flag
1442 NOT_LP64(__ get_thread(thread));
1443 __ movbool(do_not_unlock_if_synchronized, false);
1444
1445 // check for synchronized methods
1446 // Must happen AFTER invocation_counter check and stack overflow check,
1447 // so method is not locked if overflows.
1448 if (synchronized) {
1449 // Allocate monitor and lock method
1450 lock_method();
1451 } else {
1452 // no synchronization necessary
1453 #ifdef ASSERT
1454 {
1455 Label L;
1456 __ movl(rax, access_flags);
1457 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1458 __ jcc(Assembler::zero, L);
1459 __ stop("method needs synchronization");
1460 __ bind(L);
1461 }
1462 #endif
1463 }
1464
1465 // start execution
1466 #ifdef ASSERT
1467 {
1468 Label L;
1469 const Address monitor_block_top (rbp,
1470 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1471 __ movptr(rax, monitor_block_top);
1472 __ cmpptr(rax, rsp);
1473 __ jcc(Assembler::equal, L);
1474 __ stop("broken stack frame setup in interpreter");
1475 __ bind(L);
1476 }
1477 #endif
1478
1479 // jvmti support
1480 __ notify_method_entry();
1481
1482 __ dispatch_next(vtos);
1483
1484 // invocation counter overflow
1485 if (inc_counter) {
1486 if (ProfileInterpreter) {
1487 // We have decided to profile this method in the interpreter
1488 __ bind(profile_method);
1489 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1490 __ set_method_data_pointer_for_bcp();
1491 __ get_method(rbx);
1492 __ jmp(profile_method_continue);
1493 }
1494 // Handle overflow of counter and compile method
1495 __ bind(invocation_counter_overflow);
1496 generate_counter_overflow(continue_after_compile);
1497 }
1498
1499 return entry_point;
1500 }
1501
1502 //-----------------------------------------------------------------------------
1503 // Exceptions
1504
1505 void TemplateInterpreterGenerator::generate_throw_exception() {
1506 // Entry point in previous activation (i.e., if the caller was
1507 // interpreted)
1508 Interpreter::_rethrow_exception_entry = __ pc();
1509 // Restore sp to interpreter_frame_last_sp even though we are going
1510 // to empty the expression stack for the exception processing.
1511 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1512 // rax: exception
1513 // rdx: return address/pc that threw exception
1514 __ restore_bcp(); // r13/rsi points to call/send
1515 __ restore_locals();
1516 LP64_ONLY(__ reinit_heapbase()); // restore r12 as heapbase.
1517 // Entry point for exceptions thrown within interpreter code
1518 Interpreter::_throw_exception_entry = __ pc();
1519 // expression stack is undefined here
1520 // rax: exception
1521 // r13/rsi: exception bcp
1522 __ verify_oop(rax);
1523 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1524 LP64_ONLY(__ mov(c_rarg1, rax));
1525
1526 // expression stack must be empty before entering the VM in case of
1527 // an exception
1528 __ empty_expression_stack();
1529 // find exception handler address and preserve exception oop
1530 __ call_VM(rdx,
1531 CAST_FROM_FN_PTR(address,
1532 InterpreterRuntime::exception_handler_for_exception),
1533 rarg);
1534 // rax: exception handler entry point
1535 // rdx: preserved exception oop
1536 // r13/rsi: bcp for exception handler
1537 __ push_ptr(rdx); // push exception which is now the only value on the stack
1538 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1539
1540 // If the exception is not handled in the current frame the frame is
1541 // removed and the exception is rethrown (i.e. exception
1542 // continuation is _rethrow_exception).
1543 //
1544 // Note: At this point the bci is still the bxi for the instruction
1545 // which caused the exception and the expression stack is
1546 // empty. Thus, for any VM calls at this point, GC will find a legal
1547 // oop map (with empty expression stack).
1548
1549 // In current activation
1550 // tos: exception
1551 // esi: exception bcp
1552
1553 //
1554 // JVMTI PopFrame support
1555 //
1556
1557 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1558 __ empty_expression_stack();
1559 // Set the popframe_processing bit in pending_popframe_condition
1560 // indicating that we are currently handling popframe, so that
1561 // call_VMs that may happen later do not trigger new popframe
1562 // handling cycles.
1563 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1564 NOT_LP64(__ get_thread(thread));
1565 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1566 __ orl(rdx, JavaThread::popframe_processing_bit);
1567 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1568
1569 {
1570 // Check to see whether we are returning to a deoptimized frame.
1571 // (The PopFrame call ensures that the caller of the popped frame is
1572 // either interpreted or compiled and deoptimizes it if compiled.)
1573 // In this case, we can't call dispatch_next() after the frame is
1574 // popped, but instead must save the incoming arguments and restore
1575 // them after deoptimization has occurred.
1576 //
1577 // Note that we don't compare the return PC against the
1578 // deoptimization blob's unpack entry because of the presence of
1579 // adapter frames in C2.
1580 Label caller_not_deoptimized;
1581 Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1582 __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1583 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1584 InterpreterRuntime::interpreter_contains), rarg);
1585 __ testl(rax, rax);
1586 __ jcc(Assembler::notZero, caller_not_deoptimized);
1587
1588 // Compute size of arguments for saving when returning to
1589 // deoptimized caller
1590 __ get_method(rax);
1591 __ movptr(rax, Address(rax, Method::const_offset()));
1592 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1593 size_of_parameters_offset())));
1594 __ shll(rax, Interpreter::logStackElementSize);
1595 __ restore_locals();
1596 __ subptr(rlocals, rax);
1597 __ addptr(rlocals, wordSize);
1598 // Save these arguments
1599 NOT_LP64(__ get_thread(thread));
1600 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1601 Deoptimization::
1602 popframe_preserve_args),
1603 thread, rax, rlocals);
1604
1605 __ remove_activation(vtos, rdx,
1606 /* throw_monitor_exception */ false,
1607 /* install_monitor_exception */ false,
1608 /* notify_jvmdi */ false);
1609
1610 // Inform deoptimization that it is responsible for restoring
1611 // these arguments
1612 NOT_LP64(__ get_thread(thread));
1613 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1614 JavaThread::popframe_force_deopt_reexecution_bit);
1615
1616 // Continue in deoptimization handler
1617 __ jmp(rdx);
1618
1619 __ bind(caller_not_deoptimized);
1620 }
1621
1622 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1623 /* throw_monitor_exception */ false,
1624 /* install_monitor_exception */ false,
1625 /* notify_jvmdi */ false);
1626
1627 // Finish with popframe handling
1628 // A previous I2C followed by a deoptimization might have moved the
1629 // outgoing arguments further up the stack. PopFrame expects the
1630 // mutations to those outgoing arguments to be preserved and other
1631 // constraints basically require this frame to look exactly as
1632 // though it had previously invoked an interpreted activation with
1633 // no space between the top of the expression stack (current
1634 // last_sp) and the top of stack. Rather than force deopt to
1635 // maintain this kind of invariant all the time we call a small
1636 // fixup routine to move the mutated arguments onto the top of our
1637 // expression stack if necessary.
1638 #ifndef _LP64
1639 __ mov(rax, rsp);
1640 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1641 __ get_thread(thread);
1642 // PC must point into interpreter here
1643 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1644 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1645 __ get_thread(thread);
1646 #else
1647 __ mov(c_rarg1, rsp);
1648 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1649 // PC must point into interpreter here
1650 __ set_last_Java_frame(noreg, rbp, __ pc());
1651 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1652 #endif
1653 __ reset_last_Java_frame(thread, true);
1654
1655 // Restore the last_sp and null it out
1656 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1657 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1658
1659 __ restore_bcp();
1660 __ restore_locals();
1661 // The method data pointer was incremented already during
1662 // call profiling. We have to restore the mdp for the current bcp.
1663 if (ProfileInterpreter) {
1664 __ set_method_data_pointer_for_bcp();
1665 }
1666
1667 // Clear the popframe condition flag
1668 NOT_LP64(__ get_thread(thread));
1669 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1670 JavaThread::popframe_inactive);
1671
1672 #if INCLUDE_JVMTI
1673 {
1674 Label L_done;
1675 const Register local0 = rlocals;
1676
1677 __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1678 __ jcc(Assembler::notEqual, L_done);
1679
1680 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1681 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1682
1683 __ get_method(rdx);
1684 __ movptr(rax, Address(local0, 0));
1685 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1686
1687 __ testptr(rax, rax);
1688 __ jcc(Assembler::zero, L_done);
1689
1690 __ movptr(Address(rbx, 0), rax);
1691 __ bind(L_done);
1692 }
1693 #endif // INCLUDE_JVMTI
1694
1695 __ dispatch_next(vtos);
1696 // end of PopFrame support
1697
1698 Interpreter::_remove_activation_entry = __ pc();
1699
1700 // preserve exception over this code sequence
1701 __ pop_ptr(rax);
1702 NOT_LP64(__ get_thread(thread));
1703 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1704 // remove the activation (without doing throws on illegalMonitorExceptions)
1705 __ remove_activation(vtos, rdx, false, true, false);
1706 // restore exception
1707 NOT_LP64(__ get_thread(thread));
1708 __ get_vm_result(rax, thread);
1709
1710 // In between activations - previous activation type unknown yet
1711 // compute continuation point - the continuation point expects the
1712 // following registers set up:
1713 //
1714 // rax: exception
1715 // rdx: return address/pc that threw exception
1716 // rsp: expression stack of caller
1717 // rbp: ebp of caller
1718 __ push(rax); // save exception
1719 __ push(rdx); // save return address
1720 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1721 SharedRuntime::exception_handler_for_return_address),
1722 thread, rdx);
1723 __ mov(rbx, rax); // save exception handler
1724 __ pop(rdx); // restore return address
1725 __ pop(rax); // restore exception
1726 // Note that an "issuing PC" is actually the next PC after the call
1727 __ jmp(rbx); // jump to exception
1728 // handler of caller
1729 }
1730
1731
1732 //
1733 // JVMTI ForceEarlyReturn support
1734 //
1735 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1736 address entry = __ pc();
1737
1738 __ restore_bcp();
1739 __ restore_locals();
1740 __ empty_expression_stack();
1741 __ load_earlyret_value(state); // 32 bits returns value in rdx, so don't reuse
1742
1743 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1744 NOT_LP64(__ get_thread(thread));
1745 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1746 Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1747
1748 // Clear the earlyret state
1749 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1750
1751 __ remove_activation(state, rsi,
1752 false, /* throw_monitor_exception */
1753 false, /* install_monitor_exception */
1754 true); /* notify_jvmdi */
1755 __ jmp(rsi);
1756
1757 return entry;
1758 } // end of ForceEarlyReturn support
1759
1760
1761 //-----------------------------------------------------------------------------
1762 // Helper for vtos entry point generation
1763
1764 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1765 address& bep,
1766 address& cep,
1767 address& sep,
1768 address& aep,
1769 address& iep,
1770 address& lep,
1771 address& fep,
1772 address& dep,
1773 address& vep) {
1774 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1775 Label L;
1776 aep = __ pc(); __ push_ptr(); __ jmp(L);
1777 #ifndef _LP64
1778 fep = __ pc(); __ push(ftos); __ jmp(L);
1779 dep = __ pc(); __ push(dtos); __ jmp(L);
1780 #else
1781 fep = __ pc(); __ push_f(xmm0); __ jmp(L);
1782 dep = __ pc(); __ push_d(xmm0); __ jmp(L);
1783 #endif // _LP64
1784 lep = __ pc(); __ push_l(); __ jmp(L);
1785 bep = cep = sep =
1786 iep = __ pc(); __ push_i();
1787 vep = __ pc();
1788 __ bind(L);
1789 generate_and_dispatch(t);
1790 }
1791
1792 //-----------------------------------------------------------------------------
1793
1794 // Non-product code
1795 #ifndef PRODUCT
1796
1797 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1798 address entry = __ pc();
1799
1800 #ifndef _LP64
1801 // prepare expression stack
1802 __ pop(rcx); // pop return address so expression stack is 'pure'
1803 __ push(state); // save tosca
1804
1805 // pass tosca registers as arguments & call tracer
1806 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1807 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1808 __ pop(state); // restore tosca
1809
1810 // return
1811 __ jmp(rcx);
1812 #else
1813 __ push(state);
1814 __ push(c_rarg0);
1815 __ push(c_rarg1);
1816 __ push(c_rarg2);
1817 __ push(c_rarg3);
1818 __ mov(c_rarg2, rax); // Pass itos
1819 #ifdef _WIN64
1820 __ movflt(xmm3, xmm0); // Pass ftos
1821 #endif
1822 __ call_VM(noreg,
1823 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1824 c_rarg1, c_rarg2, c_rarg3);
1825 __ pop(c_rarg3);
1826 __ pop(c_rarg2);
1827 __ pop(c_rarg1);
1828 __ pop(c_rarg0);
1829 __ pop(state);
1830 __ ret(0); // return from result handler
1831 #endif // _LP64
1832
1833 return entry;
1834 }
1835
1836 void TemplateInterpreterGenerator::count_bytecode() {
1837 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1838 }
1839
1840 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1841 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1842 }
1843
1844 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1845 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1846 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1847 __ orl(rbx,
1848 ((int) t->bytecode()) <<
1849 BytecodePairHistogram::log2_number_of_codes);
1850 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1851 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1852 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1853 }
1854
1855
1856 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1857 // Call a little run-time stub to avoid blow-up for each bytecode.
1858 // The run-time runtime saves the right registers, depending on
1859 // the tosca in-state for the given template.
1860
1861 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1862 "entry must have been generated");
1863 #ifndef _LP64
1864 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1865 #else
1866 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1867 __ andptr(rsp, -16); // align stack as required by ABI
1868 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1869 __ mov(rsp, r12); // restore sp
1870 __ reinit_heapbase();
1871 #endif // _LP64
1872 }
1873
1874
1875 void TemplateInterpreterGenerator::stop_interpreter_at() {
1876 Label L;
1877 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1878 StopInterpreterAt);
1879 __ jcc(Assembler::notEqual, L);
1880 __ int3();
1881 __ bind(L);
1882 }
1883 #endif // !PRODUCT