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 // The expression stack must be empty before entering the VM if an
108 // exception happened.
109 __ empty_expression_stack();
110
111 // Setup parameters.
112 // ??? convention: expect aberrant index in register ebx/rbx.
113 // Pass array to create more detailed exceptions.
114 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
115 __ call_VM(noreg,
116 CAST_FROM_FN_PTR(address,
117 InterpreterRuntime::
118 throw_ArrayIndexOutOfBoundsException),
119 rarg, rbx);
120 return entry;
121 }
122
123 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
124 address entry = __ pc();
125
126 // object is at TOS
127 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
128 __ pop(rarg);
129
130 // expression stack must be empty before entering the VM if an
131 // exception happened
132 __ empty_expression_stack();
133
134 __ call_VM(noreg,
135 CAST_FROM_FN_PTR(address,
136 InterpreterRuntime::
137 throw_ClassCastException),
138 rarg);
139 return entry;
140 }
141
142 address TemplateInterpreterGenerator::generate_exception_handler_common(
143 const char* name, const char* message, bool pass_oop) {
144 assert(!pass_oop || message == NULL, "either oop or message but not both");
145 address entry = __ pc();
146
147 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
148 Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
149
150 if (pass_oop) {
151 // object is at TOS
152 __ pop(rarg2);
153 }
154 // expression stack must be empty before entering the VM if an
155 // exception happened
156 __ empty_expression_stack();
157 // setup parameters
158 __ lea(rarg, ExternalAddress((address)name));
159 if (pass_oop) {
160 __ call_VM(rax, CAST_FROM_FN_PTR(address,
161 InterpreterRuntime::
162 create_klass_exception),
163 rarg, rarg2);
164 } else {
165 __ lea(rarg2, ExternalAddress((address)message));
166 __ call_VM(rax,
167 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
168 rarg, rarg2);
169 }
170 // throw exception
171 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
172 return entry;
173 }
174
175 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
176 address entry = __ pc();
177
178 #ifndef _LP64
179 #ifdef COMPILER2
180 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
181 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
182 for (int i = 1; i < 8; i++) {
183 __ ffree(i);
184 }
185 } else if (UseSSE < 2) {
186 __ empty_FPU_stack();
187 }
188 #endif // COMPILER2
189 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
190 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
191 } else {
192 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
193 }
194
195 if (state == ftos) {
196 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
197 } else if (state == dtos) {
198 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
199 }
200 #endif // _LP64
201
202 // Restore stack bottom in case i2c adjusted stack
203 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
204 // and NULL it as marker that esp is now tos until next java call
205 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
206
207 __ restore_bcp();
208 __ restore_locals();
209
210 if (state == atos) {
211 Register mdp = rbx;
212 Register tmp = rcx;
213 __ profile_return_type(mdp, rax, tmp);
214 }
215
216 const Register cache = rbx;
217 const Register index = rcx;
218 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
219
220 const Register flags = cache;
221 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
222 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
223 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
224
225 const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
226 if (JvmtiExport::can_pop_frame()) {
227 NOT_LP64(__ get_thread(java_thread));
228 __ check_and_handle_popframe(java_thread);
229 }
230 if (JvmtiExport::can_force_early_return()) {
231 NOT_LP64(__ get_thread(java_thread));
232 __ check_and_handle_earlyret(java_thread);
233 }
234
235 __ dispatch_next(state, step);
236
237 return entry;
238 }
239
240
241 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
242 address entry = __ pc();
243
244 #ifndef _LP64
245 if (state == ftos) {
246 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
247 } else if (state == dtos) {
248 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
249 }
250 #endif // _LP64
251
252 // NULL last_sp until next java call
253 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
254 __ restore_bcp();
255 __ restore_locals();
256 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
257 NOT_LP64(__ get_thread(thread));
258 #if INCLUDE_JVMCI
259 // Check if we need to take lock at entry of synchronized method. This can
260 // only occur on method entry so emit it only for vtos with step 0.
261 if ((EnableJVMCI || UseAOT) && state == vtos && step == 0) {
262 Label L;
263 __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
264 __ jcc(Assembler::zero, L);
265 // Clear flag.
266 __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
267 // Satisfy calling convention for lock_method().
268 __ get_method(rbx);
269 // Take lock.
270 lock_method();
271 __ bind(L);
272 } else {
273 #ifdef ASSERT
274 if (EnableJVMCI) {
275 Label L;
276 __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
277 __ jccb(Assembler::zero, L);
278 __ stop("unexpected pending monitor in deopt entry");
279 __ bind(L);
280 }
281 #endif
282 }
283 #endif
284 // handle exceptions
285 {
286 Label L;
287 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
288 __ jcc(Assembler::zero, L);
289 __ call_VM(noreg,
290 CAST_FROM_FN_PTR(address,
291 InterpreterRuntime::throw_pending_exception));
292 __ should_not_reach_here();
293 __ bind(L);
294 }
295 if (continuation == NULL) {
296 __ dispatch_next(state, step);
297 } else {
298 __ jump_to_entry(continuation);
299 }
300 return entry;
301 }
302
303 address TemplateInterpreterGenerator::generate_result_handler_for(
304 BasicType type) {
305 address entry = __ pc();
306 switch (type) {
307 case T_BOOLEAN: __ c2bool(rax); break;
308 #ifndef _LP64
309 case T_CHAR : __ andptr(rax, 0xFFFF); break;
310 #else
311 case T_CHAR : __ movzwl(rax, rax); break;
312 #endif // _LP64
313 case T_BYTE : __ sign_extend_byte(rax); break;
314 case T_SHORT : __ sign_extend_short(rax); break;
315 case T_INT : /* nothing to do */ break;
316 case T_LONG : /* nothing to do */ break;
317 case T_VOID : /* nothing to do */ break;
318 #ifndef _LP64
319 case T_DOUBLE :
320 case T_FLOAT :
321 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
322 __ pop(t); // remove return address first
323 // Must return a result for interpreter or compiler. In SSE
324 // mode, results are returned in xmm0 and the FPU stack must
325 // be empty.
326 if (type == T_FLOAT && UseSSE >= 1) {
327 // Load ST0
328 __ fld_d(Address(rsp, 0));
329 // Store as float and empty fpu stack
330 __ fstp_s(Address(rsp, 0));
331 // and reload
332 __ movflt(xmm0, Address(rsp, 0));
333 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
334 __ movdbl(xmm0, Address(rsp, 0));
335 } else {
336 // restore ST0
337 __ fld_d(Address(rsp, 0));
338 }
339 // and pop the temp
340 __ addptr(rsp, 2 * wordSize);
341 __ push(t); // restore return address
342 }
343 break;
344 #else
345 case T_FLOAT : /* nothing to do */ break;
346 case T_DOUBLE : /* nothing to do */ break;
347 #endif // _LP64
348
349 case T_OBJECT :
350 // retrieve result from frame
351 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
352 // and verify it
353 __ verify_oop(rax);
354 break;
355 default : ShouldNotReachHere();
356 }
357 __ ret(0); // return from result handler
358 return entry;
359 }
360
361 address TemplateInterpreterGenerator::generate_safept_entry_for(
362 TosState state,
363 address runtime_entry) {
364 address entry = __ pc();
365 __ push(state);
366 __ call_VM(noreg, runtime_entry);
367 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
368 return entry;
369 }
370
371
372
373 // Helpers for commoning out cases in the various type of method entries.
374 //
375
376
377 // increment invocation count & check for overflow
378 //
379 // Note: checking for negative value instead of overflow
380 // so we have a 'sticky' overflow test
381 //
382 // rbx: method
383 // rcx: invocation counter
384 //
385 void TemplateInterpreterGenerator::generate_counter_incr(
386 Label* overflow,
387 Label* profile_method,
388 Label* profile_method_continue) {
389 Label done;
390 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
391 if (TieredCompilation) {
392 int increment = InvocationCounter::count_increment;
393 Label no_mdo;
394 if (ProfileInterpreter) {
395 // Are we profiling?
396 __ movptr(rax, Address(rbx, Method::method_data_offset()));
397 __ testptr(rax, rax);
398 __ jccb(Assembler::zero, no_mdo);
399 // Increment counter in the MDO
400 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
401 in_bytes(InvocationCounter::counter_offset()));
402 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
403 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
404 __ jmp(done);
405 }
406 __ bind(no_mdo);
407 // Increment counter in MethodCounters
408 const Address invocation_counter(rax,
409 MethodCounters::invocation_counter_offset() +
410 InvocationCounter::counter_offset());
411 __ get_method_counters(rbx, rax, done);
412 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
413 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
414 false, Assembler::zero, overflow);
415 __ bind(done);
416 } else { // not TieredCompilation
417 const Address backedge_counter(rax,
418 MethodCounters::backedge_counter_offset() +
419 InvocationCounter::counter_offset());
420 const Address invocation_counter(rax,
421 MethodCounters::invocation_counter_offset() +
422 InvocationCounter::counter_offset());
423
424 __ get_method_counters(rbx, rax, done);
425
426 if (ProfileInterpreter) {
427 __ incrementl(Address(rax,
428 MethodCounters::interpreter_invocation_counter_offset()));
429 }
430 // Update standard invocation counters
431 __ movl(rcx, invocation_counter);
432 __ incrementl(rcx, InvocationCounter::count_increment);
433 __ movl(invocation_counter, rcx); // save invocation count
434
435 __ movl(rax, backedge_counter); // load backedge counter
436 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
437
438 __ addl(rcx, rax); // add both counters
439
440 // profile_method is non-null only for interpreted method so
441 // profile_method != NULL == !native_call
442
443 if (ProfileInterpreter && profile_method != NULL) {
444 // Test to see if we should create a method data oop
445 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
446 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
447 __ jcc(Assembler::less, *profile_method_continue);
448
449 // if no method data exists, go to profile_method
450 __ test_method_data_pointer(rax, *profile_method);
451 }
452
453 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
454 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
455 __ jcc(Assembler::aboveEqual, *overflow);
456 __ bind(done);
457 }
458 }
459
460 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
461
462 // Asm interpreter on entry
463 // r14/rdi - locals
464 // r13/rsi - bcp
465 // rbx - method
466 // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
467 // rbp - interpreter frame
468
469 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
470 // Everything as it was on entry
471 // rdx is not restored. Doesn't appear to really be set.
472
473 // InterpreterRuntime::frequency_counter_overflow takes two
474 // arguments, the first (thread) is passed by call_VM, the second
475 // indicates if the counter overflow occurs at a backwards branch
476 // (NULL bcp). We pass zero for it. The call returns the address
477 // of the verified entry point for the method or NULL if the
478 // compilation did not complete (either went background or bailed
479 // out).
480 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
481 __ movl(rarg, 0);
482 __ call_VM(noreg,
483 CAST_FROM_FN_PTR(address,
484 InterpreterRuntime::frequency_counter_overflow),
485 rarg);
486
487 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
488 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
489 // and jump to the interpreted entry.
490 __ jmp(do_continue, relocInfo::none);
491 }
492
493 // See if we've got enough room on the stack for locals plus overhead below
494 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
495 // without going through the signal handler, i.e., reserved and yellow zones
496 // will not be made usable. The shadow zone must suffice to handle the
497 // overflow.
498 // The expression stack grows down incrementally, so the normal guard
499 // page mechanism will work for that.
500 //
501 // NOTE: Since the additional locals are also always pushed (wasn't
502 // obvious in generate_fixed_frame) so the guard should work for them
503 // too.
504 //
505 // Args:
506 // rdx: number of additional locals this frame needs (what we must check)
507 // rbx: Method*
508 //
509 // Kills:
510 // rax
511 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
512
513 // monitor entry size: see picture of stack in frame_x86.hpp
514 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
515
516 // total overhead size: entry_size + (saved rbp through expr stack
517 // bottom). be sure to change this if you add/subtract anything
518 // to/from the overhead area
519 const int overhead_size =
520 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
521
522 const int page_size = os::vm_page_size();
523
524 Label after_frame_check;
525
526 // see if the frame is greater than one page in size. If so,
527 // then we need to verify there is enough stack space remaining
528 // for the additional locals.
529 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
530 __ jcc(Assembler::belowEqual, after_frame_check);
531
532 // compute rsp as if this were going to be the last frame on
533 // the stack before the red zone
534
535 Label after_frame_check_pop;
536 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
537 #ifndef _LP64
538 __ push(thread);
539 __ get_thread(thread);
540 #endif
541
542 const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
543
544 // locals + overhead, in bytes
545 __ mov(rax, rdx);
546 __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
547 __ addptr(rax, overhead_size);
548
549 #ifdef ASSERT
550 Label limit_okay;
551 // Verify that thread stack overflow limit is non-zero.
552 __ cmpptr(stack_limit, (int32_t)NULL_WORD);
553 __ jcc(Assembler::notEqual, limit_okay);
554 __ stop("stack overflow limit is zero");
555 __ bind(limit_okay);
556 #endif
557
558 // Add locals/frame size to stack limit.
559 __ addptr(rax, stack_limit);
560
561 // Check against the current stack bottom.
562 __ cmpptr(rsp, rax);
563
564 __ jcc(Assembler::above, after_frame_check_pop);
565 NOT_LP64(__ pop(rsi)); // get saved bcp
566
567 // Restore sender's sp as SP. This is necessary if the sender's
568 // frame is an extended compiled frame (see gen_c2i_adapter())
569 // and safer anyway in case of JSR292 adaptations.
570
571 __ pop(rax); // return address must be moved if SP is changed
572 __ mov(rsp, rbcp);
573 __ push(rax);
574
575 // Note: the restored frame is not necessarily interpreted.
576 // Use the shared runtime version of the StackOverflowError.
577 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
578 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
579 // all done with frame size check
580 __ bind(after_frame_check_pop);
581 NOT_LP64(__ pop(rsi));
582
583 // all done with frame size check
584 __ bind(after_frame_check);
585 }
586
587 // Allocate monitor and lock method (asm interpreter)
588 //
589 // Args:
590 // rbx: Method*
591 // r14/rdi: locals
592 //
593 // Kills:
594 // rax
595 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
596 // rscratch1, rscratch2 (scratch regs)
597 void TemplateInterpreterGenerator::lock_method() {
598 // synchronize method
599 const Address access_flags(rbx, Method::access_flags_offset());
600 const Address monitor_block_top(
601 rbp,
602 frame::interpreter_frame_monitor_block_top_offset * wordSize);
603 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
604
605 #ifdef ASSERT
606 {
607 Label L;
608 __ movl(rax, access_flags);
609 __ testl(rax, JVM_ACC_SYNCHRONIZED);
610 __ jcc(Assembler::notZero, L);
611 __ stop("method doesn't need synchronization");
612 __ bind(L);
613 }
614 #endif // ASSERT
615
616 // get synchronization object
617 {
618 Label done;
619 __ movl(rax, access_flags);
620 __ testl(rax, JVM_ACC_STATIC);
621 // get receiver (assume this is frequent case)
622 __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
623 __ jcc(Assembler::zero, done);
624 __ load_mirror(rax, rbx);
625
626 #ifdef ASSERT
627 {
628 Label L;
629 __ testptr(rax, rax);
630 __ jcc(Assembler::notZero, L);
631 __ stop("synchronization object is NULL");
632 __ bind(L);
633 }
634 #endif // ASSERT
635
636 __ bind(done);
637 __ resolve_for_write(OOP_NOT_NULL, rax);
638 }
639
640 // add space for monitor & lock
641 __ subptr(rsp, entry_size); // add space for a monitor entry
642 __ movptr(monitor_block_top, rsp); // set new monitor block top
643 // store object
644 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
645 const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
646 __ movptr(lockreg, rsp); // object address
647 __ lock_object(lockreg);
648 }
649
650 // Generate a fixed interpreter frame. This is identical setup for
651 // interpreted methods and for native methods hence the shared code.
652 //
653 // Args:
654 // rax: return address
655 // rbx: Method*
656 // r14/rdi: pointer to locals
657 // r13/rsi: sender sp
658 // rdx: cp cache
659 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
660 // initialize fixed part of activation frame
661 __ push(rax); // save return address
662 __ enter(); // save old & set new rbp
663 __ push(rbcp); // set sender sp
664 __ push((int)NULL_WORD); // leave last_sp as null
665 __ movptr(rbcp, Address(rbx, Method::const_offset())); // get ConstMethod*
666 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
667 __ push(rbx); // save Method*
668 // Get mirror and store it in the frame as GC root for this Method*
669 __ load_mirror(rdx, rbx);
670 __ push(rdx);
671 if (ProfileInterpreter) {
672 Label method_data_continue;
673 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
674 __ testptr(rdx, rdx);
675 __ jcc(Assembler::zero, method_data_continue);
676 __ addptr(rdx, in_bytes(MethodData::data_offset()));
677 __ bind(method_data_continue);
678 __ push(rdx); // set the mdp (method data pointer)
679 } else {
680 __ push(0);
681 }
682
683 __ movptr(rdx, Address(rbx, Method::const_offset()));
684 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
685 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
686 __ push(rdx); // set constant pool cache
687 __ push(rlocals); // set locals pointer
688 if (native_call) {
689 __ push(0); // no bcp
690 } else {
691 __ push(rbcp); // set bcp
692 }
693 __ push(0); // reserve word for pointer to expression stack bottom
694 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
695 }
696
697 // End of helpers
698
699 // Method entry for java.lang.ref.Reference.get.
700 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
701 // Code: _aload_0, _getfield, _areturn
702 // parameter size = 1
703 //
704 // The code that gets generated by this routine is split into 2 parts:
705 // 1. The "intrinsified" code performing an ON_WEAK_OOP_REF load,
706 // 2. The slow path - which is an expansion of the regular method entry.
707 //
708 // Notes:-
709 // * An intrinsic is always executed, where an ON_WEAK_OOP_REF load is performed.
710 // * We may jump to the slow path iff the receiver is null. If the
711 // Reference object is null then we no longer perform an ON_WEAK_OOP_REF load
712 // Thus we can use the regular method entry code to generate the NPE.
713 //
714 // rbx: Method*
715
716 // r13: senderSP must preserve for slow path, set SP to it on fast path
717
718 address entry = __ pc();
719
720 const int referent_offset = java_lang_ref_Reference::referent_offset;
721 guarantee(referent_offset > 0, "referent offset not initialized");
722
723 Label slow_path;
724 // rbx: method
725
726 // Check if local 0 != NULL
727 // If the receiver is null then it is OK to jump to the slow path.
728 __ movptr(rax, Address(rsp, wordSize));
729
730 __ testptr(rax, rax);
731 __ jcc(Assembler::zero, slow_path);
732
733 // rax: local 0
734 // rbx: method (but can be used as scratch now)
735 // rdx: scratch
736 // rdi: scratch
737
738 // Preserve the sender sp in case the load barrier
739 // calls the runtime
740 NOT_LP64(__ push(rsi));
741
742 // Load the value of the referent field.
743 const Address field_address(rax, referent_offset);
744 __ load_heap_oop(rax, field_address, /*tmp1*/ rbx, /*tmp_thread*/ rdx, ON_WEAK_OOP_REF);
745
746 // _areturn
747 const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
748 NOT_LP64(__ pop(rsi)); // get sender sp
749 __ pop(rdi); // get return address
750 __ mov(rsp, sender_sp); // set sp to sender sp
751 __ jmp(rdi);
752 __ ret(0);
753
754 // generate a vanilla interpreter entry as the slow path
755 __ bind(slow_path);
756 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
757 return entry;
758 }
759
760 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
761 // Quick & dirty stack overflow checking: bang the stack & handle trap.
762 // Note that we do the banging after the frame is setup, since the exception
763 // handling code expects to find a valid interpreter frame on the stack.
764 // Doing the banging earlier fails if the caller frame is not an interpreter
765 // frame.
766 // (Also, the exception throwing code expects to unlock any synchronized
767 // method receiever, so do the banging after locking the receiver.)
768
769 // Bang each page in the shadow zone. We can't assume it's been done for
770 // an interpreter frame with greater than a page of locals, so each page
771 // needs to be checked. Only true for non-native.
772 if (UseStackBanging) {
773 const int page_size = os::vm_page_size();
774 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
775 const int start_page = native_call ? n_shadow_pages : 1;
776 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
777 __ bang_stack_with_offset(pages*page_size);
778 }
779 }
780 }
781
782 // Interpreter stub for calling a native method. (asm interpreter)
783 // This sets up a somewhat different looking stack for calling the
784 // native method than the typical interpreter frame setup.
785 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
786 // determine code generation flags
787 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
788
789 // rbx: Method*
790 // rbcp: sender sp
791
792 address entry_point = __ pc();
793
794 const Address constMethod (rbx, Method::const_offset());
795 const Address access_flags (rbx, Method::access_flags_offset());
796 const Address size_of_parameters(rcx, ConstMethod::
797 size_of_parameters_offset());
798
799
800 // get parameter size (always needed)
801 __ movptr(rcx, constMethod);
802 __ load_unsigned_short(rcx, size_of_parameters);
803
804 // native calls don't need the stack size check since they have no
805 // expression stack and the arguments are already on the stack and
806 // we only add a handful of words to the stack
807
808 // rbx: Method*
809 // rcx: size of parameters
810 // rbcp: sender sp
811 __ pop(rax); // get return address
812
813 // for natives the size of locals is zero
814
815 // compute beginning of parameters
816 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
817
818 // add 2 zero-initialized slots for native calls
819 // initialize result_handler slot
820 __ push((int) NULL_WORD);
821 // slot for oop temp
822 // (static native method holder mirror/jni oop result)
823 __ push((int) NULL_WORD);
824
825 // initialize fixed part of activation frame
826 generate_fixed_frame(true);
827
828 // make sure method is native & not abstract
829 #ifdef ASSERT
830 __ movl(rax, access_flags);
831 {
832 Label L;
833 __ testl(rax, JVM_ACC_NATIVE);
834 __ jcc(Assembler::notZero, L);
835 __ stop("tried to execute non-native method as native");
836 __ bind(L);
837 }
838 {
839 Label L;
840 __ testl(rax, JVM_ACC_ABSTRACT);
841 __ jcc(Assembler::zero, L);
842 __ stop("tried to execute abstract method in interpreter");
843 __ bind(L);
844 }
845 #endif
846
847 // Since at this point in the method invocation the exception handler
848 // would try to exit the monitor of synchronized methods which hasn't
849 // been entered yet, we set the thread local variable
850 // _do_not_unlock_if_synchronized to true. The remove_activation will
851 // check this flag.
852
853 const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
854 NOT_LP64(__ get_thread(thread1));
855 const Address do_not_unlock_if_synchronized(thread1,
856 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
857 __ movbool(do_not_unlock_if_synchronized, true);
858
859 // increment invocation count & check for overflow
860 Label invocation_counter_overflow;
861 if (inc_counter) {
862 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
863 }
864
865 Label continue_after_compile;
866 __ bind(continue_after_compile);
867
868 bang_stack_shadow_pages(true);
869
870 // reset the _do_not_unlock_if_synchronized flag
871 NOT_LP64(__ get_thread(thread1));
872 __ movbool(do_not_unlock_if_synchronized, false);
873
874 // check for synchronized methods
875 // Must happen AFTER invocation_counter check and stack overflow check,
876 // so method is not locked if overflows.
877 if (synchronized) {
878 lock_method();
879 } else {
880 // no synchronization necessary
881 #ifdef ASSERT
882 {
883 Label L;
884 __ movl(rax, access_flags);
885 __ testl(rax, JVM_ACC_SYNCHRONIZED);
886 __ jcc(Assembler::zero, L);
887 __ stop("method needs synchronization");
888 __ bind(L);
889 }
890 #endif
891 }
892
893 // start execution
894 #ifdef ASSERT
895 {
896 Label L;
897 const Address monitor_block_top(rbp,
898 frame::interpreter_frame_monitor_block_top_offset * wordSize);
899 __ movptr(rax, monitor_block_top);
900 __ cmpptr(rax, rsp);
901 __ jcc(Assembler::equal, L);
902 __ stop("broken stack frame setup in interpreter");
903 __ bind(L);
904 }
905 #endif
906
907 // jvmti support
908 __ notify_method_entry();
909
910 // work registers
911 const Register method = rbx;
912 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
913 const Register t = NOT_LP64(rcx) LP64_ONLY(r11);
914
915 // allocate space for parameters
916 __ get_method(method);
917 __ movptr(t, Address(method, Method::const_offset()));
918 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
919
920 #ifndef _LP64
921 __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
922 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
923 __ subptr(rsp, t);
924 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
925 #else
926 __ shll(t, Interpreter::logStackElementSize);
927
928 __ subptr(rsp, t);
929 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
930 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
931 #endif // _LP64
932
933 // get signature handler
934 {
935 Label L;
936 __ movptr(t, Address(method, Method::signature_handler_offset()));
937 __ testptr(t, t);
938 __ jcc(Assembler::notZero, L);
939 __ call_VM(noreg,
940 CAST_FROM_FN_PTR(address,
941 InterpreterRuntime::prepare_native_call),
942 method);
943 __ get_method(method);
944 __ movptr(t, Address(method, Method::signature_handler_offset()));
945 __ bind(L);
946 }
947
948 // call signature handler
949 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
950 "adjust this code");
951 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
952 "adjust this code");
953 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
954 "adjust this code");
955
956 // The generated handlers do not touch RBX (the method oop).
957 // However, large signatures cannot be cached and are generated
958 // each time here. The slow-path generator can do a GC on return,
959 // so we must reload it after the call.
960 __ call(t);
961 __ get_method(method); // slow path can do a GC, reload RBX
962
963
964 // result handler is in rax
965 // set result handler
966 __ movptr(Address(rbp,
967 (frame::interpreter_frame_result_handler_offset) * wordSize),
968 rax);
969
970 // pass mirror handle if static call
971 {
972 Label L;
973 __ movl(t, Address(method, Method::access_flags_offset()));
974 __ testl(t, JVM_ACC_STATIC);
975 __ jcc(Assembler::zero, L);
976 // get mirror
977 __ load_mirror(t, method, rax);
978 // copy mirror into activation frame
979 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
980 t);
981 // pass handle to mirror
982 #ifndef _LP64
983 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
984 __ movptr(Address(rsp, wordSize), t);
985 #else
986 __ lea(c_rarg1,
987 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
988 #endif // _LP64
989 __ bind(L);
990 }
991
992 // get native function entry point
993 {
994 Label L;
995 __ movptr(rax, Address(method, Method::native_function_offset()));
996 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
997 __ cmpptr(rax, unsatisfied.addr());
998 __ jcc(Assembler::notEqual, L);
999 __ call_VM(noreg,
1000 CAST_FROM_FN_PTR(address,
1001 InterpreterRuntime::prepare_native_call),
1002 method);
1003 __ get_method(method);
1004 __ movptr(rax, Address(method, Method::native_function_offset()));
1005 __ bind(L);
1006 }
1007
1008 // pass JNIEnv
1009 #ifndef _LP64
1010 __ get_thread(thread);
1011 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1012 __ movptr(Address(rsp, 0), t);
1013
1014 // set_last_Java_frame_before_call
1015 // It is enough that the pc()
1016 // points into the right code segment. It does not have to be the correct return pc.
1017 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1018 #else
1019 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1020
1021 // It is enough that the pc() points into the right code
1022 // segment. It does not have to be the correct return pc.
1023 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1024 #endif // _LP64
1025
1026 // change thread state
1027 #ifdef ASSERT
1028 {
1029 Label L;
1030 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1031 __ cmpl(t, _thread_in_Java);
1032 __ jcc(Assembler::equal, L);
1033 __ stop("Wrong thread state in native stub");
1034 __ bind(L);
1035 }
1036 #endif
1037
1038 // Change state to native
1039
1040 __ movl(Address(thread, JavaThread::thread_state_offset()),
1041 _thread_in_native);
1042
1043 // Call the native method.
1044 __ call(rax);
1045 // 32: result potentially in rdx:rax or ST0
1046 // 64: result potentially in rax or xmm0
1047
1048 // Verify or restore cpu control state after JNI call
1049 __ restore_cpu_control_state_after_jni();
1050
1051 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1052 // in order to extract the result of a method call. If the order of these
1053 // pushes change or anything else is added to the stack then the code in
1054 // interpreter_frame_result must also change.
1055
1056 #ifndef _LP64
1057 // save potential result in ST(0) & rdx:rax
1058 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1059 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1060 // It is safe to do this push because state is _thread_in_native and return address will be found
1061 // via _last_native_pc and not via _last_jave_sp
1062
1063 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1064 // If the order changes or anything else is added to the stack the code in
1065 // interpreter_frame_result will have to be changed.
1066
1067 { Label L;
1068 Label push_double;
1069 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1070 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1071 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1072 float_handler.addr());
1073 __ jcc(Assembler::equal, push_double);
1074 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1075 double_handler.addr());
1076 __ jcc(Assembler::notEqual, L);
1077 __ bind(push_double);
1078 __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1079 __ bind(L);
1080 }
1081 #else
1082 __ push(dtos);
1083 #endif // _LP64
1084
1085 __ push(ltos);
1086
1087 // change thread state
1088 NOT_LP64(__ get_thread(thread));
1089 __ movl(Address(thread, JavaThread::thread_state_offset()),
1090 _thread_in_native_trans);
1091
1092 if (os::is_MP()) {
1093 if (UseMembar) {
1094 // Force this write out before the read below
1095 __ membar(Assembler::Membar_mask_bits(
1096 Assembler::LoadLoad | Assembler::LoadStore |
1097 Assembler::StoreLoad | Assembler::StoreStore));
1098 } else {
1099 // Write serialization page so VM thread can do a pseudo remote membar.
1100 // We use the current thread pointer to calculate a thread specific
1101 // offset to write to within the page. This minimizes bus traffic
1102 // due to cache line collision.
1103 __ serialize_memory(thread, rcx);
1104 }
1105 }
1106
1107 #ifndef _LP64
1108 if (AlwaysRestoreFPU) {
1109 // Make sure the control word is correct.
1110 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1111 }
1112 #endif // _LP64
1113
1114 // check for safepoint operation in progress and/or pending suspend requests
1115 {
1116 Label Continue;
1117 Label slow_path;
1118
1119 #ifndef _LP64
1120 __ safepoint_poll(slow_path, thread, noreg);
1121 #else
1122 __ safepoint_poll(slow_path, r15_thread, rscratch1);
1123 #endif
1124
1125 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1126 __ jcc(Assembler::equal, Continue);
1127 __ bind(slow_path);
1128
1129 // Don't use call_VM as it will see a possible pending exception
1130 // and forward it and never return here preventing us from
1131 // clearing _last_native_pc down below. Also can't use
1132 // call_VM_leaf either as it will check to see if r13 & r14 are
1133 // preserved and correspond to the bcp/locals pointers. So we do a
1134 // runtime call by hand.
1135 //
1136 #ifndef _LP64
1137 __ push(thread);
1138 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1139 JavaThread::check_special_condition_for_native_trans)));
1140 __ increment(rsp, wordSize);
1141 __ get_thread(thread);
1142 #else
1143 __ mov(c_rarg0, r15_thread);
1144 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1145 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1146 __ andptr(rsp, -16); // align stack as required by ABI
1147 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1148 __ mov(rsp, r12); // restore sp
1149 __ reinit_heapbase();
1150 #endif // _LP64
1151 __ bind(Continue);
1152 }
1153
1154 // change thread state
1155 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1156
1157 // reset_last_Java_frame
1158 __ reset_last_Java_frame(thread, true);
1159
1160 if (CheckJNICalls) {
1161 // clear_pending_jni_exception_check
1162 __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1163 }
1164
1165 // reset handle block
1166 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1167 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1168
1169 // If result is an oop unbox and store it in frame where gc will see it
1170 // and result handler will pick it up
1171
1172 {
1173 Label no_oop, not_weak, store_result;
1174 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1175 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1176 __ jcc(Assembler::notEqual, no_oop);
1177 // retrieve result
1178 __ pop(ltos);
1179 // Unbox oop result, e.g. JNIHandles::resolve value.
1180 __ resolve_jobject(rax /* value */,
1181 thread /* thread */,
1182 t /* tmp */);
1183 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1184 // keep stack depth as expected by pushing oop which will eventually be discarded
1185 __ push(ltos);
1186 __ bind(no_oop);
1187 }
1188
1189
1190 {
1191 Label no_reguard;
1192 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1193 JavaThread::stack_guard_yellow_reserved_disabled);
1194 __ jcc(Assembler::notEqual, no_reguard);
1195
1196 __ pusha(); // XXX only save smashed registers
1197 #ifndef _LP64
1198 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1199 __ popa();
1200 #else
1201 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1202 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1203 __ andptr(rsp, -16); // align stack as required by ABI
1204 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1205 __ mov(rsp, r12); // restore sp
1206 __ popa(); // XXX only restore smashed registers
1207 __ reinit_heapbase();
1208 #endif // _LP64
1209
1210 __ bind(no_reguard);
1211 }
1212
1213
1214 // The method register is junk from after the thread_in_native transition
1215 // until here. Also can't call_VM until the bcp has been
1216 // restored. Need bcp for throwing exception below so get it now.
1217 __ get_method(method);
1218
1219 // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1220 __ movptr(rbcp, Address(method, Method::const_offset())); // get ConstMethod*
1221 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
1222
1223 // handle exceptions (exception handling will handle unlocking!)
1224 {
1225 Label L;
1226 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1227 __ jcc(Assembler::zero, L);
1228 // Note: At some point we may want to unify this with the code
1229 // used in call_VM_base(); i.e., we should use the
1230 // StubRoutines::forward_exception code. For now this doesn't work
1231 // here because the rsp is not correctly set at this point.
1232 __ MacroAssembler::call_VM(noreg,
1233 CAST_FROM_FN_PTR(address,
1234 InterpreterRuntime::throw_pending_exception));
1235 __ should_not_reach_here();
1236 __ bind(L);
1237 }
1238
1239 // do unlocking if necessary
1240 {
1241 Label L;
1242 __ movl(t, Address(method, Method::access_flags_offset()));
1243 __ testl(t, JVM_ACC_SYNCHRONIZED);
1244 __ jcc(Assembler::zero, L);
1245 // the code below should be shared with interpreter macro
1246 // assembler implementation
1247 {
1248 Label unlock;
1249 // BasicObjectLock will be first in list, since this is a
1250 // synchronized method. However, need to check that the object
1251 // has not been unlocked by an explicit monitorexit bytecode.
1252 const Address monitor(rbp,
1253 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1254 wordSize - (int)sizeof(BasicObjectLock)));
1255
1256 const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1257
1258 // monitor expect in c_rarg1 for slow unlock path
1259 __ lea(regmon, monitor); // address of first monitor
1260
1261 __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1262 __ testptr(t, t);
1263 __ jcc(Assembler::notZero, unlock);
1264
1265 // Entry already unlocked, need to throw exception
1266 __ MacroAssembler::call_VM(noreg,
1267 CAST_FROM_FN_PTR(address,
1268 InterpreterRuntime::throw_illegal_monitor_state_exception));
1269 __ should_not_reach_here();
1270
1271 __ bind(unlock);
1272 __ unlock_object(regmon);
1273 }
1274 __ bind(L);
1275 }
1276
1277 // jvmti support
1278 // Note: This must happen _after_ handling/throwing any exceptions since
1279 // the exception handler code notifies the runtime of method exits
1280 // too. If this happens before, method entry/exit notifications are
1281 // not properly paired (was bug - gri 11/22/99).
1282 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1283
1284 // restore potential result in edx:eax, call result handler to
1285 // restore potential result in ST0 & handle result
1286
1287 __ pop(ltos);
1288 LP64_ONLY( __ pop(dtos));
1289
1290 __ movptr(t, Address(rbp,
1291 (frame::interpreter_frame_result_handler_offset) * wordSize));
1292 __ call(t);
1293
1294 // remove activation
1295 __ movptr(t, Address(rbp,
1296 frame::interpreter_frame_sender_sp_offset *
1297 wordSize)); // get sender sp
1298 __ leave(); // remove frame anchor
1299 __ pop(rdi); // get return address
1300 __ mov(rsp, t); // set sp to sender sp
1301 __ jmp(rdi);
1302
1303 if (inc_counter) {
1304 // Handle overflow of counter and compile method
1305 __ bind(invocation_counter_overflow);
1306 generate_counter_overflow(continue_after_compile);
1307 }
1308
1309 return entry_point;
1310 }
1311
1312 // Abstract method entry
1313 // Attempt to execute abstract method. Throw exception
1314 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1315
1316 address entry_point = __ pc();
1317
1318 // abstract method entry
1319
1320 // pop return address, reset last_sp to NULL
1321 __ empty_expression_stack();
1322 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1323 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1324
1325 // throw exception
1326 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod), rbx);
1327 // the call_VM checks for exception, so we should never return here.
1328 __ should_not_reach_here();
1329
1330 return entry_point;
1331 }
1332
1333 //
1334 // Generic interpreted method entry to (asm) interpreter
1335 //
1336 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1337 // determine code generation flags
1338 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1339
1340 // ebx: Method*
1341 // rbcp: sender sp
1342 address entry_point = __ pc();
1343
1344 const Address constMethod(rbx, Method::const_offset());
1345 const Address access_flags(rbx, Method::access_flags_offset());
1346 const Address size_of_parameters(rdx,
1347 ConstMethod::size_of_parameters_offset());
1348 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1349
1350
1351 // get parameter size (always needed)
1352 __ movptr(rdx, constMethod);
1353 __ load_unsigned_short(rcx, size_of_parameters);
1354
1355 // rbx: Method*
1356 // rcx: size of parameters
1357 // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1358
1359 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1360 __ subl(rdx, rcx); // rdx = no. of additional locals
1361
1362 // YYY
1363 // __ incrementl(rdx);
1364 // __ andl(rdx, -2);
1365
1366 // see if we've got enough room on the stack for locals plus overhead.
1367 generate_stack_overflow_check();
1368
1369 // get return address
1370 __ pop(rax);
1371
1372 // compute beginning of parameters
1373 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1374
1375 // rdx - # of additional locals
1376 // allocate space for locals
1377 // explicitly initialize locals
1378 {
1379 Label exit, loop;
1380 __ testl(rdx, rdx);
1381 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1382 __ bind(loop);
1383 __ push((int) NULL_WORD); // initialize local variables
1384 __ decrementl(rdx); // until everything initialized
1385 __ jcc(Assembler::greater, loop);
1386 __ bind(exit);
1387 }
1388
1389 // initialize fixed part of activation frame
1390 generate_fixed_frame(false);
1391
1392 // make sure method is not native & not abstract
1393 #ifdef ASSERT
1394 __ movl(rax, access_flags);
1395 {
1396 Label L;
1397 __ testl(rax, JVM_ACC_NATIVE);
1398 __ jcc(Assembler::zero, L);
1399 __ stop("tried to execute native method as non-native");
1400 __ bind(L);
1401 }
1402 {
1403 Label L;
1404 __ testl(rax, JVM_ACC_ABSTRACT);
1405 __ jcc(Assembler::zero, L);
1406 __ stop("tried to execute abstract method in interpreter");
1407 __ bind(L);
1408 }
1409 #endif
1410
1411 // Since at this point in the method invocation the exception
1412 // handler would try to exit the monitor of synchronized methods
1413 // which hasn't been entered yet, we set the thread local variable
1414 // _do_not_unlock_if_synchronized to true. The remove_activation
1415 // will check this flag.
1416
1417 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1418 NOT_LP64(__ get_thread(thread));
1419 const Address do_not_unlock_if_synchronized(thread,
1420 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1421 __ movbool(do_not_unlock_if_synchronized, true);
1422
1423 __ profile_parameters_type(rax, rcx, rdx);
1424 // increment invocation count & check for overflow
1425 Label invocation_counter_overflow;
1426 Label profile_method;
1427 Label profile_method_continue;
1428 if (inc_counter) {
1429 generate_counter_incr(&invocation_counter_overflow,
1430 &profile_method,
1431 &profile_method_continue);
1432 if (ProfileInterpreter) {
1433 __ bind(profile_method_continue);
1434 }
1435 }
1436
1437 Label continue_after_compile;
1438 __ bind(continue_after_compile);
1439
1440 // check for synchronized interpreted methods
1441 bang_stack_shadow_pages(false);
1442
1443 // reset the _do_not_unlock_if_synchronized flag
1444 NOT_LP64(__ get_thread(thread));
1445 __ movbool(do_not_unlock_if_synchronized, false);
1446
1447 // check for synchronized methods
1448 // Must happen AFTER invocation_counter check and stack overflow check,
1449 // so method is not locked if overflows.
1450 if (synchronized) {
1451 // Allocate monitor and lock method
1452 lock_method();
1453 } else {
1454 // no synchronization necessary
1455 #ifdef ASSERT
1456 {
1457 Label L;
1458 __ movl(rax, access_flags);
1459 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1460 __ jcc(Assembler::zero, L);
1461 __ stop("method needs synchronization");
1462 __ bind(L);
1463 }
1464 #endif
1465 }
1466
1467 // start execution
1468 #ifdef ASSERT
1469 {
1470 Label L;
1471 const Address monitor_block_top (rbp,
1472 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1473 __ movptr(rax, monitor_block_top);
1474 __ cmpptr(rax, rsp);
1475 __ jcc(Assembler::equal, L);
1476 __ stop("broken stack frame setup in interpreter");
1477 __ bind(L);
1478 }
1479 #endif
1480
1481 // jvmti support
1482 __ notify_method_entry();
1483
1484 __ dispatch_next(vtos);
1485
1486 // invocation counter overflow
1487 if (inc_counter) {
1488 if (ProfileInterpreter) {
1489 // We have decided to profile this method in the interpreter
1490 __ bind(profile_method);
1491 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1492 __ set_method_data_pointer_for_bcp();
1493 __ get_method(rbx);
1494 __ jmp(profile_method_continue);
1495 }
1496 // Handle overflow of counter and compile method
1497 __ bind(invocation_counter_overflow);
1498 generate_counter_overflow(continue_after_compile);
1499 }
1500
1501 return entry_point;
1502 }
1503
1504 //-----------------------------------------------------------------------------
1505 // Exceptions
1506
1507 void TemplateInterpreterGenerator::generate_throw_exception() {
1508 // Entry point in previous activation (i.e., if the caller was
1509 // interpreted)
1510 Interpreter::_rethrow_exception_entry = __ pc();
1511 // Restore sp to interpreter_frame_last_sp even though we are going
1512 // to empty the expression stack for the exception processing.
1513 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1514 // rax: exception
1515 // rdx: return address/pc that threw exception
1516 __ restore_bcp(); // r13/rsi points to call/send
1517 __ restore_locals();
1518 LP64_ONLY(__ reinit_heapbase()); // restore r12 as heapbase.
1519 // Entry point for exceptions thrown within interpreter code
1520 Interpreter::_throw_exception_entry = __ pc();
1521 // expression stack is undefined here
1522 // rax: exception
1523 // r13/rsi: exception bcp
1524 __ verify_oop(rax);
1525 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1526 LP64_ONLY(__ mov(c_rarg1, rax));
1527
1528 // expression stack must be empty before entering the VM in case of
1529 // an exception
1530 __ empty_expression_stack();
1531 // find exception handler address and preserve exception oop
1532 __ call_VM(rdx,
1533 CAST_FROM_FN_PTR(address,
1534 InterpreterRuntime::exception_handler_for_exception),
1535 rarg);
1536 // rax: exception handler entry point
1537 // rdx: preserved exception oop
1538 // r13/rsi: bcp for exception handler
1539 __ push_ptr(rdx); // push exception which is now the only value on the stack
1540 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1541
1542 // If the exception is not handled in the current frame the frame is
1543 // removed and the exception is rethrown (i.e. exception
1544 // continuation is _rethrow_exception).
1545 //
1546 // Note: At this point the bci is still the bxi for the instruction
1547 // which caused the exception and the expression stack is
1548 // empty. Thus, for any VM calls at this point, GC will find a legal
1549 // oop map (with empty expression stack).
1550
1551 // In current activation
1552 // tos: exception
1553 // esi: exception bcp
1554
1555 //
1556 // JVMTI PopFrame support
1557 //
1558
1559 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1560 __ empty_expression_stack();
1561 // Set the popframe_processing bit in pending_popframe_condition
1562 // indicating that we are currently handling popframe, so that
1563 // call_VMs that may happen later do not trigger new popframe
1564 // handling cycles.
1565 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1566 NOT_LP64(__ get_thread(thread));
1567 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1568 __ orl(rdx, JavaThread::popframe_processing_bit);
1569 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1570
1571 {
1572 // Check to see whether we are returning to a deoptimized frame.
1573 // (The PopFrame call ensures that the caller of the popped frame is
1574 // either interpreted or compiled and deoptimizes it if compiled.)
1575 // In this case, we can't call dispatch_next() after the frame is
1576 // popped, but instead must save the incoming arguments and restore
1577 // them after deoptimization has occurred.
1578 //
1579 // Note that we don't compare the return PC against the
1580 // deoptimization blob's unpack entry because of the presence of
1581 // adapter frames in C2.
1582 Label caller_not_deoptimized;
1583 Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1584 __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1585 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1586 InterpreterRuntime::interpreter_contains), rarg);
1587 __ testl(rax, rax);
1588 __ jcc(Assembler::notZero, caller_not_deoptimized);
1589
1590 // Compute size of arguments for saving when returning to
1591 // deoptimized caller
1592 __ get_method(rax);
1593 __ movptr(rax, Address(rax, Method::const_offset()));
1594 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1595 size_of_parameters_offset())));
1596 __ shll(rax, Interpreter::logStackElementSize);
1597 __ restore_locals();
1598 __ subptr(rlocals, rax);
1599 __ addptr(rlocals, wordSize);
1600 // Save these arguments
1601 NOT_LP64(__ get_thread(thread));
1602 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1603 Deoptimization::
1604 popframe_preserve_args),
1605 thread, rax, rlocals);
1606
1607 __ remove_activation(vtos, rdx,
1608 /* throw_monitor_exception */ false,
1609 /* install_monitor_exception */ false,
1610 /* notify_jvmdi */ false);
1611
1612 // Inform deoptimization that it is responsible for restoring
1613 // these arguments
1614 NOT_LP64(__ get_thread(thread));
1615 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1616 JavaThread::popframe_force_deopt_reexecution_bit);
1617
1618 // Continue in deoptimization handler
1619 __ jmp(rdx);
1620
1621 __ bind(caller_not_deoptimized);
1622 }
1623
1624 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1625 /* throw_monitor_exception */ false,
1626 /* install_monitor_exception */ false,
1627 /* notify_jvmdi */ false);
1628
1629 // Finish with popframe handling
1630 // A previous I2C followed by a deoptimization might have moved the
1631 // outgoing arguments further up the stack. PopFrame expects the
1632 // mutations to those outgoing arguments to be preserved and other
1633 // constraints basically require this frame to look exactly as
1634 // though it had previously invoked an interpreted activation with
1635 // no space between the top of the expression stack (current
1636 // last_sp) and the top of stack. Rather than force deopt to
1637 // maintain this kind of invariant all the time we call a small
1638 // fixup routine to move the mutated arguments onto the top of our
1639 // expression stack if necessary.
1640 #ifndef _LP64
1641 __ mov(rax, rsp);
1642 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1643 __ get_thread(thread);
1644 // PC must point into interpreter here
1645 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1646 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1647 __ get_thread(thread);
1648 #else
1649 __ mov(c_rarg1, rsp);
1650 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1651 // PC must point into interpreter here
1652 __ set_last_Java_frame(noreg, rbp, __ pc());
1653 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1654 #endif
1655 __ reset_last_Java_frame(thread, true);
1656
1657 // Restore the last_sp and null it out
1658 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1659 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1660
1661 __ restore_bcp();
1662 __ restore_locals();
1663 // The method data pointer was incremented already during
1664 // call profiling. We have to restore the mdp for the current bcp.
1665 if (ProfileInterpreter) {
1666 __ set_method_data_pointer_for_bcp();
1667 }
1668
1669 // Clear the popframe condition flag
1670 NOT_LP64(__ get_thread(thread));
1671 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1672 JavaThread::popframe_inactive);
1673
1674 #if INCLUDE_JVMTI
1675 {
1676 Label L_done;
1677 const Register local0 = rlocals;
1678
1679 __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1680 __ jcc(Assembler::notEqual, L_done);
1681
1682 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1683 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1684
1685 __ get_method(rdx);
1686 __ movptr(rax, Address(local0, 0));
1687 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1688
1689 __ testptr(rax, rax);
1690 __ jcc(Assembler::zero, L_done);
1691
1692 __ movptr(Address(rbx, 0), rax);
1693 __ bind(L_done);
1694 }
1695 #endif // INCLUDE_JVMTI
1696
1697 __ dispatch_next(vtos);
1698 // end of PopFrame support
1699
1700 Interpreter::_remove_activation_entry = __ pc();
1701
1702 // preserve exception over this code sequence
1703 __ pop_ptr(rax);
1704 NOT_LP64(__ get_thread(thread));
1705 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1706 // remove the activation (without doing throws on illegalMonitorExceptions)
1707 __ remove_activation(vtos, rdx, false, true, false);
1708 // restore exception
1709 NOT_LP64(__ get_thread(thread));
1710 __ get_vm_result(rax, thread);
1711
1712 // In between activations - previous activation type unknown yet
1713 // compute continuation point - the continuation point expects the
1714 // following registers set up:
1715 //
1716 // rax: exception
1717 // rdx: return address/pc that threw exception
1718 // rsp: expression stack of caller
1719 // rbp: ebp of caller
1720 __ push(rax); // save exception
1721 __ push(rdx); // save return address
1722 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1723 SharedRuntime::exception_handler_for_return_address),
1724 thread, rdx);
1725 __ mov(rbx, rax); // save exception handler
1726 __ pop(rdx); // restore return address
1727 __ pop(rax); // restore exception
1728 // Note that an "issuing PC" is actually the next PC after the call
1729 __ jmp(rbx); // jump to exception
1730 // handler of caller
1731 }
1732
1733
1734 //
1735 // JVMTI ForceEarlyReturn support
1736 //
1737 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1738 address entry = __ pc();
1739
1740 __ restore_bcp();
1741 __ restore_locals();
1742 __ empty_expression_stack();
1743 __ load_earlyret_value(state); // 32 bits returns value in rdx, so don't reuse
1744
1745 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1746 NOT_LP64(__ get_thread(thread));
1747 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1748 Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1749
1750 // Clear the earlyret state
1751 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1752
1753 __ remove_activation(state, rsi,
1754 false, /* throw_monitor_exception */
1755 false, /* install_monitor_exception */
1756 true); /* notify_jvmdi */
1757 __ jmp(rsi);
1758
1759 return entry;
1760 } // end of ForceEarlyReturn support
1761
1762
1763 //-----------------------------------------------------------------------------
1764 // Helper for vtos entry point generation
1765
1766 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1767 address& bep,
1768 address& cep,
1769 address& sep,
1770 address& aep,
1771 address& iep,
1772 address& lep,
1773 address& fep,
1774 address& dep,
1775 address& vep) {
1776 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1777 Label L;
1778 aep = __ pc(); __ push_ptr(); __ jmp(L);
1779 #ifndef _LP64
1780 fep = __ pc(); __ push(ftos); __ jmp(L);
1781 dep = __ pc(); __ push(dtos); __ jmp(L);
1782 #else
1783 fep = __ pc(); __ push_f(xmm0); __ jmp(L);
1784 dep = __ pc(); __ push_d(xmm0); __ jmp(L);
1785 #endif // _LP64
1786 lep = __ pc(); __ push_l(); __ jmp(L);
1787 bep = cep = sep =
1788 iep = __ pc(); __ push_i();
1789 vep = __ pc();
1790 __ bind(L);
1791 generate_and_dispatch(t);
1792 }
1793
1794 //-----------------------------------------------------------------------------
1795
1796 // Non-product code
1797 #ifndef PRODUCT
1798
1799 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1800 address entry = __ pc();
1801
1802 #ifndef _LP64
1803 // prepare expression stack
1804 __ pop(rcx); // pop return address so expression stack is 'pure'
1805 __ push(state); // save tosca
1806
1807 // pass tosca registers as arguments & call tracer
1808 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1809 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1810 __ pop(state); // restore tosca
1811
1812 // return
1813 __ jmp(rcx);
1814 #else
1815 __ push(state);
1816 __ push(c_rarg0);
1817 __ push(c_rarg1);
1818 __ push(c_rarg2);
1819 __ push(c_rarg3);
1820 __ mov(c_rarg2, rax); // Pass itos
1821 #ifdef _WIN64
1822 __ movflt(xmm3, xmm0); // Pass ftos
1823 #endif
1824 __ call_VM(noreg,
1825 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1826 c_rarg1, c_rarg2, c_rarg3);
1827 __ pop(c_rarg3);
1828 __ pop(c_rarg2);
1829 __ pop(c_rarg1);
1830 __ pop(c_rarg0);
1831 __ pop(state);
1832 __ ret(0); // return from result handler
1833 #endif // _LP64
1834
1835 return entry;
1836 }
1837
1838 void TemplateInterpreterGenerator::count_bytecode() {
1839 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1840 }
1841
1842 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1843 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1844 }
1845
1846 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1847 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1848 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1849 __ orl(rbx,
1850 ((int) t->bytecode()) <<
1851 BytecodePairHistogram::log2_number_of_codes);
1852 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1853 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1854 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1855 }
1856
1857
1858 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1859 // Call a little run-time stub to avoid blow-up for each bytecode.
1860 // The run-time runtime saves the right registers, depending on
1861 // the tosca in-state for the given template.
1862
1863 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1864 "entry must have been generated");
1865 #ifndef _LP64
1866 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1867 #else
1868 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1869 __ andptr(rsp, -16); // align stack as required by ABI
1870 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1871 __ mov(rsp, r12); // restore sp
1872 __ reinit_heapbase();
1873 #endif // _LP64
1874 }
1875
1876
1877 void TemplateInterpreterGenerator::stop_interpreter_at() {
1878 Label L;
1879 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1880 StopInterpreterAt);
1881 __ jcc(Assembler::notEqual, L);
1882 __ int3();
1883 __ bind(L);
1884 }
1885 #endif // !PRODUCT