1 /*
2 * Copyright (c) 2003, 2016, 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 "interpreter/bytecodeHistogram.hpp"
28 #include "interpreter/interp_masm.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/templateInterpreterGenerator.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "oops/arrayOop.hpp"
34 #include "oops/methodData.hpp"
35 #include "oops/method.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/jvmtiExport.hpp"
38 #include "prims/jvmtiThreadState.hpp"
39 #include "runtime/arguments.hpp"
40 #include "runtime/deoptimization.hpp"
41 #include "runtime/frame.inline.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/synchronizer.hpp"
45 #include "runtime/timer.hpp"
46 #include "runtime/vframeArray.hpp"
47 #include "utilities/debug.hpp"
48 #include "utilities/macros.hpp"
49
50 #define __ _masm->
51
52 // Size of interpreter code. Increase if too small. Interpreter will
53 // fail with a guarantee ("not enough space for interpreter generation");
54 // if too small.
55 // Run with +PrintInterpreter to get the VM to print out the size.
56 // Max size with JVMTI
57 #ifdef AMD64
58 int TemplateInterpreter::InterpreterCodeSize = 256 * 1024;
59 #else
60 int TemplateInterpreter::InterpreterCodeSize = 224 * 1024;
61 #endif // AMD64
62
63 // Global Register Names
64 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
65 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
66
67 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
68 const int bcp_offset = frame::interpreter_frame_bcp_offset * wordSize;
69 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
70
71
72 //-----------------------------------------------------------------------------
73
74 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
75 address entry = __ pc();
76
77 #ifdef ASSERT
78 {
79 Label L;
80 __ lea(rax, Address(rbp,
81 frame::interpreter_frame_monitor_block_top_offset *
82 wordSize));
83 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
84 // grows negative)
85 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
86 __ stop ("interpreter frame not set up");
87 __ bind(L);
88 }
89 #endif // ASSERT
90 // Restore bcp under the assumption that the current frame is still
91 // interpreted
92 __ restore_bcp();
93
94 // expression stack must be empty before entering the VM if an
95 // exception happened
96 __ empty_expression_stack();
97 // throw exception
98 __ call_VM(noreg,
99 CAST_FROM_FN_PTR(address,
100 InterpreterRuntime::throw_StackOverflowError));
101 return entry;
102 }
103
104 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
105 const char* name) {
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 __ lea(rarg, ExternalAddress((address)name));
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
175 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
176 address entry = __ pc();
177 // NULL last_sp until next java call
178 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
179 __ dispatch_next(state);
180 return entry;
181 }
182
183
184 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
185 address entry = __ pc();
186
187 #ifndef _LP64
188 #ifdef COMPILER2
189 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
190 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
191 for (int i = 1; i < 8; i++) {
192 __ ffree(i);
193 }
194 } else if (UseSSE < 2) {
195 __ empty_FPU_stack();
196 }
197 #endif // COMPILER2
198 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
199 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
200 } else {
201 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
202 }
203
204 if (state == ftos) {
205 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
206 } else if (state == dtos) {
207 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
208 }
209 #endif // _LP64
210
211 // Restore stack bottom in case i2c adjusted stack
212 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
213 // and NULL it as marker that esp is now tos until next java call
214 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
215
216 __ restore_bcp();
217 __ restore_locals();
218
219 if (state == atos) {
220 Register mdp = rbx;
221 Register tmp = rcx;
222 __ profile_return_type(mdp, rax, tmp);
223 }
224
225 const Register cache = rbx;
226 const Register index = rcx;
227 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
228
229 const Register flags = cache;
230 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
231 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
232 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
233 __ dispatch_next(state, step);
234
235 return entry;
236 }
237
238
239 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
240 address entry = __ pc();
241
242 #ifndef _LP64
243 if (state == ftos) {
244 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
245 } else if (state == dtos) {
246 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
247 }
248 #endif // _LP64
249
250 // NULL last_sp until next java call
251 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
252 __ restore_bcp();
253 __ restore_locals();
254 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
255 NOT_LP64(__ get_thread(thread));
256 #if INCLUDE_JVMCI
257 // Check if we need to take lock at entry of synchronized method.
258 if (UseJVMCICompiler) {
259 Label L;
260 __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
261 __ jcc(Assembler::zero, L);
262 // Clear flag.
263 __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
264 // Satisfy calling convention for lock_method().
265 __ get_method(rbx);
266 // Take lock.
267 lock_method();
268 __ bind(L);
269 }
270 #endif
271 // handle exceptions
272 {
273 Label L;
274 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
275 __ jcc(Assembler::zero, L);
276 __ call_VM(noreg,
277 CAST_FROM_FN_PTR(address,
278 InterpreterRuntime::throw_pending_exception));
279 __ should_not_reach_here();
280 __ bind(L);
281 }
282 __ dispatch_next(state, step);
283 return entry;
284 }
285
286 address TemplateInterpreterGenerator::generate_result_handler_for(
287 BasicType type) {
288 address entry = __ pc();
289 switch (type) {
290 case T_BOOLEAN: __ c2bool(rax); break;
291 #ifndef _LP64
292 case T_CHAR : __ andptr(rax, 0xFFFF); break;
293 #else
294 case T_CHAR : __ movzwl(rax, rax); break;
295 #endif // _LP64
296 case T_BYTE : __ sign_extend_byte(rax); break;
297 case T_SHORT : __ sign_extend_short(rax); break;
298 case T_INT : /* nothing to do */ break;
299 case T_LONG : /* nothing to do */ break;
300 case T_VOID : /* nothing to do */ break;
301 #ifndef _LP64
302 case T_DOUBLE :
303 case T_FLOAT :
304 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
305 __ pop(t); // remove return address first
306 // Must return a result for interpreter or compiler. In SSE
307 // mode, results are returned in xmm0 and the FPU stack must
308 // be empty.
309 if (type == T_FLOAT && UseSSE >= 1) {
310 // Load ST0
311 __ fld_d(Address(rsp, 0));
312 // Store as float and empty fpu stack
313 __ fstp_s(Address(rsp, 0));
314 // and reload
315 __ movflt(xmm0, Address(rsp, 0));
316 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
317 __ movdbl(xmm0, Address(rsp, 0));
318 } else {
319 // restore ST0
320 __ fld_d(Address(rsp, 0));
321 }
322 // and pop the temp
323 __ addptr(rsp, 2 * wordSize);
324 __ push(t); // restore return address
325 }
326 break;
327 #else
328 case T_FLOAT : /* nothing to do */ break;
329 case T_DOUBLE : /* nothing to do */ break;
330 #endif // _LP64
331
332 case T_OBJECT :
333 // retrieve result from frame
334 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
335 // and verify it
336 __ verify_oop(rax);
337 break;
338 default : ShouldNotReachHere();
339 }
340 __ ret(0); // return from result handler
341 return entry;
342 }
343
344 address TemplateInterpreterGenerator::generate_safept_entry_for(
345 TosState state,
346 address runtime_entry) {
347 address entry = __ pc();
348 __ push(state);
349 __ call_VM(noreg, runtime_entry);
350 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
351 return entry;
352 }
353
354
355
356 // Helpers for commoning out cases in the various type of method entries.
357 //
358
359
360 // increment invocation count & check for overflow
361 //
362 // Note: checking for negative value instead of overflow
363 // so we have a 'sticky' overflow test
364 //
365 // rbx: method
366 // rcx: invocation counter
367 //
368 void TemplateInterpreterGenerator::generate_counter_incr(
369 Label* overflow,
370 Label* profile_method,
371 Label* profile_method_continue) {
372 Label done;
373 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
374 if (TieredCompilation) {
375 int increment = InvocationCounter::count_increment;
376 Label no_mdo;
377 if (ProfileInterpreter) {
378 // Are we profiling?
379 __ movptr(rax, Address(rbx, Method::method_data_offset()));
380 __ testptr(rax, rax);
381 __ jccb(Assembler::zero, no_mdo);
382 // Increment counter in the MDO
383 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
384 in_bytes(InvocationCounter::counter_offset()));
385 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
386 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
387 __ jmp(done);
388 }
389 __ bind(no_mdo);
390 // Increment counter in MethodCounters
391 const Address invocation_counter(rax,
392 MethodCounters::invocation_counter_offset() +
393 InvocationCounter::counter_offset());
394 __ get_method_counters(rbx, rax, done);
395 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
396 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
397 false, Assembler::zero, overflow);
398 __ bind(done);
399 } else { // not TieredCompilation
400 const Address backedge_counter(rax,
401 MethodCounters::backedge_counter_offset() +
402 InvocationCounter::counter_offset());
403 const Address invocation_counter(rax,
404 MethodCounters::invocation_counter_offset() +
405 InvocationCounter::counter_offset());
406
407 __ get_method_counters(rbx, rax, done);
408
409 if (ProfileInterpreter) {
410 __ incrementl(Address(rax,
411 MethodCounters::interpreter_invocation_counter_offset()));
412 }
413 // Update standard invocation counters
414 __ movl(rcx, invocation_counter);
415 __ incrementl(rcx, InvocationCounter::count_increment);
416 __ movl(invocation_counter, rcx); // save invocation count
417
418 __ movl(rax, backedge_counter); // load backedge counter
419 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
420
421 __ addl(rcx, rax); // add both counters
422
423 // profile_method is non-null only for interpreted method so
424 // profile_method != NULL == !native_call
425
426 if (ProfileInterpreter && profile_method != NULL) {
427 // Test to see if we should create a method data oop
428 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
429 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
430 __ jcc(Assembler::less, *profile_method_continue);
431
432 // if no method data exists, go to profile_method
433 __ test_method_data_pointer(rax, *profile_method);
434 }
435
436 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
437 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
438 __ jcc(Assembler::aboveEqual, *overflow);
439 __ bind(done);
440 }
441 }
442
443 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
444
445 // Asm interpreter on entry
446 // r14/rdi - locals
447 // r13/rsi - bcp
448 // rbx - method
449 // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
450 // rbp - interpreter frame
451
452 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
453 // Everything as it was on entry
454 // rdx is not restored. Doesn't appear to really be set.
455
456 // InterpreterRuntime::frequency_counter_overflow takes two
457 // arguments, the first (thread) is passed by call_VM, the second
458 // indicates if the counter overflow occurs at a backwards branch
459 // (NULL bcp). We pass zero for it. The call returns the address
460 // of the verified entry point for the method or NULL if the
461 // compilation did not complete (either went background or bailed
462 // out).
463 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
464 __ movl(rarg, 0);
465 __ call_VM(noreg,
466 CAST_FROM_FN_PTR(address,
467 InterpreterRuntime::frequency_counter_overflow),
468 rarg);
469
470 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
471 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
472 // and jump to the interpreted entry.
473 __ jmp(do_continue, relocInfo::none);
474 }
475
476 // See if we've got enough room on the stack for locals plus overhead below
477 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
478 // without going through the signal handler, i.e., reserved and yellow zones
479 // will not be made usable. The shadow zone must suffice to handle the
480 // overflow.
481 // The expression stack grows down incrementally, so the normal guard
482 // page mechanism will work for that.
483 //
484 // NOTE: Since the additional locals are also always pushed (wasn't
485 // obvious in generate_fixed_frame) so the guard should work for them
486 // too.
487 //
488 // Args:
489 // rdx: number of additional locals this frame needs (what we must check)
490 // rbx: Method*
491 //
492 // Kills:
493 // rax
494 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
495
496 // monitor entry size: see picture of stack in frame_x86.hpp
497 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
498
499 // total overhead size: entry_size + (saved rbp through expr stack
500 // bottom). be sure to change this if you add/subtract anything
501 // to/from the overhead area
502 const int overhead_size =
503 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
504
505 const int page_size = os::vm_page_size();
506
507 Label after_frame_check;
508
509 // see if the frame is greater than one page in size. If so,
510 // then we need to verify there is enough stack space remaining
511 // for the additional locals.
512 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
513 __ jcc(Assembler::belowEqual, after_frame_check);
514
515 // compute rsp as if this were going to be the last frame on
516 // the stack before the red zone
517
518 Label after_frame_check_pop;
519 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
520 #ifndef _LP64
521 __ push(thread);
522 __ get_thread(thread);
523 #endif
524
525 const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
526
527 // locals + overhead, in bytes
528 __ mov(rax, rdx);
529 __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
530 __ addptr(rax, overhead_size);
531
532 #ifdef ASSERT
533 Label limit_okay;
534 // Verify that thread stack overflow limit is non-zero.
535 __ cmpptr(stack_limit, (int32_t)NULL_WORD);
536 __ jcc(Assembler::notEqual, limit_okay);
537 __ stop("stack overflow limit is zero");
538 __ bind(limit_okay);
539 #endif
540
541 // Add locals/frame size to stack limit.
542 __ addptr(rax, stack_limit);
543
544 // Check against the current stack bottom.
545 __ cmpptr(rsp, rax);
546
547 __ jcc(Assembler::above, after_frame_check_pop);
548 NOT_LP64(__ pop(rsi)); // get saved bcp
549
550 // Restore sender's sp as SP. This is necessary if the sender's
551 // frame is an extended compiled frame (see gen_c2i_adapter())
552 // and safer anyway in case of JSR292 adaptations.
553
554 __ pop(rax); // return address must be moved if SP is changed
555 __ mov(rsp, rbcp);
556 __ push(rax);
557
558 // Note: the restored frame is not necessarily interpreted.
559 // Use the shared runtime version of the StackOverflowError.
560 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
561 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
562 // all done with frame size check
563 __ bind(after_frame_check_pop);
564 NOT_LP64(__ pop(rsi));
565
566 // all done with frame size check
567 __ bind(after_frame_check);
568 }
569
570 // Allocate monitor and lock method (asm interpreter)
571 //
572 // Args:
573 // rbx: Method*
574 // r14/rdi: locals
575 //
576 // Kills:
577 // rax
578 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
579 // rscratch1, rscratch2 (scratch regs)
580 void TemplateInterpreterGenerator::lock_method() {
581 // synchronize method
582 const Address access_flags(rbx, Method::access_flags_offset());
583 const Address monitor_block_top(
584 rbp,
585 frame::interpreter_frame_monitor_block_top_offset * wordSize);
586 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
587
588 #ifdef ASSERT
589 {
590 Label L;
591 __ movl(rax, access_flags);
592 __ testl(rax, JVM_ACC_SYNCHRONIZED);
593 __ jcc(Assembler::notZero, L);
594 __ stop("method doesn't need synchronization");
595 __ bind(L);
596 }
597 #endif // ASSERT
598
599 // get synchronization object
600 {
601 Label done;
602 __ movl(rax, access_flags);
603 __ testl(rax, JVM_ACC_STATIC);
604 // get receiver (assume this is frequent case)
605 __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
606 __ jcc(Assembler::zero, done);
607 __ load_mirror(rax, rbx);
608
609 #ifdef ASSERT
610 {
611 Label L;
612 __ testptr(rax, rax);
613 __ jcc(Assembler::notZero, L);
614 __ stop("synchronization object is NULL");
615 __ bind(L);
616 }
617 #endif // ASSERT
618
619 __ bind(done);
620 }
621
622 // add space for monitor & lock
623 __ subptr(rsp, entry_size); // add space for a monitor entry
624 __ movptr(monitor_block_top, rsp); // set new monitor block top
625 // store object
626 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
627 const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
628 __ movptr(lockreg, rsp); // object address
629 __ lock_object(lockreg);
630 }
631
632 // Generate a fixed interpreter frame. This is identical setup for
633 // interpreted methods and for native methods hence the shared code.
634 //
635 // Args:
636 // rax: return address
637 // rbx: Method*
638 // r14/rdi: pointer to locals
639 // r13/rsi: sender sp
640 // rdx: cp cache
641 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
642 // initialize fixed part of activation frame
643 __ push(rax); // save return address
644 __ enter(); // save old & set new rbp
645 __ push(rbcp); // set sender sp
646 __ push((int)NULL_WORD); // leave last_sp as null
647 __ movptr(rbcp, Address(rbx, Method::const_offset())); // get ConstMethod*
648 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
649 __ push(rbx); // save Method*
650 // Get mirror and store it in the frame as GC root for this Method*
651 __ load_mirror(rdx, rbx);
652 __ push(rdx);
653 if (ProfileInterpreter) {
654 Label method_data_continue;
655 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
656 __ testptr(rdx, rdx);
657 __ jcc(Assembler::zero, method_data_continue);
658 __ addptr(rdx, in_bytes(MethodData::data_offset()));
659 __ bind(method_data_continue);
660 __ push(rdx); // set the mdp (method data pointer)
661 } else {
662 __ push(0);
663 }
664
665 __ movptr(rdx, Address(rbx, Method::const_offset()));
666 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
667 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
668 __ push(rdx); // set constant pool cache
669 __ push(rlocals); // set locals pointer
670 if (native_call) {
671 __ push(0); // no bcp
672 } else {
673 __ push(rbcp); // set bcp
674 }
675 __ push(0); // reserve word for pointer to expression stack bottom
676 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
677 }
678
679 // End of helpers
680
681 // Method entry for java.lang.ref.Reference.get.
682 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
683 #if INCLUDE_ALL_GCS
684 // Code: _aload_0, _getfield, _areturn
685 // parameter size = 1
686 //
687 // The code that gets generated by this routine is split into 2 parts:
688 // 1. The "intrinsified" code for G1 (or any SATB based GC),
689 // 2. The slow path - which is an expansion of the regular method entry.
690 //
691 // Notes:-
692 // * In the G1 code we do not check whether we need to block for
693 // a safepoint. If G1 is enabled then we must execute the specialized
694 // code for Reference.get (except when the Reference object is null)
695 // so that we can log the value in the referent field with an SATB
696 // update buffer.
697 // If the code for the getfield template is modified so that the
698 // G1 pre-barrier code is executed when the current method is
699 // Reference.get() then going through the normal method entry
700 // will be fine.
701 // * The G1 code can, however, check the receiver object (the instance
702 // of java.lang.Reference) and jump to the slow path if null. If the
703 // Reference object is null then we obviously cannot fetch the referent
704 // and so we don't need to call the G1 pre-barrier. Thus we can use the
705 // regular method entry code to generate the NPE.
706 //
707 // rbx: Method*
708
709 // r13: senderSP must preserve for slow path, set SP to it on fast path
710
711 address entry = __ pc();
712
713 const int referent_offset = java_lang_ref_Reference::referent_offset;
714 guarantee(referent_offset > 0, "referent offset not initialized");
715
716 if (UseG1GC) {
717 Label slow_path;
718 // rbx: method
719
720 // Check if local 0 != NULL
721 // If the receiver is null then it is OK to jump to the slow path.
722 __ movptr(rax, Address(rsp, wordSize));
723
724 __ testptr(rax, rax);
725 __ jcc(Assembler::zero, slow_path);
726
727 // rax: local 0
728 // rbx: method (but can be used as scratch now)
729 // rdx: scratch
730 // rdi: scratch
731
732 // Preserve the sender sp in case the pre-barrier
733 // calls the runtime
734 NOT_LP64(__ push(rsi));
735
736 // Generate the G1 pre-barrier code to log the value of
737 // the referent field in an SATB buffer.
738
739 // Load the value of the referent field.
740 const Address field_address(rax, referent_offset);
741 __ load_heap_oop(rax, field_address);
742
743 const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
744 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
745 NOT_LP64(__ get_thread(thread));
746
747 // Generate the G1 pre-barrier code to log the value of
748 // the referent field in an SATB buffer.
749 __ g1_write_barrier_pre(noreg /* obj */,
750 rax /* pre_val */,
751 thread /* thread */,
752 rbx /* tmp */,
753 true /* tosca_live */,
754 true /* expand_call */);
755
756 // _areturn
757 NOT_LP64(__ pop(rsi)); // get sender sp
758 __ pop(rdi); // get return address
759 __ mov(rsp, sender_sp); // set sp to sender sp
760 __ jmp(rdi);
761 __ ret(0);
762
763 // generate a vanilla interpreter entry as the slow path
764 __ bind(slow_path);
765 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
766 return entry;
767 }
768 #endif // INCLUDE_ALL_GCS
769
770 // If G1 is not enabled then attempt to go through the accessor entry point
771 // Reference.get is an accessor
772 return NULL;
773 }
774
775 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
776 // Quick & dirty stack overflow checking: bang the stack & handle trap.
777 // Note that we do the banging after the frame is setup, since the exception
778 // handling code expects to find a valid interpreter frame on the stack.
779 // Doing the banging earlier fails if the caller frame is not an interpreter
780 // frame.
781 // (Also, the exception throwing code expects to unlock any synchronized
782 // method receiever, so do the banging after locking the receiver.)
783
784 // Bang each page in the shadow zone. We can't assume it's been done for
785 // an interpreter frame with greater than a page of locals, so each page
786 // needs to be checked. Only true for non-native.
787 if (UseStackBanging) {
788 const int page_size = os::vm_page_size();
789 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
790 const int start_page = native_call ? n_shadow_pages : 1;
791 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
792 __ bang_stack_with_offset(pages*page_size);
793 }
794 }
795 }
796
797 // Interpreter stub for calling a native method. (asm interpreter)
798 // This sets up a somewhat different looking stack for calling the
799 // native method than the typical interpreter frame setup.
800 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
801 // determine code generation flags
802 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
803
804 // rbx: Method*
805 // rbcp: sender sp
806
807 address entry_point = __ pc();
808
809 const Address constMethod (rbx, Method::const_offset());
810 const Address access_flags (rbx, Method::access_flags_offset());
811 const Address size_of_parameters(rcx, ConstMethod::
812 size_of_parameters_offset());
813
814
815 // get parameter size (always needed)
816 __ movptr(rcx, constMethod);
817 __ load_unsigned_short(rcx, size_of_parameters);
818
819 // native calls don't need the stack size check since they have no
820 // expression stack and the arguments are already on the stack and
821 // we only add a handful of words to the stack
822
823 // rbx: Method*
824 // rcx: size of parameters
825 // rbcp: sender sp
826 __ pop(rax); // get return address
827
828 // for natives the size of locals is zero
829
830 // compute beginning of parameters
831 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
832
833 // add 2 zero-initialized slots for native calls
834 // initialize result_handler slot
835 __ push((int) NULL_WORD);
836 // slot for oop temp
837 // (static native method holder mirror/jni oop result)
838 __ push((int) NULL_WORD);
839
840 // initialize fixed part of activation frame
841 generate_fixed_frame(true);
842
843 // make sure method is native & not abstract
844 #ifdef ASSERT
845 __ movl(rax, access_flags);
846 {
847 Label L;
848 __ testl(rax, JVM_ACC_NATIVE);
849 __ jcc(Assembler::notZero, L);
850 __ stop("tried to execute non-native method as native");
851 __ bind(L);
852 }
853 {
854 Label L;
855 __ testl(rax, JVM_ACC_ABSTRACT);
856 __ jcc(Assembler::zero, L);
857 __ stop("tried to execute abstract method in interpreter");
858 __ bind(L);
859 }
860 #endif
861
862 // Since at this point in the method invocation the exception handler
863 // would try to exit the monitor of synchronized methods which hasn't
864 // been entered yet, we set the thread local variable
865 // _do_not_unlock_if_synchronized to true. The remove_activation will
866 // check this flag.
867
868 const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
869 NOT_LP64(__ get_thread(thread1));
870 const Address do_not_unlock_if_synchronized(thread1,
871 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
872 __ movbool(do_not_unlock_if_synchronized, true);
873
874 // increment invocation count & check for overflow
875 Label invocation_counter_overflow;
876 if (inc_counter) {
877 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
878 }
879
880 Label continue_after_compile;
881 __ bind(continue_after_compile);
882
883 bang_stack_shadow_pages(true);
884
885 // reset the _do_not_unlock_if_synchronized flag
886 NOT_LP64(__ get_thread(thread1));
887 __ movbool(do_not_unlock_if_synchronized, false);
888
889 // check for synchronized methods
890 // Must happen AFTER invocation_counter check and stack overflow check,
891 // so method is not locked if overflows.
892 if (synchronized) {
893 lock_method();
894 } else {
895 // no synchronization necessary
896 #ifdef ASSERT
897 {
898 Label L;
899 __ movl(rax, access_flags);
900 __ testl(rax, JVM_ACC_SYNCHRONIZED);
901 __ jcc(Assembler::zero, L);
902 __ stop("method needs synchronization");
903 __ bind(L);
904 }
905 #endif
906 }
907
908 // start execution
909 #ifdef ASSERT
910 {
911 Label L;
912 const Address monitor_block_top(rbp,
913 frame::interpreter_frame_monitor_block_top_offset * wordSize);
914 __ movptr(rax, monitor_block_top);
915 __ cmpptr(rax, rsp);
916 __ jcc(Assembler::equal, L);
917 __ stop("broken stack frame setup in interpreter");
918 __ bind(L);
919 }
920 #endif
921
922 // jvmti support
923 __ notify_method_entry();
924
925 // work registers
926 const Register method = rbx;
927 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
928 const Register t = NOT_LP64(rcx) LP64_ONLY(r11);
929
930 // allocate space for parameters
931 __ get_method(method);
932 __ movptr(t, Address(method, Method::const_offset()));
933 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
934
935 #ifndef _LP64
936 __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
937 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
938 __ subptr(rsp, t);
939 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
940 #else
941 __ shll(t, Interpreter::logStackElementSize);
942
943 __ subptr(rsp, t);
944 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
945 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
946 #endif // _LP64
947
948 // get signature handler
949 {
950 Label L;
951 __ movptr(t, Address(method, Method::signature_handler_offset()));
952 __ testptr(t, t);
953 __ jcc(Assembler::notZero, L);
954 __ call_VM(noreg,
955 CAST_FROM_FN_PTR(address,
956 InterpreterRuntime::prepare_native_call),
957 method);
958 __ get_method(method);
959 __ movptr(t, Address(method, Method::signature_handler_offset()));
960 __ bind(L);
961 }
962
963 // call signature handler
964 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
965 "adjust this code");
966 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
967 "adjust this code");
968 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
969 "adjust this code");
970
971 // The generated handlers do not touch RBX (the method oop).
972 // However, large signatures cannot be cached and are generated
973 // each time here. The slow-path generator can do a GC on return,
974 // so we must reload it after the call.
975 __ call(t);
976 __ get_method(method); // slow path can do a GC, reload RBX
977
978
979 // result handler is in rax
980 // set result handler
981 __ movptr(Address(rbp,
982 (frame::interpreter_frame_result_handler_offset) * wordSize),
983 rax);
984
985 // pass mirror handle if static call
986 {
987 Label L;
988 __ movl(t, Address(method, Method::access_flags_offset()));
989 __ testl(t, JVM_ACC_STATIC);
990 __ jcc(Assembler::zero, L);
991 // get mirror
992 __ load_mirror(t, method);
993 // copy mirror into activation frame
994 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
995 t);
996 // pass handle to mirror
997 #ifndef _LP64
998 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
999 __ movptr(Address(rsp, wordSize), t);
1000 #else
1001 __ lea(c_rarg1,
1002 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1003 #endif // _LP64
1004 __ bind(L);
1005 }
1006
1007 // get native function entry point
1008 {
1009 Label L;
1010 __ movptr(rax, Address(method, Method::native_function_offset()));
1011 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1012 __ cmpptr(rax, unsatisfied.addr());
1013 __ jcc(Assembler::notEqual, L);
1014 __ call_VM(noreg,
1015 CAST_FROM_FN_PTR(address,
1016 InterpreterRuntime::prepare_native_call),
1017 method);
1018 __ get_method(method);
1019 __ movptr(rax, Address(method, Method::native_function_offset()));
1020 __ bind(L);
1021 }
1022
1023 // pass JNIEnv
1024 #ifndef _LP64
1025 __ get_thread(thread);
1026 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1027 __ movptr(Address(rsp, 0), t);
1028
1029 // set_last_Java_frame_before_call
1030 // It is enough that the pc()
1031 // points into the right code segment. It does not have to be the correct return pc.
1032 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1033 #else
1034 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1035
1036 // It is enough that the pc() points into the right code
1037 // segment. It does not have to be the correct return pc.
1038 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1039 #endif // _LP64
1040
1041 // change thread state
1042 #ifdef ASSERT
1043 {
1044 Label L;
1045 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1046 __ cmpl(t, _thread_in_Java);
1047 __ jcc(Assembler::equal, L);
1048 __ stop("Wrong thread state in native stub");
1049 __ bind(L);
1050 }
1051 #endif
1052
1053 // Change state to native
1054
1055 __ movl(Address(thread, JavaThread::thread_state_offset()),
1056 _thread_in_native);
1057
1058 // Call the native method.
1059 __ call(rax);
1060 // 32: result potentially in rdx:rax or ST0
1061 // 64: result potentially in rax or xmm0
1062
1063 // Verify or restore cpu control state after JNI call
1064 __ restore_cpu_control_state_after_jni();
1065
1066 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1067 // in order to extract the result of a method call. If the order of these
1068 // pushes change or anything else is added to the stack then the code in
1069 // interpreter_frame_result must also change.
1070
1071 #ifndef _LP64
1072 // save potential result in ST(0) & rdx:rax
1073 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1074 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1075 // It is safe to do this push because state is _thread_in_native and return address will be found
1076 // via _last_native_pc and not via _last_jave_sp
1077
1078 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1079 // If the order changes or anything else is added to the stack the code in
1080 // interpreter_frame_result will have to be changed.
1081
1082 { Label L;
1083 Label push_double;
1084 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1085 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1086 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1087 float_handler.addr());
1088 __ jcc(Assembler::equal, push_double);
1089 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1090 double_handler.addr());
1091 __ jcc(Assembler::notEqual, L);
1092 __ bind(push_double);
1093 __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1094 __ bind(L);
1095 }
1096 #else
1097 __ push(dtos);
1098 #endif // _LP64
1099
1100 __ push(ltos);
1101
1102 // change thread state
1103 NOT_LP64(__ get_thread(thread));
1104 __ movl(Address(thread, JavaThread::thread_state_offset()),
1105 _thread_in_native_trans);
1106
1107 if (os::is_MP()) {
1108 if (UseMembar) {
1109 // Force this write out before the read below
1110 __ membar(Assembler::Membar_mask_bits(
1111 Assembler::LoadLoad | Assembler::LoadStore |
1112 Assembler::StoreLoad | Assembler::StoreStore));
1113 } else {
1114 // Write serialization page so VM thread can do a pseudo remote membar.
1115 // We use the current thread pointer to calculate a thread specific
1116 // offset to write to within the page. This minimizes bus traffic
1117 // due to cache line collision.
1118 __ serialize_memory(thread, rcx);
1119 }
1120 }
1121
1122 #ifndef _LP64
1123 if (AlwaysRestoreFPU) {
1124 // Make sure the control word is correct.
1125 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1126 }
1127 #endif // _LP64
1128
1129 // check for safepoint operation in progress and/or pending suspend requests
1130 {
1131 Label Continue;
1132 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1133 SafepointSynchronize::_not_synchronized);
1134
1135 Label L;
1136 __ jcc(Assembler::notEqual, L);
1137 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1138 __ jcc(Assembler::equal, Continue);
1139 __ bind(L);
1140
1141 // Don't use call_VM as it will see a possible pending exception
1142 // and forward it and never return here preventing us from
1143 // clearing _last_native_pc down below. Also can't use
1144 // call_VM_leaf either as it will check to see if r13 & r14 are
1145 // preserved and correspond to the bcp/locals pointers. So we do a
1146 // runtime call by hand.
1147 //
1148 #ifndef _LP64
1149 __ push(thread);
1150 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1151 JavaThread::check_special_condition_for_native_trans)));
1152 __ increment(rsp, wordSize);
1153 __ get_thread(thread);
1154 #else
1155 __ mov(c_rarg0, r15_thread);
1156 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1157 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1158 __ andptr(rsp, -16); // align stack as required by ABI
1159 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1160 __ mov(rsp, r12); // restore sp
1161 __ reinit_heapbase();
1162 #endif // _LP64
1163 __ bind(Continue);
1164 }
1165
1166 // change thread state
1167 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1168
1169 // reset_last_Java_frame
1170 __ reset_last_Java_frame(thread, true, true);
1171
1172 // reset handle block
1173 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1174 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1175
1176 // If result is an oop unbox and store it in frame where gc will see it
1177 // and result handler will pick it up
1178
1179 {
1180 Label no_oop, store_result;
1181 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1182 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1183 __ jcc(Assembler::notEqual, no_oop);
1184 // retrieve result
1185 __ pop(ltos);
1186 __ testptr(rax, rax);
1187 __ jcc(Assembler::zero, store_result);
1188 __ movptr(rax, Address(rax, 0));
1189 __ bind(store_result);
1190 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1191 // keep stack depth as expected by pushing oop which will eventually be discarded
1192 __ push(ltos);
1193 __ bind(no_oop);
1194 }
1195
1196
1197 {
1198 Label no_reguard;
1199 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1200 JavaThread::stack_guard_yellow_reserved_disabled);
1201 __ jcc(Assembler::notEqual, no_reguard);
1202
1203 __ pusha(); // XXX only save smashed registers
1204 #ifndef _LP64
1205 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1206 __ popa();
1207 #else
1208 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1209 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1210 __ andptr(rsp, -16); // align stack as required by ABI
1211 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1212 __ mov(rsp, r12); // restore sp
1213 __ popa(); // XXX only restore smashed registers
1214 __ reinit_heapbase();
1215 #endif // _LP64
1216
1217 __ bind(no_reguard);
1218 }
1219
1220
1221 // The method register is junk from after the thread_in_native transition
1222 // until here. Also can't call_VM until the bcp has been
1223 // restored. Need bcp for throwing exception below so get it now.
1224 __ get_method(method);
1225
1226 // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1227 __ movptr(rbcp, Address(method, Method::const_offset())); // get ConstMethod*
1228 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
1229
1230 // handle exceptions (exception handling will handle unlocking!)
1231 {
1232 Label L;
1233 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1234 __ jcc(Assembler::zero, L);
1235 // Note: At some point we may want to unify this with the code
1236 // used in call_VM_base(); i.e., we should use the
1237 // StubRoutines::forward_exception code. For now this doesn't work
1238 // here because the rsp is not correctly set at this point.
1239 __ MacroAssembler::call_VM(noreg,
1240 CAST_FROM_FN_PTR(address,
1241 InterpreterRuntime::throw_pending_exception));
1242 __ should_not_reach_here();
1243 __ bind(L);
1244 }
1245
1246 // do unlocking if necessary
1247 {
1248 Label L;
1249 __ movl(t, Address(method, Method::access_flags_offset()));
1250 __ testl(t, JVM_ACC_SYNCHRONIZED);
1251 __ jcc(Assembler::zero, L);
1252 // the code below should be shared with interpreter macro
1253 // assembler implementation
1254 {
1255 Label unlock;
1256 // BasicObjectLock will be first in list, since this is a
1257 // synchronized method. However, need to check that the object
1258 // has not been unlocked by an explicit monitorexit bytecode.
1259 const Address monitor(rbp,
1260 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1261 wordSize - (int)sizeof(BasicObjectLock)));
1262
1263 const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1264
1265 // monitor expect in c_rarg1 for slow unlock path
1266 __ lea(regmon, monitor); // address of first monitor
1267
1268 __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1269 __ testptr(t, t);
1270 __ jcc(Assembler::notZero, unlock);
1271
1272 // Entry already unlocked, need to throw exception
1273 __ MacroAssembler::call_VM(noreg,
1274 CAST_FROM_FN_PTR(address,
1275 InterpreterRuntime::throw_illegal_monitor_state_exception));
1276 __ should_not_reach_here();
1277
1278 __ bind(unlock);
1279 __ unlock_object(regmon);
1280 }
1281 __ bind(L);
1282 }
1283
1284 // jvmti support
1285 // Note: This must happen _after_ handling/throwing any exceptions since
1286 // the exception handler code notifies the runtime of method exits
1287 // too. If this happens before, method entry/exit notifications are
1288 // not properly paired (was bug - gri 11/22/99).
1289 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1290
1291 // restore potential result in edx:eax, call result handler to
1292 // restore potential result in ST0 & handle result
1293
1294 __ pop(ltos);
1295 LP64_ONLY( __ pop(dtos));
1296
1297 __ movptr(t, Address(rbp,
1298 (frame::interpreter_frame_result_handler_offset) * wordSize));
1299 __ call(t);
1300
1301 // remove activation
1302 __ movptr(t, Address(rbp,
1303 frame::interpreter_frame_sender_sp_offset *
1304 wordSize)); // get sender sp
1305 __ leave(); // remove frame anchor
1306 __ pop(rdi); // get return address
1307 __ mov(rsp, t); // set sp to sender sp
1308 __ jmp(rdi);
1309
1310 if (inc_counter) {
1311 // Handle overflow of counter and compile method
1312 __ bind(invocation_counter_overflow);
1313 generate_counter_overflow(continue_after_compile);
1314 }
1315
1316 return entry_point;
1317 }
1318
1319 // Abstract method entry
1320 // Attempt to execute abstract method. Throw exception
1321 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1322
1323 address entry_point = __ pc();
1324
1325 // abstract method entry
1326
1327 // pop return address, reset last_sp to NULL
1328 __ empty_expression_stack();
1329 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1330 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1331
1332 // throw exception
1333 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
1334 // the call_VM checks for exception, so we should never return here.
1335 __ should_not_reach_here();
1336
1337 return entry_point;
1338 }
1339
1340 //
1341 // Generic interpreted method entry to (asm) interpreter
1342 //
1343 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1344 // determine code generation flags
1345 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1346
1347 // ebx: Method*
1348 // rbcp: sender sp
1349 address entry_point = __ pc();
1350
1351 const Address constMethod(rbx, Method::const_offset());
1352 const Address access_flags(rbx, Method::access_flags_offset());
1353 const Address size_of_parameters(rdx,
1354 ConstMethod::size_of_parameters_offset());
1355 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1356
1357
1358 // get parameter size (always needed)
1359 __ movptr(rdx, constMethod);
1360 __ load_unsigned_short(rcx, size_of_parameters);
1361
1362 // rbx: Method*
1363 // rcx: size of parameters
1364 // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1365
1366 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1367 __ subl(rdx, rcx); // rdx = no. of additional locals
1368
1369 // YYY
1370 // __ incrementl(rdx);
1371 // __ andl(rdx, -2);
1372
1373 // see if we've got enough room on the stack for locals plus overhead.
1374 generate_stack_overflow_check();
1375
1376 // get return address
1377 __ pop(rax);
1378
1379 // compute beginning of parameters
1380 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1381
1382 // rdx - # of additional locals
1383 // allocate space for locals
1384 // explicitly initialize locals
1385 {
1386 Label exit, loop;
1387 __ testl(rdx, rdx);
1388 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1389 __ bind(loop);
1390 __ push((int) NULL_WORD); // initialize local variables
1391 __ decrementl(rdx); // until everything initialized
1392 __ jcc(Assembler::greater, loop);
1393 __ bind(exit);
1394 }
1395
1396 // initialize fixed part of activation frame
1397 generate_fixed_frame(false);
1398
1399 // make sure method is not native & not abstract
1400 #ifdef ASSERT
1401 __ movl(rax, access_flags);
1402 {
1403 Label L;
1404 __ testl(rax, JVM_ACC_NATIVE);
1405 __ jcc(Assembler::zero, L);
1406 __ stop("tried to execute native method as non-native");
1407 __ bind(L);
1408 }
1409 {
1410 Label L;
1411 __ testl(rax, JVM_ACC_ABSTRACT);
1412 __ jcc(Assembler::zero, L);
1413 __ stop("tried to execute abstract method in interpreter");
1414 __ bind(L);
1415 }
1416 #endif
1417
1418 // Since at this point in the method invocation the exception
1419 // handler would try to exit the monitor of synchronized methods
1420 // which hasn't been entered yet, we set the thread local variable
1421 // _do_not_unlock_if_synchronized to true. The remove_activation
1422 // will check this flag.
1423
1424 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1425 NOT_LP64(__ get_thread(thread));
1426 const Address do_not_unlock_if_synchronized(thread,
1427 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1428 __ movbool(do_not_unlock_if_synchronized, true);
1429
1430 __ profile_parameters_type(rax, rcx, rdx);
1431 // increment invocation count & check for overflow
1432 Label invocation_counter_overflow;
1433 Label profile_method;
1434 Label profile_method_continue;
1435 if (inc_counter) {
1436 generate_counter_incr(&invocation_counter_overflow,
1437 &profile_method,
1438 &profile_method_continue);
1439 if (ProfileInterpreter) {
1440 __ bind(profile_method_continue);
1441 }
1442 }
1443
1444 Label continue_after_compile;
1445 __ bind(continue_after_compile);
1446
1447 // check for synchronized interpreted methods
1448 bang_stack_shadow_pages(false);
1449
1450 // reset the _do_not_unlock_if_synchronized flag
1451 NOT_LP64(__ get_thread(thread));
1452 __ movbool(do_not_unlock_if_synchronized, false);
1453
1454 // check for synchronized methods
1455 // Must happen AFTER invocation_counter check and stack overflow check,
1456 // so method is not locked if overflows.
1457 if (synchronized) {
1458 // Allocate monitor and lock method
1459 lock_method();
1460 } else {
1461 // no synchronization necessary
1462 #ifdef ASSERT
1463 {
1464 Label L;
1465 __ movl(rax, access_flags);
1466 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1467 __ jcc(Assembler::zero, L);
1468 __ stop("method needs synchronization");
1469 __ bind(L);
1470 }
1471 #endif
1472 }
1473
1474 // start execution
1475 #ifdef ASSERT
1476 {
1477 Label L;
1478 const Address monitor_block_top (rbp,
1479 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1480 __ movptr(rax, monitor_block_top);
1481 __ cmpptr(rax, rsp);
1482 __ jcc(Assembler::equal, L);
1483 __ stop("broken stack frame setup in interpreter");
1484 __ bind(L);
1485 }
1486 #endif
1487
1488 // jvmti support
1489 __ notify_method_entry();
1490
1491 __ dispatch_next(vtos);
1492
1493 // invocation counter overflow
1494 if (inc_counter) {
1495 if (ProfileInterpreter) {
1496 // We have decided to profile this method in the interpreter
1497 __ bind(profile_method);
1498 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1499 __ set_method_data_pointer_for_bcp();
1500 __ get_method(rbx);
1501 __ jmp(profile_method_continue);
1502 }
1503 // Handle overflow of counter and compile method
1504 __ bind(invocation_counter_overflow);
1505 generate_counter_overflow(continue_after_compile);
1506 }
1507
1508 return entry_point;
1509 }
1510
1511 //-----------------------------------------------------------------------------
1512 // Exceptions
1513
1514 void TemplateInterpreterGenerator::generate_throw_exception() {
1515 // Entry point in previous activation (i.e., if the caller was
1516 // interpreted)
1517 Interpreter::_rethrow_exception_entry = __ pc();
1518 // Restore sp to interpreter_frame_last_sp even though we are going
1519 // to empty the expression stack for the exception processing.
1520 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1521 // rax: exception
1522 // rdx: return address/pc that threw exception
1523 __ restore_bcp(); // r13/rsi points to call/send
1524 __ restore_locals();
1525 LP64_ONLY(__ reinit_heapbase()); // restore r12 as heapbase.
1526 // Entry point for exceptions thrown within interpreter code
1527 Interpreter::_throw_exception_entry = __ pc();
1528 // expression stack is undefined here
1529 // rax: exception
1530 // r13/rsi: exception bcp
1531 __ verify_oop(rax);
1532 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1533 LP64_ONLY(__ mov(c_rarg1, rax));
1534
1535 // expression stack must be empty before entering the VM in case of
1536 // an exception
1537 __ empty_expression_stack();
1538 // find exception handler address and preserve exception oop
1539 __ call_VM(rdx,
1540 CAST_FROM_FN_PTR(address,
1541 InterpreterRuntime::exception_handler_for_exception),
1542 rarg);
1543 // rax: exception handler entry point
1544 // rdx: preserved exception oop
1545 // r13/rsi: bcp for exception handler
1546 __ push_ptr(rdx); // push exception which is now the only value on the stack
1547 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1548
1549 // If the exception is not handled in the current frame the frame is
1550 // removed and the exception is rethrown (i.e. exception
1551 // continuation is _rethrow_exception).
1552 //
1553 // Note: At this point the bci is still the bxi for the instruction
1554 // which caused the exception and the expression stack is
1555 // empty. Thus, for any VM calls at this point, GC will find a legal
1556 // oop map (with empty expression stack).
1557
1558 // In current activation
1559 // tos: exception
1560 // esi: exception bcp
1561
1562 //
1563 // JVMTI PopFrame support
1564 //
1565
1566 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1567 __ empty_expression_stack();
1568 // Set the popframe_processing bit in pending_popframe_condition
1569 // indicating that we are currently handling popframe, so that
1570 // call_VMs that may happen later do not trigger new popframe
1571 // handling cycles.
1572 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1573 NOT_LP64(__ get_thread(thread));
1574 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1575 __ orl(rdx, JavaThread::popframe_processing_bit);
1576 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1577
1578 {
1579 // Check to see whether we are returning to a deoptimized frame.
1580 // (The PopFrame call ensures that the caller of the popped frame is
1581 // either interpreted or compiled and deoptimizes it if compiled.)
1582 // In this case, we can't call dispatch_next() after the frame is
1583 // popped, but instead must save the incoming arguments and restore
1584 // them after deoptimization has occurred.
1585 //
1586 // Note that we don't compare the return PC against the
1587 // deoptimization blob's unpack entry because of the presence of
1588 // adapter frames in C2.
1589 Label caller_not_deoptimized;
1590 Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1591 __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1592 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1593 InterpreterRuntime::interpreter_contains), rarg);
1594 __ testl(rax, rax);
1595 __ jcc(Assembler::notZero, caller_not_deoptimized);
1596
1597 // Compute size of arguments for saving when returning to
1598 // deoptimized caller
1599 __ get_method(rax);
1600 __ movptr(rax, Address(rax, Method::const_offset()));
1601 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1602 size_of_parameters_offset())));
1603 __ shll(rax, Interpreter::logStackElementSize);
1604 __ restore_locals();
1605 __ subptr(rlocals, rax);
1606 __ addptr(rlocals, wordSize);
1607 // Save these arguments
1608 NOT_LP64(__ get_thread(thread));
1609 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1610 Deoptimization::
1611 popframe_preserve_args),
1612 thread, rax, rlocals);
1613
1614 __ remove_activation(vtos, rdx,
1615 /* throw_monitor_exception */ false,
1616 /* install_monitor_exception */ false,
1617 /* notify_jvmdi */ false);
1618
1619 // Inform deoptimization that it is responsible for restoring
1620 // these arguments
1621 NOT_LP64(__ get_thread(thread));
1622 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1623 JavaThread::popframe_force_deopt_reexecution_bit);
1624
1625 // Continue in deoptimization handler
1626 __ jmp(rdx);
1627
1628 __ bind(caller_not_deoptimized);
1629 }
1630
1631 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1632 /* throw_monitor_exception */ false,
1633 /* install_monitor_exception */ false,
1634 /* notify_jvmdi */ false);
1635
1636 // Finish with popframe handling
1637 // A previous I2C followed by a deoptimization might have moved the
1638 // outgoing arguments further up the stack. PopFrame expects the
1639 // mutations to those outgoing arguments to be preserved and other
1640 // constraints basically require this frame to look exactly as
1641 // though it had previously invoked an interpreted activation with
1642 // no space between the top of the expression stack (current
1643 // last_sp) and the top of stack. Rather than force deopt to
1644 // maintain this kind of invariant all the time we call a small
1645 // fixup routine to move the mutated arguments onto the top of our
1646 // expression stack if necessary.
1647 #ifndef _LP64
1648 __ mov(rax, rsp);
1649 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1650 __ get_thread(thread);
1651 // PC must point into interpreter here
1652 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1653 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1654 __ get_thread(thread);
1655 #else
1656 __ mov(c_rarg1, rsp);
1657 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1658 // PC must point into interpreter here
1659 __ set_last_Java_frame(noreg, rbp, __ pc());
1660 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1661 #endif
1662 __ reset_last_Java_frame(thread, true, true);
1663
1664 // Restore the last_sp and null it out
1665 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1666 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1667
1668 __ restore_bcp();
1669 __ restore_locals();
1670 // The method data pointer was incremented already during
1671 // call profiling. We have to restore the mdp for the current bcp.
1672 if (ProfileInterpreter) {
1673 __ set_method_data_pointer_for_bcp();
1674 }
1675
1676 // Clear the popframe condition flag
1677 NOT_LP64(__ get_thread(thread));
1678 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1679 JavaThread::popframe_inactive);
1680
1681 #if INCLUDE_JVMTI
1682 {
1683 Label L_done;
1684 const Register local0 = rlocals;
1685
1686 __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1687 __ jcc(Assembler::notEqual, L_done);
1688
1689 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1690 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1691
1692 __ get_method(rdx);
1693 __ movptr(rax, Address(local0, 0));
1694 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1695
1696 __ testptr(rax, rax);
1697 __ jcc(Assembler::zero, L_done);
1698
1699 __ movptr(Address(rbx, 0), rax);
1700 __ bind(L_done);
1701 }
1702 #endif // INCLUDE_JVMTI
1703
1704 __ dispatch_next(vtos);
1705 // end of PopFrame support
1706
1707 Interpreter::_remove_activation_entry = __ pc();
1708
1709 // preserve exception over this code sequence
1710 __ pop_ptr(rax);
1711 NOT_LP64(__ get_thread(thread));
1712 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1713 // remove the activation (without doing throws on illegalMonitorExceptions)
1714 __ remove_activation(vtos, rdx, false, true, false);
1715 // restore exception
1716 NOT_LP64(__ get_thread(thread));
1717 __ get_vm_result(rax, thread);
1718
1719 // In between activations - previous activation type unknown yet
1720 // compute continuation point - the continuation point expects the
1721 // following registers set up:
1722 //
1723 // rax: exception
1724 // rdx: return address/pc that threw exception
1725 // rsp: expression stack of caller
1726 // rbp: ebp of caller
1727 __ push(rax); // save exception
1728 __ push(rdx); // save return address
1729 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1730 SharedRuntime::exception_handler_for_return_address),
1731 thread, rdx);
1732 __ mov(rbx, rax); // save exception handler
1733 __ pop(rdx); // restore return address
1734 __ pop(rax); // restore exception
1735 // Note that an "issuing PC" is actually the next PC after the call
1736 __ jmp(rbx); // jump to exception
1737 // handler of caller
1738 }
1739
1740
1741 //
1742 // JVMTI ForceEarlyReturn support
1743 //
1744 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1745 address entry = __ pc();
1746
1747 __ restore_bcp();
1748 __ restore_locals();
1749 __ empty_expression_stack();
1750 __ load_earlyret_value(state); // 32 bits returns value in rdx, so don't reuse
1751
1752 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1753 NOT_LP64(__ get_thread(thread));
1754 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1755 Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1756
1757 // Clear the earlyret state
1758 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1759
1760 __ remove_activation(state, rsi,
1761 false, /* throw_monitor_exception */
1762 false, /* install_monitor_exception */
1763 true); /* notify_jvmdi */
1764 __ jmp(rsi);
1765
1766 return entry;
1767 } // end of ForceEarlyReturn support
1768
1769
1770 //-----------------------------------------------------------------------------
1771 // Helper for vtos entry point generation
1772
1773 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1774 address& bep,
1775 address& cep,
1776 address& sep,
1777 address& aep,
1778 address& iep,
1779 address& lep,
1780 address& fep,
1781 address& dep,
1782 address& vep) {
1783 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1784 Label L;
1785 aep = __ pc(); __ push_ptr(); __ jmp(L);
1786 #ifndef _LP64
1787 fep = __ pc(); __ push(ftos); __ jmp(L);
1788 dep = __ pc(); __ push(dtos); __ jmp(L);
1789 #else
1790 fep = __ pc(); __ push_f(xmm0); __ jmp(L);
1791 dep = __ pc(); __ push_d(xmm0); __ jmp(L);
1792 #endif // _LP64
1793 lep = __ pc(); __ push_l(); __ jmp(L);
1794 bep = cep = sep =
1795 iep = __ pc(); __ push_i();
1796 vep = __ pc();
1797 __ bind(L);
1798 generate_and_dispatch(t);
1799 }
1800
1801 //-----------------------------------------------------------------------------
1802
1803 // Non-product code
1804 #ifndef PRODUCT
1805
1806 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1807 address entry = __ pc();
1808
1809 #ifndef _LP64
1810 // prepare expression stack
1811 __ pop(rcx); // pop return address so expression stack is 'pure'
1812 __ push(state); // save tosca
1813
1814 // pass tosca registers as arguments & call tracer
1815 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1816 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1817 __ pop(state); // restore tosca
1818
1819 // return
1820 __ jmp(rcx);
1821 #else
1822 __ push(state);
1823 __ push(c_rarg0);
1824 __ push(c_rarg1);
1825 __ push(c_rarg2);
1826 __ push(c_rarg3);
1827 __ mov(c_rarg2, rax); // Pass itos
1828 #ifdef _WIN64
1829 __ movflt(xmm3, xmm0); // Pass ftos
1830 #endif
1831 __ call_VM(noreg,
1832 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1833 c_rarg1, c_rarg2, c_rarg3);
1834 __ pop(c_rarg3);
1835 __ pop(c_rarg2);
1836 __ pop(c_rarg1);
1837 __ pop(c_rarg0);
1838 __ pop(state);
1839 __ ret(0); // return from result handler
1840 #endif // _LP64
1841
1842 return entry;
1843 }
1844
1845 void TemplateInterpreterGenerator::count_bytecode() {
1846 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1847 }
1848
1849 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1850 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1851 }
1852
1853 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1854 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1855 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1856 __ orl(rbx,
1857 ((int) t->bytecode()) <<
1858 BytecodePairHistogram::log2_number_of_codes);
1859 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1860 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1861 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1862 }
1863
1864
1865 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1866 // Call a little run-time stub to avoid blow-up for each bytecode.
1867 // The run-time runtime saves the right registers, depending on
1868 // the tosca in-state for the given template.
1869
1870 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1871 "entry must have been generated");
1872 #ifndef _LP64
1873 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1874 #else
1875 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1876 __ andptr(rsp, -16); // align stack as required by ABI
1877 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1878 __ mov(rsp, r12); // restore sp
1879 __ reinit_heapbase();
1880 #endif // _LP64
1881 }
1882
1883
1884 void TemplateInterpreterGenerator::stop_interpreter_at() {
1885 Label L;
1886 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1887 StopInterpreterAt);
1888 __ jcc(Assembler::notEqual, L);
1889 __ int3();
1890 __ bind(L);
1891 }
1892 #endif // !PRODUCT