1 /*
2 * Copyright (c) 2003, 2017, 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 = JVMCI_ONLY(268) NOT_JVMCI(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 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
175 address entry = __ pc();
176
177 #ifndef _LP64
178 #ifdef COMPILER2
179 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
180 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
181 for (int i = 1; i < 8; i++) {
182 __ ffree(i);
183 }
184 } else if (UseSSE < 2) {
185 __ empty_FPU_stack();
186 }
187 #endif // COMPILER2
188 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
189 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
190 } else {
191 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
192 }
193
194 if (state == ftos) {
195 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
196 } else if (state == dtos) {
197 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
198 }
199 #endif // _LP64
200
201 // Restore stack bottom in case i2c adjusted stack
202 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
203 // and NULL it as marker that esp is now tos until next java call
204 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
205
206 __ restore_bcp();
207 __ restore_locals();
208
209 if (state == atos) {
210 Register mdp = rbx;
211 Register tmp = rcx;
212 __ profile_return_type(mdp, rax, tmp);
213 }
214
215 const Register cache = rbx;
216 const Register index = rcx;
217 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
218
219 const Register flags = cache;
220 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
221 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
222 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
223
224 const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
225 if (JvmtiExport::can_pop_frame()) {
226 NOT_LP64(__ get_thread(java_thread));
227 __ check_and_handle_popframe(java_thread);
228 }
229 if (JvmtiExport::can_force_early_return()) {
230 NOT_LP64(__ get_thread(java_thread));
231 __ check_and_handle_earlyret(java_thread);
232 }
233
234 __ dispatch_next(state, step);
235
236 return entry;
237 }
238
239
240 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
241 address entry = __ pc();
242
243 #ifndef _LP64
244 if (state == ftos) {
245 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
246 } else if (state == dtos) {
247 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
248 }
249 #endif // _LP64
250
251 // NULL last_sp until next java call
252 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
253 __ restore_bcp();
254 __ restore_locals();
255 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
256 NOT_LP64(__ get_thread(thread));
257 #if INCLUDE_JVMCI
258 // Check if we need to take lock at entry of synchronized method. This can
259 // only occur on method entry so emit it only for vtos with step 0.
260 if ((UseJVMCICompiler || UseAOT) && state == vtos && step == 0) {
261 Label L;
262 __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
263 __ jcc(Assembler::zero, L);
264 // Clear flag.
265 __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
266 // Satisfy calling convention for lock_method().
267 __ get_method(rbx);
268 // Take lock.
269 lock_method();
270 __ bind(L);
271 } else {
272 #ifdef ASSERT
273 if (UseJVMCICompiler) {
274 Label L;
275 __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
276 __ jccb(Assembler::zero, L);
277 __ stop("unexpected pending monitor in deopt entry");
278 __ bind(L);
279 }
280 #endif
281 }
282 #endif
283 // handle exceptions
284 {
285 Label L;
286 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
287 __ jcc(Assembler::zero, L);
288 __ call_VM(noreg,
289 CAST_FROM_FN_PTR(address,
290 InterpreterRuntime::throw_pending_exception));
291 __ should_not_reach_here();
292 __ bind(L);
293 }
294 __ dispatch_next(state, step);
295 return entry;
296 }
297
298 address TemplateInterpreterGenerator::generate_result_handler_for(
299 BasicType type) {
300 address entry = __ pc();
301 switch (type) {
302 case T_BOOLEAN: __ c2bool(rax); break;
303 #ifndef _LP64
304 case T_CHAR : __ andptr(rax, 0xFFFF); break;
305 #else
306 case T_CHAR : __ movzwl(rax, rax); break;
307 #endif // _LP64
308 case T_BYTE : __ sign_extend_byte(rax); break;
309 case T_SHORT : __ sign_extend_short(rax); break;
310 case T_INT : /* nothing to do */ break;
311 case T_LONG : /* nothing to do */ break;
312 case T_VOID : /* nothing to do */ break;
313 #ifndef _LP64
314 case T_DOUBLE :
315 case T_FLOAT :
316 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
317 __ pop(t); // remove return address first
318 // Must return a result for interpreter or compiler. In SSE
319 // mode, results are returned in xmm0 and the FPU stack must
320 // be empty.
321 if (type == T_FLOAT && UseSSE >= 1) {
322 // Load ST0
323 __ fld_d(Address(rsp, 0));
324 // Store as float and empty fpu stack
325 __ fstp_s(Address(rsp, 0));
326 // and reload
327 __ movflt(xmm0, Address(rsp, 0));
328 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
329 __ movdbl(xmm0, Address(rsp, 0));
330 } else {
331 // restore ST0
332 __ fld_d(Address(rsp, 0));
333 }
334 // and pop the temp
335 __ addptr(rsp, 2 * wordSize);
336 __ push(t); // restore return address
337 }
338 break;
339 #else
340 case T_FLOAT : /* nothing to do */ break;
341 case T_DOUBLE : /* nothing to do */ break;
342 #endif // _LP64
343
344 case T_VALUETYPE: // fall through (value types are handled with oops)
345 case T_OBJECT :
346 // retrieve result from frame
347 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
348 // and verify it
349 __ verify_oop(rax);
350 break;
351 default : ShouldNotReachHere();
352 }
353 __ ret(0); // return from result handler
354 return entry;
355 }
356
357 address TemplateInterpreterGenerator::generate_safept_entry_for(
358 TosState state,
359 address runtime_entry) {
360 address entry = __ pc();
361 __ push(state);
362 __ call_VM(noreg, runtime_entry);
363 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
364 return entry;
365 }
366
367
368
369 // Helpers for commoning out cases in the various type of method entries.
370 //
371
372
373 // increment invocation count & check for overflow
374 //
375 // Note: checking for negative value instead of overflow
376 // so we have a 'sticky' overflow test
377 //
378 // rbx: method
379 // rcx: invocation counter
380 //
381 void TemplateInterpreterGenerator::generate_counter_incr(
382 Label* overflow,
383 Label* profile_method,
384 Label* profile_method_continue) {
385 Label done;
386 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
387 if (TieredCompilation) {
388 int increment = InvocationCounter::count_increment;
389 Label no_mdo;
390 if (ProfileInterpreter) {
391 // Are we profiling?
392 __ movptr(rax, Address(rbx, Method::method_data_offset()));
393 __ testptr(rax, rax);
394 __ jccb(Assembler::zero, no_mdo);
395 // Increment counter in the MDO
396 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
397 in_bytes(InvocationCounter::counter_offset()));
398 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
399 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
400 __ jmp(done);
401 }
402 __ bind(no_mdo);
403 // Increment counter in MethodCounters
404 const Address invocation_counter(rax,
405 MethodCounters::invocation_counter_offset() +
406 InvocationCounter::counter_offset());
407 __ get_method_counters(rbx, rax, done);
408 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
409 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
410 false, Assembler::zero, overflow);
411 __ bind(done);
412 } else { // not TieredCompilation
413 const Address backedge_counter(rax,
414 MethodCounters::backedge_counter_offset() +
415 InvocationCounter::counter_offset());
416 const Address invocation_counter(rax,
417 MethodCounters::invocation_counter_offset() +
418 InvocationCounter::counter_offset());
419
420 __ get_method_counters(rbx, rax, done);
421
422 if (ProfileInterpreter) {
423 __ incrementl(Address(rax,
424 MethodCounters::interpreter_invocation_counter_offset()));
425 }
426 // Update standard invocation counters
427 __ movl(rcx, invocation_counter);
428 __ incrementl(rcx, InvocationCounter::count_increment);
429 __ movl(invocation_counter, rcx); // save invocation count
430
431 __ movl(rax, backedge_counter); // load backedge counter
432 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
433
434 __ addl(rcx, rax); // add both counters
435
436 // profile_method is non-null only for interpreted method so
437 // profile_method != NULL == !native_call
438
439 if (ProfileInterpreter && profile_method != NULL) {
440 // Test to see if we should create a method data oop
441 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
442 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
443 __ jcc(Assembler::less, *profile_method_continue);
444
445 // if no method data exists, go to profile_method
446 __ test_method_data_pointer(rax, *profile_method);
447 }
448
449 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
450 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
451 __ jcc(Assembler::aboveEqual, *overflow);
452 __ bind(done);
453 }
454 }
455
456 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
457
458 // Asm interpreter on entry
459 // r14/rdi - locals
460 // r13/rsi - bcp
461 // rbx - method
462 // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
463 // rbp - interpreter frame
464
465 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
466 // Everything as it was on entry
467 // rdx is not restored. Doesn't appear to really be set.
468
469 // InterpreterRuntime::frequency_counter_overflow takes two
470 // arguments, the first (thread) is passed by call_VM, the second
471 // indicates if the counter overflow occurs at a backwards branch
472 // (NULL bcp). We pass zero for it. The call returns the address
473 // of the verified entry point for the method or NULL if the
474 // compilation did not complete (either went background or bailed
475 // out).
476 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
477 __ movl(rarg, 0);
478 __ call_VM(noreg,
479 CAST_FROM_FN_PTR(address,
480 InterpreterRuntime::frequency_counter_overflow),
481 rarg);
482
483 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
484 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
485 // and jump to the interpreted entry.
486 __ jmp(do_continue, relocInfo::none);
487 }
488
489 // See if we've got enough room on the stack for locals plus overhead below
490 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
491 // without going through the signal handler, i.e., reserved and yellow zones
492 // will not be made usable. The shadow zone must suffice to handle the
493 // overflow.
494 // The expression stack grows down incrementally, so the normal guard
495 // page mechanism will work for that.
496 //
497 // NOTE: Since the additional locals are also always pushed (wasn't
498 // obvious in generate_fixed_frame) so the guard should work for them
499 // too.
500 //
501 // Args:
502 // rdx: number of additional locals this frame needs (what we must check)
503 // rbx: Method*
504 //
505 // Kills:
506 // rax
507 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
508
509 // monitor entry size: see picture of stack in frame_x86.hpp
510 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
511
512 // total overhead size: entry_size + (saved rbp through expr stack
513 // bottom). be sure to change this if you add/subtract anything
514 // to/from the overhead area
515 const int overhead_size =
516 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
517
518 const int page_size = os::vm_page_size();
519
520 Label after_frame_check;
521
522 // see if the frame is greater than one page in size. If so,
523 // then we need to verify there is enough stack space remaining
524 // for the additional locals.
525 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
526 __ jcc(Assembler::belowEqual, after_frame_check);
527
528 // compute rsp as if this were going to be the last frame on
529 // the stack before the red zone
530
531 Label after_frame_check_pop;
532 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
533 #ifndef _LP64
534 __ push(thread);
535 __ get_thread(thread);
536 #endif
537
538 const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
539
540 // locals + overhead, in bytes
541 __ mov(rax, rdx);
542 __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
543 __ addptr(rax, overhead_size);
544
545 #ifdef ASSERT
546 Label limit_okay;
547 // Verify that thread stack overflow limit is non-zero.
548 __ cmpptr(stack_limit, (int32_t)NULL_WORD);
549 __ jcc(Assembler::notEqual, limit_okay);
550 __ stop("stack overflow limit is zero");
551 __ bind(limit_okay);
552 #endif
553
554 // Add locals/frame size to stack limit.
555 __ addptr(rax, stack_limit);
556
557 // Check against the current stack bottom.
558 __ cmpptr(rsp, rax);
559
560 __ jcc(Assembler::above, after_frame_check_pop);
561 NOT_LP64(__ pop(rsi)); // get saved bcp
562
563 // Restore sender's sp as SP. This is necessary if the sender's
564 // frame is an extended compiled frame (see gen_c2i_adapter())
565 // and safer anyway in case of JSR292 adaptations.
566
567 __ pop(rax); // return address must be moved if SP is changed
568 __ mov(rsp, rbcp);
569 __ push(rax);
570
571 // Note: the restored frame is not necessarily interpreted.
572 // Use the shared runtime version of the StackOverflowError.
573 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
574 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
575 // all done with frame size check
576 __ bind(after_frame_check_pop);
577 NOT_LP64(__ pop(rsi));
578
579 // all done with frame size check
580 __ bind(after_frame_check);
581 }
582
583 // Allocate monitor and lock method (asm interpreter)
584 //
585 // Args:
586 // rbx: Method*
587 // r14/rdi: locals
588 //
589 // Kills:
590 // rax
591 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
592 // rscratch1, rscratch2 (scratch regs)
593 void TemplateInterpreterGenerator::lock_method() {
594 // synchronize method
595 const Address access_flags(rbx, Method::access_flags_offset());
596 const Address monitor_block_top(
597 rbp,
598 frame::interpreter_frame_monitor_block_top_offset * wordSize);
599 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
600
601 #ifdef ASSERT
602 {
603 Label L;
604 __ movl(rax, access_flags);
605 __ testl(rax, JVM_ACC_SYNCHRONIZED);
606 __ jcc(Assembler::notZero, L);
607 __ stop("method doesn't need synchronization");
608 __ bind(L);
609 }
610 #endif // ASSERT
611
612 // get synchronization object
613 {
614 Label done;
615 __ movl(rax, access_flags);
616 __ testl(rax, JVM_ACC_STATIC);
617 // get receiver (assume this is frequent case)
618 __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
619 __ jcc(Assembler::zero, done);
620 __ load_mirror(rax, rbx);
621
622 #ifdef ASSERT
623 {
624 Label L;
625 __ testptr(rax, rax);
626 __ jcc(Assembler::notZero, L);
627 __ stop("synchronization object is NULL");
628 __ bind(L);
629 }
630 #endif // ASSERT
631
632 __ bind(done);
633 }
634
635 // add space for monitor & lock
636 __ subptr(rsp, entry_size); // add space for a monitor entry
637 __ movptr(monitor_block_top, rsp); // set new monitor block top
638 // store object
639 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
640 const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
641 __ movptr(lockreg, rsp); // object address
642 __ lock_object(lockreg);
643 }
644
645 // Generate a fixed interpreter frame. This is identical setup for
646 // interpreted methods and for native methods hence the shared code.
647 //
648 // Args:
649 // rax: return address
650 // rbx: Method*
651 // r14/rdi: pointer to locals
652 // r13/rsi: sender sp
653 // rdx: cp cache
654 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
655 // initialize fixed part of activation frame
656 __ push(rax); // save return address
657 __ enter(); // save old & set new rbp
658 __ push(rbcp); // set sender sp
659 __ push((int)NULL_WORD); // leave last_sp as null
660 __ movptr(rbcp, Address(rbx, Method::const_offset())); // get ConstMethod*
661 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
662 __ push(rbx); // save Method*
663 // Get mirror and store it in the frame as GC root for this Method*
664 __ load_mirror(rdx, rbx);
665 __ push(rdx);
666 if (ProfileInterpreter) {
667 Label method_data_continue;
668 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
669 __ testptr(rdx, rdx);
670 __ jcc(Assembler::zero, method_data_continue);
671 __ addptr(rdx, in_bytes(MethodData::data_offset()));
672 __ bind(method_data_continue);
673 __ push(rdx); // set the mdp (method data pointer)
674 } else {
675 __ push(0);
676 }
677
678 __ movptr(rdx, Address(rbx, Method::const_offset()));
679 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
680 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
681 __ push(rdx); // set constant pool cache
682 __ push(rlocals); // set locals pointer
683 if (native_call) {
684 __ push(0); // no bcp
685 } else {
686 __ push(rbcp); // set bcp
687 }
688 __ push(0); // reserve word for pointer to expression stack bottom
689 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
690 }
691
692 // End of helpers
693
694 // Method entry for java.lang.ref.Reference.get.
695 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
696 #if INCLUDE_ALL_GCS
697 // Code: _aload_0, _getfield, _areturn
698 // parameter size = 1
699 //
700 // The code that gets generated by this routine is split into 2 parts:
701 // 1. The "intrinsified" code for G1 (or any SATB based GC),
702 // 2. The slow path - which is an expansion of the regular method entry.
703 //
704 // Notes:-
705 // * In the G1 code we do not check whether we need to block for
706 // a safepoint. If G1 is enabled then we must execute the specialized
707 // code for Reference.get (except when the Reference object is null)
708 // so that we can log the value in the referent field with an SATB
709 // update buffer.
710 // If the code for the getfield template is modified so that the
711 // G1 pre-barrier code is executed when the current method is
712 // Reference.get() then going through the normal method entry
713 // will be fine.
714 // * The G1 code can, however, check the receiver object (the instance
715 // of java.lang.Reference) and jump to the slow path if null. If the
716 // Reference object is null then we obviously cannot fetch the referent
717 // and so we don't need to call the G1 pre-barrier. Thus we can use the
718 // regular method entry code to generate the NPE.
719 //
720 // rbx: Method*
721
722 // r13: senderSP must preserve for slow path, set SP to it on fast path
723
724 address entry = __ pc();
725
726 const int referent_offset = java_lang_ref_Reference::referent_offset;
727 guarantee(referent_offset > 0, "referent offset not initialized");
728
729 if (UseG1GC) {
730 Label slow_path;
731 // rbx: method
732
733 // Check if local 0 != NULL
734 // If the receiver is null then it is OK to jump to the slow path.
735 __ movptr(rax, Address(rsp, wordSize));
736
737 __ testptr(rax, rax);
738 __ jcc(Assembler::zero, slow_path);
739
740 // rax: local 0
741 // rbx: method (but can be used as scratch now)
742 // rdx: scratch
743 // rdi: scratch
744
745 // Preserve the sender sp in case the pre-barrier
746 // calls the runtime
747 NOT_LP64(__ push(rsi));
748
749 // Generate the G1 pre-barrier code to log the value of
750 // the referent field in an SATB buffer.
751
752 // Load the value of the referent field.
753 const Address field_address(rax, referent_offset);
754 __ load_heap_oop(rax, field_address);
755
756 const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
757 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
758 NOT_LP64(__ get_thread(thread));
759
760 // Generate the G1 pre-barrier code to log the value of
761 // the referent field in an SATB buffer.
762 __ g1_write_barrier_pre(noreg /* obj */,
763 rax /* pre_val */,
764 thread /* thread */,
765 rbx /* tmp */,
766 true /* tosca_live */,
767 true /* expand_call */);
768
769 // _areturn
770 NOT_LP64(__ pop(rsi)); // get sender sp
771 __ pop(rdi); // get return address
772 __ mov(rsp, sender_sp); // set sp to sender sp
773 __ jmp(rdi);
774 __ ret(0);
775
776 // generate a vanilla interpreter entry as the slow path
777 __ bind(slow_path);
778 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
779 return entry;
780 }
781 #endif // INCLUDE_ALL_GCS
782
783 // If G1 is not enabled then attempt to go through the accessor entry point
784 // Reference.get is an accessor
785 return NULL;
786 }
787
788 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
789 // Quick & dirty stack overflow checking: bang the stack & handle trap.
790 // Note that we do the banging after the frame is setup, since the exception
791 // handling code expects to find a valid interpreter frame on the stack.
792 // Doing the banging earlier fails if the caller frame is not an interpreter
793 // frame.
794 // (Also, the exception throwing code expects to unlock any synchronized
795 // method receiever, so do the banging after locking the receiver.)
796
797 // Bang each page in the shadow zone. We can't assume it's been done for
798 // an interpreter frame with greater than a page of locals, so each page
799 // needs to be checked. Only true for non-native.
800 if (UseStackBanging) {
801 const int page_size = os::vm_page_size();
802 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
803 const int start_page = native_call ? n_shadow_pages : 1;
804 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
805 __ bang_stack_with_offset(pages*page_size);
806 }
807 }
808 }
809
810 // Interpreter stub for calling a native method. (asm interpreter)
811 // This sets up a somewhat different looking stack for calling the
812 // native method than the typical interpreter frame setup.
813 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
814 // determine code generation flags
815 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
816
817 // rbx: Method*
818 // rbcp: sender sp
819
820 address entry_point = __ pc();
821
822 const Address constMethod (rbx, Method::const_offset());
823 const Address access_flags (rbx, Method::access_flags_offset());
824 const Address size_of_parameters(rcx, ConstMethod::
825 size_of_parameters_offset());
826
827
828 // get parameter size (always needed)
829 __ movptr(rcx, constMethod);
830 __ load_unsigned_short(rcx, size_of_parameters);
831
832 // native calls don't need the stack size check since they have no
833 // expression stack and the arguments are already on the stack and
834 // we only add a handful of words to the stack
835
836 // rbx: Method*
837 // rcx: size of parameters
838 // rbcp: sender sp
839 __ pop(rax); // get return address
840
841 // for natives the size of locals is zero
842
843 // compute beginning of parameters
844 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
845
846 // add 2 zero-initialized slots for native calls
847 // initialize result_handler slot
848 __ push((int) NULL_WORD);
849 // slot for oop temp
850 // (static native method holder mirror/jni oop result)
851 __ push((int) NULL_WORD);
852
853 // initialize fixed part of activation frame
854 generate_fixed_frame(true);
855
856 // make sure method is native & not abstract
857 #ifdef ASSERT
858 __ movl(rax, access_flags);
859 {
860 Label L;
861 __ testl(rax, JVM_ACC_NATIVE);
862 __ jcc(Assembler::notZero, L);
863 __ stop("tried to execute non-native method as native");
864 __ bind(L);
865 }
866 {
867 Label L;
868 __ testl(rax, JVM_ACC_ABSTRACT);
869 __ jcc(Assembler::zero, L);
870 __ stop("tried to execute abstract method in interpreter");
871 __ bind(L);
872 }
873 #endif
874
875 // Since at this point in the method invocation the exception handler
876 // would try to exit the monitor of synchronized methods which hasn't
877 // been entered yet, we set the thread local variable
878 // _do_not_unlock_if_synchronized to true. The remove_activation will
879 // check this flag.
880
881 const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
882 NOT_LP64(__ get_thread(thread1));
883 const Address do_not_unlock_if_synchronized(thread1,
884 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
885 __ movbool(do_not_unlock_if_synchronized, true);
886
887 // increment invocation count & check for overflow
888 Label invocation_counter_overflow;
889 if (inc_counter) {
890 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
891 }
892
893 Label continue_after_compile;
894 __ bind(continue_after_compile);
895
896 bang_stack_shadow_pages(true);
897
898 // reset the _do_not_unlock_if_synchronized flag
899 NOT_LP64(__ get_thread(thread1));
900 __ movbool(do_not_unlock_if_synchronized, false);
901
902 // check for synchronized methods
903 // Must happen AFTER invocation_counter check and stack overflow check,
904 // so method is not locked if overflows.
905 if (synchronized) {
906 lock_method();
907 } else {
908 // no synchronization necessary
909 #ifdef ASSERT
910 {
911 Label L;
912 __ movl(rax, access_flags);
913 __ testl(rax, JVM_ACC_SYNCHRONIZED);
914 __ jcc(Assembler::zero, L);
915 __ stop("method needs synchronization");
916 __ bind(L);
917 }
918 #endif
919 }
920
921 // start execution
922 #ifdef ASSERT
923 {
924 Label L;
925 const Address monitor_block_top(rbp,
926 frame::interpreter_frame_monitor_block_top_offset * wordSize);
927 __ movptr(rax, monitor_block_top);
928 __ cmpptr(rax, rsp);
929 __ jcc(Assembler::equal, L);
930 __ stop("broken stack frame setup in interpreter");
931 __ bind(L);
932 }
933 #endif
934
935 // jvmti support
936 __ notify_method_entry();
937
938 // work registers
939 const Register method = rbx;
940 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
941 const Register t = NOT_LP64(rcx) LP64_ONLY(r11);
942
943 // allocate space for parameters
944 __ get_method(method);
945 __ movptr(t, Address(method, Method::const_offset()));
946 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
947
948 #ifndef _LP64
949 __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
950 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
951 __ subptr(rsp, t);
952 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
953 #else
954 __ shll(t, Interpreter::logStackElementSize);
955
956 __ subptr(rsp, t);
957 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
958 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
959 #endif // _LP64
960
961 // get signature handler
962 {
963 Label L;
964 __ movptr(t, Address(method, Method::signature_handler_offset()));
965 __ testptr(t, t);
966 __ jcc(Assembler::notZero, L);
967 __ call_VM(noreg,
968 CAST_FROM_FN_PTR(address,
969 InterpreterRuntime::prepare_native_call),
970 method);
971 __ get_method(method);
972 __ movptr(t, Address(method, Method::signature_handler_offset()));
973 __ bind(L);
974 }
975
976 // call signature handler
977 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
978 "adjust this code");
979 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
980 "adjust this code");
981 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
982 "adjust this code");
983
984 // The generated handlers do not touch RBX (the method oop).
985 // However, large signatures cannot be cached and are generated
986 // each time here. The slow-path generator can do a GC on return,
987 // so we must reload it after the call.
988 __ call(t);
989 __ get_method(method); // slow path can do a GC, reload RBX
990
991
992 // result handler is in rax
993 // set result handler
994 __ movptr(Address(rbp,
995 (frame::interpreter_frame_result_handler_offset) * wordSize),
996 rax);
997
998 // pass mirror handle if static call
999 {
1000 Label L;
1001 __ movl(t, Address(method, Method::access_flags_offset()));
1002 __ testl(t, JVM_ACC_STATIC);
1003 __ jcc(Assembler::zero, L);
1004 // get mirror
1005 __ load_mirror(t, method);
1006 // copy mirror into activation frame
1007 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1008 t);
1009 // pass handle to mirror
1010 #ifndef _LP64
1011 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1012 __ movptr(Address(rsp, wordSize), t);
1013 #else
1014 __ lea(c_rarg1,
1015 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1016 #endif // _LP64
1017 __ bind(L);
1018 }
1019
1020 // get native function entry point
1021 {
1022 Label L;
1023 __ movptr(rax, Address(method, Method::native_function_offset()));
1024 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1025 __ cmpptr(rax, unsatisfied.addr());
1026 __ jcc(Assembler::notEqual, L);
1027 __ call_VM(noreg,
1028 CAST_FROM_FN_PTR(address,
1029 InterpreterRuntime::prepare_native_call),
1030 method);
1031 __ get_method(method);
1032 __ movptr(rax, Address(method, Method::native_function_offset()));
1033 __ bind(L);
1034 }
1035
1036 // pass JNIEnv
1037 #ifndef _LP64
1038 __ get_thread(thread);
1039 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1040 __ movptr(Address(rsp, 0), t);
1041
1042 // set_last_Java_frame_before_call
1043 // It is enough that the pc()
1044 // points into the right code segment. It does not have to be the correct return pc.
1045 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1046 #else
1047 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1048
1049 // It is enough that the pc() points into the right code
1050 // segment. It does not have to be the correct return pc.
1051 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1052 #endif // _LP64
1053
1054 // change thread state
1055 #ifdef ASSERT
1056 {
1057 Label L;
1058 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1059 __ cmpl(t, _thread_in_Java);
1060 __ jcc(Assembler::equal, L);
1061 __ stop("Wrong thread state in native stub");
1062 __ bind(L);
1063 }
1064 #endif
1065
1066 // Change state to native
1067
1068 __ movl(Address(thread, JavaThread::thread_state_offset()),
1069 _thread_in_native);
1070
1071 // Call the native method.
1072 __ call(rax);
1073 // 32: result potentially in rdx:rax or ST0
1074 // 64: result potentially in rax or xmm0
1075
1076 // Verify or restore cpu control state after JNI call
1077 __ restore_cpu_control_state_after_jni();
1078
1079 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1080 // in order to extract the result of a method call. If the order of these
1081 // pushes change or anything else is added to the stack then the code in
1082 // interpreter_frame_result must also change.
1083
1084 #ifndef _LP64
1085 // save potential result in ST(0) & rdx:rax
1086 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1087 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1088 // It is safe to do this push because state is _thread_in_native and return address will be found
1089 // via _last_native_pc and not via _last_jave_sp
1090
1091 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1092 // If the order changes or anything else is added to the stack the code in
1093 // interpreter_frame_result will have to be changed.
1094
1095 { Label L;
1096 Label push_double;
1097 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1098 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1099 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1100 float_handler.addr());
1101 __ jcc(Assembler::equal, push_double);
1102 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1103 double_handler.addr());
1104 __ jcc(Assembler::notEqual, L);
1105 __ bind(push_double);
1106 __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1107 __ bind(L);
1108 }
1109 #else
1110 __ push(dtos);
1111 #endif // _LP64
1112
1113 __ push(ltos);
1114
1115 // change thread state
1116 NOT_LP64(__ get_thread(thread));
1117 __ movl(Address(thread, JavaThread::thread_state_offset()),
1118 _thread_in_native_trans);
1119
1120 if (os::is_MP()) {
1121 if (UseMembar) {
1122 // Force this write out before the read below
1123 __ membar(Assembler::Membar_mask_bits(
1124 Assembler::LoadLoad | Assembler::LoadStore |
1125 Assembler::StoreLoad | Assembler::StoreStore));
1126 } else {
1127 // Write serialization page so VM thread can do a pseudo remote membar.
1128 // We use the current thread pointer to calculate a thread specific
1129 // offset to write to within the page. This minimizes bus traffic
1130 // due to cache line collision.
1131 __ serialize_memory(thread, rcx);
1132 }
1133 }
1134
1135 #ifndef _LP64
1136 if (AlwaysRestoreFPU) {
1137 // Make sure the control word is correct.
1138 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1139 }
1140 #endif // _LP64
1141
1142 // check for safepoint operation in progress and/or pending suspend requests
1143 {
1144 Label Continue;
1145 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1146 SafepointSynchronize::_not_synchronized);
1147
1148 Label L;
1149 __ jcc(Assembler::notEqual, L);
1150 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1151 __ jcc(Assembler::equal, Continue);
1152 __ bind(L);
1153
1154 // Don't use call_VM as it will see a possible pending exception
1155 // and forward it and never return here preventing us from
1156 // clearing _last_native_pc down below. Also can't use
1157 // call_VM_leaf either as it will check to see if r13 & r14 are
1158 // preserved and correspond to the bcp/locals pointers. So we do a
1159 // runtime call by hand.
1160 //
1161 #ifndef _LP64
1162 __ push(thread);
1163 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1164 JavaThread::check_special_condition_for_native_trans)));
1165 __ increment(rsp, wordSize);
1166 __ get_thread(thread);
1167 #else
1168 __ mov(c_rarg0, r15_thread);
1169 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1170 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1171 __ andptr(rsp, -16); // align stack as required by ABI
1172 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1173 __ mov(rsp, r12); // restore sp
1174 __ reinit_heapbase();
1175 #endif // _LP64
1176 __ bind(Continue);
1177 }
1178
1179 // change thread state
1180 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1181
1182 // reset_last_Java_frame
1183 __ reset_last_Java_frame(thread, true);
1184
1185 if (CheckJNICalls) {
1186 // clear_pending_jni_exception_check
1187 __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1188 }
1189
1190 // reset handle block
1191 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1192 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1193
1194 // If result is an oop unbox and store it in frame where gc will see it
1195 // and result handler will pick it up
1196
1197 {
1198 Label no_oop, not_weak, store_result;
1199 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1200 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1201 __ jcc(Assembler::notEqual, no_oop);
1202 // retrieve result
1203 __ pop(ltos);
1204 // Unbox oop result, e.g. JNIHandles::resolve value.
1205 __ resolve_jobject(rax /* value */,
1206 thread /* thread */,
1207 t /* tmp */);
1208 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1209 // keep stack depth as expected by pushing oop which will eventually be discarded
1210 __ push(ltos);
1211 __ bind(no_oop);
1212 }
1213
1214
1215 {
1216 Label no_reguard;
1217 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1218 JavaThread::stack_guard_yellow_reserved_disabled);
1219 __ jcc(Assembler::notEqual, no_reguard);
1220
1221 __ pusha(); // XXX only save smashed registers
1222 #ifndef _LP64
1223 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1224 __ popa();
1225 #else
1226 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1227 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1228 __ andptr(rsp, -16); // align stack as required by ABI
1229 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1230 __ mov(rsp, r12); // restore sp
1231 __ popa(); // XXX only restore smashed registers
1232 __ reinit_heapbase();
1233 #endif // _LP64
1234
1235 __ bind(no_reguard);
1236 }
1237
1238
1239 // The method register is junk from after the thread_in_native transition
1240 // until here. Also can't call_VM until the bcp has been
1241 // restored. Need bcp for throwing exception below so get it now.
1242 __ get_method(method);
1243
1244 // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1245 __ movptr(rbcp, Address(method, Method::const_offset())); // get ConstMethod*
1246 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
1247
1248 // handle exceptions (exception handling will handle unlocking!)
1249 {
1250 Label L;
1251 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1252 __ jcc(Assembler::zero, L);
1253 // Note: At some point we may want to unify this with the code
1254 // used in call_VM_base(); i.e., we should use the
1255 // StubRoutines::forward_exception code. For now this doesn't work
1256 // here because the rsp is not correctly set at this point.
1257 __ MacroAssembler::call_VM(noreg,
1258 CAST_FROM_FN_PTR(address,
1259 InterpreterRuntime::throw_pending_exception));
1260 __ should_not_reach_here();
1261 __ bind(L);
1262 }
1263
1264 // do unlocking if necessary
1265 {
1266 Label L;
1267 __ movl(t, Address(method, Method::access_flags_offset()));
1268 __ testl(t, JVM_ACC_SYNCHRONIZED);
1269 __ jcc(Assembler::zero, L);
1270 // the code below should be shared with interpreter macro
1271 // assembler implementation
1272 {
1273 Label unlock;
1274 // BasicObjectLock will be first in list, since this is a
1275 // synchronized method. However, need to check that the object
1276 // has not been unlocked by an explicit monitorexit bytecode.
1277 const Address monitor(rbp,
1278 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1279 wordSize - (int)sizeof(BasicObjectLock)));
1280
1281 const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1282
1283 // monitor expect in c_rarg1 for slow unlock path
1284 __ lea(regmon, monitor); // address of first monitor
1285
1286 __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1287 __ testptr(t, t);
1288 __ jcc(Assembler::notZero, unlock);
1289
1290 // Entry already unlocked, need to throw exception
1291 __ MacroAssembler::call_VM(noreg,
1292 CAST_FROM_FN_PTR(address,
1293 InterpreterRuntime::throw_illegal_monitor_state_exception));
1294 __ should_not_reach_here();
1295
1296 __ bind(unlock);
1297 __ unlock_object(regmon);
1298 }
1299 __ bind(L);
1300 }
1301
1302 // jvmti support
1303 // Note: This must happen _after_ handling/throwing any exceptions since
1304 // the exception handler code notifies the runtime of method exits
1305 // too. If this happens before, method entry/exit notifications are
1306 // not properly paired (was bug - gri 11/22/99).
1307 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1308
1309 // restore potential result in edx:eax, call result handler to
1310 // restore potential result in ST0 & handle result
1311
1312 __ pop(ltos);
1313 LP64_ONLY( __ pop(dtos));
1314
1315 __ movptr(t, Address(rbp,
1316 (frame::interpreter_frame_result_handler_offset) * wordSize));
1317 __ call(t);
1318
1319 // remove activation
1320 __ movptr(t, Address(rbp,
1321 frame::interpreter_frame_sender_sp_offset *
1322 wordSize)); // get sender sp
1323 __ leave(); // remove frame anchor
1324 __ pop(rdi); // get return address
1325 __ mov(rsp, t); // set sp to sender sp
1326 __ jmp(rdi);
1327
1328 if (inc_counter) {
1329 // Handle overflow of counter and compile method
1330 __ bind(invocation_counter_overflow);
1331 generate_counter_overflow(continue_after_compile);
1332 }
1333
1334 return entry_point;
1335 }
1336
1337 // Abstract method entry
1338 // Attempt to execute abstract method. Throw exception
1339 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1340
1341 address entry_point = __ pc();
1342
1343 // abstract method entry
1344
1345 // pop return address, reset last_sp to NULL
1346 __ empty_expression_stack();
1347 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1348 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1349
1350 // throw exception
1351 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
1352 // the call_VM checks for exception, so we should never return here.
1353 __ should_not_reach_here();
1354
1355 return entry_point;
1356 }
1357
1358 //
1359 // Generic interpreted method entry to (asm) interpreter
1360 //
1361 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1362 // determine code generation flags
1363 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1364
1365 // ebx: Method*
1366 // rbcp: sender sp
1367 address entry_point = __ pc();
1368
1369 const Address constMethod(rbx, Method::const_offset());
1370 const Address access_flags(rbx, Method::access_flags_offset());
1371 const Address size_of_parameters(rdx,
1372 ConstMethod::size_of_parameters_offset());
1373 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1374
1375
1376 // get parameter size (always needed)
1377 __ movptr(rdx, constMethod);
1378 __ load_unsigned_short(rcx, size_of_parameters);
1379
1380 // rbx: Method*
1381 // rcx: size of parameters
1382 // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1383
1384 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1385 __ subl(rdx, rcx); // rdx = no. of additional locals
1386
1387 // YYY
1388 // __ incrementl(rdx);
1389 // __ andl(rdx, -2);
1390
1391 // see if we've got enough room on the stack for locals plus overhead.
1392 generate_stack_overflow_check();
1393
1394 // get return address
1395 __ pop(rax);
1396
1397 // compute beginning of parameters
1398 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1399
1400 // rdx - # of additional locals
1401 // allocate space for locals
1402 // explicitly initialize locals
1403 {
1404 Label exit, loop;
1405 __ testl(rdx, rdx);
1406 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1407 __ bind(loop);
1408 __ push((int) NULL_WORD); // initialize local variables
1409 __ decrementl(rdx); // until everything initialized
1410 __ jcc(Assembler::greater, loop);
1411 __ bind(exit);
1412 }
1413
1414 // initialize fixed part of activation frame
1415 generate_fixed_frame(false);
1416
1417 // make sure method is not native & not abstract
1418 #ifdef ASSERT
1419 __ movl(rax, access_flags);
1420 {
1421 Label L;
1422 __ testl(rax, JVM_ACC_NATIVE);
1423 __ jcc(Assembler::zero, L);
1424 __ stop("tried to execute native method as non-native");
1425 __ bind(L);
1426 }
1427 {
1428 Label L;
1429 __ testl(rax, JVM_ACC_ABSTRACT);
1430 __ jcc(Assembler::zero, L);
1431 __ stop("tried to execute abstract method in interpreter");
1432 __ bind(L);
1433 }
1434 #endif
1435
1436 // Since at this point in the method invocation the exception
1437 // handler would try to exit the monitor of synchronized methods
1438 // which hasn't been entered yet, we set the thread local variable
1439 // _do_not_unlock_if_synchronized to true. The remove_activation
1440 // will check this flag.
1441
1442 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1443 NOT_LP64(__ get_thread(thread));
1444 const Address do_not_unlock_if_synchronized(thread,
1445 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1446 __ movbool(do_not_unlock_if_synchronized, true);
1447
1448 __ profile_parameters_type(rax, rcx, rdx);
1449 // increment invocation count & check for overflow
1450 Label invocation_counter_overflow;
1451 Label profile_method;
1452 Label profile_method_continue;
1453 if (inc_counter) {
1454 generate_counter_incr(&invocation_counter_overflow,
1455 &profile_method,
1456 &profile_method_continue);
1457 if (ProfileInterpreter) {
1458 __ bind(profile_method_continue);
1459 }
1460 }
1461
1462 Label continue_after_compile;
1463 __ bind(continue_after_compile);
1464
1465 // check for synchronized interpreted methods
1466 bang_stack_shadow_pages(false);
1467
1468 // reset the _do_not_unlock_if_synchronized flag
1469 NOT_LP64(__ get_thread(thread));
1470 __ movbool(do_not_unlock_if_synchronized, false);
1471
1472 // check for synchronized methods
1473 // Must happen AFTER invocation_counter check and stack overflow check,
1474 // so method is not locked if overflows.
1475 if (synchronized) {
1476 // Allocate monitor and lock method
1477 lock_method();
1478 } else {
1479 // no synchronization necessary
1480 #ifdef ASSERT
1481 {
1482 Label L;
1483 __ movl(rax, access_flags);
1484 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1485 __ jcc(Assembler::zero, L);
1486 __ stop("method needs synchronization");
1487 __ bind(L);
1488 }
1489 #endif
1490 }
1491
1492 // start execution
1493 #ifdef ASSERT
1494 {
1495 Label L;
1496 const Address monitor_block_top (rbp,
1497 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1498 __ movptr(rax, monitor_block_top);
1499 __ cmpptr(rax, rsp);
1500 __ jcc(Assembler::equal, L);
1501 __ stop("broken stack frame setup in interpreter");
1502 __ bind(L);
1503 }
1504 #endif
1505
1506 // jvmti support
1507 __ notify_method_entry();
1508
1509 __ dispatch_next(vtos);
1510
1511 // invocation counter overflow
1512 if (inc_counter) {
1513 if (ProfileInterpreter) {
1514 // We have decided to profile this method in the interpreter
1515 __ bind(profile_method);
1516 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1517 __ set_method_data_pointer_for_bcp();
1518 __ get_method(rbx);
1519 __ jmp(profile_method_continue);
1520 }
1521 // Handle overflow of counter and compile method
1522 __ bind(invocation_counter_overflow);
1523 generate_counter_overflow(continue_after_compile);
1524 }
1525
1526 return entry_point;
1527 }
1528
1529 //-----------------------------------------------------------------------------
1530 // Exceptions
1531
1532 void TemplateInterpreterGenerator::generate_throw_exception() {
1533 // Entry point in previous activation (i.e., if the caller was
1534 // interpreted)
1535 Interpreter::_rethrow_exception_entry = __ pc();
1536 // Restore sp to interpreter_frame_last_sp even though we are going
1537 // to empty the expression stack for the exception processing.
1538 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1539 // rax: exception
1540 // rdx: return address/pc that threw exception
1541 __ restore_bcp(); // r13/rsi points to call/send
1542 __ restore_locals();
1543 LP64_ONLY(__ reinit_heapbase()); // restore r12 as heapbase.
1544 // Entry point for exceptions thrown within interpreter code
1545 Interpreter::_throw_exception_entry = __ pc();
1546 // expression stack is undefined here
1547 // rax: exception
1548 // r13/rsi: exception bcp
1549 __ verify_oop(rax);
1550 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1551 LP64_ONLY(__ mov(c_rarg1, rax));
1552
1553 // expression stack must be empty before entering the VM in case of
1554 // an exception
1555 __ empty_expression_stack();
1556 // find exception handler address and preserve exception oop
1557 __ call_VM(rdx,
1558 CAST_FROM_FN_PTR(address,
1559 InterpreterRuntime::exception_handler_for_exception),
1560 rarg);
1561 // rax: exception handler entry point
1562 // rdx: preserved exception oop
1563 // r13/rsi: bcp for exception handler
1564 __ push_ptr(rdx); // push exception which is now the only value on the stack
1565 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1566
1567 // If the exception is not handled in the current frame the frame is
1568 // removed and the exception is rethrown (i.e. exception
1569 // continuation is _rethrow_exception).
1570 //
1571 // Note: At this point the bci is still the bxi for the instruction
1572 // which caused the exception and the expression stack is
1573 // empty. Thus, for any VM calls at this point, GC will find a legal
1574 // oop map (with empty expression stack).
1575
1576 // In current activation
1577 // tos: exception
1578 // esi: exception bcp
1579
1580 //
1581 // JVMTI PopFrame support
1582 //
1583
1584 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1585 __ empty_expression_stack();
1586 // Set the popframe_processing bit in pending_popframe_condition
1587 // indicating that we are currently handling popframe, so that
1588 // call_VMs that may happen later do not trigger new popframe
1589 // handling cycles.
1590 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1591 NOT_LP64(__ get_thread(thread));
1592 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1593 __ orl(rdx, JavaThread::popframe_processing_bit);
1594 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1595
1596 {
1597 // Check to see whether we are returning to a deoptimized frame.
1598 // (The PopFrame call ensures that the caller of the popped frame is
1599 // either interpreted or compiled and deoptimizes it if compiled.)
1600 // In this case, we can't call dispatch_next() after the frame is
1601 // popped, but instead must save the incoming arguments and restore
1602 // them after deoptimization has occurred.
1603 //
1604 // Note that we don't compare the return PC against the
1605 // deoptimization blob's unpack entry because of the presence of
1606 // adapter frames in C2.
1607 Label caller_not_deoptimized;
1608 Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1609 __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1610 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1611 InterpreterRuntime::interpreter_contains), rarg);
1612 __ testl(rax, rax);
1613 __ jcc(Assembler::notZero, caller_not_deoptimized);
1614
1615 // Compute size of arguments for saving when returning to
1616 // deoptimized caller
1617 __ get_method(rax);
1618 __ movptr(rax, Address(rax, Method::const_offset()));
1619 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1620 size_of_parameters_offset())));
1621 __ shll(rax, Interpreter::logStackElementSize);
1622 __ restore_locals();
1623 __ subptr(rlocals, rax);
1624 __ addptr(rlocals, wordSize);
1625 // Save these arguments
1626 NOT_LP64(__ get_thread(thread));
1627 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1628 Deoptimization::
1629 popframe_preserve_args),
1630 thread, rax, rlocals);
1631
1632 __ remove_activation(vtos, rdx,
1633 /* throw_monitor_exception */ false,
1634 /* install_monitor_exception */ false,
1635 /* notify_jvmdi */ false);
1636
1637 // Inform deoptimization that it is responsible for restoring
1638 // these arguments
1639 NOT_LP64(__ get_thread(thread));
1640 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1641 JavaThread::popframe_force_deopt_reexecution_bit);
1642
1643 // Continue in deoptimization handler
1644 __ jmp(rdx);
1645
1646 __ bind(caller_not_deoptimized);
1647 }
1648
1649 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1650 /* throw_monitor_exception */ false,
1651 /* install_monitor_exception */ false,
1652 /* notify_jvmdi */ false);
1653
1654 // Finish with popframe handling
1655 // A previous I2C followed by a deoptimization might have moved the
1656 // outgoing arguments further up the stack. PopFrame expects the
1657 // mutations to those outgoing arguments to be preserved and other
1658 // constraints basically require this frame to look exactly as
1659 // though it had previously invoked an interpreted activation with
1660 // no space between the top of the expression stack (current
1661 // last_sp) and the top of stack. Rather than force deopt to
1662 // maintain this kind of invariant all the time we call a small
1663 // fixup routine to move the mutated arguments onto the top of our
1664 // expression stack if necessary.
1665 #ifndef _LP64
1666 __ mov(rax, rsp);
1667 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1668 __ get_thread(thread);
1669 // PC must point into interpreter here
1670 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1671 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1672 __ get_thread(thread);
1673 #else
1674 __ mov(c_rarg1, rsp);
1675 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1676 // PC must point into interpreter here
1677 __ set_last_Java_frame(noreg, rbp, __ pc());
1678 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1679 #endif
1680 __ reset_last_Java_frame(thread, true);
1681
1682 // Restore the last_sp and null it out
1683 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1684 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1685
1686 __ restore_bcp();
1687 __ restore_locals();
1688 // The method data pointer was incremented already during
1689 // call profiling. We have to restore the mdp for the current bcp.
1690 if (ProfileInterpreter) {
1691 __ set_method_data_pointer_for_bcp();
1692 }
1693
1694 // Clear the popframe condition flag
1695 NOT_LP64(__ get_thread(thread));
1696 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1697 JavaThread::popframe_inactive);
1698
1699 #if INCLUDE_JVMTI
1700 {
1701 Label L_done;
1702 const Register local0 = rlocals;
1703
1704 __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1705 __ jcc(Assembler::notEqual, L_done);
1706
1707 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1708 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1709
1710 __ get_method(rdx);
1711 __ movptr(rax, Address(local0, 0));
1712 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1713
1714 __ testptr(rax, rax);
1715 __ jcc(Assembler::zero, L_done);
1716
1717 __ movptr(Address(rbx, 0), rax);
1718 __ bind(L_done);
1719 }
1720 #endif // INCLUDE_JVMTI
1721
1722 __ dispatch_next(vtos);
1723 // end of PopFrame support
1724
1725 Interpreter::_remove_activation_entry = __ pc();
1726
1727 // preserve exception over this code sequence
1728 __ pop_ptr(rax);
1729 NOT_LP64(__ get_thread(thread));
1730 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1731 // remove the activation (without doing throws on illegalMonitorExceptions)
1732 __ remove_activation(vtos, rdx, false, true, false);
1733 // restore exception
1734 NOT_LP64(__ get_thread(thread));
1735 __ get_vm_result(rax, thread);
1736
1737 // In between activations - previous activation type unknown yet
1738 // compute continuation point - the continuation point expects the
1739 // following registers set up:
1740 //
1741 // rax: exception
1742 // rdx: return address/pc that threw exception
1743 // rsp: expression stack of caller
1744 // rbp: ebp of caller
1745 __ push(rax); // save exception
1746 __ push(rdx); // save return address
1747 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1748 SharedRuntime::exception_handler_for_return_address),
1749 thread, rdx);
1750 __ mov(rbx, rax); // save exception handler
1751 __ pop(rdx); // restore return address
1752 __ pop(rax); // restore exception
1753 // Note that an "issuing PC" is actually the next PC after the call
1754 __ jmp(rbx); // jump to exception
1755 // handler of caller
1756 }
1757
1758
1759 //
1760 // JVMTI ForceEarlyReturn support
1761 //
1762 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1763 address entry = __ pc();
1764
1765 __ restore_bcp();
1766 __ restore_locals();
1767 __ empty_expression_stack();
1768 __ load_earlyret_value(state); // 32 bits returns value in rdx, so don't reuse
1769
1770 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1771 NOT_LP64(__ get_thread(thread));
1772 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1773 Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1774
1775 // Clear the earlyret state
1776 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1777
1778 __ remove_activation(state, rsi,
1779 false, /* throw_monitor_exception */
1780 false, /* install_monitor_exception */
1781 true); /* notify_jvmdi */
1782 __ jmp(rsi);
1783
1784 return entry;
1785 } // end of ForceEarlyReturn support
1786
1787
1788 //-----------------------------------------------------------------------------
1789 // Helper for vtos entry point generation
1790
1791 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1792 address& bep,
1793 address& cep,
1794 address& sep,
1795 address& aep,
1796 address& iep,
1797 address& lep,
1798 address& fep,
1799 address& dep,
1800 address& qep,
1801 address& vep) {
1802 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1803 Label L;
1804 aep = __ pc(); __ push_ptr(); __ jmp(L);
1805 qep = __ pc(); __ push_ptr(); __ jmp(L);
1806 #ifndef _LP64
1807 fep = __ pc(); __ push(ftos); __ jmp(L);
1808 dep = __ pc(); __ push(dtos); __ jmp(L);
1809 #else
1810 fep = __ pc(); __ push_f(xmm0); __ jmp(L);
1811 dep = __ pc(); __ push_d(xmm0); __ jmp(L);
1812 #endif // _LP64
1813 lep = __ pc(); __ push_l(); __ jmp(L);
1814 bep = cep = sep =
1815 iep = __ pc(); __ push_i();
1816 vep = __ pc();
1817 __ bind(L);
1818 generate_and_dispatch(t);
1819 }
1820
1821 //-----------------------------------------------------------------------------
1822
1823 // Non-product code
1824 #ifndef PRODUCT
1825
1826 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1827 address entry = __ pc();
1828
1829 #ifndef _LP64
1830 // prepare expression stack
1831 __ pop(rcx); // pop return address so expression stack is 'pure'
1832 __ push(state); // save tosca
1833
1834 // pass tosca registers as arguments & call tracer
1835 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1836 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1837 __ pop(state); // restore tosca
1838
1839 // return
1840 __ jmp(rcx);
1841 #else
1842 __ push(state);
1843 __ push(c_rarg0);
1844 __ push(c_rarg1);
1845 __ push(c_rarg2);
1846 __ push(c_rarg3);
1847 __ mov(c_rarg2, rax); // Pass itos
1848 #ifdef _WIN64
1849 __ movflt(xmm3, xmm0); // Pass ftos
1850 #endif
1851 __ call_VM(noreg,
1852 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1853 c_rarg1, c_rarg2, c_rarg3);
1854 __ pop(c_rarg3);
1855 __ pop(c_rarg2);
1856 __ pop(c_rarg1);
1857 __ pop(c_rarg0);
1858 __ pop(state);
1859 __ ret(0); // return from result handler
1860 #endif // _LP64
1861
1862 return entry;
1863 }
1864
1865 void TemplateInterpreterGenerator::count_bytecode() {
1866 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1867 }
1868
1869 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1870 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1871 }
1872
1873 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1874 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1875 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1876 __ orl(rbx,
1877 ((int) t->bytecode()) <<
1878 BytecodePairHistogram::log2_number_of_codes);
1879 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1880 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1881 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1882 }
1883
1884
1885 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1886 // Call a little run-time stub to avoid blow-up for each bytecode.
1887 // The run-time runtime saves the right registers, depending on
1888 // the tosca in-state for the given template.
1889
1890 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1891 "entry must have been generated");
1892 #ifndef _LP64
1893 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1894 #else
1895 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1896 __ andptr(rsp, -16); // align stack as required by ABI
1897 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1898 __ mov(rsp, r12); // restore sp
1899 __ reinit_heapbase();
1900 #endif // _LP64
1901 }
1902
1903
1904 void TemplateInterpreterGenerator::stop_interpreter_at() {
1905 Label L;
1906 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1907 StopInterpreterAt);
1908 __ jcc(Assembler::notEqual, L);
1909 __ int3();
1910 __ bind(L);
1911 }
1912 #endif // !PRODUCT