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 if (state == qtos && ValueTypeReturnedAsFields) {
207 // A value type is being returned. If fields are in registers we
208 // need to allocate a value type instance and initialize it with
209 // the value of the fields.
210 __ super_call_VM_leaf(StubRoutines::store_value_type_fields_to_buf());
211 }
212
213 __ restore_bcp();
214 __ restore_locals();
215
216 if (state == atos) {
217 Register mdp = rbx;
218 Register tmp = rcx;
219 __ profile_return_type(mdp, rax, tmp);
220 }
221
222 const Register cache = rbx;
223 const Register index = rcx;
224 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
225
226 const Register flags = cache;
227 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
228 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
229 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
230
231 const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
232 if (JvmtiExport::can_pop_frame()) {
233 NOT_LP64(__ get_thread(java_thread));
234 __ check_and_handle_popframe(java_thread);
235 }
236 if (JvmtiExport::can_force_early_return()) {
237 NOT_LP64(__ get_thread(java_thread));
238 __ check_and_handle_earlyret(java_thread);
239 }
240
241 __ dispatch_next(state, step);
242
243 return entry;
244 }
245
246
247 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
248 address entry = __ pc();
249
250 #ifndef _LP64
251 if (state == ftos) {
252 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
253 } else if (state == dtos) {
254 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
255 }
256 #endif // _LP64
257
258 // NULL last_sp until next java call
259 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
260 __ restore_bcp();
261 __ restore_locals();
262 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
263 NOT_LP64(__ get_thread(thread));
264 #if INCLUDE_JVMCI
265 // Check if we need to take lock at entry of synchronized method. This can
266 // only occur on method entry so emit it only for vtos with step 0.
267 if ((UseJVMCICompiler || UseAOT) && state == vtos && step == 0) {
268 Label L;
269 __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
270 __ jcc(Assembler::zero, L);
271 // Clear flag.
272 __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
273 // Satisfy calling convention for lock_method().
274 __ get_method(rbx);
275 // Take lock.
276 lock_method();
277 __ bind(L);
278 } else {
279 #ifdef ASSERT
280 if (UseJVMCICompiler) {
281 Label L;
282 __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
283 __ jccb(Assembler::zero, L);
284 __ stop("unexpected pending monitor in deopt entry");
285 __ bind(L);
286 }
287 #endif
288 }
289 #endif
290 // handle exceptions
291 {
292 Label L;
293 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
294 __ jcc(Assembler::zero, L);
295 __ call_VM(noreg,
296 CAST_FROM_FN_PTR(address,
297 InterpreterRuntime::throw_pending_exception));
298 __ should_not_reach_here();
299 __ bind(L);
300 }
301 __ dispatch_next(state, step);
302 return entry;
303 }
304
305 address TemplateInterpreterGenerator::generate_result_handler_for(
306 BasicType type) {
307 address entry = __ pc();
308 switch (type) {
309 case T_BOOLEAN: __ c2bool(rax); break;
310 #ifndef _LP64
311 case T_CHAR : __ andptr(rax, 0xFFFF); break;
312 #else
313 case T_CHAR : __ movzwl(rax, rax); break;
314 #endif // _LP64
315 case T_BYTE : __ sign_extend_byte(rax); break;
316 case T_SHORT : __ sign_extend_short(rax); break;
317 case T_INT : /* nothing to do */ break;
318 case T_LONG : /* nothing to do */ break;
319 case T_VOID : /* nothing to do */ break;
320 #ifndef _LP64
321 case T_DOUBLE :
322 case T_FLOAT :
323 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
324 __ pop(t); // remove return address first
325 // Must return a result for interpreter or compiler. In SSE
326 // mode, results are returned in xmm0 and the FPU stack must
327 // be empty.
328 if (type == T_FLOAT && UseSSE >= 1) {
329 // Load ST0
330 __ fld_d(Address(rsp, 0));
331 // Store as float and empty fpu stack
332 __ fstp_s(Address(rsp, 0));
333 // and reload
334 __ movflt(xmm0, Address(rsp, 0));
335 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
336 __ movdbl(xmm0, Address(rsp, 0));
337 } else {
338 // restore ST0
339 __ fld_d(Address(rsp, 0));
340 }
341 // and pop the temp
342 __ addptr(rsp, 2 * wordSize);
343 __ push(t); // restore return address
344 }
345 break;
346 #else
347 case T_FLOAT : /* nothing to do */ break;
348 case T_DOUBLE : /* nothing to do */ break;
349 #endif // _LP64
350
351 case T_VALUETYPE: // fall through (value types are handled with oops)
352 case T_OBJECT :
353 // retrieve result from frame
354 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
355 // and verify it
356 __ verify_oop(rax);
357 break;
358 default : ShouldNotReachHere();
359 }
360 __ ret(0); // return from result handler
361 return entry;
362 }
363
364 address TemplateInterpreterGenerator::generate_safept_entry_for(
365 TosState state,
366 address runtime_entry) {
367 address entry = __ pc();
368 __ push(state);
369 __ call_VM(noreg, runtime_entry);
370 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
371 return entry;
372 }
373
374
375
376 // Helpers for commoning out cases in the various type of method entries.
377 //
378
379
380 // increment invocation count & check for overflow
381 //
382 // Note: checking for negative value instead of overflow
383 // so we have a 'sticky' overflow test
384 //
385 // rbx: method
386 // rcx: invocation counter
387 //
388 void TemplateInterpreterGenerator::generate_counter_incr(
389 Label* overflow,
390 Label* profile_method,
391 Label* profile_method_continue) {
392 Label done;
393 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
394 if (TieredCompilation) {
395 int increment = InvocationCounter::count_increment;
396 Label no_mdo;
397 if (ProfileInterpreter) {
398 // Are we profiling?
399 __ movptr(rax, Address(rbx, Method::method_data_offset()));
400 __ testptr(rax, rax);
401 __ jccb(Assembler::zero, no_mdo);
402 // Increment counter in the MDO
403 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
404 in_bytes(InvocationCounter::counter_offset()));
405 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
406 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
407 __ jmp(done);
408 }
409 __ bind(no_mdo);
410 // Increment counter in MethodCounters
411 const Address invocation_counter(rax,
412 MethodCounters::invocation_counter_offset() +
413 InvocationCounter::counter_offset());
414 __ get_method_counters(rbx, rax, done);
415 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
416 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
417 false, Assembler::zero, overflow);
418 __ bind(done);
419 } else { // not TieredCompilation
420 const Address backedge_counter(rax,
421 MethodCounters::backedge_counter_offset() +
422 InvocationCounter::counter_offset());
423 const Address invocation_counter(rax,
424 MethodCounters::invocation_counter_offset() +
425 InvocationCounter::counter_offset());
426
427 __ get_method_counters(rbx, rax, done);
428
429 if (ProfileInterpreter) {
430 __ incrementl(Address(rax,
431 MethodCounters::interpreter_invocation_counter_offset()));
432 }
433 // Update standard invocation counters
434 __ movl(rcx, invocation_counter);
435 __ incrementl(rcx, InvocationCounter::count_increment);
436 __ movl(invocation_counter, rcx); // save invocation count
437
438 __ movl(rax, backedge_counter); // load backedge counter
439 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
440
441 __ addl(rcx, rax); // add both counters
442
443 // profile_method is non-null only for interpreted method so
444 // profile_method != NULL == !native_call
445
446 if (ProfileInterpreter && profile_method != NULL) {
447 // Test to see if we should create a method data oop
448 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
449 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
450 __ jcc(Assembler::less, *profile_method_continue);
451
452 // if no method data exists, go to profile_method
453 __ test_method_data_pointer(rax, *profile_method);
454 }
455
456 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
457 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
458 __ jcc(Assembler::aboveEqual, *overflow);
459 __ bind(done);
460 }
461 }
462
463 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
464
465 // Asm interpreter on entry
466 // r14/rdi - locals
467 // r13/rsi - bcp
468 // rbx - method
469 // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
470 // rbp - interpreter frame
471
472 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
473 // Everything as it was on entry
474 // rdx is not restored. Doesn't appear to really be set.
475
476 // InterpreterRuntime::frequency_counter_overflow takes two
477 // arguments, the first (thread) is passed by call_VM, the second
478 // indicates if the counter overflow occurs at a backwards branch
479 // (NULL bcp). We pass zero for it. The call returns the address
480 // of the verified entry point for the method or NULL if the
481 // compilation did not complete (either went background or bailed
482 // out).
483 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
484 __ movl(rarg, 0);
485 __ call_VM(noreg,
486 CAST_FROM_FN_PTR(address,
487 InterpreterRuntime::frequency_counter_overflow),
488 rarg);
489
490 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
491 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
492 // and jump to the interpreted entry.
493 __ jmp(do_continue, relocInfo::none);
494 }
495
496 // See if we've got enough room on the stack for locals plus overhead below
497 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
498 // without going through the signal handler, i.e., reserved and yellow zones
499 // will not be made usable. The shadow zone must suffice to handle the
500 // overflow.
501 // The expression stack grows down incrementally, so the normal guard
502 // page mechanism will work for that.
503 //
504 // NOTE: Since the additional locals are also always pushed (wasn't
505 // obvious in generate_fixed_frame) so the guard should work for them
506 // too.
507 //
508 // Args:
509 // rdx: number of additional locals this frame needs (what we must check)
510 // rbx: Method*
511 //
512 // Kills:
513 // rax
514 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
515
516 // monitor entry size: see picture of stack in frame_x86.hpp
517 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
518
519 // total overhead size: entry_size + (saved rbp through expr stack
520 // bottom). be sure to change this if you add/subtract anything
521 // to/from the overhead area
522 const int overhead_size =
523 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
524
525 const int page_size = os::vm_page_size();
526
527 Label after_frame_check;
528
529 // see if the frame is greater than one page in size. If so,
530 // then we need to verify there is enough stack space remaining
531 // for the additional locals.
532 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
533 __ jcc(Assembler::belowEqual, after_frame_check);
534
535 // compute rsp as if this were going to be the last frame on
536 // the stack before the red zone
537
538 Label after_frame_check_pop;
539 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
540 #ifndef _LP64
541 __ push(thread);
542 __ get_thread(thread);
543 #endif
544
545 const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
546
547 // locals + overhead, in bytes
548 __ mov(rax, rdx);
549 __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
550 __ addptr(rax, overhead_size);
551
552 #ifdef ASSERT
553 Label limit_okay;
554 // Verify that thread stack overflow limit is non-zero.
555 __ cmpptr(stack_limit, (int32_t)NULL_WORD);
556 __ jcc(Assembler::notEqual, limit_okay);
557 __ stop("stack overflow limit is zero");
558 __ bind(limit_okay);
559 #endif
560
561 // Add locals/frame size to stack limit.
562 __ addptr(rax, stack_limit);
563
564 // Check against the current stack bottom.
565 __ cmpptr(rsp, rax);
566
567 __ jcc(Assembler::above, after_frame_check_pop);
568 NOT_LP64(__ pop(rsi)); // get saved bcp
569
570 // Restore sender's sp as SP. This is necessary if the sender's
571 // frame is an extended compiled frame (see gen_c2i_adapter())
572 // and safer anyway in case of JSR292 adaptations.
573
574 __ pop(rax); // return address must be moved if SP is changed
575 __ mov(rsp, rbcp);
576 __ push(rax);
577
578 // Note: the restored frame is not necessarily interpreted.
579 // Use the shared runtime version of the StackOverflowError.
580 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
581 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
582 // all done with frame size check
583 __ bind(after_frame_check_pop);
584 NOT_LP64(__ pop(rsi));
585
586 // all done with frame size check
587 __ bind(after_frame_check);
588 }
589
590 // Allocate monitor and lock method (asm interpreter)
591 //
592 // Args:
593 // rbx: Method*
594 // r14/rdi: locals
595 //
596 // Kills:
597 // rax
598 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
599 // rscratch1, rscratch2 (scratch regs)
600 void TemplateInterpreterGenerator::lock_method() {
601 // synchronize method
602 const Address access_flags(rbx, Method::access_flags_offset());
603 const Address monitor_block_top(
604 rbp,
605 frame::interpreter_frame_monitor_block_top_offset * wordSize);
606 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
607
608 #ifdef ASSERT
609 {
610 Label L;
611 __ movl(rax, access_flags);
612 __ testl(rax, JVM_ACC_SYNCHRONIZED);
613 __ jcc(Assembler::notZero, L);
614 __ stop("method doesn't need synchronization");
615 __ bind(L);
616 }
617 #endif // ASSERT
618
619 // get synchronization object
620 {
621 Label done;
622 __ movl(rax, access_flags);
623 __ testl(rax, JVM_ACC_STATIC);
624 // get receiver (assume this is frequent case)
625 __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
626 __ jcc(Assembler::zero, done);
627 __ load_mirror(rax, rbx);
628
629 #ifdef ASSERT
630 {
631 Label L;
632 __ testptr(rax, rax);
633 __ jcc(Assembler::notZero, L);
634 __ stop("synchronization object is NULL");
635 __ bind(L);
636 }
637 #endif // ASSERT
638
639 __ bind(done);
640 }
641
642 // add space for monitor & lock
643 __ subptr(rsp, entry_size); // add space for a monitor entry
644 __ movptr(monitor_block_top, rsp); // set new monitor block top
645 // store object
646 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
647 const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
648 __ movptr(lockreg, rsp); // object address
649 __ lock_object(lockreg);
650 }
651
652 // Generate a fixed interpreter frame. This is identical setup for
653 // interpreted methods and for native methods hence the shared code.
654 //
655 // Args:
656 // rax: return address
657 // rbx: Method*
658 // r14/rdi: pointer to locals
659 // r13/rsi: sender sp
660 // rdx: cp cache
661 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
662 // initialize fixed part of activation frame
663 __ push(rax); // save return address
664 __ enter(); // save old & set new rbp
665 __ push(rbcp); // set sender sp
666 __ push((int)NULL_WORD); // leave last_sp as null
667 __ movptr(rbcp, Address(rbx, Method::const_offset())); // get ConstMethod*
668 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
669 __ push(rbx); // save Method*
670 // Get mirror and store it in the frame as GC root for this Method*
671 __ load_mirror(rdx, rbx);
672 __ push(rdx);
673 if (ProfileInterpreter) {
674 Label method_data_continue;
675 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
676 __ testptr(rdx, rdx);
677 __ jcc(Assembler::zero, method_data_continue);
678 __ addptr(rdx, in_bytes(MethodData::data_offset()));
679 __ bind(method_data_continue);
680 __ push(rdx); // set the mdp (method data pointer)
681 } else {
682 __ push(0);
683 }
684
685 __ movptr(rdx, Address(rbx, Method::const_offset()));
686 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
687 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
688 __ push(rdx); // set constant pool cache
689 const Register thread1 = NOT_LP64(rdx) LP64_ONLY(r15_thread);
690 NOT_LP64(__ get_thread(thread1));
691 __ movptr(rdx, Address(thread1, JavaThread::vt_alloc_ptr_offset()));
692 __ push(rdx); // value type allocation pointer when activation is created
693 __ push(rlocals); // set locals pointer
694 if (native_call) {
695 __ push(0); // no bcp
696 } else {
697 __ push(rbcp); // set bcp
698 }
699 __ push(0); // reserve word for pointer to expression stack bottom
700 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
701 }
702
703 // End of helpers
704
705 // Method entry for java.lang.ref.Reference.get.
706 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
707 #if INCLUDE_ALL_GCS
708 // Code: _aload_0, _getfield, _areturn
709 // parameter size = 1
710 //
711 // The code that gets generated by this routine is split into 2 parts:
712 // 1. The "intrinsified" code for G1 (or any SATB based GC),
713 // 2. The slow path - which is an expansion of the regular method entry.
714 //
715 // Notes:-
716 // * In the G1 code we do not check whether we need to block for
717 // a safepoint. If G1 is enabled then we must execute the specialized
718 // code for Reference.get (except when the Reference object is null)
719 // so that we can log the value in the referent field with an SATB
720 // update buffer.
721 // If the code for the getfield template is modified so that the
722 // G1 pre-barrier code is executed when the current method is
723 // Reference.get() then going through the normal method entry
724 // will be fine.
725 // * The G1 code can, however, check the receiver object (the instance
726 // of java.lang.Reference) and jump to the slow path if null. If the
727 // Reference object is null then we obviously cannot fetch the referent
728 // and so we don't need to call the G1 pre-barrier. Thus we can use the
729 // regular method entry code to generate the NPE.
730 //
731 // rbx: Method*
732
733 // r13: senderSP must preserve for slow path, set SP to it on fast path
734
735 address entry = __ pc();
736
737 const int referent_offset = java_lang_ref_Reference::referent_offset;
738 guarantee(referent_offset > 0, "referent offset not initialized");
739
740 if (UseG1GC) {
741 Label slow_path;
742 // rbx: method
743
744 // Check if local 0 != NULL
745 // If the receiver is null then it is OK to jump to the slow path.
746 __ movptr(rax, Address(rsp, wordSize));
747
748 __ testptr(rax, rax);
749 __ jcc(Assembler::zero, slow_path);
750
751 // rax: local 0
752 // rbx: method (but can be used as scratch now)
753 // rdx: scratch
754 // rdi: scratch
755
756 // Preserve the sender sp in case the pre-barrier
757 // calls the runtime
758 NOT_LP64(__ push(rsi));
759
760 // Generate the G1 pre-barrier code to log the value of
761 // the referent field in an SATB buffer.
762
763 // Load the value of the referent field.
764 const Address field_address(rax, referent_offset);
765 __ load_heap_oop(rax, field_address);
766
767 const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
768 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
769 NOT_LP64(__ get_thread(thread));
770
771 // Generate the G1 pre-barrier code to log the value of
772 // the referent field in an SATB buffer.
773 __ g1_write_barrier_pre(noreg /* obj */,
774 rax /* pre_val */,
775 thread /* thread */,
776 rbx /* tmp */,
777 true /* tosca_live */,
778 true /* expand_call */);
779
780 // _areturn
781 NOT_LP64(__ pop(rsi)); // get sender sp
782 __ pop(rdi); // get return address
783 __ mov(rsp, sender_sp); // set sp to sender sp
784 __ jmp(rdi);
785 __ ret(0);
786
787 // generate a vanilla interpreter entry as the slow path
788 __ bind(slow_path);
789 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
790 return entry;
791 }
792 #endif // INCLUDE_ALL_GCS
793
794 // If G1 is not enabled then attempt to go through the accessor entry point
795 // Reference.get is an accessor
796 return NULL;
797 }
798
799 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
800 // Quick & dirty stack overflow checking: bang the stack & handle trap.
801 // Note that we do the banging after the frame is setup, since the exception
802 // handling code expects to find a valid interpreter frame on the stack.
803 // Doing the banging earlier fails if the caller frame is not an interpreter
804 // frame.
805 // (Also, the exception throwing code expects to unlock any synchronized
806 // method receiever, so do the banging after locking the receiver.)
807
808 // Bang each page in the shadow zone. We can't assume it's been done for
809 // an interpreter frame with greater than a page of locals, so each page
810 // needs to be checked. Only true for non-native.
811 if (UseStackBanging) {
812 const int page_size = os::vm_page_size();
813 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
814 const int start_page = native_call ? n_shadow_pages : 1;
815 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
816 __ bang_stack_with_offset(pages*page_size);
817 }
818 }
819 }
820
821 // Interpreter stub for calling a native method. (asm interpreter)
822 // This sets up a somewhat different looking stack for calling the
823 // native method than the typical interpreter frame setup.
824 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
825 // determine code generation flags
826 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
827
828 // rbx: Method*
829 // rbcp: sender sp
830
831 address entry_point = __ pc();
832
833 const Address constMethod (rbx, Method::const_offset());
834 const Address access_flags (rbx, Method::access_flags_offset());
835 const Address size_of_parameters(rcx, ConstMethod::
836 size_of_parameters_offset());
837
838
839 // get parameter size (always needed)
840 __ movptr(rcx, constMethod);
841 __ load_unsigned_short(rcx, size_of_parameters);
842
843 // native calls don't need the stack size check since they have no
844 // expression stack and the arguments are already on the stack and
845 // we only add a handful of words to the stack
846
847 // rbx: Method*
848 // rcx: size of parameters
849 // rbcp: sender sp
850 __ pop(rax); // get return address
851
852 // for natives the size of locals is zero
853
854 // compute beginning of parameters
855 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
856
857 // add 2 zero-initialized slots for native calls
858 // initialize result_handler slot
859 __ push((int) NULL_WORD);
860 // slot for oop temp
861 // (static native method holder mirror/jni oop result)
862 __ push((int) NULL_WORD);
863
864 // initialize fixed part of activation frame
865 generate_fixed_frame(true);
866
867 // make sure method is native & not abstract
868 #ifdef ASSERT
869 __ movl(rax, access_flags);
870 {
871 Label L;
872 __ testl(rax, JVM_ACC_NATIVE);
873 __ jcc(Assembler::notZero, L);
874 __ stop("tried to execute non-native method as native");
875 __ bind(L);
876 }
877 {
878 Label L;
879 __ testl(rax, JVM_ACC_ABSTRACT);
880 __ jcc(Assembler::zero, L);
881 __ stop("tried to execute abstract method in interpreter");
882 __ bind(L);
883 }
884 #endif
885
886 // Since at this point in the method invocation the exception handler
887 // would try to exit the monitor of synchronized methods which hasn't
888 // been entered yet, we set the thread local variable
889 // _do_not_unlock_if_synchronized to true. The remove_activation will
890 // check this flag.
891
892 const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
893 NOT_LP64(__ get_thread(thread1));
894 const Address do_not_unlock_if_synchronized(thread1,
895 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
896 __ movbool(do_not_unlock_if_synchronized, true);
897
898 // increment invocation count & check for overflow
899 Label invocation_counter_overflow;
900 if (inc_counter) {
901 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
902 }
903
904 Label continue_after_compile;
905 __ bind(continue_after_compile);
906
907 bang_stack_shadow_pages(true);
908
909 // reset the _do_not_unlock_if_synchronized flag
910 NOT_LP64(__ get_thread(thread1));
911 __ movbool(do_not_unlock_if_synchronized, false);
912
913 // check for synchronized methods
914 // Must happen AFTER invocation_counter check and stack overflow check,
915 // so method is not locked if overflows.
916 if (synchronized) {
917 lock_method();
918 } else {
919 // no synchronization necessary
920 #ifdef ASSERT
921 {
922 Label L;
923 __ movl(rax, access_flags);
924 __ testl(rax, JVM_ACC_SYNCHRONIZED);
925 __ jcc(Assembler::zero, L);
926 __ stop("method needs synchronization");
927 __ bind(L);
928 }
929 #endif
930 }
931
932 // start execution
933 #ifdef ASSERT
934 {
935 Label L;
936 const Address monitor_block_top(rbp,
937 frame::interpreter_frame_monitor_block_top_offset * wordSize);
938 __ movptr(rax, monitor_block_top);
939 __ cmpptr(rax, rsp);
940 __ jcc(Assembler::equal, L);
941 __ stop("broken stack frame setup in interpreter");
942 __ bind(L);
943 }
944 #endif
945
946 // jvmti support
947 __ notify_method_entry();
948
949 // work registers
950 const Register method = rbx;
951 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
952 const Register t = NOT_LP64(rcx) LP64_ONLY(r11);
953
954 // allocate space for parameters
955 __ get_method(method);
956 __ movptr(t, Address(method, Method::const_offset()));
957 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
958
959 #ifndef _LP64
960 __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
961 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
962 __ subptr(rsp, t);
963 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
964 #else
965 __ shll(t, Interpreter::logStackElementSize);
966
967 __ subptr(rsp, t);
968 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
969 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
970 #endif // _LP64
971
972 // get signature handler
973 {
974 Label L;
975 __ movptr(t, Address(method, Method::signature_handler_offset()));
976 __ testptr(t, t);
977 __ jcc(Assembler::notZero, L);
978 __ call_VM(noreg,
979 CAST_FROM_FN_PTR(address,
980 InterpreterRuntime::prepare_native_call),
981 method);
982 __ get_method(method);
983 __ movptr(t, Address(method, Method::signature_handler_offset()));
984 __ bind(L);
985 }
986
987 // call signature handler
988 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
989 "adjust this code");
990 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
991 "adjust this code");
992 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
993 "adjust this code");
994
995 // The generated handlers do not touch RBX (the method oop).
996 // However, large signatures cannot be cached and are generated
997 // each time here. The slow-path generator can do a GC on return,
998 // so we must reload it after the call.
999 __ call(t);
1000 __ get_method(method); // slow path can do a GC, reload RBX
1001
1002
1003 // result handler is in rax
1004 // set result handler
1005 __ movptr(Address(rbp,
1006 (frame::interpreter_frame_result_handler_offset) * wordSize),
1007 rax);
1008
1009 // pass mirror handle if static call
1010 {
1011 Label L;
1012 __ movl(t, Address(method, Method::access_flags_offset()));
1013 __ testl(t, JVM_ACC_STATIC);
1014 __ jcc(Assembler::zero, L);
1015 // get mirror
1016 __ load_mirror(t, method);
1017 // copy mirror into activation frame
1018 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1019 t);
1020 // pass handle to mirror
1021 #ifndef _LP64
1022 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1023 __ movptr(Address(rsp, wordSize), t);
1024 #else
1025 __ lea(c_rarg1,
1026 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1027 #endif // _LP64
1028 __ bind(L);
1029 }
1030
1031 // get native function entry point
1032 {
1033 Label L;
1034 __ movptr(rax, Address(method, Method::native_function_offset()));
1035 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1036 __ cmpptr(rax, unsatisfied.addr());
1037 __ jcc(Assembler::notEqual, L);
1038 __ call_VM(noreg,
1039 CAST_FROM_FN_PTR(address,
1040 InterpreterRuntime::prepare_native_call),
1041 method);
1042 __ get_method(method);
1043 __ movptr(rax, Address(method, Method::native_function_offset()));
1044 __ bind(L);
1045 }
1046
1047 // pass JNIEnv
1048 #ifndef _LP64
1049 __ get_thread(thread);
1050 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1051 __ movptr(Address(rsp, 0), t);
1052
1053 // set_last_Java_frame_before_call
1054 // It is enough that the pc()
1055 // points into the right code segment. It does not have to be the correct return pc.
1056 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1057 #else
1058 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1059
1060 // It is enough that the pc() points into the right code
1061 // segment. It does not have to be the correct return pc.
1062 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1063 #endif // _LP64
1064
1065 // change thread state
1066 #ifdef ASSERT
1067 {
1068 Label L;
1069 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1070 __ cmpl(t, _thread_in_Java);
1071 __ jcc(Assembler::equal, L);
1072 __ stop("Wrong thread state in native stub");
1073 __ bind(L);
1074 }
1075 #endif
1076
1077 // Change state to native
1078
1079 __ movl(Address(thread, JavaThread::thread_state_offset()),
1080 _thread_in_native);
1081
1082 // Call the native method.
1083 __ call(rax);
1084 // 32: result potentially in rdx:rax or ST0
1085 // 64: result potentially in rax or xmm0
1086
1087 // Verify or restore cpu control state after JNI call
1088 __ restore_cpu_control_state_after_jni();
1089
1090 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1091 // in order to extract the result of a method call. If the order of these
1092 // pushes change or anything else is added to the stack then the code in
1093 // interpreter_frame_result must also change.
1094
1095 #ifndef _LP64
1096 // save potential result in ST(0) & rdx:rax
1097 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1098 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1099 // It is safe to do this push because state is _thread_in_native and return address will be found
1100 // via _last_native_pc and not via _last_jave_sp
1101
1102 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1103 // If the order changes or anything else is added to the stack the code in
1104 // interpreter_frame_result will have to be changed.
1105
1106 { Label L;
1107 Label push_double;
1108 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1109 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1110 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1111 float_handler.addr());
1112 __ jcc(Assembler::equal, push_double);
1113 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1114 double_handler.addr());
1115 __ jcc(Assembler::notEqual, L);
1116 __ bind(push_double);
1117 __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1118 __ bind(L);
1119 }
1120 #else
1121 __ push(dtos);
1122 #endif // _LP64
1123
1124 __ push(ltos);
1125
1126 // change thread state
1127 NOT_LP64(__ get_thread(thread));
1128 __ movl(Address(thread, JavaThread::thread_state_offset()),
1129 _thread_in_native_trans);
1130
1131 if (os::is_MP()) {
1132 if (UseMembar) {
1133 // Force this write out before the read below
1134 __ membar(Assembler::Membar_mask_bits(
1135 Assembler::LoadLoad | Assembler::LoadStore |
1136 Assembler::StoreLoad | Assembler::StoreStore));
1137 } else {
1138 // Write serialization page so VM thread can do a pseudo remote membar.
1139 // We use the current thread pointer to calculate a thread specific
1140 // offset to write to within the page. This minimizes bus traffic
1141 // due to cache line collision.
1142 __ serialize_memory(thread, rcx);
1143 }
1144 }
1145
1146 #ifndef _LP64
1147 if (AlwaysRestoreFPU) {
1148 // Make sure the control word is correct.
1149 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1150 }
1151 #endif // _LP64
1152
1153 // check for safepoint operation in progress and/or pending suspend requests
1154 {
1155 Label Continue;
1156 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1157 SafepointSynchronize::_not_synchronized);
1158
1159 Label L;
1160 __ jcc(Assembler::notEqual, L);
1161 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1162 __ jcc(Assembler::equal, Continue);
1163 __ bind(L);
1164
1165 // Don't use call_VM as it will see a possible pending exception
1166 // and forward it and never return here preventing us from
1167 // clearing _last_native_pc down below. Also can't use
1168 // call_VM_leaf either as it will check to see if r13 & r14 are
1169 // preserved and correspond to the bcp/locals pointers. So we do a
1170 // runtime call by hand.
1171 //
1172 #ifndef _LP64
1173 __ push(thread);
1174 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1175 JavaThread::check_special_condition_for_native_trans)));
1176 __ increment(rsp, wordSize);
1177 __ get_thread(thread);
1178 #else
1179 __ mov(c_rarg0, r15_thread);
1180 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1181 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1182 __ andptr(rsp, -16); // align stack as required by ABI
1183 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1184 __ mov(rsp, r12); // restore sp
1185 __ reinit_heapbase();
1186 #endif // _LP64
1187 __ bind(Continue);
1188 }
1189
1190 // change thread state
1191 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1192
1193 // reset_last_Java_frame
1194 __ reset_last_Java_frame(thread, true);
1195
1196 if (CheckJNICalls) {
1197 // clear_pending_jni_exception_check
1198 __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1199 }
1200
1201 // reset handle block
1202 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1203 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1204
1205 // If result is an oop unbox and store it in frame where gc will see it
1206 // and result handler will pick it up
1207
1208 {
1209 Label no_oop, not_weak, store_result;
1210 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1211 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1212 __ jcc(Assembler::notEqual, no_oop);
1213 // retrieve result
1214 __ pop(ltos);
1215 // Unbox oop result, e.g. JNIHandles::resolve value.
1216 __ resolve_jobject(rax /* value */,
1217 thread /* thread */,
1218 t /* tmp */);
1219 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1220 // keep stack depth as expected by pushing oop which will eventually be discarded
1221 __ push(ltos);
1222 __ bind(no_oop);
1223 }
1224
1225
1226 {
1227 Label no_reguard;
1228 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1229 JavaThread::stack_guard_yellow_reserved_disabled);
1230 __ jcc(Assembler::notEqual, no_reguard);
1231
1232 __ pusha(); // XXX only save smashed registers
1233 #ifndef _LP64
1234 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1235 __ popa();
1236 #else
1237 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1238 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1239 __ andptr(rsp, -16); // align stack as required by ABI
1240 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1241 __ mov(rsp, r12); // restore sp
1242 __ popa(); // XXX only restore smashed registers
1243 __ reinit_heapbase();
1244 #endif // _LP64
1245
1246 __ bind(no_reguard);
1247 }
1248
1249
1250 // The method register is junk from after the thread_in_native transition
1251 // until here. Also can't call_VM until the bcp has been
1252 // restored. Need bcp for throwing exception below so get it now.
1253 __ get_method(method);
1254
1255 // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1256 __ movptr(rbcp, Address(method, Method::const_offset())); // get ConstMethod*
1257 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
1258
1259 // handle exceptions (exception handling will handle unlocking!)
1260 {
1261 Label L;
1262 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1263 __ jcc(Assembler::zero, L);
1264 // Note: At some point we may want to unify this with the code
1265 // used in call_VM_base(); i.e., we should use the
1266 // StubRoutines::forward_exception code. For now this doesn't work
1267 // here because the rsp is not correctly set at this point.
1268 __ MacroAssembler::call_VM(noreg,
1269 CAST_FROM_FN_PTR(address,
1270 InterpreterRuntime::throw_pending_exception));
1271 __ should_not_reach_here();
1272 __ bind(L);
1273 }
1274
1275 // do unlocking if necessary
1276 {
1277 Label L;
1278 __ movl(t, Address(method, Method::access_flags_offset()));
1279 __ testl(t, JVM_ACC_SYNCHRONIZED);
1280 __ jcc(Assembler::zero, L);
1281 // the code below should be shared with interpreter macro
1282 // assembler implementation
1283 {
1284 Label unlock;
1285 // BasicObjectLock will be first in list, since this is a
1286 // synchronized method. However, need to check that the object
1287 // has not been unlocked by an explicit monitorexit bytecode.
1288 const Address monitor(rbp,
1289 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1290 wordSize - (int)sizeof(BasicObjectLock)));
1291
1292 const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1293
1294 // monitor expect in c_rarg1 for slow unlock path
1295 __ lea(regmon, monitor); // address of first monitor
1296
1297 __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1298 __ testptr(t, t);
1299 __ jcc(Assembler::notZero, unlock);
1300
1301 // Entry already unlocked, need to throw exception
1302 __ MacroAssembler::call_VM(noreg,
1303 CAST_FROM_FN_PTR(address,
1304 InterpreterRuntime::throw_illegal_monitor_state_exception));
1305 __ should_not_reach_here();
1306
1307 __ bind(unlock);
1308 __ unlock_object(regmon);
1309 }
1310 __ bind(L);
1311 }
1312
1313 // jvmti support
1314 // Note: This must happen _after_ handling/throwing any exceptions since
1315 // the exception handler code notifies the runtime of method exits
1316 // too. If this happens before, method entry/exit notifications are
1317 // not properly paired (was bug - gri 11/22/99).
1318 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1319
1320 // restore potential result in edx:eax, call result handler to
1321 // restore potential result in ST0 & handle result
1322
1323 __ pop(ltos);
1324 LP64_ONLY( __ pop(dtos));
1325
1326 __ movptr(t, Address(rbp,
1327 (frame::interpreter_frame_result_handler_offset) * wordSize));
1328 __ call(t);
1329
1330 // remove activation
1331 __ movptr(t, Address(rbp,
1332 frame::interpreter_frame_sender_sp_offset *
1333 wordSize)); // get sender sp
1334 __ leave(); // remove frame anchor
1335 __ pop(rdi); // get return address
1336 __ mov(rsp, t); // set sp to sender sp
1337 __ jmp(rdi);
1338
1339 if (inc_counter) {
1340 // Handle overflow of counter and compile method
1341 __ bind(invocation_counter_overflow);
1342 generate_counter_overflow(continue_after_compile);
1343 }
1344
1345 return entry_point;
1346 }
1347
1348 // Abstract method entry
1349 // Attempt to execute abstract method. Throw exception
1350 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1351
1352 address entry_point = __ pc();
1353
1354 // abstract method entry
1355
1356 // pop return address, reset last_sp to NULL
1357 __ empty_expression_stack();
1358 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1359 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1360
1361 // throw exception
1362 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
1363 // the call_VM checks for exception, so we should never return here.
1364 __ should_not_reach_here();
1365
1366 return entry_point;
1367 }
1368
1369 //
1370 // Generic interpreted method entry to (asm) interpreter
1371 //
1372 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1373 // determine code generation flags
1374 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1375
1376 // ebx: Method*
1377 // rbcp: sender sp
1378 address entry_point = __ pc();
1379
1380 const Address constMethod(rbx, Method::const_offset());
1381 const Address access_flags(rbx, Method::access_flags_offset());
1382 const Address size_of_parameters(rdx,
1383 ConstMethod::size_of_parameters_offset());
1384 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1385
1386
1387 // get parameter size (always needed)
1388 __ movptr(rdx, constMethod);
1389 __ load_unsigned_short(rcx, size_of_parameters);
1390
1391 // rbx: Method*
1392 // rcx: size of parameters
1393 // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1394
1395 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1396 __ subl(rdx, rcx); // rdx = no. of additional locals
1397
1398 // YYY
1399 // __ incrementl(rdx);
1400 // __ andl(rdx, -2);
1401
1402 // see if we've got enough room on the stack for locals plus overhead.
1403 generate_stack_overflow_check();
1404
1405 // get return address
1406 __ pop(rax);
1407
1408 // compute beginning of parameters
1409 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1410
1411 // rdx - # of additional locals
1412 // allocate space for locals
1413 // explicitly initialize locals
1414 {
1415 Label exit, loop;
1416 __ testl(rdx, rdx);
1417 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1418 __ bind(loop);
1419 __ push((int) NULL_WORD); // initialize local variables
1420 __ decrementl(rdx); // until everything initialized
1421 __ jcc(Assembler::greater, loop);
1422 __ bind(exit);
1423 }
1424
1425 // initialize fixed part of activation frame
1426 generate_fixed_frame(false);
1427
1428 // make sure method is not native & not abstract
1429 #ifdef ASSERT
1430 __ movl(rax, access_flags);
1431 {
1432 Label L;
1433 __ testl(rax, JVM_ACC_NATIVE);
1434 __ jcc(Assembler::zero, L);
1435 __ stop("tried to execute native method as non-native");
1436 __ bind(L);
1437 }
1438 {
1439 Label L;
1440 __ testl(rax, JVM_ACC_ABSTRACT);
1441 __ jcc(Assembler::zero, L);
1442 __ stop("tried to execute abstract method in interpreter");
1443 __ bind(L);
1444 }
1445 #endif
1446
1447 // Since at this point in the method invocation the exception
1448 // handler would try to exit the monitor of synchronized methods
1449 // which hasn't been entered yet, we set the thread local variable
1450 // _do_not_unlock_if_synchronized to true. The remove_activation
1451 // will check this flag.
1452
1453 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1454 NOT_LP64(__ get_thread(thread));
1455 const Address do_not_unlock_if_synchronized(thread,
1456 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1457 __ movbool(do_not_unlock_if_synchronized, true);
1458
1459 __ profile_parameters_type(rax, rcx, rdx);
1460 // increment invocation count & check for overflow
1461 Label invocation_counter_overflow;
1462 Label profile_method;
1463 Label profile_method_continue;
1464 if (inc_counter) {
1465 generate_counter_incr(&invocation_counter_overflow,
1466 &profile_method,
1467 &profile_method_continue);
1468 if (ProfileInterpreter) {
1469 __ bind(profile_method_continue);
1470 }
1471 }
1472
1473 Label continue_after_compile;
1474 __ bind(continue_after_compile);
1475
1476 // check for synchronized interpreted methods
1477 bang_stack_shadow_pages(false);
1478
1479 // reset the _do_not_unlock_if_synchronized flag
1480 NOT_LP64(__ get_thread(thread));
1481 __ movbool(do_not_unlock_if_synchronized, false);
1482
1483 // check for synchronized methods
1484 // Must happen AFTER invocation_counter check and stack overflow check,
1485 // so method is not locked if overflows.
1486 if (synchronized) {
1487 // Allocate monitor and lock method
1488 lock_method();
1489 } else {
1490 // no synchronization necessary
1491 #ifdef ASSERT
1492 {
1493 Label L;
1494 __ movl(rax, access_flags);
1495 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1496 __ jcc(Assembler::zero, L);
1497 __ stop("method needs synchronization");
1498 __ bind(L);
1499 }
1500 #endif
1501 }
1502
1503 // start execution
1504 #ifdef ASSERT
1505 {
1506 Label L;
1507 const Address monitor_block_top (rbp,
1508 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1509 __ movptr(rax, monitor_block_top);
1510 __ cmpptr(rax, rsp);
1511 __ jcc(Assembler::equal, L);
1512 __ stop("broken stack frame setup in interpreter");
1513 __ bind(L);
1514 }
1515 #endif
1516
1517 // jvmti support
1518 __ notify_method_entry();
1519
1520 __ dispatch_next(vtos);
1521
1522 // invocation counter overflow
1523 if (inc_counter) {
1524 if (ProfileInterpreter) {
1525 // We have decided to profile this method in the interpreter
1526 __ bind(profile_method);
1527 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1528 __ set_method_data_pointer_for_bcp();
1529 __ get_method(rbx);
1530 __ jmp(profile_method_continue);
1531 }
1532 // Handle overflow of counter and compile method
1533 __ bind(invocation_counter_overflow);
1534 generate_counter_overflow(continue_after_compile);
1535 }
1536
1537 return entry_point;
1538 }
1539
1540 //-----------------------------------------------------------------------------
1541 // Exceptions
1542
1543 void TemplateInterpreterGenerator::generate_throw_exception() {
1544 // Entry point in previous activation (i.e., if the caller was
1545 // interpreted)
1546 Interpreter::_rethrow_exception_entry = __ pc();
1547 // Restore sp to interpreter_frame_last_sp even though we are going
1548 // to empty the expression stack for the exception processing.
1549 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1550 // rax: exception
1551 // rdx: return address/pc that threw exception
1552 __ restore_bcp(); // r13/rsi points to call/send
1553 __ restore_locals();
1554 LP64_ONLY(__ reinit_heapbase()); // restore r12 as heapbase.
1555 // Entry point for exceptions thrown within interpreter code
1556 Interpreter::_throw_exception_entry = __ pc();
1557 // expression stack is undefined here
1558 // rax: exception
1559 // r13/rsi: exception bcp
1560 __ verify_oop(rax);
1561 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1562 LP64_ONLY(__ mov(c_rarg1, rax));
1563
1564 // expression stack must be empty before entering the VM in case of
1565 // an exception
1566 __ empty_expression_stack();
1567 // find exception handler address and preserve exception oop
1568 __ call_VM(rdx,
1569 CAST_FROM_FN_PTR(address,
1570 InterpreterRuntime::exception_handler_for_exception),
1571 rarg);
1572 // rax: exception handler entry point
1573 // rdx: preserved exception oop
1574 // r13/rsi: bcp for exception handler
1575 __ push_ptr(rdx); // push exception which is now the only value on the stack
1576 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1577
1578 // If the exception is not handled in the current frame the frame is
1579 // removed and the exception is rethrown (i.e. exception
1580 // continuation is _rethrow_exception).
1581 //
1582 // Note: At this point the bci is still the bxi for the instruction
1583 // which caused the exception and the expression stack is
1584 // empty. Thus, for any VM calls at this point, GC will find a legal
1585 // oop map (with empty expression stack).
1586
1587 // In current activation
1588 // tos: exception
1589 // esi: exception bcp
1590
1591 //
1592 // JVMTI PopFrame support
1593 //
1594
1595 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1596 __ empty_expression_stack();
1597 // Set the popframe_processing bit in pending_popframe_condition
1598 // indicating that we are currently handling popframe, so that
1599 // call_VMs that may happen later do not trigger new popframe
1600 // handling cycles.
1601 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1602 NOT_LP64(__ get_thread(thread));
1603 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1604 __ orl(rdx, JavaThread::popframe_processing_bit);
1605 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1606
1607 {
1608 // Check to see whether we are returning to a deoptimized frame.
1609 // (The PopFrame call ensures that the caller of the popped frame is
1610 // either interpreted or compiled and deoptimizes it if compiled.)
1611 // In this case, we can't call dispatch_next() after the frame is
1612 // popped, but instead must save the incoming arguments and restore
1613 // them after deoptimization has occurred.
1614 //
1615 // Note that we don't compare the return PC against the
1616 // deoptimization blob's unpack entry because of the presence of
1617 // adapter frames in C2.
1618 Label caller_not_deoptimized;
1619 Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1620 __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1621 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1622 InterpreterRuntime::interpreter_contains), rarg);
1623 __ testl(rax, rax);
1624 __ jcc(Assembler::notZero, caller_not_deoptimized);
1625
1626 // Compute size of arguments for saving when returning to
1627 // deoptimized caller
1628 __ get_method(rax);
1629 __ movptr(rax, Address(rax, Method::const_offset()));
1630 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1631 size_of_parameters_offset())));
1632 __ shll(rax, Interpreter::logStackElementSize);
1633 __ restore_locals();
1634 __ subptr(rlocals, rax);
1635 __ addptr(rlocals, wordSize);
1636 // Save these arguments
1637 NOT_LP64(__ get_thread(thread));
1638 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1639 Deoptimization::
1640 popframe_preserve_args),
1641 thread, rax, rlocals);
1642
1643 __ remove_activation(vtos, rdx,
1644 /* throw_monitor_exception */ false,
1645 /* install_monitor_exception */ false,
1646 /* notify_jvmdi */ false);
1647
1648 // Inform deoptimization that it is responsible for restoring
1649 // these arguments
1650 NOT_LP64(__ get_thread(thread));
1651 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1652 JavaThread::popframe_force_deopt_reexecution_bit);
1653
1654 // Continue in deoptimization handler
1655 __ jmp(rdx);
1656
1657 __ bind(caller_not_deoptimized);
1658 }
1659
1660 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1661 /* throw_monitor_exception */ false,
1662 /* install_monitor_exception */ false,
1663 /* notify_jvmdi */ false);
1664
1665 // Finish with popframe handling
1666 // A previous I2C followed by a deoptimization might have moved the
1667 // outgoing arguments further up the stack. PopFrame expects the
1668 // mutations to those outgoing arguments to be preserved and other
1669 // constraints basically require this frame to look exactly as
1670 // though it had previously invoked an interpreted activation with
1671 // no space between the top of the expression stack (current
1672 // last_sp) and the top of stack. Rather than force deopt to
1673 // maintain this kind of invariant all the time we call a small
1674 // fixup routine to move the mutated arguments onto the top of our
1675 // expression stack if necessary.
1676 #ifndef _LP64
1677 __ mov(rax, rsp);
1678 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1679 __ get_thread(thread);
1680 // PC must point into interpreter here
1681 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1682 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1683 __ get_thread(thread);
1684 #else
1685 __ mov(c_rarg1, rsp);
1686 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1687 // PC must point into interpreter here
1688 __ set_last_Java_frame(noreg, rbp, __ pc());
1689 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1690 #endif
1691 __ reset_last_Java_frame(thread, true);
1692
1693 // Restore the last_sp and null it out
1694 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1695 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1696
1697 __ restore_bcp();
1698 __ restore_locals();
1699 // The method data pointer was incremented already during
1700 // call profiling. We have to restore the mdp for the current bcp.
1701 if (ProfileInterpreter) {
1702 __ set_method_data_pointer_for_bcp();
1703 }
1704
1705 // Clear the popframe condition flag
1706 NOT_LP64(__ get_thread(thread));
1707 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1708 JavaThread::popframe_inactive);
1709
1710 #if INCLUDE_JVMTI
1711 {
1712 Label L_done;
1713 const Register local0 = rlocals;
1714
1715 __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1716 __ jcc(Assembler::notEqual, L_done);
1717
1718 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1719 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1720
1721 __ get_method(rdx);
1722 __ movptr(rax, Address(local0, 0));
1723 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1724
1725 __ testptr(rax, rax);
1726 __ jcc(Assembler::zero, L_done);
1727
1728 __ movptr(Address(rbx, 0), rax);
1729 __ bind(L_done);
1730 }
1731 #endif // INCLUDE_JVMTI
1732
1733 __ dispatch_next(vtos);
1734 // end of PopFrame support
1735
1736 Interpreter::_remove_activation_entry = __ pc();
1737
1738 // preserve exception over this code sequence
1739 __ pop_ptr(rax);
1740 NOT_LP64(__ get_thread(thread));
1741 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1742 // remove the activation (without doing throws on illegalMonitorExceptions)
1743 __ remove_activation(vtos, rdx, false, true, false);
1744 // restore exception
1745 NOT_LP64(__ get_thread(thread));
1746 __ get_vm_result(rax, thread);
1747
1748 // In between activations - previous activation type unknown yet
1749 // compute continuation point - the continuation point expects the
1750 // following registers set up:
1751 //
1752 // rax: exception
1753 // rdx: return address/pc that threw exception
1754 // rsp: expression stack of caller
1755 // rbp: ebp of caller
1756 __ push(rax); // save exception
1757 __ push(rdx); // save return address
1758 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1759 SharedRuntime::exception_handler_for_return_address),
1760 thread, rdx);
1761 __ mov(rbx, rax); // save exception handler
1762 __ pop(rdx); // restore return address
1763 __ pop(rax); // restore exception
1764 // Note that an "issuing PC" is actually the next PC after the call
1765 __ jmp(rbx); // jump to exception
1766 // handler of caller
1767 }
1768
1769
1770 //
1771 // JVMTI ForceEarlyReturn support
1772 //
1773 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1774 address entry = __ pc();
1775
1776 __ restore_bcp();
1777 __ restore_locals();
1778 __ empty_expression_stack();
1779 __ load_earlyret_value(state); // 32 bits returns value in rdx, so don't reuse
1780
1781 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1782 NOT_LP64(__ get_thread(thread));
1783 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1784 Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1785
1786 // Clear the earlyret state
1787 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1788
1789 __ remove_activation(state, rsi,
1790 false, /* throw_monitor_exception */
1791 false, /* install_monitor_exception */
1792 true); /* notify_jvmdi */
1793 __ jmp(rsi);
1794
1795 return entry;
1796 } // end of ForceEarlyReturn support
1797
1798
1799 //-----------------------------------------------------------------------------
1800 // Helper for vtos entry point generation
1801
1802 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1803 address& bep,
1804 address& cep,
1805 address& sep,
1806 address& aep,
1807 address& iep,
1808 address& lep,
1809 address& fep,
1810 address& dep,
1811 address& qep,
1812 address& vep) {
1813 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1814 Label L;
1815 aep = __ pc(); __ push_ptr(); __ jmp(L);
1816 qep = __ pc(); __ push_ptr(); __ jmp(L);
1817 #ifndef _LP64
1818 fep = __ pc(); __ push(ftos); __ jmp(L);
1819 dep = __ pc(); __ push(dtos); __ jmp(L);
1820 #else
1821 fep = __ pc(); __ push_f(xmm0); __ jmp(L);
1822 dep = __ pc(); __ push_d(xmm0); __ jmp(L);
1823 #endif // _LP64
1824 lep = __ pc(); __ push_l(); __ jmp(L);
1825 bep = cep = sep =
1826 iep = __ pc(); __ push_i();
1827 vep = __ pc();
1828 __ bind(L);
1829 generate_and_dispatch(t);
1830 }
1831
1832 //-----------------------------------------------------------------------------
1833
1834 // Non-product code
1835 #ifndef PRODUCT
1836
1837 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1838 address entry = __ pc();
1839
1840 #ifndef _LP64
1841 // prepare expression stack
1842 __ pop(rcx); // pop return address so expression stack is 'pure'
1843 __ push(state); // save tosca
1844
1845 // pass tosca registers as arguments & call tracer
1846 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1847 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1848 __ pop(state); // restore tosca
1849
1850 // return
1851 __ jmp(rcx);
1852 #else
1853 __ push(state);
1854 __ push(c_rarg0);
1855 __ push(c_rarg1);
1856 __ push(c_rarg2);
1857 __ push(c_rarg3);
1858 __ mov(c_rarg2, rax); // Pass itos
1859 #ifdef _WIN64
1860 __ movflt(xmm3, xmm0); // Pass ftos
1861 #endif
1862 __ call_VM(noreg,
1863 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1864 c_rarg1, c_rarg2, c_rarg3);
1865 __ pop(c_rarg3);
1866 __ pop(c_rarg2);
1867 __ pop(c_rarg1);
1868 __ pop(c_rarg0);
1869 __ pop(state);
1870 __ ret(0); // return from result handler
1871 #endif // _LP64
1872
1873 return entry;
1874 }
1875
1876 void TemplateInterpreterGenerator::count_bytecode() {
1877 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1878 }
1879
1880 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1881 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1882 }
1883
1884 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1885 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1886 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1887 __ orl(rbx,
1888 ((int) t->bytecode()) <<
1889 BytecodePairHistogram::log2_number_of_codes);
1890 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1891 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1892 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1893 }
1894
1895
1896 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1897 // Call a little run-time stub to avoid blow-up for each bytecode.
1898 // The run-time runtime saves the right registers, depending on
1899 // the tosca in-state for the given template.
1900
1901 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1902 "entry must have been generated");
1903 #ifndef _LP64
1904 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1905 #else
1906 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1907 __ andptr(rsp, -16); // align stack as required by ABI
1908 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1909 __ mov(rsp, r12); // restore sp
1910 __ reinit_heapbase();
1911 #endif // _LP64
1912 }
1913
1914
1915 void TemplateInterpreterGenerator::stop_interpreter_at() {
1916 Label L;
1917 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1918 StopInterpreterAt);
1919 __ jcc(Assembler::notEqual, L);
1920 __ int3();
1921 __ bind(L);
1922 }
1923 #endif // !PRODUCT