1 /*
2 * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "compiler/disassembler.hpp"
28 #include "gc/shared/barrierSetAssembler.hpp"
29 #include "interpreter/bytecodeHistogram.hpp"
30 #include "interpreter/interp_masm.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "interpreter/templateInterpreterGenerator.hpp"
34 #include "interpreter/templateTable.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/methodData.hpp"
37 #include "oops/method.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "oops/valueKlass.hpp"
40 #include "prims/jvmtiExport.hpp"
41 #include "prims/jvmtiThreadState.hpp"
42 #include "runtime/arguments.hpp"
43 #include "runtime/deoptimization.hpp"
44 #include "runtime/frame.inline.hpp"
45 #include "runtime/sharedRuntime.hpp"
46 #include "runtime/stubRoutines.hpp"
47 #include "runtime/synchronizer.hpp"
48 #include "runtime/timer.hpp"
49 #include "runtime/vframeArray.hpp"
50 #include "utilities/debug.hpp"
51 #include "utilities/macros.hpp"
52
53 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
54
55 // Size of interpreter code. Increase if too small. Interpreter will
56 // fail with a guarantee ("not enough space for interpreter generation");
57 // if too small.
58 // Run with +PrintInterpreter to get the VM to print out the size.
59 // Max size with JVMTI
60 #ifdef AMD64
61 int TemplateInterpreter::InterpreterCodeSize = JVMCI_ONLY(280) NOT_JVMCI(268) * 1024;
62 #else
63 int TemplateInterpreter::InterpreterCodeSize = 224 * 1024;
64 #endif // AMD64
65
66 // Global Register Names
67 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
68 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
69
70 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
71 const int bcp_offset = frame::interpreter_frame_bcp_offset * wordSize;
72 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
73
74
75 //-----------------------------------------------------------------------------
76
77 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
78 address entry = __ pc();
79
80 #ifdef ASSERT
81 {
82 Label L;
83 __ lea(rax, Address(rbp,
84 frame::interpreter_frame_monitor_block_top_offset *
85 wordSize));
86 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
87 // grows negative)
88 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
89 __ stop ("interpreter frame not set up");
90 __ bind(L);
91 }
92 #endif // ASSERT
93 // Restore bcp under the assumption that the current frame is still
94 // interpreted
95 __ restore_bcp();
96
97 // expression stack must be empty before entering the VM if an
98 // exception happened
99 __ empty_expression_stack();
100 // throw exception
101 __ call_VM(noreg,
102 CAST_FROM_FN_PTR(address,
103 InterpreterRuntime::throw_StackOverflowError));
104 return entry;
105 }
106
107 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
108 address entry = __ pc();
109 // The expression stack must be empty before entering the VM if an
110 // exception happened.
111 __ empty_expression_stack();
112
113 // Setup parameters.
114 // ??? convention: expect aberrant index in register ebx/rbx.
115 // Pass array to create more detailed exceptions.
116 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
117 __ call_VM(noreg,
118 CAST_FROM_FN_PTR(address,
119 InterpreterRuntime::
120 throw_ArrayIndexOutOfBoundsException),
121 rarg, rbx);
122 return entry;
123 }
124
125 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
126 address entry = __ pc();
127
128 // object is at TOS
129 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
130 __ pop(rarg);
131
132 // expression stack must be empty before entering the VM if an
133 // exception happened
134 __ empty_expression_stack();
135
136 __ call_VM(noreg,
137 CAST_FROM_FN_PTR(address,
138 InterpreterRuntime::
139 throw_ClassCastException),
140 rarg);
141 return entry;
142 }
143
144 address TemplateInterpreterGenerator::generate_exception_handler_common(
145 const char* name, const char* message, bool pass_oop) {
146 assert(!pass_oop || message == NULL, "either oop or message but not both");
147 address entry = __ pc();
148
149 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
150 Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
151
152 if (pass_oop) {
153 // object is at TOS
154 __ pop(rarg2);
155 }
156 // expression stack must be empty before entering the VM if an
157 // exception happened
158 __ empty_expression_stack();
159 // setup parameters
160 __ lea(rarg, ExternalAddress((address)name));
161 if (pass_oop) {
162 __ call_VM(rax, CAST_FROM_FN_PTR(address,
163 InterpreterRuntime::
164 create_klass_exception),
165 rarg, rarg2);
166 } else {
167 __ lea(rarg2, ExternalAddress((address)message));
168 __ call_VM(rax,
169 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
170 rarg, rarg2);
171 }
172 // throw exception
173 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
174 return entry;
175 }
176
177 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
178 address entry = __ pc();
179
180 #ifndef _LP64
181 #ifdef COMPILER2
182 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
183 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
184 for (int i = 1; i < 8; i++) {
185 __ ffree(i);
186 }
187 } else if (UseSSE < 2) {
188 __ empty_FPU_stack();
189 }
190 #endif // COMPILER2
191 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
192 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
193 } else {
194 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
195 }
196
197 if (state == ftos) {
198 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
199 } else if (state == dtos) {
200 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
201 }
202 #endif // _LP64
203
204 // Restore stack bottom in case i2c adjusted stack
205 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
206 // and NULL it as marker that esp is now tos until next java call
207 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
208
209 if (state == atos && ValueTypeReturnedAsFields) {
210 #ifndef _LP64
211 __ super_call_VM_leaf(StubRoutines::store_value_type_fields_to_buf());
212 #else
213 // A value type might be returned. If fields are in registers we
214 // need to allocate a value type instance and initialize it with
215 // the value of the fields.
216 Label skip, slow_case;
217 // We only need a new buffered value if a new one is not returned
218 __ testptr(rax, 1);
219 __ jcc(Assembler::zero, skip);
220
221 // Try to allocate a new buffered value (from the heap)
222 if (UseTLAB) {
223 __ mov(rbx, rax);
224 __ andptr(rbx, -2);
225
226 __ movl(r14, Address(rbx, Klass::layout_helper_offset()));
227
228 __ movptr(r13, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
229 __ lea(r14, Address(r13, r14, Address::times_1));
230 __ cmpptr(r14, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
231 __ jcc(Assembler::above, slow_case);
232 __ movptr(Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())), r14);
233 __ movptr(Address(r13, oopDesc::mark_offset_in_bytes()), (intptr_t)markOopDesc::always_locked_prototype());
234
235 __ xorl(rax, rax); // use zero reg to clear memory (shorter code)
236 __ store_klass_gap(r13, rax); // zero klass gap for compressed oops
237 __ mov(rax, rbx);
238 __ store_klass(r13, rbx); // klass
239
240 // We have our new buffered value, initialize its fields with a
241 // value class specific handler
242 __ movptr(rbx, Address(rax, InstanceKlass::adr_valueklass_fixed_block_offset()));
243 __ movptr(rbx, Address(rbx, ValueKlass::pack_handler_offset()));
244 __ mov(rax, r13);
245 __ call(rbx);
246 __ jmp(skip);
247 }
248
249 __ bind(slow_case);
250 // We failed to allocate a new value, fall back to a runtime
251 // call. Some oop field may be live in some registers but we can't
252 // tell. That runtime call will take care of preserving them
253 // across a GC if there's one.
254 __ super_call_VM_leaf(StubRoutines::store_value_type_fields_to_buf());
255 __ bind(skip);
256 #endif
257 }
258
259 __ restore_bcp();
260 __ restore_locals();
261
262 if (state == atos) {
263 Register mdp = rbx;
264 Register tmp = rcx;
265 __ profile_return_type(mdp, rax, tmp);
266 }
267
268 const Register cache = rbx;
269 const Register index = rcx;
270 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
271
272 const Register flags = cache;
273 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
274 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
275 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
276
277 const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
278 if (JvmtiExport::can_pop_frame()) {
279 NOT_LP64(__ get_thread(java_thread));
280 __ check_and_handle_popframe(java_thread);
281 }
282 if (JvmtiExport::can_force_early_return()) {
283 NOT_LP64(__ get_thread(java_thread));
284 __ check_and_handle_earlyret(java_thread);
285 }
286
287 __ dispatch_next(state, step);
288
289 return entry;
290 }
291
292
293 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
294 address entry = __ pc();
295
296 #ifndef _LP64
297 if (state == ftos) {
298 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
299 } else if (state == dtos) {
300 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
301 }
302 #endif // _LP64
303
304 // NULL last_sp until next java call
305 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
306 __ restore_bcp();
307 __ restore_locals();
308 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
309 NOT_LP64(__ get_thread(thread));
310 #if INCLUDE_JVMCI
311 // Check if we need to take lock at entry of synchronized method. This can
312 // only occur on method entry so emit it only for vtos with step 0.
313 if ((EnableJVMCI || UseAOT) && state == vtos && step == 0) {
314 Label L;
315 __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
316 __ jcc(Assembler::zero, L);
317 // Clear flag.
318 __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
319 // Satisfy calling convention for lock_method().
320 __ get_method(rbx);
321 // Take lock.
322 lock_method();
323 __ bind(L);
324 } else {
325 #ifdef ASSERT
326 if (EnableJVMCI) {
327 Label L;
328 __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
329 __ jcc(Assembler::zero, L);
330 __ stop("unexpected pending monitor in deopt entry");
331 __ bind(L);
332 }
333 #endif
334 }
335 #endif
336 // handle exceptions
337 {
338 Label L;
339 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
340 __ jcc(Assembler::zero, L);
341 __ call_VM(noreg,
342 CAST_FROM_FN_PTR(address,
343 InterpreterRuntime::throw_pending_exception));
344 __ should_not_reach_here();
345 __ bind(L);
346 }
347 if (continuation == NULL) {
348 __ dispatch_next(state, step);
349 } else {
350 __ jump_to_entry(continuation);
351 }
352 return entry;
353 }
354
355 address TemplateInterpreterGenerator::generate_result_handler_for(
356 BasicType type) {
357 address entry = __ pc();
358 switch (type) {
359 case T_BOOLEAN: __ c2bool(rax); break;
360 #ifndef _LP64
361 case T_CHAR : __ andptr(rax, 0xFFFF); break;
362 #else
363 case T_CHAR : __ movzwl(rax, rax); break;
364 #endif // _LP64
365 case T_BYTE : __ sign_extend_byte(rax); break;
366 case T_SHORT : __ sign_extend_short(rax); break;
367 case T_INT : /* nothing to do */ break;
368 case T_LONG : /* nothing to do */ break;
369 case T_VOID : /* nothing to do */ break;
370 #ifndef _LP64
371 case T_DOUBLE :
372 case T_FLOAT :
373 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
374 __ pop(t); // remove return address first
375 // Must return a result for interpreter or compiler. In SSE
376 // mode, results are returned in xmm0 and the FPU stack must
377 // be empty.
378 if (type == T_FLOAT && UseSSE >= 1) {
379 // Load ST0
380 __ fld_d(Address(rsp, 0));
381 // Store as float and empty fpu stack
382 __ fstp_s(Address(rsp, 0));
383 // and reload
384 __ movflt(xmm0, Address(rsp, 0));
385 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
386 __ movdbl(xmm0, Address(rsp, 0));
387 } else {
388 // restore ST0
389 __ fld_d(Address(rsp, 0));
390 }
391 // and pop the temp
392 __ addptr(rsp, 2 * wordSize);
393 __ push(t); // restore return address
394 }
395 break;
396 #else
397 case T_FLOAT : /* nothing to do */ break;
398 case T_DOUBLE : /* nothing to do */ break;
399 #endif // _LP64
400
401 case T_VALUETYPE: // fall through (value types are handled with oops)
402 case T_OBJECT :
403 // retrieve result from frame
404 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
405 // and verify it
406 __ verify_oop(rax);
407 break;
408 default : ShouldNotReachHere();
409 }
410 __ ret(0); // return from result handler
411 return entry;
412 }
413
414 address TemplateInterpreterGenerator::generate_safept_entry_for(
415 TosState state,
416 address runtime_entry) {
417 address entry = __ pc();
418 __ push(state);
419 __ call_VM(noreg, runtime_entry);
420 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
421 return entry;
422 }
423
424
425
426 // Helpers for commoning out cases in the various type of method entries.
427 //
428
429
430 // increment invocation count & check for overflow
431 //
432 // Note: checking for negative value instead of overflow
433 // so we have a 'sticky' overflow test
434 //
435 // rbx: method
436 // rcx: invocation counter
437 //
438 void TemplateInterpreterGenerator::generate_counter_incr(
439 Label* overflow,
440 Label* profile_method,
441 Label* profile_method_continue) {
442 Label done;
443 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
444 if (TieredCompilation) {
445 int increment = InvocationCounter::count_increment;
446 Label no_mdo;
447 if (ProfileInterpreter) {
448 // Are we profiling?
449 __ movptr(rax, Address(rbx, Method::method_data_offset()));
450 __ testptr(rax, rax);
451 __ jccb(Assembler::zero, no_mdo);
452 // Increment counter in the MDO
453 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
454 in_bytes(InvocationCounter::counter_offset()));
455 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
456 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
457 __ jmp(done);
458 }
459 __ bind(no_mdo);
460 // Increment counter in MethodCounters
461 const Address invocation_counter(rax,
462 MethodCounters::invocation_counter_offset() +
463 InvocationCounter::counter_offset());
464 __ get_method_counters(rbx, rax, done);
465 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
466 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
467 false, Assembler::zero, overflow);
468 __ bind(done);
469 } else { // not TieredCompilation
470 const Address backedge_counter(rax,
471 MethodCounters::backedge_counter_offset() +
472 InvocationCounter::counter_offset());
473 const Address invocation_counter(rax,
474 MethodCounters::invocation_counter_offset() +
475 InvocationCounter::counter_offset());
476
477 __ get_method_counters(rbx, rax, done);
478
479 if (ProfileInterpreter) {
480 __ incrementl(Address(rax,
481 MethodCounters::interpreter_invocation_counter_offset()));
482 }
483 // Update standard invocation counters
484 __ movl(rcx, invocation_counter);
485 __ incrementl(rcx, InvocationCounter::count_increment);
486 __ movl(invocation_counter, rcx); // save invocation count
487
488 __ movl(rax, backedge_counter); // load backedge counter
489 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
490
491 __ addl(rcx, rax); // add both counters
492
493 // profile_method is non-null only for interpreted method so
494 // profile_method != NULL == !native_call
495
496 if (ProfileInterpreter && profile_method != NULL) {
497 // Test to see if we should create a method data oop
498 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
499 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
500 __ jcc(Assembler::less, *profile_method_continue);
501
502 // if no method data exists, go to profile_method
503 __ test_method_data_pointer(rax, *profile_method);
504 }
505
506 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
507 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
508 __ jcc(Assembler::aboveEqual, *overflow);
509 __ bind(done);
510 }
511 }
512
513 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
514
515 // Asm interpreter on entry
516 // r14/rdi - locals
517 // r13/rsi - bcp
518 // rbx - method
519 // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
520 // rbp - interpreter frame
521
522 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
523 // Everything as it was on entry
524 // rdx is not restored. Doesn't appear to really be set.
525
526 // InterpreterRuntime::frequency_counter_overflow takes two
527 // arguments, the first (thread) is passed by call_VM, the second
528 // indicates if the counter overflow occurs at a backwards branch
529 // (NULL bcp). We pass zero for it. The call returns the address
530 // of the verified entry point for the method or NULL if the
531 // compilation did not complete (either went background or bailed
532 // out).
533 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
534 __ movl(rarg, 0);
535 __ call_VM(noreg,
536 CAST_FROM_FN_PTR(address,
537 InterpreterRuntime::frequency_counter_overflow),
538 rarg);
539
540 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
541 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
542 // and jump to the interpreted entry.
543 __ jmp(do_continue, relocInfo::none);
544 }
545
546 // See if we've got enough room on the stack for locals plus overhead below
547 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
548 // without going through the signal handler, i.e., reserved and yellow zones
549 // will not be made usable. The shadow zone must suffice to handle the
550 // overflow.
551 // The expression stack grows down incrementally, so the normal guard
552 // page mechanism will work for that.
553 //
554 // NOTE: Since the additional locals are also always pushed (wasn't
555 // obvious in generate_fixed_frame) so the guard should work for them
556 // too.
557 //
558 // Args:
559 // rdx: number of additional locals this frame needs (what we must check)
560 // rbx: Method*
561 //
562 // Kills:
563 // rax
564 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
565
566 // monitor entry size: see picture of stack in frame_x86.hpp
567 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
568
569 // total overhead size: entry_size + (saved rbp through expr stack
570 // bottom). be sure to change this if you add/subtract anything
571 // to/from the overhead area
572 const int overhead_size =
573 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
574
575 const int page_size = os::vm_page_size();
576
577 Label after_frame_check;
578
579 // see if the frame is greater than one page in size. If so,
580 // then we need to verify there is enough stack space remaining
581 // for the additional locals.
582 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
583 __ jcc(Assembler::belowEqual, after_frame_check);
584
585 // compute rsp as if this were going to be the last frame on
586 // the stack before the red zone
587
588 Label after_frame_check_pop;
589 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
590 #ifndef _LP64
591 __ push(thread);
592 __ get_thread(thread);
593 #endif
594
595 const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
596
597 // locals + overhead, in bytes
598 __ mov(rax, rdx);
599 __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
600 __ addptr(rax, overhead_size);
601
602 #ifdef ASSERT
603 Label limit_okay;
604 // Verify that thread stack overflow limit is non-zero.
605 __ cmpptr(stack_limit, (int32_t)NULL_WORD);
606 __ jcc(Assembler::notEqual, limit_okay);
607 __ stop("stack overflow limit is zero");
608 __ bind(limit_okay);
609 #endif
610
611 // Add locals/frame size to stack limit.
612 __ addptr(rax, stack_limit);
613
614 // Check against the current stack bottom.
615 __ cmpptr(rsp, rax);
616
617 __ jcc(Assembler::above, after_frame_check_pop);
618 NOT_LP64(__ pop(rsi)); // get saved bcp
619
620 // Restore sender's sp as SP. This is necessary if the sender's
621 // frame is an extended compiled frame (see gen_c2i_adapter())
622 // and safer anyway in case of JSR292 adaptations.
623
624 __ pop(rax); // return address must be moved if SP is changed
625 __ mov(rsp, rbcp);
626 __ push(rax);
627
628 // Note: the restored frame is not necessarily interpreted.
629 // Use the shared runtime version of the StackOverflowError.
630 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
631 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
632 // all done with frame size check
633 __ bind(after_frame_check_pop);
634 NOT_LP64(__ pop(rsi));
635
636 // all done with frame size check
637 __ bind(after_frame_check);
638 }
639
640 // Allocate monitor and lock method (asm interpreter)
641 //
642 // Args:
643 // rbx: Method*
644 // r14/rdi: locals
645 //
646 // Kills:
647 // rax
648 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
649 // rscratch1, rscratch2 (scratch regs)
650 void TemplateInterpreterGenerator::lock_method() {
651 // synchronize method
652 const Address access_flags(rbx, Method::access_flags_offset());
653 const Address monitor_block_top(
654 rbp,
655 frame::interpreter_frame_monitor_block_top_offset * wordSize);
656 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
657
658 #ifdef ASSERT
659 {
660 Label L;
661 __ movl(rax, access_flags);
662 __ testl(rax, JVM_ACC_SYNCHRONIZED);
663 __ jcc(Assembler::notZero, L);
664 __ stop("method doesn't need synchronization");
665 __ bind(L);
666 }
667 #endif // ASSERT
668
669 // get synchronization object
670 {
671 Label done;
672 __ movl(rax, access_flags);
673 __ testl(rax, JVM_ACC_STATIC);
674 // get receiver (assume this is frequent case)
675 __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
676 __ jcc(Assembler::zero, done);
677 __ load_mirror(rax, rbx);
678
679 #ifdef ASSERT
680 {
681 Label L;
682 __ testptr(rax, rax);
683 __ jcc(Assembler::notZero, L);
684 __ stop("synchronization object is NULL");
685 __ bind(L);
686 }
687 #endif // ASSERT
688
689 __ bind(done);
690 __ resolve(IS_NOT_NULL, rax);
691 }
692
693 // add space for monitor & lock
694 __ subptr(rsp, entry_size); // add space for a monitor entry
695 __ movptr(monitor_block_top, rsp); // set new monitor block top
696 // store object
697 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
698 const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
699 __ movptr(lockreg, rsp); // object address
700 __ lock_object(lockreg);
701 }
702
703 // Generate a fixed interpreter frame. This is identical setup for
704 // interpreted methods and for native methods hence the shared code.
705 //
706 // Args:
707 // rax: return address
708 // rbx: Method*
709 // r14/rdi: pointer to locals
710 // r13/rsi: sender sp
711 // rdx: cp cache
712 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
713 // initialize fixed part of activation frame
714 __ push(rax); // save return address
715 __ enter(); // save old & set new rbp
716 __ push(rbcp); // set sender sp
717 __ push((int)NULL_WORD); // leave last_sp as null
718 __ movptr(rbcp, Address(rbx, Method::const_offset())); // get ConstMethod*
719 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
720 __ push(rbx); // save Method*
721 // Get mirror and store it in the frame as GC root for this Method*
722 __ load_mirror(rdx, rbx);
723 __ push(rdx);
724 if (ProfileInterpreter) {
725 Label method_data_continue;
726 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
727 __ testptr(rdx, rdx);
728 __ jcc(Assembler::zero, method_data_continue);
729 __ addptr(rdx, in_bytes(MethodData::data_offset()));
730 __ bind(method_data_continue);
731 __ push(rdx); // set the mdp (method data pointer)
732 } else {
733 __ push(0);
734 }
735
736 __ movptr(rdx, Address(rbx, Method::const_offset()));
737 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
738 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
739 __ push(rdx); // set constant pool cache
740 __ push(rlocals); // set locals pointer
741 if (native_call) {
742 __ push(0); // no bcp
743 } else {
744 __ push(rbcp); // set bcp
745 }
746 __ push(0); // reserve word for pointer to expression stack bottom
747 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
748 }
749
750 // End of helpers
751
752 // Method entry for java.lang.ref.Reference.get.
753 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
754 // Code: _aload_0, _getfield, _areturn
755 // parameter size = 1
756 //
757 // The code that gets generated by this routine is split into 2 parts:
758 // 1. The "intrinsified" code performing an ON_WEAK_OOP_REF load,
759 // 2. The slow path - which is an expansion of the regular method entry.
760 //
761 // Notes:-
762 // * An intrinsic is always executed, where an ON_WEAK_OOP_REF load is performed.
763 // * We may jump to the slow path iff the receiver is null. If the
764 // Reference object is null then we no longer perform an ON_WEAK_OOP_REF load
765 // Thus we can use the regular method entry code to generate the NPE.
766 //
767 // rbx: Method*
768
769 // r13: senderSP must preserve for slow path, set SP to it on fast path
770
771 address entry = __ pc();
772
773 const int referent_offset = java_lang_ref_Reference::referent_offset;
774 guarantee(referent_offset > 0, "referent offset not initialized");
775
776 Label slow_path;
777 // rbx: method
778
779 // Check if local 0 != NULL
780 // If the receiver is null then it is OK to jump to the slow path.
781 __ movptr(rax, Address(rsp, wordSize));
782
783 __ testptr(rax, rax);
784 __ jcc(Assembler::zero, slow_path);
785
786 // rax: local 0
787 // rbx: method (but can be used as scratch now)
788 // rdx: scratch
789 // rdi: scratch
790
791 // Preserve the sender sp in case the load barrier
792 // calls the runtime
793 NOT_LP64(__ push(rsi));
794
795 // Load the value of the referent field.
796 const Address field_address(rax, referent_offset);
797 __ load_heap_oop(rax, field_address, /*tmp1*/ rbx, /*tmp_thread*/ rdx, ON_WEAK_OOP_REF);
798
799 // _areturn
800 const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
801 NOT_LP64(__ pop(rsi)); // get sender sp
802 __ pop(rdi); // get return address
803 __ mov(rsp, sender_sp); // set sp to sender sp
804 __ jmp(rdi);
805 __ ret(0);
806
807 // generate a vanilla interpreter entry as the slow path
808 __ bind(slow_path);
809 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
810 return entry;
811 }
812
813 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
814 // Quick & dirty stack overflow checking: bang the stack & handle trap.
815 // Note that we do the banging after the frame is setup, since the exception
816 // handling code expects to find a valid interpreter frame on the stack.
817 // Doing the banging earlier fails if the caller frame is not an interpreter
818 // frame.
819 // (Also, the exception throwing code expects to unlock any synchronized
820 // method receiever, so do the banging after locking the receiver.)
821
822 // Bang each page in the shadow zone. We can't assume it's been done for
823 // an interpreter frame with greater than a page of locals, so each page
824 // needs to be checked. Only true for non-native.
825 if (UseStackBanging) {
826 const int page_size = os::vm_page_size();
827 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
828 const int start_page = native_call ? n_shadow_pages : 1;
829 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
830 __ bang_stack_with_offset(pages*page_size);
831 }
832 }
833 }
834
835 // Interpreter stub for calling a native method. (asm interpreter)
836 // This sets up a somewhat different looking stack for calling the
837 // native method than the typical interpreter frame setup.
838 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
839 // determine code generation flags
840 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
841
842 // rbx: Method*
843 // rbcp: sender sp
844
845 address entry_point = __ pc();
846
847 const Address constMethod (rbx, Method::const_offset());
848 const Address access_flags (rbx, Method::access_flags_offset());
849 const Address size_of_parameters(rcx, ConstMethod::
850 size_of_parameters_offset());
851
852
853 // get parameter size (always needed)
854 __ movptr(rcx, constMethod);
855 __ load_unsigned_short(rcx, size_of_parameters);
856
857 // native calls don't need the stack size check since they have no
858 // expression stack and the arguments are already on the stack and
859 // we only add a handful of words to the stack
860
861 // rbx: Method*
862 // rcx: size of parameters
863 // rbcp: sender sp
864 __ pop(rax); // get return address
865
866 // for natives the size of locals is zero
867
868 // compute beginning of parameters
869 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
870
871 // add 2 zero-initialized slots for native calls
872 // initialize result_handler slot
873 __ push((int) NULL_WORD);
874 // slot for oop temp
875 // (static native method holder mirror/jni oop result)
876 __ push((int) NULL_WORD);
877
878 // initialize fixed part of activation frame
879 generate_fixed_frame(true);
880
881 // make sure method is native & not abstract
882 #ifdef ASSERT
883 __ movl(rax, access_flags);
884 {
885 Label L;
886 __ testl(rax, JVM_ACC_NATIVE);
887 __ jcc(Assembler::notZero, L);
888 __ stop("tried to execute non-native method as native");
889 __ bind(L);
890 }
891 {
892 Label L;
893 __ testl(rax, JVM_ACC_ABSTRACT);
894 __ jcc(Assembler::zero, L);
895 __ stop("tried to execute abstract method in interpreter");
896 __ bind(L);
897 }
898 #endif
899
900 // Since at this point in the method invocation the exception handler
901 // would try to exit the monitor of synchronized methods which hasn't
902 // been entered yet, we set the thread local variable
903 // _do_not_unlock_if_synchronized to true. The remove_activation will
904 // check this flag.
905
906 const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
907 NOT_LP64(__ get_thread(thread1));
908 const Address do_not_unlock_if_synchronized(thread1,
909 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
910 __ movbool(do_not_unlock_if_synchronized, true);
911
912 // increment invocation count & check for overflow
913 Label invocation_counter_overflow;
914 if (inc_counter) {
915 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
916 }
917
918 Label continue_after_compile;
919 __ bind(continue_after_compile);
920
921 bang_stack_shadow_pages(true);
922
923 // reset the _do_not_unlock_if_synchronized flag
924 NOT_LP64(__ get_thread(thread1));
925 __ movbool(do_not_unlock_if_synchronized, false);
926
927 // check for synchronized methods
928 // Must happen AFTER invocation_counter check and stack overflow check,
929 // so method is not locked if overflows.
930 if (synchronized) {
931 lock_method();
932 } else {
933 // no synchronization necessary
934 #ifdef ASSERT
935 {
936 Label L;
937 __ movl(rax, access_flags);
938 __ testl(rax, JVM_ACC_SYNCHRONIZED);
939 __ jcc(Assembler::zero, L);
940 __ stop("method needs synchronization");
941 __ bind(L);
942 }
943 #endif
944 }
945
946 // start execution
947 #ifdef ASSERT
948 {
949 Label L;
950 const Address monitor_block_top(rbp,
951 frame::interpreter_frame_monitor_block_top_offset * wordSize);
952 __ movptr(rax, monitor_block_top);
953 __ cmpptr(rax, rsp);
954 __ jcc(Assembler::equal, L);
955 __ stop("broken stack frame setup in interpreter");
956 __ bind(L);
957 }
958 #endif
959
960 // jvmti support
961 __ notify_method_entry();
962
963 // work registers
964 const Register method = rbx;
965 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
966 const Register t = NOT_LP64(rcx) LP64_ONLY(r11);
967
968 // allocate space for parameters
969 __ get_method(method);
970 __ movptr(t, Address(method, Method::const_offset()));
971 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
972
973 #ifndef _LP64
974 __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
975 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
976 __ subptr(rsp, t);
977 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
978 #else
979 __ shll(t, Interpreter::logStackElementSize);
980
981 __ subptr(rsp, t);
982 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
983 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
984 #endif // _LP64
985
986 // get signature handler
987 {
988 Label L;
989 __ movptr(t, Address(method, Method::signature_handler_offset()));
990 __ testptr(t, t);
991 __ jcc(Assembler::notZero, L);
992 __ call_VM(noreg,
993 CAST_FROM_FN_PTR(address,
994 InterpreterRuntime::prepare_native_call),
995 method);
996 __ get_method(method);
997 __ movptr(t, Address(method, Method::signature_handler_offset()));
998 __ bind(L);
999 }
1000
1001 // call signature handler
1002 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1003 "adjust this code");
1004 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
1005 "adjust this code");
1006 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
1007 "adjust this code");
1008
1009 // The generated handlers do not touch RBX (the method oop).
1010 // However, large signatures cannot be cached and are generated
1011 // each time here. The slow-path generator can do a GC on return,
1012 // so we must reload it after the call.
1013 __ call(t);
1014 __ get_method(method); // slow path can do a GC, reload RBX
1015
1016
1017 // result handler is in rax
1018 // set result handler
1019 __ movptr(Address(rbp,
1020 (frame::interpreter_frame_result_handler_offset) * wordSize),
1021 rax);
1022
1023 // pass mirror handle if static call
1024 {
1025 Label L;
1026 __ movl(t, Address(method, Method::access_flags_offset()));
1027 __ testl(t, JVM_ACC_STATIC);
1028 __ jcc(Assembler::zero, L);
1029 // get mirror
1030 __ load_mirror(t, method, rax);
1031 // copy mirror into activation frame
1032 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1033 t);
1034 // pass handle to mirror
1035 #ifndef _LP64
1036 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1037 __ movptr(Address(rsp, wordSize), t);
1038 #else
1039 __ lea(c_rarg1,
1040 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1041 #endif // _LP64
1042 __ bind(L);
1043 }
1044
1045 // get native function entry point
1046 {
1047 Label L;
1048 __ movptr(rax, Address(method, Method::native_function_offset()));
1049 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1050 __ cmpptr(rax, unsatisfied.addr());
1051 __ jcc(Assembler::notEqual, L);
1052 __ call_VM(noreg,
1053 CAST_FROM_FN_PTR(address,
1054 InterpreterRuntime::prepare_native_call),
1055 method);
1056 __ get_method(method);
1057 __ movptr(rax, Address(method, Method::native_function_offset()));
1058 __ bind(L);
1059 }
1060
1061 // pass JNIEnv
1062 #ifndef _LP64
1063 __ get_thread(thread);
1064 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1065 __ movptr(Address(rsp, 0), t);
1066
1067 // set_last_Java_frame_before_call
1068 // It is enough that the pc()
1069 // points into the right code segment. It does not have to be the correct return pc.
1070 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1071 #else
1072 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1073
1074 // It is enough that the pc() points into the right code
1075 // segment. It does not have to be the correct return pc.
1076 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1077 #endif // _LP64
1078
1079 // change thread state
1080 #ifdef ASSERT
1081 {
1082 Label L;
1083 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1084 __ cmpl(t, _thread_in_Java);
1085 __ jcc(Assembler::equal, L);
1086 __ stop("Wrong thread state in native stub");
1087 __ bind(L);
1088 }
1089 #endif
1090
1091 // Change state to native
1092
1093 __ movl(Address(thread, JavaThread::thread_state_offset()),
1094 _thread_in_native);
1095
1096 // Call the native method.
1097 __ call(rax);
1098 // 32: result potentially in rdx:rax or ST0
1099 // 64: result potentially in rax or xmm0
1100
1101 // Verify or restore cpu control state after JNI call
1102 __ restore_cpu_control_state_after_jni();
1103
1104 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1105 // in order to extract the result of a method call. If the order of these
1106 // pushes change or anything else is added to the stack then the code in
1107 // interpreter_frame_result must also change.
1108
1109 #ifndef _LP64
1110 // save potential result in ST(0) & rdx:rax
1111 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1112 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1113 // It is safe to do this push because state is _thread_in_native and return address will be found
1114 // via _last_native_pc and not via _last_jave_sp
1115
1116 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1117 // If the order changes or anything else is added to the stack the code in
1118 // interpreter_frame_result will have to be changed.
1119
1120 { Label L;
1121 Label push_double;
1122 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1123 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1124 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1125 float_handler.addr());
1126 __ jcc(Assembler::equal, push_double);
1127 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1128 double_handler.addr());
1129 __ jcc(Assembler::notEqual, L);
1130 __ bind(push_double);
1131 __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1132 __ bind(L);
1133 }
1134 #else
1135 __ push(dtos);
1136 #endif // _LP64
1137
1138 __ push(ltos);
1139
1140 // change thread state
1141 NOT_LP64(__ get_thread(thread));
1142 __ movl(Address(thread, JavaThread::thread_state_offset()),
1143 _thread_in_native_trans);
1144
1145 if (os::is_MP()) {
1146 if (UseMembar) {
1147 // Force this write out before the read below
1148 __ membar(Assembler::Membar_mask_bits(
1149 Assembler::LoadLoad | Assembler::LoadStore |
1150 Assembler::StoreLoad | Assembler::StoreStore));
1151 } else {
1152 // Write serialization page so VM thread can do a pseudo remote membar.
1153 // We use the current thread pointer to calculate a thread specific
1154 // offset to write to within the page. This minimizes bus traffic
1155 // due to cache line collision.
1156 __ serialize_memory(thread, rcx);
1157 }
1158 }
1159
1160 #ifndef _LP64
1161 if (AlwaysRestoreFPU) {
1162 // Make sure the control word is correct.
1163 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1164 }
1165 #endif // _LP64
1166
1167 // check for safepoint operation in progress and/or pending suspend requests
1168 {
1169 Label Continue;
1170 Label slow_path;
1171
1172 #ifndef _LP64
1173 __ safepoint_poll(slow_path, thread, noreg);
1174 #else
1175 __ safepoint_poll(slow_path, r15_thread, rscratch1);
1176 #endif
1177
1178 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1179 __ jcc(Assembler::equal, Continue);
1180 __ bind(slow_path);
1181
1182 // Don't use call_VM as it will see a possible pending exception
1183 // and forward it and never return here preventing us from
1184 // clearing _last_native_pc down below. Also can't use
1185 // call_VM_leaf either as it will check to see if r13 & r14 are
1186 // preserved and correspond to the bcp/locals pointers. So we do a
1187 // runtime call by hand.
1188 //
1189 #ifndef _LP64
1190 __ push(thread);
1191 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1192 JavaThread::check_special_condition_for_native_trans)));
1193 __ increment(rsp, wordSize);
1194 __ get_thread(thread);
1195 #else
1196 __ mov(c_rarg0, r15_thread);
1197 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1198 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1199 __ andptr(rsp, -16); // align stack as required by ABI
1200 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1201 __ mov(rsp, r12); // restore sp
1202 __ reinit_heapbase();
1203 #endif // _LP64
1204 __ bind(Continue);
1205 }
1206
1207 // change thread state
1208 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1209
1210 // reset_last_Java_frame
1211 __ reset_last_Java_frame(thread, true);
1212
1213 if (CheckJNICalls) {
1214 // clear_pending_jni_exception_check
1215 __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1216 }
1217
1218 // reset handle block
1219 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1220 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1221
1222 // If result is an oop unbox and store it in frame where gc will see it
1223 // and result handler will pick it up
1224
1225 {
1226 Label no_oop, not_weak, store_result;
1227 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1228 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1229 __ jcc(Assembler::notEqual, no_oop);
1230 // retrieve result
1231 __ pop(ltos);
1232 // Unbox oop result, e.g. JNIHandles::resolve value.
1233 __ resolve_jobject(rax /* value */,
1234 thread /* thread */,
1235 t /* tmp */);
1236 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1237 // keep stack depth as expected by pushing oop which will eventually be discarded
1238 __ push(ltos);
1239 __ bind(no_oop);
1240 }
1241
1242
1243 {
1244 Label no_reguard;
1245 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1246 JavaThread::stack_guard_yellow_reserved_disabled);
1247 __ jcc(Assembler::notEqual, no_reguard);
1248
1249 __ pusha(); // XXX only save smashed registers
1250 #ifndef _LP64
1251 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1252 __ popa();
1253 #else
1254 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1255 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1256 __ andptr(rsp, -16); // align stack as required by ABI
1257 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1258 __ mov(rsp, r12); // restore sp
1259 __ popa(); // XXX only restore smashed registers
1260 __ reinit_heapbase();
1261 #endif // _LP64
1262
1263 __ bind(no_reguard);
1264 }
1265
1266
1267 // The method register is junk from after the thread_in_native transition
1268 // until here. Also can't call_VM until the bcp has been
1269 // restored. Need bcp for throwing exception below so get it now.
1270 __ get_method(method);
1271
1272 // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1273 __ movptr(rbcp, Address(method, Method::const_offset())); // get ConstMethod*
1274 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
1275
1276 // handle exceptions (exception handling will handle unlocking!)
1277 {
1278 Label L;
1279 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1280 __ jcc(Assembler::zero, L);
1281 // Note: At some point we may want to unify this with the code
1282 // used in call_VM_base(); i.e., we should use the
1283 // StubRoutines::forward_exception code. For now this doesn't work
1284 // here because the rsp is not correctly set at this point.
1285 __ MacroAssembler::call_VM(noreg,
1286 CAST_FROM_FN_PTR(address,
1287 InterpreterRuntime::throw_pending_exception));
1288 __ should_not_reach_here();
1289 __ bind(L);
1290 }
1291
1292 // do unlocking if necessary
1293 {
1294 Label L;
1295 __ movl(t, Address(method, Method::access_flags_offset()));
1296 __ testl(t, JVM_ACC_SYNCHRONIZED);
1297 __ jcc(Assembler::zero, L);
1298 // the code below should be shared with interpreter macro
1299 // assembler implementation
1300 {
1301 Label unlock;
1302 // BasicObjectLock will be first in list, since this is a
1303 // synchronized method. However, need to check that the object
1304 // has not been unlocked by an explicit monitorexit bytecode.
1305 const Address monitor(rbp,
1306 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1307 wordSize - (int)sizeof(BasicObjectLock)));
1308
1309 const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1310
1311 // monitor expect in c_rarg1 for slow unlock path
1312 __ lea(regmon, monitor); // address of first monitor
1313
1314 __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1315 __ testptr(t, t);
1316 __ jcc(Assembler::notZero, unlock);
1317
1318 // Entry already unlocked, need to throw exception
1319 __ MacroAssembler::call_VM(noreg,
1320 CAST_FROM_FN_PTR(address,
1321 InterpreterRuntime::throw_illegal_monitor_state_exception));
1322 __ should_not_reach_here();
1323
1324 __ bind(unlock);
1325 __ unlock_object(regmon);
1326 }
1327 __ bind(L);
1328 }
1329
1330 // jvmti support
1331 // Note: This must happen _after_ handling/throwing any exceptions since
1332 // the exception handler code notifies the runtime of method exits
1333 // too. If this happens before, method entry/exit notifications are
1334 // not properly paired (was bug - gri 11/22/99).
1335 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1336
1337 // restore potential result in edx:eax, call result handler to
1338 // restore potential result in ST0 & handle result
1339
1340 __ pop(ltos);
1341 LP64_ONLY( __ pop(dtos));
1342
1343 __ movptr(t, Address(rbp,
1344 (frame::interpreter_frame_result_handler_offset) * wordSize));
1345 __ call(t);
1346
1347 // remove activation
1348 __ movptr(t, Address(rbp,
1349 frame::interpreter_frame_sender_sp_offset *
1350 wordSize)); // get sender sp
1351 __ leave(); // remove frame anchor
1352 __ pop(rdi); // get return address
1353 __ mov(rsp, t); // set sp to sender sp
1354 __ jmp(rdi);
1355
1356 if (inc_counter) {
1357 // Handle overflow of counter and compile method
1358 __ bind(invocation_counter_overflow);
1359 generate_counter_overflow(continue_after_compile);
1360 }
1361
1362 return entry_point;
1363 }
1364
1365 // Abstract method entry
1366 // Attempt to execute abstract method. Throw exception
1367 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1368
1369 address entry_point = __ pc();
1370
1371 // abstract method entry
1372
1373 // pop return address, reset last_sp to NULL
1374 __ empty_expression_stack();
1375 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1376 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1377
1378 // throw exception
1379 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod), rbx);
1380 // the call_VM checks for exception, so we should never return here.
1381 __ should_not_reach_here();
1382
1383 return entry_point;
1384 }
1385
1386 //
1387 // Generic interpreted method entry to (asm) interpreter
1388 //
1389 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1390 // determine code generation flags
1391 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1392
1393 // ebx: Method*
1394 // rbcp: sender sp
1395 address entry_point = __ pc();
1396
1397 const Address constMethod(rbx, Method::const_offset());
1398 const Address access_flags(rbx, Method::access_flags_offset());
1399 const Address size_of_parameters(rdx,
1400 ConstMethod::size_of_parameters_offset());
1401 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1402
1403
1404 // get parameter size (always needed)
1405 __ movptr(rdx, constMethod);
1406 __ load_unsigned_short(rcx, size_of_parameters);
1407
1408 // rbx: Method*
1409 // rcx: size of parameters
1410 // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1411
1412 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1413 __ subl(rdx, rcx); // rdx = no. of additional locals
1414
1415 // YYY
1416 // __ incrementl(rdx);
1417 // __ andl(rdx, -2);
1418
1419 // see if we've got enough room on the stack for locals plus overhead.
1420 generate_stack_overflow_check();
1421
1422 // get return address
1423 __ pop(rax);
1424
1425 // compute beginning of parameters
1426 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1427
1428 // rdx - # of additional locals
1429 // allocate space for locals
1430 // explicitly initialize locals
1431 {
1432 Label exit, loop;
1433 __ testl(rdx, rdx);
1434 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1435 __ bind(loop);
1436 __ push((int) NULL_WORD); // initialize local variables
1437 __ decrementl(rdx); // until everything initialized
1438 __ jcc(Assembler::greater, loop);
1439 __ bind(exit);
1440 }
1441
1442 // initialize fixed part of activation frame
1443 generate_fixed_frame(false);
1444
1445 // make sure method is not native & not abstract
1446 #ifdef ASSERT
1447 __ movl(rax, access_flags);
1448 {
1449 Label L;
1450 __ testl(rax, JVM_ACC_NATIVE);
1451 __ jcc(Assembler::zero, L);
1452 __ stop("tried to execute native method as non-native");
1453 __ bind(L);
1454 }
1455 {
1456 Label L;
1457 __ testl(rax, JVM_ACC_ABSTRACT);
1458 __ jcc(Assembler::zero, L);
1459 __ stop("tried to execute abstract method in interpreter");
1460 __ bind(L);
1461 }
1462 #endif
1463
1464 // Since at this point in the method invocation the exception
1465 // handler would try to exit the monitor of synchronized methods
1466 // which hasn't been entered yet, we set the thread local variable
1467 // _do_not_unlock_if_synchronized to true. The remove_activation
1468 // will check this flag.
1469
1470 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1471 NOT_LP64(__ get_thread(thread));
1472 const Address do_not_unlock_if_synchronized(thread,
1473 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1474 __ movbool(do_not_unlock_if_synchronized, true);
1475
1476 __ profile_parameters_type(rax, rcx, rdx);
1477 // increment invocation count & check for overflow
1478 Label invocation_counter_overflow;
1479 Label profile_method;
1480 Label profile_method_continue;
1481 if (inc_counter) {
1482 generate_counter_incr(&invocation_counter_overflow,
1483 &profile_method,
1484 &profile_method_continue);
1485 if (ProfileInterpreter) {
1486 __ bind(profile_method_continue);
1487 }
1488 }
1489
1490 Label continue_after_compile;
1491 __ bind(continue_after_compile);
1492
1493 // check for synchronized interpreted methods
1494 bang_stack_shadow_pages(false);
1495
1496 // reset the _do_not_unlock_if_synchronized flag
1497 NOT_LP64(__ get_thread(thread));
1498 __ movbool(do_not_unlock_if_synchronized, false);
1499
1500 // check for synchronized methods
1501 // Must happen AFTER invocation_counter check and stack overflow check,
1502 // so method is not locked if overflows.
1503 if (synchronized) {
1504 // Allocate monitor and lock method
1505 lock_method();
1506 } else {
1507 // no synchronization necessary
1508 #ifdef ASSERT
1509 {
1510 Label L;
1511 __ movl(rax, access_flags);
1512 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1513 __ jcc(Assembler::zero, L);
1514 __ stop("method needs synchronization");
1515 __ bind(L);
1516 }
1517 #endif
1518 }
1519
1520 // start execution
1521 #ifdef ASSERT
1522 {
1523 Label L;
1524 const Address monitor_block_top (rbp,
1525 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1526 __ movptr(rax, monitor_block_top);
1527 __ cmpptr(rax, rsp);
1528 __ jcc(Assembler::equal, L);
1529 __ stop("broken stack frame setup in interpreter");
1530 __ bind(L);
1531 }
1532 #endif
1533
1534 // jvmti support
1535 __ notify_method_entry();
1536
1537 __ dispatch_next(vtos);
1538
1539 // invocation counter overflow
1540 if (inc_counter) {
1541 if (ProfileInterpreter) {
1542 // We have decided to profile this method in the interpreter
1543 __ bind(profile_method);
1544 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1545 __ set_method_data_pointer_for_bcp();
1546 __ get_method(rbx);
1547 __ jmp(profile_method_continue);
1548 }
1549 // Handle overflow of counter and compile method
1550 __ bind(invocation_counter_overflow);
1551 generate_counter_overflow(continue_after_compile);
1552 }
1553
1554 return entry_point;
1555 }
1556
1557 //-----------------------------------------------------------------------------
1558 // Exceptions
1559
1560 void TemplateInterpreterGenerator::generate_throw_exception() {
1561 // Entry point in previous activation (i.e., if the caller was
1562 // interpreted)
1563 Interpreter::_rethrow_exception_entry = __ pc();
1564 // Restore sp to interpreter_frame_last_sp even though we are going
1565 // to empty the expression stack for the exception processing.
1566 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1567 // rax: exception
1568 // rdx: return address/pc that threw exception
1569 __ restore_bcp(); // r13/rsi points to call/send
1570 __ restore_locals();
1571 LP64_ONLY(__ reinit_heapbase()); // restore r12 as heapbase.
1572 // Entry point for exceptions thrown within interpreter code
1573 Interpreter::_throw_exception_entry = __ pc();
1574 // expression stack is undefined here
1575 // rax: exception
1576 // r13/rsi: exception bcp
1577 __ verify_oop(rax);
1578 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1579 LP64_ONLY(__ mov(c_rarg1, rax));
1580
1581 // expression stack must be empty before entering the VM in case of
1582 // an exception
1583 __ empty_expression_stack();
1584 // find exception handler address and preserve exception oop
1585 __ call_VM(rdx,
1586 CAST_FROM_FN_PTR(address,
1587 InterpreterRuntime::exception_handler_for_exception),
1588 rarg);
1589 // rax: exception handler entry point
1590 // rdx: preserved exception oop
1591 // r13/rsi: bcp for exception handler
1592 __ push_ptr(rdx); // push exception which is now the only value on the stack
1593 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1594
1595 // If the exception is not handled in the current frame the frame is
1596 // removed and the exception is rethrown (i.e. exception
1597 // continuation is _rethrow_exception).
1598 //
1599 // Note: At this point the bci is still the bxi for the instruction
1600 // which caused the exception and the expression stack is
1601 // empty. Thus, for any VM calls at this point, GC will find a legal
1602 // oop map (with empty expression stack).
1603
1604 // In current activation
1605 // tos: exception
1606 // esi: exception bcp
1607
1608 //
1609 // JVMTI PopFrame support
1610 //
1611
1612 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1613 __ empty_expression_stack();
1614 // Set the popframe_processing bit in pending_popframe_condition
1615 // indicating that we are currently handling popframe, so that
1616 // call_VMs that may happen later do not trigger new popframe
1617 // handling cycles.
1618 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1619 NOT_LP64(__ get_thread(thread));
1620 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1621 __ orl(rdx, JavaThread::popframe_processing_bit);
1622 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1623
1624 {
1625 // Check to see whether we are returning to a deoptimized frame.
1626 // (The PopFrame call ensures that the caller of the popped frame is
1627 // either interpreted or compiled and deoptimizes it if compiled.)
1628 // In this case, we can't call dispatch_next() after the frame is
1629 // popped, but instead must save the incoming arguments and restore
1630 // them after deoptimization has occurred.
1631 //
1632 // Note that we don't compare the return PC against the
1633 // deoptimization blob's unpack entry because of the presence of
1634 // adapter frames in C2.
1635 Label caller_not_deoptimized;
1636 Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1637 __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1638 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1639 InterpreterRuntime::interpreter_contains), rarg);
1640 __ testl(rax, rax);
1641 __ jcc(Assembler::notZero, caller_not_deoptimized);
1642
1643 // Compute size of arguments for saving when returning to
1644 // deoptimized caller
1645 __ get_method(rax);
1646 __ movptr(rax, Address(rax, Method::const_offset()));
1647 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1648 size_of_parameters_offset())));
1649 __ shll(rax, Interpreter::logStackElementSize);
1650 __ restore_locals();
1651 __ subptr(rlocals, rax);
1652 __ addptr(rlocals, wordSize);
1653 // Save these arguments
1654 NOT_LP64(__ get_thread(thread));
1655 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1656 Deoptimization::
1657 popframe_preserve_args),
1658 thread, rax, rlocals);
1659
1660 __ remove_activation(vtos, rdx,
1661 /* throw_monitor_exception */ false,
1662 /* install_monitor_exception */ false,
1663 /* notify_jvmdi */ false);
1664
1665 // Inform deoptimization that it is responsible for restoring
1666 // these arguments
1667 NOT_LP64(__ get_thread(thread));
1668 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1669 JavaThread::popframe_force_deopt_reexecution_bit);
1670
1671 // Continue in deoptimization handler
1672 __ jmp(rdx);
1673
1674 __ bind(caller_not_deoptimized);
1675 }
1676
1677 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1678 /* throw_monitor_exception */ false,
1679 /* install_monitor_exception */ false,
1680 /* notify_jvmdi */ false);
1681
1682 // Finish with popframe handling
1683 // A previous I2C followed by a deoptimization might have moved the
1684 // outgoing arguments further up the stack. PopFrame expects the
1685 // mutations to those outgoing arguments to be preserved and other
1686 // constraints basically require this frame to look exactly as
1687 // though it had previously invoked an interpreted activation with
1688 // no space between the top of the expression stack (current
1689 // last_sp) and the top of stack. Rather than force deopt to
1690 // maintain this kind of invariant all the time we call a small
1691 // fixup routine to move the mutated arguments onto the top of our
1692 // expression stack if necessary.
1693 #ifndef _LP64
1694 __ mov(rax, rsp);
1695 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1696 __ get_thread(thread);
1697 // PC must point into interpreter here
1698 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1699 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1700 __ get_thread(thread);
1701 #else
1702 __ mov(c_rarg1, rsp);
1703 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1704 // PC must point into interpreter here
1705 __ set_last_Java_frame(noreg, rbp, __ pc());
1706 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1707 #endif
1708 __ reset_last_Java_frame(thread, true);
1709
1710 // Restore the last_sp and null it out
1711 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1712 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1713
1714 __ restore_bcp();
1715 __ restore_locals();
1716 // The method data pointer was incremented already during
1717 // call profiling. We have to restore the mdp for the current bcp.
1718 if (ProfileInterpreter) {
1719 __ set_method_data_pointer_for_bcp();
1720 }
1721
1722 // Clear the popframe condition flag
1723 NOT_LP64(__ get_thread(thread));
1724 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1725 JavaThread::popframe_inactive);
1726
1727 #if INCLUDE_JVMTI
1728 {
1729 Label L_done;
1730 const Register local0 = rlocals;
1731
1732 __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1733 __ jcc(Assembler::notEqual, L_done);
1734
1735 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1736 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1737
1738 __ get_method(rdx);
1739 __ movptr(rax, Address(local0, 0));
1740 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1741
1742 __ testptr(rax, rax);
1743 __ jcc(Assembler::zero, L_done);
1744
1745 __ movptr(Address(rbx, 0), rax);
1746 __ bind(L_done);
1747 }
1748 #endif // INCLUDE_JVMTI
1749
1750 __ dispatch_next(vtos);
1751 // end of PopFrame support
1752
1753 Interpreter::_remove_activation_entry = __ pc();
1754
1755 // preserve exception over this code sequence
1756 __ pop_ptr(rax);
1757 NOT_LP64(__ get_thread(thread));
1758 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1759 // remove the activation (without doing throws on illegalMonitorExceptions)
1760 __ remove_activation(vtos, rdx, false, true, false);
1761 // restore exception
1762 NOT_LP64(__ get_thread(thread));
1763 __ get_vm_result(rax, thread);
1764
1765 // In between activations - previous activation type unknown yet
1766 // compute continuation point - the continuation point expects the
1767 // following registers set up:
1768 //
1769 // rax: exception
1770 // rdx: return address/pc that threw exception
1771 // rsp: expression stack of caller
1772 // rbp: ebp of caller
1773 __ push(rax); // save exception
1774 __ push(rdx); // save return address
1775 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1776 SharedRuntime::exception_handler_for_return_address),
1777 thread, rdx);
1778 __ mov(rbx, rax); // save exception handler
1779 __ pop(rdx); // restore return address
1780 __ pop(rax); // restore exception
1781 // Note that an "issuing PC" is actually the next PC after the call
1782 __ jmp(rbx); // jump to exception
1783 // handler of caller
1784 }
1785
1786
1787 //
1788 // JVMTI ForceEarlyReturn support
1789 //
1790 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1791 address entry = __ pc();
1792
1793 __ restore_bcp();
1794 __ restore_locals();
1795 __ empty_expression_stack();
1796 __ load_earlyret_value(state); // 32 bits returns value in rdx, so don't reuse
1797
1798 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1799 NOT_LP64(__ get_thread(thread));
1800 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1801 Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1802
1803 // Clear the earlyret state
1804 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1805
1806 __ remove_activation(state, rsi,
1807 false, /* throw_monitor_exception */
1808 false, /* install_monitor_exception */
1809 true); /* notify_jvmdi */
1810 __ jmp(rsi);
1811
1812 return entry;
1813 } // end of ForceEarlyReturn support
1814
1815
1816 //-----------------------------------------------------------------------------
1817 // Helper for vtos entry point generation
1818
1819 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1820 address& bep,
1821 address& cep,
1822 address& sep,
1823 address& aep,
1824 address& iep,
1825 address& lep,
1826 address& fep,
1827 address& dep,
1828 address& vep) {
1829 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1830 Label L;
1831 aep = __ pc(); // atos entry point
1832 __ push_ptr();
1833 __ jmp(L);
1834 #ifndef _LP64
1835 fep = __ pc(); // ftos entry point
1836 __ push(ftos);
1837 __ jmp(L);
1838 dep = __ pc(); // dtos entry point
1839 __ push(dtos);
1840 __ jmp(L);
1841 #else
1842 fep = __ pc(); // ftos entry point
1843 __ push_f(xmm0);
1844 __ jmp(L);
1845 dep = __ pc(); // dtos entry point
1846 __ push_d(xmm0);
1847 __ jmp(L);
1848 #endif // _LP64
1849 lep = __ pc(); // ltos entry point
1850 __ push_l();
1851 __ jmp(L);
1852 bep = cep = sep = iep = __ pc(); // [bcsi]tos entry point
1853 __ push_i();
1854 vep = __ pc(); // vtos entry point
1855 __ bind(L);
1856 generate_and_dispatch(t);
1857 }
1858
1859 //-----------------------------------------------------------------------------
1860
1861 // Non-product code
1862 #ifndef PRODUCT
1863
1864 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1865 address entry = __ pc();
1866
1867 #ifndef _LP64
1868 // prepare expression stack
1869 __ pop(rcx); // pop return address so expression stack is 'pure'
1870 __ push(state); // save tosca
1871
1872 // pass tosca registers as arguments & call tracer
1873 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1874 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1875 __ pop(state); // restore tosca
1876
1877 // return
1878 __ jmp(rcx);
1879 #else
1880 __ push(state);
1881 __ push(c_rarg0);
1882 __ push(c_rarg1);
1883 __ push(c_rarg2);
1884 __ push(c_rarg3);
1885 __ mov(c_rarg2, rax); // Pass itos
1886 #ifdef _WIN64
1887 __ movflt(xmm3, xmm0); // Pass ftos
1888 #endif
1889 __ call_VM(noreg,
1890 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1891 c_rarg1, c_rarg2, c_rarg3);
1892 __ pop(c_rarg3);
1893 __ pop(c_rarg2);
1894 __ pop(c_rarg1);
1895 __ pop(c_rarg0);
1896 __ pop(state);
1897 __ ret(0); // return from result handler
1898 #endif // _LP64
1899
1900 return entry;
1901 }
1902
1903 void TemplateInterpreterGenerator::count_bytecode() {
1904 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1905 }
1906
1907 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1908 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1909 }
1910
1911 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1912 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1913 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1914 __ orl(rbx,
1915 ((int) t->bytecode()) <<
1916 BytecodePairHistogram::log2_number_of_codes);
1917 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1918 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1919 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1920 }
1921
1922
1923 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1924 // Call a little run-time stub to avoid blow-up for each bytecode.
1925 // The run-time runtime saves the right registers, depending on
1926 // the tosca in-state for the given template.
1927
1928 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1929 "entry must have been generated");
1930 #ifndef _LP64
1931 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1932 #else
1933 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1934 __ andptr(rsp, -16); // align stack as required by ABI
1935 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1936 __ mov(rsp, r12); // restore sp
1937 __ reinit_heapbase();
1938 #endif // _LP64
1939 }
1940
1941
1942 void TemplateInterpreterGenerator::stop_interpreter_at() {
1943 Label L;
1944 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1945 StopInterpreterAt);
1946 __ jcc(Assembler::notEqual, L);
1947 __ int3();
1948 __ bind(L);
1949 }
1950 #endif // !PRODUCT