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 "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 "oops/valueKlass.hpp"
38 #include "prims/jvmtiExport.hpp"
39 #include "prims/jvmtiThreadState.hpp"
40 #include "runtime/arguments.hpp"
41 #include "runtime/deoptimization.hpp"
42 #include "runtime/frame.inline.hpp"
43 #include "runtime/sharedRuntime.hpp"
44 #include "runtime/stubRoutines.hpp"
45 #include "runtime/synchronizer.hpp"
46 #include "runtime/timer.hpp"
47 #include "runtime/vframeArray.hpp"
48 #include "utilities/debug.hpp"
49 #include "utilities/macros.hpp"
50
51 #define __ _masm->
52
53 // Size of interpreter code. Increase if too small. Interpreter will
54 // fail with a guarantee ("not enough space for interpreter generation");
55 // if too small.
56 // Run with +PrintInterpreter to get the VM to print out the size.
57 // Max size with JVMTI
58 #ifdef AMD64
59 int TemplateInterpreter::InterpreterCodeSize = JVMCI_ONLY(280) NOT_JVMCI(268) * 1024;
60 #else
61 int TemplateInterpreter::InterpreterCodeSize = 224 * 1024;
62 #endif // AMD64
63
64 // Global Register Names
65 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
66 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
67
68 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
69 const int bcp_offset = frame::interpreter_frame_bcp_offset * wordSize;
70 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
71
72
73 //-----------------------------------------------------------------------------
74
75 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
76 address entry = __ pc();
77
78 #ifdef ASSERT
79 {
80 Label L;
81 __ lea(rax, Address(rbp,
82 frame::interpreter_frame_monitor_block_top_offset *
83 wordSize));
84 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
85 // grows negative)
86 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
87 __ stop ("interpreter frame not set up");
88 __ bind(L);
89 }
90 #endif // ASSERT
91 // Restore bcp under the assumption that the current frame is still
92 // interpreted
93 __ restore_bcp();
94
95 // expression stack must be empty before entering the VM if an
96 // exception happened
97 __ empty_expression_stack();
98 // throw exception
99 __ call_VM(noreg,
100 CAST_FROM_FN_PTR(address,
101 InterpreterRuntime::throw_StackOverflowError));
102 return entry;
103 }
104
105 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
106 const char* name) {
107 address entry = __ pc();
108 // expression stack must be empty before entering the VM if an
109 // exception happened
110 __ empty_expression_stack();
111 // setup parameters
112 // ??? convention: expect aberrant index in register ebx
113 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
114 __ lea(rarg, ExternalAddress((address)name));
115 __ call_VM(noreg,
116 CAST_FROM_FN_PTR(address,
117 InterpreterRuntime::
118 throw_ArrayIndexOutOfBoundsException),
119 rarg, rbx);
120 return entry;
121 }
122
123 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
124 address entry = __ pc();
125
126 // object is at TOS
127 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
128 __ pop(rarg);
129
130 // expression stack must be empty before entering the VM if an
131 // exception happened
132 __ empty_expression_stack();
133
134 __ call_VM(noreg,
135 CAST_FROM_FN_PTR(address,
136 InterpreterRuntime::
137 throw_ClassCastException),
138 rarg);
139 return entry;
140 }
141
142 address TemplateInterpreterGenerator::generate_exception_handler_common(
143 const char* name, const char* message, bool pass_oop) {
144 assert(!pass_oop || message == NULL, "either oop or message but not both");
145 address entry = __ pc();
146
147 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
148 Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
149
150 if (pass_oop) {
151 // object is at TOS
152 __ pop(rarg2);
153 }
154 // expression stack must be empty before entering the VM if an
155 // exception happened
156 __ empty_expression_stack();
157 // setup parameters
158 __ lea(rarg, ExternalAddress((address)name));
159 if (pass_oop) {
160 __ call_VM(rax, CAST_FROM_FN_PTR(address,
161 InterpreterRuntime::
162 create_klass_exception),
163 rarg, rarg2);
164 } else {
165 __ lea(rarg2, ExternalAddress((address)message));
166 __ call_VM(rax,
167 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
168 rarg, rarg2);
169 }
170 // throw exception
171 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
172 return entry;
173 }
174
175 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
176 address entry = __ pc();
177
178 #ifndef _LP64
179 #ifdef COMPILER2
180 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
181 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
182 for (int i = 1; i < 8; i++) {
183 __ ffree(i);
184 }
185 } else if (UseSSE < 2) {
186 __ empty_FPU_stack();
187 }
188 #endif // COMPILER2
189 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
190 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
191 } else {
192 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
193 }
194
195 if (state == ftos) {
196 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
197 } else if (state == dtos) {
198 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
199 }
200 #endif // _LP64
201
202 // Restore stack bottom in case i2c adjusted stack
203 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
204 // and NULL it as marker that esp is now tos until next java call
205 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
206
207 if (/*state == qtos*/ false && ValueTypeReturnedAsFields) {
208 #ifndef _LP64
209 __ super_call_VM_leaf(StubRoutines::store_value_type_fields_to_buf());
210 #else
211 // A value type is being returned. If fields are in registers we
212 // need to allocate a value type instance and initialize it with
213 // the value of the fields.
214 Label skip, slow_case;
215 // We only need a new buffered value if a new one is not returned
216 __ testptr(rax, 1);
217 __ jcc(Assembler::zero, skip);
218
219 // Try to allocate a new buffered value (from the heap)
220 if (UseTLAB) {
221 __ mov(rbx, rax);
222 __ andptr(rbx, -2);
223
224 __ movl(r14, Address(rbx, Klass::layout_helper_offset()));
225
226 __ movptr(r13, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
227 __ lea(r14, Address(r13, r14, Address::times_1));
228 __ cmpptr(r14, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
229 __ jcc(Assembler::above, slow_case);
230 __ movptr(Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())), r14);
231
232 if (UseBiasedLocking) {
233 __ movptr(rax, Address(rbx, Klass::prototype_header_offset()));
234 __ movptr(Address(r13, oopDesc::mark_offset_in_bytes ()), rax);
235 } else {
236 __ movptr(Address(r13, oopDesc::mark_offset_in_bytes ()),
237 (intptr_t)markOopDesc::prototype());
238 }
239 __ xorl(rax, rax); // use zero reg to clear memory (shorter code)
240 __ store_klass_gap(r13, rax); // zero klass gap for compressed oops
241 __ mov(rax, rbx);
242 __ store_klass(r13, rbx); // klass
243
244 // We have our new buffered value, initialize its fields with a
245 // value class specific handler
246 __ movptr(rbx, Address(rax, ValueKlass::pack_handler_offset()));
247 __ mov(rax, r13);
248 __ call(rbx);
249 __ jmp(skip);
250 }
251
252 __ bind(slow_case);
253 // We failed to allocate a new value, fall back to a runtime
254 // call. Some oop field may be live in some registers but we can't
255 // tell. That runtime call will take care of preserving them
256 // across a GC if there's one.
257 __ super_call_VM_leaf(StubRoutines::store_value_type_fields_to_buf());
258 __ bind(skip);
259 #endif
260 }
261
262 __ restore_bcp();
263 __ restore_locals();
264
265 if (state == atos) {
266 Register mdp = rbx;
267 Register tmp = rcx;
268 __ profile_return_type(mdp, rax, tmp);
269 }
270
271 const Register cache = rbx;
272 const Register index = rcx;
273 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
274
275 const Register flags = cache;
276 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
277 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
278 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
279
280 const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
281 if (JvmtiExport::can_pop_frame()) {
282 NOT_LP64(__ get_thread(java_thread));
283 __ check_and_handle_popframe(java_thread);
284 }
285 if (JvmtiExport::can_force_early_return()) {
286 NOT_LP64(__ get_thread(java_thread));
287 __ check_and_handle_earlyret(java_thread);
288 }
289
290 __ dispatch_next(state, step);
291
292 return entry;
293 }
294
295
296 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
297 address entry = __ pc();
298
299 #ifndef _LP64
300 if (state == ftos) {
301 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
302 } else if (state == dtos) {
303 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
304 }
305 #endif // _LP64
306
307 // NULL last_sp until next java call
308 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
309 __ restore_bcp();
310 __ restore_locals();
311 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
312 NOT_LP64(__ get_thread(thread));
313 #if INCLUDE_JVMCI
314 // Check if we need to take lock at entry of synchronized method. This can
315 // only occur on method entry so emit it only for vtos with step 0.
316 if ((EnableJVMCI || UseAOT) && state == vtos && step == 0) {
317 Label L;
318 __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
319 __ jcc(Assembler::zero, L);
320 // Clear flag.
321 __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
322 // Satisfy calling convention for lock_method().
323 __ get_method(rbx);
324 // Take lock.
325 lock_method();
326 __ bind(L);
327 } else {
328 #ifdef ASSERT
329 if (EnableJVMCI) {
330 Label L;
331 __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
332 __ jccb(Assembler::zero, L);
333 __ stop("unexpected pending monitor in deopt entry");
334 __ bind(L);
335 }
336 #endif
337 }
338 #endif
339 // handle exceptions
340 {
341 Label L;
342 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
343 __ jcc(Assembler::zero, L);
344 __ call_VM(noreg,
345 CAST_FROM_FN_PTR(address,
346 InterpreterRuntime::throw_pending_exception));
347 __ should_not_reach_here();
348 __ bind(L);
349 }
350 if (continuation == NULL) {
351 __ dispatch_next(state, step);
352 } else {
353 __ jump_to_entry(continuation);
354 }
355 return entry;
356 }
357
358 address TemplateInterpreterGenerator::generate_result_handler_for(
359 BasicType type) {
360 address entry = __ pc();
361 switch (type) {
362 case T_BOOLEAN: __ c2bool(rax); break;
363 #ifndef _LP64
364 case T_CHAR : __ andptr(rax, 0xFFFF); break;
365 #else
366 case T_CHAR : __ movzwl(rax, rax); break;
367 #endif // _LP64
368 case T_BYTE : __ sign_extend_byte(rax); break;
369 case T_SHORT : __ sign_extend_short(rax); break;
370 case T_INT : /* nothing to do */ break;
371 case T_LONG : /* nothing to do */ break;
372 case T_VOID : /* nothing to do */ break;
373 #ifndef _LP64
374 case T_DOUBLE :
375 case T_FLOAT :
376 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
377 __ pop(t); // remove return address first
378 // Must return a result for interpreter or compiler. In SSE
379 // mode, results are returned in xmm0 and the FPU stack must
380 // be empty.
381 if (type == T_FLOAT && UseSSE >= 1) {
382 // Load ST0
383 __ fld_d(Address(rsp, 0));
384 // Store as float and empty fpu stack
385 __ fstp_s(Address(rsp, 0));
386 // and reload
387 __ movflt(xmm0, Address(rsp, 0));
388 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
389 __ movdbl(xmm0, Address(rsp, 0));
390 } else {
391 // restore ST0
392 __ fld_d(Address(rsp, 0));
393 }
394 // and pop the temp
395 __ addptr(rsp, 2 * wordSize);
396 __ push(t); // restore return address
397 }
398 break;
399 #else
400 case T_FLOAT : /* nothing to do */ break;
401 case T_DOUBLE : /* nothing to do */ break;
402 #endif // _LP64
403
404 case T_VALUETYPE: // fall through (value types are handled with oops)
405 case T_OBJECT :
406 // retrieve result from frame
407 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
408 // and verify it
409 __ verify_oop(rax);
410 break;
411 default : ShouldNotReachHere();
412 }
413 __ ret(0); // return from result handler
414 return entry;
415 }
416
417 address TemplateInterpreterGenerator::generate_safept_entry_for(
418 TosState state,
419 address runtime_entry) {
420 address entry = __ pc();
421 __ push(state);
422 __ call_VM(noreg, runtime_entry);
423 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
424 return entry;
425 }
426
427
428
429 // Helpers for commoning out cases in the various type of method entries.
430 //
431
432
433 // increment invocation count & check for overflow
434 //
435 // Note: checking for negative value instead of overflow
436 // so we have a 'sticky' overflow test
437 //
438 // rbx: method
439 // rcx: invocation counter
440 //
441 void TemplateInterpreterGenerator::generate_counter_incr(
442 Label* overflow,
443 Label* profile_method,
444 Label* profile_method_continue) {
445 Label done;
446 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
447 if (TieredCompilation) {
448 int increment = InvocationCounter::count_increment;
449 Label no_mdo;
450 if (ProfileInterpreter) {
451 // Are we profiling?
452 __ movptr(rax, Address(rbx, Method::method_data_offset()));
453 __ testptr(rax, rax);
454 __ jccb(Assembler::zero, no_mdo);
455 // Increment counter in the MDO
456 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
457 in_bytes(InvocationCounter::counter_offset()));
458 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
459 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
460 __ jmp(done);
461 }
462 __ bind(no_mdo);
463 // Increment counter in MethodCounters
464 const Address invocation_counter(rax,
465 MethodCounters::invocation_counter_offset() +
466 InvocationCounter::counter_offset());
467 __ get_method_counters(rbx, rax, done);
468 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
469 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
470 false, Assembler::zero, overflow);
471 __ bind(done);
472 } else { // not TieredCompilation
473 const Address backedge_counter(rax,
474 MethodCounters::backedge_counter_offset() +
475 InvocationCounter::counter_offset());
476 const Address invocation_counter(rax,
477 MethodCounters::invocation_counter_offset() +
478 InvocationCounter::counter_offset());
479
480 __ get_method_counters(rbx, rax, done);
481
482 if (ProfileInterpreter) {
483 __ incrementl(Address(rax,
484 MethodCounters::interpreter_invocation_counter_offset()));
485 }
486 // Update standard invocation counters
487 __ movl(rcx, invocation_counter);
488 __ incrementl(rcx, InvocationCounter::count_increment);
489 __ movl(invocation_counter, rcx); // save invocation count
490
491 __ movl(rax, backedge_counter); // load backedge counter
492 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
493
494 __ addl(rcx, rax); // add both counters
495
496 // profile_method is non-null only for interpreted method so
497 // profile_method != NULL == !native_call
498
499 if (ProfileInterpreter && profile_method != NULL) {
500 // Test to see if we should create a method data oop
501 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
502 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
503 __ jcc(Assembler::less, *profile_method_continue);
504
505 // if no method data exists, go to profile_method
506 __ test_method_data_pointer(rax, *profile_method);
507 }
508
509 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
510 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
511 __ jcc(Assembler::aboveEqual, *overflow);
512 __ bind(done);
513 }
514 }
515
516 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
517
518 // Asm interpreter on entry
519 // r14/rdi - locals
520 // r13/rsi - bcp
521 // rbx - method
522 // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
523 // rbp - interpreter frame
524
525 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
526 // Everything as it was on entry
527 // rdx is not restored. Doesn't appear to really be set.
528
529 // InterpreterRuntime::frequency_counter_overflow takes two
530 // arguments, the first (thread) is passed by call_VM, the second
531 // indicates if the counter overflow occurs at a backwards branch
532 // (NULL bcp). We pass zero for it. The call returns the address
533 // of the verified entry point for the method or NULL if the
534 // compilation did not complete (either went background or bailed
535 // out).
536 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
537 __ movl(rarg, 0);
538 __ call_VM(noreg,
539 CAST_FROM_FN_PTR(address,
540 InterpreterRuntime::frequency_counter_overflow),
541 rarg);
542
543 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
544 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
545 // and jump to the interpreted entry.
546 __ jmp(do_continue, relocInfo::none);
547 }
548
549 // See if we've got enough room on the stack for locals plus overhead below
550 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
551 // without going through the signal handler, i.e., reserved and yellow zones
552 // will not be made usable. The shadow zone must suffice to handle the
553 // overflow.
554 // The expression stack grows down incrementally, so the normal guard
555 // page mechanism will work for that.
556 //
557 // NOTE: Since the additional locals are also always pushed (wasn't
558 // obvious in generate_fixed_frame) so the guard should work for them
559 // too.
560 //
561 // Args:
562 // rdx: number of additional locals this frame needs (what we must check)
563 // rbx: Method*
564 //
565 // Kills:
566 // rax
567 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
568
569 // monitor entry size: see picture of stack in frame_x86.hpp
570 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
571
572 // total overhead size: entry_size + (saved rbp through expr stack
573 // bottom). be sure to change this if you add/subtract anything
574 // to/from the overhead area
575 const int overhead_size =
576 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
577
578 const int page_size = os::vm_page_size();
579
580 Label after_frame_check;
581
582 // see if the frame is greater than one page in size. If so,
583 // then we need to verify there is enough stack space remaining
584 // for the additional locals.
585 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
586 __ jcc(Assembler::belowEqual, after_frame_check);
587
588 // compute rsp as if this were going to be the last frame on
589 // the stack before the red zone
590
591 Label after_frame_check_pop;
592 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
593 #ifndef _LP64
594 __ push(thread);
595 __ get_thread(thread);
596 #endif
597
598 const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
599
600 // locals + overhead, in bytes
601 __ mov(rax, rdx);
602 __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
603 __ addptr(rax, overhead_size);
604
605 #ifdef ASSERT
606 Label limit_okay;
607 // Verify that thread stack overflow limit is non-zero.
608 __ cmpptr(stack_limit, (int32_t)NULL_WORD);
609 __ jcc(Assembler::notEqual, limit_okay);
610 __ stop("stack overflow limit is zero");
611 __ bind(limit_okay);
612 #endif
613
614 // Add locals/frame size to stack limit.
615 __ addptr(rax, stack_limit);
616
617 // Check against the current stack bottom.
618 __ cmpptr(rsp, rax);
619
620 __ jcc(Assembler::above, after_frame_check_pop);
621 NOT_LP64(__ pop(rsi)); // get saved bcp
622
623 // Restore sender's sp as SP. This is necessary if the sender's
624 // frame is an extended compiled frame (see gen_c2i_adapter())
625 // and safer anyway in case of JSR292 adaptations.
626
627 __ pop(rax); // return address must be moved if SP is changed
628 __ mov(rsp, rbcp);
629 __ push(rax);
630
631 // Note: the restored frame is not necessarily interpreted.
632 // Use the shared runtime version of the StackOverflowError.
633 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
634 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
635 // all done with frame size check
636 __ bind(after_frame_check_pop);
637 NOT_LP64(__ pop(rsi));
638
639 // all done with frame size check
640 __ bind(after_frame_check);
641 }
642
643 // Allocate monitor and lock method (asm interpreter)
644 //
645 // Args:
646 // rbx: Method*
647 // r14/rdi: locals
648 //
649 // Kills:
650 // rax
651 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
652 // rscratch1, rscratch2 (scratch regs)
653 void TemplateInterpreterGenerator::lock_method() {
654 // synchronize method
655 const Address access_flags(rbx, Method::access_flags_offset());
656 const Address monitor_block_top(
657 rbp,
658 frame::interpreter_frame_monitor_block_top_offset * wordSize);
659 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
660
661 #ifdef ASSERT
662 {
663 Label L;
664 __ movl(rax, access_flags);
665 __ testl(rax, JVM_ACC_SYNCHRONIZED);
666 __ jcc(Assembler::notZero, L);
667 __ stop("method doesn't need synchronization");
668 __ bind(L);
669 }
670 #endif // ASSERT
671
672 // get synchronization object
673 {
674 Label check_value_recv, done;
675 __ movl(rax, access_flags);
676 __ testl(rax, JVM_ACC_STATIC);
677 // get receiver (assume this is frequent case)
678 __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
679 __ jcc(Assembler::zero, (EnableValhalla) ? check_value_recv : done);
680 __ load_mirror(rax, rbx);
681
682 #ifdef ASSERT
683 {
684 Label L;
685 __ testptr(rax, rax);
686 __ jcc(Assembler::notZero, L);
687 __ stop("synchronization object is NULL");
688 __ bind(L);
689 }
690 #endif // ASSERT
691 __ jmp(done);
692 if (EnableValhalla) {
693 __ bind(check_value_recv);
694 __ test_oop_is_not_value(rax, rbx, done);
695 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
696 InterpreterRuntime::throw_illegal_monitor_state_exception));
697 __ should_not_reach_here();
698 }
699 __ bind(done);
700 }
701
702 // add space for monitor & lock
703 __ subptr(rsp, entry_size); // add space for a monitor entry
704 __ movptr(monitor_block_top, rsp); // set new monitor block top
705 // store object
706 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
707 const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
708 __ movptr(lockreg, rsp); // object address
709 __ lock_object(lockreg);
710 }
711
712 // Generate a fixed interpreter frame. This is identical setup for
713 // interpreted methods and for native methods hence the shared code.
714 //
715 // Args:
716 // rax: return address
717 // rbx: Method*
718 // r14/rdi: pointer to locals
719 // r13/rsi: sender sp
720 // rdx: cp cache
721 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
722 // initialize fixed part of activation frame
723 __ push(rax); // save return address
724 __ enter(); // save old & set new rbp
725 __ push(rbcp); // set sender sp
726 __ push((int)NULL_WORD); // leave last_sp as null
727 __ movptr(rbcp, Address(rbx, Method::const_offset())); // get ConstMethod*
728 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
729 __ push(rbx); // save Method*
730 // Get mirror and store it in the frame as GC root for this Method*
731 __ load_mirror(rdx, rbx);
732 __ push(rdx);
733 if (ProfileInterpreter) {
734 Label method_data_continue;
735 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
736 __ testptr(rdx, rdx);
737 __ jcc(Assembler::zero, method_data_continue);
738 __ addptr(rdx, in_bytes(MethodData::data_offset()));
739 __ bind(method_data_continue);
740 __ push(rdx); // set the mdp (method data pointer)
741 } else {
742 __ push(0);
743 }
744
745 __ movptr(rdx, Address(rbx, Method::const_offset()));
746 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
747 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
748 __ push(rdx); // set constant pool cache
749 const Register thread1 = NOT_LP64(rdx) LP64_ONLY(r15_thread);
750 NOT_LP64(__ get_thread(thread1));
751 __ movptr(rdx, Address(thread1, JavaThread::vt_alloc_ptr_offset()));
752 __ push(rdx); // value type allocation pointer when activation is created
753 __ push(rlocals); // set locals pointer
754 if (native_call) {
755 __ push(0); // no bcp
756 } else {
757 __ push(rbcp); // set bcp
758 }
759 __ push(0); // reserve word for pointer to expression stack bottom
760 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
761 }
762
763 // End of helpers
764
765 // Method entry for java.lang.ref.Reference.get.
766 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
767 #if INCLUDE_ALL_GCS
768 // Code: _aload_0, _getfield, _areturn
769 // parameter size = 1
770 //
771 // The code that gets generated by this routine is split into 2 parts:
772 // 1. The "intrinsified" code for G1 (or any SATB based GC),
773 // 2. The slow path - which is an expansion of the regular method entry.
774 //
775 // Notes:-
776 // * In the G1 code we do not check whether we need to block for
777 // a safepoint. If G1 is enabled then we must execute the specialized
778 // code for Reference.get (except when the Reference object is null)
779 // so that we can log the value in the referent field with an SATB
780 // update buffer.
781 // If the code for the getfield template is modified so that the
782 // G1 pre-barrier code is executed when the current method is
783 // Reference.get() then going through the normal method entry
784 // will be fine.
785 // * The G1 code can, however, check the receiver object (the instance
786 // of java.lang.Reference) and jump to the slow path if null. If the
787 // Reference object is null then we obviously cannot fetch the referent
788 // and so we don't need to call the G1 pre-barrier. Thus we can use the
789 // regular method entry code to generate the NPE.
790 //
791 // rbx: Method*
792
793 // r13: senderSP must preserve for slow path, set SP to it on fast path
794
795 address entry = __ pc();
796
797 const int referent_offset = java_lang_ref_Reference::referent_offset;
798 guarantee(referent_offset > 0, "referent offset not initialized");
799
800 if (UseG1GC) {
801 Label slow_path;
802 // rbx: method
803
804 // Check if local 0 != NULL
805 // If the receiver is null then it is OK to jump to the slow path.
806 __ movptr(rax, Address(rsp, wordSize));
807
808 __ testptr(rax, rax);
809 __ jcc(Assembler::zero, slow_path);
810
811 // rax: local 0
812 // rbx: method (but can be used as scratch now)
813 // rdx: scratch
814 // rdi: scratch
815
816 // Preserve the sender sp in case the pre-barrier
817 // calls the runtime
818 NOT_LP64(__ push(rsi));
819
820 // Generate the G1 pre-barrier code to log the value of
821 // the referent field in an SATB buffer.
822
823 // Load the value of the referent field.
824 const Address field_address(rax, referent_offset);
825 __ load_heap_oop(rax, field_address);
826
827 const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
828 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
829 NOT_LP64(__ get_thread(thread));
830
831 // Generate the G1 pre-barrier code to log the value of
832 // the referent field in an SATB buffer.
833 __ g1_write_barrier_pre(noreg /* obj */,
834 rax /* pre_val */,
835 thread /* thread */,
836 rbx /* tmp */,
837 true /* tosca_live */,
838 true /* expand_call */);
839
840 // _areturn
841 NOT_LP64(__ pop(rsi)); // get sender sp
842 __ pop(rdi); // get return address
843 __ mov(rsp, sender_sp); // set sp to sender sp
844 __ jmp(rdi);
845 __ ret(0);
846
847 // generate a vanilla interpreter entry as the slow path
848 __ bind(slow_path);
849 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
850 return entry;
851 }
852 #endif // INCLUDE_ALL_GCS
853
854 // If G1 is not enabled then attempt to go through the accessor entry point
855 // Reference.get is an accessor
856 return NULL;
857 }
858
859 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
860 // Quick & dirty stack overflow checking: bang the stack & handle trap.
861 // Note that we do the banging after the frame is setup, since the exception
862 // handling code expects to find a valid interpreter frame on the stack.
863 // Doing the banging earlier fails if the caller frame is not an interpreter
864 // frame.
865 // (Also, the exception throwing code expects to unlock any synchronized
866 // method receiever, so do the banging after locking the receiver.)
867
868 // Bang each page in the shadow zone. We can't assume it's been done for
869 // an interpreter frame with greater than a page of locals, so each page
870 // needs to be checked. Only true for non-native.
871 if (UseStackBanging) {
872 const int page_size = os::vm_page_size();
873 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
874 const int start_page = native_call ? n_shadow_pages : 1;
875 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
876 __ bang_stack_with_offset(pages*page_size);
877 }
878 }
879 }
880
881 // Interpreter stub for calling a native method. (asm interpreter)
882 // This sets up a somewhat different looking stack for calling the
883 // native method than the typical interpreter frame setup.
884 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
885 // determine code generation flags
886 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
887
888 // rbx: Method*
889 // rbcp: sender sp
890
891 address entry_point = __ pc();
892
893 const Address constMethod (rbx, Method::const_offset());
894 const Address access_flags (rbx, Method::access_flags_offset());
895 const Address size_of_parameters(rcx, ConstMethod::
896 size_of_parameters_offset());
897
898
899 // get parameter size (always needed)
900 __ movptr(rcx, constMethod);
901 __ load_unsigned_short(rcx, size_of_parameters);
902
903 // native calls don't need the stack size check since they have no
904 // expression stack and the arguments are already on the stack and
905 // we only add a handful of words to the stack
906
907 // rbx: Method*
908 // rcx: size of parameters
909 // rbcp: sender sp
910 __ pop(rax); // get return address
911
912 // for natives the size of locals is zero
913
914 // compute beginning of parameters
915 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
916
917 // add 2 zero-initialized slots for native calls
918 // initialize result_handler slot
919 __ push((int) NULL_WORD);
920 // slot for oop temp
921 // (static native method holder mirror/jni oop result)
922 __ push((int) NULL_WORD);
923
924 // initialize fixed part of activation frame
925 generate_fixed_frame(true);
926
927 // make sure method is native & not abstract
928 #ifdef ASSERT
929 __ movl(rax, access_flags);
930 {
931 Label L;
932 __ testl(rax, JVM_ACC_NATIVE);
933 __ jcc(Assembler::notZero, L);
934 __ stop("tried to execute non-native method as native");
935 __ bind(L);
936 }
937 {
938 Label L;
939 __ testl(rax, JVM_ACC_ABSTRACT);
940 __ jcc(Assembler::zero, L);
941 __ stop("tried to execute abstract method in interpreter");
942 __ bind(L);
943 }
944 #endif
945
946 // Since at this point in the method invocation the exception handler
947 // would try to exit the monitor of synchronized methods which hasn't
948 // been entered yet, we set the thread local variable
949 // _do_not_unlock_if_synchronized to true. The remove_activation will
950 // check this flag.
951
952 const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
953 NOT_LP64(__ get_thread(thread1));
954 const Address do_not_unlock_if_synchronized(thread1,
955 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
956 __ movbool(do_not_unlock_if_synchronized, true);
957
958 // increment invocation count & check for overflow
959 Label invocation_counter_overflow;
960 if (inc_counter) {
961 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
962 }
963
964 Label continue_after_compile;
965 __ bind(continue_after_compile);
966
967 bang_stack_shadow_pages(true);
968
969 // reset the _do_not_unlock_if_synchronized flag
970 NOT_LP64(__ get_thread(thread1));
971 __ movbool(do_not_unlock_if_synchronized, false);
972
973 // check for synchronized methods
974 // Must happen AFTER invocation_counter check and stack overflow check,
975 // so method is not locked if overflows.
976 if (synchronized) {
977 lock_method();
978 } else {
979 // no synchronization necessary
980 #ifdef ASSERT
981 {
982 Label L;
983 __ movl(rax, access_flags);
984 __ testl(rax, JVM_ACC_SYNCHRONIZED);
985 __ jcc(Assembler::zero, L);
986 __ stop("method needs synchronization");
987 __ bind(L);
988 }
989 #endif
990 }
991
992 // start execution
993 #ifdef ASSERT
994 {
995 Label L;
996 const Address monitor_block_top(rbp,
997 frame::interpreter_frame_monitor_block_top_offset * wordSize);
998 __ movptr(rax, monitor_block_top);
999 __ cmpptr(rax, rsp);
1000 __ jcc(Assembler::equal, L);
1001 __ stop("broken stack frame setup in interpreter");
1002 __ bind(L);
1003 }
1004 #endif
1005
1006 // jvmti support
1007 __ notify_method_entry();
1008
1009 // work registers
1010 const Register method = rbx;
1011 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
1012 const Register t = NOT_LP64(rcx) LP64_ONLY(r11);
1013
1014 // allocate space for parameters
1015 __ get_method(method);
1016 __ movptr(t, Address(method, Method::const_offset()));
1017 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1018
1019 #ifndef _LP64
1020 __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
1021 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
1022 __ subptr(rsp, t);
1023 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
1024 #else
1025 __ shll(t, Interpreter::logStackElementSize);
1026
1027 __ subptr(rsp, t);
1028 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1029 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
1030 #endif // _LP64
1031
1032 // get signature handler
1033 {
1034 Label L;
1035 __ movptr(t, Address(method, Method::signature_handler_offset()));
1036 __ testptr(t, t);
1037 __ jcc(Assembler::notZero, L);
1038 __ call_VM(noreg,
1039 CAST_FROM_FN_PTR(address,
1040 InterpreterRuntime::prepare_native_call),
1041 method);
1042 __ get_method(method);
1043 __ movptr(t, Address(method, Method::signature_handler_offset()));
1044 __ bind(L);
1045 }
1046
1047 // call signature handler
1048 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1049 "adjust this code");
1050 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
1051 "adjust this code");
1052 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
1053 "adjust this code");
1054
1055 // The generated handlers do not touch RBX (the method oop).
1056 // However, large signatures cannot be cached and are generated
1057 // each time here. The slow-path generator can do a GC on return,
1058 // so we must reload it after the call.
1059 __ call(t);
1060 __ get_method(method); // slow path can do a GC, reload RBX
1061
1062
1063 // result handler is in rax
1064 // set result handler
1065 __ movptr(Address(rbp,
1066 (frame::interpreter_frame_result_handler_offset) * wordSize),
1067 rax);
1068
1069 // pass mirror handle if static call
1070 {
1071 Label L;
1072 __ movl(t, Address(method, Method::access_flags_offset()));
1073 __ testl(t, JVM_ACC_STATIC);
1074 __ jcc(Assembler::zero, L);
1075 // get mirror
1076 __ load_mirror(t, method);
1077 // copy mirror into activation frame
1078 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1079 t);
1080 // pass handle to mirror
1081 #ifndef _LP64
1082 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1083 __ movptr(Address(rsp, wordSize), t);
1084 #else
1085 __ lea(c_rarg1,
1086 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1087 #endif // _LP64
1088 __ bind(L);
1089 }
1090
1091 // get native function entry point
1092 {
1093 Label L;
1094 __ movptr(rax, Address(method, Method::native_function_offset()));
1095 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1096 __ cmpptr(rax, unsatisfied.addr());
1097 __ jcc(Assembler::notEqual, L);
1098 __ call_VM(noreg,
1099 CAST_FROM_FN_PTR(address,
1100 InterpreterRuntime::prepare_native_call),
1101 method);
1102 __ get_method(method);
1103 __ movptr(rax, Address(method, Method::native_function_offset()));
1104 __ bind(L);
1105 }
1106
1107 // pass JNIEnv
1108 #ifndef _LP64
1109 __ get_thread(thread);
1110 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1111 __ movptr(Address(rsp, 0), t);
1112
1113 // set_last_Java_frame_before_call
1114 // It is enough that the pc()
1115 // points into the right code segment. It does not have to be the correct return pc.
1116 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1117 #else
1118 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1119
1120 // It is enough that the pc() points into the right code
1121 // segment. It does not have to be the correct return pc.
1122 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1123 #endif // _LP64
1124
1125 // change thread state
1126 #ifdef ASSERT
1127 {
1128 Label L;
1129 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1130 __ cmpl(t, _thread_in_Java);
1131 __ jcc(Assembler::equal, L);
1132 __ stop("Wrong thread state in native stub");
1133 __ bind(L);
1134 }
1135 #endif
1136
1137 // Change state to native
1138
1139 __ movl(Address(thread, JavaThread::thread_state_offset()),
1140 _thread_in_native);
1141
1142 // Call the native method.
1143 __ call(rax);
1144 // 32: result potentially in rdx:rax or ST0
1145 // 64: result potentially in rax or xmm0
1146
1147 // Verify or restore cpu control state after JNI call
1148 __ restore_cpu_control_state_after_jni();
1149
1150 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1151 // in order to extract the result of a method call. If the order of these
1152 // pushes change or anything else is added to the stack then the code in
1153 // interpreter_frame_result must also change.
1154
1155 #ifndef _LP64
1156 // save potential result in ST(0) & rdx:rax
1157 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1158 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1159 // It is safe to do this push because state is _thread_in_native and return address will be found
1160 // via _last_native_pc and not via _last_jave_sp
1161
1162 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1163 // If the order changes or anything else is added to the stack the code in
1164 // interpreter_frame_result will have to be changed.
1165
1166 { Label L;
1167 Label push_double;
1168 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1169 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1170 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1171 float_handler.addr());
1172 __ jcc(Assembler::equal, push_double);
1173 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1174 double_handler.addr());
1175 __ jcc(Assembler::notEqual, L);
1176 __ bind(push_double);
1177 __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1178 __ bind(L);
1179 }
1180 #else
1181 __ push(dtos);
1182 #endif // _LP64
1183
1184 __ push(ltos);
1185
1186 // change thread state
1187 NOT_LP64(__ get_thread(thread));
1188 __ movl(Address(thread, JavaThread::thread_state_offset()),
1189 _thread_in_native_trans);
1190
1191 if (os::is_MP()) {
1192 if (UseMembar) {
1193 // Force this write out before the read below
1194 __ membar(Assembler::Membar_mask_bits(
1195 Assembler::LoadLoad | Assembler::LoadStore |
1196 Assembler::StoreLoad | Assembler::StoreStore));
1197 } else {
1198 // Write serialization page so VM thread can do a pseudo remote membar.
1199 // We use the current thread pointer to calculate a thread specific
1200 // offset to write to within the page. This minimizes bus traffic
1201 // due to cache line collision.
1202 __ serialize_memory(thread, rcx);
1203 }
1204 }
1205
1206 #ifndef _LP64
1207 if (AlwaysRestoreFPU) {
1208 // Make sure the control word is correct.
1209 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1210 }
1211 #endif // _LP64
1212
1213 // check for safepoint operation in progress and/or pending suspend requests
1214 {
1215 Label Continue;
1216 Label slow_path;
1217
1218 #ifndef _LP64
1219 __ safepoint_poll(slow_path, thread, noreg);
1220 #else
1221 __ safepoint_poll(slow_path, r15_thread, rscratch1);
1222 #endif
1223
1224 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1225 __ jcc(Assembler::equal, Continue);
1226 __ bind(slow_path);
1227
1228 // Don't use call_VM as it will see a possible pending exception
1229 // and forward it and never return here preventing us from
1230 // clearing _last_native_pc down below. Also can't use
1231 // call_VM_leaf either as it will check to see if r13 & r14 are
1232 // preserved and correspond to the bcp/locals pointers. So we do a
1233 // runtime call by hand.
1234 //
1235 #ifndef _LP64
1236 __ push(thread);
1237 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1238 JavaThread::check_special_condition_for_native_trans)));
1239 __ increment(rsp, wordSize);
1240 __ get_thread(thread);
1241 #else
1242 __ mov(c_rarg0, r15_thread);
1243 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1244 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1245 __ andptr(rsp, -16); // align stack as required by ABI
1246 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1247 __ mov(rsp, r12); // restore sp
1248 __ reinit_heapbase();
1249 #endif // _LP64
1250 __ bind(Continue);
1251 }
1252
1253 // change thread state
1254 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1255
1256 // reset_last_Java_frame
1257 __ reset_last_Java_frame(thread, true);
1258
1259 if (CheckJNICalls) {
1260 // clear_pending_jni_exception_check
1261 __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1262 }
1263
1264 // reset handle block
1265 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1266 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1267
1268 // If result is an oop unbox and store it in frame where gc will see it
1269 // and result handler will pick it up
1270
1271 {
1272 Label no_oop, not_weak, store_result;
1273 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1274 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1275 __ jcc(Assembler::notEqual, no_oop);
1276 // retrieve result
1277 __ pop(ltos);
1278 // Unbox oop result, e.g. JNIHandles::resolve value.
1279 __ resolve_jobject(rax /* value */,
1280 thread /* thread */,
1281 t /* tmp */);
1282 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1283 // keep stack depth as expected by pushing oop which will eventually be discarded
1284 __ push(ltos);
1285 __ bind(no_oop);
1286 }
1287
1288
1289 {
1290 Label no_reguard;
1291 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1292 JavaThread::stack_guard_yellow_reserved_disabled);
1293 __ jcc(Assembler::notEqual, no_reguard);
1294
1295 __ pusha(); // XXX only save smashed registers
1296 #ifndef _LP64
1297 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1298 __ popa();
1299 #else
1300 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1301 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1302 __ andptr(rsp, -16); // align stack as required by ABI
1303 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1304 __ mov(rsp, r12); // restore sp
1305 __ popa(); // XXX only restore smashed registers
1306 __ reinit_heapbase();
1307 #endif // _LP64
1308
1309 __ bind(no_reguard);
1310 }
1311
1312
1313 // The method register is junk from after the thread_in_native transition
1314 // until here. Also can't call_VM until the bcp has been
1315 // restored. Need bcp for throwing exception below so get it now.
1316 __ get_method(method);
1317
1318 // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1319 __ movptr(rbcp, Address(method, Method::const_offset())); // get ConstMethod*
1320 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
1321
1322 // handle exceptions (exception handling will handle unlocking!)
1323 {
1324 Label L;
1325 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1326 __ jcc(Assembler::zero, L);
1327 // Note: At some point we may want to unify this with the code
1328 // used in call_VM_base(); i.e., we should use the
1329 // StubRoutines::forward_exception code. For now this doesn't work
1330 // here because the rsp is not correctly set at this point.
1331 __ MacroAssembler::call_VM(noreg,
1332 CAST_FROM_FN_PTR(address,
1333 InterpreterRuntime::throw_pending_exception));
1334 __ should_not_reach_here();
1335 __ bind(L);
1336 }
1337
1338 // do unlocking if necessary
1339 {
1340 Label L;
1341 __ movl(t, Address(method, Method::access_flags_offset()));
1342 __ testl(t, JVM_ACC_SYNCHRONIZED);
1343 __ jcc(Assembler::zero, L);
1344 // the code below should be shared with interpreter macro
1345 // assembler implementation
1346 {
1347 Label unlock;
1348 // BasicObjectLock will be first in list, since this is a
1349 // synchronized method. However, need to check that the object
1350 // has not been unlocked by an explicit monitorexit bytecode.
1351 const Address monitor(rbp,
1352 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1353 wordSize - (int)sizeof(BasicObjectLock)));
1354
1355 const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1356
1357 // monitor expect in c_rarg1 for slow unlock path
1358 __ lea(regmon, monitor); // address of first monitor
1359
1360 __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1361 __ testptr(t, t);
1362 __ jcc(Assembler::notZero, unlock);
1363
1364 // Entry already unlocked, need to throw exception
1365 __ MacroAssembler::call_VM(noreg,
1366 CAST_FROM_FN_PTR(address,
1367 InterpreterRuntime::throw_illegal_monitor_state_exception));
1368 __ should_not_reach_here();
1369
1370 __ bind(unlock);
1371 __ unlock_object(regmon);
1372 }
1373 __ bind(L);
1374 }
1375
1376 // jvmti support
1377 // Note: This must happen _after_ handling/throwing any exceptions since
1378 // the exception handler code notifies the runtime of method exits
1379 // too. If this happens before, method entry/exit notifications are
1380 // not properly paired (was bug - gri 11/22/99).
1381 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1382
1383 // restore potential result in edx:eax, call result handler to
1384 // restore potential result in ST0 & handle result
1385
1386 __ pop(ltos);
1387 LP64_ONLY( __ pop(dtos));
1388
1389 __ movptr(t, Address(rbp,
1390 (frame::interpreter_frame_result_handler_offset) * wordSize));
1391 __ call(t);
1392
1393 // remove activation
1394 __ movptr(t, Address(rbp,
1395 frame::interpreter_frame_sender_sp_offset *
1396 wordSize)); // get sender sp
1397 __ leave(); // remove frame anchor
1398 __ pop(rdi); // get return address
1399 __ mov(rsp, t); // set sp to sender sp
1400 __ jmp(rdi);
1401
1402 if (inc_counter) {
1403 // Handle overflow of counter and compile method
1404 __ bind(invocation_counter_overflow);
1405 generate_counter_overflow(continue_after_compile);
1406 }
1407
1408 return entry_point;
1409 }
1410
1411 // Abstract method entry
1412 // Attempt to execute abstract method. Throw exception
1413 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1414
1415 address entry_point = __ pc();
1416
1417 // abstract method entry
1418
1419 // pop return address, reset last_sp to NULL
1420 __ empty_expression_stack();
1421 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1422 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1423
1424 // throw exception
1425 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
1426 // the call_VM checks for exception, so we should never return here.
1427 __ should_not_reach_here();
1428
1429 return entry_point;
1430 }
1431
1432 //
1433 // Generic interpreted method entry to (asm) interpreter
1434 //
1435 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1436 // determine code generation flags
1437 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1438
1439 // ebx: Method*
1440 // rbcp: sender sp
1441 address entry_point = __ pc();
1442
1443 const Address constMethod(rbx, Method::const_offset());
1444 const Address access_flags(rbx, Method::access_flags_offset());
1445 const Address size_of_parameters(rdx,
1446 ConstMethod::size_of_parameters_offset());
1447 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1448
1449
1450 // get parameter size (always needed)
1451 __ movptr(rdx, constMethod);
1452 __ load_unsigned_short(rcx, size_of_parameters);
1453
1454 // rbx: Method*
1455 // rcx: size of parameters
1456 // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1457
1458 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1459 __ subl(rdx, rcx); // rdx = no. of additional locals
1460
1461 // YYY
1462 // __ incrementl(rdx);
1463 // __ andl(rdx, -2);
1464
1465 // see if we've got enough room on the stack for locals plus overhead.
1466 generate_stack_overflow_check();
1467
1468 // get return address
1469 __ pop(rax);
1470
1471 // compute beginning of parameters
1472 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1473
1474 // rdx - # of additional locals
1475 // allocate space for locals
1476 // explicitly initialize locals
1477 {
1478 Label exit, loop;
1479 __ testl(rdx, rdx);
1480 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1481 __ bind(loop);
1482 __ push((int) NULL_WORD); // initialize local variables
1483 __ decrementl(rdx); // until everything initialized
1484 __ jcc(Assembler::greater, loop);
1485 __ bind(exit);
1486 }
1487
1488 // initialize fixed part of activation frame
1489 generate_fixed_frame(false);
1490
1491 // make sure method is not native & not abstract
1492 #ifdef ASSERT
1493 __ movl(rax, access_flags);
1494 {
1495 Label L;
1496 __ testl(rax, JVM_ACC_NATIVE);
1497 __ jcc(Assembler::zero, L);
1498 __ stop("tried to execute native method as non-native");
1499 __ bind(L);
1500 }
1501 {
1502 Label L;
1503 __ testl(rax, JVM_ACC_ABSTRACT);
1504 __ jcc(Assembler::zero, L);
1505 __ stop("tried to execute abstract method in interpreter");
1506 __ bind(L);
1507 }
1508 #endif
1509
1510 // Since at this point in the method invocation the exception
1511 // handler would try to exit the monitor of synchronized methods
1512 // which hasn't been entered yet, we set the thread local variable
1513 // _do_not_unlock_if_synchronized to true. The remove_activation
1514 // will check this flag.
1515
1516 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1517 NOT_LP64(__ get_thread(thread));
1518 const Address do_not_unlock_if_synchronized(thread,
1519 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1520 __ movbool(do_not_unlock_if_synchronized, true);
1521
1522 __ profile_parameters_type(rax, rcx, rdx);
1523 // increment invocation count & check for overflow
1524 Label invocation_counter_overflow;
1525 Label profile_method;
1526 Label profile_method_continue;
1527 if (inc_counter) {
1528 generate_counter_incr(&invocation_counter_overflow,
1529 &profile_method,
1530 &profile_method_continue);
1531 if (ProfileInterpreter) {
1532 __ bind(profile_method_continue);
1533 }
1534 }
1535
1536 Label continue_after_compile;
1537 __ bind(continue_after_compile);
1538
1539 // check for synchronized interpreted methods
1540 bang_stack_shadow_pages(false);
1541
1542 // reset the _do_not_unlock_if_synchronized flag
1543 NOT_LP64(__ get_thread(thread));
1544 __ movbool(do_not_unlock_if_synchronized, false);
1545
1546 // check for synchronized methods
1547 // Must happen AFTER invocation_counter check and stack overflow check,
1548 // so method is not locked if overflows.
1549 if (synchronized) {
1550 // Allocate monitor and lock method
1551 lock_method();
1552 } else {
1553 // no synchronization necessary
1554 #ifdef ASSERT
1555 {
1556 Label L;
1557 __ movl(rax, access_flags);
1558 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1559 __ jcc(Assembler::zero, L);
1560 __ stop("method needs synchronization");
1561 __ bind(L);
1562 }
1563 #endif
1564 }
1565
1566 // start execution
1567 #ifdef ASSERT
1568 {
1569 Label L;
1570 const Address monitor_block_top (rbp,
1571 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1572 __ movptr(rax, monitor_block_top);
1573 __ cmpptr(rax, rsp);
1574 __ jcc(Assembler::equal, L);
1575 __ stop("broken stack frame setup in interpreter");
1576 __ bind(L);
1577 }
1578 #endif
1579
1580 // jvmti support
1581 __ notify_method_entry();
1582
1583 __ dispatch_next(vtos);
1584
1585 // invocation counter overflow
1586 if (inc_counter) {
1587 if (ProfileInterpreter) {
1588 // We have decided to profile this method in the interpreter
1589 __ bind(profile_method);
1590 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1591 __ set_method_data_pointer_for_bcp();
1592 __ get_method(rbx);
1593 __ jmp(profile_method_continue);
1594 }
1595 // Handle overflow of counter and compile method
1596 __ bind(invocation_counter_overflow);
1597 generate_counter_overflow(continue_after_compile);
1598 }
1599
1600 return entry_point;
1601 }
1602
1603 //-----------------------------------------------------------------------------
1604 // Exceptions
1605
1606 void TemplateInterpreterGenerator::generate_throw_exception() {
1607 // Entry point in previous activation (i.e., if the caller was
1608 // interpreted)
1609 Interpreter::_rethrow_exception_entry = __ pc();
1610 // Restore sp to interpreter_frame_last_sp even though we are going
1611 // to empty the expression stack for the exception processing.
1612 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1613 // rax: exception
1614 // rdx: return address/pc that threw exception
1615 __ restore_bcp(); // r13/rsi points to call/send
1616 __ restore_locals();
1617 LP64_ONLY(__ reinit_heapbase()); // restore r12 as heapbase.
1618 // Entry point for exceptions thrown within interpreter code
1619 Interpreter::_throw_exception_entry = __ pc();
1620 // expression stack is undefined here
1621 // rax: exception
1622 // r13/rsi: exception bcp
1623 __ verify_oop(rax);
1624 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1625 LP64_ONLY(__ mov(c_rarg1, rax));
1626
1627 // expression stack must be empty before entering the VM in case of
1628 // an exception
1629 __ empty_expression_stack();
1630 // find exception handler address and preserve exception oop
1631 __ call_VM(rdx,
1632 CAST_FROM_FN_PTR(address,
1633 InterpreterRuntime::exception_handler_for_exception),
1634 rarg);
1635 // rax: exception handler entry point
1636 // rdx: preserved exception oop
1637 // r13/rsi: bcp for exception handler
1638 __ push_ptr(rdx); // push exception which is now the only value on the stack
1639 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1640
1641 // If the exception is not handled in the current frame the frame is
1642 // removed and the exception is rethrown (i.e. exception
1643 // continuation is _rethrow_exception).
1644 //
1645 // Note: At this point the bci is still the bxi for the instruction
1646 // which caused the exception and the expression stack is
1647 // empty. Thus, for any VM calls at this point, GC will find a legal
1648 // oop map (with empty expression stack).
1649
1650 // In current activation
1651 // tos: exception
1652 // esi: exception bcp
1653
1654 //
1655 // JVMTI PopFrame support
1656 //
1657
1658 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1659 __ empty_expression_stack();
1660 // Set the popframe_processing bit in pending_popframe_condition
1661 // indicating that we are currently handling popframe, so that
1662 // call_VMs that may happen later do not trigger new popframe
1663 // handling cycles.
1664 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1665 NOT_LP64(__ get_thread(thread));
1666 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1667 __ orl(rdx, JavaThread::popframe_processing_bit);
1668 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1669
1670 {
1671 // Check to see whether we are returning to a deoptimized frame.
1672 // (The PopFrame call ensures that the caller of the popped frame is
1673 // either interpreted or compiled and deoptimizes it if compiled.)
1674 // In this case, we can't call dispatch_next() after the frame is
1675 // popped, but instead must save the incoming arguments and restore
1676 // them after deoptimization has occurred.
1677 //
1678 // Note that we don't compare the return PC against the
1679 // deoptimization blob's unpack entry because of the presence of
1680 // adapter frames in C2.
1681 Label caller_not_deoptimized;
1682 Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1683 __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1684 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1685 InterpreterRuntime::interpreter_contains), rarg);
1686 __ testl(rax, rax);
1687 __ jcc(Assembler::notZero, caller_not_deoptimized);
1688
1689 // Compute size of arguments for saving when returning to
1690 // deoptimized caller
1691 __ get_method(rax);
1692 __ movptr(rax, Address(rax, Method::const_offset()));
1693 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1694 size_of_parameters_offset())));
1695 __ shll(rax, Interpreter::logStackElementSize);
1696 __ restore_locals();
1697 __ subptr(rlocals, rax);
1698 __ addptr(rlocals, wordSize);
1699 // Save these arguments
1700 NOT_LP64(__ get_thread(thread));
1701 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1702 Deoptimization::
1703 popframe_preserve_args),
1704 thread, rax, rlocals);
1705
1706 __ remove_activation(vtos, rdx,
1707 /* throw_monitor_exception */ false,
1708 /* install_monitor_exception */ false,
1709 /* notify_jvmdi */ false);
1710
1711 // Inform deoptimization that it is responsible for restoring
1712 // these arguments
1713 NOT_LP64(__ get_thread(thread));
1714 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1715 JavaThread::popframe_force_deopt_reexecution_bit);
1716
1717 // Continue in deoptimization handler
1718 __ jmp(rdx);
1719
1720 __ bind(caller_not_deoptimized);
1721 }
1722
1723 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1724 /* throw_monitor_exception */ false,
1725 /* install_monitor_exception */ false,
1726 /* notify_jvmdi */ false);
1727
1728 // Finish with popframe handling
1729 // A previous I2C followed by a deoptimization might have moved the
1730 // outgoing arguments further up the stack. PopFrame expects the
1731 // mutations to those outgoing arguments to be preserved and other
1732 // constraints basically require this frame to look exactly as
1733 // though it had previously invoked an interpreted activation with
1734 // no space between the top of the expression stack (current
1735 // last_sp) and the top of stack. Rather than force deopt to
1736 // maintain this kind of invariant all the time we call a small
1737 // fixup routine to move the mutated arguments onto the top of our
1738 // expression stack if necessary.
1739 #ifndef _LP64
1740 __ mov(rax, rsp);
1741 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1742 __ get_thread(thread);
1743 // PC must point into interpreter here
1744 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1745 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1746 __ get_thread(thread);
1747 #else
1748 __ mov(c_rarg1, rsp);
1749 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1750 // PC must point into interpreter here
1751 __ set_last_Java_frame(noreg, rbp, __ pc());
1752 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1753 #endif
1754 __ reset_last_Java_frame(thread, true);
1755
1756 // Restore the last_sp and null it out
1757 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1758 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1759
1760 __ restore_bcp();
1761 __ restore_locals();
1762 // The method data pointer was incremented already during
1763 // call profiling. We have to restore the mdp for the current bcp.
1764 if (ProfileInterpreter) {
1765 __ set_method_data_pointer_for_bcp();
1766 }
1767
1768 // Clear the popframe condition flag
1769 NOT_LP64(__ get_thread(thread));
1770 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1771 JavaThread::popframe_inactive);
1772
1773 #if INCLUDE_JVMTI
1774 {
1775 Label L_done;
1776 const Register local0 = rlocals;
1777
1778 __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1779 __ jcc(Assembler::notEqual, L_done);
1780
1781 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1782 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1783
1784 __ get_method(rdx);
1785 __ movptr(rax, Address(local0, 0));
1786 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1787
1788 __ testptr(rax, rax);
1789 __ jcc(Assembler::zero, L_done);
1790
1791 __ movptr(Address(rbx, 0), rax);
1792 __ bind(L_done);
1793 }
1794 #endif // INCLUDE_JVMTI
1795
1796 __ dispatch_next(vtos);
1797 // end of PopFrame support
1798
1799 Interpreter::_remove_activation_entry = __ pc();
1800
1801 // preserve exception over this code sequence
1802 __ pop_ptr(rax);
1803 NOT_LP64(__ get_thread(thread));
1804 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1805 // remove the activation (without doing throws on illegalMonitorExceptions)
1806 __ remove_activation(vtos, rdx, false, true, false);
1807 // restore exception
1808 NOT_LP64(__ get_thread(thread));
1809 __ get_vm_result(rax, thread);
1810
1811 // In between activations - previous activation type unknown yet
1812 // compute continuation point - the continuation point expects the
1813 // following registers set up:
1814 //
1815 // rax: exception
1816 // rdx: return address/pc that threw exception
1817 // rsp: expression stack of caller
1818 // rbp: ebp of caller
1819 __ push(rax); // save exception
1820 __ push(rdx); // save return address
1821 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1822 SharedRuntime::exception_handler_for_return_address),
1823 thread, rdx);
1824 __ mov(rbx, rax); // save exception handler
1825 __ pop(rdx); // restore return address
1826 __ pop(rax); // restore exception
1827 // Note that an "issuing PC" is actually the next PC after the call
1828 __ jmp(rbx); // jump to exception
1829 // handler of caller
1830 }
1831
1832
1833 //
1834 // JVMTI ForceEarlyReturn support
1835 //
1836 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1837 address entry = __ pc();
1838
1839 __ restore_bcp();
1840 __ restore_locals();
1841 __ empty_expression_stack();
1842 __ load_earlyret_value(state); // 32 bits returns value in rdx, so don't reuse
1843
1844 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1845 NOT_LP64(__ get_thread(thread));
1846 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1847 Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1848
1849 // Clear the earlyret state
1850 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1851
1852 __ remove_activation(state, rsi,
1853 false, /* throw_monitor_exception */
1854 false, /* install_monitor_exception */
1855 true); /* notify_jvmdi */
1856 __ jmp(rsi);
1857
1858 return entry;
1859 } // end of ForceEarlyReturn support
1860
1861
1862 //-----------------------------------------------------------------------------
1863 // Helper for vtos entry point generation
1864
1865 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1866 address& bep,
1867 address& cep,
1868 address& sep,
1869 address& aep,
1870 address& iep,
1871 address& lep,
1872 address& fep,
1873 address& dep,
1874 address& vep) {
1875 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1876 Label L;
1877 aep = __ pc(); __ push_ptr(); __ jmp(L);
1878 #ifndef _LP64
1879 fep = __ pc(); __ push(ftos); __ jmp(L);
1880 dep = __ pc(); __ push(dtos); __ jmp(L);
1881 #else
1882 fep = __ pc(); __ push_f(xmm0); __ jmp(L);
1883 dep = __ pc(); __ push_d(xmm0); __ jmp(L);
1884 #endif // _LP64
1885 lep = __ pc(); __ push_l(); __ jmp(L);
1886 bep = cep = sep =
1887 iep = __ pc(); __ push_i();
1888 vep = __ pc();
1889 __ bind(L);
1890 generate_and_dispatch(t);
1891 }
1892
1893 //-----------------------------------------------------------------------------
1894
1895 // Non-product code
1896 #ifndef PRODUCT
1897
1898 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1899 address entry = __ pc();
1900
1901 #ifndef _LP64
1902 // prepare expression stack
1903 __ pop(rcx); // pop return address so expression stack is 'pure'
1904 __ push(state); // save tosca
1905
1906 // pass tosca registers as arguments & call tracer
1907 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1908 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1909 __ pop(state); // restore tosca
1910
1911 // return
1912 __ jmp(rcx);
1913 #else
1914 __ push(state);
1915 __ push(c_rarg0);
1916 __ push(c_rarg1);
1917 __ push(c_rarg2);
1918 __ push(c_rarg3);
1919 __ mov(c_rarg2, rax); // Pass itos
1920 #ifdef _WIN64
1921 __ movflt(xmm3, xmm0); // Pass ftos
1922 #endif
1923 __ call_VM(noreg,
1924 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1925 c_rarg1, c_rarg2, c_rarg3);
1926 __ pop(c_rarg3);
1927 __ pop(c_rarg2);
1928 __ pop(c_rarg1);
1929 __ pop(c_rarg0);
1930 __ pop(state);
1931 __ ret(0); // return from result handler
1932 #endif // _LP64
1933
1934 return entry;
1935 }
1936
1937 void TemplateInterpreterGenerator::count_bytecode() {
1938 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1939 }
1940
1941 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1942 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1943 }
1944
1945 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1946 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1947 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1948 __ orl(rbx,
1949 ((int) t->bytecode()) <<
1950 BytecodePairHistogram::log2_number_of_codes);
1951 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1952 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1953 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1954 }
1955
1956
1957 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1958 // Call a little run-time stub to avoid blow-up for each bytecode.
1959 // The run-time runtime saves the right registers, depending on
1960 // the tosca in-state for the given template.
1961
1962 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1963 "entry must have been generated");
1964 #ifndef _LP64
1965 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1966 #else
1967 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1968 __ andptr(rsp, -16); // align stack as required by ABI
1969 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1970 __ mov(rsp, r12); // restore sp
1971 __ reinit_heapbase();
1972 #endif // _LP64
1973 }
1974
1975
1976 void TemplateInterpreterGenerator::stop_interpreter_at() {
1977 Label L;
1978 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1979 StopInterpreterAt);
1980 __ jcc(Assembler::notEqual, L);
1981 __ int3();
1982 __ bind(L);
1983 }
1984 #endif // !PRODUCT