1 /*
2 * Copyright (c) 2003, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/bytecodeHistogram.hpp"
28 #include "interpreter/interp_masm.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/templateInterpreterGenerator.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "oops/arrayOop.hpp"
34 #include "oops/methodData.hpp"
35 #include "oops/method.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/jvmtiExport.hpp"
38 #include "prims/jvmtiThreadState.hpp"
39 #include "runtime/arguments.hpp"
40 #include "runtime/deoptimization.hpp"
41 #include "runtime/frame.inline.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/synchronizer.hpp"
45 #include "runtime/timer.hpp"
46 #include "runtime/vframeArray.hpp"
47 #include "utilities/debug.hpp"
48 #include "utilities/macros.hpp"
49
50 #define __ _masm->
51
52 // Size of interpreter code. Increase if too small. Interpreter will
53 // fail with a guarantee ("not enough space for interpreter generation");
54 // if too small.
55 // Run with +PrintInterpreter to get the VM to print out the size.
56 // Max size with JVMTI
57 #ifdef AMD64
58 int TemplateInterpreter::InterpreterCodeSize = 256 * 1024;
59 #else
60 int TemplateInterpreter::InterpreterCodeSize = 224 * 1024;
61 #endif // AMD64
62
63 // Global Register Names
64 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
65 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
66
67 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
68 const int bcp_offset = frame::interpreter_frame_bcp_offset * wordSize;
69 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
70
71
72 //-----------------------------------------------------------------------------
73
74 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
75 address entry = __ pc();
76
77 #ifdef ASSERT
78 {
79 Label L;
80 __ lea(rax, Address(rbp,
81 frame::interpreter_frame_monitor_block_top_offset *
82 wordSize));
83 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
84 // grows negative)
85 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
86 __ stop ("interpreter frame not set up");
87 __ bind(L);
88 }
89 #endif // ASSERT
90 // Restore bcp under the assumption that the current frame is still
91 // interpreted
92 __ restore_bcp();
93
94 // expression stack must be empty before entering the VM if an
95 // exception happened
96 __ empty_expression_stack();
97 // throw exception
98 __ call_VM(noreg,
99 CAST_FROM_FN_PTR(address,
100 InterpreterRuntime::throw_StackOverflowError));
101 return entry;
102 }
103
104 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
105 const char* name) {
106 address entry = __ pc();
107 // expression stack must be empty before entering the VM if an
108 // exception happened
109 __ empty_expression_stack();
110 // setup parameters
111 // ??? convention: expect aberrant index in register ebx
112 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
113 __ lea(rarg, ExternalAddress((address)name));
114 __ call_VM(noreg,
115 CAST_FROM_FN_PTR(address,
116 InterpreterRuntime::
117 throw_ArrayIndexOutOfBoundsException),
118 rarg, rbx);
119 return entry;
120 }
121
122 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
123 address entry = __ pc();
124
125 // object is at TOS
126 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
127 __ pop(rarg);
128
129 // expression stack must be empty before entering the VM if an
130 // exception happened
131 __ empty_expression_stack();
132
133 __ call_VM(noreg,
134 CAST_FROM_FN_PTR(address,
135 InterpreterRuntime::
136 throw_ClassCastException),
137 rarg);
138 return entry;
139 }
140
141 address TemplateInterpreterGenerator::generate_exception_handler_common(
142 const char* name, const char* message, bool pass_oop) {
143 assert(!pass_oop || message == NULL, "either oop or message but not both");
144 address entry = __ pc();
145
146 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
147 Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
148
149 if (pass_oop) {
150 // object is at TOS
151 __ pop(rarg2);
152 }
153 // expression stack must be empty before entering the VM if an
154 // exception happened
155 __ empty_expression_stack();
156 // setup parameters
157 __ lea(rarg, ExternalAddress((address)name));
158 if (pass_oop) {
159 __ call_VM(rax, CAST_FROM_FN_PTR(address,
160 InterpreterRuntime::
161 create_klass_exception),
162 rarg, rarg2);
163 } else {
164 __ lea(rarg2, ExternalAddress((address)message));
165 __ call_VM(rax,
166 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
167 rarg, rarg2);
168 }
169 // throw exception
170 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
171 return entry;
172 }
173
174
175 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
176 address entry = __ pc();
177 // NULL last_sp until next java call
178 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
179 __ dispatch_next(state);
180 return entry;
181 }
182
183
184 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
185 address entry = __ pc();
186
187 #ifndef _LP64
188 #ifdef COMPILER2
189 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
190 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
191 for (int i = 1; i < 8; i++) {
192 __ ffree(i);
193 }
194 } else if (UseSSE < 2) {
195 __ empty_FPU_stack();
196 }
197 #endif // COMPILER2
198 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
199 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
200 } else {
201 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
202 }
203
204 if (state == ftos) {
205 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
206 } else if (state == dtos) {
207 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
208 }
209 #endif // _LP64
210
211 // Restore stack bottom in case i2c adjusted stack
212 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
213 // and NULL it as marker that esp is now tos until next java call
214 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
215
216 __ restore_bcp();
217 __ restore_locals();
218
219 if (state == atos) {
220 Register mdp = rbx;
221 Register tmp = rcx;
222 __ profile_return_type(mdp, rax, tmp);
223 }
224
225 const Register cache = rbx;
226 const Register index = rcx;
227 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
228
229 const Register flags = cache;
230 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
231 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
232 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
233 __ dispatch_next(state, step);
234
235 return entry;
236 }
237
238
239 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
240 address entry = __ pc();
241
242 #ifndef _LP64
243 if (state == ftos) {
244 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
245 } else if (state == dtos) {
246 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
247 }
248 #endif // _LP64
249
250 // NULL last_sp until next java call
251 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
252 __ restore_bcp();
253 __ restore_locals();
254 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
255 NOT_LP64(__ get_thread(thread));
256 #if INCLUDE_JVMCI
257 // Check if we need to take lock at entry of synchronized method.
258 if (UseJVMCICompiler) {
259 Label L;
260 __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
261 __ jcc(Assembler::zero, L);
262 // Clear flag.
263 __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
264 // Satisfy calling convention for lock_method().
265 __ get_method(rbx);
266 // Take lock.
267 lock_method();
268 __ bind(L);
269 }
270 #endif
271 // handle exceptions
272 {
273 Label L;
274 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
275 __ jcc(Assembler::zero, L);
276 __ call_VM(noreg,
277 CAST_FROM_FN_PTR(address,
278 InterpreterRuntime::throw_pending_exception));
279 __ should_not_reach_here();
280 __ bind(L);
281 }
282 __ dispatch_next(state, step);
283 return entry;
284 }
285
286 address TemplateInterpreterGenerator::generate_result_handler_for(
287 BasicType type) {
288 address entry = __ pc();
289 switch (type) {
290 case T_BOOLEAN: __ c2bool(rax); break;
291 #ifndef _LP64
292 case T_CHAR : __ andptr(rax, 0xFFFF); break;
293 #else
294 case T_CHAR : __ movzwl(rax, rax); break;
295 #endif // _LP64
296 case T_BYTE : __ sign_extend_byte(rax); break;
297 case T_SHORT : __ sign_extend_short(rax); break;
298 case T_INT : /* nothing to do */ break;
299 case T_LONG : /* nothing to do */ break;
300 case T_VOID : /* nothing to do */ break;
301 #ifndef _LP64
302 case T_DOUBLE :
303 case T_FLOAT :
304 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
305 __ pop(t); // remove return address first
306 // Must return a result for interpreter or compiler. In SSE
307 // mode, results are returned in xmm0 and the FPU stack must
308 // be empty.
309 if (type == T_FLOAT && UseSSE >= 1) {
310 // Load ST0
311 __ fld_d(Address(rsp, 0));
312 // Store as float and empty fpu stack
313 __ fstp_s(Address(rsp, 0));
314 // and reload
315 __ movflt(xmm0, Address(rsp, 0));
316 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
317 __ movdbl(xmm0, Address(rsp, 0));
318 } else {
319 // restore ST0
320 __ fld_d(Address(rsp, 0));
321 }
322 // and pop the temp
323 __ addptr(rsp, 2 * wordSize);
324 __ push(t); // restore return address
325 }
326 break;
327 #else
328 case T_FLOAT : /* nothing to do */ break;
329 case T_DOUBLE : /* nothing to do */ break;
330 #endif // _LP64
331
332 case T_OBJECT :
333 // retrieve result from frame
334 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
335 // and verify it
336 __ verify_oop(rax);
337 break;
338 default : ShouldNotReachHere();
339 }
340 __ ret(0); // return from result handler
341 return entry;
342 }
343
344 address TemplateInterpreterGenerator::generate_safept_entry_for(
345 TosState state,
346 address runtime_entry) {
347 address entry = __ pc();
348 __ push(state);
349 __ call_VM(noreg, runtime_entry);
350 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
351 return entry;
352 }
353
354
355
356 // Helpers for commoning out cases in the various type of method entries.
357 //
358
359
360 // increment invocation count & check for overflow
361 //
362 // Note: checking for negative value instead of overflow
363 // so we have a 'sticky' overflow test
364 //
365 // rbx: method
366 // rcx: invocation counter
367 //
368 void TemplateInterpreterGenerator::generate_counter_incr(
369 Label* overflow,
370 Label* profile_method,
371 Label* profile_method_continue) {
372 Label done;
373 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
374 if (TieredCompilation) {
375 int increment = InvocationCounter::count_increment;
376 Label no_mdo;
377 if (ProfileInterpreter) {
378 // Are we profiling?
379 __ movptr(rax, Address(rbx, Method::method_data_offset()));
380 __ testptr(rax, rax);
381 __ jccb(Assembler::zero, no_mdo);
382 // Increment counter in the MDO
383 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
384 in_bytes(InvocationCounter::counter_offset()));
385 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
386 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
387 __ jmp(done);
388 }
389 __ bind(no_mdo);
390 // Increment counter in MethodCounters
391 const Address invocation_counter(rax,
392 MethodCounters::invocation_counter_offset() +
393 InvocationCounter::counter_offset());
394 __ get_method_counters(rbx, rax, done);
395 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
396 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
397 false, Assembler::zero, overflow);
398 __ bind(done);
399 } else { // not TieredCompilation
400 const Address backedge_counter(rax,
401 MethodCounters::backedge_counter_offset() +
402 InvocationCounter::counter_offset());
403 const Address invocation_counter(rax,
404 MethodCounters::invocation_counter_offset() +
405 InvocationCounter::counter_offset());
406
407 __ get_method_counters(rbx, rax, done);
408
409 if (ProfileInterpreter) {
410 __ incrementl(Address(rax,
411 MethodCounters::interpreter_invocation_counter_offset()));
412 }
413 // Update standard invocation counters
414 __ movl(rcx, invocation_counter);
415 __ incrementl(rcx, InvocationCounter::count_increment);
416 __ movl(invocation_counter, rcx); // save invocation count
417
418 __ movl(rax, backedge_counter); // load backedge counter
419 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
420
421 __ addl(rcx, rax); // add both counters
422
423 // profile_method is non-null only for interpreted method so
424 // profile_method != NULL == !native_call
425
426 if (ProfileInterpreter && profile_method != NULL) {
427 // Test to see if we should create a method data oop
428 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
429 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
430 __ jcc(Assembler::less, *profile_method_continue);
431
432 // if no method data exists, go to profile_method
433 __ test_method_data_pointer(rax, *profile_method);
434 }
435
436 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
437 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
438 __ jcc(Assembler::aboveEqual, *overflow);
439 __ bind(done);
440 }
441 }
442
443 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
444
445 // Asm interpreter on entry
446 // r14/rdi - locals
447 // r13/rsi - bcp
448 // rbx - method
449 // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
450 // rbp - interpreter frame
451
452 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
453 // Everything as it was on entry
454 // rdx is not restored. Doesn't appear to really be set.
455
456 // InterpreterRuntime::frequency_counter_overflow takes two
457 // arguments, the first (thread) is passed by call_VM, the second
458 // indicates if the counter overflow occurs at a backwards branch
459 // (NULL bcp). We pass zero for it. The call returns the address
460 // of the verified entry point for the method or NULL if the
461 // compilation did not complete (either went background or bailed
462 // out).
463 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
464 __ movl(rarg, 0);
465 __ call_VM(noreg,
466 CAST_FROM_FN_PTR(address,
467 InterpreterRuntime::frequency_counter_overflow),
468 rarg);
469
470 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
471 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
472 // and jump to the interpreted entry.
473 __ jmp(do_continue, relocInfo::none);
474 }
475
476 // See if we've got enough room on the stack for locals plus overhead below
477 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
478 // without going through the signal handler, i.e., reserved and yellow zones
479 // will not be made usable. The shadow zone must suffice to handle the
480 // overflow.
481 // The expression stack grows down incrementally, so the normal guard
482 // page mechanism will work for that.
483 //
484 // NOTE: Since the additional locals are also always pushed (wasn't
485 // obvious in generate_fixed_frame) so the guard should work for them
486 // too.
487 //
488 // Args:
489 // rdx: number of additional locals this frame needs (what we must check)
490 // rbx: Method*
491 //
492 // Kills:
493 // rax
494 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
495
496 // monitor entry size: see picture of stack in frame_x86.hpp
497 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
498
499 // total overhead size: entry_size + (saved rbp through expr stack
500 // bottom). be sure to change this if you add/subtract anything
501 // to/from the overhead area
502 const int overhead_size =
503 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
504
505 const int page_size = os::vm_page_size();
506
507 Label after_frame_check;
508
509 // see if the frame is greater than one page in size. If so,
510 // then we need to verify there is enough stack space remaining
511 // for the additional locals.
512 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
513 __ jcc(Assembler::belowEqual, after_frame_check);
514
515 // compute rsp as if this were going to be the last frame on
516 // the stack before the red zone
517
518 Label after_frame_check_pop;
519 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
520 #ifndef _LP64
521 __ push(thread);
522 __ get_thread(thread);
523 #endif
524
525 const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
526
527 // locals + overhead, in bytes
528 __ mov(rax, rdx);
529 __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
530 __ addptr(rax, overhead_size);
531
532 #ifdef ASSERT
533 Label limit_okay;
534 // Verify that thread stack overflow limit is non-zero.
535 __ cmpptr(stack_limit, (int32_t)NULL_WORD);
536 __ jcc(Assembler::notEqual, limit_okay);
537 __ stop("stack overflow limit is zero");
538 __ bind(limit_okay);
539 #endif
540
541 // Add locals/frame size to stack limit.
542 __ addptr(rax, stack_limit);
543
544 // Check against the current stack bottom.
545 __ cmpptr(rsp, rax);
546
547 __ jcc(Assembler::above, after_frame_check_pop);
548 NOT_LP64(__ pop(rsi)); // get saved bcp
549
550 // Restore sender's sp as SP. This is necessary if the sender's
551 // frame is an extended compiled frame (see gen_c2i_adapter())
552 // and safer anyway in case of JSR292 adaptations.
553
554 __ pop(rax); // return address must be moved if SP is changed
555 __ mov(rsp, rbcp);
556 __ push(rax);
557
558 // Note: the restored frame is not necessarily interpreted.
559 // Use the shared runtime version of the StackOverflowError.
560 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
561 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
562 // all done with frame size check
563 __ bind(after_frame_check_pop);
564 NOT_LP64(__ pop(rsi));
565
566 // all done with frame size check
567 __ bind(after_frame_check);
568 }
569
570 // Allocate monitor and lock method (asm interpreter)
571 //
572 // Args:
573 // rbx: Method*
574 // r14/rdi: locals
575 //
576 // Kills:
577 // rax
578 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
579 // rscratch1, rscratch2 (scratch regs)
580 void TemplateInterpreterGenerator::lock_method() {
581 // synchronize method
582 const Address access_flags(rbx, Method::access_flags_offset());
583 const Address monitor_block_top(
584 rbp,
585 frame::interpreter_frame_monitor_block_top_offset * wordSize);
586 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
587
588 #ifdef ASSERT
589 {
590 Label L;
591 __ movl(rax, access_flags);
592 __ testl(rax, JVM_ACC_SYNCHRONIZED);
593 __ jcc(Assembler::notZero, L);
594 __ stop("method doesn't need synchronization");
595 __ bind(L);
596 }
597 #endif // ASSERT
598
599 // get synchronization object
600 {
601 Label done;
602 __ movl(rax, access_flags);
603 __ testl(rax, JVM_ACC_STATIC);
604 // get receiver (assume this is frequent case)
605 __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
606 __ jcc(Assembler::zero, done);
607 __ load_mirror(rax, rbx);
608
609 #ifdef ASSERT
610 {
611 Label L;
612 __ testptr(rax, rax);
613 __ jcc(Assembler::notZero, L);
614 __ stop("synchronization object is NULL");
615 __ bind(L);
616 }
617 #endif // ASSERT
618
619 __ bind(done);
620 oopDesc::bs()->interpreter_write_barrier(_masm, rax);
621 }
622
623 // add space for monitor & lock
624 __ subptr(rsp, entry_size); // add space for a monitor entry
625 __ movptr(monitor_block_top, rsp); // set new monitor block top
626 // store object
627 __ shenandoah_store_addr_check(rax);
628 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
629 const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
630 __ movptr(lockreg, rsp); // object address
631 __ lock_object(lockreg);
632 }
633
634 // Generate a fixed interpreter frame. This is identical setup for
635 // interpreted methods and for native methods hence the shared code.
636 //
637 // Args:
638 // rax: return address
639 // rbx: Method*
640 // r14/rdi: pointer to locals
641 // r13/rsi: sender sp
642 // rdx: cp cache
643 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
644 // initialize fixed part of activation frame
645 __ push(rax); // save return address
646 __ enter(); // save old & set new rbp
647 __ push(rbcp); // set sender sp
648 __ push((int)NULL_WORD); // leave last_sp as null
649 __ movptr(rbcp, Address(rbx, Method::const_offset())); // get ConstMethod*
650 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
651 __ push(rbx); // save Method*
652 // Get mirror and store it in the frame as GC root for this Method*
653 __ load_mirror(rdx, rbx);
654 __ push(rdx);
655 if (ProfileInterpreter) {
656 Label method_data_continue;
657 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
658 __ testptr(rdx, rdx);
659 __ jcc(Assembler::zero, method_data_continue);
660 __ addptr(rdx, in_bytes(MethodData::data_offset()));
661 __ bind(method_data_continue);
662 __ push(rdx); // set the mdp (method data pointer)
663 } else {
664 __ push(0);
665 }
666
667 __ movptr(rdx, Address(rbx, Method::const_offset()));
668 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
669 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
670 __ push(rdx); // set constant pool cache
671 __ push(rlocals); // set locals pointer
672 if (native_call) {
673 __ push(0); // no bcp
674 } else {
675 __ push(rbcp); // set bcp
676 }
677 __ push(0); // reserve word for pointer to expression stack bottom
678 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
679 }
680
681 // End of helpers
682
683 // Method entry for java.lang.ref.Reference.get.
684 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
685 #if INCLUDE_ALL_GCS
686 // Code: _aload_0, _getfield, _areturn
687 // parameter size = 1
688 //
689 // The code that gets generated by this routine is split into 2 parts:
690 // 1. The "intrinsified" code for G1 (or any SATB based GC),
691 // 2. The slow path - which is an expansion of the regular method entry.
692 //
693 // Notes:-
694 // * In the G1 code we do not check whether we need to block for
695 // a safepoint. If G1 is enabled then we must execute the specialized
696 // code for Reference.get (except when the Reference object is null)
697 // so that we can log the value in the referent field with an SATB
698 // update buffer.
699 // If the code for the getfield template is modified so that the
700 // G1 pre-barrier code is executed when the current method is
701 // Reference.get() then going through the normal method entry
702 // will be fine.
703 // * The G1 code can, however, check the receiver object (the instance
704 // of java.lang.Reference) and jump to the slow path if null. If the
705 // Reference object is null then we obviously cannot fetch the referent
706 // and so we don't need to call the G1 pre-barrier. Thus we can use the
707 // regular method entry code to generate the NPE.
708 //
709 // rbx: Method*
710
711 // r13: senderSP must preserve for slow path, set SP to it on fast path
712
713 address entry = __ pc();
714
715 const int referent_offset = java_lang_ref_Reference::referent_offset;
716 guarantee(referent_offset > 0, "referent offset not initialized");
717
718 if (UseG1GC || UseShenandoahGC) {
719 Label slow_path;
720 // rbx: method
721
722 // Check if local 0 != NULL
723 // If the receiver is null then it is OK to jump to the slow path.
724 __ movptr(rax, Address(rsp, wordSize));
725
726 __ testptr(rax, rax);
727 __ jcc(Assembler::zero, slow_path);
728
729 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rax);
730
731 // rax: local 0
732 // rbx: method (but can be used as scratch now)
733 // rdx: scratch
734 // rdi: scratch
735
736 // Preserve the sender sp in case the pre-barrier
737 // calls the runtime
738 NOT_LP64(__ push(rsi));
739
740 // Generate the G1 pre-barrier code to log the value of
741 // the referent field in an SATB buffer.
742
743 // Load the value of the referent field.
744 const Address field_address(rax, referent_offset);
745 __ load_heap_oop(rax, field_address);
746
747 const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
748 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
749 NOT_LP64(__ get_thread(thread));
750
751 // Generate the G1 pre-barrier code to log the value of
752 // the referent field in an SATB buffer.
753 __ g1_write_barrier_pre(noreg /* obj */,
754 rax /* pre_val */,
755 thread /* thread */,
756 rbx /* tmp */,
757 true /* tosca_live */,
758 true /* expand_call */);
759
760 // _areturn
761 NOT_LP64(__ pop(rsi)); // get sender sp
762 __ pop(rdi); // get return address
763 __ mov(rsp, sender_sp); // set sp to sender sp
764 __ jmp(rdi);
765 __ ret(0);
766
767 // generate a vanilla interpreter entry as the slow path
768 __ bind(slow_path);
769 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
770 return entry;
771 }
772 #endif // INCLUDE_ALL_GCS
773
774 // If G1 is not enabled then attempt to go through the accessor entry point
775 // Reference.get is an accessor
776 return NULL;
777 }
778
779 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
780 // Quick & dirty stack overflow checking: bang the stack & handle trap.
781 // Note that we do the banging after the frame is setup, since the exception
782 // handling code expects to find a valid interpreter frame on the stack.
783 // Doing the banging earlier fails if the caller frame is not an interpreter
784 // frame.
785 // (Also, the exception throwing code expects to unlock any synchronized
786 // method receiever, so do the banging after locking the receiver.)
787
788 // Bang each page in the shadow zone. We can't assume it's been done for
789 // an interpreter frame with greater than a page of locals, so each page
790 // needs to be checked. Only true for non-native.
791 if (UseStackBanging) {
792 const int page_size = os::vm_page_size();
793 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
794 const int start_page = native_call ? n_shadow_pages : 1;
795 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
796 __ bang_stack_with_offset(pages*page_size);
797 }
798 }
799 }
800
801 // Interpreter stub for calling a native method. (asm interpreter)
802 // This sets up a somewhat different looking stack for calling the
803 // native method than the typical interpreter frame setup.
804 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
805 // determine code generation flags
806 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
807
808 // rbx: Method*
809 // rbcp: sender sp
810
811 address entry_point = __ pc();
812
813 const Address constMethod (rbx, Method::const_offset());
814 const Address access_flags (rbx, Method::access_flags_offset());
815 const Address size_of_parameters(rcx, ConstMethod::
816 size_of_parameters_offset());
817
818
819 // get parameter size (always needed)
820 __ movptr(rcx, constMethod);
821 __ load_unsigned_short(rcx, size_of_parameters);
822
823 // native calls don't need the stack size check since they have no
824 // expression stack and the arguments are already on the stack and
825 // we only add a handful of words to the stack
826
827 // rbx: Method*
828 // rcx: size of parameters
829 // rbcp: sender sp
830 __ pop(rax); // get return address
831
832 // for natives the size of locals is zero
833
834 // compute beginning of parameters
835 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
836
837 // add 2 zero-initialized slots for native calls
838 // initialize result_handler slot
839 __ push((int) NULL_WORD);
840 // slot for oop temp
841 // (static native method holder mirror/jni oop result)
842 __ push((int) NULL_WORD);
843
844 // initialize fixed part of activation frame
845 generate_fixed_frame(true);
846
847 // make sure method is native & not abstract
848 #ifdef ASSERT
849 __ movl(rax, access_flags);
850 {
851 Label L;
852 __ testl(rax, JVM_ACC_NATIVE);
853 __ jcc(Assembler::notZero, L);
854 __ stop("tried to execute non-native method as native");
855 __ bind(L);
856 }
857 {
858 Label L;
859 __ testl(rax, JVM_ACC_ABSTRACT);
860 __ jcc(Assembler::zero, L);
861 __ stop("tried to execute abstract method in interpreter");
862 __ bind(L);
863 }
864 #endif
865
866 // Since at this point in the method invocation the exception handler
867 // would try to exit the monitor of synchronized methods which hasn't
868 // been entered yet, we set the thread local variable
869 // _do_not_unlock_if_synchronized to true. The remove_activation will
870 // check this flag.
871
872 const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
873 NOT_LP64(__ get_thread(thread1));
874 const Address do_not_unlock_if_synchronized(thread1,
875 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
876 __ movbool(do_not_unlock_if_synchronized, true);
877
878 // increment invocation count & check for overflow
879 Label invocation_counter_overflow;
880 if (inc_counter) {
881 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
882 }
883
884 Label continue_after_compile;
885 __ bind(continue_after_compile);
886
887 bang_stack_shadow_pages(true);
888
889 // reset the _do_not_unlock_if_synchronized flag
890 NOT_LP64(__ get_thread(thread1));
891 __ movbool(do_not_unlock_if_synchronized, false);
892
893 // check for synchronized methods
894 // Must happen AFTER invocation_counter check and stack overflow check,
895 // so method is not locked if overflows.
896 if (synchronized) {
897 lock_method();
898 } else {
899 // no synchronization necessary
900 #ifdef ASSERT
901 {
902 Label L;
903 __ movl(rax, access_flags);
904 __ testl(rax, JVM_ACC_SYNCHRONIZED);
905 __ jcc(Assembler::zero, L);
906 __ stop("method needs synchronization");
907 __ bind(L);
908 }
909 #endif
910 }
911
912 // start execution
913 #ifdef ASSERT
914 {
915 Label L;
916 const Address monitor_block_top(rbp,
917 frame::interpreter_frame_monitor_block_top_offset * wordSize);
918 __ movptr(rax, monitor_block_top);
919 __ cmpptr(rax, rsp);
920 __ jcc(Assembler::equal, L);
921 __ stop("broken stack frame setup in interpreter");
922 __ bind(L);
923 }
924 #endif
925
926 // jvmti support
927 __ notify_method_entry();
928
929 // work registers
930 const Register method = rbx;
931 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
932 const Register t = NOT_LP64(rcx) LP64_ONLY(r11);
933
934 // allocate space for parameters
935 __ get_method(method);
936 __ movptr(t, Address(method, Method::const_offset()));
937 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
938
939 #ifndef _LP64
940 __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
941 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
942 __ subptr(rsp, t);
943 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
944 #else
945 __ shll(t, Interpreter::logStackElementSize);
946
947 __ subptr(rsp, t);
948 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
949 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
950 #endif // _LP64
951
952 // get signature handler
953 {
954 Label L;
955 __ movptr(t, Address(method, Method::signature_handler_offset()));
956 __ testptr(t, t);
957 __ jcc(Assembler::notZero, L);
958 __ call_VM(noreg,
959 CAST_FROM_FN_PTR(address,
960 InterpreterRuntime::prepare_native_call),
961 method);
962 __ get_method(method);
963 __ movptr(t, Address(method, Method::signature_handler_offset()));
964 __ bind(L);
965 }
966
967 // call signature handler
968 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
969 "adjust this code");
970 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
971 "adjust this code");
972 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
973 "adjust this code");
974
975 // The generated handlers do not touch RBX (the method oop).
976 // However, large signatures cannot be cached and are generated
977 // each time here. The slow-path generator can do a GC on return,
978 // so we must reload it after the call.
979 __ call(t);
980 __ get_method(method); // slow path can do a GC, reload RBX
981
982
983 // result handler is in rax
984 // set result handler
985 __ movptr(Address(rbp,
986 (frame::interpreter_frame_result_handler_offset) * wordSize),
987 rax);
988
989 // pass mirror handle if static call
990 {
991 Label L;
992 __ movl(t, Address(method, Method::access_flags_offset()));
993 __ testl(t, JVM_ACC_STATIC);
994 __ jcc(Assembler::zero, L);
995 // get mirror
996 __ load_mirror(t, method);
997 // copy mirror into activation frame
998 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
999 t);
1000 // pass handle to mirror
1001 #ifndef _LP64
1002 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1003 __ movptr(Address(rsp, wordSize), t);
1004 #else
1005 __ lea(c_rarg1,
1006 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1007 #endif // _LP64
1008 __ bind(L);
1009 }
1010
1011 // get native function entry point
1012 {
1013 Label L;
1014 __ movptr(rax, Address(method, Method::native_function_offset()));
1015 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1016 __ cmpptr(rax, unsatisfied.addr());
1017 __ jcc(Assembler::notEqual, L);
1018 __ call_VM(noreg,
1019 CAST_FROM_FN_PTR(address,
1020 InterpreterRuntime::prepare_native_call),
1021 method);
1022 __ get_method(method);
1023 __ movptr(rax, Address(method, Method::native_function_offset()));
1024 __ bind(L);
1025 }
1026
1027 // pass JNIEnv
1028 #ifndef _LP64
1029 __ get_thread(thread);
1030 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1031 __ movptr(Address(rsp, 0), t);
1032
1033 // set_last_Java_frame_before_call
1034 // It is enough that the pc()
1035 // points into the right code segment. It does not have to be the correct return pc.
1036 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1037 #else
1038 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1039
1040 // It is enough that the pc() points into the right code
1041 // segment. It does not have to be the correct return pc.
1042 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1043 #endif // _LP64
1044
1045 // change thread state
1046 #ifdef ASSERT
1047 {
1048 Label L;
1049 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1050 __ cmpl(t, _thread_in_Java);
1051 __ jcc(Assembler::equal, L);
1052 __ stop("Wrong thread state in native stub");
1053 __ bind(L);
1054 }
1055 #endif
1056
1057 // Change state to native
1058
1059 __ movl(Address(thread, JavaThread::thread_state_offset()),
1060 _thread_in_native);
1061
1062 // Call the native method.
1063 __ call(rax);
1064 // 32: result potentially in rdx:rax or ST0
1065 // 64: result potentially in rax or xmm0
1066
1067 // Verify or restore cpu control state after JNI call
1068 __ restore_cpu_control_state_after_jni();
1069
1070 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1071 // in order to extract the result of a method call. If the order of these
1072 // pushes change or anything else is added to the stack then the code in
1073 // interpreter_frame_result must also change.
1074
1075 #ifndef _LP64
1076 // save potential result in ST(0) & rdx:rax
1077 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1078 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1079 // It is safe to do this push because state is _thread_in_native and return address will be found
1080 // via _last_native_pc and not via _last_jave_sp
1081
1082 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1083 // If the order changes or anything else is added to the stack the code in
1084 // interpreter_frame_result will have to be changed.
1085
1086 { Label L;
1087 Label push_double;
1088 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1089 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1090 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1091 float_handler.addr());
1092 __ jcc(Assembler::equal, push_double);
1093 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1094 double_handler.addr());
1095 __ jcc(Assembler::notEqual, L);
1096 __ bind(push_double);
1097 __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1098 __ bind(L);
1099 }
1100 #else
1101 __ push(dtos);
1102 #endif // _LP64
1103
1104 __ push(ltos);
1105
1106 // change thread state
1107 NOT_LP64(__ get_thread(thread));
1108 __ movl(Address(thread, JavaThread::thread_state_offset()),
1109 _thread_in_native_trans);
1110
1111 if (os::is_MP()) {
1112 if (UseMembar) {
1113 // Force this write out before the read below
1114 __ membar(Assembler::Membar_mask_bits(
1115 Assembler::LoadLoad | Assembler::LoadStore |
1116 Assembler::StoreLoad | Assembler::StoreStore));
1117 } else {
1118 // Write serialization page so VM thread can do a pseudo remote membar.
1119 // We use the current thread pointer to calculate a thread specific
1120 // offset to write to within the page. This minimizes bus traffic
1121 // due to cache line collision.
1122 __ serialize_memory(thread, rcx);
1123 }
1124 }
1125
1126 #ifndef _LP64
1127 if (AlwaysRestoreFPU) {
1128 // Make sure the control word is correct.
1129 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1130 }
1131 #endif // _LP64
1132
1133 // check for safepoint operation in progress and/or pending suspend requests
1134 {
1135 Label Continue;
1136 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1137 SafepointSynchronize::_not_synchronized);
1138
1139 Label L;
1140 __ jcc(Assembler::notEqual, L);
1141 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1142 __ jcc(Assembler::equal, Continue);
1143 __ bind(L);
1144
1145 // Don't use call_VM as it will see a possible pending exception
1146 // and forward it and never return here preventing us from
1147 // clearing _last_native_pc down below. Also can't use
1148 // call_VM_leaf either as it will check to see if r13 & r14 are
1149 // preserved and correspond to the bcp/locals pointers. So we do a
1150 // runtime call by hand.
1151 //
1152 #ifndef _LP64
1153 __ push(thread);
1154 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1155 JavaThread::check_special_condition_for_native_trans)));
1156 __ increment(rsp, wordSize);
1157 __ get_thread(thread);
1158 #else
1159 __ mov(c_rarg0, r15_thread);
1160 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1161 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1162 __ andptr(rsp, -16); // align stack as required by ABI
1163 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1164 __ mov(rsp, r12); // restore sp
1165 __ reinit_heapbase();
1166 #endif // _LP64
1167 __ bind(Continue);
1168 }
1169
1170 // change thread state
1171 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1172
1173 // reset_last_Java_frame
1174 __ reset_last_Java_frame(thread, true);
1175
1176 // reset handle block
1177 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1178 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1179
1180 // If result is an oop unbox and store it in frame where gc will see it
1181 // and result handler will pick it up
1182
1183 {
1184 Label no_oop, store_result;
1185 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1186 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1187 __ jcc(Assembler::notEqual, no_oop);
1188 // retrieve result
1189 __ pop(ltos);
1190 __ testptr(rax, rax);
1191 __ jcc(Assembler::zero, store_result);
1192 __ movptr(rax, Address(rax, 0));
1193 __ bind(store_result);
1194 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1195 // keep stack depth as expected by pushing oop which will eventually be discarded
1196 __ push(ltos);
1197 __ bind(no_oop);
1198 }
1199
1200
1201 {
1202 Label no_reguard;
1203 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1204 JavaThread::stack_guard_yellow_reserved_disabled);
1205 __ jcc(Assembler::notEqual, no_reguard);
1206
1207 __ pusha(); // XXX only save smashed registers
1208 #ifndef _LP64
1209 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1210 __ popa();
1211 #else
1212 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1213 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1214 __ andptr(rsp, -16); // align stack as required by ABI
1215 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1216 __ mov(rsp, r12); // restore sp
1217 __ popa(); // XXX only restore smashed registers
1218 __ reinit_heapbase();
1219 #endif // _LP64
1220
1221 __ bind(no_reguard);
1222 }
1223
1224
1225 // The method register is junk from after the thread_in_native transition
1226 // until here. Also can't call_VM until the bcp has been
1227 // restored. Need bcp for throwing exception below so get it now.
1228 __ get_method(method);
1229
1230 // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1231 __ movptr(rbcp, Address(method, Method::const_offset())); // get ConstMethod*
1232 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
1233
1234 // handle exceptions (exception handling will handle unlocking!)
1235 {
1236 Label L;
1237 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1238 __ jcc(Assembler::zero, L);
1239 // Note: At some point we may want to unify this with the code
1240 // used in call_VM_base(); i.e., we should use the
1241 // StubRoutines::forward_exception code. For now this doesn't work
1242 // here because the rsp is not correctly set at this point.
1243 __ MacroAssembler::call_VM(noreg,
1244 CAST_FROM_FN_PTR(address,
1245 InterpreterRuntime::throw_pending_exception));
1246 __ should_not_reach_here();
1247 __ bind(L);
1248 }
1249
1250 // do unlocking if necessary
1251 {
1252 Label L;
1253 __ movl(t, Address(method, Method::access_flags_offset()));
1254 __ testl(t, JVM_ACC_SYNCHRONIZED);
1255 __ jcc(Assembler::zero, L);
1256 // the code below should be shared with interpreter macro
1257 // assembler implementation
1258 {
1259 Label unlock;
1260 // BasicObjectLock will be first in list, since this is a
1261 // synchronized method. However, need to check that the object
1262 // has not been unlocked by an explicit monitorexit bytecode.
1263 const Address monitor(rbp,
1264 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1265 wordSize - (int)sizeof(BasicObjectLock)));
1266
1267 const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1268
1269 // monitor expect in c_rarg1 for slow unlock path
1270 __ lea(regmon, monitor); // address of first monitor
1271
1272 __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1273 __ shenandoah_store_addr_check(t); // Invariant
1274 __ testptr(t, t);
1275 __ jcc(Assembler::notZero, unlock);
1276
1277 // Entry already unlocked, need to throw exception
1278 __ MacroAssembler::call_VM(noreg,
1279 CAST_FROM_FN_PTR(address,
1280 InterpreterRuntime::throw_illegal_monitor_state_exception));
1281 __ should_not_reach_here();
1282
1283 __ bind(unlock);
1284 __ unlock_object(regmon);
1285 }
1286 __ bind(L);
1287 }
1288
1289 // jvmti support
1290 // Note: This must happen _after_ handling/throwing any exceptions since
1291 // the exception handler code notifies the runtime of method exits
1292 // too. If this happens before, method entry/exit notifications are
1293 // not properly paired (was bug - gri 11/22/99).
1294 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1295
1296 // restore potential result in edx:eax, call result handler to
1297 // restore potential result in ST0 & handle result
1298
1299 __ pop(ltos);
1300 LP64_ONLY( __ pop(dtos));
1301
1302 __ movptr(t, Address(rbp,
1303 (frame::interpreter_frame_result_handler_offset) * wordSize));
1304 __ call(t);
1305
1306 // remove activation
1307 __ movptr(t, Address(rbp,
1308 frame::interpreter_frame_sender_sp_offset *
1309 wordSize)); // get sender sp
1310 __ leave(); // remove frame anchor
1311 __ pop(rdi); // get return address
1312 __ mov(rsp, t); // set sp to sender sp
1313 __ jmp(rdi);
1314
1315 if (inc_counter) {
1316 // Handle overflow of counter and compile method
1317 __ bind(invocation_counter_overflow);
1318 generate_counter_overflow(continue_after_compile);
1319 }
1320
1321 return entry_point;
1322 }
1323
1324 // Abstract method entry
1325 // Attempt to execute abstract method. Throw exception
1326 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1327
1328 address entry_point = __ pc();
1329
1330 // abstract method entry
1331
1332 // pop return address, reset last_sp to NULL
1333 __ empty_expression_stack();
1334 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1335 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1336
1337 // throw exception
1338 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
1339 // the call_VM checks for exception, so we should never return here.
1340 __ should_not_reach_here();
1341
1342 return entry_point;
1343 }
1344
1345 //
1346 // Generic interpreted method entry to (asm) interpreter
1347 //
1348 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1349 // determine code generation flags
1350 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1351
1352 // ebx: Method*
1353 // rbcp: sender sp
1354 address entry_point = __ pc();
1355
1356 const Address constMethod(rbx, Method::const_offset());
1357 const Address access_flags(rbx, Method::access_flags_offset());
1358 const Address size_of_parameters(rdx,
1359 ConstMethod::size_of_parameters_offset());
1360 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1361
1362
1363 // get parameter size (always needed)
1364 __ movptr(rdx, constMethod);
1365 __ load_unsigned_short(rcx, size_of_parameters);
1366
1367 // rbx: Method*
1368 // rcx: size of parameters
1369 // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1370
1371 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1372 __ subl(rdx, rcx); // rdx = no. of additional locals
1373
1374 // YYY
1375 // __ incrementl(rdx);
1376 // __ andl(rdx, -2);
1377
1378 // see if we've got enough room on the stack for locals plus overhead.
1379 generate_stack_overflow_check();
1380
1381 // get return address
1382 __ pop(rax);
1383
1384 // compute beginning of parameters
1385 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1386
1387 // rdx - # of additional locals
1388 // allocate space for locals
1389 // explicitly initialize locals
1390 {
1391 Label exit, loop;
1392 __ testl(rdx, rdx);
1393 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1394 __ bind(loop);
1395 __ push((int) NULL_WORD); // initialize local variables
1396 __ decrementl(rdx); // until everything initialized
1397 __ jcc(Assembler::greater, loop);
1398 __ bind(exit);
1399 }
1400
1401 // initialize fixed part of activation frame
1402 generate_fixed_frame(false);
1403
1404 // make sure method is not native & not abstract
1405 #ifdef ASSERT
1406 __ movl(rax, access_flags);
1407 {
1408 Label L;
1409 __ testl(rax, JVM_ACC_NATIVE);
1410 __ jcc(Assembler::zero, L);
1411 __ stop("tried to execute native method as non-native");
1412 __ bind(L);
1413 }
1414 {
1415 Label L;
1416 __ testl(rax, JVM_ACC_ABSTRACT);
1417 __ jcc(Assembler::zero, L);
1418 __ stop("tried to execute abstract method in interpreter");
1419 __ bind(L);
1420 }
1421 #endif
1422
1423 // Since at this point in the method invocation the exception
1424 // handler would try to exit the monitor of synchronized methods
1425 // which hasn't been entered yet, we set the thread local variable
1426 // _do_not_unlock_if_synchronized to true. The remove_activation
1427 // will check this flag.
1428
1429 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1430 NOT_LP64(__ get_thread(thread));
1431 const Address do_not_unlock_if_synchronized(thread,
1432 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1433 __ movbool(do_not_unlock_if_synchronized, true);
1434
1435 __ profile_parameters_type(rax, rcx, rdx);
1436 // increment invocation count & check for overflow
1437 Label invocation_counter_overflow;
1438 Label profile_method;
1439 Label profile_method_continue;
1440 if (inc_counter) {
1441 generate_counter_incr(&invocation_counter_overflow,
1442 &profile_method,
1443 &profile_method_continue);
1444 if (ProfileInterpreter) {
1445 __ bind(profile_method_continue);
1446 }
1447 }
1448
1449 Label continue_after_compile;
1450 __ bind(continue_after_compile);
1451
1452 // check for synchronized interpreted methods
1453 bang_stack_shadow_pages(false);
1454
1455 // reset the _do_not_unlock_if_synchronized flag
1456 NOT_LP64(__ get_thread(thread));
1457 __ movbool(do_not_unlock_if_synchronized, false);
1458
1459 // check for synchronized methods
1460 // Must happen AFTER invocation_counter check and stack overflow check,
1461 // so method is not locked if overflows.
1462 if (synchronized) {
1463 // Allocate monitor and lock method
1464 lock_method();
1465 } else {
1466 // no synchronization necessary
1467 #ifdef ASSERT
1468 {
1469 Label L;
1470 __ movl(rax, access_flags);
1471 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1472 __ jcc(Assembler::zero, L);
1473 __ stop("method needs synchronization");
1474 __ bind(L);
1475 }
1476 #endif
1477 }
1478
1479 // start execution
1480 #ifdef ASSERT
1481 {
1482 Label L;
1483 const Address monitor_block_top (rbp,
1484 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1485 __ movptr(rax, monitor_block_top);
1486 __ cmpptr(rax, rsp);
1487 __ jcc(Assembler::equal, L);
1488 __ stop("broken stack frame setup in interpreter");
1489 __ bind(L);
1490 }
1491 #endif
1492
1493 // jvmti support
1494 __ notify_method_entry();
1495
1496 __ dispatch_next(vtos);
1497
1498 // invocation counter overflow
1499 if (inc_counter) {
1500 if (ProfileInterpreter) {
1501 // We have decided to profile this method in the interpreter
1502 __ bind(profile_method);
1503 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1504 __ set_method_data_pointer_for_bcp();
1505 __ get_method(rbx);
1506 __ jmp(profile_method_continue);
1507 }
1508 // Handle overflow of counter and compile method
1509 __ bind(invocation_counter_overflow);
1510 generate_counter_overflow(continue_after_compile);
1511 }
1512
1513 return entry_point;
1514 }
1515
1516 //-----------------------------------------------------------------------------
1517 // Exceptions
1518
1519 void TemplateInterpreterGenerator::generate_throw_exception() {
1520 // Entry point in previous activation (i.e., if the caller was
1521 // interpreted)
1522 Interpreter::_rethrow_exception_entry = __ pc();
1523 // Restore sp to interpreter_frame_last_sp even though we are going
1524 // to empty the expression stack for the exception processing.
1525 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1526 // rax: exception
1527 // rdx: return address/pc that threw exception
1528 __ restore_bcp(); // r13/rsi points to call/send
1529 __ restore_locals();
1530 LP64_ONLY(__ reinit_heapbase()); // restore r12 as heapbase.
1531 // Entry point for exceptions thrown within interpreter code
1532 Interpreter::_throw_exception_entry = __ pc();
1533 // expression stack is undefined here
1534 // rax: exception
1535 // r13/rsi: exception bcp
1536 __ verify_oop(rax);
1537 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1538 LP64_ONLY(__ mov(c_rarg1, rax));
1539
1540 // expression stack must be empty before entering the VM in case of
1541 // an exception
1542 __ empty_expression_stack();
1543 // find exception handler address and preserve exception oop
1544 __ call_VM(rdx,
1545 CAST_FROM_FN_PTR(address,
1546 InterpreterRuntime::exception_handler_for_exception),
1547 rarg);
1548 // rax: exception handler entry point
1549 // rdx: preserved exception oop
1550 // r13/rsi: bcp for exception handler
1551 __ push_ptr(rdx); // push exception which is now the only value on the stack
1552 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1553
1554 // If the exception is not handled in the current frame the frame is
1555 // removed and the exception is rethrown (i.e. exception
1556 // continuation is _rethrow_exception).
1557 //
1558 // Note: At this point the bci is still the bxi for the instruction
1559 // which caused the exception and the expression stack is
1560 // empty. Thus, for any VM calls at this point, GC will find a legal
1561 // oop map (with empty expression stack).
1562
1563 // In current activation
1564 // tos: exception
1565 // esi: exception bcp
1566
1567 //
1568 // JVMTI PopFrame support
1569 //
1570
1571 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1572 __ empty_expression_stack();
1573 // Set the popframe_processing bit in pending_popframe_condition
1574 // indicating that we are currently handling popframe, so that
1575 // call_VMs that may happen later do not trigger new popframe
1576 // handling cycles.
1577 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1578 NOT_LP64(__ get_thread(thread));
1579 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1580 __ orl(rdx, JavaThread::popframe_processing_bit);
1581 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1582
1583 {
1584 // Check to see whether we are returning to a deoptimized frame.
1585 // (The PopFrame call ensures that the caller of the popped frame is
1586 // either interpreted or compiled and deoptimizes it if compiled.)
1587 // In this case, we can't call dispatch_next() after the frame is
1588 // popped, but instead must save the incoming arguments and restore
1589 // them after deoptimization has occurred.
1590 //
1591 // Note that we don't compare the return PC against the
1592 // deoptimization blob's unpack entry because of the presence of
1593 // adapter frames in C2.
1594 Label caller_not_deoptimized;
1595 Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1596 __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1597 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1598 InterpreterRuntime::interpreter_contains), rarg);
1599 __ testl(rax, rax);
1600 __ jcc(Assembler::notZero, caller_not_deoptimized);
1601
1602 // Compute size of arguments for saving when returning to
1603 // deoptimized caller
1604 __ get_method(rax);
1605 __ movptr(rax, Address(rax, Method::const_offset()));
1606 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1607 size_of_parameters_offset())));
1608 __ shll(rax, Interpreter::logStackElementSize);
1609 __ restore_locals();
1610 __ subptr(rlocals, rax);
1611 __ addptr(rlocals, wordSize);
1612 // Save these arguments
1613 NOT_LP64(__ get_thread(thread));
1614 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1615 Deoptimization::
1616 popframe_preserve_args),
1617 thread, rax, rlocals);
1618
1619 __ remove_activation(vtos, rdx,
1620 /* throw_monitor_exception */ false,
1621 /* install_monitor_exception */ false,
1622 /* notify_jvmdi */ false);
1623
1624 // Inform deoptimization that it is responsible for restoring
1625 // these arguments
1626 NOT_LP64(__ get_thread(thread));
1627 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1628 JavaThread::popframe_force_deopt_reexecution_bit);
1629
1630 // Continue in deoptimization handler
1631 __ jmp(rdx);
1632
1633 __ bind(caller_not_deoptimized);
1634 }
1635
1636 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1637 /* throw_monitor_exception */ false,
1638 /* install_monitor_exception */ false,
1639 /* notify_jvmdi */ false);
1640
1641 // Finish with popframe handling
1642 // A previous I2C followed by a deoptimization might have moved the
1643 // outgoing arguments further up the stack. PopFrame expects the
1644 // mutations to those outgoing arguments to be preserved and other
1645 // constraints basically require this frame to look exactly as
1646 // though it had previously invoked an interpreted activation with
1647 // no space between the top of the expression stack (current
1648 // last_sp) and the top of stack. Rather than force deopt to
1649 // maintain this kind of invariant all the time we call a small
1650 // fixup routine to move the mutated arguments onto the top of our
1651 // expression stack if necessary.
1652 #ifndef _LP64
1653 __ mov(rax, rsp);
1654 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1655 __ get_thread(thread);
1656 // PC must point into interpreter here
1657 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1658 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1659 __ get_thread(thread);
1660 #else
1661 __ mov(c_rarg1, rsp);
1662 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1663 // PC must point into interpreter here
1664 __ set_last_Java_frame(noreg, rbp, __ pc());
1665 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1666 #endif
1667 __ reset_last_Java_frame(thread, true);
1668
1669 // Restore the last_sp and null it out
1670 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1671 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1672
1673 __ restore_bcp();
1674 __ restore_locals();
1675 // The method data pointer was incremented already during
1676 // call profiling. We have to restore the mdp for the current bcp.
1677 if (ProfileInterpreter) {
1678 __ set_method_data_pointer_for_bcp();
1679 }
1680
1681 // Clear the popframe condition flag
1682 NOT_LP64(__ get_thread(thread));
1683 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1684 JavaThread::popframe_inactive);
1685
1686 #if INCLUDE_JVMTI
1687 {
1688 Label L_done;
1689 const Register local0 = rlocals;
1690
1691 __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1692 __ jcc(Assembler::notEqual, L_done);
1693
1694 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1695 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1696
1697 __ get_method(rdx);
1698 __ movptr(rax, Address(local0, 0));
1699 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1700
1701 __ testptr(rax, rax);
1702 __ jcc(Assembler::zero, L_done);
1703
1704 __ movptr(Address(rbx, 0), rax);
1705 __ bind(L_done);
1706 }
1707 #endif // INCLUDE_JVMTI
1708
1709 __ dispatch_next(vtos);
1710 // end of PopFrame support
1711
1712 Interpreter::_remove_activation_entry = __ pc();
1713
1714 // preserve exception over this code sequence
1715 __ pop_ptr(rax);
1716 NOT_LP64(__ get_thread(thread));
1717 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1718 // remove the activation (without doing throws on illegalMonitorExceptions)
1719 __ remove_activation(vtos, rdx, false, true, false);
1720 // restore exception
1721 NOT_LP64(__ get_thread(thread));
1722 __ get_vm_result(rax, thread);
1723
1724 // In between activations - previous activation type unknown yet
1725 // compute continuation point - the continuation point expects the
1726 // following registers set up:
1727 //
1728 // rax: exception
1729 // rdx: return address/pc that threw exception
1730 // rsp: expression stack of caller
1731 // rbp: ebp of caller
1732 __ push(rax); // save exception
1733 __ push(rdx); // save return address
1734 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1735 SharedRuntime::exception_handler_for_return_address),
1736 thread, rdx);
1737 __ mov(rbx, rax); // save exception handler
1738 __ pop(rdx); // restore return address
1739 __ pop(rax); // restore exception
1740 // Note that an "issuing PC" is actually the next PC after the call
1741 __ jmp(rbx); // jump to exception
1742 // handler of caller
1743 }
1744
1745
1746 //
1747 // JVMTI ForceEarlyReturn support
1748 //
1749 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1750 address entry = __ pc();
1751
1752 __ restore_bcp();
1753 __ restore_locals();
1754 __ empty_expression_stack();
1755 __ load_earlyret_value(state); // 32 bits returns value in rdx, so don't reuse
1756
1757 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1758 NOT_LP64(__ get_thread(thread));
1759 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1760 Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1761
1762 // Clear the earlyret state
1763 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1764
1765 __ remove_activation(state, rsi,
1766 false, /* throw_monitor_exception */
1767 false, /* install_monitor_exception */
1768 true); /* notify_jvmdi */
1769 __ jmp(rsi);
1770
1771 return entry;
1772 } // end of ForceEarlyReturn support
1773
1774
1775 //-----------------------------------------------------------------------------
1776 // Helper for vtos entry point generation
1777
1778 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1779 address& bep,
1780 address& cep,
1781 address& sep,
1782 address& aep,
1783 address& iep,
1784 address& lep,
1785 address& fep,
1786 address& dep,
1787 address& vep) {
1788 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1789 Label L;
1790 aep = __ pc(); __ push_ptr(); __ jmp(L);
1791 #ifndef _LP64
1792 fep = __ pc(); __ push(ftos); __ jmp(L);
1793 dep = __ pc(); __ push(dtos); __ jmp(L);
1794 #else
1795 fep = __ pc(); __ push_f(xmm0); __ jmp(L);
1796 dep = __ pc(); __ push_d(xmm0); __ jmp(L);
1797 #endif // _LP64
1798 lep = __ pc(); __ push_l(); __ jmp(L);
1799 bep = cep = sep =
1800 iep = __ pc(); __ push_i();
1801 vep = __ pc();
1802 __ bind(L);
1803 generate_and_dispatch(t);
1804 }
1805
1806 //-----------------------------------------------------------------------------
1807
1808 // Non-product code
1809 #ifndef PRODUCT
1810
1811 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1812 address entry = __ pc();
1813
1814 #ifndef _LP64
1815 // prepare expression stack
1816 __ pop(rcx); // pop return address so expression stack is 'pure'
1817 __ push(state); // save tosca
1818
1819 // pass tosca registers as arguments & call tracer
1820 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1821 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1822 __ pop(state); // restore tosca
1823
1824 // return
1825 __ jmp(rcx);
1826 #else
1827 __ push(state);
1828 __ push(c_rarg0);
1829 __ push(c_rarg1);
1830 __ push(c_rarg2);
1831 __ push(c_rarg3);
1832 __ mov(c_rarg2, rax); // Pass itos
1833 #ifdef _WIN64
1834 __ movflt(xmm3, xmm0); // Pass ftos
1835 #endif
1836 __ call_VM(noreg,
1837 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1838 c_rarg1, c_rarg2, c_rarg3);
1839 __ pop(c_rarg3);
1840 __ pop(c_rarg2);
1841 __ pop(c_rarg1);
1842 __ pop(c_rarg0);
1843 __ pop(state);
1844 __ ret(0); // return from result handler
1845 #endif // _LP64
1846
1847 return entry;
1848 }
1849
1850 void TemplateInterpreterGenerator::count_bytecode() {
1851 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1852 }
1853
1854 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1855 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1856 }
1857
1858 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1859 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1860 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1861 __ orl(rbx,
1862 ((int) t->bytecode()) <<
1863 BytecodePairHistogram::log2_number_of_codes);
1864 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1865 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1866 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1867 }
1868
1869
1870 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1871 // Call a little run-time stub to avoid blow-up for each bytecode.
1872 // The run-time runtime saves the right registers, depending on
1873 // the tosca in-state for the given template.
1874
1875 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1876 "entry must have been generated");
1877 #ifndef _LP64
1878 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1879 #else
1880 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1881 __ andptr(rsp, -16); // align stack as required by ABI
1882 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1883 __ mov(rsp, r12); // restore sp
1884 __ reinit_heapbase();
1885 #endif // _LP64
1886 }
1887
1888
1889 void TemplateInterpreterGenerator::stop_interpreter_at() {
1890 Label L;
1891 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1892 StopInterpreterAt);
1893 __ jcc(Assembler::notEqual, L);
1894 __ int3();
1895 __ bind(L);
1896 }
1897 #endif // !PRODUCT