1 /*
2 * Copyright (c) 2003, 2025, 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 "asm/macroAssembler.hpp"
26 #include "classfile/javaClasses.hpp"
27 #include "compiler/compiler_globals.hpp"
28 #include "compiler/disassembler.hpp"
29 #include "gc/shared/barrierSetAssembler.hpp"
30 #include "interpreter/bytecodeHistogram.hpp"
31 #include "interpreter/interp_masm.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "interpreter/interpreterRuntime.hpp"
34 #include "interpreter/templateInterpreterGenerator.hpp"
35 #include "interpreter/templateTable.hpp"
36 #include "oops/arrayOop.hpp"
37 #include "oops/methodCounters.hpp"
38 #include "oops/methodData.hpp"
39 #include "oops/method.hpp"
40 #include "oops/oop.inline.hpp"
41 #include "oops/resolvedIndyEntry.hpp"
42 #include "oops/resolvedMethodEntry.hpp"
43 #include "prims/jvmtiExport.hpp"
44 #include "prims/jvmtiThreadState.hpp"
45 #include "runtime/continuation.hpp"
46 #include "runtime/deoptimization.hpp"
47 #include "runtime/frame.inline.hpp"
48 #include "runtime/globals.hpp"
49 #include "runtime/jniHandles.hpp"
50 #include "runtime/sharedRuntime.hpp"
51 #include "runtime/stubRoutines.hpp"
52 #include "runtime/synchronizer.hpp"
53 #include "runtime/timer.hpp"
54 #include "runtime/vframeArray.hpp"
55 #include "utilities/checkedCast.hpp"
56 #include "utilities/debug.hpp"
57 #include "utilities/macros.hpp"
58
59 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
60
61 // Size of interpreter code. Increase if too small. Interpreter will
62 // fail with a guarantee ("not enough space for interpreter generation");
63 // if too small.
64 // Run with +PrintInterpreter to get the VM to print out the size.
65 // Max size with JVMTI
66 #ifdef AMD64
67 int TemplateInterpreter::InterpreterCodeSize = JVMCI_ONLY(268) NOT_JVMCI(256) * 1024;
68 #else
69 int TemplateInterpreter::InterpreterCodeSize = 224 * 1024;
70 #endif // AMD64
71
72 // Global Register Names
73 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
74 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
75
76 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
77 const int bcp_offset = frame::interpreter_frame_bcp_offset * wordSize;
78 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
79
80
81 //-----------------------------------------------------------------------------
82
83 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
84 address entry = __ pc();
85
86 #ifdef ASSERT
87 {
88 Label L;
89 __ movptr(rax, Address(rbp,
90 frame::interpreter_frame_monitor_block_top_offset *
91 wordSize));
92 __ lea(rax, Address(rbp, rax, Address::times_ptr));
93 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
94 // grows negative)
95 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
96 __ stop ("interpreter frame not set up");
97 __ bind(L);
98 }
99 #endif // ASSERT
100 // Restore bcp under the assumption that the current frame is still
101 // interpreted
102 __ restore_bcp();
103
104 // expression stack must be empty before entering the VM if an
105 // exception happened
106 __ empty_expression_stack();
107 // throw exception
108 __ call_VM(noreg,
109 CAST_FROM_FN_PTR(address,
110 InterpreterRuntime::throw_StackOverflowError));
111 return entry;
112 }
113
114 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
115 address entry = __ pc();
116 // The expression stack must be empty before entering the VM if an
117 // exception happened.
118 __ empty_expression_stack();
119
120 // Setup parameters.
121 // ??? convention: expect aberrant index in register ebx/rbx.
122 // Pass array to create more detailed exceptions.
123 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
124 __ call_VM(noreg,
125 CAST_FROM_FN_PTR(address,
126 InterpreterRuntime::
127 throw_ArrayIndexOutOfBoundsException),
128 rarg, rbx);
129 return entry;
130 }
131
132 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
133 address entry = __ pc();
134
135 // object is at TOS
136 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
137 __ pop(rarg);
138
139 // expression stack must be empty before entering the VM if an
140 // exception happened
141 __ empty_expression_stack();
142
143 __ call_VM(noreg,
144 CAST_FROM_FN_PTR(address,
145 InterpreterRuntime::
146 throw_ClassCastException),
147 rarg);
148 return entry;
149 }
150
151 address TemplateInterpreterGenerator::generate_exception_handler_common(
152 const char* name, const char* message, bool pass_oop) {
153 assert(!pass_oop || message == nullptr, "either oop or message but not both");
154 address entry = __ pc();
155
156 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
157 Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
158
159 if (pass_oop) {
160 // object is at TOS
161 __ pop(rarg2);
162 }
163 // expression stack must be empty before entering the VM if an
164 // exception happened
165 __ empty_expression_stack();
166 // setup parameters
167 __ lea(rarg, ExternalAddress((address)name));
168 if (pass_oop) {
169 __ call_VM(rax, CAST_FROM_FN_PTR(address,
170 InterpreterRuntime::
171 create_klass_exception),
172 rarg, rarg2);
173 } else {
174 __ lea(rarg2, ExternalAddress((address)message));
175 __ call_VM(rax,
176 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
177 rarg, rarg2);
178 }
179 // throw exception
180 __ jump(RuntimeAddress(Interpreter::throw_exception_entry()));
181 return entry;
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(rcx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
213 __ lea(rsp, Address(rbp, rcx, Address::times_ptr));
214 // and null it as marker that esp is now tos until next java call
215 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
216
217 __ restore_bcp();
218 __ restore_locals();
219
220 if (state == atos) {
221 Register mdp = rbx;
222 Register tmp = rcx;
223 __ profile_return_type(mdp, rax, tmp);
224 }
225
226 const Register cache = rbx;
227 const Register index = rcx;
228 if (index_size == sizeof(u4)) {
229 __ load_resolved_indy_entry(cache, index);
230 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
231 __ lea(rsp, Address(rsp, cache, Interpreter::stackElementScale()));
232 } else {
233 assert(index_size == sizeof(u2), "Can only be u2");
234 __ load_method_entry(cache, index);
235 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
236 __ lea(rsp, Address(rsp, cache, Interpreter::stackElementScale()));
237 }
238
239 const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
240 if (JvmtiExport::can_pop_frame()) {
241 NOT_LP64(__ get_thread(java_thread));
242 __ check_and_handle_popframe(java_thread);
243 }
244 if (JvmtiExport::can_force_early_return()) {
245 NOT_LP64(__ get_thread(java_thread));
246 __ check_and_handle_earlyret(java_thread);
247 }
248
249 __ dispatch_next(state, step);
250
251 return entry;
252 }
253
254
255 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
256 address entry = __ pc();
257
258 #ifndef _LP64
259 if (state == ftos) {
260 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
261 } else if (state == dtos) {
262 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
263 }
264 #endif // _LP64
265
266 // null last_sp until next java call
267 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
268 __ restore_bcp();
269 __ restore_locals();
270 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
271 NOT_LP64(__ get_thread(thread));
272 #if INCLUDE_JVMCI
273 // Check if we need to take lock at entry of synchronized method. This can
274 // only occur on method entry so emit it only for vtos with step 0.
275 if (EnableJVMCI && state == vtos && step == 0) {
276 Label L;
277 __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
278 __ jcc(Assembler::zero, L);
279 // Clear flag.
280 __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
281 // Satisfy calling convention for lock_method().
282 __ get_method(rbx);
283 // Take lock.
284 lock_method();
285 __ bind(L);
286 } else {
287 #ifdef ASSERT
288 if (EnableJVMCI) {
289 Label L;
290 __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
291 __ jcc(Assembler::zero, L);
292 __ stop("unexpected pending monitor in deopt entry");
293 __ bind(L);
294 }
295 #endif
296 }
297 #endif
298 // handle exceptions
299 {
300 Label L;
301 __ cmpptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
302 __ jcc(Assembler::zero, L);
303 __ call_VM(noreg,
304 CAST_FROM_FN_PTR(address,
305 InterpreterRuntime::throw_pending_exception));
306 __ should_not_reach_here();
307 __ bind(L);
308 }
309 if (continuation == nullptr) {
310 __ dispatch_next(state, step);
311 } else {
312 __ jump_to_entry(continuation);
313 }
314 return entry;
315 }
316
317 address TemplateInterpreterGenerator::generate_result_handler_for(
318 BasicType type) {
319 address entry = __ pc();
320 switch (type) {
321 case T_BOOLEAN: __ c2bool(rax); break;
322 #ifndef _LP64
323 case T_CHAR : __ andptr(rax, 0xFFFF); break;
324 #else
325 case T_CHAR : __ movzwl(rax, rax); break;
326 #endif // _LP64
327 case T_BYTE : __ sign_extend_byte(rax); break;
328 case T_SHORT : __ sign_extend_short(rax); break;
329 case T_INT : /* nothing to do */ break;
330 case T_LONG : /* nothing to do */ break;
331 case T_VOID : /* nothing to do */ break;
332 #ifndef _LP64
333 case T_DOUBLE :
334 case T_FLOAT :
335 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
336 __ pop(t); // remove return address first
337 // Must return a result for interpreter or compiler. In SSE
338 // mode, results are returned in xmm0 and the FPU stack must
339 // be empty.
340 if (type == T_FLOAT && UseSSE >= 1) {
341 // Load ST0
342 __ fld_d(Address(rsp, 0));
343 // Store as float and empty fpu stack
344 __ fstp_s(Address(rsp, 0));
345 // and reload
346 __ movflt(xmm0, Address(rsp, 0));
347 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
348 __ movdbl(xmm0, Address(rsp, 0));
349 } else {
350 // restore ST0
351 __ fld_d(Address(rsp, 0));
352 }
353 // and pop the temp
354 __ addptr(rsp, 2 * wordSize);
355 __ push(t); // restore return address
356 }
357 break;
358 #else
359 case T_FLOAT : /* nothing to do */ break;
360 case T_DOUBLE : /* nothing to do */ break;
361 #endif // _LP64
362
363 case T_OBJECT :
364 // retrieve result from frame
365 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
366 // and verify it
367 __ verify_oop(rax);
368 break;
369 default : ShouldNotReachHere();
370 }
371 __ ret(0); // return from result handler
372 return entry;
373 }
374
375 address TemplateInterpreterGenerator::generate_safept_entry_for(
376 TosState state,
377 address runtime_entry) {
378 address entry = __ pc();
379
380 __ push(state);
381 __ push_cont_fastpath();
382 __ call_VM(noreg, runtime_entry);
383 __ pop_cont_fastpath();
384
385 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
386 return entry;
387 }
388
389 address TemplateInterpreterGenerator::generate_cont_resume_interpreter_adapter() {
390 if (!Continuations::enabled()) return nullptr;
391 address start = __ pc();
392
393 __ restore_bcp();
394 __ restore_locals();
395
396 // Get return address before adjusting rsp
397 __ movptr(rax, Address(rsp, 0));
398
399 // Restore stack bottom
400 __ movptr(rcx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
401 __ lea(rsp, Address(rbp, rcx, Address::times_ptr));
402 // and null it as marker that esp is now tos until next java call
403 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
404
405 __ jmp(rax);
406
407 return start;
408 }
409
410
411 // Helpers for commoning out cases in the various type of method entries.
412 //
413
414
415 // increment invocation count & check for overflow
416 //
417 // Note: checking for negative value instead of overflow
418 // so we have a 'sticky' overflow test
419 //
420 // rbx: method
421 // rcx: invocation counter
422 //
423 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
424 Label done;
425 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
426 Label no_mdo;
427 if (ProfileInterpreter) {
428 // Are we profiling?
429 __ movptr(rax, Address(rbx, Method::method_data_offset()));
430 __ testptr(rax, rax);
431 __ jccb(Assembler::zero, no_mdo);
432 // Increment counter in the MDO
433 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
434 in_bytes(InvocationCounter::counter_offset()));
435 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
436 __ increment_mask_and_jump(mdo_invocation_counter, mask, rcx, overflow);
437 __ jmp(done);
438 }
439 __ bind(no_mdo);
440 // Increment counter in MethodCounters
441 const Address invocation_counter(rax,
442 MethodCounters::invocation_counter_offset() +
443 InvocationCounter::counter_offset());
444 __ get_method_counters(rbx, rax, done);
445 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
446 __ increment_mask_and_jump(invocation_counter, mask, rcx, overflow);
447 __ bind(done);
448 }
449
450 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
451
452 // Asm interpreter on entry
453 // r14/rdi - locals
454 // r13/rsi - bcp
455 // rbx - method
456 // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
457 // rbp - interpreter frame
458
459 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
460 // Everything as it was on entry
461 // rdx is not restored. Doesn't appear to really be set.
462
463 // InterpreterRuntime::frequency_counter_overflow takes two
464 // arguments, the first (thread) is passed by call_VM, the second
465 // indicates if the counter overflow occurs at a backwards branch
466 // (null bcp). We pass zero for it. The call returns the address
467 // of the verified entry point for the method or null if the
468 // compilation did not complete (either went background or bailed
469 // out).
470 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
471 __ movl(rarg, 0);
472 __ call_VM(noreg,
473 CAST_FROM_FN_PTR(address,
474 InterpreterRuntime::frequency_counter_overflow),
475 rarg);
476
477 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
478 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
479 // and jump to the interpreted entry.
480 __ jmp(do_continue, relocInfo::none);
481 }
482
483 // See if we've got enough room on the stack for locals plus overhead below
484 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
485 // without going through the signal handler, i.e., reserved and yellow zones
486 // will not be made usable. The shadow zone must suffice to handle the
487 // overflow.
488 // The expression stack grows down incrementally, so the normal guard
489 // page mechanism will work for that.
490 //
491 // NOTE: Since the additional locals are also always pushed (wasn't
492 // obvious in generate_fixed_frame) so the guard should work for them
493 // too.
494 //
495 // Args:
496 // rdx: number of additional locals this frame needs (what we must check)
497 // rbx: Method*
498 //
499 // Kills:
500 // rax
501 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
502
503 // monitor entry size: see picture of stack in frame_x86.hpp
504 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
505
506 // total overhead size: entry_size + (saved rbp through expr stack
507 // bottom). be sure to change this if you add/subtract anything
508 // to/from the overhead area
509 const int overhead_size =
510 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
511
512 const int page_size = (int)os::vm_page_size();
513
514 Label after_frame_check;
515
516 // see if the frame is greater than one page in size. If so,
517 // then we need to verify there is enough stack space remaining
518 // for the additional locals.
519 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
520 __ jcc(Assembler::belowEqual, after_frame_check);
521
522 // compute rsp as if this were going to be the last frame on
523 // the stack before the red zone
524
525 Label after_frame_check_pop;
526 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
527 #ifndef _LP64
528 __ push(thread);
529 __ get_thread(thread);
530 #endif
531
532 const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
533
534 // locals + overhead, in bytes
535 __ mov(rax, rdx);
536 __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
537 __ addptr(rax, overhead_size);
538
539 #ifdef ASSERT
540 Label limit_okay;
541 // Verify that thread stack overflow limit is non-zero.
542 __ cmpptr(stack_limit, NULL_WORD);
543 __ jcc(Assembler::notEqual, limit_okay);
544 __ stop("stack overflow limit is zero");
545 __ bind(limit_okay);
546 #endif
547
548 // Add locals/frame size to stack limit.
549 __ addptr(rax, stack_limit);
550
551 // Check against the current stack bottom.
552 __ cmpptr(rsp, rax);
553
554 __ jcc(Assembler::above, after_frame_check_pop);
555 NOT_LP64(__ pop(rsi)); // get saved bcp
556
557 // Restore sender's sp as SP. This is necessary if the sender's
558 // frame is an extended compiled frame (see gen_c2i_adapter())
559 // and safer anyway in case of JSR292 adaptations.
560
561 __ pop(rax); // return address must be moved if SP is changed
562 __ mov(rsp, rbcp);
563 __ push(rax);
564
565 // Note: the restored frame is not necessarily interpreted.
566 // Use the shared runtime version of the StackOverflowError.
567 assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
568 __ jump(RuntimeAddress(SharedRuntime::throw_StackOverflowError_entry()));
569 // all done with frame size check
570 __ bind(after_frame_check_pop);
571 NOT_LP64(__ pop(rsi));
572
573 // all done with frame size check
574 __ bind(after_frame_check);
575 }
576
577 // Allocate monitor and lock method (asm interpreter)
578 //
579 // Args:
580 // rbx: Method*
581 // r14/rdi: locals
582 //
583 // Kills:
584 // rax
585 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
586 // rscratch1, rscratch2 (scratch regs)
587 void TemplateInterpreterGenerator::lock_method() {
588 // synchronize method
589 const Address access_flags(rbx, Method::access_flags_offset());
590 const Address monitor_block_top(
591 rbp,
592 frame::interpreter_frame_monitor_block_top_offset * wordSize);
593 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
594
595 #ifdef ASSERT
596 {
597 Label L;
598 __ load_unsigned_short(rax, access_flags);
599 __ testl(rax, JVM_ACC_SYNCHRONIZED);
600 __ jcc(Assembler::notZero, L);
601 __ stop("method doesn't need synchronization");
602 __ bind(L);
603 }
604 #endif // ASSERT
605
606 // get synchronization object
607 {
608 Label done;
609 __ load_unsigned_short(rax, access_flags);
610 __ testl(rax, JVM_ACC_STATIC);
611 // get receiver (assume this is frequent case)
612 __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
613 __ jcc(Assembler::zero, done);
614 __ load_mirror(rax, rbx, rscratch2);
615
616 #ifdef ASSERT
617 {
618 Label L;
619 __ testptr(rax, rax);
620 __ jcc(Assembler::notZero, L);
621 __ stop("synchronization object is null");
622 __ bind(L);
623 }
624 #endif // ASSERT
625
626 __ bind(done);
627 }
628
629 // add space for monitor & lock
630 __ subptr(rsp, entry_size); // add space for a monitor entry
631 __ subptr(monitor_block_top, entry_size / wordSize); // set new monitor block top
632 // store object
633 __ movptr(Address(rsp, BasicObjectLock::obj_offset()), rax);
634 const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
635 __ movptr(lockreg, rsp); // object address
636 __ lock_object(lockreg);
637 }
638
639 // Generate a fixed interpreter frame. This is identical setup for
640 // interpreted methods and for native methods hence the shared code.
641 //
642 // Args:
643 // rax: return address
644 // rbx: Method*
645 // r14/rdi: pointer to locals
646 // r13/rsi: sender sp
647 // rdx: cp cache
648 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
649 // initialize fixed part of activation frame
650 __ push(rax); // save return address
651 __ enter(); // save old & set new rbp
652 __ push(rbcp); // set sender sp
653 __ push(NULL_WORD); // leave last_sp as null
654 __ movptr(rbcp, Address(rbx, Method::const_offset())); // get ConstMethod*
655 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
656 __ push(rbx); // save Method*
657 // Get mirror and store it in the frame as GC root for this Method*
658 __ load_mirror(rdx, rbx, rscratch2);
659 __ push(rdx);
660 if (ProfileInterpreter) {
661 Label method_data_continue;
662 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
663 __ testptr(rdx, rdx);
664 __ jcc(Assembler::zero, method_data_continue);
665 __ addptr(rdx, in_bytes(MethodData::data_offset()));
666 __ bind(method_data_continue);
667 __ push(rdx); // set the mdp (method data pointer)
668 } else {
669 __ push(0);
670 }
671
672 __ movptr(rdx, Address(rbx, Method::const_offset()));
673 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
674 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset()));
675 __ push(rdx); // set constant pool cache
676
677 __ movptr(rax, rlocals);
678 __ subptr(rax, rbp);
679 __ shrptr(rax, Interpreter::logStackElementSize); // rax = rlocals - fp();
680 __ push(rax); // set relativized rlocals, see frame::interpreter_frame_locals()
681
682 if (native_call) {
683 __ push(0); // no bcp
684 } else {
685 __ push(rbcp); // set bcp
686 }
687 // initialize relativized pointer to expression stack bottom
688 __ push(frame::interpreter_frame_initial_sp_offset);
689 }
690
691 // End of helpers
692
693 // Method entry for java.lang.ref.Reference.get.
694 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
695 // Code: _aload_0, _getfield, _areturn
696 // parameter size = 1
697 //
698 // The code that gets generated by this routine is split into 2 parts:
699 // 1. The "intrinsified" code performing an ON_WEAK_OOP_REF load,
700 // 2. The slow path - which is an expansion of the regular method entry.
701 //
702 // Notes:-
703 // * An intrinsic is always executed, where an ON_WEAK_OOP_REF load is performed.
704 // * We may jump to the slow path iff the receiver is null. If the
705 // Reference object is null then we no longer perform an ON_WEAK_OOP_REF load
706 // Thus we can use the regular method entry code to generate the NPE.
707 //
708 // rbx: Method*
709
710 // r13: senderSP must preserve for slow path, set SP to it on fast path
711
712 address entry = __ pc();
713
714 const int referent_offset = java_lang_ref_Reference::referent_offset();
715
716 Label slow_path;
717 // rbx: method
718
719 // Check if local 0 != null
720 // If the receiver is null then it is OK to jump to the slow path.
721 __ movptr(rax, Address(rsp, wordSize));
722
723 __ testptr(rax, rax);
724 __ jcc(Assembler::zero, slow_path);
725
726 // rax: local 0
727 // rbx: method (but can be used as scratch now)
728 // rdx: scratch
729 // rdi: scratch
730
731 // Preserve the sender sp in case the load barrier
732 // calls the runtime
733 NOT_LP64(__ push(rsi));
734
735 // Load the value of the referent field.
736 const Address field_address(rax, referent_offset);
737 __ load_heap_oop(rax, field_address, /*tmp1*/ rbx, /*tmp_thread*/ rdx, ON_WEAK_OOP_REF);
738
739 // _areturn
740 const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
741 NOT_LP64(__ pop(rsi)); // get sender sp
742 __ pop(rdi); // get return address
743 __ mov(rsp, sender_sp); // set sp to sender sp
744 __ jmp(rdi);
745 __ ret(0);
746
747 // generate a vanilla interpreter entry as the slow path
748 __ bind(slow_path);
749 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
750 return entry;
751 }
752
753 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
754 // See more discussion in stackOverflow.hpp.
755
756 // Note that we do the banging after the frame is setup, since the exception
757 // handling code expects to find a valid interpreter frame on the stack.
758 // Doing the banging earlier fails if the caller frame is not an interpreter
759 // frame.
760 // (Also, the exception throwing code expects to unlock any synchronized
761 // method receiver, so do the banging after locking the receiver.)
762
763 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
764 const int page_size = (int)os::vm_page_size();
765 const int n_shadow_pages = shadow_zone_size / page_size;
766
767 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
768 #ifndef _LP64
769 __ push(thread);
770 __ get_thread(thread);
771 #endif
772
773 #ifdef ASSERT
774 Label L_good_limit;
775 __ cmpptr(Address(thread, JavaThread::shadow_zone_safe_limit()), NULL_WORD);
776 __ jcc(Assembler::notEqual, L_good_limit);
777 __ stop("shadow zone safe limit is not initialized");
778 __ bind(L_good_limit);
779
780 Label L_good_watermark;
781 __ cmpptr(Address(thread, JavaThread::shadow_zone_growth_watermark()), NULL_WORD);
782 __ jcc(Assembler::notEqual, L_good_watermark);
783 __ stop("shadow zone growth watermark is not initialized");
784 __ bind(L_good_watermark);
785 #endif
786
787 Label L_done;
788
789 __ cmpptr(rsp, Address(thread, JavaThread::shadow_zone_growth_watermark()));
790 __ jcc(Assembler::above, L_done);
791
792 for (int p = 1; p <= n_shadow_pages; p++) {
793 __ bang_stack_with_offset(p*page_size);
794 }
795
796 // Record the new watermark, but only if update is above the safe limit.
797 // Otherwise, the next time around the check above would pass the safe limit.
798 __ cmpptr(rsp, Address(thread, JavaThread::shadow_zone_safe_limit()));
799 __ jccb(Assembler::belowEqual, L_done);
800 __ movptr(Address(thread, JavaThread::shadow_zone_growth_watermark()), rsp);
801
802 __ bind(L_done);
803
804 #ifndef _LP64
805 __ pop(thread);
806 #endif
807 }
808
809 // Interpreter stub for calling a native method. (asm interpreter)
810 // This sets up a somewhat different looking stack for calling the
811 // native method than the typical interpreter frame setup.
812 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
813 // determine code generation flags
814 bool inc_counter = UseCompiler || CountCompiledCalls;
815
816 // rbx: Method*
817 // rbcp: sender sp
818
819 address entry_point = __ pc();
820
821 const Address constMethod (rbx, Method::const_offset());
822 const Address access_flags (rbx, Method::access_flags_offset());
823 const Address size_of_parameters(rcx, ConstMethod::
824 size_of_parameters_offset());
825
826
827 // get parameter size (always needed)
828 __ movptr(rcx, constMethod);
829 __ load_unsigned_short(rcx, size_of_parameters);
830
831 // native calls don't need the stack size check since they have no
832 // expression stack and the arguments are already on the stack and
833 // we only add a handful of words to the stack
834
835 // rbx: Method*
836 // rcx: size of parameters
837 // rbcp: sender sp
838 __ pop(rax); // get return address
839
840 // for natives the size of locals is zero
841
842 // compute beginning of parameters
843 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
844
845 // add 2 zero-initialized slots for native calls
846 // initialize result_handler slot
847 __ push(NULL_WORD);
848 // slot for oop temp
849 // (static native method holder mirror/jni oop result)
850 __ push(NULL_WORD);
851
852 // initialize fixed part of activation frame
853 generate_fixed_frame(true);
854
855 // make sure method is native & not abstract
856 #ifdef ASSERT
857 __ load_unsigned_short(rax, access_flags);
858 {
859 Label L;
860 __ testl(rax, JVM_ACC_NATIVE);
861 __ jcc(Assembler::notZero, L);
862 __ stop("tried to execute non-native method as native");
863 __ bind(L);
864 }
865 {
866 Label L;
867 __ testl(rax, JVM_ACC_ABSTRACT);
868 __ jcc(Assembler::zero, L);
869 __ stop("tried to execute abstract method in interpreter");
870 __ bind(L);
871 }
872 #endif
873
874 // Since at this point in the method invocation the exception handler
875 // would try to exit the monitor of synchronized methods which hasn't
876 // been entered yet, we set the thread local variable
877 // _do_not_unlock_if_synchronized to true. The remove_activation will
878 // check this flag.
879
880 const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
881 NOT_LP64(__ get_thread(thread1));
882 const Address do_not_unlock_if_synchronized(thread1,
883 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
884 __ movbool(do_not_unlock_if_synchronized, true);
885
886 // increment invocation count & check for overflow
887 Label invocation_counter_overflow;
888 if (inc_counter) {
889 generate_counter_incr(&invocation_counter_overflow);
890 }
891
892 Label continue_after_compile;
893 __ bind(continue_after_compile);
894
895 bang_stack_shadow_pages(true);
896
897 // reset the _do_not_unlock_if_synchronized flag
898 NOT_LP64(__ get_thread(thread1));
899 __ movbool(do_not_unlock_if_synchronized, false);
900
901 // check for synchronized methods
902 // Must happen AFTER invocation_counter check and stack overflow check,
903 // so method is not locked if overflows.
904 if (synchronized) {
905 lock_method();
906 } else {
907 // no synchronization necessary
908 #ifdef ASSERT
909 {
910 Label L;
911 __ load_unsigned_short(rax, access_flags);
912 __ testl(rax, JVM_ACC_SYNCHRONIZED);
913 __ jcc(Assembler::zero, L);
914 __ stop("method needs synchronization");
915 __ bind(L);
916 }
917 #endif
918 }
919
920 // start execution
921 #ifdef ASSERT
922 {
923 Label L;
924 const Address monitor_block_top(rbp,
925 frame::interpreter_frame_monitor_block_top_offset * wordSize);
926 __ movptr(rax, monitor_block_top);
927 __ lea(rax, Address(rbp, rax, Address::times_ptr));
928 __ cmpptr(rax, rsp);
929 __ jcc(Assembler::equal, L);
930 __ stop("broken stack frame setup in interpreter 5");
931 __ bind(L);
932 }
933 #endif
934
935 // jvmti support
936 __ notify_method_entry();
937
938 // work registers
939 const Register method = rbx;
940 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
941 const Register t = NOT_LP64(rcx) LP64_ONLY(r11);
942
943 // allocate space for parameters
944 __ get_method(method);
945 __ movptr(t, Address(method, Method::const_offset()));
946 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
947
948 #ifndef _LP64
949 __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
950 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
951 __ subptr(rsp, t);
952 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
953 #else
954 __ shll(t, Interpreter::logStackElementSize);
955
956 __ subptr(rsp, t);
957 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
958 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
959 #endif // _LP64
960
961 // get signature handler
962 {
963 Label L;
964 __ movptr(t, Address(method, Method::signature_handler_offset()));
965 __ testptr(t, t);
966 __ jcc(Assembler::notZero, L);
967 __ call_VM(noreg,
968 CAST_FROM_FN_PTR(address,
969 InterpreterRuntime::prepare_native_call),
970 method);
971 __ get_method(method);
972 __ movptr(t, Address(method, Method::signature_handler_offset()));
973 __ bind(L);
974 }
975
976 // call signature handler
977 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
978 "adjust this code");
979 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
980 "adjust this code");
981 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
982 "adjust this code");
983
984 // The generated handlers do not touch RBX (the method).
985 // However, large signatures cannot be cached and are generated
986 // each time here. The slow-path generator can do a GC on return,
987 // so we must reload it after the call.
988 __ call(t);
989 __ get_method(method); // slow path can do a GC, reload RBX
990
991
992 // result handler is in rax
993 // set result handler
994 __ movptr(Address(rbp,
995 (frame::interpreter_frame_result_handler_offset) * wordSize),
996 rax);
997
998 // pass mirror handle if static call
999 {
1000 Label L;
1001 __ load_unsigned_short(t, Address(method, Method::access_flags_offset()));
1002 __ testl(t, JVM_ACC_STATIC);
1003 __ jcc(Assembler::zero, L);
1004 // get mirror
1005 __ load_mirror(t, method, rax);
1006 // copy mirror into activation frame
1007 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1008 t);
1009 // pass handle to mirror
1010 #ifndef _LP64
1011 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1012 __ movptr(Address(rsp, wordSize), t);
1013 #else
1014 __ lea(c_rarg1,
1015 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1016 #endif // _LP64
1017 __ bind(L);
1018 }
1019
1020 // get native function entry point
1021 {
1022 Label L;
1023 __ movptr(rax, Address(method, Method::native_function_offset()));
1024 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1025 __ cmpptr(rax, unsatisfied.addr(), rscratch1);
1026 __ jcc(Assembler::notEqual, L);
1027 __ call_VM(noreg,
1028 CAST_FROM_FN_PTR(address,
1029 InterpreterRuntime::prepare_native_call),
1030 method);
1031 __ get_method(method);
1032 __ movptr(rax, Address(method, Method::native_function_offset()));
1033 __ bind(L);
1034 }
1035
1036 // pass JNIEnv
1037 #ifndef _LP64
1038 __ get_thread(thread);
1039 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1040 __ movptr(Address(rsp, 0), t);
1041
1042 // set_last_Java_frame_before_call
1043 // It is enough that the pc()
1044 // points into the right code segment. It does not have to be the correct return pc.
1045 __ set_last_Java_frame(thread, noreg, rbp, __ pc(), noreg);
1046 #else
1047 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1048
1049 // It is enough that the pc() points into the right code
1050 // segment. It does not have to be the correct return pc.
1051 // For convenience we use the pc we want to resume to in
1052 // case of preemption on Object.wait.
1053 Label native_return;
1054 __ set_last_Java_frame(rsp, rbp, native_return, rscratch1);
1055 #endif // _LP64
1056
1057 // change thread state
1058 #ifdef ASSERT
1059 {
1060 Label L;
1061 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1062 __ cmpl(t, _thread_in_Java);
1063 __ jcc(Assembler::equal, L);
1064 __ stop("Wrong thread state in native stub");
1065 __ bind(L);
1066 }
1067 #endif
1068
1069 // Change state to native
1070
1071 __ movl(Address(thread, JavaThread::thread_state_offset()),
1072 _thread_in_native);
1073
1074 __ push_cont_fastpath();
1075
1076 // Call the native method.
1077 __ call(rax);
1078 // 32: result potentially in rdx:rax or ST0
1079 // 64: result potentially in rax or xmm0
1080
1081 __ pop_cont_fastpath();
1082
1083 // Verify or restore cpu control state after JNI call
1084 __ restore_cpu_control_state_after_jni(rscratch1);
1085
1086 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1087 // in order to extract the result of a method call. If the order of these
1088 // pushes change or anything else is added to the stack then the code in
1089 // interpreter_frame_result must also change.
1090
1091 #ifndef _LP64
1092 // save potential result in ST(0) & rdx:rax
1093 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1094 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1095 // It is safe to do this push because state is _thread_in_native and return address will be found
1096 // via _last_native_pc and not via _last_jave_sp
1097
1098 // NOTE: the order of these push(es) is known to frame::interpreter_frame_result.
1099 // If the order changes or anything else is added to the stack the code in
1100 // interpreter_frame_result will have to be changed.
1101
1102 { Label L;
1103 Label push_double;
1104 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1105 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1106 __ cmpptr(Address(rbp, (frame::interpreter_frame_result_handler_offset)*wordSize),
1107 float_handler.addr(), noreg);
1108 __ jcc(Assembler::equal, push_double);
1109 __ cmpptr(Address(rbp, (frame::interpreter_frame_result_handler_offset)*wordSize),
1110 double_handler.addr(), noreg);
1111 __ jcc(Assembler::notEqual, L);
1112 __ bind(push_double);
1113 __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1114 __ bind(L);
1115 }
1116 #else
1117 __ push(dtos);
1118 #endif // _LP64
1119
1120 __ push(ltos);
1121
1122 // change thread state
1123 NOT_LP64(__ get_thread(thread));
1124 __ movl(Address(thread, JavaThread::thread_state_offset()),
1125 _thread_in_native_trans);
1126
1127 // Force this write out before the read below
1128 if (!UseSystemMemoryBarrier) {
1129 __ membar(Assembler::Membar_mask_bits(
1130 Assembler::LoadLoad | Assembler::LoadStore |
1131 Assembler::StoreLoad | Assembler::StoreStore));
1132 }
1133 #ifndef _LP64
1134 if (AlwaysRestoreFPU) {
1135 // Make sure the control word is correct.
1136 __ fldcw(ExternalAddress(StubRoutines::x86::addr_fpu_cntrl_wrd_std()));
1137 }
1138 #endif // _LP64
1139
1140 // check for safepoint operation in progress and/or pending suspend requests
1141 {
1142 Label Continue;
1143 Label slow_path;
1144
1145 __ safepoint_poll(slow_path, thread, true /* at_return */, false /* in_nmethod */);
1146
1147 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1148 __ jcc(Assembler::equal, Continue);
1149 __ bind(slow_path);
1150
1151 // Don't use call_VM as it will see a possible pending exception
1152 // and forward it and never return here preventing us from
1153 // clearing _last_native_pc down below. Also can't use
1154 // call_VM_leaf either as it will check to see if r13 & r14 are
1155 // preserved and correspond to the bcp/locals pointers. So we do a
1156 // runtime call by hand.
1157 //
1158 #ifndef _LP64
1159 __ push(thread);
1160 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1161 JavaThread::check_special_condition_for_native_trans)));
1162 __ increment(rsp, wordSize);
1163 __ get_thread(thread);
1164 #else
1165 __ mov(c_rarg0, r15_thread);
1166 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1167 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1168 __ andptr(rsp, -16); // align stack as required by ABI
1169 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1170 __ mov(rsp, r12); // restore sp
1171 __ reinit_heapbase();
1172 #endif // _LP64
1173 __ bind(Continue);
1174 }
1175
1176 // change thread state
1177 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1178
1179 #ifdef _LP64
1180 if (LockingMode != LM_LEGACY) {
1181 // Check preemption for Object.wait()
1182 Label not_preempted;
1183 __ movptr(rscratch1, Address(r15_thread, JavaThread::preempt_alternate_return_offset()));
1184 __ cmpptr(rscratch1, NULL_WORD);
1185 __ jccb(Assembler::equal, not_preempted);
1186 __ movptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
1187 __ jmp(rscratch1);
1188 __ bind(native_return);
1189 __ restore_after_resume(true /* is_native */);
1190 __ bind(not_preempted);
1191 } else {
1192 // any pc will do so just use this one for LM_LEGACY to keep code together.
1193 __ bind(native_return);
1194 }
1195 #endif // _LP64
1196
1197 // reset_last_Java_frame
1198 __ reset_last_Java_frame(thread, true);
1199
1200 if (CheckJNICalls) {
1201 // clear_pending_jni_exception_check
1202 __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1203 }
1204
1205 // reset handle block
1206 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1207 __ movl(Address(t, JNIHandleBlock::top_offset()), NULL_WORD);
1208
1209 // If result is an oop unbox and store it in frame where gc will see it
1210 // and result handler will pick it up
1211
1212 {
1213 Label no_oop;
1214 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1215 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1216 __ jcc(Assembler::notEqual, no_oop);
1217 // retrieve result
1218 __ pop(ltos);
1219 // Unbox oop result, e.g. JNIHandles::resolve value.
1220 __ resolve_jobject(rax /* value */,
1221 thread /* thread */,
1222 t /* tmp */);
1223 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1224 // keep stack depth as expected by pushing oop which will eventually be discarded
1225 __ push(ltos);
1226 __ bind(no_oop);
1227 }
1228
1229
1230 {
1231 Label no_reguard;
1232 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1233 StackOverflow::stack_guard_yellow_reserved_disabled);
1234 __ jcc(Assembler::notEqual, no_reguard);
1235
1236 __ pusha(); // XXX only save smashed registers
1237 #ifndef _LP64
1238 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1239 __ popa();
1240 #else
1241 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1242 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1243 __ andptr(rsp, -16); // align stack as required by ABI
1244 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1245 __ mov(rsp, r12); // restore sp
1246 __ popa(); // XXX only restore smashed registers
1247 __ reinit_heapbase();
1248 #endif // _LP64
1249
1250 __ bind(no_reguard);
1251 }
1252
1253
1254 // The method register is junk from after the thread_in_native transition
1255 // until here. Also can't call_VM until the bcp has been
1256 // restored. Need bcp for throwing exception below so get it now.
1257 __ get_method(method);
1258
1259 // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1260 __ movptr(rbcp, Address(method, Method::const_offset())); // get ConstMethod*
1261 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
1262
1263 // handle exceptions (exception handling will handle unlocking!)
1264 {
1265 Label L;
1266 __ cmpptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
1267 __ jcc(Assembler::zero, L);
1268 // Note: At some point we may want to unify this with the code
1269 // used in call_VM_base(); i.e., we should use the
1270 // StubRoutines::forward_exception code. For now this doesn't work
1271 // here because the rsp is not correctly set at this point.
1272 __ MacroAssembler::call_VM(noreg,
1273 CAST_FROM_FN_PTR(address,
1274 InterpreterRuntime::throw_pending_exception));
1275 __ should_not_reach_here();
1276 __ bind(L);
1277 }
1278
1279 // do unlocking if necessary
1280 {
1281 Label L;
1282 __ load_unsigned_short(t, Address(method, Method::access_flags_offset()));
1283 __ testl(t, JVM_ACC_SYNCHRONIZED);
1284 __ jcc(Assembler::zero, L);
1285 // the code below should be shared with interpreter macro
1286 // assembler implementation
1287 {
1288 Label unlock;
1289 // BasicObjectLock will be first in list, since this is a
1290 // synchronized method. However, need to check that the object
1291 // has not been unlocked by an explicit monitorexit bytecode.
1292 const Address monitor(rbp,
1293 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1294 wordSize - (int)sizeof(BasicObjectLock)));
1295
1296 const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1297
1298 // monitor expect in c_rarg1 for slow unlock path
1299 __ lea(regmon, monitor); // address of first monitor
1300
1301 __ movptr(t, Address(regmon, BasicObjectLock::obj_offset()));
1302 __ testptr(t, t);
1303 __ jcc(Assembler::notZero, unlock);
1304
1305 // Entry already unlocked, need to throw exception
1306 __ MacroAssembler::call_VM(noreg,
1307 CAST_FROM_FN_PTR(address,
1308 InterpreterRuntime::throw_illegal_monitor_state_exception));
1309 __ should_not_reach_here();
1310
1311 __ bind(unlock);
1312 __ unlock_object(regmon);
1313 }
1314 __ bind(L);
1315 }
1316
1317 // jvmti support
1318 // Note: This must happen _after_ handling/throwing any exceptions since
1319 // the exception handler code notifies the runtime of method exits
1320 // too. If this happens before, method entry/exit notifications are
1321 // not properly paired (was bug - gri 11/22/99).
1322 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1323
1324 // restore potential result in edx:eax, call result handler to
1325 // restore potential result in ST0 & handle result
1326
1327 __ pop(ltos);
1328 LP64_ONLY( __ pop(dtos));
1329
1330 __ movptr(t, Address(rbp,
1331 (frame::interpreter_frame_result_handler_offset) * wordSize));
1332 __ call(t);
1333
1334 // remove activation
1335 __ movptr(t, Address(rbp,
1336 frame::interpreter_frame_sender_sp_offset *
1337 wordSize)); // get sender sp
1338 __ leave(); // remove frame anchor
1339 __ pop(rdi); // get return address
1340 __ mov(rsp, t); // set sp to sender sp
1341 __ jmp(rdi);
1342
1343 if (inc_counter) {
1344 // Handle overflow of counter and compile method
1345 __ bind(invocation_counter_overflow);
1346 generate_counter_overflow(continue_after_compile);
1347 }
1348
1349 return entry_point;
1350 }
1351
1352 // Abstract method entry
1353 // Attempt to execute abstract method. Throw exception
1354 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1355
1356 address entry_point = __ pc();
1357
1358 // abstract method entry
1359
1360 // pop return address, reset last_sp to null
1361 __ empty_expression_stack();
1362 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1363 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1364
1365 // throw exception
1366 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod), rbx);
1367 // the call_VM checks for exception, so we should never return here.
1368 __ should_not_reach_here();
1369
1370 return entry_point;
1371 }
1372
1373 //
1374 // Generic interpreted method entry to (asm) interpreter
1375 //
1376 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1377 // determine code generation flags
1378 bool inc_counter = UseCompiler || CountCompiledCalls;
1379
1380 // ebx: Method*
1381 // rbcp: sender sp (set in InterpreterMacroAssembler::prepare_to_jump_from_interpreted / generate_call_stub)
1382 address entry_point = __ pc();
1383
1384 const Address constMethod(rbx, Method::const_offset());
1385 const Address access_flags(rbx, Method::access_flags_offset());
1386 const Address size_of_parameters(rdx,
1387 ConstMethod::size_of_parameters_offset());
1388 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1389
1390
1391 // get parameter size (always needed)
1392 __ movptr(rdx, constMethod);
1393 __ load_unsigned_short(rcx, size_of_parameters);
1394
1395 // rbx: Method*
1396 // rcx: size of parameters
1397 // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1398
1399 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1400 __ subl(rdx, rcx); // rdx = no. of additional locals
1401
1402 // YYY
1403 // __ incrementl(rdx);
1404 // __ andl(rdx, -2);
1405
1406 // see if we've got enough room on the stack for locals plus overhead.
1407 generate_stack_overflow_check();
1408
1409 // get return address
1410 __ pop(rax);
1411
1412 // compute beginning of parameters
1413 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1414
1415 // rdx - # of additional locals
1416 // allocate space for locals
1417 // explicitly initialize locals
1418 {
1419 Label exit, loop;
1420 __ testl(rdx, rdx);
1421 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1422 __ bind(loop);
1423 __ push(NULL_WORD); // initialize local variables
1424 __ decrementl(rdx); // until everything initialized
1425 __ jcc(Assembler::greater, loop);
1426 __ bind(exit);
1427 }
1428
1429 // initialize fixed part of activation frame
1430 generate_fixed_frame(false);
1431
1432 // make sure method is not native & not abstract
1433 #ifdef ASSERT
1434 __ load_unsigned_short(rax, access_flags);
1435 {
1436 Label L;
1437 __ testl(rax, JVM_ACC_NATIVE);
1438 __ jcc(Assembler::zero, L);
1439 __ stop("tried to execute native method as non-native");
1440 __ bind(L);
1441 }
1442 {
1443 Label L;
1444 __ testl(rax, JVM_ACC_ABSTRACT);
1445 __ jcc(Assembler::zero, L);
1446 __ stop("tried to execute abstract method in interpreter");
1447 __ bind(L);
1448 }
1449 #endif
1450
1451 // Since at this point in the method invocation the exception
1452 // handler would try to exit the monitor of synchronized methods
1453 // which hasn't been entered yet, we set the thread local variable
1454 // _do_not_unlock_if_synchronized to true. The remove_activation
1455 // will check this flag.
1456
1457 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1458 NOT_LP64(__ get_thread(thread));
1459 const Address do_not_unlock_if_synchronized(thread,
1460 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1461 __ movbool(do_not_unlock_if_synchronized, true);
1462
1463 __ profile_parameters_type(rax, rcx, rdx);
1464 // increment invocation count & check for overflow
1465 Label invocation_counter_overflow;
1466 if (inc_counter) {
1467 generate_counter_incr(&invocation_counter_overflow);
1468 }
1469
1470 Label continue_after_compile;
1471 __ bind(continue_after_compile);
1472
1473 // check for synchronized interpreted methods
1474 bang_stack_shadow_pages(false);
1475
1476 // reset the _do_not_unlock_if_synchronized flag
1477 NOT_LP64(__ get_thread(thread));
1478 __ movbool(do_not_unlock_if_synchronized, false);
1479
1480 // check for synchronized methods
1481 // Must happen AFTER invocation_counter check and stack overflow check,
1482 // so method is not locked if overflows.
1483 if (synchronized) {
1484 // Allocate monitor and lock method
1485 lock_method();
1486 } else {
1487 // no synchronization necessary
1488 #ifdef ASSERT
1489 {
1490 Label L;
1491 __ load_unsigned_short(rax, access_flags);
1492 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1493 __ jcc(Assembler::zero, L);
1494 __ stop("method needs synchronization");
1495 __ bind(L);
1496 }
1497 #endif
1498 }
1499
1500 // start execution
1501 #ifdef ASSERT
1502 {
1503 Label L;
1504 const Address monitor_block_top (rbp,
1505 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1506 __ movptr(rax, monitor_block_top);
1507 __ lea(rax, Address(rbp, rax, Address::times_ptr));
1508 __ cmpptr(rax, rsp);
1509 __ jcc(Assembler::equal, L);
1510 __ stop("broken stack frame setup in interpreter 6");
1511 __ bind(L);
1512 }
1513 #endif
1514
1515 // jvmti support
1516 __ notify_method_entry();
1517
1518 __ dispatch_next(vtos);
1519
1520 // invocation counter overflow
1521 if (inc_counter) {
1522 // Handle overflow of counter and compile method
1523 __ bind(invocation_counter_overflow);
1524 generate_counter_overflow(continue_after_compile);
1525 }
1526
1527 return entry_point;
1528 }
1529
1530 //-----------------------------------------------------------------------------
1531 // Exceptions
1532
1533 void TemplateInterpreterGenerator::generate_throw_exception() {
1534 // Entry point in previous activation (i.e., if the caller was
1535 // interpreted)
1536 Interpreter::_rethrow_exception_entry = __ pc();
1537 // Restore sp to interpreter_frame_last_sp even though we are going
1538 // to empty the expression stack for the exception processing.
1539 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1540 // rax: exception
1541 // rdx: return address/pc that threw exception
1542 __ restore_bcp(); // r13/rsi points to call/send
1543 __ restore_locals();
1544 LP64_ONLY(__ reinit_heapbase()); // restore r12 as heapbase.
1545 // Entry point for exceptions thrown within interpreter code
1546 Interpreter::_throw_exception_entry = __ pc();
1547 // expression stack is undefined here
1548 // rax: exception
1549 // r13/rsi: exception bcp
1550 __ verify_oop(rax);
1551 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1552 LP64_ONLY(__ mov(c_rarg1, rax));
1553
1554 // expression stack must be empty before entering the VM in case of
1555 // an exception
1556 __ empty_expression_stack();
1557 // find exception handler address and preserve exception oop
1558 __ call_VM(rdx,
1559 CAST_FROM_FN_PTR(address,
1560 InterpreterRuntime::exception_handler_for_exception),
1561 rarg);
1562 // rax: exception handler entry point
1563 // rdx: preserved exception oop
1564 // r13/rsi: bcp for exception handler
1565 __ push_ptr(rdx); // push exception which is now the only value on the stack
1566 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1567
1568 // If the exception is not handled in the current frame the frame is
1569 // removed and the exception is rethrown (i.e. exception
1570 // continuation is _rethrow_exception).
1571 //
1572 // Note: At this point the bci is still the bxi for the instruction
1573 // which caused the exception and the expression stack is
1574 // empty. Thus, for any VM calls at this point, GC will find a legal
1575 // oop map (with empty expression stack).
1576
1577 // In current activation
1578 // tos: exception
1579 // esi: exception bcp
1580
1581 //
1582 // JVMTI PopFrame support
1583 //
1584
1585 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1586 __ empty_expression_stack();
1587 // Set the popframe_processing bit in pending_popframe_condition
1588 // indicating that we are currently handling popframe, so that
1589 // call_VMs that may happen later do not trigger new popframe
1590 // handling cycles.
1591 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1592 NOT_LP64(__ get_thread(thread));
1593 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1594 __ orl(rdx, JavaThread::popframe_processing_bit);
1595 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1596
1597 {
1598 // Check to see whether we are returning to a deoptimized frame.
1599 // (The PopFrame call ensures that the caller of the popped frame is
1600 // either interpreted or compiled and deoptimizes it if compiled.)
1601 // In this case, we can't call dispatch_next() after the frame is
1602 // popped, but instead must save the incoming arguments and restore
1603 // them after deoptimization has occurred.
1604 //
1605 // Note that we don't compare the return PC against the
1606 // deoptimization blob's unpack entry because of the presence of
1607 // adapter frames in C2.
1608 Label caller_not_deoptimized;
1609 Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1610 __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1611 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1612 InterpreterRuntime::interpreter_contains), rarg);
1613 __ testl(rax, rax);
1614 __ jcc(Assembler::notZero, caller_not_deoptimized);
1615
1616 // Compute size of arguments for saving when returning to
1617 // deoptimized caller
1618 __ get_method(rax);
1619 __ movptr(rax, Address(rax, Method::const_offset()));
1620 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1621 size_of_parameters_offset())));
1622 __ shll(rax, Interpreter::logStackElementSize);
1623 __ restore_locals();
1624 __ subptr(rlocals, rax);
1625 __ addptr(rlocals, wordSize);
1626 // Save these arguments
1627 NOT_LP64(__ get_thread(thread));
1628 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1629 Deoptimization::
1630 popframe_preserve_args),
1631 thread, rax, rlocals);
1632
1633 __ remove_activation(vtos, rdx,
1634 /* throw_monitor_exception */ false,
1635 /* install_monitor_exception */ false,
1636 /* notify_jvmdi */ false);
1637
1638 // Inform deoptimization that it is responsible for restoring
1639 // these arguments
1640 NOT_LP64(__ get_thread(thread));
1641 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1642 JavaThread::popframe_force_deopt_reexecution_bit);
1643
1644 // Continue in deoptimization handler
1645 __ jmp(rdx);
1646
1647 __ bind(caller_not_deoptimized);
1648 }
1649
1650 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1651 /* throw_monitor_exception */ false,
1652 /* install_monitor_exception */ false,
1653 /* notify_jvmdi */ false);
1654
1655 // Finish with popframe handling
1656 // A previous I2C followed by a deoptimization might have moved the
1657 // outgoing arguments further up the stack. PopFrame expects the
1658 // mutations to those outgoing arguments to be preserved and other
1659 // constraints basically require this frame to look exactly as
1660 // though it had previously invoked an interpreted activation with
1661 // no space between the top of the expression stack (current
1662 // last_sp) and the top of stack. Rather than force deopt to
1663 // maintain this kind of invariant all the time we call a small
1664 // fixup routine to move the mutated arguments onto the top of our
1665 // expression stack if necessary.
1666 #ifndef _LP64
1667 __ mov(rax, rsp);
1668 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1669 __ lea(rbx, Address(rbp, rbx, Address::times_ptr));
1670 __ get_thread(thread);
1671 // PC must point into interpreter here
1672 __ set_last_Java_frame(thread, noreg, rbp, __ pc(), noreg);
1673 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1674 __ get_thread(thread);
1675 #else
1676 __ mov(c_rarg1, rsp);
1677 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1678 __ lea(c_rarg2, Address(rbp, c_rarg2, Address::times_ptr));
1679 // PC must point into interpreter here
1680 __ set_last_Java_frame(noreg, rbp, __ pc(), rscratch1);
1681 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1682 #endif
1683 __ reset_last_Java_frame(thread, true);
1684
1685 // Restore the last_sp and null it out
1686 __ movptr(rcx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1687 __ lea(rsp, Address(rbp, rcx, Address::times_ptr));
1688 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1689
1690 __ restore_bcp();
1691 __ restore_locals();
1692 // The method data pointer was incremented already during
1693 // call profiling. We have to restore the mdp for the current bcp.
1694 if (ProfileInterpreter) {
1695 __ set_method_data_pointer_for_bcp();
1696 }
1697
1698 // Clear the popframe condition flag
1699 NOT_LP64(__ get_thread(thread));
1700 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1701 JavaThread::popframe_inactive);
1702
1703 #if INCLUDE_JVMTI
1704 {
1705 Label L_done;
1706 const Register local0 = rlocals;
1707
1708 __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1709 __ jcc(Assembler::notEqual, L_done);
1710
1711 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1712 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
1713
1714 __ get_method(rdx);
1715 __ movptr(rax, Address(local0, 0));
1716 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1717
1718 __ testptr(rax, rax);
1719 __ jcc(Assembler::zero, L_done);
1720
1721 __ movptr(Address(rbx, 0), rax);
1722 __ bind(L_done);
1723 }
1724 #endif // INCLUDE_JVMTI
1725
1726 __ dispatch_next(vtos);
1727 // end of PopFrame support
1728
1729 Interpreter::_remove_activation_entry = __ pc();
1730
1731 // preserve exception over this code sequence
1732 __ pop_ptr(rax);
1733 NOT_LP64(__ get_thread(thread));
1734 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1735 // remove the activation (without doing throws on illegalMonitorExceptions)
1736 __ remove_activation(vtos, rdx, false, true, false);
1737 // restore exception
1738 NOT_LP64(__ get_thread(thread));
1739 __ get_vm_result(rax, thread);
1740
1741 // In between activations - previous activation type unknown yet
1742 // compute continuation point - the continuation point expects the
1743 // following registers set up:
1744 //
1745 // rax: exception
1746 // rdx: return address/pc that threw exception
1747 // rsp: expression stack of caller
1748 // rbp: ebp of caller
1749 __ push(rax); // save exception
1750 __ push(rdx); // save return address
1751 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1752 SharedRuntime::exception_handler_for_return_address),
1753 thread, rdx);
1754 __ mov(rbx, rax); // save exception handler
1755 __ pop(rdx); // restore return address
1756 __ pop(rax); // restore exception
1757 // Note that an "issuing PC" is actually the next PC after the call
1758 __ jmp(rbx); // jump to exception
1759 // handler of caller
1760 }
1761
1762
1763 //
1764 // JVMTI ForceEarlyReturn support
1765 //
1766 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1767 address entry = __ pc();
1768
1769 __ restore_bcp();
1770 __ restore_locals();
1771 __ empty_expression_stack();
1772 __ load_earlyret_value(state); // 32 bits returns value in rdx, so don't reuse
1773
1774 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1775 NOT_LP64(__ get_thread(thread));
1776 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1777 Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1778
1779 // Clear the earlyret state
1780 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1781
1782 __ remove_activation(state, rsi,
1783 false, /* throw_monitor_exception */
1784 false, /* install_monitor_exception */
1785 true); /* notify_jvmdi */
1786 __ jmp(rsi);
1787
1788 return entry;
1789 } // end of ForceEarlyReturn support
1790
1791
1792 //-----------------------------------------------------------------------------
1793 // Helper for vtos entry point generation
1794
1795 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1796 address& bep,
1797 address& cep,
1798 address& sep,
1799 address& aep,
1800 address& iep,
1801 address& lep,
1802 address& fep,
1803 address& dep,
1804 address& vep) {
1805 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1806 Label L;
1807 #ifndef _LP64
1808 fep = __ pc(); // ftos entry point
1809 __ push(ftos);
1810 __ jmpb(L);
1811 dep = __ pc(); // dtos entry point
1812 __ push(dtos);
1813 __ jmpb(L);
1814 #else
1815 fep = __ pc(); // ftos entry point
1816 __ push_f(xmm0);
1817 __ jmpb(L);
1818 dep = __ pc(); // dtos entry point
1819 __ push_d(xmm0);
1820 __ jmpb(L);
1821 #endif // _LP64
1822 lep = __ pc(); // ltos entry point
1823 __ push_l();
1824 __ jmpb(L);
1825 aep = bep = cep = sep = iep = __ pc(); // [abcsi]tos entry point
1826 __ push_i_or_ptr();
1827 vep = __ pc(); // vtos entry point
1828 __ bind(L);
1829 generate_and_dispatch(t);
1830 }
1831
1832 //-----------------------------------------------------------------------------
1833
1834 // Non-product code
1835 #ifndef PRODUCT
1836
1837 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1838 address entry = __ pc();
1839
1840 #ifndef _LP64
1841 // prepare expression stack
1842 __ pop(rcx); // pop return address so expression stack is 'pure'
1843 __ push(state); // save tosca
1844
1845 // pass tosca registers as arguments & call tracer
1846 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1847 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1848 __ pop(state); // restore tosca
1849
1850 // return
1851 __ jmp(rcx);
1852 #else
1853 __ push(state);
1854 __ push(c_rarg0);
1855 __ push(c_rarg1);
1856 __ push(c_rarg2);
1857 __ push(c_rarg3);
1858 __ mov(c_rarg2, rax); // Pass itos
1859 #ifdef _WIN64
1860 __ movflt(xmm3, xmm0); // Pass ftos
1861 #endif
1862 __ call_VM(noreg,
1863 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1864 c_rarg1, c_rarg2, c_rarg3);
1865 __ pop(c_rarg3);
1866 __ pop(c_rarg2);
1867 __ pop(c_rarg1);
1868 __ pop(c_rarg0);
1869 __ pop(state);
1870 __ ret(0); // return from result handler
1871 #endif // _LP64
1872
1873 return entry;
1874 }
1875
1876 void TemplateInterpreterGenerator::count_bytecode() {
1877 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value), rscratch1);
1878 }
1879
1880 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1881 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]), rscratch1);
1882 }
1883
1884 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1885 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1886 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1887 __ orl(rbx,
1888 ((int) t->bytecode()) <<
1889 BytecodePairHistogram::log2_number_of_codes);
1890 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx, rscratch1);
1891 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1892 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1893 }
1894
1895
1896 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1897 // Call a little run-time stub to avoid blow-up for each bytecode.
1898 // The run-time runtime saves the right registers, depending on
1899 // the tosca in-state for the given template.
1900
1901 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
1902 "entry must have been generated");
1903 #ifndef _LP64
1904 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1905 #else
1906 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1907 __ andptr(rsp, -16); // align stack as required by ABI
1908 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1909 __ mov(rsp, r12); // restore sp
1910 __ reinit_heapbase();
1911 #endif // _LP64
1912 }
1913
1914
1915 void TemplateInterpreterGenerator::stop_interpreter_at() {
1916 Label L;
1917 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1918 StopInterpreterAt,
1919 rscratch1);
1920 __ jcc(Assembler::notEqual, L);
1921 __ int3();
1922 __ bind(L);
1923 }
1924 #endif // !PRODUCT