1 /*
2 * Copyright (c) 2003, 2023, 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 "classfile/javaClasses.hpp"
28 #include "compiler/compiler_globals.hpp"
29 #include "compiler/disassembler.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interpreter/bytecodeHistogram.hpp"
32 #include "interpreter/interp_masm.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "interpreter/interpreterRuntime.hpp"
35 #include "interpreter/templateInterpreterGenerator.hpp"
36 #include "interpreter/templateTable.hpp"
37 #include "oops/arrayOop.hpp"
38 #include "oops/methodCounters.hpp"
39 #include "oops/methodData.hpp"
40 #include "oops/method.hpp"
41 #include "oops/oop.inline.hpp"
42 #include "oops/resolvedIndyEntry.hpp"
43 #include "oops/resolvedMethodEntry.hpp"
44 #include "prims/jvmtiExport.hpp"
45 #include "prims/jvmtiThreadState.hpp"
46 #include "runtime/continuation.hpp"
47 #include "runtime/deoptimization.hpp"
48 #include "runtime/frame.inline.hpp"
49 #include "runtime/globals.hpp"
50 #include "runtime/jniHandles.hpp"
51 #include "runtime/sharedRuntime.hpp"
52 #include "runtime/stubRoutines.hpp"
53 #include "runtime/synchronizer.hpp"
54 #include "runtime/timer.hpp"
55 #include "runtime/vframeArray.hpp"
56 #include "utilities/checkedCast.hpp"
57 #include "utilities/debug.hpp"
58 #include "utilities/macros.hpp"
59
60 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
61
62 // Size of interpreter code. Increase if too small. Interpreter will
63 // fail with a guarantee ("not enough space for interpreter generation");
64 // if too small.
65 // Run with +PrintInterpreter to get the VM to print out the size.
66 // Max size with JVMTI
67 #ifdef AMD64
68 int TemplateInterpreter::InterpreterCodeSize = JVMCI_ONLY(268) NOT_JVMCI(256) * 1024;
69 #else
70 int TemplateInterpreter::InterpreterCodeSize = 224 * 1024;
71 #endif // AMD64
72
73 // Global Register Names
74 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
75 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
76
77 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
78 const int bcp_offset = frame::interpreter_frame_bcp_offset * wordSize;
79 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
80
81
82 //-----------------------------------------------------------------------------
83
84 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
85 address entry = __ pc();
86
87 #ifdef ASSERT
88 {
89 Label L;
90 __ movptr(rax, Address(rbp,
91 frame::interpreter_frame_monitor_block_top_offset *
92 wordSize));
93 __ lea(rax, Address(rbp, rax, Address::times_ptr));
94 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
95 // grows negative)
96 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
97 __ stop ("interpreter frame not set up");
98 __ bind(L);
99 }
100 #endif // ASSERT
101 // Restore bcp under the assumption that the current frame is still
102 // interpreted
103 __ restore_bcp();
104
105 // expression stack must be empty before entering the VM if an
106 // exception happened
107 __ empty_expression_stack();
108 // throw exception
109 __ call_VM(noreg,
110 CAST_FROM_FN_PTR(address,
111 InterpreterRuntime::throw_StackOverflowError));
112 return entry;
113 }
114
115 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
116 address entry = __ pc();
117 // The expression stack must be empty before entering the VM if an
118 // exception happened.
119 __ empty_expression_stack();
120
121 // Setup parameters.
122 // ??? convention: expect aberrant index in register ebx/rbx.
123 // Pass array to create more detailed exceptions.
124 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
125 __ call_VM(noreg,
126 CAST_FROM_FN_PTR(address,
127 InterpreterRuntime::
128 throw_ArrayIndexOutOfBoundsException),
129 rarg, rbx);
130 return entry;
131 }
132
133 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
134 address entry = __ pc();
135
136 // object is at TOS
137 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
138 __ pop(rarg);
139
140 // expression stack must be empty before entering the VM if an
141 // exception happened
142 __ empty_expression_stack();
143
144 __ call_VM(noreg,
145 CAST_FROM_FN_PTR(address,
146 InterpreterRuntime::
147 throw_ClassCastException),
148 rarg);
149 return entry;
150 }
151
152 address TemplateInterpreterGenerator::generate_exception_handler_common(
153 const char* name, const char* message, bool pass_oop) {
154 assert(!pass_oop || message == nullptr, "either oop or message but not both");
155 address entry = __ pc();
156
157 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
158 Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
159
160 if (pass_oop) {
161 // object is at TOS
162 __ pop(rarg2);
163 }
164 // expression stack must be empty before entering the VM if an
165 // exception happened
166 __ empty_expression_stack();
167 // setup parameters
168 __ lea(rarg, ExternalAddress((address)name));
169 if (pass_oop) {
170 __ call_VM(rax, CAST_FROM_FN_PTR(address,
171 InterpreterRuntime::
172 create_klass_exception),
173 rarg, rarg2);
174 } else {
175 __ lea(rarg2, ExternalAddress((address)message));
176 __ call_VM(rax,
177 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
178 rarg, rarg2);
179 }
180 // throw exception
181 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
182 return entry;
183 }
184
185 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
186 address entry = __ pc();
187
188 #ifndef _LP64
189 #ifdef COMPILER2
190 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
191 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
192 for (int i = 1; i < 8; i++) {
193 __ ffree(i);
194 }
195 } else if (UseSSE < 2) {
196 __ empty_FPU_stack();
197 }
198 #endif // COMPILER2
199 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
200 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
201 } else {
202 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
203 }
204
205 if (state == ftos) {
206 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
207 } else if (state == dtos) {
208 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
209 }
210 #endif // _LP64
211
212 // Restore stack bottom in case i2c adjusted stack
213 __ movptr(rcx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
214 __ lea(rsp, Address(rbp, rcx, Address::times_ptr));
215 // and null it as marker that esp is now tos until next java call
216 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
217
218 __ restore_bcp();
219 __ restore_locals();
220
221 if (state == atos) {
222 Register mdp = rbx;
223 Register tmp = rcx;
224 __ profile_return_type(mdp, rax, tmp);
225 }
226
227 const Register cache = rbx;
228 const Register index = rcx;
229 if (index_size == sizeof(u4)) {
230 __ load_resolved_indy_entry(cache, index);
231 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
232 __ lea(rsp, Address(rsp, cache, Interpreter::stackElementScale()));
233 } else {
234 assert(index_size == sizeof(u2), "Can only be u2");
235 __ load_method_entry(cache, index);
236 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
237 __ lea(rsp, Address(rsp, cache, Interpreter::stackElementScale()));
238 }
239
240 const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
241 if (JvmtiExport::can_pop_frame()) {
242 NOT_LP64(__ get_thread(java_thread));
243 __ check_and_handle_popframe(java_thread);
244 }
245 if (JvmtiExport::can_force_early_return()) {
246 NOT_LP64(__ get_thread(java_thread));
247 __ check_and_handle_earlyret(java_thread);
248 }
249
250 __ dispatch_next(state, step);
251
252 return entry;
253 }
254
255
256 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
257 address entry = __ pc();
258
259 #ifndef _LP64
260 if (state == ftos) {
261 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
262 } else if (state == dtos) {
263 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
264 }
265 #endif // _LP64
266
267 // null last_sp until next java call
268 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
269 __ restore_bcp();
270 __ restore_locals();
271 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
272 NOT_LP64(__ get_thread(thread));
273 #if INCLUDE_JVMCI
274 // Check if we need to take lock at entry of synchronized method. This can
275 // only occur on method entry so emit it only for vtos with step 0.
276 if (EnableJVMCI && state == vtos && step == 0) {
277 Label L;
278 __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
279 __ jcc(Assembler::zero, L);
280 // Clear flag.
281 __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
282 // Satisfy calling convention for lock_method().
283 __ get_method(rbx);
284 // Take lock.
285 lock_method();
286 __ bind(L);
287 } else {
288 #ifdef ASSERT
289 if (EnableJVMCI) {
290 Label L;
291 __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
292 __ jcc(Assembler::zero, L);
293 __ stop("unexpected pending monitor in deopt entry");
294 __ bind(L);
295 }
296 #endif
297 }
298 #endif
299 // handle exceptions
300 {
301 Label L;
302 __ cmpptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
303 __ jcc(Assembler::zero, L);
304 __ call_VM(noreg,
305 CAST_FROM_FN_PTR(address,
306 InterpreterRuntime::throw_pending_exception));
307 __ should_not_reach_here();
308 __ bind(L);
309 }
310 if (continuation == nullptr) {
311 __ dispatch_next(state, step);
312 } else {
313 __ jump_to_entry(continuation);
314 }
315 return entry;
316 }
317
318 address TemplateInterpreterGenerator::generate_result_handler_for(
319 BasicType type) {
320 address entry = __ pc();
321 switch (type) {
322 case T_BOOLEAN: __ c2bool(rax); break;
323 #ifndef _LP64
324 case T_CHAR : __ andptr(rax, 0xFFFF); break;
325 #else
326 case T_CHAR : __ movzwl(rax, rax); break;
327 #endif // _LP64
328 case T_BYTE : __ sign_extend_byte(rax); break;
329 case T_SHORT : __ sign_extend_short(rax); break;
330 case T_INT : /* nothing to do */ break;
331 case T_LONG : /* nothing to do */ break;
332 case T_VOID : /* nothing to do */ break;
333 #ifndef _LP64
334 case T_DOUBLE :
335 case T_FLOAT :
336 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
337 __ pop(t); // remove return address first
338 // Must return a result for interpreter or compiler. In SSE
339 // mode, results are returned in xmm0 and the FPU stack must
340 // be empty.
341 if (type == T_FLOAT && UseSSE >= 1) {
342 // Load ST0
343 __ fld_d(Address(rsp, 0));
344 // Store as float and empty fpu stack
345 __ fstp_s(Address(rsp, 0));
346 // and reload
347 __ movflt(xmm0, Address(rsp, 0));
348 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
349 __ movdbl(xmm0, Address(rsp, 0));
350 } else {
351 // restore ST0
352 __ fld_d(Address(rsp, 0));
353 }
354 // and pop the temp
355 __ addptr(rsp, 2 * wordSize);
356 __ push(t); // restore return address
357 }
358 break;
359 #else
360 case T_FLOAT : /* nothing to do */ break;
361 case T_DOUBLE : /* nothing to do */ break;
362 #endif // _LP64
363
364 case T_OBJECT :
365 // retrieve result from frame
366 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
367 // and verify it
368 __ verify_oop(rax);
369 break;
370 default : ShouldNotReachHere();
371 }
372 __ ret(0); // return from result handler
373 return entry;
374 }
375
376 address TemplateInterpreterGenerator::generate_safept_entry_for(
377 TosState state,
378 address runtime_entry) {
379 address entry = __ pc();
380
381 __ push(state);
382 __ push_cont_fastpath();
383 __ call_VM(noreg, runtime_entry);
384 __ pop_cont_fastpath();
385
386 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
387 return entry;
388 }
389
390 address TemplateInterpreterGenerator::generate_cont_preempt_rerun_interpreter_adapter() {
391 if (!Continuations::enabled()) return nullptr;
392 address start = __ pc();
393
394 __ pop(rbp);
395
396 // We will return to the intermediate call made in call_VM skipping the restoration
397 // of bcp and locals done in InterpreterMacroAssembler::call_VM_base, so fix them here.
398 __ restore_bcp();
399 __ restore_locals();
400
401 // Get return address before adjusting rsp
402 __ movptr(rax, Address(rsp, 0));
403
404 // Restore stack bottom
405 __ movptr(rcx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
406 __ lea(rsp, Address(rbp, rcx, Address::times_ptr));
407 // and NULL it as marker that esp is now tos until next java call
408 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
409
410 __ jmp(rax);
411
412 return start;
413 }
414
415
416 // Helpers for commoning out cases in the various type of method entries.
417 //
418
419
420 // increment invocation count & check for overflow
421 //
422 // Note: checking for negative value instead of overflow
423 // so we have a 'sticky' overflow test
424 //
425 // rbx: method
426 // rcx: invocation counter
427 //
428 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
429 Label done;
430 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
431 Label no_mdo;
432 if (ProfileInterpreter) {
433 // Are we profiling?
434 __ movptr(rax, Address(rbx, Method::method_data_offset()));
435 __ testptr(rax, rax);
436 __ jccb(Assembler::zero, no_mdo);
437 // Increment counter in the MDO
438 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
439 in_bytes(InvocationCounter::counter_offset()));
440 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
441 __ increment_mask_and_jump(mdo_invocation_counter, mask, rcx, overflow);
442 __ jmp(done);
443 }
444 __ bind(no_mdo);
445 // Increment counter in MethodCounters
446 const Address invocation_counter(rax,
447 MethodCounters::invocation_counter_offset() +
448 InvocationCounter::counter_offset());
449 __ get_method_counters(rbx, rax, done);
450 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
451 __ increment_mask_and_jump(invocation_counter, mask, rcx, overflow);
452 __ bind(done);
453 }
454
455 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
456
457 // Asm interpreter on entry
458 // r14/rdi - locals
459 // r13/rsi - bcp
460 // rbx - method
461 // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
462 // rbp - interpreter frame
463
464 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
465 // Everything as it was on entry
466 // rdx is not restored. Doesn't appear to really be set.
467
468 // InterpreterRuntime::frequency_counter_overflow takes two
469 // arguments, the first (thread) is passed by call_VM, the second
470 // indicates if the counter overflow occurs at a backwards branch
471 // (null bcp). We pass zero for it. The call returns the address
472 // of the verified entry point for the method or null if the
473 // compilation did not complete (either went background or bailed
474 // out).
475 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
476 __ movl(rarg, 0);
477 __ call_VM(noreg,
478 CAST_FROM_FN_PTR(address,
479 InterpreterRuntime::frequency_counter_overflow),
480 rarg);
481
482 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
483 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
484 // and jump to the interpreted entry.
485 __ jmp(do_continue, relocInfo::none);
486 }
487
488 // See if we've got enough room on the stack for locals plus overhead below
489 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
490 // without going through the signal handler, i.e., reserved and yellow zones
491 // will not be made usable. The shadow zone must suffice to handle the
492 // overflow.
493 // The expression stack grows down incrementally, so the normal guard
494 // page mechanism will work for that.
495 //
496 // NOTE: Since the additional locals are also always pushed (wasn't
497 // obvious in generate_fixed_frame) so the guard should work for them
498 // too.
499 //
500 // Args:
501 // rdx: number of additional locals this frame needs (what we must check)
502 // rbx: Method*
503 //
504 // Kills:
505 // rax
506 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
507
508 // monitor entry size: see picture of stack in frame_x86.hpp
509 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
510
511 // total overhead size: entry_size + (saved rbp through expr stack
512 // bottom). be sure to change this if you add/subtract anything
513 // to/from the overhead area
514 const int overhead_size =
515 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
516
517 const int page_size = (int)os::vm_page_size();
518
519 Label after_frame_check;
520
521 // see if the frame is greater than one page in size. If so,
522 // then we need to verify there is enough stack space remaining
523 // for the additional locals.
524 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
525 __ jcc(Assembler::belowEqual, after_frame_check);
526
527 // compute rsp as if this were going to be the last frame on
528 // the stack before the red zone
529
530 Label after_frame_check_pop;
531 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
532 #ifndef _LP64
533 __ push(thread);
534 __ get_thread(thread);
535 #endif
536
537 const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
538
539 // locals + overhead, in bytes
540 __ mov(rax, rdx);
541 __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
542 __ addptr(rax, overhead_size);
543
544 #ifdef ASSERT
545 Label limit_okay;
546 // Verify that thread stack overflow limit is non-zero.
547 __ cmpptr(stack_limit, NULL_WORD);
548 __ jcc(Assembler::notEqual, limit_okay);
549 __ stop("stack overflow limit is zero");
550 __ bind(limit_okay);
551 #endif
552
553 // Add locals/frame size to stack limit.
554 __ addptr(rax, stack_limit);
555
556 // Check against the current stack bottom.
557 __ cmpptr(rsp, rax);
558
559 __ jcc(Assembler::above, after_frame_check_pop);
560 NOT_LP64(__ pop(rsi)); // get saved bcp
561
562 // Restore sender's sp as SP. This is necessary if the sender's
563 // frame is an extended compiled frame (see gen_c2i_adapter())
564 // and safer anyway in case of JSR292 adaptations.
565
566 __ pop(rax); // return address must be moved if SP is changed
567 __ mov(rsp, rbcp);
568 __ push(rax);
569
570 // Note: the restored frame is not necessarily interpreted.
571 // Use the shared runtime version of the StackOverflowError.
572 assert(StubRoutines::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
573 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
574 // all done with frame size check
575 __ bind(after_frame_check_pop);
576 NOT_LP64(__ pop(rsi));
577
578 // all done with frame size check
579 __ bind(after_frame_check);
580 }
581
582 // Allocate monitor and lock method (asm interpreter)
583 //
584 // Args:
585 // rbx: Method*
586 // r14/rdi: locals
587 //
588 // Kills:
589 // rax
590 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
591 // rscratch1, rscratch2 (scratch regs)
592 void TemplateInterpreterGenerator::lock_method() {
593 // synchronize method
594 const Address access_flags(rbx, Method::access_flags_offset());
595 const Address monitor_block_top(
596 rbp,
597 frame::interpreter_frame_monitor_block_top_offset * wordSize);
598 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
599
600 #ifdef ASSERT
601 {
602 Label L;
603 __ movl(rax, access_flags);
604 __ testl(rax, JVM_ACC_SYNCHRONIZED);
605 __ jcc(Assembler::notZero, L);
606 __ stop("method doesn't need synchronization");
607 __ bind(L);
608 }
609 #endif // ASSERT
610
611 // get synchronization object
612 {
613 Label done;
614 __ movl(rax, access_flags);
615 __ testl(rax, JVM_ACC_STATIC);
616 // get receiver (assume this is frequent case)
617 __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
618 __ jcc(Assembler::zero, done);
619 __ load_mirror(rax, rbx, rscratch2);
620
621 #ifdef ASSERT
622 {
623 Label L;
624 __ testptr(rax, rax);
625 __ jcc(Assembler::notZero, L);
626 __ stop("synchronization object is null");
627 __ bind(L);
628 }
629 #endif // ASSERT
630
631 __ bind(done);
632 }
633
634 // add space for monitor & lock
635 __ subptr(rsp, entry_size); // add space for a monitor entry
636 __ subptr(monitor_block_top, entry_size / wordSize); // set new monitor block top
637 // store object
638 __ movptr(Address(rsp, BasicObjectLock::obj_offset()), rax);
639 const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
640 __ movptr(lockreg, rsp); // object address
641 __ lock_object(lockreg);
642 }
643
644 // Generate a fixed interpreter frame. This is identical setup for
645 // interpreted methods and for native methods hence the shared code.
646 //
647 // Args:
648 // rax: return address
649 // rbx: Method*
650 // r14/rdi: pointer to locals
651 // r13/rsi: sender sp
652 // rdx: cp cache
653 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
654 // initialize fixed part of activation frame
655 __ push(rax); // save return address
656 __ enter(); // save old & set new rbp
657 __ push(rbcp); // set sender sp
658 __ push(NULL_WORD); // leave last_sp as null
659 __ movptr(rbcp, Address(rbx, Method::const_offset())); // get ConstMethod*
660 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
661 __ push(rbx); // save Method*
662 // Get mirror and store it in the frame as GC root for this Method*
663 __ load_mirror(rdx, rbx, rscratch2);
664 __ push(rdx);
665 if (ProfileInterpreter) {
666 Label method_data_continue;
667 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
668 __ testptr(rdx, rdx);
669 __ jcc(Assembler::zero, method_data_continue);
670 __ addptr(rdx, in_bytes(MethodData::data_offset()));
671 __ bind(method_data_continue);
672 __ push(rdx); // set the mdp (method data pointer)
673 } else {
674 __ push(0);
675 }
676
677 __ movptr(rdx, Address(rbx, Method::const_offset()));
678 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
679 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset()));
680 __ push(rdx); // set constant pool cache
681
682 __ movptr(rax, rlocals);
683 __ subptr(rax, rbp);
684 __ shrptr(rax, Interpreter::logStackElementSize); // rax = rlocals - fp();
685 __ push(rax); // set relativized rlocals, see frame::interpreter_frame_locals()
686
687 if (native_call) {
688 __ push(0); // no bcp
689 } else {
690 __ push(rbcp); // set bcp
691 }
692 // initialize relativized pointer to expression stack bottom
693 __ push(frame::interpreter_frame_initial_sp_offset);
694 }
695
696 // End of helpers
697
698 // Method entry for java.lang.ref.Reference.get.
699 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
700 // Code: _aload_0, _getfield, _areturn
701 // parameter size = 1
702 //
703 // The code that gets generated by this routine is split into 2 parts:
704 // 1. The "intrinsified" code performing an ON_WEAK_OOP_REF load,
705 // 2. The slow path - which is an expansion of the regular method entry.
706 //
707 // Notes:-
708 // * An intrinsic is always executed, where an ON_WEAK_OOP_REF load is performed.
709 // * We may jump to the slow path iff the receiver is null. If the
710 // Reference object is null then we no longer perform an ON_WEAK_OOP_REF load
711 // Thus we can use the regular method entry code to generate the NPE.
712 //
713 // rbx: Method*
714
715 // r13: senderSP must preserve for slow path, set SP to it on fast path
716
717 address entry = __ pc();
718
719 const int referent_offset = java_lang_ref_Reference::referent_offset();
720
721 Label slow_path;
722 // rbx: method
723
724 // Check if local 0 != null
725 // If the receiver is null then it is OK to jump to the slow path.
726 __ movptr(rax, Address(rsp, wordSize));
727
728 __ testptr(rax, rax);
729 __ jcc(Assembler::zero, slow_path);
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 load barrier
737 // calls the runtime
738 NOT_LP64(__ push(rsi));
739
740 // Load the value of the referent field.
741 const Address field_address(rax, referent_offset);
742 __ load_heap_oop(rax, field_address, /*tmp1*/ rbx, /*tmp_thread*/ rdx, ON_WEAK_OOP_REF);
743
744 // _areturn
745 const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
746 NOT_LP64(__ pop(rsi)); // get sender sp
747 __ pop(rdi); // get return address
748 __ mov(rsp, sender_sp); // set sp to sender sp
749 __ jmp(rdi);
750 __ ret(0);
751
752 // generate a vanilla interpreter entry as the slow path
753 __ bind(slow_path);
754 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
755 return entry;
756 }
757
758 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
759 // See more discussion in stackOverflow.hpp.
760
761 // Note that we do the banging after the frame is setup, since the exception
762 // handling code expects to find a valid interpreter frame on the stack.
763 // Doing the banging earlier fails if the caller frame is not an interpreter
764 // frame.
765 // (Also, the exception throwing code expects to unlock any synchronized
766 // method receiver, so do the banging after locking the receiver.)
767
768 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
769 const int page_size = (int)os::vm_page_size();
770 const int n_shadow_pages = shadow_zone_size / page_size;
771
772 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
773 #ifndef _LP64
774 __ push(thread);
775 __ get_thread(thread);
776 #endif
777
778 #ifdef ASSERT
779 Label L_good_limit;
780 __ cmpptr(Address(thread, JavaThread::shadow_zone_safe_limit()), NULL_WORD);
781 __ jcc(Assembler::notEqual, L_good_limit);
782 __ stop("shadow zone safe limit is not initialized");
783 __ bind(L_good_limit);
784
785 Label L_good_watermark;
786 __ cmpptr(Address(thread, JavaThread::shadow_zone_growth_watermark()), NULL_WORD);
787 __ jcc(Assembler::notEqual, L_good_watermark);
788 __ stop("shadow zone growth watermark is not initialized");
789 __ bind(L_good_watermark);
790 #endif
791
792 Label L_done;
793
794 __ cmpptr(rsp, Address(thread, JavaThread::shadow_zone_growth_watermark()));
795 __ jcc(Assembler::above, L_done);
796
797 for (int p = 1; p <= n_shadow_pages; p++) {
798 __ bang_stack_with_offset(p*page_size);
799 }
800
801 // Record the new watermark, but only if update is above the safe limit.
802 // Otherwise, the next time around the check above would pass the safe limit.
803 __ cmpptr(rsp, Address(thread, JavaThread::shadow_zone_safe_limit()));
804 __ jccb(Assembler::belowEqual, L_done);
805 __ movptr(Address(thread, JavaThread::shadow_zone_growth_watermark()), rsp);
806
807 __ bind(L_done);
808
809 #ifndef _LP64
810 __ pop(thread);
811 #endif
812 }
813
814 // Interpreter stub for calling a native method. (asm interpreter)
815 // This sets up a somewhat different looking stack for calling the
816 // native method than the typical interpreter frame setup.
817 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
818 // determine code generation flags
819 bool inc_counter = UseCompiler || CountCompiledCalls;
820
821 // rbx: Method*
822 // rbcp: sender sp
823
824 address entry_point = __ pc();
825
826 const Address constMethod (rbx, Method::const_offset());
827 const Address access_flags (rbx, Method::access_flags_offset());
828 const Address size_of_parameters(rcx, ConstMethod::
829 size_of_parameters_offset());
830
831
832 // get parameter size (always needed)
833 __ movptr(rcx, constMethod);
834 __ load_unsigned_short(rcx, size_of_parameters);
835
836 // native calls don't need the stack size check since they have no
837 // expression stack and the arguments are already on the stack and
838 // we only add a handful of words to the stack
839
840 // rbx: Method*
841 // rcx: size of parameters
842 // rbcp: sender sp
843 __ pop(rax); // get return address
844
845 // for natives the size of locals is zero
846
847 // compute beginning of parameters
848 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
849
850 // add 2 zero-initialized slots for native calls
851 // initialize result_handler slot
852 __ push(NULL_WORD);
853 // slot for oop temp
854 // (static native method holder mirror/jni oop result)
855 __ push(NULL_WORD);
856
857 // initialize fixed part of activation frame
858 generate_fixed_frame(true);
859
860 // make sure method is native & not abstract
861 #ifdef ASSERT
862 __ movl(rax, access_flags);
863 {
864 Label L;
865 __ testl(rax, JVM_ACC_NATIVE);
866 __ jcc(Assembler::notZero, L);
867 __ stop("tried to execute non-native method as native");
868 __ bind(L);
869 }
870 {
871 Label L;
872 __ testl(rax, JVM_ACC_ABSTRACT);
873 __ jcc(Assembler::zero, L);
874 __ stop("tried to execute abstract method in interpreter");
875 __ bind(L);
876 }
877 #endif
878
879 // Since at this point in the method invocation the exception handler
880 // would try to exit the monitor of synchronized methods which hasn't
881 // been entered yet, we set the thread local variable
882 // _do_not_unlock_if_synchronized to true. The remove_activation will
883 // check this flag.
884
885 const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
886 NOT_LP64(__ get_thread(thread1));
887 const Address do_not_unlock_if_synchronized(thread1,
888 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
889 __ movbool(do_not_unlock_if_synchronized, true);
890
891 // increment invocation count & check for overflow
892 Label invocation_counter_overflow;
893 if (inc_counter) {
894 generate_counter_incr(&invocation_counter_overflow);
895 }
896
897 Label continue_after_compile;
898 __ bind(continue_after_compile);
899
900 bang_stack_shadow_pages(true);
901
902 // reset the _do_not_unlock_if_synchronized flag
903 NOT_LP64(__ get_thread(thread1));
904 __ movbool(do_not_unlock_if_synchronized, false);
905
906 // check for synchronized methods
907 // Must happen AFTER invocation_counter check and stack overflow check,
908 // so method is not locked if overflows.
909 if (synchronized) {
910 lock_method();
911 } else {
912 // no synchronization necessary
913 #ifdef ASSERT
914 {
915 Label L;
916 __ movl(rax, access_flags);
917 __ testl(rax, JVM_ACC_SYNCHRONIZED);
918 __ jcc(Assembler::zero, L);
919 __ stop("method needs synchronization");
920 __ bind(L);
921 }
922 #endif
923 }
924
925 // start execution
926 #ifdef ASSERT
927 {
928 Label L;
929 const Address monitor_block_top(rbp,
930 frame::interpreter_frame_monitor_block_top_offset * wordSize);
931 __ movptr(rax, monitor_block_top);
932 __ lea(rax, Address(rbp, rax, Address::times_ptr));
933 __ cmpptr(rax, rsp);
934 __ jcc(Assembler::equal, L);
935 __ stop("broken stack frame setup in interpreter 5");
936 __ bind(L);
937 }
938 #endif
939
940 // jvmti support
941 __ notify_method_entry();
942
943 // work registers
944 const Register method = rbx;
945 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
946 const Register t = NOT_LP64(rcx) LP64_ONLY(r11);
947
948 // allocate space for parameters
949 __ get_method(method);
950 __ movptr(t, Address(method, Method::const_offset()));
951 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
952
953 #ifndef _LP64
954 __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
955 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
956 __ subptr(rsp, t);
957 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
958 #else
959 __ shll(t, Interpreter::logStackElementSize);
960
961 __ subptr(rsp, t);
962 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
963 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
964 #endif // _LP64
965
966 // get signature handler
967 {
968 Label L;
969 __ movptr(t, Address(method, Method::signature_handler_offset()));
970 __ testptr(t, t);
971 __ jcc(Assembler::notZero, L);
972 __ call_VM(noreg,
973 CAST_FROM_FN_PTR(address,
974 InterpreterRuntime::prepare_native_call),
975 method);
976 __ get_method(method);
977 __ movptr(t, Address(method, Method::signature_handler_offset()));
978 __ bind(L);
979 }
980
981 // call signature handler
982 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
983 "adjust this code");
984 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
985 "adjust this code");
986 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
987 "adjust this code");
988
989 // The generated handlers do not touch RBX (the method).
990 // However, large signatures cannot be cached and are generated
991 // each time here. The slow-path generator can do a GC on return,
992 // so we must reload it after the call.
993 __ call(t);
994 __ get_method(method); // slow path can do a GC, reload RBX
995
996
997 // result handler is in rax
998 // set result handler
999 __ movptr(Address(rbp,
1000 (frame::interpreter_frame_result_handler_offset) * wordSize),
1001 rax);
1002
1003 // pass mirror handle if static call
1004 {
1005 Label L;
1006 __ movl(t, Address(method, Method::access_flags_offset()));
1007 __ testl(t, JVM_ACC_STATIC);
1008 __ jcc(Assembler::zero, L);
1009 // get mirror
1010 __ load_mirror(t, method, rax);
1011 // copy mirror into activation frame
1012 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1013 t);
1014 // pass handle to mirror
1015 #ifndef _LP64
1016 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1017 __ movptr(Address(rsp, wordSize), t);
1018 #else
1019 __ lea(c_rarg1,
1020 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1021 #endif // _LP64
1022 __ bind(L);
1023 }
1024
1025 // get native function entry point
1026 {
1027 Label L;
1028 __ movptr(rax, Address(method, Method::native_function_offset()));
1029 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1030 __ cmpptr(rax, unsatisfied.addr(), rscratch1);
1031 __ jcc(Assembler::notEqual, L);
1032 __ call_VM(noreg,
1033 CAST_FROM_FN_PTR(address,
1034 InterpreterRuntime::prepare_native_call),
1035 method);
1036 __ get_method(method);
1037 __ movptr(rax, Address(method, Method::native_function_offset()));
1038 __ bind(L);
1039 }
1040
1041 // pass JNIEnv
1042 #ifndef _LP64
1043 __ get_thread(thread);
1044 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1045 __ movptr(Address(rsp, 0), t);
1046
1047 // set_last_Java_frame_before_call
1048 // It is enough that the pc()
1049 // points into the right code segment. It does not have to be the correct return pc.
1050 __ set_last_Java_frame(thread, noreg, rbp, __ pc(), noreg);
1051 #else
1052 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1053
1054 // It is enough that the pc() points into the right code
1055 // segment. It does not have to be the correct return pc.
1056 __ set_last_Java_frame(rsp, rbp, (address) __ pc(), rscratch1);
1057 #endif // _LP64
1058
1059 // change thread state
1060 #ifdef ASSERT
1061 {
1062 Label L;
1063 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1064 __ cmpl(t, _thread_in_Java);
1065 __ jcc(Assembler::equal, L);
1066 __ stop("Wrong thread state in native stub");
1067 __ bind(L);
1068 }
1069 #endif
1070
1071 // Change state to native
1072
1073 __ movl(Address(thread, JavaThread::thread_state_offset()),
1074 _thread_in_native);
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 // Verify or restore cpu control state after JNI call
1082 __ restore_cpu_control_state_after_jni(rscratch1);
1083
1084 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1085 // in order to extract the result of a method call. If the order of these
1086 // pushes change or anything else is added to the stack then the code in
1087 // interpreter_frame_result must also change.
1088
1089 #ifndef _LP64
1090 // save potential result in ST(0) & rdx:rax
1091 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1092 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1093 // It is safe to do this push because state is _thread_in_native and return address will be found
1094 // via _last_native_pc and not via _last_jave_sp
1095
1096 // NOTE: the order of these push(es) is known to frame::interpreter_frame_result.
1097 // If the order changes or anything else is added to the stack the code in
1098 // interpreter_frame_result will have to be changed.
1099
1100 { Label L;
1101 Label push_double;
1102 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1103 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1104 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1105 float_handler.addr(), noreg);
1106 __ jcc(Assembler::equal, push_double);
1107 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1108 double_handler.addr(), noreg);
1109 __ jcc(Assembler::notEqual, L);
1110 __ bind(push_double);
1111 __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1112 __ bind(L);
1113 }
1114 #else
1115 __ push(dtos);
1116 #endif // _LP64
1117
1118 __ push(ltos);
1119
1120 // change thread state
1121 NOT_LP64(__ get_thread(thread));
1122 __ movl(Address(thread, JavaThread::thread_state_offset()),
1123 _thread_in_native_trans);
1124
1125 // Force this write out before the read below
1126 if (!UseSystemMemoryBarrier) {
1127 __ membar(Assembler::Membar_mask_bits(
1128 Assembler::LoadLoad | Assembler::LoadStore |
1129 Assembler::StoreLoad | Assembler::StoreStore));
1130 }
1131 #ifndef _LP64
1132 if (AlwaysRestoreFPU) {
1133 // Make sure the control word is correct.
1134 __ fldcw(ExternalAddress(StubRoutines::x86::addr_fpu_cntrl_wrd_std()));
1135 }
1136 #endif // _LP64
1137
1138 // check for safepoint operation in progress and/or pending suspend requests
1139 {
1140 Label Continue;
1141 Label slow_path;
1142
1143 __ safepoint_poll(slow_path, thread, true /* at_return */, false /* in_nmethod */);
1144
1145 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1146 __ jcc(Assembler::equal, Continue);
1147 __ bind(slow_path);
1148
1149 // Don't use call_VM as it will see a possible pending exception
1150 // and forward it and never return here preventing us from
1151 // clearing _last_native_pc down below. Also can't use
1152 // call_VM_leaf either as it will check to see if r13 & r14 are
1153 // preserved and correspond to the bcp/locals pointers. So we do a
1154 // runtime call by hand.
1155 //
1156 #ifndef _LP64
1157 __ push(thread);
1158 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1159 JavaThread::check_special_condition_for_native_trans)));
1160 __ increment(rsp, wordSize);
1161 __ get_thread(thread);
1162 #else
1163 __ mov(c_rarg0, r15_thread);
1164 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1165 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1166 __ andptr(rsp, -16); // align stack as required by ABI
1167 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1168 __ mov(rsp, r12); // restore sp
1169 __ reinit_heapbase();
1170 #endif // _LP64
1171 __ bind(Continue);
1172 }
1173
1174 // change thread state
1175 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1176
1177 // reset_last_Java_frame
1178 __ reset_last_Java_frame(thread, true);
1179
1180 if (CheckJNICalls) {
1181 // clear_pending_jni_exception_check
1182 __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1183 }
1184
1185 // reset handle block
1186 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1187 __ movl(Address(t, JNIHandleBlock::top_offset()), NULL_WORD);
1188
1189 // If result is an oop unbox and store it in frame where gc will see it
1190 // and result handler will pick it up
1191
1192 {
1193 Label no_oop;
1194 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1195 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1196 __ jcc(Assembler::notEqual, no_oop);
1197 // retrieve result
1198 __ pop(ltos);
1199 // Unbox oop result, e.g. JNIHandles::resolve value.
1200 __ resolve_jobject(rax /* value */,
1201 thread /* thread */,
1202 t /* tmp */);
1203 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1204 // keep stack depth as expected by pushing oop which will eventually be discarded
1205 __ push(ltos);
1206 __ bind(no_oop);
1207 }
1208
1209
1210 {
1211 Label no_reguard;
1212 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1213 StackOverflow::stack_guard_yellow_reserved_disabled);
1214 __ jcc(Assembler::notEqual, no_reguard);
1215
1216 __ pusha(); // XXX only save smashed registers
1217 #ifndef _LP64
1218 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1219 __ popa();
1220 #else
1221 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1222 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1223 __ andptr(rsp, -16); // align stack as required by ABI
1224 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1225 __ mov(rsp, r12); // restore sp
1226 __ popa(); // XXX only restore smashed registers
1227 __ reinit_heapbase();
1228 #endif // _LP64
1229
1230 __ bind(no_reguard);
1231 }
1232
1233
1234 // The method register is junk from after the thread_in_native transition
1235 // until here. Also can't call_VM until the bcp has been
1236 // restored. Need bcp for throwing exception below so get it now.
1237 __ get_method(method);
1238
1239 // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1240 __ movptr(rbcp, Address(method, Method::const_offset())); // get ConstMethod*
1241 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
1242
1243 // handle exceptions (exception handling will handle unlocking!)
1244 {
1245 Label L;
1246 __ cmpptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
1247 __ jcc(Assembler::zero, L);
1248 // Note: At some point we may want to unify this with the code
1249 // used in call_VM_base(); i.e., we should use the
1250 // StubRoutines::forward_exception code. For now this doesn't work
1251 // here because the rsp is not correctly set at this point.
1252 __ MacroAssembler::call_VM(noreg,
1253 CAST_FROM_FN_PTR(address,
1254 InterpreterRuntime::throw_pending_exception));
1255 __ should_not_reach_here();
1256 __ bind(L);
1257 }
1258
1259 // do unlocking if necessary
1260 {
1261 Label L;
1262 __ movl(t, Address(method, Method::access_flags_offset()));
1263 __ testl(t, JVM_ACC_SYNCHRONIZED);
1264 __ jcc(Assembler::zero, L);
1265 // the code below should be shared with interpreter macro
1266 // assembler implementation
1267 {
1268 Label unlock;
1269 // BasicObjectLock will be first in list, since this is a
1270 // synchronized method. However, need to check that the object
1271 // has not been unlocked by an explicit monitorexit bytecode.
1272 const Address monitor(rbp,
1273 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1274 wordSize - (int)sizeof(BasicObjectLock)));
1275
1276 const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1277
1278 // monitor expect in c_rarg1 for slow unlock path
1279 __ lea(regmon, monitor); // address of first monitor
1280
1281 __ movptr(t, Address(regmon, BasicObjectLock::obj_offset()));
1282 __ testptr(t, t);
1283 __ jcc(Assembler::notZero, unlock);
1284
1285 // Entry already unlocked, need to throw exception
1286 __ MacroAssembler::call_VM(noreg,
1287 CAST_FROM_FN_PTR(address,
1288 InterpreterRuntime::throw_illegal_monitor_state_exception));
1289 __ should_not_reach_here();
1290
1291 __ bind(unlock);
1292 __ unlock_object(regmon);
1293 }
1294 __ bind(L);
1295 }
1296
1297 // jvmti support
1298 // Note: This must happen _after_ handling/throwing any exceptions since
1299 // the exception handler code notifies the runtime of method exits
1300 // too. If this happens before, method entry/exit notifications are
1301 // not properly paired (was bug - gri 11/22/99).
1302 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1303
1304 // restore potential result in edx:eax, call result handler to
1305 // restore potential result in ST0 & handle result
1306
1307 __ pop(ltos);
1308 LP64_ONLY( __ pop(dtos));
1309
1310 __ movptr(t, Address(rbp,
1311 (frame::interpreter_frame_result_handler_offset) * wordSize));
1312 __ call(t);
1313
1314 // remove activation
1315 __ movptr(t, Address(rbp,
1316 frame::interpreter_frame_sender_sp_offset *
1317 wordSize)); // get sender sp
1318 __ leave(); // remove frame anchor
1319 __ pop(rdi); // get return address
1320 __ mov(rsp, t); // set sp to sender sp
1321 __ jmp(rdi);
1322
1323 if (inc_counter) {
1324 // Handle overflow of counter and compile method
1325 __ bind(invocation_counter_overflow);
1326 generate_counter_overflow(continue_after_compile);
1327 }
1328
1329 return entry_point;
1330 }
1331
1332 // Abstract method entry
1333 // Attempt to execute abstract method. Throw exception
1334 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1335
1336 address entry_point = __ pc();
1337
1338 // abstract method entry
1339
1340 // pop return address, reset last_sp to null
1341 __ empty_expression_stack();
1342 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1343 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1344
1345 // throw exception
1346 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod), rbx);
1347 // the call_VM checks for exception, so we should never return here.
1348 __ should_not_reach_here();
1349
1350 return entry_point;
1351 }
1352
1353 //
1354 // Generic interpreted method entry to (asm) interpreter
1355 //
1356 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1357 // determine code generation flags
1358 bool inc_counter = UseCompiler || CountCompiledCalls;
1359
1360 // ebx: Method*
1361 // rbcp: sender sp (set in InterpreterMacroAssembler::prepare_to_jump_from_interpreted / generate_call_stub)
1362 address entry_point = __ pc();
1363
1364 const Address constMethod(rbx, Method::const_offset());
1365 const Address access_flags(rbx, Method::access_flags_offset());
1366 const Address size_of_parameters(rdx,
1367 ConstMethod::size_of_parameters_offset());
1368 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1369
1370
1371 // get parameter size (always needed)
1372 __ movptr(rdx, constMethod);
1373 __ load_unsigned_short(rcx, size_of_parameters);
1374
1375 // rbx: Method*
1376 // rcx: size of parameters
1377 // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1378
1379 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1380 __ subl(rdx, rcx); // rdx = no. of additional locals
1381
1382 // YYY
1383 // __ incrementl(rdx);
1384 // __ andl(rdx, -2);
1385
1386 // see if we've got enough room on the stack for locals plus overhead.
1387 generate_stack_overflow_check();
1388
1389 // get return address
1390 __ pop(rax);
1391
1392 // compute beginning of parameters
1393 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1394
1395 // rdx - # of additional locals
1396 // allocate space for locals
1397 // explicitly initialize locals
1398 {
1399 Label exit, loop;
1400 __ testl(rdx, rdx);
1401 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1402 __ bind(loop);
1403 __ push(NULL_WORD); // initialize local variables
1404 __ decrementl(rdx); // until everything initialized
1405 __ jcc(Assembler::greater, loop);
1406 __ bind(exit);
1407 }
1408
1409 // initialize fixed part of activation frame
1410 generate_fixed_frame(false);
1411
1412 // make sure method is not native & not abstract
1413 #ifdef ASSERT
1414 __ movl(rax, access_flags);
1415 {
1416 Label L;
1417 __ testl(rax, JVM_ACC_NATIVE);
1418 __ jcc(Assembler::zero, L);
1419 __ stop("tried to execute native method as non-native");
1420 __ bind(L);
1421 }
1422 {
1423 Label L;
1424 __ testl(rax, JVM_ACC_ABSTRACT);
1425 __ jcc(Assembler::zero, L);
1426 __ stop("tried to execute abstract method in interpreter");
1427 __ bind(L);
1428 }
1429 #endif
1430
1431 // Since at this point in the method invocation the exception
1432 // handler would try to exit the monitor of synchronized methods
1433 // which hasn't been entered yet, we set the thread local variable
1434 // _do_not_unlock_if_synchronized to true. The remove_activation
1435 // will check this flag.
1436
1437 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1438 NOT_LP64(__ get_thread(thread));
1439 const Address do_not_unlock_if_synchronized(thread,
1440 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1441 __ movbool(do_not_unlock_if_synchronized, true);
1442
1443 __ profile_parameters_type(rax, rcx, rdx);
1444 // increment invocation count & check for overflow
1445 Label invocation_counter_overflow;
1446 if (inc_counter) {
1447 generate_counter_incr(&invocation_counter_overflow);
1448 }
1449
1450 Label continue_after_compile;
1451 __ bind(continue_after_compile);
1452
1453 // check for synchronized interpreted methods
1454 bang_stack_shadow_pages(false);
1455
1456 // reset the _do_not_unlock_if_synchronized flag
1457 NOT_LP64(__ get_thread(thread));
1458 __ movbool(do_not_unlock_if_synchronized, false);
1459
1460 // check for synchronized methods
1461 // Must happen AFTER invocation_counter check and stack overflow check,
1462 // so method is not locked if overflows.
1463 if (synchronized) {
1464 // Allocate monitor and lock method
1465 lock_method();
1466 } else {
1467 // no synchronization necessary
1468 #ifdef ASSERT
1469 {
1470 Label L;
1471 __ movl(rax, access_flags);
1472 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1473 __ jcc(Assembler::zero, L);
1474 __ stop("method needs synchronization");
1475 __ bind(L);
1476 }
1477 #endif
1478 }
1479
1480 // start execution
1481 #ifdef ASSERT
1482 {
1483 Label L;
1484 const Address monitor_block_top (rbp,
1485 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1486 __ movptr(rax, monitor_block_top);
1487 __ lea(rax, Address(rbp, rax, Address::times_ptr));
1488 __ cmpptr(rax, rsp);
1489 __ jcc(Assembler::equal, L);
1490 __ stop("broken stack frame setup in interpreter 6");
1491 __ bind(L);
1492 }
1493 #endif
1494
1495 // jvmti support
1496 __ notify_method_entry();
1497
1498 __ dispatch_next(vtos);
1499
1500 // invocation counter overflow
1501 if (inc_counter) {
1502 // Handle overflow of counter and compile method
1503 __ bind(invocation_counter_overflow);
1504 generate_counter_overflow(continue_after_compile);
1505 }
1506
1507 return entry_point;
1508 }
1509
1510 //-----------------------------------------------------------------------------
1511 // Exceptions
1512
1513 void TemplateInterpreterGenerator::generate_throw_exception() {
1514 // Entry point in previous activation (i.e., if the caller was
1515 // interpreted)
1516 Interpreter::_rethrow_exception_entry = __ pc();
1517 // Restore sp to interpreter_frame_last_sp even though we are going
1518 // to empty the expression stack for the exception processing.
1519 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1520 // rax: exception
1521 // rdx: return address/pc that threw exception
1522 __ restore_bcp(); // r13/rsi points to call/send
1523 __ restore_locals();
1524 LP64_ONLY(__ reinit_heapbase()); // restore r12 as heapbase.
1525 // Entry point for exceptions thrown within interpreter code
1526 Interpreter::_throw_exception_entry = __ pc();
1527 // expression stack is undefined here
1528 // rax: exception
1529 // r13/rsi: exception bcp
1530 __ verify_oop(rax);
1531 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1532 LP64_ONLY(__ mov(c_rarg1, rax));
1533
1534 // expression stack must be empty before entering the VM in case of
1535 // an exception
1536 __ empty_expression_stack();
1537 // find exception handler address and preserve exception oop
1538 __ call_VM(rdx,
1539 CAST_FROM_FN_PTR(address,
1540 InterpreterRuntime::exception_handler_for_exception),
1541 rarg);
1542 // rax: exception handler entry point
1543 // rdx: preserved exception oop
1544 // r13/rsi: bcp for exception handler
1545 __ push_ptr(rdx); // push exception which is now the only value on the stack
1546 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1547
1548 // If the exception is not handled in the current frame the frame is
1549 // removed and the exception is rethrown (i.e. exception
1550 // continuation is _rethrow_exception).
1551 //
1552 // Note: At this point the bci is still the bxi for the instruction
1553 // which caused the exception and the expression stack is
1554 // empty. Thus, for any VM calls at this point, GC will find a legal
1555 // oop map (with empty expression stack).
1556
1557 // In current activation
1558 // tos: exception
1559 // esi: exception bcp
1560
1561 //
1562 // JVMTI PopFrame support
1563 //
1564
1565 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1566 __ empty_expression_stack();
1567 // Set the popframe_processing bit in pending_popframe_condition
1568 // indicating that we are currently handling popframe, so that
1569 // call_VMs that may happen later do not trigger new popframe
1570 // handling cycles.
1571 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1572 NOT_LP64(__ get_thread(thread));
1573 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1574 __ orl(rdx, JavaThread::popframe_processing_bit);
1575 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1576
1577 {
1578 // Check to see whether we are returning to a deoptimized frame.
1579 // (The PopFrame call ensures that the caller of the popped frame is
1580 // either interpreted or compiled and deoptimizes it if compiled.)
1581 // In this case, we can't call dispatch_next() after the frame is
1582 // popped, but instead must save the incoming arguments and restore
1583 // them after deoptimization has occurred.
1584 //
1585 // Note that we don't compare the return PC against the
1586 // deoptimization blob's unpack entry because of the presence of
1587 // adapter frames in C2.
1588 Label caller_not_deoptimized;
1589 Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1590 __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1591 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1592 InterpreterRuntime::interpreter_contains), rarg);
1593 __ testl(rax, rax);
1594 __ jcc(Assembler::notZero, caller_not_deoptimized);
1595
1596 // Compute size of arguments for saving when returning to
1597 // deoptimized caller
1598 __ get_method(rax);
1599 __ movptr(rax, Address(rax, Method::const_offset()));
1600 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1601 size_of_parameters_offset())));
1602 __ shll(rax, Interpreter::logStackElementSize);
1603 __ restore_locals();
1604 __ subptr(rlocals, rax);
1605 __ addptr(rlocals, wordSize);
1606 // Save these arguments
1607 NOT_LP64(__ get_thread(thread));
1608 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1609 Deoptimization::
1610 popframe_preserve_args),
1611 thread, rax, rlocals);
1612
1613 __ remove_activation(vtos, rdx,
1614 /* throw_monitor_exception */ false,
1615 /* install_monitor_exception */ false,
1616 /* notify_jvmdi */ false);
1617
1618 // Inform deoptimization that it is responsible for restoring
1619 // these arguments
1620 NOT_LP64(__ get_thread(thread));
1621 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1622 JavaThread::popframe_force_deopt_reexecution_bit);
1623
1624 // Continue in deoptimization handler
1625 __ jmp(rdx);
1626
1627 __ bind(caller_not_deoptimized);
1628 }
1629
1630 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1631 /* throw_monitor_exception */ false,
1632 /* install_monitor_exception */ false,
1633 /* notify_jvmdi */ false);
1634
1635 // Finish with popframe handling
1636 // A previous I2C followed by a deoptimization might have moved the
1637 // outgoing arguments further up the stack. PopFrame expects the
1638 // mutations to those outgoing arguments to be preserved and other
1639 // constraints basically require this frame to look exactly as
1640 // though it had previously invoked an interpreted activation with
1641 // no space between the top of the expression stack (current
1642 // last_sp) and the top of stack. Rather than force deopt to
1643 // maintain this kind of invariant all the time we call a small
1644 // fixup routine to move the mutated arguments onto the top of our
1645 // expression stack if necessary.
1646 #ifndef _LP64
1647 __ mov(rax, rsp);
1648 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1649 __ lea(rbx, Address(rbp, rbx, Address::times_ptr));
1650 __ get_thread(thread);
1651 // PC must point into interpreter here
1652 __ set_last_Java_frame(thread, noreg, rbp, __ pc(), noreg);
1653 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1654 __ get_thread(thread);
1655 #else
1656 __ mov(c_rarg1, rsp);
1657 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1658 __ lea(c_rarg2, Address(rbp, c_rarg2, Address::times_ptr));
1659 // PC must point into interpreter here
1660 __ set_last_Java_frame(noreg, rbp, __ pc(), rscratch1);
1661 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1662 #endif
1663 __ reset_last_Java_frame(thread, true);
1664
1665 // Restore the last_sp and null it out
1666 __ movptr(rcx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1667 __ lea(rsp, Address(rbp, rcx, Address::times_ptr));
1668 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1669
1670 __ restore_bcp();
1671 __ restore_locals();
1672 // The method data pointer was incremented already during
1673 // call profiling. We have to restore the mdp for the current bcp.
1674 if (ProfileInterpreter) {
1675 __ set_method_data_pointer_for_bcp();
1676 }
1677
1678 // Clear the popframe condition flag
1679 NOT_LP64(__ get_thread(thread));
1680 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1681 JavaThread::popframe_inactive);
1682
1683 #if INCLUDE_JVMTI
1684 {
1685 Label L_done;
1686 const Register local0 = rlocals;
1687
1688 __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1689 __ jcc(Assembler::notEqual, L_done);
1690
1691 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1692 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
1693
1694 __ get_method(rdx);
1695 __ movptr(rax, Address(local0, 0));
1696 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1697
1698 __ testptr(rax, rax);
1699 __ jcc(Assembler::zero, L_done);
1700
1701 __ movptr(Address(rbx, 0), rax);
1702 __ bind(L_done);
1703 }
1704 #endif // INCLUDE_JVMTI
1705
1706 __ dispatch_next(vtos);
1707 // end of PopFrame support
1708
1709 Interpreter::_remove_activation_entry = __ pc();
1710
1711 // preserve exception over this code sequence
1712 __ pop_ptr(rax);
1713 NOT_LP64(__ get_thread(thread));
1714 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1715 // remove the activation (without doing throws on illegalMonitorExceptions)
1716 __ remove_activation(vtos, rdx, false, true, false);
1717 // restore exception
1718 NOT_LP64(__ get_thread(thread));
1719 __ get_vm_result(rax, thread);
1720
1721 // In between activations - previous activation type unknown yet
1722 // compute continuation point - the continuation point expects the
1723 // following registers set up:
1724 //
1725 // rax: exception
1726 // rdx: return address/pc that threw exception
1727 // rsp: expression stack of caller
1728 // rbp: ebp of caller
1729 __ push(rax); // save exception
1730 __ push(rdx); // save return address
1731 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1732 SharedRuntime::exception_handler_for_return_address),
1733 thread, rdx);
1734 __ mov(rbx, rax); // save exception handler
1735 __ pop(rdx); // restore return address
1736 __ pop(rax); // restore exception
1737 // Note that an "issuing PC" is actually the next PC after the call
1738 __ jmp(rbx); // jump to exception
1739 // handler of caller
1740 }
1741
1742
1743 //
1744 // JVMTI ForceEarlyReturn support
1745 //
1746 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1747 address entry = __ pc();
1748
1749 __ restore_bcp();
1750 __ restore_locals();
1751 __ empty_expression_stack();
1752 __ load_earlyret_value(state); // 32 bits returns value in rdx, so don't reuse
1753
1754 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1755 NOT_LP64(__ get_thread(thread));
1756 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1757 Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1758
1759 // Clear the earlyret state
1760 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1761
1762 __ remove_activation(state, rsi,
1763 false, /* throw_monitor_exception */
1764 false, /* install_monitor_exception */
1765 true); /* notify_jvmdi */
1766 __ jmp(rsi);
1767
1768 return entry;
1769 } // end of ForceEarlyReturn support
1770
1771
1772 //-----------------------------------------------------------------------------
1773 // Helper for vtos entry point generation
1774
1775 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1776 address& bep,
1777 address& cep,
1778 address& sep,
1779 address& aep,
1780 address& iep,
1781 address& lep,
1782 address& fep,
1783 address& dep,
1784 address& vep) {
1785 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1786 Label L;
1787 #ifndef _LP64
1788 fep = __ pc(); // ftos entry point
1789 __ push(ftos);
1790 __ jmpb(L);
1791 dep = __ pc(); // dtos entry point
1792 __ push(dtos);
1793 __ jmpb(L);
1794 #else
1795 fep = __ pc(); // ftos entry point
1796 __ push_f(xmm0);
1797 __ jmpb(L);
1798 dep = __ pc(); // dtos entry point
1799 __ push_d(xmm0);
1800 __ jmpb(L);
1801 #endif // _LP64
1802 lep = __ pc(); // ltos entry point
1803 __ push_l();
1804 __ jmpb(L);
1805 aep = bep = cep = sep = iep = __ pc(); // [abcsi]tos entry point
1806 __ push_i_or_ptr();
1807 vep = __ pc(); // vtos entry point
1808 __ bind(L);
1809 generate_and_dispatch(t);
1810 }
1811
1812 //-----------------------------------------------------------------------------
1813
1814 // Non-product code
1815 #ifndef PRODUCT
1816
1817 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1818 address entry = __ pc();
1819
1820 #ifndef _LP64
1821 // prepare expression stack
1822 __ pop(rcx); // pop return address so expression stack is 'pure'
1823 __ push(state); // save tosca
1824
1825 // pass tosca registers as arguments & call tracer
1826 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1827 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1828 __ pop(state); // restore tosca
1829
1830 // return
1831 __ jmp(rcx);
1832 #else
1833 __ push(state);
1834 __ push(c_rarg0);
1835 __ push(c_rarg1);
1836 __ push(c_rarg2);
1837 __ push(c_rarg3);
1838 __ mov(c_rarg2, rax); // Pass itos
1839 #ifdef _WIN64
1840 __ movflt(xmm3, xmm0); // Pass ftos
1841 #endif
1842 __ call_VM(noreg,
1843 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1844 c_rarg1, c_rarg2, c_rarg3);
1845 __ pop(c_rarg3);
1846 __ pop(c_rarg2);
1847 __ pop(c_rarg1);
1848 __ pop(c_rarg0);
1849 __ pop(state);
1850 __ ret(0); // return from result handler
1851 #endif // _LP64
1852
1853 return entry;
1854 }
1855
1856 void TemplateInterpreterGenerator::count_bytecode() {
1857 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value), rscratch1);
1858 }
1859
1860 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1861 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]), rscratch1);
1862 }
1863
1864 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1865 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1866 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1867 __ orl(rbx,
1868 ((int) t->bytecode()) <<
1869 BytecodePairHistogram::log2_number_of_codes);
1870 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx, rscratch1);
1871 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1872 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1873 }
1874
1875
1876 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1877 // Call a little run-time stub to avoid blow-up for each bytecode.
1878 // The run-time runtime saves the right registers, depending on
1879 // the tosca in-state for the given template.
1880
1881 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
1882 "entry must have been generated");
1883 #ifndef _LP64
1884 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1885 #else
1886 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1887 __ andptr(rsp, -16); // align stack as required by ABI
1888 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1889 __ mov(rsp, r12); // restore sp
1890 __ reinit_heapbase();
1891 #endif // _LP64
1892 }
1893
1894
1895 void TemplateInterpreterGenerator::stop_interpreter_at() {
1896 Label L;
1897 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1898 StopInterpreterAt,
1899 rscratch1);
1900 __ jcc(Assembler::notEqual, L);
1901 __ int3();
1902 __ bind(L);
1903 }
1904 #endif // !PRODUCT