1 /*
2 * Copyright (c) 2003, 2024, 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(RuntimeAddress(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_resume_interpreter_adapter() {
391 if (!Continuations::enabled()) return nullptr;
392 address start = __ pc();
393
394 __ restore_bcp();
395 __ restore_locals();
396
397 // Get return address before adjusting rsp
398 __ movptr(rax, Address(rsp, 0));
399
400 // Restore stack bottom
401 __ movptr(rcx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
402 __ lea(rsp, Address(rbp, rcx, Address::times_ptr));
403 // and NULL it as marker that esp is now tos until next java call
404 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
405
406 __ jmp(rax);
407
408 return start;
409 }
410
411
412 // Helpers for commoning out cases in the various type of method entries.
413 //
414
415
416 // increment invocation count & check for overflow
417 //
418 // Note: checking for negative value instead of overflow
419 // so we have a 'sticky' overflow test
420 //
421 // rbx: method
422 // rcx: invocation counter
423 //
424 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
425 Label done;
426 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
427 Label no_mdo;
428 if (ProfileInterpreter) {
429 // Are we profiling?
430 __ movptr(rax, Address(rbx, Method::method_data_offset()));
431 __ testptr(rax, rax);
432 __ jccb(Assembler::zero, no_mdo);
433 // Increment counter in the MDO
434 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
435 in_bytes(InvocationCounter::counter_offset()));
436 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
437 __ increment_mask_and_jump(mdo_invocation_counter, mask, rcx, overflow);
438 __ jmp(done);
439 }
440 __ bind(no_mdo);
441 // Increment counter in MethodCounters
442 const Address invocation_counter(rax,
443 MethodCounters::invocation_counter_offset() +
444 InvocationCounter::counter_offset());
445 __ get_method_counters(rbx, rax, done);
446 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
447 __ increment_mask_and_jump(invocation_counter, mask, rcx, overflow);
448 __ bind(done);
449 }
450
451 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
452
453 // Asm interpreter on entry
454 // r14/rdi - locals
455 // r13/rsi - bcp
456 // rbx - method
457 // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
458 // rbp - interpreter frame
459
460 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
461 // Everything as it was on entry
462 // rdx is not restored. Doesn't appear to really be set.
463
464 // InterpreterRuntime::frequency_counter_overflow takes two
465 // arguments, the first (thread) is passed by call_VM, the second
466 // indicates if the counter overflow occurs at a backwards branch
467 // (null bcp). We pass zero for it. The call returns the address
468 // of the verified entry point for the method or null if the
469 // compilation did not complete (either went background or bailed
470 // out).
471 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
472 __ movl(rarg, 0);
473 __ call_VM(noreg,
474 CAST_FROM_FN_PTR(address,
475 InterpreterRuntime::frequency_counter_overflow),
476 rarg);
477
478 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
479 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
480 // and jump to the interpreted entry.
481 __ jmp(do_continue, relocInfo::none);
482 }
483
484 // See if we've got enough room on the stack for locals plus overhead below
485 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
486 // without going through the signal handler, i.e., reserved and yellow zones
487 // will not be made usable. The shadow zone must suffice to handle the
488 // overflow.
489 // The expression stack grows down incrementally, so the normal guard
490 // page mechanism will work for that.
491 //
492 // NOTE: Since the additional locals are also always pushed (wasn't
493 // obvious in generate_fixed_frame) so the guard should work for them
494 // too.
495 //
496 // Args:
497 // rdx: number of additional locals this frame needs (what we must check)
498 // rbx: Method*
499 //
500 // Kills:
501 // rax
502 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
503
504 // monitor entry size: see picture of stack in frame_x86.hpp
505 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
506
507 // total overhead size: entry_size + (saved rbp through expr stack
508 // bottom). be sure to change this if you add/subtract anything
509 // to/from the overhead area
510 const int overhead_size =
511 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
512
513 const int page_size = (int)os::vm_page_size();
514
515 Label after_frame_check;
516
517 // see if the frame is greater than one page in size. If so,
518 // then we need to verify there is enough stack space remaining
519 // for the additional locals.
520 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
521 __ jcc(Assembler::belowEqual, after_frame_check);
522
523 // compute rsp as if this were going to be the last frame on
524 // the stack before the red zone
525
526 Label after_frame_check_pop;
527 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
528 #ifndef _LP64
529 __ push(thread);
530 __ get_thread(thread);
531 #endif
532
533 const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
534
535 // locals + overhead, in bytes
536 __ mov(rax, rdx);
537 __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
538 __ addptr(rax, overhead_size);
539
540 #ifdef ASSERT
541 Label limit_okay;
542 // Verify that thread stack overflow limit is non-zero.
543 __ cmpptr(stack_limit, NULL_WORD);
544 __ jcc(Assembler::notEqual, limit_okay);
545 __ stop("stack overflow limit is zero");
546 __ bind(limit_okay);
547 #endif
548
549 // Add locals/frame size to stack limit.
550 __ addptr(rax, stack_limit);
551
552 // Check against the current stack bottom.
553 __ cmpptr(rsp, rax);
554
555 __ jcc(Assembler::above, after_frame_check_pop);
556 NOT_LP64(__ pop(rsi)); // get saved bcp
557
558 // Restore sender's sp as SP. This is necessary if the sender's
559 // frame is an extended compiled frame (see gen_c2i_adapter())
560 // and safer anyway in case of JSR292 adaptations.
561
562 __ pop(rax); // return address must be moved if SP is changed
563 __ mov(rsp, rbcp);
564 __ push(rax);
565
566 // Note: the restored frame is not necessarily interpreted.
567 // Use the shared runtime version of the StackOverflowError.
568 assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
569 __ jump(RuntimeAddress(SharedRuntime::throw_StackOverflowError_entry()));
570 // all done with frame size check
571 __ bind(after_frame_check_pop);
572 NOT_LP64(__ pop(rsi));
573
574 // all done with frame size check
575 __ bind(after_frame_check);
576 }
577
578 // Allocate monitor and lock method (asm interpreter)
579 //
580 // Args:
581 // rbx: Method*
582 // r14/rdi: locals
583 //
584 // Kills:
585 // rax
586 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
587 // rscratch1, rscratch2 (scratch regs)
588 void TemplateInterpreterGenerator::lock_method() {
589 // synchronize method
590 const Address access_flags(rbx, Method::access_flags_offset());
591 const Address monitor_block_top(
592 rbp,
593 frame::interpreter_frame_monitor_block_top_offset * wordSize);
594 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
595
596 #ifdef ASSERT
597 {
598 Label L;
599 __ movl(rax, access_flags);
600 __ testl(rax, JVM_ACC_SYNCHRONIZED);
601 __ jcc(Assembler::notZero, L);
602 __ stop("method doesn't need synchronization");
603 __ bind(L);
604 }
605 #endif // ASSERT
606
607 // get synchronization object
608 {
609 Label done;
610 __ movl(rax, access_flags);
611 __ testl(rax, JVM_ACC_STATIC);
612 // get receiver (assume this is frequent case)
613 __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
614 __ jcc(Assembler::zero, done);
615 __ load_mirror(rax, rbx, rscratch2);
616
617 #ifdef ASSERT
618 {
619 Label L;
620 __ testptr(rax, rax);
621 __ jcc(Assembler::notZero, L);
622 __ stop("synchronization object is null");
623 __ bind(L);
624 }
625 #endif // ASSERT
626
627 __ bind(done);
628 }
629
630 // add space for monitor & lock
631 __ subptr(rsp, entry_size); // add space for a monitor entry
632 __ subptr(monitor_block_top, entry_size / wordSize); // set new monitor block top
633 // store object
634 __ movptr(Address(rsp, BasicObjectLock::obj_offset()), rax);
635 const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
636 __ movptr(lockreg, rsp); // object address
637 __ lock_object(lockreg);
638 }
639
640 // Generate a fixed interpreter frame. This is identical setup for
641 // interpreted methods and for native methods hence the shared code.
642 //
643 // Args:
644 // rax: return address
645 // rbx: Method*
646 // r14/rdi: pointer to locals
647 // r13/rsi: sender sp
648 // rdx: cp cache
649 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
650 // initialize fixed part of activation frame
651 __ push(rax); // save return address
652 __ enter(); // save old & set new rbp
653 __ push(rbcp); // set sender sp
654 __ push(NULL_WORD); // leave last_sp as null
655 __ movptr(rbcp, Address(rbx, Method::const_offset())); // get ConstMethod*
656 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
657 __ push(rbx); // save Method*
658 // Get mirror and store it in the frame as GC root for this Method*
659 __ load_mirror(rdx, rbx, rscratch2);
660 __ push(rdx);
661 if (ProfileInterpreter) {
662 Label method_data_continue;
663 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
664 __ testptr(rdx, rdx);
665 __ jcc(Assembler::zero, method_data_continue);
666 __ addptr(rdx, in_bytes(MethodData::data_offset()));
667 __ bind(method_data_continue);
668 __ push(rdx); // set the mdp (method data pointer)
669 } else {
670 __ push(0);
671 }
672
673 __ movptr(rdx, Address(rbx, Method::const_offset()));
674 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
675 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset()));
676 __ push(rdx); // set constant pool cache
677
678 __ movptr(rax, rlocals);
679 __ subptr(rax, rbp);
680 __ shrptr(rax, Interpreter::logStackElementSize); // rax = rlocals - fp();
681 __ push(rax); // set relativized rlocals, see frame::interpreter_frame_locals()
682
683 if (native_call) {
684 __ push(0); // no bcp
685 } else {
686 __ push(rbcp); // set bcp
687 }
688 // initialize relativized pointer to expression stack bottom
689 __ push(frame::interpreter_frame_initial_sp_offset);
690 }
691
692 // End of helpers
693
694 // Method entry for java.lang.ref.Reference.get.
695 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
696 // Code: _aload_0, _getfield, _areturn
697 // parameter size = 1
698 //
699 // The code that gets generated by this routine is split into 2 parts:
700 // 1. The "intrinsified" code performing an ON_WEAK_OOP_REF load,
701 // 2. The slow path - which is an expansion of the regular method entry.
702 //
703 // Notes:-
704 // * An intrinsic is always executed, where an ON_WEAK_OOP_REF load is performed.
705 // * We may jump to the slow path iff the receiver is null. If the
706 // Reference object is null then we no longer perform an ON_WEAK_OOP_REF load
707 // Thus we can use the regular method entry code to generate the NPE.
708 //
709 // rbx: Method*
710
711 // r13: senderSP must preserve for slow path, set SP to it on fast path
712
713 address entry = __ pc();
714
715 const int referent_offset = java_lang_ref_Reference::referent_offset();
716
717 Label slow_path;
718 // rbx: method
719
720 // Check if local 0 != null
721 // If the receiver is null then it is OK to jump to the slow path.
722 __ movptr(rax, Address(rsp, wordSize));
723
724 __ testptr(rax, rax);
725 __ jcc(Assembler::zero, slow_path);
726
727 // rax: local 0
728 // rbx: method (but can be used as scratch now)
729 // rdx: scratch
730 // rdi: scratch
731
732 // Preserve the sender sp in case the load barrier
733 // calls the runtime
734 NOT_LP64(__ push(rsi));
735
736 // Load the value of the referent field.
737 const Address field_address(rax, referent_offset);
738 __ load_heap_oop(rax, field_address, /*tmp1*/ rbx, /*tmp_thread*/ rdx, ON_WEAK_OOP_REF);
739
740 // _areturn
741 const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
742 NOT_LP64(__ pop(rsi)); // get sender sp
743 __ pop(rdi); // get return address
744 __ mov(rsp, sender_sp); // set sp to sender sp
745 __ jmp(rdi);
746 __ ret(0);
747
748 // generate a vanilla interpreter entry as the slow path
749 __ bind(slow_path);
750 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
751 return entry;
752 }
753
754 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
755 // See more discussion in stackOverflow.hpp.
756
757 // Note that we do the banging after the frame is setup, since the exception
758 // handling code expects to find a valid interpreter frame on the stack.
759 // Doing the banging earlier fails if the caller frame is not an interpreter
760 // frame.
761 // (Also, the exception throwing code expects to unlock any synchronized
762 // method receiver, so do the banging after locking the receiver.)
763
764 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
765 const int page_size = (int)os::vm_page_size();
766 const int n_shadow_pages = shadow_zone_size / page_size;
767
768 const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
769 #ifndef _LP64
770 __ push(thread);
771 __ get_thread(thread);
772 #endif
773
774 #ifdef ASSERT
775 Label L_good_limit;
776 __ cmpptr(Address(thread, JavaThread::shadow_zone_safe_limit()), NULL_WORD);
777 __ jcc(Assembler::notEqual, L_good_limit);
778 __ stop("shadow zone safe limit is not initialized");
779 __ bind(L_good_limit);
780
781 Label L_good_watermark;
782 __ cmpptr(Address(thread, JavaThread::shadow_zone_growth_watermark()), NULL_WORD);
783 __ jcc(Assembler::notEqual, L_good_watermark);
784 __ stop("shadow zone growth watermark is not initialized");
785 __ bind(L_good_watermark);
786 #endif
787
788 Label L_done;
789
790 __ cmpptr(rsp, Address(thread, JavaThread::shadow_zone_growth_watermark()));
791 __ jcc(Assembler::above, L_done);
792
793 for (int p = 1; p <= n_shadow_pages; p++) {
794 __ bang_stack_with_offset(p*page_size);
795 }
796
797 // Record the new watermark, but only if update is above the safe limit.
798 // Otherwise, the next time around the check above would pass the safe limit.
799 __ cmpptr(rsp, Address(thread, JavaThread::shadow_zone_safe_limit()));
800 __ jccb(Assembler::belowEqual, L_done);
801 __ movptr(Address(thread, JavaThread::shadow_zone_growth_watermark()), rsp);
802
803 __ bind(L_done);
804
805 #ifndef _LP64
806 __ pop(thread);
807 #endif
808 }
809
810 // Interpreter stub for calling a native method. (asm interpreter)
811 // This sets up a somewhat different looking stack for calling the
812 // native method than the typical interpreter frame setup.
813 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized, bool runtime_upcalls) {
814 // determine code generation flags
815 bool inc_counter = UseCompiler || CountCompiledCalls;
816
817 // rbx: Method*
818 // rbcp: sender sp
819
820 address entry_point = __ pc();
821
822 const Address constMethod (rbx, Method::const_offset());
823 const Address access_flags (rbx, Method::access_flags_offset());
824 const Address size_of_parameters(rcx, ConstMethod::
825 size_of_parameters_offset());
826
827
828 // get parameter size (always needed)
829 __ movptr(rcx, constMethod);
830 __ load_unsigned_short(rcx, size_of_parameters);
831
832 // native calls don't need the stack size check since they have no
833 // expression stack and the arguments are already on the stack and
834 // we only add a handful of words to the stack
835
836 // rbx: Method*
837 // rcx: size of parameters
838 // rbcp: sender sp
839 __ pop(rax); // get return address
840
841 // for natives the size of locals is zero
842
843 // compute beginning of parameters
844 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
845
846 // add 2 zero-initialized slots for native calls
847 // initialize result_handler slot
848 __ push(NULL_WORD);
849 // slot for oop temp
850 // (static native method holder mirror/jni oop result)
851 __ push(NULL_WORD);
852
853 // initialize fixed part of activation frame
854 generate_fixed_frame(true);
855
856 // make sure method is native & not abstract
857 #ifdef ASSERT
858 __ movl(rax, access_flags);
859 {
860 Label L;
861 __ testl(rax, JVM_ACC_NATIVE);
862 __ jcc(Assembler::notZero, L);
863 __ stop("tried to execute non-native method as native");
864 __ bind(L);
865 }
866 {
867 Label L;
868 __ testl(rax, JVM_ACC_ABSTRACT);
869 __ jcc(Assembler::zero, L);
870 __ stop("tried to execute abstract method in interpreter");
871 __ bind(L);
872 }
873 #endif
874
875 // Since at this point in the method invocation the exception handler
876 // would try to exit the monitor of synchronized methods which hasn't
877 // been entered yet, we set the thread local variable
878 // _do_not_unlock_if_synchronized to true. The remove_activation will
879 // check this flag.
880
881 const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
882 NOT_LP64(__ get_thread(thread1));
883 const Address do_not_unlock_if_synchronized(thread1,
884 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
885 __ movbool(do_not_unlock_if_synchronized, true);
886
887 // increment invocation count & check for overflow
888 Label invocation_counter_overflow;
889 if (inc_counter) {
890 generate_counter_incr(&invocation_counter_overflow);
891 }
892
893 Label continue_after_compile;
894 __ bind(continue_after_compile);
895
896 bang_stack_shadow_pages(true);
897
898 // reset the _do_not_unlock_if_synchronized flag
899 NOT_LP64(__ get_thread(thread1));
900 __ movbool(do_not_unlock_if_synchronized, false);
901
902 // check for synchronized methods
903 // Must happen AFTER invocation_counter check and stack overflow check,
904 // so method is not locked if overflows.
905 if (synchronized) {
906 lock_method();
907 } else {
908 // no synchronization necessary
909 #ifdef ASSERT
910 {
911 Label L;
912 __ movl(rax, access_flags);
913 __ testl(rax, JVM_ACC_SYNCHRONIZED);
914 __ jcc(Assembler::zero, L);
915 __ stop("method needs synchronization");
916 __ bind(L);
917 }
918 #endif
919 }
920
921 // start execution
922 #ifdef ASSERT
923 {
924 Label L;
925 const Address monitor_block_top(rbp,
926 frame::interpreter_frame_monitor_block_top_offset * wordSize);
927 __ movptr(rax, monitor_block_top);
928 __ lea(rax, Address(rbp, rax, Address::times_ptr));
929 __ cmpptr(rax, rsp);
930 __ jcc(Assembler::equal, L);
931 __ stop("broken stack frame setup in interpreter 5");
932 __ bind(L);
933 }
934 #endif
935
936 // jvmti support
937 __ notify_method_entry();
938
939 // runtime upcalls
940 if (runtime_upcalls) {
941 __ generate_runtime_upcalls_on_method_entry();
942 }
943
944 // work registers
945 const Register method = rbx;
946 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
947 const Register t = NOT_LP64(rcx) LP64_ONLY(r11);
948
949 // allocate space for parameters
950 __ get_method(method);
951 __ movptr(t, Address(method, Method::const_offset()));
952 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
953
954 #ifndef _LP64
955 __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
956 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
957 __ subptr(rsp, t);
958 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
959 #else
960 __ shll(t, Interpreter::logStackElementSize);
961
962 __ subptr(rsp, t);
963 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
964 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
965 #endif // _LP64
966
967 // get signature handler
968 {
969 Label L;
970 __ movptr(t, Address(method, Method::signature_handler_offset()));
971 __ testptr(t, t);
972 __ jcc(Assembler::notZero, L);
973 __ call_VM(noreg,
974 CAST_FROM_FN_PTR(address,
975 InterpreterRuntime::prepare_native_call),
976 method);
977 __ get_method(method);
978 __ movptr(t, Address(method, Method::signature_handler_offset()));
979 __ bind(L);
980 }
981
982 // call signature handler
983 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
984 "adjust this code");
985 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
986 "adjust this code");
987 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
988 "adjust this code");
989
990 // The generated handlers do not touch RBX (the method).
991 // However, large signatures cannot be cached and are generated
992 // each time here. The slow-path generator can do a GC on return,
993 // so we must reload it after the call.
994 __ call(t);
995 __ get_method(method); // slow path can do a GC, reload RBX
996
997
998 // result handler is in rax
999 // set result handler
1000 __ movptr(Address(rbp,
1001 (frame::interpreter_frame_result_handler_offset) * wordSize),
1002 rax);
1003
1004 // pass mirror handle if static call
1005 {
1006 Label L;
1007 __ movl(t, Address(method, Method::access_flags_offset()));
1008 __ testl(t, JVM_ACC_STATIC);
1009 __ jcc(Assembler::zero, L);
1010 // get mirror
1011 __ load_mirror(t, method, rax);
1012 // copy mirror into activation frame
1013 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1014 t);
1015 // pass handle to mirror
1016 #ifndef _LP64
1017 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1018 __ movptr(Address(rsp, wordSize), t);
1019 #else
1020 __ lea(c_rarg1,
1021 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1022 #endif // _LP64
1023 __ bind(L);
1024 }
1025
1026 // get native function entry point
1027 {
1028 Label L;
1029 __ movptr(rax, Address(method, Method::native_function_offset()));
1030 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1031 __ cmpptr(rax, unsatisfied.addr(), rscratch1);
1032 __ jcc(Assembler::notEqual, L);
1033 __ call_VM(noreg,
1034 CAST_FROM_FN_PTR(address,
1035 InterpreterRuntime::prepare_native_call),
1036 method);
1037 __ get_method(method);
1038 __ movptr(rax, Address(method, Method::native_function_offset()));
1039 __ bind(L);
1040 }
1041
1042 // pass JNIEnv
1043 #ifndef _LP64
1044 __ get_thread(thread);
1045 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1046 __ movptr(Address(rsp, 0), t);
1047
1048 // set_last_Java_frame_before_call
1049 // It is enough that the pc()
1050 // points into the right code segment. It does not have to be the correct return pc.
1051 __ set_last_Java_frame(thread, noreg, rbp, __ pc(), noreg);
1052 #else
1053 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1054
1055 // It is enough that the pc() points into the right code
1056 // segment. It does not have to be the correct return pc.
1057 // For convenience we use the pc we want to resume to in
1058 // case of preemption on Object.wait.
1059 Label native_return;
1060 __ set_last_Java_frame(rsp, rbp, native_return, rscratch1);
1061 #endif // _LP64
1062
1063 // change thread state
1064 #ifdef ASSERT
1065 {
1066 Label L;
1067 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1068 __ cmpl(t, _thread_in_Java);
1069 __ jcc(Assembler::equal, L);
1070 __ stop("Wrong thread state in native stub");
1071 __ bind(L);
1072 }
1073 #endif
1074
1075 // Change state to native
1076
1077 __ movl(Address(thread, JavaThread::thread_state_offset()),
1078 _thread_in_native);
1079
1080 __ push_cont_fastpath();
1081
1082 // Call the native method.
1083 __ call(rax);
1084 // 32: result potentially in rdx:rax or ST0
1085 // 64: result potentially in rax or xmm0
1086
1087 __ pop_cont_fastpath();
1088
1089 // Verify or restore cpu control state after JNI call
1090 __ restore_cpu_control_state_after_jni(rscratch1);
1091
1092 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1093 // in order to extract the result of a method call. If the order of these
1094 // pushes change or anything else is added to the stack then the code in
1095 // interpreter_frame_result must also change.
1096
1097 #ifndef _LP64
1098 // save potential result in ST(0) & rdx:rax
1099 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1100 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1101 // It is safe to do this push because state is _thread_in_native and return address will be found
1102 // via _last_native_pc and not via _last_jave_sp
1103
1104 // NOTE: the order of these push(es) is known to frame::interpreter_frame_result.
1105 // If the order changes or anything else is added to the stack the code in
1106 // interpreter_frame_result will have to be changed.
1107
1108 { Label L;
1109 Label push_double;
1110 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1111 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1112 __ cmpptr(Address(rbp, (frame::interpreter_frame_result_handler_offset)*wordSize),
1113 float_handler.addr(), noreg);
1114 __ jcc(Assembler::equal, push_double);
1115 __ cmpptr(Address(rbp, (frame::interpreter_frame_result_handler_offset)*wordSize),
1116 double_handler.addr(), noreg);
1117 __ jcc(Assembler::notEqual, L);
1118 __ bind(push_double);
1119 __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1120 __ bind(L);
1121 }
1122 #else
1123 __ push(dtos);
1124 #endif // _LP64
1125
1126 __ push(ltos);
1127
1128 // change thread state
1129 NOT_LP64(__ get_thread(thread));
1130 __ movl(Address(thread, JavaThread::thread_state_offset()),
1131 _thread_in_native_trans);
1132
1133 // Force this write out before the read below
1134 if (!UseSystemMemoryBarrier) {
1135 __ membar(Assembler::Membar_mask_bits(
1136 Assembler::LoadLoad | Assembler::LoadStore |
1137 Assembler::StoreLoad | Assembler::StoreStore));
1138 }
1139 #ifndef _LP64
1140 if (AlwaysRestoreFPU) {
1141 // Make sure the control word is correct.
1142 __ fldcw(ExternalAddress(StubRoutines::x86::addr_fpu_cntrl_wrd_std()));
1143 }
1144 #endif // _LP64
1145
1146 // check for safepoint operation in progress and/or pending suspend requests
1147 {
1148 Label Continue;
1149 Label slow_path;
1150
1151 __ safepoint_poll(slow_path, thread, true /* at_return */, false /* in_nmethod */);
1152
1153 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1154 __ jcc(Assembler::equal, Continue);
1155 __ bind(slow_path);
1156
1157 // Don't use call_VM as it will see a possible pending exception
1158 // and forward it and never return here preventing us from
1159 // clearing _last_native_pc down below. Also can't use
1160 // call_VM_leaf either as it will check to see if r13 & r14 are
1161 // preserved and correspond to the bcp/locals pointers. So we do a
1162 // runtime call by hand.
1163 //
1164 #ifndef _LP64
1165 __ push(thread);
1166 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1167 JavaThread::check_special_condition_for_native_trans)));
1168 __ increment(rsp, wordSize);
1169 __ get_thread(thread);
1170 #else
1171 __ mov(c_rarg0, r15_thread);
1172 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1173 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1174 __ andptr(rsp, -16); // align stack as required by ABI
1175 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1176 __ mov(rsp, r12); // restore sp
1177 __ reinit_heapbase();
1178 #endif // _LP64
1179 __ bind(Continue);
1180 }
1181
1182 // change thread state
1183 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1184
1185 #ifdef _LP64
1186 if (LockingMode != LM_LEGACY) {
1187 // Check preemption for Object.wait()
1188 Label not_preempted;
1189 __ movptr(rscratch1, Address(r15_thread, JavaThread::preempt_alternate_return_offset()));
1190 __ cmpptr(rscratch1, NULL_WORD);
1191 __ jccb(Assembler::equal, not_preempted);
1192 __ movptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
1193 __ jmp(rscratch1);
1194 __ bind(native_return);
1195 __ restore_after_resume(true /* is_native */);
1196 __ bind(not_preempted);
1197 } else {
1198 // any pc will do so just use this one for LM_LEGACY to keep code together.
1199 __ bind(native_return);
1200 }
1201 #endif // _LP64
1202
1203 // reset_last_Java_frame
1204 __ reset_last_Java_frame(thread, true);
1205
1206 if (CheckJNICalls) {
1207 // clear_pending_jni_exception_check
1208 __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1209 }
1210
1211 // reset handle block
1212 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1213 __ movl(Address(t, JNIHandleBlock::top_offset()), NULL_WORD);
1214
1215 // If result is an oop unbox and store it in frame where gc will see it
1216 // and result handler will pick it up
1217
1218 {
1219 Label no_oop;
1220 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1221 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1222 __ jcc(Assembler::notEqual, no_oop);
1223 // retrieve result
1224 __ pop(ltos);
1225 // Unbox oop result, e.g. JNIHandles::resolve value.
1226 __ resolve_jobject(rax /* value */,
1227 thread /* thread */,
1228 t /* tmp */);
1229 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1230 // keep stack depth as expected by pushing oop which will eventually be discarded
1231 __ push(ltos);
1232 __ bind(no_oop);
1233 }
1234
1235
1236 {
1237 Label no_reguard;
1238 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1239 StackOverflow::stack_guard_yellow_reserved_disabled);
1240 __ jcc(Assembler::notEqual, no_reguard);
1241
1242 __ pusha(); // XXX only save smashed registers
1243 #ifndef _LP64
1244 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1245 __ popa();
1246 #else
1247 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1248 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1249 __ andptr(rsp, -16); // align stack as required by ABI
1250 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1251 __ mov(rsp, r12); // restore sp
1252 __ popa(); // XXX only restore smashed registers
1253 __ reinit_heapbase();
1254 #endif // _LP64
1255
1256 __ bind(no_reguard);
1257 }
1258
1259
1260 // The method register is junk from after the thread_in_native transition
1261 // until here. Also can't call_VM until the bcp has been
1262 // restored. Need bcp for throwing exception below so get it now.
1263 __ get_method(method);
1264
1265 // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1266 __ movptr(rbcp, Address(method, Method::const_offset())); // get ConstMethod*
1267 __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
1268
1269 // handle exceptions (exception handling will handle unlocking!)
1270 {
1271 Label L;
1272 __ cmpptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
1273 __ jcc(Assembler::zero, L);
1274 // Note: At some point we may want to unify this with the code
1275 // used in call_VM_base(); i.e., we should use the
1276 // StubRoutines::forward_exception code. For now this doesn't work
1277 // here because the rsp is not correctly set at this point.
1278 __ MacroAssembler::call_VM(noreg,
1279 CAST_FROM_FN_PTR(address,
1280 InterpreterRuntime::throw_pending_exception));
1281 __ should_not_reach_here();
1282 __ bind(L);
1283 }
1284
1285 // do unlocking if necessary
1286 {
1287 Label L;
1288 __ movl(t, Address(method, Method::access_flags_offset()));
1289 __ testl(t, JVM_ACC_SYNCHRONIZED);
1290 __ jcc(Assembler::zero, L);
1291 // the code below should be shared with interpreter macro
1292 // assembler implementation
1293 {
1294 Label unlock;
1295 // BasicObjectLock will be first in list, since this is a
1296 // synchronized method. However, need to check that the object
1297 // has not been unlocked by an explicit monitorexit bytecode.
1298 const Address monitor(rbp,
1299 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1300 wordSize - (int)sizeof(BasicObjectLock)));
1301
1302 const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1303
1304 // monitor expect in c_rarg1 for slow unlock path
1305 __ lea(regmon, monitor); // address of first monitor
1306
1307 __ movptr(t, Address(regmon, BasicObjectLock::obj_offset()));
1308 __ testptr(t, t);
1309 __ jcc(Assembler::notZero, unlock);
1310
1311 // Entry already unlocked, need to throw exception
1312 __ MacroAssembler::call_VM(noreg,
1313 CAST_FROM_FN_PTR(address,
1314 InterpreterRuntime::throw_illegal_monitor_state_exception));
1315 __ should_not_reach_here();
1316
1317 __ bind(unlock);
1318 __ unlock_object(regmon);
1319 }
1320 __ bind(L);
1321 }
1322
1323 // jvmti support
1324 // Note: This must happen _after_ handling/throwing any exceptions since
1325 // the exception handler code notifies the runtime of method exits
1326 // too. If this happens before, method entry/exit notifications are
1327 // not properly paired (was bug - gri 11/22/99).
1328 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1329
1330 // restore potential result in edx:eax, call result handler to
1331 // restore potential result in ST0 & handle result
1332
1333 __ pop(ltos);
1334 LP64_ONLY( __ pop(dtos));
1335
1336 __ movptr(t, Address(rbp,
1337 (frame::interpreter_frame_result_handler_offset) * wordSize));
1338 __ call(t);
1339
1340 // remove activation
1341 __ movptr(t, Address(rbp,
1342 frame::interpreter_frame_sender_sp_offset *
1343 wordSize)); // get sender sp
1344 __ leave(); // remove frame anchor
1345 __ pop(rdi); // get return address
1346 __ mov(rsp, t); // set sp to sender sp
1347 __ jmp(rdi);
1348
1349 if (inc_counter) {
1350 // Handle overflow of counter and compile method
1351 __ bind(invocation_counter_overflow);
1352 generate_counter_overflow(continue_after_compile);
1353 }
1354
1355 return entry_point;
1356 }
1357
1358 // Abstract method entry
1359 // Attempt to execute abstract method. Throw exception
1360 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1361
1362 address entry_point = __ pc();
1363
1364 // abstract method entry
1365
1366 // pop return address, reset last_sp to null
1367 __ empty_expression_stack();
1368 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1369 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1370
1371 // throw exception
1372 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod), rbx);
1373 // the call_VM checks for exception, so we should never return here.
1374 __ should_not_reach_here();
1375
1376 return entry_point;
1377 }
1378
1379 //
1380 // Generic interpreted method entry to (asm) interpreter
1381 //
1382 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized, bool runtime_upcalls) {
1383 // determine code generation flags
1384 bool inc_counter = UseCompiler || CountCompiledCalls;
1385
1386 // ebx: Method*
1387 // rbcp: sender sp (set in InterpreterMacroAssembler::prepare_to_jump_from_interpreted / generate_call_stub)
1388 address entry_point = __ pc();
1389
1390 const Address constMethod(rbx, Method::const_offset());
1391 const Address access_flags(rbx, Method::access_flags_offset());
1392 const Address size_of_parameters(rdx,
1393 ConstMethod::size_of_parameters_offset());
1394 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1395
1396
1397 // get parameter size (always needed)
1398 __ movptr(rdx, constMethod);
1399 __ load_unsigned_short(rcx, size_of_parameters);
1400
1401 // rbx: Method*
1402 // rcx: size of parameters
1403 // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1404
1405 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1406 __ subl(rdx, rcx); // rdx = no. of additional locals
1407
1408 // YYY
1409 // __ incrementl(rdx);
1410 // __ andl(rdx, -2);
1411
1412 // see if we've got enough room on the stack for locals plus overhead.
1413 generate_stack_overflow_check();
1414
1415 // get return address
1416 __ pop(rax);
1417
1418 // compute beginning of parameters
1419 __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1420
1421 // rdx - # of additional locals
1422 // allocate space for locals
1423 // explicitly initialize locals
1424 {
1425 Label exit, loop;
1426 __ testl(rdx, rdx);
1427 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1428 __ bind(loop);
1429 __ push(NULL_WORD); // initialize local variables
1430 __ decrementl(rdx); // until everything initialized
1431 __ jcc(Assembler::greater, loop);
1432 __ bind(exit);
1433 }
1434
1435 // initialize fixed part of activation frame
1436 generate_fixed_frame(false);
1437
1438 // make sure method is not native & not abstract
1439 #ifdef ASSERT
1440 __ movl(rax, access_flags);
1441 {
1442 Label L;
1443 __ testl(rax, JVM_ACC_NATIVE);
1444 __ jcc(Assembler::zero, L);
1445 __ stop("tried to execute native method as non-native");
1446 __ bind(L);
1447 }
1448 {
1449 Label L;
1450 __ testl(rax, JVM_ACC_ABSTRACT);
1451 __ jcc(Assembler::zero, L);
1452 __ stop("tried to execute abstract method in interpreter");
1453 __ bind(L);
1454 }
1455 #endif
1456
1457 // Since at this point in the method invocation the exception
1458 // handler would try to exit the monitor of synchronized methods
1459 // which hasn't been entered yet, we set the thread local variable
1460 // _do_not_unlock_if_synchronized to true. The remove_activation
1461 // will check this flag.
1462
1463 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1464 NOT_LP64(__ get_thread(thread));
1465 const Address do_not_unlock_if_synchronized(thread,
1466 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1467 __ movbool(do_not_unlock_if_synchronized, true);
1468
1469 __ profile_parameters_type(rax, rcx, rdx);
1470 // increment invocation count & check for overflow
1471 Label invocation_counter_overflow;
1472 if (inc_counter) {
1473 generate_counter_incr(&invocation_counter_overflow);
1474 }
1475
1476 Label continue_after_compile;
1477 __ bind(continue_after_compile);
1478
1479 // check for synchronized interpreted methods
1480 bang_stack_shadow_pages(false);
1481
1482 // reset the _do_not_unlock_if_synchronized flag
1483 NOT_LP64(__ get_thread(thread));
1484 __ movbool(do_not_unlock_if_synchronized, false);
1485
1486 // check for synchronized methods
1487 // Must happen AFTER invocation_counter check and stack overflow check,
1488 // so method is not locked if overflows.
1489 if (synchronized) {
1490 // Allocate monitor and lock method
1491 lock_method();
1492 } else {
1493 // no synchronization necessary
1494 #ifdef ASSERT
1495 {
1496 Label L;
1497 __ movl(rax, access_flags);
1498 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1499 __ jcc(Assembler::zero, L);
1500 __ stop("method needs synchronization");
1501 __ bind(L);
1502 }
1503 #endif
1504 }
1505
1506 // start execution
1507 #ifdef ASSERT
1508 {
1509 Label L;
1510 const Address monitor_block_top (rbp,
1511 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1512 __ movptr(rax, monitor_block_top);
1513 __ lea(rax, Address(rbp, rax, Address::times_ptr));
1514 __ cmpptr(rax, rsp);
1515 __ jcc(Assembler::equal, L);
1516 __ stop("broken stack frame setup in interpreter 6");
1517 __ bind(L);
1518 }
1519 #endif
1520
1521 // jvmti support
1522 __ notify_method_entry();
1523
1524 if (runtime_upcalls) {
1525 __ generate_runtime_upcalls_on_method_entry();
1526 }
1527
1528 __ dispatch_next(vtos);
1529
1530 // invocation counter overflow
1531 if (inc_counter) {
1532 // Handle overflow of counter and compile method
1533 __ bind(invocation_counter_overflow);
1534 generate_counter_overflow(continue_after_compile);
1535 }
1536
1537 return entry_point;
1538 }
1539
1540 //-----------------------------------------------------------------------------
1541 // Exceptions
1542
1543 void TemplateInterpreterGenerator::generate_throw_exception() {
1544 // Entry point in previous activation (i.e., if the caller was
1545 // interpreted)
1546 Interpreter::_rethrow_exception_entry = __ pc();
1547 // Restore sp to interpreter_frame_last_sp even though we are going
1548 // to empty the expression stack for the exception processing.
1549 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1550 // rax: exception
1551 // rdx: return address/pc that threw exception
1552 __ restore_bcp(); // r13/rsi points to call/send
1553 __ restore_locals();
1554 LP64_ONLY(__ reinit_heapbase()); // restore r12 as heapbase.
1555 // Entry point for exceptions thrown within interpreter code
1556 Interpreter::_throw_exception_entry = __ pc();
1557 // expression stack is undefined here
1558 // rax: exception
1559 // r13/rsi: exception bcp
1560 __ verify_oop(rax);
1561 Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1562 LP64_ONLY(__ mov(c_rarg1, rax));
1563
1564 // expression stack must be empty before entering the VM in case of
1565 // an exception
1566 __ empty_expression_stack();
1567 // find exception handler address and preserve exception oop
1568 __ call_VM(rdx,
1569 CAST_FROM_FN_PTR(address,
1570 InterpreterRuntime::exception_handler_for_exception),
1571 rarg);
1572 // rax: exception handler entry point
1573 // rdx: preserved exception oop
1574 // r13/rsi: bcp for exception handler
1575 __ push_ptr(rdx); // push exception which is now the only value on the stack
1576 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1577
1578 // If the exception is not handled in the current frame the frame is
1579 // removed and the exception is rethrown (i.e. exception
1580 // continuation is _rethrow_exception).
1581 //
1582 // Note: At this point the bci is still the bxi for the instruction
1583 // which caused the exception and the expression stack is
1584 // empty. Thus, for any VM calls at this point, GC will find a legal
1585 // oop map (with empty expression stack).
1586
1587 // In current activation
1588 // tos: exception
1589 // esi: exception bcp
1590
1591 //
1592 // JVMTI PopFrame support
1593 //
1594
1595 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1596 __ empty_expression_stack();
1597 // Set the popframe_processing bit in pending_popframe_condition
1598 // indicating that we are currently handling popframe, so that
1599 // call_VMs that may happen later do not trigger new popframe
1600 // handling cycles.
1601 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1602 NOT_LP64(__ get_thread(thread));
1603 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1604 __ orl(rdx, JavaThread::popframe_processing_bit);
1605 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1606
1607 {
1608 // Check to see whether we are returning to a deoptimized frame.
1609 // (The PopFrame call ensures that the caller of the popped frame is
1610 // either interpreted or compiled and deoptimizes it if compiled.)
1611 // In this case, we can't call dispatch_next() after the frame is
1612 // popped, but instead must save the incoming arguments and restore
1613 // them after deoptimization has occurred.
1614 //
1615 // Note that we don't compare the return PC against the
1616 // deoptimization blob's unpack entry because of the presence of
1617 // adapter frames in C2.
1618 Label caller_not_deoptimized;
1619 Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1620 __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1621 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1622 InterpreterRuntime::interpreter_contains), rarg);
1623 __ testl(rax, rax);
1624 __ jcc(Assembler::notZero, caller_not_deoptimized);
1625
1626 // Compute size of arguments for saving when returning to
1627 // deoptimized caller
1628 __ get_method(rax);
1629 __ movptr(rax, Address(rax, Method::const_offset()));
1630 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1631 size_of_parameters_offset())));
1632 __ shll(rax, Interpreter::logStackElementSize);
1633 __ restore_locals();
1634 __ subptr(rlocals, rax);
1635 __ addptr(rlocals, wordSize);
1636 // Save these arguments
1637 NOT_LP64(__ get_thread(thread));
1638 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1639 Deoptimization::
1640 popframe_preserve_args),
1641 thread, rax, rlocals);
1642
1643 __ remove_activation(vtos, rdx,
1644 /* throw_monitor_exception */ false,
1645 /* install_monitor_exception */ false,
1646 /* notify_jvmdi */ false);
1647
1648 // Inform deoptimization that it is responsible for restoring
1649 // these arguments
1650 NOT_LP64(__ get_thread(thread));
1651 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1652 JavaThread::popframe_force_deopt_reexecution_bit);
1653
1654 // Continue in deoptimization handler
1655 __ jmp(rdx);
1656
1657 __ bind(caller_not_deoptimized);
1658 }
1659
1660 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1661 /* throw_monitor_exception */ false,
1662 /* install_monitor_exception */ false,
1663 /* notify_jvmdi */ false);
1664
1665 // Finish with popframe handling
1666 // A previous I2C followed by a deoptimization might have moved the
1667 // outgoing arguments further up the stack. PopFrame expects the
1668 // mutations to those outgoing arguments to be preserved and other
1669 // constraints basically require this frame to look exactly as
1670 // though it had previously invoked an interpreted activation with
1671 // no space between the top of the expression stack (current
1672 // last_sp) and the top of stack. Rather than force deopt to
1673 // maintain this kind of invariant all the time we call a small
1674 // fixup routine to move the mutated arguments onto the top of our
1675 // expression stack if necessary.
1676 #ifndef _LP64
1677 __ mov(rax, rsp);
1678 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1679 __ lea(rbx, Address(rbp, rbx, Address::times_ptr));
1680 __ get_thread(thread);
1681 // PC must point into interpreter here
1682 __ set_last_Java_frame(thread, noreg, rbp, __ pc(), noreg);
1683 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1684 __ get_thread(thread);
1685 #else
1686 __ mov(c_rarg1, rsp);
1687 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1688 __ lea(c_rarg2, Address(rbp, c_rarg2, Address::times_ptr));
1689 // PC must point into interpreter here
1690 __ set_last_Java_frame(noreg, rbp, __ pc(), rscratch1);
1691 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1692 #endif
1693 __ reset_last_Java_frame(thread, true);
1694
1695 // Restore the last_sp and null it out
1696 __ movptr(rcx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1697 __ lea(rsp, Address(rbp, rcx, Address::times_ptr));
1698 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1699
1700 __ restore_bcp();
1701 __ restore_locals();
1702 // The method data pointer was incremented already during
1703 // call profiling. We have to restore the mdp for the current bcp.
1704 if (ProfileInterpreter) {
1705 __ set_method_data_pointer_for_bcp();
1706 }
1707
1708 // Clear the popframe condition flag
1709 NOT_LP64(__ get_thread(thread));
1710 __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1711 JavaThread::popframe_inactive);
1712
1713 #if INCLUDE_JVMTI
1714 {
1715 Label L_done;
1716 const Register local0 = rlocals;
1717
1718 __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1719 __ jcc(Assembler::notEqual, L_done);
1720
1721 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1722 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
1723
1724 __ get_method(rdx);
1725 __ movptr(rax, Address(local0, 0));
1726 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1727
1728 __ testptr(rax, rax);
1729 __ jcc(Assembler::zero, L_done);
1730
1731 __ movptr(Address(rbx, 0), rax);
1732 __ bind(L_done);
1733 }
1734 #endif // INCLUDE_JVMTI
1735
1736 __ dispatch_next(vtos);
1737 // end of PopFrame support
1738
1739 Interpreter::_remove_activation_entry = __ pc();
1740
1741 // preserve exception over this code sequence
1742 __ pop_ptr(rax);
1743 NOT_LP64(__ get_thread(thread));
1744 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1745 // remove the activation (without doing throws on illegalMonitorExceptions)
1746 __ remove_activation(vtos, rdx, false, true, false);
1747 // restore exception
1748 NOT_LP64(__ get_thread(thread));
1749 __ get_vm_result(rax, thread);
1750
1751 // In between activations - previous activation type unknown yet
1752 // compute continuation point - the continuation point expects the
1753 // following registers set up:
1754 //
1755 // rax: exception
1756 // rdx: return address/pc that threw exception
1757 // rsp: expression stack of caller
1758 // rbp: ebp of caller
1759 __ push(rax); // save exception
1760 __ push(rdx); // save return address
1761 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1762 SharedRuntime::exception_handler_for_return_address),
1763 thread, rdx);
1764 __ mov(rbx, rax); // save exception handler
1765 __ pop(rdx); // restore return address
1766 __ pop(rax); // restore exception
1767 // Note that an "issuing PC" is actually the next PC after the call
1768 __ jmp(rbx); // jump to exception
1769 // handler of caller
1770 }
1771
1772
1773 //
1774 // JVMTI ForceEarlyReturn support
1775 //
1776 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1777 address entry = __ pc();
1778
1779 __ restore_bcp();
1780 __ restore_locals();
1781 __ empty_expression_stack();
1782 __ load_earlyret_value(state); // 32 bits returns value in rdx, so don't reuse
1783
1784 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1785 NOT_LP64(__ get_thread(thread));
1786 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1787 Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1788
1789 // Clear the earlyret state
1790 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1791
1792 __ remove_activation(state, rsi,
1793 false, /* throw_monitor_exception */
1794 false, /* install_monitor_exception */
1795 true); /* notify_jvmdi */
1796 __ jmp(rsi);
1797
1798 return entry;
1799 } // end of ForceEarlyReturn support
1800
1801
1802 //-----------------------------------------------------------------------------
1803 // Helper for vtos entry point generation
1804
1805 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1806 address& bep,
1807 address& cep,
1808 address& sep,
1809 address& aep,
1810 address& iep,
1811 address& lep,
1812 address& fep,
1813 address& dep,
1814 address& vep) {
1815 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1816 Label L;
1817 #ifndef _LP64
1818 fep = __ pc(); // ftos entry point
1819 __ push(ftos);
1820 __ jmpb(L);
1821 dep = __ pc(); // dtos entry point
1822 __ push(dtos);
1823 __ jmpb(L);
1824 #else
1825 fep = __ pc(); // ftos entry point
1826 __ push_f(xmm0);
1827 __ jmpb(L);
1828 dep = __ pc(); // dtos entry point
1829 __ push_d(xmm0);
1830 __ jmpb(L);
1831 #endif // _LP64
1832 lep = __ pc(); // ltos entry point
1833 __ push_l();
1834 __ jmpb(L);
1835 aep = bep = cep = sep = iep = __ pc(); // [abcsi]tos entry point
1836 __ push_i_or_ptr();
1837 vep = __ pc(); // vtos entry point
1838 __ bind(L);
1839 generate_and_dispatch(t);
1840 }
1841
1842 //-----------------------------------------------------------------------------
1843
1844 void TemplateInterpreterGenerator::count_bytecode() {
1845 #ifdef _LP64
1846 __ incrementq(ExternalAddress((address) &BytecodeCounter::_counter_value), rscratch1);
1847 #else
1848 Unimplemented();
1849 #endif
1850 }
1851
1852 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1853 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]), rscratch1);
1854 }
1855
1856 // Non-product code
1857 #ifndef PRODUCT
1858
1859 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1860 address entry = __ pc();
1861
1862 #ifndef _LP64
1863 // prepare expression stack
1864 __ pop(rcx); // pop return address so expression stack is 'pure'
1865 __ push(state); // save tosca
1866
1867 // pass tosca registers as arguments & call tracer
1868 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1869 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1870 __ pop(state); // restore tosca
1871
1872 // return
1873 __ jmp(rcx);
1874 #else
1875 __ push(state);
1876 __ push(c_rarg0);
1877 __ push(c_rarg1);
1878 __ push(c_rarg2);
1879 __ push(c_rarg3);
1880 __ mov(c_rarg2, rax); // Pass itos
1881 #ifdef _WIN64
1882 __ movflt(xmm3, xmm0); // Pass ftos
1883 #endif
1884 __ call_VM(noreg,
1885 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1886 c_rarg1, c_rarg2, c_rarg3);
1887 __ pop(c_rarg3);
1888 __ pop(c_rarg2);
1889 __ pop(c_rarg1);
1890 __ pop(c_rarg0);
1891 __ pop(state);
1892 __ ret(0); // return from result handler
1893 #endif // _LP64
1894
1895 return entry;
1896 }
1897
1898 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1899 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1900 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1901 __ orl(rbx,
1902 ((int) t->bytecode()) <<
1903 BytecodePairHistogram::log2_number_of_codes);
1904 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx, rscratch1);
1905 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1906 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1907 }
1908
1909
1910 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1911 // Call a little run-time stub to avoid blow-up for each bytecode.
1912 // The run-time runtime saves the right registers, depending on
1913 // the tosca in-state for the given template.
1914
1915 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
1916 "entry must have been generated");
1917 #ifndef _LP64
1918 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1919 #else
1920 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1921 __ andptr(rsp, -16); // align stack as required by ABI
1922 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1923 __ mov(rsp, r12); // restore sp
1924 __ reinit_heapbase();
1925 #endif // _LP64
1926 }
1927
1928
1929 void TemplateInterpreterGenerator::stop_interpreter_at() {
1930 Label L;
1931 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1932 StopInterpreterAt,
1933 rscratch1);
1934 __ jcc(Assembler::notEqual, L);
1935 __ int3();
1936 __ bind(L);
1937 }
1938 #endif // !PRODUCT