1 /*
2 * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/bytecodeHistogram.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterGenerator.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/interp_masm.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "oops/arrayOop.hpp"
34 #include "oops/methodData.hpp"
35 #include "oops/method.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/jvmtiExport.hpp"
38 #include "prims/jvmtiThreadState.hpp"
39 #include "runtime/arguments.hpp"
40 #include "runtime/deoptimization.hpp"
41 #include "runtime/frame.inline.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/synchronizer.hpp"
45 #include "runtime/timer.hpp"
46 #include "runtime/vframeArray.hpp"
47 #include "utilities/debug.hpp"
48 #include "utilities/macros.hpp"
49
50 #define __ _masm->
51
52
53 #ifndef CC_INTERP
54 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
55 const int bcp_offset = frame::interpreter_frame_bcp_offset * wordSize;
56 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
57
58 //------------------------------------------------------------------------------------------------------------------------
59
60 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
61 address entry = __ pc();
62
63 // Note: There should be a minimal interpreter frame set up when stack
64 // overflow occurs since we check explicitly for it now.
65 //
66 #ifdef ASSERT
67 { Label L;
68 __ lea(rax, Address(rbp,
69 frame::interpreter_frame_monitor_block_top_offset * wordSize));
70 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp,
71 // (stack grows negative)
72 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
73 __ stop ("interpreter frame not set up");
74 __ bind(L);
75 }
76 #endif // ASSERT
77 // Restore bcp under the assumption that the current frame is still
78 // interpreted
79 __ restore_bcp();
80
81 // expression stack must be empty before entering the VM if an exception
82 // happened
83 __ empty_expression_stack();
84 __ empty_FPU_stack();
85 // throw exception
86 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
87 return entry;
88 }
89
90 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
91 address entry = __ pc();
92 // expression stack must be empty before entering the VM if an exception happened
93 __ empty_expression_stack();
94 __ empty_FPU_stack();
95 // setup parameters
96 // ??? convention: expect aberrant index in register rbx,
97 __ lea(rax, ExternalAddress((address)name));
98 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
99 return entry;
100 }
101
102 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
103 address entry = __ pc();
104 // object is at TOS
105 __ pop(rax);
106 // expression stack must be empty before entering the VM if an exception
107 // happened
108 __ empty_expression_stack();
109 __ empty_FPU_stack();
110 __ call_VM(noreg,
111 CAST_FROM_FN_PTR(address,
112 InterpreterRuntime::throw_ClassCastException),
113 rax);
114 return entry;
115 }
116
117 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
118 assert(!pass_oop || message == NULL, "either oop or message but not both");
119 address entry = __ pc();
120 if (pass_oop) {
121 // object is at TOS
122 __ pop(rbx);
123 }
124 // expression stack must be empty before entering the VM if an exception happened
125 __ empty_expression_stack();
126 __ empty_FPU_stack();
127 // setup parameters
128 __ lea(rax, ExternalAddress((address)name));
129 if (pass_oop) {
130 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
131 } else {
132 if (message != NULL) {
133 __ lea(rbx, ExternalAddress((address)message));
134 } else {
135 __ movptr(rbx, NULL_WORD);
136 }
137 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
138 }
139 // throw exception
140 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
141 return entry;
142 }
143
144
145 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
146 address entry = __ pc();
147 // NULL last_sp until next java call
148 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
149 __ dispatch_next(state);
150 return entry;
151 }
152
153
154 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
155 address entry = __ pc();
156
157 #ifdef COMPILER2
158 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
159 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
160 for (int i = 1; i < 8; i++) {
161 __ ffree(i);
162 }
163 } else if (UseSSE < 2) {
164 __ empty_FPU_stack();
165 }
166 #endif
167 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
168 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
169 } else {
170 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
171 }
172
173 // In SSE mode, interpreter returns FP results in xmm0 but they need
174 // to end up back on the FPU so it can operate on them.
175 if (state == ftos && UseSSE >= 1) {
176 __ subptr(rsp, wordSize);
177 __ movflt(Address(rsp, 0), xmm0);
178 __ fld_s(Address(rsp, 0));
179 __ addptr(rsp, wordSize);
180 } else if (state == dtos && UseSSE >= 2) {
181 __ subptr(rsp, 2*wordSize);
182 __ movdbl(Address(rsp, 0), xmm0);
183 __ fld_d(Address(rsp, 0));
184 __ addptr(rsp, 2*wordSize);
185 }
186
187 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
188
189 // Restore stack bottom in case i2c adjusted stack
190 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
191 // and NULL it as marker that rsp is now tos until next java call
192 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
193
194 __ restore_bcp();
195 __ restore_locals();
196
197 if (state == atos) {
198 Register mdp = rbx;
199 Register tmp = rcx;
200 __ profile_return_type(mdp, rax, tmp);
201 }
202
203 const Register cache = rbx;
204 const Register index = rcx;
205 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
206
207 const Register flags = cache;
208 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
209 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
210 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
211 __ dispatch_next(state, step);
212
213 return entry;
214 }
215
216
217 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
218 address entry = __ pc();
219
220 // In SSE mode, FP results are in xmm0
221 if (state == ftos && UseSSE > 0) {
222 __ subptr(rsp, wordSize);
223 __ movflt(Address(rsp, 0), xmm0);
224 __ fld_s(Address(rsp, 0));
225 __ addptr(rsp, wordSize);
226 } else if (state == dtos && UseSSE >= 2) {
227 __ subptr(rsp, 2*wordSize);
228 __ movdbl(Address(rsp, 0), xmm0);
229 __ fld_d(Address(rsp, 0));
230 __ addptr(rsp, 2*wordSize);
231 }
232
233 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
234
235 // The stack is not extended by deopt but we must NULL last_sp as this
236 // entry is like a "return".
237 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
238 __ restore_bcp();
239 __ restore_locals();
240 // handle exceptions
241 { Label L;
242 const Register thread = rcx;
243 __ get_thread(thread);
244 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
245 __ jcc(Assembler::zero, L);
246 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
247 __ should_not_reach_here();
248 __ bind(L);
249 }
250 __ dispatch_next(state, step);
251 return entry;
252 }
253
254
255 int AbstractInterpreter::BasicType_as_index(BasicType type) {
256 int i = 0;
257 switch (type) {
258 case T_BOOLEAN: i = 0; break;
259 case T_CHAR : i = 1; break;
260 case T_BYTE : i = 2; break;
261 case T_SHORT : i = 3; break;
262 case T_INT : // fall through
263 case T_LONG : // fall through
264 case T_VOID : i = 4; break;
265 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
266 case T_DOUBLE : i = 6; break;
267 case T_OBJECT : // fall through
268 case T_ARRAY : i = 7; break;
269 default : ShouldNotReachHere();
270 }
271 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
272 return i;
273 }
274
275
276 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
277 address entry = __ pc();
278 switch (type) {
279 case T_BOOLEAN: __ c2bool(rax); break;
280 case T_CHAR : __ andptr(rax, 0xFFFF); break;
281 case T_BYTE : __ sign_extend_byte (rax); break;
282 case T_SHORT : __ sign_extend_short(rax); break;
283 case T_INT : /* nothing to do */ break;
284 case T_LONG : /* nothing to do */ break;
285 case T_VOID : /* nothing to do */ break;
286 case T_DOUBLE :
287 case T_FLOAT :
288 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
289 __ pop(t); // remove return address first
290 // Must return a result for interpreter or compiler. In SSE
291 // mode, results are returned in xmm0 and the FPU stack must
292 // be empty.
293 if (type == T_FLOAT && UseSSE >= 1) {
294 // Load ST0
295 __ fld_d(Address(rsp, 0));
296 // Store as float and empty fpu stack
297 __ fstp_s(Address(rsp, 0));
298 // and reload
299 __ movflt(xmm0, Address(rsp, 0));
300 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
301 __ movdbl(xmm0, Address(rsp, 0));
302 } else {
303 // restore ST0
304 __ fld_d(Address(rsp, 0));
305 }
306 // and pop the temp
307 __ addptr(rsp, 2 * wordSize);
308 __ push(t); // restore return address
309 }
310 break;
311 case T_OBJECT :
312 // retrieve result from frame
313 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
314 // and verify it
315 __ verify_oop(rax);
316 break;
317 default : ShouldNotReachHere();
318 }
319 __ ret(0); // return from result handler
320 return entry;
321 }
322
323 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
324 address entry = __ pc();
325 __ push(state);
326 __ call_VM(noreg, runtime_entry);
327 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
328 return entry;
329 }
330
331
332 // Helpers for commoning out cases in the various type of method entries.
333 //
334
335 // increment invocation count & check for overflow
336 //
337 // Note: checking for negative value instead of overflow
338 // so we have a 'sticky' overflow test
339 //
340 // rbx,: method
341 // rcx: invocation counter
342 //
343 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
344 Label done;
345 // Note: In tiered we increment either counters in MethodCounters* or in MDO
346 // depending if we're profiling or not.
347 if (TieredCompilation) {
348 int increment = InvocationCounter::count_increment;
349 Label no_mdo;
350 if (ProfileInterpreter) {
351 // Are we profiling?
352 __ movptr(rax, Address(rbx, Method::method_data_offset()));
353 __ testptr(rax, rax);
354 __ jccb(Assembler::zero, no_mdo);
355 // Increment counter in the MDO
356 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
357 in_bytes(InvocationCounter::counter_offset()));
358 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
359 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
360 __ jmp(done);
361 }
362 __ bind(no_mdo);
363 // Increment counter in MethodCounters
364 const Address invocation_counter(rax,
365 MethodCounters::invocation_counter_offset() +
366 InvocationCounter::counter_offset());
367
368 __ get_method_counters(rbx, rax, done);
369 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
370 __ increment_mask_and_jump(invocation_counter, increment, mask,
371 rcx, false, Assembler::zero, overflow);
372 __ bind(done);
373 } else { // not TieredCompilation
374 const Address backedge_counter(rax,
375 MethodCounters::backedge_counter_offset() +
376 InvocationCounter::counter_offset());
377 const Address invocation_counter(rax,
378 MethodCounters::invocation_counter_offset() +
379 InvocationCounter::counter_offset());
380
381 __ get_method_counters(rbx, rax, done);
382
383 if (ProfileInterpreter) {
384 __ incrementl(Address(rax,
385 MethodCounters::interpreter_invocation_counter_offset()));
386 }
387
388 // Update standard invocation counters
389 __ movl(rcx, invocation_counter);
390 __ incrementl(rcx, InvocationCounter::count_increment);
391 __ movl(invocation_counter, rcx); // save invocation count
392
393 __ movl(rax, backedge_counter); // load backedge counter
394 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
395
396 __ addl(rcx, rax); // add both counters
397
398 // profile_method is non-null only for interpreted method so
399 // profile_method != NULL == !native_call
400 // BytecodeInterpreter only calls for native so code is elided.
401
402 if (ProfileInterpreter && profile_method != NULL) {
403 // Test to see if we should create a method data oop
404 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
405 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
406 __ jcc(Assembler::less, *profile_method_continue);
407
408 // if no method data exists, go to profile_method
409 __ test_method_data_pointer(rax, *profile_method);
410 }
411
412 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
413 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
414 __ jcc(Assembler::aboveEqual, *overflow);
415 __ bind(done);
416 }
417 }
418
419 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
420
421 // Asm interpreter on entry
422 // rdi - locals
423 // rsi - bcp
424 // rbx, - method
425 // rdx - cpool
426 // rbp, - interpreter frame
427
428 // C++ interpreter on entry
429 // rsi - new interpreter state pointer
430 // rbp - interpreter frame pointer
431 // rbx - method
432
433 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
434 // rbx, - method
435 // rcx - rcvr (assuming there is one)
436 // top of stack return address of interpreter caller
437 // rsp - sender_sp
438
439 // C++ interpreter only
440 // rsi - previous interpreter state pointer
441
442 // InterpreterRuntime::frequency_counter_overflow takes one argument
443 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
444 // The call returns the address of the verified entry point for the method or NULL
445 // if the compilation did not complete (either went background or bailed out).
446 __ movptr(rax, (intptr_t)false);
447 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
448
449 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
450
451 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
452 // and jump to the interpreted entry.
453 __ jmp(*do_continue, relocInfo::none);
454
455 }
456
457 void InterpreterGenerator::generate_stack_overflow_check(void) {
458 // see if we've got enough room on the stack for locals plus overhead.
459 // the expression stack grows down incrementally, so the normal guard
460 // page mechanism will work for that.
461 //
462 // Registers live on entry:
463 //
464 // Asm interpreter
465 // rdx: number of additional locals this frame needs (what we must check)
466 // rbx,: Method*
467
468 // destroyed on exit
469 // rax,
470
471 // NOTE: since the additional locals are also always pushed (wasn't obvious in
472 // generate_fixed_frame) so the guard should work for them too.
473 //
474
475 // monitor entry size: see picture of stack in frame_x86.hpp
476 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
477
478 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
479 // be sure to change this if you add/subtract anything to/from the overhead area
480 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
481
482 const int page_size = os::vm_page_size();
483
484 Label after_frame_check;
485
486 // see if the frame is greater than one page in size. If so,
487 // then we need to verify there is enough stack space remaining
488 // for the additional locals.
489 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
490 __ jcc(Assembler::belowEqual, after_frame_check);
491
492 // compute rsp as if this were going to be the last frame on
493 // the stack before the red zone
494
495 Label after_frame_check_pop;
496
497 __ push(rsi);
498
499 const Register thread = rsi;
500
501 __ get_thread(thread);
502
503 const Address stack_base(thread, Thread::stack_base_offset());
504 const Address stack_size(thread, Thread::stack_size_offset());
505
506 // locals + overhead, in bytes
507 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
508
509 #ifdef ASSERT
510 Label stack_base_okay, stack_size_okay;
511 // verify that thread stack base is non-zero
512 __ cmpptr(stack_base, (int32_t)NULL_WORD);
513 __ jcc(Assembler::notEqual, stack_base_okay);
514 __ stop("stack base is zero");
515 __ bind(stack_base_okay);
516 // verify that thread stack size is non-zero
517 __ cmpptr(stack_size, 0);
518 __ jcc(Assembler::notEqual, stack_size_okay);
519 __ stop("stack size is zero");
520 __ bind(stack_size_okay);
521 #endif
522
523 // Add stack base to locals and subtract stack size
524 __ addptr(rax, stack_base);
525 __ subptr(rax, stack_size);
526
527 // Use the maximum number of pages we might bang.
528 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
529 (StackRedPages+StackYellowPages);
530 __ addptr(rax, max_pages * page_size);
531
532 // check against the current stack bottom
533 __ cmpptr(rsp, rax);
534 __ jcc(Assembler::above, after_frame_check_pop);
535
536 __ pop(rsi); // get saved bcp / (c++ prev state ).
537
538 // Restore sender's sp as SP. This is necessary if the sender's
539 // frame is an extended compiled frame (see gen_c2i_adapter())
540 // and safer anyway in case of JSR292 adaptations.
541
542 __ pop(rax); // return address must be moved if SP is changed
543 __ mov(rsp, rsi);
544 __ push(rax);
545
546 // Note: the restored frame is not necessarily interpreted.
547 // Use the shared runtime version of the StackOverflowError.
548 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
549 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
550 // all done with frame size check
551 __ bind(after_frame_check_pop);
552 __ pop(rsi);
553
554 __ bind(after_frame_check);
555 }
556
557 // Allocate monitor and lock method (asm interpreter)
558 // rbx, - Method*
559 //
560 void InterpreterGenerator::lock_method(void) {
561 // synchronize method
562 const Address access_flags (rbx, Method::access_flags_offset());
563 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
564 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
565
566 #ifdef ASSERT
567 { Label L;
568 __ movl(rax, access_flags);
569 __ testl(rax, JVM_ACC_SYNCHRONIZED);
570 __ jcc(Assembler::notZero, L);
571 __ stop("method doesn't need synchronization");
572 __ bind(L);
573 }
574 #endif // ASSERT
575 // get synchronization object
576 { Label done;
577 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
578 __ movl(rax, access_flags);
579 __ testl(rax, JVM_ACC_STATIC);
580 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
581 __ jcc(Assembler::zero, done);
582 __ movptr(rax, Address(rbx, Method::const_offset()));
583 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
584 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
585 __ movptr(rax, Address(rax, mirror_offset));
586 __ bind(done);
587 }
588 // add space for monitor & lock
589 __ subptr(rsp, entry_size); // add space for a monitor entry
590 __ movptr(monitor_block_top, rsp); // set new monitor block top
591 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
592 __ mov(rdx, rsp); // object address
593 __ lock_object(rdx);
594 }
595
596 //
597 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
598 // and for native methods hence the shared code.
599
600 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
601 // initialize fixed part of activation frame
602 __ push(rax); // save return address
603 __ enter(); // save old & set new rbp,
604
605
606 __ push(rsi); // set sender sp
607 __ push((int32_t)NULL_WORD); // leave last_sp as null
608 __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod*
609 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
610 __ push(rbx); // save Method*
611 if (ProfileInterpreter) {
612 Label method_data_continue;
613 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
614 __ testptr(rdx, rdx);
615 __ jcc(Assembler::zero, method_data_continue);
616 __ addptr(rdx, in_bytes(MethodData::data_offset()));
617 __ bind(method_data_continue);
618 __ push(rdx); // set the mdp (method data pointer)
619 } else {
620 __ push(0);
621 }
622
623 __ movptr(rdx, Address(rbx, Method::const_offset()));
624 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
625 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
626 __ push(rdx); // set constant pool cache
627 __ push(rdi); // set locals pointer
628 if (native_call) {
629 __ push(0); // no bcp
630 } else {
631 __ push(rsi); // set bcp
632 }
633 __ push(0); // reserve word for pointer to expression stack bottom
634 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
635 }
636
637
638 // Method entry for java.lang.ref.Reference.get.
639 address InterpreterGenerator::generate_Reference_get_entry(void) {
640 #if INCLUDE_ALL_GCS
641 // Code: _aload_0, _getfield, _areturn
642 // parameter size = 1
643 //
644 // The code that gets generated by this routine is split into 2 parts:
645 // 1. The "intrinsified" code for G1 (or any SATB based GC),
646 // 2. The slow path - which is an expansion of the regular method entry.
647 //
648 // Notes:-
649 // * In the G1 code we do not check whether we need to block for
650 // a safepoint. If G1 is enabled then we must execute the specialized
651 // code for Reference.get (except when the Reference object is null)
652 // so that we can log the value in the referent field with an SATB
653 // update buffer.
654 // If the code for the getfield template is modified so that the
655 // G1 pre-barrier code is executed when the current method is
656 // Reference.get() then going through the normal method entry
657 // will be fine.
658 // * The G1 code below can, however, check the receiver object (the instance
659 // of java.lang.Reference) and jump to the slow path if null. If the
660 // Reference object is null then we obviously cannot fetch the referent
661 // and so we don't need to call the G1 pre-barrier. Thus we can use the
662 // regular method entry code to generate the NPE.
663 //
664 // This code is based on generate_accessor_enty.
665
666 // rbx,: Method*
667 // rcx: receiver (preserve for slow entry into asm interpreter)
668
669 // rsi: senderSP must preserved for slow path, set SP to it on fast path
670
671 address entry = __ pc();
672
673 const int referent_offset = java_lang_ref_Reference::referent_offset;
674 guarantee(referent_offset > 0, "referent offset not initialized");
675
676 if (UseG1GC) {
677 Label slow_path;
678
679 // Check if local 0 != NULL
680 // If the receiver is null then it is OK to jump to the slow path.
681 __ movptr(rax, Address(rsp, wordSize));
682 __ testptr(rax, rax);
683 __ jcc(Assembler::zero, slow_path);
684
685 // rax: local 0 (must be preserved across the G1 barrier call)
686 //
687 // rbx: method (at this point it's scratch)
688 // rcx: receiver (at this point it's scratch)
689 // rdx: scratch
690 // rdi: scratch
691 //
692 // rsi: sender sp
693
694 // Preserve the sender sp in case the pre-barrier
695 // calls the runtime
696 __ push(rsi);
697
698 // Load the value of the referent field.
699 const Address field_address(rax, referent_offset);
700 __ movptr(rax, field_address);
701
702 // Generate the G1 pre-barrier code to log the value of
703 // the referent field in an SATB buffer.
704 __ get_thread(rcx);
705 __ g1_write_barrier_pre(noreg /* obj */,
706 rax /* pre_val */,
707 rcx /* thread */,
708 rbx /* tmp */,
709 true /* tosca_save */,
710 true /* expand_call */);
711
712 // _areturn
713 __ pop(rsi); // get sender sp
714 __ pop(rdi); // get return address
715 __ mov(rsp, rsi); // set sp to sender sp
716 __ jmp(rdi);
717
718 __ bind(slow_path);
719 (void) generate_normal_entry(false);
720
721 return entry;
722 }
723 #endif // INCLUDE_ALL_GCS
724
725 // If G1 is not enabled then attempt to go through the accessor entry point
726 // Reference.get is an accessor
727 return generate_jump_to_normal_entry();
728 }
729
730 /**
731 * Method entry for static native methods:
732 * int java.util.zip.CRC32.update(int crc, int b)
733 */
734 address InterpreterGenerator::generate_CRC32_update_entry() {
735 if (UseCRC32Intrinsics) {
736 address entry = __ pc();
737
738 // rbx,: Method*
739 // rsi: senderSP must preserved for slow path, set SP to it on fast path
740 // rdx: scratch
741 // rdi: scratch
742
743 Label slow_path;
744 // If we need a safepoint check, generate full interpreter entry.
745 ExternalAddress state(SafepointSynchronize::address_of_state());
746 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
747 SafepointSynchronize::_not_synchronized);
748 __ jcc(Assembler::notEqual, slow_path);
749
750 // We don't generate local frame and don't align stack because
751 // we call stub code and there is no safepoint on this path.
752
753 // Load parameters
754 const Register crc = rax; // crc
755 const Register val = rdx; // source java byte value
756 const Register tbl = rdi; // scratch
757
758 // Arguments are reversed on java expression stack
759 __ movl(val, Address(rsp, wordSize)); // byte value
760 __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
761
762 __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
763 __ notl(crc); // ~crc
764 __ update_byte_crc32(crc, val, tbl);
765 __ notl(crc); // ~crc
766 // result in rax
767
768 // _areturn
769 __ pop(rdi); // get return address
770 __ mov(rsp, rsi); // set sp to sender sp
771 __ jmp(rdi);
772
773 // generate a vanilla native entry as the slow path
774 __ bind(slow_path);
775
776 (void) generate_native_entry(false);
777
778 return entry;
779 }
780 return generate_native_entry(false);
781 }
782
783 /**
784 * Method entry for static native methods:
785 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
786 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
787 */
788 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
789 if (UseCRC32Intrinsics) {
790 address entry = __ pc();
791
792 // rbx,: Method*
793 // rsi: senderSP must preserved for slow path, set SP to it on fast path
794 // rdx: scratch
795 // rdi: scratch
796
797 Label slow_path;
798 // If we need a safepoint check, generate full interpreter entry.
799 ExternalAddress state(SafepointSynchronize::address_of_state());
800 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
801 SafepointSynchronize::_not_synchronized);
802 __ jcc(Assembler::notEqual, slow_path);
803
804 // We don't generate local frame and don't align stack because
805 // we call stub code and there is no safepoint on this path.
806
807 // Load parameters
808 const Register crc = rax; // crc
809 const Register buf = rdx; // source java byte array address
810 const Register len = rdi; // length
811
812 // Arguments are reversed on java expression stack
813 __ movl(len, Address(rsp, wordSize)); // Length
814 // Calculate address of start element
815 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
816 __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
817 __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
818 __ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC
819 } else {
820 __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
821 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
822 __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
823 __ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC
824 }
825
826 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
827 // result in rax
828
829 // _areturn
830 __ pop(rdi); // get return address
831 __ mov(rsp, rsi); // set sp to sender sp
832 __ jmp(rdi);
833
834 // generate a vanilla native entry as the slow path
835 __ bind(slow_path);
836
837 (void) generate_native_entry(false);
838
839 return entry;
840 }
841 return generate_native_entry(false);
842 }
843
844 //
845 // Interpreter stub for calling a native method. (asm interpreter)
846 // This sets up a somewhat different looking stack for calling the native method
847 // than the typical interpreter frame setup.
848 //
849
850 address InterpreterGenerator::generate_native_entry(bool synchronized) {
851 // determine code generation flags
852 bool inc_counter = UseCompiler || CountCompiledCalls;
853
854 // rbx,: Method*
855 // rsi: sender sp
856 // rsi: previous interpreter state (C++ interpreter) must preserve
857 address entry_point = __ pc();
858
859 const Address constMethod (rbx, Method::const_offset());
860 const Address access_flags (rbx, Method::access_flags_offset());
861 const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
862
863 // get parameter size (always needed)
864 __ movptr(rcx, constMethod);
865 __ load_unsigned_short(rcx, size_of_parameters);
866
867 // native calls don't need the stack size check since they have no expression stack
868 // and the arguments are already on the stack and we only add a handful of words
869 // to the stack
870
871 // rbx,: Method*
872 // rcx: size of parameters
873 // rsi: sender sp
874
875 __ pop(rax); // get return address
876 // for natives the size of locals is zero
877
878 // compute beginning of parameters (rdi)
879 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
880
881
882 // add 2 zero-initialized slots for native calls
883 // NULL result handler
884 __ push((int32_t)NULL_WORD);
885 // NULL oop temp (mirror or jni oop result)
886 __ push((int32_t)NULL_WORD);
887
888 // initialize fixed part of activation frame
889 generate_fixed_frame(true);
890
891 // make sure method is native & not abstract
892 #ifdef ASSERT
893 __ movl(rax, access_flags);
894 {
895 Label L;
896 __ testl(rax, JVM_ACC_NATIVE);
897 __ jcc(Assembler::notZero, L);
898 __ stop("tried to execute non-native method as native");
899 __ bind(L);
900 }
901 { Label L;
902 __ testl(rax, JVM_ACC_ABSTRACT);
903 __ jcc(Assembler::zero, L);
904 __ stop("tried to execute abstract method in interpreter");
905 __ bind(L);
906 }
907 #endif
908
909 // Since at this point in the method invocation the exception handler
910 // would try to exit the monitor of synchronized methods which hasn't
911 // been entered yet, we set the thread local variable
912 // _do_not_unlock_if_synchronized to true. The remove_activation will
913 // check this flag.
914
915 __ get_thread(rax);
916 const Address do_not_unlock_if_synchronized(rax,
917 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
918 __ movbool(do_not_unlock_if_synchronized, true);
919
920 // increment invocation count & check for overflow
921 Label invocation_counter_overflow;
922 if (inc_counter) {
923 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
924 }
925
926 Label continue_after_compile;
927 __ bind(continue_after_compile);
928
929 bang_stack_shadow_pages(true);
930
931 // reset the _do_not_unlock_if_synchronized flag
932 __ get_thread(rax);
933 __ movbool(do_not_unlock_if_synchronized, false);
934
935 // check for synchronized methods
936 // Must happen AFTER invocation_counter check and stack overflow check,
937 // so method is not locked if overflows.
938 //
939 if (synchronized) {
940 lock_method();
941 } else {
942 // no synchronization necessary
943 #ifdef ASSERT
944 { Label L;
945 __ movl(rax, access_flags);
946 __ testl(rax, JVM_ACC_SYNCHRONIZED);
947 __ jcc(Assembler::zero, L);
948 __ stop("method needs synchronization");
949 __ bind(L);
950 }
951 #endif
952 }
953
954 // start execution
955 #ifdef ASSERT
956 { Label L;
957 const Address monitor_block_top (rbp,
958 frame::interpreter_frame_monitor_block_top_offset * wordSize);
959 __ movptr(rax, monitor_block_top);
960 __ cmpptr(rax, rsp);
961 __ jcc(Assembler::equal, L);
962 __ stop("broken stack frame setup in interpreter");
963 __ bind(L);
964 }
965 #endif
966
967 // jvmti/dtrace support
968 __ notify_method_entry();
969
970 // work registers
971 const Register method = rbx;
972 const Register thread = rdi;
973 const Register t = rcx;
974
975 // allocate space for parameters
976 __ get_method(method);
977 __ movptr(t, Address(method, Method::const_offset()));
978 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
979
980 __ shlptr(t, Interpreter::logStackElementSize);
981 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
982 __ subptr(rsp, t);
983 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
984
985 // get signature handler
986 { Label L;
987 __ movptr(t, Address(method, Method::signature_handler_offset()));
988 __ testptr(t, t);
989 __ jcc(Assembler::notZero, L);
990 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
991 __ get_method(method);
992 __ movptr(t, Address(method, Method::signature_handler_offset()));
993 __ bind(L);
994 }
995
996 // call signature handler
997 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
998 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
999 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1000 // The generated handlers do not touch RBX (the method oop).
1001 // However, large signatures cannot be cached and are generated
1002 // each time here. The slow-path generator will blow RBX
1003 // sometime, so we must reload it after the call.
1004 __ call(t);
1005 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
1006
1007 // result handler is in rax,
1008 // set result handler
1009 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
1010
1011 // pass mirror handle if static call
1012 { Label L;
1013 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1014 __ movl(t, Address(method, Method::access_flags_offset()));
1015 __ testl(t, JVM_ACC_STATIC);
1016 __ jcc(Assembler::zero, L);
1017 // get mirror
1018 __ movptr(t, Address(method, Method:: const_offset()));
1019 __ movptr(t, Address(t, ConstMethod::constants_offset()));
1020 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1021 __ movptr(t, Address(t, mirror_offset));
1022 // copy mirror into activation frame
1023 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
1024 // pass handle to mirror
1025 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1026 __ movptr(Address(rsp, wordSize), t);
1027 __ bind(L);
1028 }
1029
1030 // get native function entry point
1031 { Label L;
1032 __ movptr(rax, Address(method, Method::native_function_offset()));
1033 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1034 __ cmpptr(rax, unsatisfied.addr());
1035 __ jcc(Assembler::notEqual, L);
1036 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1037 __ get_method(method);
1038 __ movptr(rax, Address(method, Method::native_function_offset()));
1039 __ bind(L);
1040 }
1041
1042 // pass JNIEnv
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());
1051
1052 // change thread state
1053 #ifdef ASSERT
1054 { Label L;
1055 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1056 __ cmpl(t, _thread_in_Java);
1057 __ jcc(Assembler::equal, L);
1058 __ stop("Wrong thread state in native stub");
1059 __ bind(L);
1060 }
1061 #endif
1062
1063 // Change state to native
1064 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1065 __ call(rax);
1066
1067 // result potentially in rdx:rax or ST0
1068
1069 // Verify or restore cpu control state after JNI call
1070 __ restore_cpu_control_state_after_jni();
1071
1072 // save potential result in ST(0) & rdx:rax
1073 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1074 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1075 // It is safe to do this push because state is _thread_in_native and return address will be found
1076 // via _last_native_pc and not via _last_jave_sp
1077
1078 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1079 // If the order changes or anything else is added to the stack the code in
1080 // interpreter_frame_result will have to be changed.
1081
1082 { Label L;
1083 Label push_double;
1084 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1085 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1086 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1087 float_handler.addr());
1088 __ jcc(Assembler::equal, push_double);
1089 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1090 double_handler.addr());
1091 __ jcc(Assembler::notEqual, L);
1092 __ bind(push_double);
1093 __ push(dtos);
1094 __ bind(L);
1095 }
1096 __ push(ltos);
1097
1098 // change thread state
1099 __ get_thread(thread);
1100 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1101 if(os::is_MP()) {
1102 if (UseMembar) {
1103 // Force this write out before the read below
1104 __ membar(Assembler::Membar_mask_bits(
1105 Assembler::LoadLoad | Assembler::LoadStore |
1106 Assembler::StoreLoad | Assembler::StoreStore));
1107 } else {
1108 // Write serialization page so VM thread can do a pseudo remote membar.
1109 // We use the current thread pointer to calculate a thread specific
1110 // offset to write to within the page. This minimizes bus traffic
1111 // due to cache line collision.
1112 __ serialize_memory(thread, rcx);
1113 }
1114 }
1115
1116 if (AlwaysRestoreFPU) {
1117 // Make sure the control word is correct.
1118 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1119 }
1120
1121 // check for safepoint operation in progress and/or pending suspend requests
1122 { Label Continue;
1123
1124 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1125 SafepointSynchronize::_not_synchronized);
1126
1127 Label L;
1128 __ jcc(Assembler::notEqual, L);
1129 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1130 __ jcc(Assembler::equal, Continue);
1131 __ bind(L);
1132
1133 // Don't use call_VM as it will see a possible pending exception and forward it
1134 // and never return here preventing us from clearing _last_native_pc down below.
1135 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1136 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1137 // by hand.
1138 //
1139 __ push(thread);
1140 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1141 JavaThread::check_special_condition_for_native_trans)));
1142 __ increment(rsp, wordSize);
1143 __ get_thread(thread);
1144
1145 __ bind(Continue);
1146 }
1147
1148 // change thread state
1149 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1150
1151 __ reset_last_Java_frame(thread, true, true);
1152
1153 // reset handle block
1154 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1155 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1156
1157 // If result was an oop then unbox and save it in the frame
1158 { Label L;
1159 Label no_oop, store_result;
1160 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1161 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1162 handler.addr());
1163 __ jcc(Assembler::notEqual, no_oop);
1164 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1165 __ pop(ltos);
1166 __ testptr(rax, rax);
1167 __ jcc(Assembler::zero, store_result);
1168 // unbox
1169 __ movptr(rax, Address(rax, 0));
1170 __ bind(store_result);
1171 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1172 // keep stack depth as expected by pushing oop which will eventually be discarded
1173 __ push(ltos);
1174 __ bind(no_oop);
1175 }
1176
1177 {
1178 Label no_reguard;
1179 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1180 __ jcc(Assembler::notEqual, no_reguard);
1181
1182 __ pusha();
1183 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1184 __ popa();
1185
1186 __ bind(no_reguard);
1187 }
1188
1189 // restore rsi to have legal interpreter frame,
1190 // i.e., bci == 0 <=> rsi == code_base()
1191 // Can't call_VM until bcp is within reasonable.
1192 __ get_method(method); // method is junk from thread_in_native to now.
1193 __ movptr(rsi, Address(method,Method::const_offset())); // get ConstMethod*
1194 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
1195
1196 // handle exceptions (exception handling will handle unlocking!)
1197 { Label L;
1198 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1199 __ jcc(Assembler::zero, L);
1200 // Note: At some point we may want to unify this with the code used in call_VM_base();
1201 // i.e., we should use the StubRoutines::forward_exception code. For now this
1202 // doesn't work here because the rsp is not correctly set at this point.
1203 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1204 __ should_not_reach_here();
1205 __ bind(L);
1206 }
1207
1208 // do unlocking if necessary
1209 { Label L;
1210 __ movl(t, Address(method, Method::access_flags_offset()));
1211 __ testl(t, JVM_ACC_SYNCHRONIZED);
1212 __ jcc(Assembler::zero, L);
1213 // the code below should be shared with interpreter macro assembler implementation
1214 { Label unlock;
1215 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1216 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1217 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1218
1219 __ lea(rdx, monitor); // address of first monitor
1220
1221 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1222 __ testptr(t, t);
1223 __ jcc(Assembler::notZero, unlock);
1224
1225 // Entry already unlocked, need to throw exception
1226 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1227 __ should_not_reach_here();
1228
1229 __ bind(unlock);
1230 __ unlock_object(rdx);
1231 }
1232 __ bind(L);
1233 }
1234
1235 // jvmti/dtrace support
1236 // Note: This must happen _after_ handling/throwing any exceptions since
1237 // the exception handler code notifies the runtime of method exits
1238 // too. If this happens before, method entry/exit notifications are
1239 // not properly paired (was bug - gri 11/22/99).
1240 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1241
1242 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1243 __ pop(ltos);
1244 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1245 __ call(t);
1246
1247 // remove activation
1248 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1249 __ leave(); // remove frame anchor
1250 __ pop(rdi); // get return address
1251 __ mov(rsp, t); // set sp to sender sp
1252 __ jmp(rdi);
1253
1254 if (inc_counter) {
1255 // Handle overflow of counter and compile method
1256 __ bind(invocation_counter_overflow);
1257 generate_counter_overflow(&continue_after_compile);
1258 }
1259
1260 return entry_point;
1261 }
1262
1263 //
1264 // Generic interpreted method entry to (asm) interpreter
1265 //
1266 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1267 // determine code generation flags
1268 bool inc_counter = UseCompiler || CountCompiledCalls;
1269
1270 // rbx,: Method*
1271 // rsi: sender sp
1272 address entry_point = __ pc();
1273
1274 const Address constMethod (rbx, Method::const_offset());
1275 const Address access_flags (rbx, Method::access_flags_offset());
1276 const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
1277 const Address size_of_locals (rdx, ConstMethod::size_of_locals_offset());
1278
1279 // get parameter size (always needed)
1280 __ movptr(rdx, constMethod);
1281 __ load_unsigned_short(rcx, size_of_parameters);
1282
1283 // rbx,: Method*
1284 // rcx: size of parameters
1285
1286 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1287
1288 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1289 __ subl(rdx, rcx); // rdx = no. of additional locals
1290
1291 // see if we've got enough room on the stack for locals plus overhead.
1292 generate_stack_overflow_check();
1293
1294 // get return address
1295 __ pop(rax);
1296
1297 // compute beginning of parameters (rdi)
1298 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1299
1300 // rdx - # of additional locals
1301 // allocate space for locals
1302 // explicitly initialize locals
1303 {
1304 Label exit, loop;
1305 __ testl(rdx, rdx);
1306 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1307 __ bind(loop);
1308 __ push((int32_t)NULL_WORD); // initialize local variables
1309 __ decrement(rdx); // until everything initialized
1310 __ jcc(Assembler::greater, loop);
1311 __ bind(exit);
1312 }
1313
1314 // initialize fixed part of activation frame
1315 generate_fixed_frame(false);
1316
1317 // make sure method is not native & not abstract
1318 #ifdef ASSERT
1319 __ movl(rax, access_flags);
1320 {
1321 Label L;
1322 __ testl(rax, JVM_ACC_NATIVE);
1323 __ jcc(Assembler::zero, L);
1324 __ stop("tried to execute native method as non-native");
1325 __ bind(L);
1326 }
1327 { Label L;
1328 __ testl(rax, JVM_ACC_ABSTRACT);
1329 __ jcc(Assembler::zero, L);
1330 __ stop("tried to execute abstract method in interpreter");
1331 __ bind(L);
1332 }
1333 #endif
1334
1335 // Since at this point in the method invocation the exception handler
1336 // would try to exit the monitor of synchronized methods which hasn't
1337 // been entered yet, we set the thread local variable
1338 // _do_not_unlock_if_synchronized to true. The remove_activation will
1339 // check this flag.
1340
1341 __ get_thread(rax);
1342 const Address do_not_unlock_if_synchronized(rax,
1343 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1344 __ movbool(do_not_unlock_if_synchronized, true);
1345
1346 __ profile_parameters_type(rax, rcx, rdx);
1347 // increment invocation count & check for overflow
1348 Label invocation_counter_overflow;
1349 Label profile_method;
1350 Label profile_method_continue;
1351 if (inc_counter) {
1352 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1353 if (ProfileInterpreter) {
1354 __ bind(profile_method_continue);
1355 }
1356 }
1357 Label continue_after_compile;
1358 __ bind(continue_after_compile);
1359
1360 bang_stack_shadow_pages(false);
1361
1362 // reset the _do_not_unlock_if_synchronized flag
1363 __ get_thread(rax);
1364 __ movbool(do_not_unlock_if_synchronized, false);
1365
1366 // check for synchronized methods
1367 // Must happen AFTER invocation_counter check and stack overflow check,
1368 // so method is not locked if overflows.
1369 //
1370 if (synchronized) {
1371 // Allocate monitor and lock method
1372 lock_method();
1373 } else {
1374 // no synchronization necessary
1375 #ifdef ASSERT
1376 { Label L;
1377 __ movl(rax, access_flags);
1378 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1379 __ jcc(Assembler::zero, L);
1380 __ stop("method needs synchronization");
1381 __ bind(L);
1382 }
1383 #endif
1384 }
1385
1386 // start execution
1387 #ifdef ASSERT
1388 { Label L;
1389 const Address monitor_block_top (rbp,
1390 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1391 __ movptr(rax, monitor_block_top);
1392 __ cmpptr(rax, rsp);
1393 __ jcc(Assembler::equal, L);
1394 __ stop("broken stack frame setup in interpreter");
1395 __ bind(L);
1396 }
1397 #endif
1398
1399 // jvmti support
1400 __ notify_method_entry();
1401
1402 __ dispatch_next(vtos);
1403
1404 // invocation counter overflow
1405 if (inc_counter) {
1406 if (ProfileInterpreter) {
1407 // We have decided to profile this method in the interpreter
1408 __ bind(profile_method);
1409 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1410 __ set_method_data_pointer_for_bcp();
1411 __ get_method(rbx);
1412 __ jmp(profile_method_continue);
1413 }
1414 // Handle overflow of counter and compile method
1415 __ bind(invocation_counter_overflow);
1416 generate_counter_overflow(&continue_after_compile);
1417 }
1418
1419 return entry_point;
1420 }
1421
1422
1423 // These should never be compiled since the interpreter will prefer
1424 // the compiled version to the intrinsic version.
1425 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1426 switch (method_kind(m)) {
1427 case Interpreter::java_lang_math_sin : // fall thru
1428 case Interpreter::java_lang_math_cos : // fall thru
1429 case Interpreter::java_lang_math_tan : // fall thru
1430 case Interpreter::java_lang_math_abs : // fall thru
1431 case Interpreter::java_lang_math_log : // fall thru
1432 case Interpreter::java_lang_math_log10 : // fall thru
1433 case Interpreter::java_lang_math_sqrt : // fall thru
1434 case Interpreter::java_lang_math_pow : // fall thru
1435 case Interpreter::java_lang_math_exp :
1436 return false;
1437 default:
1438 return true;
1439 }
1440 }
1441
1442 // How much stack a method activation needs in words.
1443 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1444
1445 const int stub_code = 4; // see generate_call_stub
1446 // Save space for one monitor to get into the interpreted method in case
1447 // the method is synchronized
1448 int monitor_size = method->is_synchronized() ?
1449 1*frame::interpreter_frame_monitor_size() : 0;
1450
1451 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1452 // be sure to change this if you add/subtract anything to/from the overhead area
1453 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1454
1455 const int method_stack = (method->max_locals() + method->max_stack()) *
1456 Interpreter::stackElementWords;
1457 return overhead_size + method_stack + stub_code;
1458 }
1459
1460 //------------------------------------------------------------------------------------------------------------------------
1461 // Exceptions
1462
1463 void TemplateInterpreterGenerator::generate_throw_exception() {
1464 // Entry point in previous activation (i.e., if the caller was interpreted)
1465 Interpreter::_rethrow_exception_entry = __ pc();
1466 const Register thread = rcx;
1467
1468 // Restore sp to interpreter_frame_last_sp even though we are going
1469 // to empty the expression stack for the exception processing.
1470 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1471 // rax,: exception
1472 // rdx: return address/pc that threw exception
1473 __ restore_bcp(); // rsi points to call/send
1474 __ restore_locals();
1475
1476 // Entry point for exceptions thrown within interpreter code
1477 Interpreter::_throw_exception_entry = __ pc();
1478 // expression stack is undefined here
1479 // rax,: exception
1480 // rsi: exception bcp
1481 __ verify_oop(rax);
1482
1483 // expression stack must be empty before entering the VM in case of an exception
1484 __ empty_expression_stack();
1485 __ empty_FPU_stack();
1486 // find exception handler address and preserve exception oop
1487 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1488 // rax,: exception handler entry point
1489 // rdx: preserved exception oop
1490 // rsi: bcp for exception handler
1491 __ push_ptr(rdx); // push exception which is now the only value on the stack
1492 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1493
1494 // If the exception is not handled in the current frame the frame is removed and
1495 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1496 //
1497 // Note: At this point the bci is still the bxi for the instruction which caused
1498 // the exception and the expression stack is empty. Thus, for any VM calls
1499 // at this point, GC will find a legal oop map (with empty expression stack).
1500
1501 // In current activation
1502 // tos: exception
1503 // rsi: exception bcp
1504
1505 //
1506 // JVMTI PopFrame support
1507 //
1508
1509 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1510 __ empty_expression_stack();
1511 __ empty_FPU_stack();
1512 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1513 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1514 // popframe handling cycles.
1515 __ get_thread(thread);
1516 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1517 __ orl(rdx, JavaThread::popframe_processing_bit);
1518 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1519
1520 {
1521 // Check to see whether we are returning to a deoptimized frame.
1522 // (The PopFrame call ensures that the caller of the popped frame is
1523 // either interpreted or compiled and deoptimizes it if compiled.)
1524 // In this case, we can't call dispatch_next() after the frame is
1525 // popped, but instead must save the incoming arguments and restore
1526 // them after deoptimization has occurred.
1527 //
1528 // Note that we don't compare the return PC against the
1529 // deoptimization blob's unpack entry because of the presence of
1530 // adapter frames in C2.
1531 Label caller_not_deoptimized;
1532 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1533 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1534 __ testl(rax, rax);
1535 __ jcc(Assembler::notZero, caller_not_deoptimized);
1536
1537 // Compute size of arguments for saving when returning to deoptimized caller
1538 __ get_method(rax);
1539 __ movptr(rax, Address(rax, Method::const_offset()));
1540 __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset()));
1541 __ shlptr(rax, Interpreter::logStackElementSize);
1542 __ restore_locals();
1543 __ subptr(rdi, rax);
1544 __ addptr(rdi, wordSize);
1545 // Save these arguments
1546 __ get_thread(thread);
1547 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1548
1549 __ remove_activation(vtos, rdx,
1550 /* throw_monitor_exception */ false,
1551 /* install_monitor_exception */ false,
1552 /* notify_jvmdi */ false);
1553
1554 // Inform deoptimization that it is responsible for restoring these arguments
1555 __ get_thread(thread);
1556 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1557
1558 // Continue in deoptimization handler
1559 __ jmp(rdx);
1560
1561 __ bind(caller_not_deoptimized);
1562 }
1563
1564 __ remove_activation(vtos, rdx,
1565 /* throw_monitor_exception */ false,
1566 /* install_monitor_exception */ false,
1567 /* notify_jvmdi */ false);
1568
1569 // Finish with popframe handling
1570 // A previous I2C followed by a deoptimization might have moved the
1571 // outgoing arguments further up the stack. PopFrame expects the
1572 // mutations to those outgoing arguments to be preserved and other
1573 // constraints basically require this frame to look exactly as
1574 // though it had previously invoked an interpreted activation with
1575 // no space between the top of the expression stack (current
1576 // last_sp) and the top of stack. Rather than force deopt to
1577 // maintain this kind of invariant all the time we call a small
1578 // fixup routine to move the mutated arguments onto the top of our
1579 // expression stack if necessary.
1580 __ mov(rax, rsp);
1581 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1582 __ get_thread(thread);
1583 // PC must point into interpreter here
1584 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1585 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1586 __ get_thread(thread);
1587 __ reset_last_Java_frame(thread, true, true);
1588 // Restore the last_sp and null it out
1589 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1590 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1591
1592 __ restore_bcp();
1593 __ restore_locals();
1594 // The method data pointer was incremented already during
1595 // call profiling. We have to restore the mdp for the current bcp.
1596 if (ProfileInterpreter) {
1597 __ set_method_data_pointer_for_bcp();
1598 }
1599
1600 // Clear the popframe condition flag
1601 __ get_thread(thread);
1602 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1603
1604 #if INCLUDE_JVMTI
1605 {
1606 Label L_done;
1607 const Register local0 = rdi;
1608
1609 __ cmpb(Address(rsi, 0), Bytecodes::_invokestatic);
1610 __ jcc(Assembler::notEqual, L_done);
1611
1612 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1613 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1614
1615 __ get_method(rdx);
1616 __ movptr(rax, Address(local0, 0));
1617 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rsi);
1618
1619 __ testptr(rax, rax);
1620 __ jcc(Assembler::zero, L_done);
1621
1622 __ movptr(Address(rbx, 0), rax);
1623 __ bind(L_done);
1624 }
1625 #endif // INCLUDE_JVMTI
1626
1627 __ dispatch_next(vtos);
1628 // end of PopFrame support
1629
1630 Interpreter::_remove_activation_entry = __ pc();
1631
1632 // preserve exception over this code sequence
1633 __ pop_ptr(rax);
1634 __ get_thread(thread);
1635 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1636 // remove the activation (without doing throws on illegalMonitorExceptions)
1637 __ remove_activation(vtos, rdx, false, true, false);
1638 // restore exception
1639 __ get_thread(thread);
1640 __ get_vm_result(rax, thread);
1641
1642 // Inbetween activations - previous activation type unknown yet
1643 // compute continuation point - the continuation point expects
1644 // the following registers set up:
1645 //
1646 // rax: exception
1647 // rdx: return address/pc that threw exception
1648 // rsp: expression stack of caller
1649 // rbp: rbp, of caller
1650 __ push(rax); // save exception
1651 __ push(rdx); // save return address
1652 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1653 __ mov(rbx, rax); // save exception handler
1654 __ pop(rdx); // restore return address
1655 __ pop(rax); // restore exception
1656 // Note that an "issuing PC" is actually the next PC after the call
1657 __ jmp(rbx); // jump to exception handler of caller
1658 }
1659
1660
1661 //
1662 // JVMTI ForceEarlyReturn support
1663 //
1664 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1665 address entry = __ pc();
1666 const Register thread = rcx;
1667
1668 __ restore_bcp();
1669 __ restore_locals();
1670 __ empty_expression_stack();
1671 __ empty_FPU_stack();
1672 __ load_earlyret_value(state);
1673
1674 __ get_thread(thread);
1675 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1676 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1677
1678 // Clear the earlyret state
1679 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1680
1681 __ remove_activation(state, rsi,
1682 false, /* throw_monitor_exception */
1683 false, /* install_monitor_exception */
1684 true); /* notify_jvmdi */
1685 __ jmp(rsi);
1686 return entry;
1687 } // end of ForceEarlyReturn support
1688
1689
1690 //------------------------------------------------------------------------------------------------------------------------
1691 // Helper for vtos entry point generation
1692
1693 void TemplateInterpreterGenerator::set_vtos_entry_points (Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) {
1694 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1695 Label L;
1696 fep = __ pc(); __ push(ftos); __ jmp(L);
1697 dep = __ pc(); __ push(dtos); __ jmp(L);
1698 lep = __ pc(); __ push(ltos); __ jmp(L);
1699 aep = __ pc(); __ push(atos); __ jmp(L);
1700 bep = cep = sep = // fall through
1701 iep = __ pc(); __ push(itos); // fall through
1702 vep = __ pc(); __ bind(L); // fall through
1703 generate_and_dispatch(t);
1704 }
1705
1706 //------------------------------------------------------------------------------------------------------------------------
1707 // Generation of individual instructions
1708
1709 // helpers for generate_and_dispatch
1710
1711
1712
1713 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1714 : TemplateInterpreterGenerator(code) {
1715 generate_all(); // down here so it can be "virtual"
1716 }
1717
1718 //------------------------------------------------------------------------------------------------------------------------
1719
1720 // Non-product code
1721 #ifndef PRODUCT
1722 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1723 address entry = __ pc();
1724
1725 // prepare expression stack
1726 __ pop(rcx); // pop return address so expression stack is 'pure'
1727 __ push(state); // save tosca
1728
1729 // pass tosca registers as arguments & call tracer
1730 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1731 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1732 __ pop(state); // restore tosca
1733
1734 // return
1735 __ jmp(rcx);
1736
1737 return entry;
1738 }
1739
1740
1741 void TemplateInterpreterGenerator::count_bytecode() {
1742 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1743 }
1744
1745
1746 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1747 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1748 }
1749
1750
1751 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1752 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1753 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1754 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1755 ExternalAddress table((address) BytecodePairHistogram::_counters);
1756 Address index(noreg, rbx, Address::times_4);
1757 __ incrementl(ArrayAddress(table, index));
1758 }
1759
1760
1761 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1762 // Call a little run-time stub to avoid blow-up for each bytecode.
1763 // The run-time runtime saves the right registers, depending on
1764 // the tosca in-state for the given template.
1765 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1766 "entry must have been generated");
1767 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1768 }
1769
1770
1771 void TemplateInterpreterGenerator::stop_interpreter_at() {
1772 Label L;
1773 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1774 StopInterpreterAt);
1775 __ jcc(Assembler::notEqual, L);
1776 __ int3();
1777 __ bind(L);
1778 }
1779 #endif // !PRODUCT
1780 #endif // CC_INTERP