1 /*
2 * Copyright (c) 2003, 2017, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2013, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.hpp"
28 #include "interpreter/bytecodeHistogram.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterGenerator.hpp"
31 #include "interpreter/interpreterRuntime.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "interpreter/bytecodeTracer.hpp"
34 #include "oops/arrayOop.hpp"
35 #include "oops/methodData.hpp"
36 #include "oops/method.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "prims/jvmtiExport.hpp"
39 #include "prims/jvmtiThreadState.hpp"
40 #include "runtime/arguments.hpp"
41 #include "runtime/deoptimization.hpp"
42 #include "runtime/frame.inline.hpp"
43 #include "runtime/sharedRuntime.hpp"
44 #include "runtime/stubRoutines.hpp"
45 #include "runtime/synchronizer.hpp"
46 #include "runtime/timer.hpp"
47 #include "runtime/vframeArray.hpp"
48 #include "utilities/debug.hpp"
49 #include <sys/types.h>
50
51 #ifndef PRODUCT
52 #include "oops/method.hpp"
53 #endif // !PRODUCT
54
55 #define __ _masm->
56
57 #ifndef CC_INTERP
58
59 //-----------------------------------------------------------------------------
60
61 extern "C" void entry(CodeBuffer*);
62
63 //-----------------------------------------------------------------------------
64
generate_StackOverflowError_handler()65 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
66 address entry = __ pc();
67
68 #ifdef ASSERT
69 {
70 Label L;
71 __ ldr(rscratch1, Address(rfp,
72 frame::interpreter_frame_monitor_block_top_offset *
73 wordSize));
74 __ mov(rscratch2, sp);
75 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
76 // grows negative)
77 __ br(Assembler::HS, L); // check if frame is complete
78 __ stop ("interpreter frame not set up");
79 __ bind(L);
80 }
81 #endif // ASSERT
82 // Restore bcp under the assumption that the current frame is still
83 // interpreted
84 __ restore_bcp();
85
86 // expression stack must be empty before entering the VM if an
87 // exception happened
88 __ empty_expression_stack();
89 // throw exception
90 __ call_VM(noreg,
91 CAST_FROM_FN_PTR(address,
92 InterpreterRuntime::throw_StackOverflowError));
93 return entry;
94 }
95
generate_ArrayIndexOutOfBounds_handler(const char * name)96 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
97 const char* name) {
98 address entry = __ pc();
99 // expression stack must be empty before entering the VM if an
100 // exception happened
101 __ empty_expression_stack();
102 // setup parameters
103 // ??? convention: expect aberrant index in register r1
104 __ movw(c_rarg2, r1);
105 __ mov(c_rarg1, (address)name);
106 __ call_VM(noreg,
107 CAST_FROM_FN_PTR(address,
108 InterpreterRuntime::
109 throw_ArrayIndexOutOfBoundsException),
110 c_rarg1, c_rarg2);
111 return entry;
112 }
113
generate_ClassCastException_handler()114 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
115 address entry = __ pc();
116
117 // object is at TOS
118 __ pop(c_rarg1);
119
120 // expression stack must be empty before entering the VM if an
121 // exception happened
122 __ empty_expression_stack();
123
124 __ call_VM(noreg,
125 CAST_FROM_FN_PTR(address,
126 InterpreterRuntime::
127 throw_ClassCastException),
128 c_rarg1);
129 return entry;
130 }
131
generate_exception_handler_common(const char * name,const char * message,bool pass_oop)132 address TemplateInterpreterGenerator::generate_exception_handler_common(
133 const char* name, const char* message, bool pass_oop) {
134 assert(!pass_oop || message == NULL, "either oop or message but not both");
135 address entry = __ pc();
136 if (pass_oop) {
137 // object is at TOS
138 __ pop(c_rarg2);
139 }
140 // expression stack must be empty before entering the VM if an
141 // exception happened
142 __ empty_expression_stack();
143 // setup parameters
144 __ lea(c_rarg1, Address((address)name));
145 if (pass_oop) {
146 __ call_VM(r0, CAST_FROM_FN_PTR(address,
147 InterpreterRuntime::
148 create_klass_exception),
149 c_rarg1, c_rarg2);
150 } else {
151 // kind of lame ExternalAddress can't take NULL because
152 // external_word_Relocation will assert.
153 if (message != NULL) {
154 __ lea(c_rarg2, Address((address)message));
155 } else {
156 __ mov(c_rarg2, NULL_WORD);
157 }
158 __ call_VM(r0,
159 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
160 c_rarg1, c_rarg2);
161 }
162 // throw exception
163 __ b(address(Interpreter::throw_exception_entry()));
164 return entry;
165 }
166
generate_continuation_for(TosState state)167 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
168 address entry = __ pc();
169 // NULL last_sp until next java call
170 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
171 __ dispatch_next(state);
172 return entry;
173 }
174
generate_return_entry_for(TosState state,int step,size_t index_size)175 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
176 address entry = __ pc();
177
178 // Restore stack bottom in case i2c adjusted stack
179 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
180 // and NULL it as marker that esp is now tos until next java call
181 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
182 __ restore_bcp();
183 __ restore_locals();
184 __ restore_constant_pool_cache();
185 __ get_method(rmethod);
186
187 // Pop N words from the stack
188 __ get_cache_and_index_at_bcp(r1, r2, 1, index_size);
189 __ ldr(r1, Address(r1, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
190 __ andr(r1, r1, ConstantPoolCacheEntry::parameter_size_mask);
191
192 __ add(esp, esp, r1, Assembler::LSL, 3);
193
194 // Restore machine SP
195 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
196 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
197 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
198 __ ldr(rscratch2,
199 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
200 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
201 __ andr(sp, rscratch1, -16);
202
203 __ get_dispatch();
204 __ dispatch_next(state, step);
205
206 return entry;
207 }
208
generate_deopt_entry_for(TosState state,int step)209 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
210 int step) {
211 address entry = __ pc();
212 __ restore_bcp();
213 __ restore_locals();
214 __ restore_constant_pool_cache();
215 __ get_method(rmethod);
216 __ get_dispatch();
217
218 // Calculate stack limit
219 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
220 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
221 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size()
222 + (EnableInvokeDynamic ? 2 : 0));
223 __ ldr(rscratch2,
224 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
225 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3);
226 __ andr(sp, rscratch1, -16);
227
228 // Restore expression stack pointer
229 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
230 // NULL last_sp until next java call
231 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
232
233 // handle exceptions
234 {
235 Label L;
236 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
237 __ cbz(rscratch1, L);
238 __ call_VM(noreg,
239 CAST_FROM_FN_PTR(address,
240 InterpreterRuntime::throw_pending_exception));
241 __ should_not_reach_here();
242 __ bind(L);
243 }
244
245 __ dispatch_next(state, step);
246 return entry;
247 }
248
249
BasicType_as_index(BasicType type)250 int AbstractInterpreter::BasicType_as_index(BasicType type) {
251 int i = 0;
252 switch (type) {
253 case T_BOOLEAN: i = 0; break;
254 case T_CHAR : i = 1; break;
255 case T_BYTE : i = 2; break;
256 case T_SHORT : i = 3; break;
257 case T_INT : i = 4; break;
258 case T_LONG : i = 5; break;
259 case T_VOID : i = 6; break;
260 case T_FLOAT : i = 7; break;
261 case T_DOUBLE : i = 8; break;
262 case T_OBJECT : i = 9; break;
263 case T_ARRAY : i = 9; break;
264 default : ShouldNotReachHere();
265 }
266 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
267 "index out of bounds");
268 return i;
269 }
270
271
generate_result_handler_for(BasicType type)272 address TemplateInterpreterGenerator::generate_result_handler_for(
273 BasicType type) {
274 address entry = __ pc();
275 switch (type) {
276 case T_BOOLEAN: __ c2bool(r0); break;
277 case T_CHAR : __ uxth(r0, r0); break;
278 case T_BYTE : __ sxtb(r0, r0); break;
279 case T_SHORT : __ sxth(r0, r0); break;
280 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this
281 case T_LONG : /* nothing to do */ break;
282 case T_VOID : /* nothing to do */ break;
283 case T_FLOAT : /* nothing to do */ break;
284 case T_DOUBLE : /* nothing to do */ break;
285 case T_OBJECT :
286 // retrieve result from frame
287 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
288 // and verify it
289 __ verify_oop(r0);
290 break;
291 default : ShouldNotReachHere();
292 }
293 __ ret(lr); // return from result handler
294 return entry;
295 }
296
generate_safept_entry_for(TosState state,address runtime_entry)297 address TemplateInterpreterGenerator::generate_safept_entry_for(
298 TosState state,
299 address runtime_entry) {
300 address entry = __ pc();
301 __ push(state);
302 __ call_VM(noreg, runtime_entry);
303 __ membar(Assembler::AnyAny);
304 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
305 return entry;
306 }
307
308 // Helpers for commoning out cases in the various type of method entries.
309 //
310
311
312 // increment invocation count & check for overflow
313 //
314 // Note: checking for negative value instead of overflow
315 // so we have a 'sticky' overflow test
316 //
317 // rmethod: method
318 //
generate_counter_incr(Label * overflow,Label * profile_method,Label * profile_method_continue)319 void InterpreterGenerator::generate_counter_incr(
320 Label* overflow,
321 Label* profile_method,
322 Label* profile_method_continue) {
323 Label done;
324 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
325 if (TieredCompilation) {
326 int increment = InvocationCounter::count_increment;
327 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
328 Label no_mdo;
329 if (ProfileInterpreter) {
330 // Are we profiling?
331 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
332 __ cbz(r0, no_mdo);
333 // Increment counter in the MDO
334 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
335 in_bytes(InvocationCounter::counter_offset()));
336 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
337 __ b(done);
338 }
339 __ bind(no_mdo);
340 // Increment counter in MethodCounters
341 const Address invocation_counter(rscratch2,
342 MethodCounters::invocation_counter_offset() +
343 InvocationCounter::counter_offset());
344 __ get_method_counters(rmethod, rscratch2, done);
345 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
346 __ bind(done);
347 } else {
348 const Address backedge_counter(rscratch2,
349 MethodCounters::backedge_counter_offset() +
350 InvocationCounter::counter_offset());
351 const Address invocation_counter(rscratch2,
352 MethodCounters::invocation_counter_offset() +
353 InvocationCounter::counter_offset());
354
355 __ get_method_counters(rmethod, rscratch2, done);
356
357 if (ProfileInterpreter) { // %%% Merge this into MethodData*
358 __ ldrw(r1, Address(rscratch2, MethodCounters::interpreter_invocation_counter_offset()));
359 __ addw(r1, r1, 1);
360 __ strw(r1, Address(rscratch2, MethodCounters::interpreter_invocation_counter_offset()));
361 }
362 // Update standard invocation counters
363 __ ldrw(r1, invocation_counter);
364 __ ldrw(r0, backedge_counter);
365
366 __ addw(r1, r1, InvocationCounter::count_increment);
367 __ andw(r0, r0, InvocationCounter::count_mask_value);
368
369 __ strw(r1, invocation_counter);
370 __ addw(r0, r0, r1); // add both counters
371
372 // profile_method is non-null only for interpreted method so
373 // profile_method != NULL == !native_call
374
375 if (ProfileInterpreter && profile_method != NULL) {
376 // Test to see if we should create a method data oop
377 unsigned long offset;
378 __ adrp(rscratch2, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit),
379 offset);
380 __ ldrw(rscratch2, Address(rscratch2, offset));
381 __ cmp(r0, rscratch2);
382 __ br(Assembler::LT, *profile_method_continue);
383
384 // if no method data exists, go to profile_method
385 __ test_method_data_pointer(rscratch2, *profile_method);
386 }
387
388 {
389 unsigned long offset;
390 __ adrp(rscratch2,
391 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit),
392 offset);
393 __ ldrw(rscratch2, Address(rscratch2, offset));
394 __ cmpw(r0, rscratch2);
395 __ br(Assembler::HS, *overflow);
396 }
397 __ bind(done);
398 }
399 }
400
generate_counter_overflow(Label * do_continue)401 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
402
403 // Asm interpreter on entry
404 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
405 // Everything as it was on entry
406
407 // InterpreterRuntime::frequency_counter_overflow takes two
408 // arguments, the first (thread) is passed by call_VM, the second
409 // indicates if the counter overflow occurs at a backwards branch
410 // (NULL bcp). We pass zero for it. The call returns the address
411 // of the verified entry point for the method or NULL if the
412 // compilation did not complete (either went background or bailed
413 // out).
414 __ mov(c_rarg1, 0);
415 __ call_VM(noreg,
416 CAST_FROM_FN_PTR(address,
417 InterpreterRuntime::frequency_counter_overflow),
418 c_rarg1);
419
420 __ b(*do_continue);
421 }
422
423 // See if we've got enough room on the stack for locals plus overhead.
424 // The expression stack grows down incrementally, so the normal guard
425 // page mechanism will work for that.
426 //
427 // NOTE: Since the additional locals are also always pushed (wasn't
428 // obvious in generate_method_entry) so the guard should work for them
429 // too.
430 //
431 // Args:
432 // r3: number of additional locals this frame needs (what we must check)
433 // rmethod: Method*
434 //
435 // Kills:
436 // r0
generate_stack_overflow_check(void)437 void InterpreterGenerator::generate_stack_overflow_check(void) {
438
439 // monitor entry size: see picture of stack set
440 // (generate_method_entry) and frame_amd64.hpp
441 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
442
443 // total overhead size: entry_size + (saved rbp through expr stack
444 // bottom). be sure to change this if you add/subtract anything
445 // to/from the overhead area
446 const int overhead_size =
447 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
448
449 const int page_size = os::vm_page_size();
450
451 Label after_frame_check;
452
453 // see if the frame is greater than one page in size. If so,
454 // then we need to verify there is enough stack space remaining
455 // for the additional locals.
456 //
457 // Note that we use SUBS rather than CMP here because the immediate
458 // field of this instruction may overflow. SUBS can cope with this
459 // because it is a macro that will expand to some number of MOV
460 // instructions and a register operation.
461 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
462 __ br(Assembler::LS, after_frame_check);
463
464 // compute rsp as if this were going to be the last frame on
465 // the stack before the red zone
466
467 const Address stack_base(rthread, Thread::stack_base_offset());
468 const Address stack_size(rthread, Thread::stack_size_offset());
469
470 // locals + overhead, in bytes
471 __ mov(r0, overhead_size);
472 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
473
474 __ ldr(rscratch1, stack_base);
475 __ ldr(rscratch2, stack_size);
476
477 #ifdef ASSERT
478 Label stack_base_okay, stack_size_okay;
479 // verify that thread stack base is non-zero
480 __ cbnz(rscratch1, stack_base_okay);
481 __ stop("stack base is zero");
482 __ bind(stack_base_okay);
483 // verify that thread stack size is non-zero
484 __ cbnz(rscratch2, stack_size_okay);
485 __ stop("stack size is zero");
486 __ bind(stack_size_okay);
487 #endif
488
489 // Add stack base to locals and subtract stack size
490 __ sub(rscratch1, rscratch1, rscratch2); // Stack limit
491 __ add(r0, r0, rscratch1);
492
493 // Use the maximum number of pages we might bang.
494 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
495 (StackRedPages+StackYellowPages);
496
497 // add in the red and yellow zone sizes
498 __ add(r0, r0, max_pages * page_size * 2);
499
500 // check against the current stack bottom
501 __ cmp(sp, r0);
502 __ br(Assembler::HI, after_frame_check);
503
504 // Remove the incoming args, peeling the machine SP back to where it
505 // was in the caller. This is not strictly necessary, but unless we
506 // do so the stack frame may have a garbage FP; this ensures a
507 // correct call stack that we can always unwind. The ANDR should be
508 // unnecessary because the sender SP in r13 is always aligned, but
509 // it doesn't hurt.
510 __ andr(sp, r13, -16);
511
512 // Note: the restored frame is not necessarily interpreted.
513 // Use the shared runtime version of the StackOverflowError.
514 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
515 __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry()));
516
517 // all done with frame size check
518 __ bind(after_frame_check);
519 }
520
521 // Allocate monitor and lock method (asm interpreter)
522 //
523 // Args:
524 // rmethod: Method*
525 // rlocals: locals
526 //
527 // Kills:
528 // r0
529 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
530 // rscratch1, rscratch2 (scratch regs)
lock_method(void)531 void InterpreterGenerator::lock_method(void) {
532 // synchronize method
533 const Address access_flags(rmethod, Method::access_flags_offset());
534 const Address monitor_block_top(
535 rfp,
536 frame::interpreter_frame_monitor_block_top_offset * wordSize);
537 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
538
539 #ifdef ASSERT
540 {
541 Label L;
542 __ ldrw(r0, access_flags);
543 __ tst(r0, JVM_ACC_SYNCHRONIZED);
544 __ br(Assembler::NE, L);
545 __ stop("method doesn't need synchronization");
546 __ bind(L);
547 }
548 #endif // ASSERT
549
550 // get synchronization object
551 {
552 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
553 Label done;
554 __ ldrw(r0, access_flags);
555 __ tst(r0, JVM_ACC_STATIC);
556 // get receiver (assume this is frequent case)
557 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
558 __ br(Assembler::EQ, done);
559 __ ldr(r0, Address(rmethod, Method::const_offset()));
560 __ ldr(r0, Address(r0, ConstMethod::constants_offset()));
561 __ ldr(r0, Address(r0,
562 ConstantPool::pool_holder_offset_in_bytes()));
563 __ ldr(r0, Address(r0, mirror_offset));
564
565 #ifdef ASSERT
566 {
567 Label L;
568 __ cbnz(r0, L);
569 __ stop("synchronization object is NULL");
570 __ bind(L);
571 }
572 #endif // ASSERT
573
574 __ bind(done);
575 }
576
577 // add space for monitor & lock
578 __ sub(sp, sp, entry_size); // add space for a monitor entry
579 __ sub(esp, esp, entry_size);
580 __ mov(rscratch1, esp);
581 __ str(rscratch1, monitor_block_top); // set new monitor block top
582 // store object
583 __ str(r0, Address(esp, BasicObjectLock::obj_offset_in_bytes()));
584 __ mov(c_rarg1, esp); // object address
585 __ lock_object(c_rarg1);
586 }
587
588 // Generate a fixed interpreter frame. This is identical setup for
589 // interpreted methods and for native methods hence the shared code.
590 //
591 // Args:
592 // lr: return address
593 // rmethod: Method*
594 // rlocals: pointer to locals
595 // rcpool: cp cache
596 // stack_pointer: previous sp
generate_fixed_frame(bool native_call)597 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
598 // initialize fixed part of activation frame
599 if (native_call) {
600 __ sub(esp, sp, 12 * wordSize);
601 __ mov(rbcp, zr);
602 __ stp(esp, zr, Address(__ pre(sp, -12 * wordSize)));
603 // add 2 zero-initialized slots for native calls
604 __ stp(zr, zr, Address(sp, 10 * wordSize));
605 } else {
606 __ sub(esp, sp, 10 * wordSize);
607 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
608 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
609 __ stp(esp, rbcp, Address(__ pre(sp, -10 * wordSize)));
610 }
611
612 if (ProfileInterpreter) {
613 Label method_data_continue;
614 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
615 __ cbz(rscratch1, method_data_continue);
616 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
617 __ bind(method_data_continue);
618 __ stp(rscratch1, rmethod, Address(sp, 4 * wordSize)); // save Method* and mdp (method data pointer)
619 } else {
620 __ stp(zr, rmethod, Address(sp, 4 * wordSize)); // save Method* (no mdp)
621 }
622
623 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
624 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
625 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset_in_bytes()));
626 __ stp(rlocals, rcpool, Address(sp, 2 * wordSize));
627
628 __ stp(rfp, lr, Address(sp, 8 * wordSize));
629 __ lea(rfp, Address(sp, 8 * wordSize));
630
631 // set sender sp
632 // leave last_sp as null
633 __ stp(zr, r13, Address(sp, 6 * wordSize));
634
635 // Move SP out of the way
636 if (! native_call) {
637 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
638 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
639 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size()
640 + (EnableInvokeDynamic ? 2 : 0));
641 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
642 __ andr(sp, rscratch1, -16);
643 }
644 }
645
646 // End of helpers
647
648 // Various method entries
649 //------------------------------------------------------------------------------------------------------------------------
650 //
651 //
652
653 // Call an accessor method (assuming it is resolved, otherwise drop
654 // into vanilla (slow path) entry
generate_accessor_entry(void)655 address InterpreterGenerator::generate_accessor_entry(void) {
656 return NULL;
657 }
658
659 // Method entry for java.lang.ref.Reference.get.
generate_Reference_get_entry(void)660 address InterpreterGenerator::generate_Reference_get_entry(void) {
661 #if INCLUDE_ALL_GCS
662 // Code: _aload_0, _getfield, _areturn
663 // parameter size = 1
664 //
665 // The code that gets generated by this routine is split into 2 parts:
666 // 1. The "intrinsified" code for G1 (or any SATB based GC),
667 // 2. The slow path - which is an expansion of the regular method entry.
668 //
669 // Notes:-
670 // * In the G1 code we do not check whether we need to block for
671 // a safepoint. If G1 is enabled then we must execute the specialized
672 // code for Reference.get (except when the Reference object is null)
673 // so that we can log the value in the referent field with an SATB
674 // update buffer.
675 // If the code for the getfield template is modified so that the
676 // G1 pre-barrier code is executed when the current method is
677 // Reference.get() then going through the normal method entry
678 // will be fine.
679 // * The G1 code can, however, check the receiver object (the instance
680 // of java.lang.Reference) and jump to the slow path if null. If the
681 // Reference object is null then we obviously cannot fetch the referent
682 // and so we don't need to call the G1 pre-barrier. Thus we can use the
683 // regular method entry code to generate the NPE.
684 //
685 // This code is based on generate_accessor_entry.
686 //
687 // rmethod: Method*
688 // r13: senderSP must preserve for slow path, set SP to it on fast path
689
690 address entry = __ pc();
691
692 const int referent_offset = java_lang_ref_Reference::referent_offset;
693 guarantee(referent_offset > 0, "referent offset not initialized");
694
695 if (UseG1GC) {
696 Label slow_path;
697 const Register local_0 = c_rarg0;
698 // Check if local 0 != NULL
699 // If the receiver is null then it is OK to jump to the slow path.
700 __ ldr(local_0, Address(esp, 0));
701 __ cbz(local_0, slow_path);
702
703 // Load the value of the referent field.
704 const Address field_address(local_0, referent_offset);
705 __ load_heap_oop(local_0, field_address);
706
707 __ mov(r19, r13); // Move senderSP to a callee-saved register
708 // Generate the G1 pre-barrier code to log the value of
709 // the referent field in an SATB buffer.
710 __ enter(); // g1_write may call runtime
711 __ g1_write_barrier_pre(noreg /* obj */,
712 local_0 /* pre_val */,
713 rthread /* thread */,
714 rscratch2 /* tmp */,
715 true /* tosca_live */,
716 true /* expand_call */);
717 __ leave();
718 // areturn
719 __ andr(sp, r19, -16); // done with stack
720 __ ret(lr);
721
722 // generate a vanilla interpreter entry as the slow path
723 __ bind(slow_path);
724 (void) generate_normal_entry(false);
725
726 return entry;
727 }
728 #endif // INCLUDE_ALL_GCS
729
730 // If G1 is not enabled then attempt to go through the accessor entry point
731 // Reference.get is an accessor
732 return generate_accessor_entry();
733 }
734
735 /**
736 * Method entry for static native methods:
737 * int java.util.zip.CRC32.update(int crc, int b)
738 */
generate_CRC32_update_entry()739 address InterpreterGenerator::generate_CRC32_update_entry() {
740 if (UseCRC32Intrinsics) {
741 address entry = __ pc();
742
743 // rmethod: Method*
744 // r13: senderSP must preserved for slow path
745 // esp: args
746
747 Label slow_path;
748 // If we need a safepoint check, generate full interpreter entry.
749 ExternalAddress state(SafepointSynchronize::address_of_state());
750 unsigned long offset;
751 __ adrp(rscratch1, ExternalAddress(SafepointSynchronize::address_of_state()), offset);
752 __ ldrw(rscratch1, Address(rscratch1, offset));
753 assert(SafepointSynchronize::_not_synchronized == 0, "rewrite this code");
754 __ cbnz(rscratch1, slow_path);
755
756 // We don't generate local frame and don't align stack because
757 // we call stub code and there is no safepoint on this path.
758
759 // Load parameters
760 const Register crc = c_rarg0; // crc
761 const Register val = c_rarg1; // source java byte value
762 const Register tbl = c_rarg2; // scratch
763
764 // Arguments are reversed on java expression stack
765 __ ldrw(val, Address(esp, 0)); // byte value
766 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
767
768 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
769 __ add(tbl, tbl, offset);
770
771 __ ornw(crc, zr, crc); // ~crc
772 __ update_byte_crc32(crc, val, tbl);
773 __ ornw(crc, zr, crc); // ~crc
774
775 // result in c_rarg0
776
777 __ andr(sp, r13, -16);
778 __ ret(lr);
779
780 // generate a vanilla native entry as the slow path
781 __ bind(slow_path);
782
783 (void) generate_native_entry(false);
784
785 return entry;
786 }
787 return generate_native_entry(false);
788 }
789
790 /**
791 * Method entry for static native methods:
792 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
793 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
794 */
generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind)795 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
796 if (UseCRC32Intrinsics) {
797 address entry = __ pc();
798
799 // rmethod,: Method*
800 // r13: senderSP must preserved for slow path
801
802 Label slow_path;
803 // If we need a safepoint check, generate full interpreter entry.
804 ExternalAddress state(SafepointSynchronize::address_of_state());
805 unsigned long offset;
806 __ adrp(rscratch1, ExternalAddress(SafepointSynchronize::address_of_state()), offset);
807 __ ldrw(rscratch1, Address(rscratch1, offset));
808 assert(SafepointSynchronize::_not_synchronized == 0, "rewrite this code");
809 __ cbnz(rscratch1, slow_path);
810
811 // We don't generate local frame and don't align stack because
812 // we call stub code and there is no safepoint on this path.
813
814 // Load parameters
815 const Register crc = c_rarg0; // crc
816 const Register buf = c_rarg1; // source java byte array address
817 const Register len = c_rarg2; // length
818 const Register off = len; // offset (never overlaps with 'len')
819
820 // Arguments are reversed on java expression stack
821 // Calculate address of start element
822 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
823 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
824 __ ldrw(off, Address(esp, wordSize)); // offset
825 __ add(buf, buf, off); // + offset
826 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
827 } else {
828 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
829 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
830 __ ldrw(off, Address(esp, wordSize)); // offset
831 __ add(buf, buf, off); // + offset
832 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
833 }
834 // Can now load 'len' since we're finished with 'off'
835 __ ldrw(len, Address(esp, 0x0)); // Length
836
837 __ andr(sp, r13, -16); // Restore the caller's SP
838
839 // We are frameless so we can just jump to the stub.
840 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
841
842 // generate a vanilla native entry as the slow path
843 __ bind(slow_path);
844
845 (void) generate_native_entry(false);
846
847 return entry;
848 }
849 return generate_native_entry(false);
850 }
851
bang_stack_shadow_pages(bool native_call)852 void InterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
853 // Bang each page in the shadow zone. We can't assume it's been done for
854 // an interpreter frame with greater than a page of locals, so each page
855 // needs to be checked. Only true for non-native.
856 if (UseStackBanging) {
857 const int start_page = native_call ? StackShadowPages : 1;
858 const int page_size = os::vm_page_size();
859 for (int pages = start_page; pages <= StackShadowPages ; pages++) {
860 __ sub(rscratch2, sp, pages*page_size);
861 __ str(zr, Address(rscratch2));
862 }
863 }
864 }
865
866
867 // Interpreter stub for calling a native method. (asm interpreter)
868 // This sets up a somewhat different looking stack for calling the
869 // native method than the typical interpreter frame setup.
generate_native_entry(bool synchronized)870 address InterpreterGenerator::generate_native_entry(bool synchronized) {
871 // determine code generation flags
872 bool inc_counter = UseCompiler || CountCompiledCalls;
873
874 // r1: Method*
875 // rscratch1: sender sp
876
877 address entry_point = __ pc();
878
879 const Address constMethod (rmethod, Method::const_offset());
880 const Address access_flags (rmethod, Method::access_flags_offset());
881 const Address size_of_parameters(r2, ConstMethod::
882 size_of_parameters_offset());
883
884 // get parameter size (always needed)
885 __ ldr(r2, constMethod);
886 __ load_unsigned_short(r2, size_of_parameters);
887
888 // native calls don't need the stack size check since they have no
889 // expression stack and the arguments are already on the stack and
890 // we only add a handful of words to the stack
891
892 // rmethod: Method*
893 // r2: size of parameters
894 // rscratch1: sender sp
895
896 // for natives the size of locals is zero
897
898 // compute beginning of parameters (rlocals)
899 __ add(rlocals, esp, r2, ext::uxtx, 3);
900 __ add(rlocals, rlocals, -wordSize);
901
902 // Pull SP back to minimum size: this avoids holes in the stack
903 __ andr(sp, esp, -16);
904
905 // initialize fixed part of activation frame
906 generate_fixed_frame(true);
907
908 // make sure method is native & not abstract
909 #ifdef ASSERT
910 __ ldrw(r0, access_flags);
911 {
912 Label L;
913 __ tst(r0, JVM_ACC_NATIVE);
914 __ br(Assembler::NE, L);
915 __ stop("tried to execute non-native method as native");
916 __ bind(L);
917 }
918 {
919 Label L;
920 __ tst(r0, JVM_ACC_ABSTRACT);
921 __ br(Assembler::EQ, L);
922 __ stop("tried to execute abstract method in interpreter");
923 __ bind(L);
924 }
925 #endif
926
927 // Since at this point in the method invocation the exception
928 // handler would try to exit the monitor of synchronized methods
929 // which hasn't been entered yet, we set the thread local variable
930 // _do_not_unlock_if_synchronized to true. The remove_activation
931 // will check this flag.
932
933 const Address do_not_unlock_if_synchronized(rthread,
934 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
935 __ mov(rscratch2, true);
936 __ strb(rscratch2, do_not_unlock_if_synchronized);
937
938 // increment invocation count & check for overflow
939 Label invocation_counter_overflow;
940 if (inc_counter) {
941 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
942 }
943
944 Label continue_after_compile;
945 __ bind(continue_after_compile);
946
947 bang_stack_shadow_pages(true);
948
949 // reset the _do_not_unlock_if_synchronized flag
950 __ strb(zr, do_not_unlock_if_synchronized);
951
952 // check for synchronized methods
953 // Must happen AFTER invocation_counter check and stack overflow check,
954 // so method is not locked if overflows.
955 if (synchronized) {
956 lock_method();
957 } else {
958 // no synchronization necessary
959 #ifdef ASSERT
960 {
961 Label L;
962 __ ldrw(r0, access_flags);
963 __ tst(r0, JVM_ACC_SYNCHRONIZED);
964 __ br(Assembler::EQ, L);
965 __ stop("method needs synchronization");
966 __ bind(L);
967 }
968 #endif
969 }
970
971 // start execution
972 #ifdef ASSERT
973 {
974 Label L;
975 const Address monitor_block_top(rfp,
976 frame::interpreter_frame_monitor_block_top_offset * wordSize);
977 __ ldr(rscratch1, monitor_block_top);
978 __ cmp(esp, rscratch1);
979 __ br(Assembler::EQ, L);
980 __ stop("broken stack frame setup in interpreter");
981 __ bind(L);
982 }
983 #endif
984
985 // jvmti support
986 __ notify_method_entry();
987
988 // work registers
989 const Register t = r17;
990 const Register result_handler = r19;
991
992 // allocate space for parameters
993 __ ldr(t, Address(rmethod, Method::const_offset()));
994 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
995
996 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
997 __ andr(sp, rscratch1, -16);
998 __ mov(esp, rscratch1);
999
1000 // get signature handler
1001 {
1002 Label L;
1003 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1004 __ cbnz(t, L);
1005 __ call_VM(noreg,
1006 CAST_FROM_FN_PTR(address,
1007 InterpreterRuntime::prepare_native_call),
1008 rmethod);
1009 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1010 __ bind(L);
1011 }
1012
1013 // call signature handler
1014 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1015 "adjust this code");
1016 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1017 "adjust this code");
1018 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1019 "adjust this code");
1020
1021 // The generated handlers do not touch rmethod (the method).
1022 // However, large signatures cannot be cached and are generated
1023 // each time here. The slow-path generator can do a GC on return,
1024 // so we must reload it after the call.
1025 __ blr(t);
1026 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1027
1028
1029 // result handler is in r0
1030 // set result handler
1031 __ mov(result_handler, r0);
1032 // pass mirror handle if static call
1033 {
1034 Label L;
1035 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1036 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1037 __ tst(t, JVM_ACC_STATIC);
1038 __ br(Assembler::EQ, L);
1039 // get mirror
1040 __ ldr(t, Address(rmethod, Method::const_offset()));
1041 __ ldr(t, Address(t, ConstMethod::constants_offset()));
1042 __ ldr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1043 __ ldr(t, Address(t, mirror_offset));
1044 // copy mirror into activation frame
1045 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1046 // pass handle to mirror
1047 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1048 __ bind(L);
1049 }
1050
1051 // get native function entry point in r10
1052 {
1053 Label L;
1054 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1055 address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1056 __ mov(rscratch2, unsatisfied);
1057 __ ldr(rscratch2, rscratch2);
1058 __ cmp(r10, rscratch2);
1059 __ br(Assembler::NE, L);
1060 __ call_VM(noreg,
1061 CAST_FROM_FN_PTR(address,
1062 InterpreterRuntime::prepare_native_call),
1063 rmethod);
1064 __ get_method(rmethod);
1065 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1066 __ bind(L);
1067 }
1068
1069 // pass JNIEnv
1070 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1071
1072 // Set the last Java PC in the frame anchor to be the return address from
1073 // the call to the native method: this will allow the debugger to
1074 // generate an accurate stack trace.
1075 Label native_return;
1076 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1077
1078 // change thread state
1079 #ifdef ASSERT
1080 {
1081 Label L;
1082 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1083 __ cmp(t, _thread_in_Java);
1084 __ br(Assembler::EQ, L);
1085 __ stop("Wrong thread state in native stub");
1086 __ bind(L);
1087 }
1088 #endif
1089
1090 // Change state to native
1091 __ mov(rscratch1, _thread_in_native);
1092 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1093 __ stlrw(rscratch1, rscratch2);
1094
1095 // Call the native method.
1096 __ blr(r10);
1097 __ bind(native_return);
1098 __ maybe_isb();
1099 __ get_method(rmethod);
1100 // result potentially in r0 or v0
1101
1102 // make room for the pushes we're about to do
1103 __ sub(rscratch1, esp, 4 * wordSize);
1104 __ andr(sp, rscratch1, -16);
1105
1106 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1107 // in order to extract the result of a method call. If the order of these
1108 // pushes change or anything else is added to the stack then the code in
1109 // interpreter_frame_result must also change.
1110 __ push(dtos);
1111 __ push(ltos);
1112
1113 // change thread state
1114 __ mov(rscratch1, _thread_in_native_trans);
1115 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1116 __ stlrw(rscratch1, rscratch2);
1117
1118 if (os::is_MP()) {
1119 if (UseMembar) {
1120 // Force this write out before the read below
1121 __ dsb(Assembler::SY);
1122 } else {
1123 // Write serialization page so VM thread can do a pseudo remote membar.
1124 // We use the current thread pointer to calculate a thread specific
1125 // offset to write to within the page. This minimizes bus traffic
1126 // due to cache line collision.
1127 __ serialize_memory(rthread, rscratch2);
1128 }
1129 }
1130
1131 // check for safepoint operation in progress and/or pending suspend requests
1132 {
1133 Label Continue;
1134 {
1135 unsigned long offset;
1136 __ adrp(rscratch2, SafepointSynchronize::address_of_state(), offset);
1137 __ ldrw(rscratch2, Address(rscratch2, offset));
1138 }
1139 assert(SafepointSynchronize::_not_synchronized == 0,
1140 "SafepointSynchronize::_not_synchronized");
1141 Label L;
1142 __ cbnz(rscratch2, L);
1143 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1144 __ cbz(rscratch2, Continue);
1145 __ bind(L);
1146
1147 // Don't use call_VM as it will see a possible pending exception
1148 // and forward it and never return here preventing us from
1149 // clearing _last_native_pc down below. Also can't use
1150 // call_VM_leaf either as it will check to see if r13 & r14 are
1151 // preserved and correspond to the bcp/locals pointers. So we do a
1152 // runtime call by hand.
1153 //
1154 __ mov(c_rarg0, rthread);
1155 __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1156 __ blr(rscratch2);
1157 __ maybe_isb();
1158 __ get_method(rmethod);
1159 __ reinit_heapbase();
1160 __ bind(Continue);
1161 }
1162
1163 // change thread state
1164 __ mov(rscratch1, _thread_in_Java);
1165 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1166 __ stlrw(rscratch1, rscratch2);
1167
1168 // reset_last_Java_frame
1169 __ reset_last_Java_frame(true);
1170
1171 // reset handle block
1172 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1173 __ str(zr, Address(t, JNIHandleBlock::top_offset_in_bytes()));
1174
1175 // If result is an oop unbox and store it in frame where gc will see it
1176 // and result handler will pick it up
1177
1178 {
1179 Label no_oop, not_weak, store_result;
1180 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1181 __ cmp(t, result_handler);
1182 __ br(Assembler::NE, no_oop);
1183 // Unbox oop result, e.g. JNIHandles::resolve result.
1184 __ pop(ltos);
1185 __ cbz(r0, store_result); // Use NULL as-is.
1186 STATIC_ASSERT(JNIHandles::weak_tag_mask == 1u);
1187 __ tbz(r0, 0, not_weak); // Test for jweak tag.
1188 // Resolve jweak.
1189 __ ldr(r0, Address(r0, -JNIHandles::weak_tag_value));
1190 #if INCLUDE_ALL_GCS
1191 if (UseG1GC) {
1192 __ enter(); // Barrier may call runtime.
1193 __ g1_write_barrier_pre(noreg /* obj */,
1194 r0 /* pre_val */,
1195 rthread /* thread */,
1196 t /* tmp */,
1197 true /* tosca_live */,
1198 true /* expand_call */);
1199 __ leave();
1200 }
1201 #endif // INCLUDE_ALL_GCS
1202 __ b(store_result);
1203 __ bind(not_weak);
1204 // Resolve (untagged) jobject.
1205 __ ldr(r0, Address(r0, 0));
1206 __ bind(store_result);
1207 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1208 // keep stack depth as expected by pushing oop which will eventually be discarded
1209 __ push(ltos);
1210 __ bind(no_oop);
1211 }
1212
1213 {
1214 Label no_reguard;
1215 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1216 __ ldrb(rscratch1, Address(rscratch1));
1217 __ cmp(rscratch1, JavaThread::stack_guard_yellow_disabled);
1218 __ br(Assembler::NE, no_reguard);
1219
1220 __ pusha(); // XXX only save smashed registers
1221 __ mov(c_rarg0, rthread);
1222 __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1223 __ blr(rscratch2);
1224 __ popa(); // XXX only restore smashed registers
1225 __ bind(no_reguard);
1226 }
1227
1228 // The method register is junk from after the thread_in_native transition
1229 // until here. Also can't call_VM until the bcp has been
1230 // restored. Need bcp for throwing exception below so get it now.
1231 __ get_method(rmethod);
1232
1233 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1234 // rbcp == code_base()
1235 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1236 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1237 // handle exceptions (exception handling will handle unlocking!)
1238 {
1239 Label L;
1240 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1241 __ cbz(rscratch1, L);
1242 // Note: At some point we may want to unify this with the code
1243 // used in call_VM_base(); i.e., we should use the
1244 // StubRoutines::forward_exception code. For now this doesn't work
1245 // here because the rsp is not correctly set at this point.
1246 __ MacroAssembler::call_VM(noreg,
1247 CAST_FROM_FN_PTR(address,
1248 InterpreterRuntime::throw_pending_exception));
1249 __ should_not_reach_here();
1250 __ bind(L);
1251 }
1252
1253 // do unlocking if necessary
1254 {
1255 Label L;
1256 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1257 __ tst(t, JVM_ACC_SYNCHRONIZED);
1258 __ br(Assembler::EQ, L);
1259 // the code below should be shared with interpreter macro
1260 // assembler implementation
1261 {
1262 Label unlock;
1263 // BasicObjectLock will be first in list, since this is a
1264 // synchronized method. However, need to check that the object
1265 // has not been unlocked by an explicit monitorexit bytecode.
1266
1267 // monitor expect in c_rarg1 for slow unlock path
1268 __ lea (c_rarg1, Address(rfp, // address of first monitor
1269 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1270 wordSize - sizeof(BasicObjectLock))));
1271
1272 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1273 __ cbnz(t, unlock);
1274
1275 // Entry already unlocked, need to throw exception
1276 __ MacroAssembler::call_VM(noreg,
1277 CAST_FROM_FN_PTR(address,
1278 InterpreterRuntime::throw_illegal_monitor_state_exception));
1279 __ should_not_reach_here();
1280
1281 __ bind(unlock);
1282 __ unlock_object(c_rarg1);
1283 }
1284 __ bind(L);
1285 }
1286
1287 // jvmti support
1288 // Note: This must happen _after_ handling/throwing any exceptions since
1289 // the exception handler code notifies the runtime of method exits
1290 // too. If this happens before, method entry/exit notifications are
1291 // not properly paired (was bug - gri 11/22/99).
1292 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1293
1294 // restore potential result in r0:d0, call result handler to
1295 // restore potential result in ST0 & handle result
1296
1297 __ pop(ltos);
1298 __ pop(dtos);
1299
1300 __ blr(result_handler);
1301
1302 // remove activation
1303 __ ldr(esp, Address(rfp,
1304 frame::interpreter_frame_sender_sp_offset *
1305 wordSize)); // get sender sp
1306 // remove frame anchor
1307 __ leave();
1308
1309 // resture sender sp
1310 __ mov(sp, esp);
1311
1312 __ ret(lr);
1313
1314 if (inc_counter) {
1315 // Handle overflow of counter and compile method
1316 __ bind(invocation_counter_overflow);
1317 generate_counter_overflow(&continue_after_compile);
1318 }
1319
1320 return entry_point;
1321 }
1322
1323 //
1324 // Generic interpreted method entry to (asm) interpreter
1325 //
generate_normal_entry(bool synchronized)1326 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1327 // determine code generation flags
1328 bool inc_counter = UseCompiler || CountCompiledCalls;
1329
1330 // rscratch1: sender sp
1331 address entry_point = __ pc();
1332
1333 const Address constMethod(rmethod, Method::const_offset());
1334 const Address access_flags(rmethod, Method::access_flags_offset());
1335 const Address size_of_parameters(r3,
1336 ConstMethod::size_of_parameters_offset());
1337 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1338
1339 // get parameter size (always needed)
1340 // need to load the const method first
1341 __ ldr(r3, constMethod);
1342 __ load_unsigned_short(r2, size_of_parameters);
1343
1344 // r2: size of parameters
1345
1346 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1347 __ sub(r3, r3, r2); // r3 = no. of additional locals
1348
1349 // see if we've got enough room on the stack for locals plus overhead.
1350 generate_stack_overflow_check();
1351
1352 // compute beginning of parameters (rlocals)
1353 __ add(rlocals, esp, r2, ext::uxtx, 3);
1354 __ sub(rlocals, rlocals, wordSize);
1355
1356 // Make room for locals
1357 __ sub(rscratch1, esp, r3, ext::uxtx, 3);
1358 __ andr(sp, rscratch1, -16);
1359
1360 // r3 - # of additional locals
1361 // allocate space for locals
1362 // explicitly initialize locals
1363 {
1364 Label exit, loop;
1365 __ ands(zr, r3, r3);
1366 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1367 __ bind(loop);
1368 __ str(zr, Address(__ post(rscratch1, wordSize)));
1369 __ sub(r3, r3, 1); // until everything initialized
1370 __ cbnz(r3, loop);
1371 __ bind(exit);
1372 }
1373
1374 // And the base dispatch table
1375 __ get_dispatch();
1376
1377 // initialize fixed part of activation frame
1378 generate_fixed_frame(false);
1379
1380 // make sure method is not native & not abstract
1381 #ifdef ASSERT
1382 __ ldrw(r0, access_flags);
1383 {
1384 Label L;
1385 __ tst(r0, JVM_ACC_NATIVE);
1386 __ br(Assembler::EQ, L);
1387 __ stop("tried to execute native method as non-native");
1388 __ bind(L);
1389 }
1390 {
1391 Label L;
1392 __ tst(r0, JVM_ACC_ABSTRACT);
1393 __ br(Assembler::EQ, L);
1394 __ stop("tried to execute abstract method in interpreter");
1395 __ bind(L);
1396 }
1397 #endif
1398
1399 // Since at this point in the method invocation the exception
1400 // handler would try to exit the monitor of synchronized methods
1401 // which hasn't been entered yet, we set the thread local variable
1402 // _do_not_unlock_if_synchronized to true. The remove_activation
1403 // will check this flag.
1404
1405 const Address do_not_unlock_if_synchronized(rthread,
1406 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1407 __ mov(rscratch2, true);
1408 __ strb(rscratch2, do_not_unlock_if_synchronized);
1409
1410 // increment invocation count & check for overflow
1411 Label invocation_counter_overflow;
1412 Label profile_method;
1413 Label profile_method_continue;
1414 if (inc_counter) {
1415 generate_counter_incr(&invocation_counter_overflow,
1416 &profile_method,
1417 &profile_method_continue);
1418 if (ProfileInterpreter) {
1419 __ bind(profile_method_continue);
1420 }
1421 }
1422
1423 Label continue_after_compile;
1424 __ bind(continue_after_compile);
1425
1426 bang_stack_shadow_pages(false);
1427
1428 // reset the _do_not_unlock_if_synchronized flag
1429 __ strb(zr, do_not_unlock_if_synchronized);
1430
1431 // check for synchronized methods
1432 // Must happen AFTER invocation_counter check and stack overflow check,
1433 // so method is not locked if overflows.
1434 if (synchronized) {
1435 // Allocate monitor and lock method
1436 lock_method();
1437 } else {
1438 // no synchronization necessary
1439 #ifdef ASSERT
1440 {
1441 Label L;
1442 __ ldrw(r0, access_flags);
1443 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1444 __ br(Assembler::EQ, L);
1445 __ stop("method needs synchronization");
1446 __ bind(L);
1447 }
1448 #endif
1449 }
1450
1451 // start execution
1452 #ifdef ASSERT
1453 {
1454 Label L;
1455 const Address monitor_block_top (rfp,
1456 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1457 __ ldr(rscratch1, monitor_block_top);
1458 __ cmp(esp, rscratch1);
1459 __ br(Assembler::EQ, L);
1460 __ stop("broken stack frame setup in interpreter");
1461 __ bind(L);
1462 }
1463 #endif
1464
1465 // jvmti support
1466 __ notify_method_entry();
1467
1468 __ dispatch_next(vtos);
1469
1470 // invocation counter overflow
1471 if (inc_counter) {
1472 if (ProfileInterpreter) {
1473 // We have decided to profile this method in the interpreter
1474 __ bind(profile_method);
1475 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1476 __ set_method_data_pointer_for_bcp();
1477 // don't think we need this
1478 __ get_method(r1);
1479 __ b(profile_method_continue);
1480 }
1481 // Handle overflow of counter and compile method
1482 __ bind(invocation_counter_overflow);
1483 generate_counter_overflow(&continue_after_compile);
1484 }
1485
1486 return entry_point;
1487 }
1488
1489 // Entry points
1490 //
1491 // Here we generate the various kind of entries into the interpreter.
1492 // The two main entry type are generic bytecode methods and native
1493 // call method. These both come in synchronized and non-synchronized
1494 // versions but the frame layout they create is very similar. The
1495 // other method entry types are really just special purpose entries
1496 // that are really entry and interpretation all in one. These are for
1497 // trivial methods like accessor, empty, or special math methods.
1498 //
1499 // When control flow reaches any of the entry types for the interpreter
1500 // the following holds ->
1501 //
1502 // Arguments:
1503 //
1504 // rmethod: Method*
1505 //
1506 // Stack layout immediately at entry
1507 //
1508 // [ return address ] <--- rsp
1509 // [ parameter n ]
1510 // ...
1511 // [ parameter 1 ]
1512 // [ expression stack ] (caller's java expression stack)
1513
1514 // Assuming that we don't go to one of the trivial specialized entries
1515 // the stack will look like below when we are ready to execute the
1516 // first bytecode (or call the native routine). The register usage
1517 // will be as the template based interpreter expects (see
1518 // interpreter_aarch64.hpp).
1519 //
1520 // local variables follow incoming parameters immediately; i.e.
1521 // the return address is moved to the end of the locals).
1522 //
1523 // [ monitor entry ] <--- esp
1524 // ...
1525 // [ monitor entry ]
1526 // [ expr. stack bottom ]
1527 // [ saved rbcp ]
1528 // [ current rlocals ]
1529 // [ Method* ]
1530 // [ saved rfp ] <--- rfp
1531 // [ return address ]
1532 // [ local variable m ]
1533 // ...
1534 // [ local variable 1 ]
1535 // [ parameter n ]
1536 // ...
1537 // [ parameter 1 ] <--- rlocals
1538
generate_method_entry(AbstractInterpreter::MethodKind kind)1539 address AbstractInterpreterGenerator::generate_method_entry(
1540 AbstractInterpreter::MethodKind kind) {
1541 // determine code generation flags
1542 bool synchronized = false;
1543 address entry_point = NULL;
1544
1545 switch (kind) {
1546 case Interpreter::zerolocals : break;
1547 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1548 case Interpreter::native : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break;
1549 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true); break;
1550 case Interpreter::empty : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry(); break;
1551 case Interpreter::accessor : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry(); break;
1552 case Interpreter::abstract : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry(); break;
1553
1554 case Interpreter::java_lang_math_sin : // fall thru
1555 case Interpreter::java_lang_math_cos : // fall thru
1556 case Interpreter::java_lang_math_tan : // fall thru
1557 case Interpreter::java_lang_math_abs : // fall thru
1558 case Interpreter::java_lang_math_log : // fall thru
1559 case Interpreter::java_lang_math_log10 : // fall thru
1560 case Interpreter::java_lang_math_sqrt : // fall thru
1561 case Interpreter::java_lang_math_pow : // fall thru
1562 case Interpreter::java_lang_math_exp : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind); break;
1563 case Interpreter::java_lang_ref_reference_get
1564 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1565 case Interpreter::java_util_zip_CRC32_update
1566 : entry_point = ((InterpreterGenerator*)this)->generate_CRC32_update_entry(); break;
1567 case Interpreter::java_util_zip_CRC32_updateBytes
1568 : // fall thru
1569 case Interpreter::java_util_zip_CRC32_updateByteBuffer
1570 : entry_point = ((InterpreterGenerator*)this)->generate_CRC32_updateBytes_entry(kind); break;
1571 default : ShouldNotReachHere(); break;
1572 }
1573
1574 if (entry_point) {
1575 return entry_point;
1576 }
1577
1578 return ((InterpreterGenerator*) this)->
1579 generate_normal_entry(synchronized);
1580 }
1581
1582
1583 // These should never be compiled since the interpreter will prefer
1584 // the compiled version to the intrinsic version.
can_be_compiled(methodHandle m)1585 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1586 switch (method_kind(m)) {
1587 case Interpreter::java_lang_math_sin : // fall thru
1588 case Interpreter::java_lang_math_cos : // fall thru
1589 case Interpreter::java_lang_math_tan : // fall thru
1590 case Interpreter::java_lang_math_abs : // fall thru
1591 case Interpreter::java_lang_math_log : // fall thru
1592 case Interpreter::java_lang_math_log10 : // fall thru
1593 case Interpreter::java_lang_math_sqrt : // fall thru
1594 case Interpreter::java_lang_math_pow : // fall thru
1595 case Interpreter::java_lang_math_exp :
1596 return false;
1597 default:
1598 return true;
1599 }
1600 }
1601
1602 // How much stack a method activation needs in words.
size_top_interpreter_activation(Method * method)1603 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1604 const int entry_size = frame::interpreter_frame_monitor_size();
1605
1606 // total overhead size: entry_size + (saved rfp thru expr stack
1607 // bottom). be sure to change this if you add/subtract anything
1608 // to/from the overhead area
1609 const int overhead_size =
1610 -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1611
1612 const int stub_code = frame::entry_frame_after_call_words;
1613 const int method_stack = (method->max_locals() + method->max_stack()) *
1614 Interpreter::stackElementWords;
1615 return (overhead_size + method_stack + stub_code);
1616 }
1617
1618 // asm based interpreter deoptimization helpers
size_activation(int max_stack,int temps,int extra_args,int monitors,int callee_params,int callee_locals,bool is_top_frame)1619 int AbstractInterpreter::size_activation(int max_stack,
1620 int temps,
1621 int extra_args,
1622 int monitors,
1623 int callee_params,
1624 int callee_locals,
1625 bool is_top_frame) {
1626 // Note: This calculation must exactly parallel the frame setup
1627 // in AbstractInterpreterGenerator::generate_method_entry.
1628
1629 // fixed size of an interpreter frame:
1630 int overhead = frame::sender_sp_offset -
1631 frame::interpreter_frame_initial_sp_offset;
1632 // Our locals were accounted for by the caller (or last_frame_adjust
1633 // on the transistion) Since the callee parameters already account
1634 // for the callee's params we only need to account for the extra
1635 // locals.
1636 int size = overhead +
1637 (callee_locals - callee_params) +
1638 monitors * frame::interpreter_frame_monitor_size() +
1639 // On the top frame, at all times SP <= ESP, and SP is
1640 // 16-aligned. We ensure this by adjusting SP on method
1641 // entry and re-entry to allow room for the maximum size of
1642 // the expression stack. When we call another method we bump
1643 // SP so that no stack space is wasted. So, only on the top
1644 // frame do we need to allow max_stack words.
1645 (is_top_frame ? max_stack : temps + extra_args);
1646
1647 // On AArch64 we always keep the stack pointer 16-aligned, so we
1648 // must round up here.
1649 size = round_to(size, 2);
1650
1651 return size;
1652 }
1653
layout_activation(Method * method,int tempcount,int popframe_extra_args,int moncount,int caller_actual_parameters,int callee_param_count,int callee_locals,frame * caller,frame * interpreter_frame,bool is_top_frame,bool is_bottom_frame)1654 void AbstractInterpreter::layout_activation(Method* method,
1655 int tempcount,
1656 int popframe_extra_args,
1657 int moncount,
1658 int caller_actual_parameters,
1659 int callee_param_count,
1660 int callee_locals,
1661 frame* caller,
1662 frame* interpreter_frame,
1663 bool is_top_frame,
1664 bool is_bottom_frame) {
1665 // The frame interpreter_frame is guaranteed to be the right size,
1666 // as determined by a previous call to the size_activation() method.
1667 // It is also guaranteed to be walkable even though it is in a
1668 // skeletal state
1669
1670 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1671 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1672 Interpreter::stackElementWords;
1673
1674 #ifdef ASSERT
1675 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable");
1676 #endif
1677
1678 interpreter_frame->interpreter_frame_set_method(method);
1679 // NOTE the difference in using sender_sp and
1680 // interpreter_frame_sender_sp interpreter_frame_sender_sp is
1681 // the original sp of the caller (the unextended_sp) and
1682 // sender_sp is fp+8/16 (32bit/64bit) XXX
1683 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1684
1685 #ifdef ASSERT
1686 if (caller->is_interpreted_frame()) {
1687 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1688 }
1689 #endif
1690
1691 interpreter_frame->interpreter_frame_set_locals(locals);
1692 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1693 BasicObjectLock* monbot = montop - moncount;
1694 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1695
1696 // Set last_sp
1697 intptr_t* esp = (intptr_t*) monbot -
1698 tempcount*Interpreter::stackElementWords -
1699 popframe_extra_args;
1700 interpreter_frame->interpreter_frame_set_last_sp(esp);
1701
1702 // All frames but the initial (oldest) interpreter frame we fill in have
1703 // a value for sender_sp that allows walking the stack but isn't
1704 // truly correct. Correct the value here.
1705 if (extra_locals != 0 &&
1706 interpreter_frame->sender_sp() ==
1707 interpreter_frame->interpreter_frame_sender_sp()) {
1708 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() +
1709 extra_locals);
1710 }
1711 *interpreter_frame->interpreter_frame_cache_addr() =
1712 method->constants()->cache();
1713 }
1714
1715
1716 //-----------------------------------------------------------------------------
1717 // Exceptions
1718
generate_throw_exception()1719 void TemplateInterpreterGenerator::generate_throw_exception() {
1720 // Entry point in previous activation (i.e., if the caller was
1721 // interpreted)
1722 Interpreter::_rethrow_exception_entry = __ pc();
1723 // Restore sp to interpreter_frame_last_sp even though we are going
1724 // to empty the expression stack for the exception processing.
1725 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1726 // r0: exception
1727 // r3: return address/pc that threw exception
1728 __ restore_bcp(); // rbcp points to call/send
1729 __ restore_locals();
1730 __ restore_constant_pool_cache();
1731 __ reinit_heapbase(); // restore rheapbase as heapbase.
1732 __ get_dispatch();
1733
1734 // Entry point for exceptions thrown within interpreter code
1735 Interpreter::_throw_exception_entry = __ pc();
1736 // If we came here via a NullPointerException on the receiver of a
1737 // method, rmethod may be corrupt.
1738 __ get_method(rmethod);
1739 // expression stack is undefined here
1740 // r0: exception
1741 // rbcp: exception bcp
1742 __ verify_oop(r0);
1743 __ mov(c_rarg1, r0);
1744
1745 // expression stack must be empty before entering the VM in case of
1746 // an exception
1747 __ empty_expression_stack();
1748 // find exception handler address and preserve exception oop
1749 __ call_VM(r3,
1750 CAST_FROM_FN_PTR(address,
1751 InterpreterRuntime::exception_handler_for_exception),
1752 c_rarg1);
1753
1754 // Calculate stack limit
1755 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1756 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1757 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size()
1758 + (EnableInvokeDynamic ? 2 : 0) + 2);
1759 __ ldr(rscratch2,
1760 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1761 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3);
1762 __ andr(sp, rscratch1, -16);
1763
1764 // r0: exception handler entry point
1765 // r3: preserved exception oop
1766 // rbcp: bcp for exception handler
1767 __ push_ptr(r3); // push exception which is now the only value on the stack
1768 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1769
1770 // If the exception is not handled in the current frame the frame is
1771 // removed and the exception is rethrown (i.e. exception
1772 // continuation is _rethrow_exception).
1773 //
1774 // Note: At this point the bci is still the bxi for the instruction
1775 // which caused the exception and the expression stack is
1776 // empty. Thus, for any VM calls at this point, GC will find a legal
1777 // oop map (with empty expression stack).
1778
1779 //
1780 // JVMTI PopFrame support
1781 //
1782
1783 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1784 __ empty_expression_stack();
1785 // Set the popframe_processing bit in pending_popframe_condition
1786 // indicating that we are currently handling popframe, so that
1787 // call_VMs that may happen later do not trigger new popframe
1788 // handling cycles.
1789 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1790 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1791 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1792
1793 {
1794 // Check to see whether we are returning to a deoptimized frame.
1795 // (The PopFrame call ensures that the caller of the popped frame is
1796 // either interpreted or compiled and deoptimizes it if compiled.)
1797 // In this case, we can't call dispatch_next() after the frame is
1798 // popped, but instead must save the incoming arguments and restore
1799 // them after deoptimization has occurred.
1800 //
1801 // Note that we don't compare the return PC against the
1802 // deoptimization blob's unpack entry because of the presence of
1803 // adapter frames in C2.
1804 Label caller_not_deoptimized;
1805 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1806 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1807 InterpreterRuntime::interpreter_contains), c_rarg1);
1808 __ cbnz(r0, caller_not_deoptimized);
1809
1810 // Compute size of arguments for saving when returning to
1811 // deoptimized caller
1812 __ get_method(r0);
1813 __ ldr(r0, Address(r0, Method::const_offset()));
1814 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1815 size_of_parameters_offset())));
1816 __ lsl(r0, r0, Interpreter::logStackElementSize);
1817 __ restore_locals(); // XXX do we need this?
1818 __ sub(rlocals, rlocals, r0);
1819 __ add(rlocals, rlocals, wordSize);
1820 // Save these arguments
1821 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1822 Deoptimization::
1823 popframe_preserve_args),
1824 rthread, r0, rlocals);
1825
1826 __ remove_activation(vtos,
1827 /* throw_monitor_exception */ false,
1828 /* install_monitor_exception */ false,
1829 /* notify_jvmdi */ false);
1830
1831 // Inform deoptimization that it is responsible for restoring
1832 // these arguments
1833 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1834 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1835
1836 // Continue in deoptimization handler
1837 __ ret(lr);
1838
1839 __ bind(caller_not_deoptimized);
1840 }
1841
1842 __ remove_activation(vtos,
1843 /* throw_monitor_exception */ false,
1844 /* install_monitor_exception */ false,
1845 /* notify_jvmdi */ false);
1846
1847 // Restore the last_sp and null it out
1848 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1849 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1850
1851 __ restore_bcp();
1852 __ restore_locals();
1853 __ restore_constant_pool_cache();
1854 __ get_method(rmethod);
1855 __ get_dispatch();
1856
1857 // The method data pointer was incremented already during
1858 // call profiling. We have to restore the mdp for the current bcp.
1859 if (ProfileInterpreter) {
1860 __ set_method_data_pointer_for_bcp();
1861 }
1862
1863 // Clear the popframe condition flag
1864 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1865 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1866
1867 #if INCLUDE_JVMTI
1868 if (EnableInvokeDynamic) {
1869 Label L_done;
1870
1871 __ ldrb(rscratch1, Address(rbcp, 0));
1872 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1873 __ br(Assembler::NE, L_done);
1874
1875 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1876 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1877
1878 __ ldr(c_rarg0, Address(rlocals, 0));
1879 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1880
1881 __ cbz(r0, L_done);
1882
1883 __ str(r0, Address(esp, 0));
1884 __ bind(L_done);
1885 }
1886 #endif // INCLUDE_JVMTI
1887
1888 // Restore machine SP
1889 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1890 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1891 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size()
1892 + (EnableInvokeDynamic ? 2 : 0));
1893 __ ldr(rscratch2,
1894 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1895 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
1896 __ andr(sp, rscratch1, -16);
1897
1898 __ dispatch_next(vtos);
1899 // end of PopFrame support
1900
1901 Interpreter::_remove_activation_entry = __ pc();
1902
1903 // preserve exception over this code sequence
1904 __ pop_ptr(r0);
1905 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1906 // remove the activation (without doing throws on illegalMonitorExceptions)
1907 __ remove_activation(vtos, false, true, false);
1908 // restore exception
1909 __ get_vm_result(r0, rthread);
1910
1911 // In between activations - previous activation type unknown yet
1912 // compute continuation point - the continuation point expects the
1913 // following registers set up:
1914 //
1915 // r0: exception
1916 // lr: return address/pc that threw exception
1917 // esp: expression stack of caller
1918 // rfp: fp of caller
1919 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
1920 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1921 SharedRuntime::exception_handler_for_return_address),
1922 rthread, lr);
1923 __ mov(r1, r0); // save exception handler
1924 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
1925 // We might be returning to a deopt handler that expects r3 to
1926 // contain the exception pc
1927 __ mov(r3, lr);
1928 // Note that an "issuing PC" is actually the next PC after the call
1929 __ br(r1); // jump to exception
1930 // handler of caller
1931 }
1932
1933
1934 //
1935 // JVMTI ForceEarlyReturn support
1936 //
generate_earlyret_entry_for(TosState state)1937 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1938 address entry = __ pc();
1939
1940 __ restore_bcp();
1941 __ restore_locals();
1942 __ empty_expression_stack();
1943 __ load_earlyret_value(state);
1944
1945 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
1946 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
1947
1948 // Clear the earlyret state
1949 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1950 __ str(zr, cond_addr);
1951
1952 __ remove_activation(state,
1953 false, /* throw_monitor_exception */
1954 false, /* install_monitor_exception */
1955 true); /* notify_jvmdi */
1956 __ ret(lr);
1957
1958 return entry;
1959 } // end of ForceEarlyReturn support
1960
1961
1962
1963 //-----------------------------------------------------------------------------
1964 // Helper for vtos entry point generation
1965
set_vtos_entry_points(Template * t,address & bep,address & cep,address & sep,address & aep,address & iep,address & lep,address & fep,address & dep,address & vep)1966 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1967 address& bep,
1968 address& cep,
1969 address& sep,
1970 address& aep,
1971 address& iep,
1972 address& lep,
1973 address& fep,
1974 address& dep,
1975 address& vep) {
1976 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1977 Label L;
1978 aep = __ pc(); __ push_ptr(); __ b(L);
1979 fep = __ pc(); __ push_f(); __ b(L);
1980 dep = __ pc(); __ push_d(); __ b(L);
1981 lep = __ pc(); __ push_l(); __ b(L);
1982 bep = cep = sep =
1983 iep = __ pc(); __ push_i();
1984 vep = __ pc();
1985 __ bind(L);
1986 generate_and_dispatch(t);
1987 }
1988
1989 //-----------------------------------------------------------------------------
1990 // Generation of individual instructions
1991
1992 // helpers for generate_and_dispatch
1993
1994
InterpreterGenerator(StubQueue * code)1995 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1996 : TemplateInterpreterGenerator(code) {
1997 generate_all(); // down here so it can be "virtual"
1998 }
1999
2000 //-----------------------------------------------------------------------------
2001
2002 // Non-product code
2003 #ifndef PRODUCT
generate_trace_code(TosState state)2004 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2005 address entry = __ pc();
2006
2007 __ push(lr);
2008 __ push(state);
2009 __ push(RegSet::range(r0, r15), sp);
2010 __ mov(c_rarg2, r0); // Pass itos
2011 __ call_VM(noreg,
2012 CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
2013 c_rarg1, c_rarg2, c_rarg3);
2014 __ pop(RegSet::range(r0, r15), sp);
2015 __ pop(state);
2016 __ pop(lr);
2017 __ ret(lr); // return from result handler
2018
2019 return entry;
2020 }
2021
count_bytecode()2022 void TemplateInterpreterGenerator::count_bytecode() {
2023 Register rscratch3 = r0;
2024 __ push(rscratch1);
2025 __ push(rscratch2);
2026 __ push(rscratch3);
2027 __ mov(rscratch3, (address) &BytecodeCounter::_counter_value);
2028 __ atomic_add(noreg, 1, rscratch3);
2029 __ pop(rscratch3);
2030 __ pop(rscratch2);
2031 __ pop(rscratch1);
2032 }
2033
histogram_bytecode(Template * t)2034 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { ; }
2035
histogram_bytecode_pair(Template * t)2036 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { ; }
2037
2038
trace_bytecode(Template * t)2039 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2040 // Call a little run-time stub to avoid blow-up for each bytecode.
2041 // The run-time runtime saves the right registers, depending on
2042 // the tosca in-state for the given template.
2043
2044 assert(Interpreter::trace_code(t->tos_in()) != NULL,
2045 "entry must have been generated");
2046 __ bl(Interpreter::trace_code(t->tos_in()));
2047 __ reinit_heapbase();
2048 }
2049
2050
stop_interpreter_at()2051 void TemplateInterpreterGenerator::stop_interpreter_at() {
2052 Label L;
2053 __ push(rscratch1);
2054 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2055 __ ldr(rscratch1, Address(rscratch1));
2056 __ mov(rscratch2, StopInterpreterAt);
2057 __ cmpw(rscratch1, rscratch2);
2058 __ br(Assembler::NE, L);
2059 __ brk(0);
2060 __ bind(L);
2061 __ pop(rscratch1);
2062 }
2063
2064 #endif // !PRODUCT
2065 #endif // ! CC_INTERP
2066