1 /*
2 * Copyright (c) 2003, 2023, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * Copyright (c) 2020, 2022, Huawei Technologies Co., Ltd. All rights reserved.
5 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 *
7 * This code is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 only, as
9 * published by the Free Software Foundation.
10 *
11 * This code is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 * version 2 for more details (a copy is included in the LICENSE file that
15 * accompanied this code).
16 *
17 * You should have received a copy of the GNU General Public License version
18 * 2 along with this work; if not, write to the Free Software Foundation,
19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 *
21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
22 * or visit www.oracle.com if you need additional information or have any
23 * questions.
24 *
25 */
26
27 #include "precompiled.hpp"
28 #include "asm/macroAssembler.inline.hpp"
29 #include "classfile/javaClasses.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interpreter/bytecodeHistogram.hpp"
32 #include "interpreter/bytecodeTracer.hpp"
33 #include "interpreter/interp_masm.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "interpreter/interpreterRuntime.hpp"
36 #include "interpreter/templateInterpreterGenerator.hpp"
37 #include "interpreter/templateTable.hpp"
38 #include "memory/resourceArea.hpp"
39 #include "oops/arrayOop.hpp"
40 #include "oops/method.inline.hpp"
41 #include "oops/methodData.hpp"
42 #include "oops/oop.inline.hpp"
43 #include "oops/resolvedIndyEntry.hpp"
44 #include "oops/resolvedMethodEntry.hpp"
45 #include "prims/jvmtiExport.hpp"
46 #include "prims/jvmtiThreadState.hpp"
47 #include "runtime/arguments.hpp"
48 #include "runtime/deoptimization.hpp"
49 #include "runtime/frame.inline.hpp"
50 #include "runtime/globals.hpp"
51 #include "runtime/jniHandles.hpp"
52 #include "runtime/sharedRuntime.hpp"
53 #include "runtime/stubRoutines.hpp"
54 #include "runtime/synchronizer.hpp"
55 #include "runtime/timer.hpp"
56 #include "runtime/vframeArray.hpp"
57 #include "utilities/checkedCast.hpp"
58 #include "utilities/debug.hpp"
59 #include "utilities/powerOfTwo.hpp"
60 #include <sys/types.h>
61
62 #ifndef PRODUCT
63 #include "oops/method.hpp"
64 #endif // !PRODUCT
65
66 // Size of interpreter code. Increase if too small. Interpreter will
67 // fail with a guarantee ("not enough space for interpreter generation");
68 // if too small.
69 // Run with +PrintInterpreter to get the VM to print out the size.
70 // Max size with JVMTI
71 int TemplateInterpreter::InterpreterCodeSize = 256 * 1024;
72
73 #define __ _masm->
74
75 //-----------------------------------------------------------------------------
76
77 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
78 address entry = __ pc();
79
80 __ andi(esp, esp, -16);
81 __ mv(c_rarg3, esp);
82 // xmethod
83 // xlocals
84 // c_rarg3: first stack arg - wordSize
85 // adjust sp
86
87 __ addi(sp, c_rarg3, -18 * wordSize);
88 __ addi(sp, sp, -2 * wordSize);
89 __ sd(ra, Address(sp, 0));
90
91 __ call_VM(noreg,
92 CAST_FROM_FN_PTR(address,
93 InterpreterRuntime::slow_signature_handler),
94 xmethod, xlocals, c_rarg3);
95
96 // x10: result handler
97
98 // Stack layout:
99 // sp: return address <- sp
100 // 1 garbage
101 // 8 integer args (if static first is unused)
102 // 1 float/double identifiers
103 // 8 double args
104 // stack args <- esp
105 // garbage
106 // expression stack bottom
107 // bcp (null)
108 // ...
109
110 // Restore ra
111 __ ld(ra, Address(sp, 0));
112 __ addi(sp, sp , 2 * wordSize);
113
114 // Do FP first so we can use c_rarg3 as temp
115 __ lwu(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
116
117 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
118 const FloatRegister r = g_FPArgReg[i];
119 Label d, done;
120
121 __ test_bit(t0, c_rarg3, i);
122 __ bnez(t0, d);
123 __ flw(r, Address(sp, (10 + i) * wordSize));
124 __ j(done);
125 __ bind(d);
126 __ fld(r, Address(sp, (10 + i) * wordSize));
127 __ bind(done);
128 }
129
130 // c_rarg0 contains the result from the call of
131 // InterpreterRuntime::slow_signature_handler so we don't touch it
132 // here. It will be loaded with the JNIEnv* later.
133 for (int i = 1; i < Argument::n_int_register_parameters_c; i++) {
134 const Register rm = g_INTArgReg[i];
135 __ ld(rm, Address(sp, i * wordSize));
136 }
137
138 __ addi(sp, sp, 18 * wordSize);
139 __ ret();
140
141 return entry;
142 }
143
144 // Various method entries
145 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
146 // xmethod: Method*
147 // x19_sender_sp: sender sp
148 // esp: args
149
150 // These don't need a safepoint check because they aren't virtually
151 // callable. We won't enter these intrinsics from compiled code.
152 // If in the future we added an intrinsic which was virtually callable
153 // we'd have to worry about how to safepoint so that this code is used.
154
155 // mathematical functions inlined by compiler
156 // (interpreter must provide identical implementation
157 // in order to avoid monotonicity bugs when switching
158 // from interpreter to compiler in the middle of some
159 // computation)
160 //
161 // stack:
162 // [ arg ] <-- esp
163 // [ arg ]
164 // retaddr in ra
165
166 address fn = nullptr;
167 address entry_point = nullptr;
168 Register continuation = ra;
169 switch (kind) {
170 case Interpreter::java_lang_math_abs:
171 entry_point = __ pc();
172 __ fld(f10, Address(esp));
173 __ fabs_d(f10, f10);
174 __ mv(sp, x19_sender_sp); // Restore caller's SP
175 break;
176 case Interpreter::java_lang_math_sqrt:
177 entry_point = __ pc();
178 __ fld(f10, Address(esp));
179 __ fsqrt_d(f10, f10);
180 __ mv(sp, x19_sender_sp);
181 break;
182 case Interpreter::java_lang_math_sin :
183 entry_point = __ pc();
184 __ fld(f10, Address(esp));
185 __ mv(sp, x19_sender_sp);
186 __ mv(x9, ra);
187 continuation = x9; // The first callee-saved register
188 if (StubRoutines::dsin() == nullptr) {
189 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
190 } else {
191 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
192 }
193 __ call(fn);
194 break;
195 case Interpreter::java_lang_math_cos :
196 entry_point = __ pc();
197 __ fld(f10, Address(esp));
198 __ mv(sp, x19_sender_sp);
199 __ mv(x9, ra);
200 continuation = x9; // The first callee-saved register
201 if (StubRoutines::dcos() == nullptr) {
202 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
203 } else {
204 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
205 }
206 __ call(fn);
207 break;
208 case Interpreter::java_lang_math_tan :
209 entry_point = __ pc();
210 __ fld(f10, Address(esp));
211 __ mv(sp, x19_sender_sp);
212 __ mv(x9, ra);
213 continuation = x9; // The first callee-saved register
214 if (StubRoutines::dtan() == nullptr) {
215 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
216 } else {
217 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
218 }
219 __ call(fn);
220 break;
221 case Interpreter::java_lang_math_log :
222 entry_point = __ pc();
223 __ fld(f10, Address(esp));
224 __ mv(sp, x19_sender_sp);
225 __ mv(x9, ra);
226 continuation = x9; // The first callee-saved register
227 if (StubRoutines::dlog() == nullptr) {
228 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
229 } else {
230 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
231 }
232 __ call(fn);
233 break;
234 case Interpreter::java_lang_math_log10 :
235 entry_point = __ pc();
236 __ fld(f10, Address(esp));
237 __ mv(sp, x19_sender_sp);
238 __ mv(x9, ra);
239 continuation = x9; // The first callee-saved register
240 if (StubRoutines::dlog10() == nullptr) {
241 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
242 } else {
243 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
244 }
245 __ call(fn);
246 break;
247 case Interpreter::java_lang_math_exp :
248 entry_point = __ pc();
249 __ fld(f10, Address(esp));
250 __ mv(sp, x19_sender_sp);
251 __ mv(x9, ra);
252 continuation = x9; // The first callee-saved register
253 if (StubRoutines::dexp() == nullptr) {
254 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
255 } else {
256 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
257 }
258 __ call(fn);
259 break;
260 case Interpreter::java_lang_math_pow :
261 entry_point = __ pc();
262 __ mv(x9, ra);
263 continuation = x9;
264 __ fld(f10, Address(esp, 2 * Interpreter::stackElementSize));
265 __ fld(f11, Address(esp));
266 __ mv(sp, x19_sender_sp);
267 if (StubRoutines::dpow() == nullptr) {
268 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
269 } else {
270 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
271 }
272 __ call(fn);
273 break;
274 case Interpreter::java_lang_math_fmaD :
275 if (UseFMA) {
276 entry_point = __ pc();
277 __ fld(f10, Address(esp, 4 * Interpreter::stackElementSize));
278 __ fld(f11, Address(esp, 2 * Interpreter::stackElementSize));
279 __ fld(f12, Address(esp));
280 __ fmadd_d(f10, f10, f11, f12);
281 __ mv(sp, x19_sender_sp); // Restore caller's SP
282 }
283 break;
284 case Interpreter::java_lang_math_fmaF :
285 if (UseFMA) {
286 entry_point = __ pc();
287 __ flw(f10, Address(esp, 2 * Interpreter::stackElementSize));
288 __ flw(f11, Address(esp, Interpreter::stackElementSize));
289 __ flw(f12, Address(esp));
290 __ fmadd_s(f10, f10, f11, f12);
291 __ mv(sp, x19_sender_sp); // Restore caller's SP
292 }
293 break;
294 default:
295 ;
296 }
297 if (entry_point != nullptr) {
298 __ jr(continuation);
299 }
300
301 return entry_point;
302 }
303
304 // Abstract method entry
305 // Attempt to execute abstract method. Throw exception
306 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
307 // xmethod: Method*
308 // x19_sender_sp: sender SP
309
310 address entry_point = __ pc();
311
312 // abstract method entry
313
314 // pop return address, reset last_sp to null
315 __ empty_expression_stack();
316 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
317 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
318
319 // throw exception
320 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
321 InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
322 xmethod);
323 // the call_VM checks for exception, so we should never return here.
324 __ should_not_reach_here();
325
326 return entry_point;
327 }
328
329 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
330 address entry = __ pc();
331
332 #ifdef ASSERT
333 {
334 Label L;
335 __ ld(t0, Address(fp, frame::interpreter_frame_monitor_block_top_offset * wordSize));
336 __ shadd(t0, t0, fp, t0, LogBytesPerWord);
337 // maximal sp for current fp (stack grows negative)
338 // check if frame is complete
339 __ bge(t0, sp, L);
340 __ stop ("interpreter frame not set up");
341 __ bind(L);
342 }
343 #endif // ASSERT
344 // Restore bcp under the assumption that the current frame is still
345 // interpreted
346 __ restore_bcp();
347
348 // expression stack must be empty before entering the VM if an
349 // exception happened
350 __ empty_expression_stack();
351 // throw exception
352 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
353 return entry;
354 }
355
356 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
357 address entry = __ pc();
358 // expression stack must be empty before entering the VM if an
359 // exception happened
360 __ empty_expression_stack();
361 // setup parameters
362
363 // convention: expect aberrant index in register x11
364 __ zero_extend(c_rarg2, x11, 32);
365 // convention: expect array in register x13
366 __ mv(c_rarg1, x13);
367 __ call_VM(noreg,
368 CAST_FROM_FN_PTR(address,
369 InterpreterRuntime::
370 throw_ArrayIndexOutOfBoundsException),
371 c_rarg1, c_rarg2);
372 return entry;
373 }
374
375 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
376 address entry = __ pc();
377
378 // object is at TOS
379 __ pop_reg(c_rarg1);
380
381 // expression stack must be empty before entering the VM if an
382 // exception happened
383 __ empty_expression_stack();
384
385 __ call_VM(noreg,
386 CAST_FROM_FN_PTR(address,
387 InterpreterRuntime::
388 throw_ClassCastException),
389 c_rarg1);
390 return entry;
391 }
392
393 address TemplateInterpreterGenerator::generate_exception_handler_common(
394 const char* name, const char* message, bool pass_oop) {
395 assert(!pass_oop || message == nullptr, "either oop or message but not both");
396 address entry = __ pc();
397 if (pass_oop) {
398 // object is at TOS
399 __ pop_reg(c_rarg2);
400 }
401 // expression stack must be empty before entering the VM if an
402 // exception happened
403 __ empty_expression_stack();
404 // setup parameters
405 __ la(c_rarg1, Address((address)name));
406 if (pass_oop) {
407 __ call_VM(x10, CAST_FROM_FN_PTR(address,
408 InterpreterRuntime::
409 create_klass_exception),
410 c_rarg1, c_rarg2);
411 } else {
412 // kind of lame ExternalAddress can't take null because
413 // external_word_Relocation will assert.
414 if (message != nullptr) {
415 __ la(c_rarg2, Address((address)message));
416 } else {
417 __ mv(c_rarg2, NULL_WORD);
418 }
419 __ call_VM(x10,
420 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
421 c_rarg1, c_rarg2);
422 }
423 // throw exception
424 __ j(address(Interpreter::throw_exception_entry()));
425 return entry;
426 }
427
428 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
429 address entry = __ pc();
430
431 // Restore stack bottom in case i2c adjusted stack
432 __ ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
433 __ shadd(esp, t0, fp, t0, LogBytesPerWord);
434 // and null it as marker that esp is now tos until next java call
435 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
436 __ restore_bcp();
437 __ restore_locals();
438 __ restore_constant_pool_cache();
439 __ get_method(xmethod);
440
441 if (state == atos) {
442 Register obj = x10;
443 Register mdp = x11;
444 Register tmp = x12;
445 __ ld(mdp, Address(xmethod, Method::method_data_offset()));
446 __ profile_return_type(mdp, obj, tmp);
447 }
448
449 const Register cache = x11;
450 const Register index = x12;
451
452 if (index_size == sizeof(u4)) {
453 __ load_resolved_indy_entry(cache, index);
454 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
455 __ shadd(esp, cache, esp, t0, 3);
456 } else {
457 // Pop N words from the stack
458 assert(index_size == sizeof(u2), "Can only be u2");
459 __ load_method_entry(cache, index);
460 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
461
462 __ shadd(esp, cache, esp, t0, 3);
463 }
464
465 // Restore machine SP
466 __ restore_sp_after_call();
467
468 __ check_and_handle_popframe(xthread);
469 __ check_and_handle_earlyret(xthread);
470
471 __ get_dispatch();
472 __ dispatch_next(state, step);
473
474 return entry;
475 }
476
477 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
478 int step,
479 address continuation) {
480 address entry = __ pc();
481 __ restore_bcp();
482 __ restore_locals();
483 __ restore_constant_pool_cache();
484 __ get_method(xmethod);
485 __ get_dispatch();
486
487 __ restore_sp_after_call(); // Restore SP to extended SP
488
489 // Restore expression stack pointer
490 __ ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
491 __ shadd(esp, t0, fp, t0, LogBytesPerWord);
492 // null last_sp until next java call
493 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
494
495 // handle exceptions
496 {
497 Label L;
498 __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
499 __ beqz(t0, L);
500 __ call_VM(noreg,
501 CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
502 __ should_not_reach_here();
503 __ bind(L);
504 }
505
506 if (continuation == nullptr) {
507 __ dispatch_next(state, step);
508 } else {
509 __ jump_to_entry(continuation);
510 }
511 return entry;
512 }
513
514 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
515 address entry = __ pc();
516 if (type == T_OBJECT) {
517 // retrieve result from frame
518 __ ld(x10, Address(fp, frame::interpreter_frame_oop_temp_offset * wordSize));
519 // and verify it
520 __ verify_oop(x10);
521 } else {
522 __ cast_primitive_type(type, x10);
523 }
524
525 __ ret(); // return from result handler
526 return entry;
527 }
528
529 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state,
530 address runtime_entry) {
531 assert_cond(runtime_entry != nullptr);
532 address entry = __ pc();
533 __ push(state);
534 __ push_cont_fastpath(xthread);
535 __ call_VM(noreg, runtime_entry);
536 __ pop_cont_fastpath(xthread);
537 __ membar(MacroAssembler::AnyAny);
538 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
539 return entry;
540 }
541
542 // Helpers for commoning out cases in the various type of method entries.
543 //
544
545
546 // increment invocation count & check for overflow
547 //
548 // Note: checking for negative value instead of overflow
549 // so we have a 'sticky' overflow test
550 //
551 // xmethod: method
552 //
553 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
554 Label done;
555 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
556 int increment = InvocationCounter::count_increment;
557 Label no_mdo;
558 if (ProfileInterpreter) {
559 // Are we profiling?
560 __ ld(x10, Address(xmethod, Method::method_data_offset()));
561 __ beqz(x10, no_mdo);
562 // Increment counter in the MDO
563 const Address mdo_invocation_counter(x10, in_bytes(MethodData::invocation_counter_offset()) +
564 in_bytes(InvocationCounter::counter_offset()));
565 const Address mask(x10, in_bytes(MethodData::invoke_mask_offset()));
566 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, t0, t1, false, overflow);
567 __ j(done);
568 }
569 __ bind(no_mdo);
570 // Increment counter in MethodCounters
571 const Address invocation_counter(t1,
572 MethodCounters::invocation_counter_offset() +
573 InvocationCounter::counter_offset());
574 __ get_method_counters(xmethod, t1, done);
575 const Address mask(t1, in_bytes(MethodCounters::invoke_mask_offset()));
576 __ increment_mask_and_jump(invocation_counter, increment, mask, t0, x11, false, overflow);
577 __ bind(done);
578 }
579
580 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
581 __ mv(c_rarg1, zr);
582 __ call_VM(noreg,
583 CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), c_rarg1);
584 __ j(do_continue);
585 }
586
587 // See if we've got enough room on the stack for locals plus overhead
588 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
589 // without going through the signal handler, i.e., reserved and yellow zones
590 // will not be made usable. The shadow zone must suffice to handle the
591 // overflow.
592 // The expression stack grows down incrementally, so the normal guard
593 // page mechanism will work for that.
594 //
595 // NOTE: Since the additional locals are also always pushed (wasn't
596 // obvious in generate_method_entry) so the guard should work for them
597 // too.
598 //
599 // Args:
600 // x13: number of additional locals this frame needs (what we must check)
601 // xmethod: Method*
602 //
603 // Kills:
604 // x10
605 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
606
607 // monitor entry size: see picture of stack set
608 // (generate_method_entry) and frame_amd64.hpp
609 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
610
611 // total overhead size: entry_size + (saved fp through expr stack
612 // bottom). be sure to change this if you add/subtract anything
613 // to/from the overhead area
614 const int overhead_size =
615 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
616
617 const int page_size = (int)os::vm_page_size();
618
619 Label after_frame_check;
620
621 // see if the frame is greater than one page in size. If so,
622 // then we need to verify there is enough stack space remaining
623 // for the additional locals.
624 __ mv(t0, (page_size - overhead_size) / Interpreter::stackElementSize);
625 __ bleu(x13, t0, after_frame_check);
626
627 // compute sp as if this were going to be the last frame on
628 // the stack before the red zone
629
630 // locals + overhead, in bytes
631 __ mv(x10, overhead_size);
632 __ shadd(x10, x13, x10, t0, Interpreter::logStackElementSize); // 2 slots per parameter.
633
634 const Address stack_limit(xthread, JavaThread::stack_overflow_limit_offset());
635 __ ld(t0, stack_limit);
636
637 #ifdef ASSERT
638 Label limit_okay;
639 // Verify that thread stack limit is non-zero.
640 __ bnez(t0, limit_okay);
641 __ stop("stack overflow limit is zero");
642 __ bind(limit_okay);
643 #endif
644
645 // Add stack limit to locals.
646 __ add(x10, x10, t0);
647
648 // Check against the current stack bottom.
649 __ bgtu(sp, x10, after_frame_check);
650
651 // Remove the incoming args, peeling the machine SP back to where it
652 // was in the caller. This is not strictly necessary, but unless we
653 // do so the stack frame may have a garbage FP; this ensures a
654 // correct call stack that we can always unwind. The ANDI should be
655 // unnecessary because the sender SP in x19 is always aligned, but
656 // it doesn't hurt.
657 __ andi(sp, x19_sender_sp, -16);
658
659 // Note: the restored frame is not necessarily interpreted.
660 // Use the shared runtime version of the StackOverflowError.
661 assert(StubRoutines::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
662 __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry()));
663
664 // all done with frame size check
665 __ bind(after_frame_check);
666 }
667
668 // Allocate monitor and lock method (asm interpreter)
669 //
670 // Args:
671 // xmethod: Method*
672 // xlocals: locals
673 //
674 // Kills:
675 // x10
676 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
677 // t0, t1 (temporary regs)
678 void TemplateInterpreterGenerator::lock_method() {
679 // synchronize method
680 const Address access_flags(xmethod, Method::access_flags_offset());
681 const Address monitor_block_top(fp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
682 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
683
684 #ifdef ASSERT
685 __ lwu(x10, access_flags);
686 __ verify_access_flags(x10, JVM_ACC_SYNCHRONIZED, "method doesn't need synchronization", false);
687 #endif // ASSERT
688
689 // get synchronization object
690 {
691 Label done;
692 __ lwu(x10, access_flags);
693 __ andi(t0, x10, JVM_ACC_STATIC);
694 // get receiver (assume this is frequent case)
695 __ ld(x10, Address(xlocals, Interpreter::local_offset_in_bytes(0)));
696 __ beqz(t0, done);
697 __ load_mirror(x10, xmethod, x15, t1);
698
699 #ifdef ASSERT
700 {
701 Label L;
702 __ bnez(x10, L);
703 __ stop("synchronization object is null");
704 __ bind(L);
705 }
706 #endif // ASSERT
707
708 __ bind(done);
709 }
710
711 // add space for monitor & lock
712 __ check_extended_sp();
713 __ add(sp, sp, - entry_size); // add space for a monitor entry
714 __ add(esp, esp, - entry_size);
715 __ sub(t0, sp, fp);
716 __ srai(t0, t0, Interpreter::logStackElementSize);
717 __ sd(t0, Address(fp, frame::interpreter_frame_extended_sp_offset * wordSize));
718 __ sub(t0, esp, fp);
719 __ srai(t0, t0, Interpreter::logStackElementSize);
720 __ sd(t0, monitor_block_top); // set new monitor block top
721 // store object
722 __ sd(x10, Address(esp, BasicObjectLock::obj_offset()));
723 __ mv(c_rarg1, esp); // object address
724 __ lock_object(c_rarg1);
725 }
726
727 // Generate a fixed interpreter frame. This is identical setup for
728 // interpreted methods and for native methods hence the shared code.
729 //
730 // Args:
731 // ra: return address
732 // xmethod: Method*
733 // xlocals: pointer to locals
734 // xcpool: cp cache
735 // stack_pointer: previous sp
736 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
737 // initialize fixed part of activation frame
738 if (native_call) {
739 __ add(esp, sp, - 14 * wordSize);
740 __ mv(xbcp, zr);
741 __ add(sp, sp, - 14 * wordSize);
742 // add 2 zero-initialized slots for native calls
743 __ sd(zr, Address(sp, 13 * wordSize));
744 __ sd(zr, Address(sp, 12 * wordSize));
745 } else {
746 __ add(esp, sp, - 12 * wordSize);
747 __ ld(t0, Address(xmethod, Method::const_offset())); // get ConstMethod
748 __ add(xbcp, t0, in_bytes(ConstMethod::codes_offset())); // get codebase
749 __ add(sp, sp, - 12 * wordSize);
750 }
751 __ sd(xbcp, Address(sp, wordSize));
752 __ mv(t0, frame::interpreter_frame_initial_sp_offset);
753 __ sd(t0, Address(sp, 0));
754
755 if (ProfileInterpreter) {
756 Label method_data_continue;
757 __ ld(t0, Address(xmethod, Method::method_data_offset()));
758 __ beqz(t0, method_data_continue);
759 __ la(t0, Address(t0, in_bytes(MethodData::data_offset())));
760 __ bind(method_data_continue);
761 }
762
763 __ sd(xmethod, Address(sp, 7 * wordSize));
764 __ sd(ProfileInterpreter ? t0 : zr, Address(sp, 6 * wordSize));
765
766 __ sd(ra, Address(sp, 11 * wordSize));
767 __ sd(fp, Address(sp, 10 * wordSize));
768 __ la(fp, Address(sp, 12 * wordSize)); // include ra & fp
769
770 __ ld(xcpool, Address(xmethod, Method::const_offset()));
771 __ ld(xcpool, Address(xcpool, ConstMethod::constants_offset()));
772 __ ld(xcpool, Address(xcpool, ConstantPool::cache_offset()));
773 __ sd(xcpool, Address(sp, 3 * wordSize));
774 __ sub(t0, xlocals, fp);
775 __ srai(t0, t0, Interpreter::logStackElementSize); // t0 = xlocals - fp();
776 // Store relativized xlocals, see frame::interpreter_frame_locals().
777 __ sd(t0, Address(sp, 2 * wordSize));
778
779 // set sender sp
780 // leave last_sp as null
781 __ sd(x19_sender_sp, Address(sp, 9 * wordSize));
782 __ sd(zr, Address(sp, 8 * wordSize));
783
784 // Get mirror and store it in the frame as GC root for this Method*
785 __ load_mirror(t2, xmethod, x15, t1);
786 __ sd(t2, Address(sp, 4 * wordSize));
787
788 if (!native_call) {
789 __ ld(t0, Address(xmethod, Method::const_offset()));
790 __ lhu(t0, Address(t0, ConstMethod::max_stack_offset()));
791 __ add(t0, t0, MAX2(3, Method::extra_stack_entries()));
792 __ slli(t0, t0, 3);
793 __ sub(t0, sp, t0);
794 __ andi(t0, t0, -16);
795 __ sub(t1, t0, fp);
796 __ srai(t1, t1, Interpreter::logStackElementSize);
797 // Store extended SP
798 __ sd(t1, Address(sp, 5 * wordSize));
799 // Move SP out of the way
800 __ mv(sp, t0);
801 } else {
802 // Make sure there is room for the exception oop pushed in case method throws
803 // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
804 __ sub(t0, sp, 2 * wordSize);
805 __ sub(t1, t0, fp);
806 __ srai(t1, t1, Interpreter::logStackElementSize);
807 __ sd(t1, Address(sp, 5 * wordSize));
808 __ mv(sp, t0);
809 }
810 }
811
812 // End of helpers
813
814 // Various method entries
815 //------------------------------------------------------------------------------------------------------------------------
816 //
817 //
818
819 // Method entry for java.lang.ref.Reference.get.
820 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
821 // Code: _aload_0, _getfield, _areturn
822 // parameter size = 1
823 //
824 // The code that gets generated by this routine is split into 2 parts:
825 // 1. The "intrinsified" code for G1 (or any SATB based GC),
826 // 2. The slow path - which is an expansion of the regular method entry.
827 //
828 // Notes:-
829 // * In the G1 code we do not check whether we need to block for
830 // a safepoint. If G1 is enabled then we must execute the specialized
831 // code for Reference.get (except when the Reference object is null)
832 // so that we can log the value in the referent field with an SATB
833 // update buffer.
834 // If the code for the getfield template is modified so that the
835 // G1 pre-barrier code is executed when the current method is
836 // Reference.get() then going through the normal method entry
837 // will be fine.
838 // * The G1 code can, however, check the receiver object (the instance
839 // of java.lang.Reference) and jump to the slow path if null. If the
840 // Reference object is null then we obviously cannot fetch the referent
841 // and so we don't need to call the G1 pre-barrier. Thus we can use the
842 // regular method entry code to generate the NPE.
843 //
844 // This code is based on generate_accessor_entry.
845 //
846 // xmethod: Method*
847 // x19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
848
849 // ra is live. It must be saved around calls.
850
851 address entry = __ pc();
852
853 const int referent_offset = java_lang_ref_Reference::referent_offset();
854 guarantee(referent_offset > 0, "referent offset not initialized");
855
856 Label slow_path;
857 const Register local_0 = c_rarg0;
858 // Check if local 0 isn't null
859 // If the receiver is null then it is OK to jump to the slow path.
860 __ ld(local_0, Address(esp, 0));
861 __ beqz(local_0, slow_path);
862
863 // Load the value of the referent field.
864 const Address field_address(local_0, referent_offset);
865 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
866 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ t0, /*tmp2*/ t1);
867
868 // areturn
869 __ andi(sp, x19_sender_sp, -16); // done with stack
870 __ ret();
871
872 // generate a vanilla interpreter entry as the slow path
873 __ bind(slow_path);
874 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
875 return entry;
876 }
877
878 /**
879 * Method entry for static native methods:
880 * int java.util.zip.CRC32.update(int crc, int b)
881 */
882 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
883 // TODO: Unimplemented generate_CRC32_update_entry
884 return nullptr;
885 }
886
887 /**
888 * Method entry for static native methods:
889 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
890 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
891 */
892 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
893 // TODO: Unimplemented generate_CRC32_updateBytes_entry
894 return nullptr;
895 }
896
897 /**
898 * Method entry for intrinsic-candidate (non-native) methods:
899 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
900 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
901 * Unlike CRC32, CRC32C does not have any methods marked as native
902 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
903 */
904 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
905 // TODO: Unimplemented generate_CRC32C_updateBytes_entry
906 return nullptr;
907 }
908
909 // Not supported
910 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
911 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
912 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
913 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
914 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() { return nullptr; }
915 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() { return nullptr; }
916
917 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
918 // See more discussion in stackOverflow.hpp.
919
920 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
921 const int page_size = (int)os::vm_page_size();
922 const int n_shadow_pages = shadow_zone_size / page_size;
923
924 #ifdef ASSERT
925 Label L_good_limit;
926 __ ld(t0, Address(xthread, JavaThread::shadow_zone_safe_limit()));
927 __ bnez(t0, L_good_limit);
928 __ stop("shadow zone safe limit is not initialized");
929 __ bind(L_good_limit);
930
931 Label L_good_watermark;
932 __ ld(t0, Address(xthread, JavaThread::shadow_zone_growth_watermark()));
933 __ bnez(t0, L_good_watermark);
934 __ stop("shadow zone growth watermark is not initialized");
935 __ bind(L_good_watermark);
936 #endif
937
938 Label L_done;
939
940 __ ld(t0, Address(xthread, JavaThread::shadow_zone_growth_watermark()));
941 __ bgtu(sp, t0, L_done);
942
943 for (int p = 1; p <= n_shadow_pages; p++) {
944 __ bang_stack_with_offset(p * page_size);
945 }
946
947 // Record the new watermark, but only if the update is above the safe limit.
948 // Otherwise, the next time around the check above would pass the safe limit.
949 __ ld(t0, Address(xthread, JavaThread::shadow_zone_safe_limit()));
950 __ bleu(sp, t0, L_done);
951 __ sd(sp, Address(xthread, JavaThread::shadow_zone_growth_watermark()));
952
953 __ bind(L_done);
954 }
955
956 // Interpreter stub for calling a native method. (asm interpreter)
957 // This sets up a somewhat different looking stack for calling the
958 // native method than the typical interpreter frame setup.
959 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
960 // determine code generation flags
961 bool inc_counter = UseCompiler || CountCompiledCalls;
962
963 // x11: Method*
964 // x30: sender sp
965
966 address entry_point = __ pc();
967
968 const Address constMethod (xmethod, Method::const_offset());
969 const Address access_flags (xmethod, Method::access_flags_offset());
970 const Address size_of_parameters(x12, ConstMethod::
971 size_of_parameters_offset());
972
973 // get parameter size (always needed)
974 __ ld(x12, constMethod);
975 __ load_unsigned_short(x12, size_of_parameters);
976
977 // Native calls don't need the stack size check since they have no
978 // expression stack and the arguments are already on the stack and
979 // we only add a handful of words to the stack.
980
981 // xmethod: Method*
982 // x12: size of parameters
983 // x30: sender sp
984
985 // for natives the size of locals is zero
986
987 // compute beginning of parameters (xlocals)
988 __ shadd(xlocals, x12, esp, xlocals, 3);
989 __ addi(xlocals, xlocals, -wordSize);
990
991 // Pull SP back to minimum size: this avoids holes in the stack
992 __ andi(sp, esp, -16);
993
994 // initialize fixed part of activation frame
995 generate_fixed_frame(true);
996
997 // make sure method is native & not abstract
998 #ifdef ASSERT
999 __ lwu(x10, access_flags);
1000 __ verify_access_flags(x10, JVM_ACC_NATIVE, "tried to execute non-native method as native", false);
1001 __ verify_access_flags(x10, JVM_ACC_ABSTRACT, "tried to execute abstract method in interpreter");
1002 #endif
1003
1004 // Since at this point in the method invocation the exception
1005 // handler would try to exit the monitor of synchronized methods
1006 // which hasn't been entered yet, we set the thread local variable
1007 // _do_not_unlock_if_synchronized to true. The remove_activation
1008 // will check this flag.
1009
1010 const Address do_not_unlock_if_synchronized(xthread,
1011 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1012 __ mv(t1, true);
1013 __ sb(t1, do_not_unlock_if_synchronized);
1014
1015 // increment invocation count & check for overflow
1016 Label invocation_counter_overflow;
1017 if (inc_counter) {
1018 generate_counter_incr(&invocation_counter_overflow);
1019 }
1020
1021 Label continue_after_compile;
1022 __ bind(continue_after_compile);
1023
1024 bang_stack_shadow_pages(true);
1025
1026 // reset the _do_not_unlock_if_synchronized flag
1027 __ sb(zr, do_not_unlock_if_synchronized);
1028
1029 // check for synchronized methods
1030 // Must happen AFTER invocation_counter check and stack overflow check,
1031 // so method is not locked if overflows.
1032 if (synchronized) {
1033 lock_method();
1034 } else {
1035 // no synchronization necessary
1036 #ifdef ASSERT
1037 __ lwu(x10, access_flags);
1038 __ verify_access_flags(x10, JVM_ACC_SYNCHRONIZED, "method needs synchronization");
1039 #endif
1040 }
1041
1042 // start execution
1043 #ifdef ASSERT
1044 __ verify_frame_setup();
1045 #endif
1046
1047 // jvmti support
1048 __ notify_method_entry();
1049
1050 // work registers
1051 const Register t = x18;
1052 const Register result_handler = x19;
1053
1054 // allocate space for parameters
1055 __ ld(t, Address(xmethod, Method::const_offset()));
1056 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1057
1058 __ slli(t, t, Interpreter::logStackElementSize);
1059 __ sub(x30, esp, t);
1060 __ andi(sp, x30, -16);
1061 __ mv(esp, x30);
1062
1063 // get signature handler
1064 {
1065 Label L;
1066 __ ld(t, Address(xmethod, Method::signature_handler_offset()));
1067 __ bnez(t, L);
1068 __ call_VM(noreg,
1069 CAST_FROM_FN_PTR(address,
1070 InterpreterRuntime::prepare_native_call),
1071 xmethod);
1072 __ ld(t, Address(xmethod, Method::signature_handler_offset()));
1073 __ bind(L);
1074 }
1075
1076 // call signature handler
1077 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == xlocals,
1078 "adjust this code");
1079 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1080 "adjust this code");
1081 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t0,
1082 "adjust this code");
1083
1084 // The generated handlers do not touch xmethod (the method).
1085 // However, large signatures cannot be cached and are generated
1086 // each time here. The slow-path generator can do a GC on return,
1087 // so we must reload it after the call.
1088 __ jalr(t);
1089 __ get_method(xmethod); // slow path can do a GC, reload xmethod
1090
1091
1092 // result handler is in x10
1093 // set result handler
1094 __ mv(result_handler, x10);
1095 // pass mirror handle if static call
1096 {
1097 Label L;
1098 __ lwu(t, Address(xmethod, Method::access_flags_offset()));
1099 __ test_bit(t0, t, exact_log2(JVM_ACC_STATIC));
1100 __ beqz(t0, L);
1101 // get mirror
1102 __ load_mirror(t, xmethod, x28, t1);
1103 // copy mirror into activation frame
1104 __ sd(t, Address(fp, frame::interpreter_frame_oop_temp_offset * wordSize));
1105 // pass handle to mirror
1106 __ addi(c_rarg1, fp, frame::interpreter_frame_oop_temp_offset * wordSize);
1107 __ bind(L);
1108 }
1109
1110 // get native function entry point in x28
1111 {
1112 Label L;
1113 __ ld(x28, Address(xmethod, Method::native_function_offset()));
1114 address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1115 __ mv(t, unsatisfied);
1116 __ load_long_misaligned(t1, Address(t, 0), t0, 2); // 2 bytes aligned, but not 4 or 8
1117
1118 __ bne(x28, t1, L);
1119 __ call_VM(noreg,
1120 CAST_FROM_FN_PTR(address,
1121 InterpreterRuntime::prepare_native_call),
1122 xmethod);
1123 __ get_method(xmethod);
1124 __ ld(x28, Address(xmethod, Method::native_function_offset()));
1125 __ bind(L);
1126 }
1127
1128 // pass JNIEnv
1129 __ add(c_rarg0, xthread, in_bytes(JavaThread::jni_environment_offset()));
1130
1131 // It is enough that the pc() points into the right code
1132 // segment. It does not have to be the correct return pc.
1133 Label native_return;
1134 __ set_last_Java_frame(esp, fp, native_return, x30);
1135
1136 // change thread state
1137 #ifdef ASSERT
1138 {
1139 Label L;
1140 __ lwu(t, Address(xthread, JavaThread::thread_state_offset()));
1141 __ addi(t0, zr, (u1)_thread_in_Java);
1142 __ beq(t, t0, L);
1143 __ stop("Wrong thread state in native stub");
1144 __ bind(L);
1145 }
1146 #endif
1147
1148 // Change state to native
1149 __ la(t1, Address(xthread, JavaThread::thread_state_offset()));
1150 __ mv(t0, _thread_in_native);
1151 __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1152 __ sw(t0, Address(t1));
1153
1154 // Call the native method.
1155 __ jalr(x28);
1156 __ bind(native_return);
1157 __ get_method(xmethod);
1158 // result potentially in x10 or f10
1159
1160 // Restore cpu control state after JNI call
1161 __ restore_cpu_control_state_after_jni(t0);
1162
1163 // make room for the pushes we're about to do
1164 __ sub(t0, esp, 4 * wordSize);
1165 __ andi(sp, t0, -16);
1166
1167 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1168 // in order to extract the result of a method call. If the order of these
1169 // pushes change or anything else is added to the stack then the code in
1170 // interpreter_frame_result must also change.
1171 __ push(dtos);
1172 __ push(ltos);
1173
1174 // change thread state
1175 // Force all preceding writes to be observed prior to thread state change
1176 __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1177
1178 __ mv(t0, _thread_in_native_trans);
1179 __ sw(t0, Address(xthread, JavaThread::thread_state_offset()));
1180
1181 // Force this write out before the read below
1182 if (!UseSystemMemoryBarrier) {
1183 __ membar(MacroAssembler::AnyAny);
1184 }
1185
1186 // check for safepoint operation in progress and/or pending suspend requests
1187 {
1188 Label L, Continue;
1189
1190 // We need an acquire here to ensure that any subsequent load of the
1191 // global SafepointSynchronize::_state flag is ordered after this load
1192 // of the thread-local polling word. We don't want this poll to
1193 // return false (i.e. not safepointing) and a later poll of the global
1194 // SafepointSynchronize::_state spuriously to return true.
1195 //
1196 // This is to avoid a race when we're in a native->Java transition
1197 // racing the code which wakes up from a safepoint.
1198 __ safepoint_poll(L, true /* at_return */, true /* acquire */, false /* in_nmethod */);
1199 __ lwu(t1, Address(xthread, JavaThread::suspend_flags_offset()));
1200 __ beqz(t1, Continue);
1201 __ bind(L);
1202
1203 // Don't use call_VM as it will see a possible pending exception
1204 // and forward it and never return here preventing us from
1205 // clearing _last_native_pc down below. So we do a runtime call by
1206 // hand.
1207 //
1208 __ mv(c_rarg0, xthread);
1209 __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1210 __ get_method(xmethod);
1211 __ reinit_heapbase();
1212 __ bind(Continue);
1213 }
1214
1215 // change thread state
1216 // Force all preceding writes to be observed prior to thread state change
1217 __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1218
1219 __ mv(t0, _thread_in_Java);
1220 __ sw(t0, Address(xthread, JavaThread::thread_state_offset()));
1221
1222 // reset_last_Java_frame
1223 __ reset_last_Java_frame(true);
1224
1225 if (CheckJNICalls) {
1226 // clear_pending_jni_exception_check
1227 __ sd(zr, Address(xthread, JavaThread::pending_jni_exception_check_fn_offset()));
1228 }
1229
1230 // reset handle block
1231 __ ld(t, Address(xthread, JavaThread::active_handles_offset()));
1232 __ sd(zr, Address(t, JNIHandleBlock::top_offset()));
1233
1234 // If result is an oop unbox and store it in frame where gc will see it
1235 // and result handler will pick it up
1236
1237 {
1238 Label no_oop;
1239 __ la(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1240 __ bne(t, result_handler, no_oop);
1241 // Unbox oop result, e.g. JNIHandles::resolve result.
1242 __ pop(ltos);
1243 __ resolve_jobject(x10, t, t1);
1244 __ sd(x10, Address(fp, frame::interpreter_frame_oop_temp_offset * wordSize));
1245 // keep stack depth as expected by pushing oop which will eventually be discarded
1246 __ push(ltos);
1247 __ bind(no_oop);
1248 }
1249
1250 {
1251 Label no_reguard;
1252 __ lwu(t0, Address(xthread, in_bytes(JavaThread::stack_guard_state_offset())));
1253 __ addi(t1, zr, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1254 __ bne(t0, t1, no_reguard);
1255
1256 __ push_call_clobbered_registers();
1257 __ mv(c_rarg0, xthread);
1258 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1259 __ pop_call_clobbered_registers();
1260 __ bind(no_reguard);
1261 }
1262
1263 // The method register is junk from after the thread_in_native transition
1264 // until here. Also can't call_VM until the bcp has been
1265 // restored. Need bcp for throwing exception below so get it now.
1266 __ get_method(xmethod);
1267
1268 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1269 // xbcp == code_base()
1270 __ ld(xbcp, Address(xmethod, Method::const_offset())); // get ConstMethod*
1271 __ add(xbcp, xbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1272 // handle exceptions (exception handling will handle unlocking!)
1273 {
1274 Label L;
1275 __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
1276 __ beqz(t0, L);
1277 // Note: At some point we may want to unify this with the code
1278 // used in call_VM_base(); i.e., we should use the
1279 // StubRoutines::forward_exception code. For now this doesn't work
1280 // here because the sp is not correctly set at this point.
1281 __ MacroAssembler::call_VM(noreg,
1282 CAST_FROM_FN_PTR(address,
1283 InterpreterRuntime::throw_pending_exception));
1284 __ should_not_reach_here();
1285 __ bind(L);
1286 }
1287
1288 // do unlocking if necessary
1289 {
1290 Label L;
1291 __ lwu(t, Address(xmethod, Method::access_flags_offset()));
1292 __ test_bit(t0, t, exact_log2(JVM_ACC_SYNCHRONIZED));
1293 __ beqz(t0, L);
1294 // the code below should be shared with interpreter macro
1295 // assembler implementation
1296 {
1297 Label unlock;
1298 // BasicObjectLock will be first in list, since this is a
1299 // synchronized method. However, need to check that the object
1300 // has not been unlocked by an explicit monitorexit bytecode.
1301
1302 // monitor expect in c_rarg1 for slow unlock path
1303 __ la(c_rarg1, Address(fp, // address of first monitor
1304 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1305 wordSize - sizeof(BasicObjectLock))));
1306
1307 __ ld(t, Address(c_rarg1, BasicObjectLock::obj_offset()));
1308 __ bnez(t, unlock);
1309
1310 // Entry already unlocked, need to throw exception
1311 __ MacroAssembler::call_VM(noreg,
1312 CAST_FROM_FN_PTR(address,
1313 InterpreterRuntime::throw_illegal_monitor_state_exception));
1314 __ should_not_reach_here();
1315
1316 __ bind(unlock);
1317 __ unlock_object(c_rarg1);
1318 }
1319 __ bind(L);
1320 }
1321
1322 // jvmti support
1323 // Note: This must happen _after_ handling/throwing any exceptions since
1324 // the exception handler code notifies the runtime of method exits
1325 // too. If this happens before, method entry/exit notifications are
1326 // not properly paired (was bug - gri 11/22/99).
1327 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1328
1329 __ pop(ltos);
1330 __ pop(dtos);
1331
1332 __ jalr(result_handler);
1333
1334 // remove activation
1335 __ ld(esp, Address(fp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1336 // remove frame anchor
1337 __ leave();
1338
1339 // restore sender sp
1340 __ mv(sp, esp);
1341
1342 __ ret();
1343
1344 if (inc_counter) {
1345 // Handle overflow of counter and compile method
1346 __ bind(invocation_counter_overflow);
1347 generate_counter_overflow(continue_after_compile);
1348 }
1349
1350 return entry_point;
1351 }
1352
1353 //
1354 // Generic interpreted method entry to (asm) interpreter
1355 //
1356 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1357
1358 // determine code generation flags
1359 const bool inc_counter = UseCompiler || CountCompiledCalls;
1360
1361 // t0: sender sp
1362 address entry_point = __ pc();
1363
1364 const Address constMethod(xmethod, Method::const_offset());
1365 const Address access_flags(xmethod, Method::access_flags_offset());
1366 const Address size_of_parameters(x13,
1367 ConstMethod::size_of_parameters_offset());
1368 const Address size_of_locals(x13, ConstMethod::size_of_locals_offset());
1369
1370 // get parameter size (always needed)
1371 // need to load the const method first
1372 __ ld(x13, constMethod);
1373 __ load_unsigned_short(x12, size_of_parameters);
1374
1375 // x12: size of parameters
1376
1377 __ load_unsigned_short(x13, size_of_locals); // get size of locals in words
1378 __ sub(x13, x13, x12); // x13 = no. of additional locals
1379
1380 // see if we've got enough room on the stack for locals plus overhead.
1381 generate_stack_overflow_check();
1382
1383 // compute beginning of parameters (xlocals)
1384 __ shadd(xlocals, x12, esp, t1, 3);
1385 __ add(xlocals, xlocals, -wordSize);
1386
1387 // Make room for additional locals
1388 __ slli(t1, x13, 3);
1389 __ sub(t0, esp, t1);
1390
1391 // Padding between locals and fixed part of activation frame to ensure
1392 // SP is always 16-byte aligned.
1393 __ andi(sp, t0, -16);
1394
1395 // x13 - # of additional locals
1396 // allocate space for locals
1397 // explicitly initialize locals
1398 {
1399 Label exit, loop;
1400 __ blez(x13, exit); // do nothing if x13 <= 0
1401 __ bind(loop);
1402 __ sd(zr, Address(t0));
1403 __ add(t0, t0, wordSize);
1404 __ add(x13, x13, -1); // until everything initialized
1405 __ bnez(x13, loop);
1406 __ bind(exit);
1407 }
1408
1409 // And the base dispatch table
1410 __ get_dispatch();
1411
1412 // initialize fixed part of activation frame
1413 generate_fixed_frame(false);
1414
1415 // make sure method is not native & not abstract
1416 #ifdef ASSERT
1417 __ lwu(x10, access_flags);
1418 __ verify_access_flags(x10, JVM_ACC_NATIVE, "tried to execute native method as non-native");
1419 __ verify_access_flags(x10, JVM_ACC_ABSTRACT, "tried to execute abstract method in interpreter");
1420 #endif
1421
1422 // Since at this point in the method invocation the exception
1423 // handler would try to exit the monitor of synchronized methods
1424 // which hasn't been entered yet, we set the thread local variable
1425 // _do_not_unlock_if_synchronized to true. The remove_activation
1426 // will check this flag.
1427
1428 const Address do_not_unlock_if_synchronized(xthread,
1429 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1430 __ mv(t1, true);
1431 __ sb(t1, do_not_unlock_if_synchronized);
1432
1433 Label no_mdp;
1434 const Register mdp = x13;
1435 __ ld(mdp, Address(xmethod, Method::method_data_offset()));
1436 __ beqz(mdp, no_mdp);
1437 __ add(mdp, mdp, in_bytes(MethodData::data_offset()));
1438 __ profile_parameters_type(mdp, x11, x12, x14); // use x11, x12, x14 as tmp registers
1439 __ bind(no_mdp);
1440
1441 // increment invocation count & check for overflow
1442 Label invocation_counter_overflow;
1443 if (inc_counter) {
1444 generate_counter_incr(&invocation_counter_overflow);
1445 }
1446
1447 Label continue_after_compile;
1448 __ bind(continue_after_compile);
1449
1450 bang_stack_shadow_pages(false);
1451
1452 // reset the _do_not_unlock_if_synchronized flag
1453 __ sb(zr, do_not_unlock_if_synchronized);
1454
1455 // check for synchronized methods
1456 // Must happen AFTER invocation_counter check and stack overflow check,
1457 // so method is not locked if overflows.
1458 if (synchronized) {
1459 // Allocate monitor and lock method
1460 lock_method();
1461 } else {
1462 // no synchronization necessary
1463 #ifdef ASSERT
1464 __ lwu(x10, access_flags);
1465 __ verify_access_flags(x10, JVM_ACC_SYNCHRONIZED, "method needs synchronization");
1466 #endif
1467 }
1468
1469 // start execution
1470 #ifdef ASSERT
1471 __ verify_frame_setup();
1472 #endif
1473
1474 // jvmti support
1475 __ notify_method_entry();
1476
1477 __ dispatch_next(vtos);
1478
1479 // invocation counter overflow
1480 if (inc_counter) {
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 // Method entry for java.lang.Thread.currentThread
1490 address TemplateInterpreterGenerator::generate_currentThread() {
1491 address entry_point = __ pc();
1492
1493 __ ld(x10, Address(xthread, JavaThread::vthread_offset()));
1494 __ resolve_oop_handle(x10, t0, t1);
1495 __ ret();
1496
1497 return entry_point;
1498 }
1499
1500 //-----------------------------------------------------------------------------
1501 // Exceptions
1502
1503 void TemplateInterpreterGenerator::generate_throw_exception() {
1504 // Entry point in previous activation (i.e., if the caller was
1505 // interpreted)
1506 Interpreter::_rethrow_exception_entry = __ pc();
1507 // Restore sp to interpreter_frame_last_sp even though we are going
1508 // to empty the expression stack for the exception processing.
1509 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1510 // x10: exception
1511 // x13: return address/pc that threw exception
1512 __ restore_bcp(); // xbcp points to call/send
1513 __ restore_locals();
1514 __ restore_constant_pool_cache();
1515 __ reinit_heapbase(); // restore xheapbase as heapbase.
1516 __ get_dispatch();
1517
1518 // Entry point for exceptions thrown within interpreter code
1519 Interpreter::_throw_exception_entry = __ pc();
1520 // If we came here via a NullPointerException on the receiver of a
1521 // method, xthread may be corrupt.
1522 __ get_method(xmethod);
1523 // expression stack is undefined here
1524 // x10: exception
1525 // xbcp: exception bcp
1526 __ verify_oop(x10);
1527 __ mv(c_rarg1, x10);
1528
1529 // expression stack must be empty before entering the VM in case of
1530 // an exception
1531 __ empty_expression_stack();
1532 // find exception handler address and preserve exception oop
1533 __ call_VM(x13,
1534 CAST_FROM_FN_PTR(address,
1535 InterpreterRuntime::exception_handler_for_exception),
1536 c_rarg1);
1537
1538 // Restore machine SP
1539 __ restore_sp_after_call();
1540
1541 // x10: exception handler entry point
1542 // x13: preserved exception oop
1543 // xbcp: bcp for exception handler
1544 __ push_ptr(x13); // push exception which is now the only value on the stack
1545 __ jr(x10); // jump to exception handler (may be _remove_activation_entry!)
1546
1547 // If the exception is not handled in the current frame the frame is
1548 // removed and the exception is rethrown (i.e. exception
1549 // continuation is _rethrow_exception).
1550 //
1551 // Note: At this point the bci is still the bxi for the instruction
1552 // which caused the exception and the expression stack is
1553 // empty. Thus, for any VM calls at this point, GC will find a legal
1554 // oop map (with empty expression stack).
1555
1556 //
1557 // JVMTI PopFrame support
1558 //
1559
1560 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1561 __ empty_expression_stack();
1562 // Set the popframe_processing bit in pending_popframe_condition
1563 // indicating that we are currently handling popframe, so that
1564 // call_VMs that may happen later do not trigger new popframe
1565 // handling cycles.
1566 __ lwu(x13, Address(xthread, JavaThread::popframe_condition_offset()));
1567 __ ori(x13, x13, JavaThread::popframe_processing_bit);
1568 __ sw(x13, Address(xthread, JavaThread::popframe_condition_offset()));
1569
1570 {
1571 // Check to see whether we are returning to a deoptimized frame.
1572 // (The PopFrame call ensures that the caller of the popped frame is
1573 // either interpreted or compiled and deoptimizes it if compiled.)
1574 // In this case, we can't call dispatch_next() after the frame is
1575 // popped, but instead must save the incoming arguments and restore
1576 // them after deoptimization has occurred.
1577 //
1578 // Note that we don't compare the return PC against the
1579 // deoptimization blob's unpack entry because of the presence of
1580 // adapter frames in C2.
1581 Label caller_not_deoptimized;
1582 __ ld(c_rarg1, Address(fp, frame::return_addr_offset * wordSize));
1583 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), c_rarg1);
1584 __ bnez(x10, caller_not_deoptimized);
1585
1586 // Compute size of arguments for saving when returning to
1587 // deoptimized caller
1588 __ get_method(x10);
1589 __ ld(x10, Address(x10, Method::const_offset()));
1590 __ load_unsigned_short(x10, Address(x10, in_bytes(ConstMethod::
1591 size_of_parameters_offset())));
1592 __ slli(x10, x10, Interpreter::logStackElementSize);
1593 __ restore_locals();
1594 __ sub(xlocals, xlocals, x10);
1595 __ add(xlocals, xlocals, wordSize);
1596 // Save these arguments
1597 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1598 Deoptimization::
1599 popframe_preserve_args),
1600 xthread, x10, xlocals);
1601
1602 __ remove_activation(vtos,
1603 /* throw_monitor_exception */ false,
1604 /* install_monitor_exception */ false,
1605 /* notify_jvmdi */ false);
1606
1607 // Inform deoptimization that it is responsible for restoring
1608 // these arguments
1609 __ mv(t0, JavaThread::popframe_force_deopt_reexecution_bit);
1610 __ sw(t0, Address(xthread, JavaThread::popframe_condition_offset()));
1611
1612 // Continue in deoptimization handler
1613 __ ret();
1614
1615 __ bind(caller_not_deoptimized);
1616 }
1617
1618 __ remove_activation(vtos,
1619 /* throw_monitor_exception */ false,
1620 /* install_monitor_exception */ false,
1621 /* notify_jvmdi */ false);
1622
1623 // Restore the last_sp and null it out
1624 __ ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1625 __ shadd(esp, t0, fp, t0, LogBytesPerWord);
1626 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1627
1628 __ restore_bcp();
1629 __ restore_locals();
1630 __ restore_constant_pool_cache();
1631 __ get_method(xmethod);
1632 __ get_dispatch();
1633
1634 // The method data pointer was incremented already during
1635 // call profiling. We have to restore the mdp for the current bcp.
1636 if (ProfileInterpreter) {
1637 __ set_method_data_pointer_for_bcp();
1638 }
1639
1640 // Clear the popframe condition flag
1641 __ sw(zr, Address(xthread, JavaThread::popframe_condition_offset()));
1642 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1643
1644 #if INCLUDE_JVMTI
1645 {
1646 Label L_done;
1647
1648 __ lbu(t0, Address(xbcp, 0));
1649 __ mv(t1, Bytecodes::_invokestatic);
1650 __ bne(t1, t0, L_done);
1651
1652 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1653 // Detect such a case in the InterpreterRuntime function and return the member name argument,or null.
1654
1655 __ ld(c_rarg0, Address(xlocals, 0));
1656 __ call_VM(x10, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null),c_rarg0, xmethod, xbcp);
1657
1658 __ beqz(x10, L_done);
1659
1660 __ sd(x10, Address(esp, 0));
1661 __ bind(L_done);
1662 }
1663 #endif // INCLUDE_JVMTI
1664
1665 // Restore machine SP
1666 __ restore_sp_after_call();
1667
1668 __ dispatch_next(vtos);
1669 // end of PopFrame support
1670
1671 Interpreter::_remove_activation_entry = __ pc();
1672
1673 // preserve exception over this code sequence
1674 __ pop_ptr(x10);
1675 __ sd(x10, Address(xthread, JavaThread::vm_result_offset()));
1676 // remove the activation (without doing throws on illegalMonitorExceptions)
1677 __ remove_activation(vtos, false, true, false);
1678 // restore exception
1679 __ get_vm_result(x10, xthread);
1680
1681 // In between activations - previous activation type unknown yet
1682 // compute continuation point - the continuation point expects the
1683 // following registers set up:
1684 //
1685 // x10: exception
1686 // ra: return address/pc that threw exception
1687 // sp: expression stack of caller
1688 // fp: fp of caller
1689 // FIXME: There's no point saving ra here because VM calls don't trash it
1690 __ sub(sp, sp, 2 * wordSize);
1691 __ sd(x10, Address(sp, 0)); // save exception
1692 __ sd(ra, Address(sp, wordSize)); // save return address
1693 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1694 SharedRuntime::exception_handler_for_return_address),
1695 xthread, ra);
1696 __ mv(x11, x10); // save exception handler
1697 __ ld(x10, Address(sp, 0)); // restore exception
1698 __ ld(ra, Address(sp, wordSize)); // restore return address
1699 __ add(sp, sp, 2 * wordSize);
1700 // We might be returning to a deopt handler that expects x13 to
1701 // contain the exception pc
1702 __ mv(x13, ra);
1703 // Note that an "issuing PC" is actually the next PC after the call
1704 __ jr(x11); // jump to exception
1705 // handler of caller
1706 }
1707
1708 //
1709 // JVMTI ForceEarlyReturn support
1710 //
1711 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1712 address entry = __ pc();
1713
1714 __ restore_bcp();
1715 __ restore_locals();
1716 __ empty_expression_stack();
1717 __ load_earlyret_value(state);
1718
1719 __ ld(t0, Address(xthread, JavaThread::jvmti_thread_state_offset()));
1720 Address cond_addr(t0, JvmtiThreadState::earlyret_state_offset());
1721
1722 // Clear the earlyret state
1723 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1724 __ sd(zr, cond_addr);
1725
1726 __ remove_activation(state,
1727 false, /* throw_monitor_exception */
1728 false, /* install_monitor_exception */
1729 true); /* notify_jvmdi */
1730 __ ret();
1731
1732 return entry;
1733 }
1734 // end of ForceEarlyReturn support
1735
1736 //-----------------------------------------------------------------------------
1737 // Helper for vtos entry point generation
1738
1739 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1740 address& bep,
1741 address& cep,
1742 address& sep,
1743 address& aep,
1744 address& iep,
1745 address& lep,
1746 address& fep,
1747 address& dep,
1748 address& vep) {
1749 assert(t != nullptr && t->is_valid() && t->tos_in() == vtos, "illegal template");
1750 Label L;
1751 aep = __ pc(); __ push_ptr(); __ j(L);
1752 fep = __ pc(); __ push_f(); __ j(L);
1753 dep = __ pc(); __ push_d(); __ j(L);
1754 lep = __ pc(); __ push_l(); __ j(L);
1755 bep = cep = sep =
1756 iep = __ pc(); __ push_i();
1757 vep = __ pc();
1758 __ bind(L);
1759 generate_and_dispatch(t);
1760 }
1761
1762 //-----------------------------------------------------------------------------
1763
1764 // Non-product code
1765 #ifndef PRODUCT
1766 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1767 address entry = __ pc();
1768
1769 __ push_reg(ra);
1770 __ push(state);
1771 __ push_reg(RegSet::range(x10, x17) + RegSet::range(x5, x7) + RegSet::range(x28, x31), sp);
1772 __ mv(c_rarg2, x10); // Pass itos
1773 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), c_rarg1, c_rarg2, c_rarg3);
1774 __ pop_reg(RegSet::range(x10, x17) + RegSet::range(x5, x7) + RegSet::range(x28, x31), sp);
1775 __ pop(state);
1776 __ pop_reg(ra);
1777 __ ret(); // return from result handler
1778
1779 return entry;
1780 }
1781
1782 void TemplateInterpreterGenerator::count_bytecode() {
1783 __ mv(x7, (address) &BytecodeCounter::_counter_value);
1784 __ atomic_addw(noreg, 1, x7);
1785 }
1786
1787 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1788 __ mv(x7, (address) &BytecodeHistogram::_counters[t->bytecode()]);
1789 __ atomic_addw(noreg, 1, x7);
1790 }
1791
1792 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1793 // Calculate new index for counter:
1794 // _index = (_index >> log2_number_of_codes) |
1795 // (bytecode << log2_number_of_codes);
1796 Register index_addr = t1;
1797 Register index = t0;
1798 __ mv(index_addr, (address) &BytecodePairHistogram::_index);
1799 __ lw(index, index_addr);
1800 __ mv(x7, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1801 __ srli(index, index, BytecodePairHistogram::log2_number_of_codes);
1802 __ orrw(index, x7, index);
1803 __ sw(index, index_addr);
1804 // Bump bucket contents:
1805 // _counters[_index] ++;
1806 Register counter_addr = t1;
1807 __ mv(x7, (address) &BytecodePairHistogram::_counters);
1808 __ shadd(counter_addr, index, x7, counter_addr, LogBytesPerInt);
1809 __ atomic_addw(noreg, 1, counter_addr);
1810 }
1811
1812 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1813 // Call a little run-time stub to avoid blow-up for each bytecode.
1814 // The run-time runtime saves the right registers, depending on
1815 // the tosca in-state for the given template.
1816
1817 assert(Interpreter::trace_code(t->tos_in()) != nullptr, "entry must have been generated");
1818 __ jal(Interpreter::trace_code(t->tos_in()));
1819 __ reinit_heapbase();
1820 }
1821
1822 void TemplateInterpreterGenerator::stop_interpreter_at() {
1823 Label L;
1824 __ push_reg(t0);
1825 __ mv(t0, (address) &BytecodeCounter::_counter_value);
1826 __ ld(t0, Address(t0));
1827 __ mv(t1, StopInterpreterAt);
1828 __ bne(t0, t1, L);
1829 __ ebreak();
1830 __ bind(L);
1831 __ pop_reg(t0);
1832 }
1833
1834 #endif // !PRODUCT