Thu, 01 Sep 2011 01:31:25 -0700
7079673: JSR 292: C1 should inline bytecoded method handle adapters
Reviewed-by: never
1 /*
2 * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "asm/codeBuffer.hpp"
27 #include "c1/c1_CodeStubs.hpp"
28 #include "c1/c1_Defs.hpp"
29 #include "c1/c1_FrameMap.hpp"
30 #include "c1/c1_LIRAssembler.hpp"
31 #include "c1/c1_MacroAssembler.hpp"
32 #include "c1/c1_Runtime1.hpp"
33 #include "classfile/systemDictionary.hpp"
34 #include "classfile/vmSymbols.hpp"
35 #include "code/codeBlob.hpp"
36 #include "code/compiledIC.hpp"
37 #include "code/pcDesc.hpp"
38 #include "code/scopeDesc.hpp"
39 #include "code/vtableStubs.hpp"
40 #include "compiler/disassembler.hpp"
41 #include "gc_interface/collectedHeap.hpp"
42 #include "interpreter/bytecode.hpp"
43 #include "interpreter/interpreter.hpp"
44 #include "memory/allocation.inline.hpp"
45 #include "memory/barrierSet.hpp"
46 #include "memory/oopFactory.hpp"
47 #include "memory/resourceArea.hpp"
48 #include "oops/objArrayKlass.hpp"
49 #include "oops/oop.inline.hpp"
50 #include "runtime/biasedLocking.hpp"
51 #include "runtime/compilationPolicy.hpp"
52 #include "runtime/interfaceSupport.hpp"
53 #include "runtime/javaCalls.hpp"
54 #include "runtime/sharedRuntime.hpp"
55 #include "runtime/threadCritical.hpp"
56 #include "runtime/vframe.hpp"
57 #include "runtime/vframeArray.hpp"
58 #include "utilities/copy.hpp"
59 #include "utilities/events.hpp"
62 // Implementation of StubAssembler
64 StubAssembler::StubAssembler(CodeBuffer* code, const char * name, int stub_id) : C1_MacroAssembler(code) {
65 _name = name;
66 _must_gc_arguments = false;
67 _frame_size = no_frame_size;
68 _num_rt_args = 0;
69 _stub_id = stub_id;
70 }
73 void StubAssembler::set_info(const char* name, bool must_gc_arguments) {
74 _name = name;
75 _must_gc_arguments = must_gc_arguments;
76 }
79 void StubAssembler::set_frame_size(int size) {
80 if (_frame_size == no_frame_size) {
81 _frame_size = size;
82 }
83 assert(_frame_size == size, "can't change the frame size");
84 }
87 void StubAssembler::set_num_rt_args(int args) {
88 if (_num_rt_args == 0) {
89 _num_rt_args = args;
90 }
91 assert(_num_rt_args == args, "can't change the number of args");
92 }
94 // Implementation of Runtime1
96 CodeBlob* Runtime1::_blobs[Runtime1::number_of_ids];
97 const char *Runtime1::_blob_names[] = {
98 RUNTIME1_STUBS(STUB_NAME, LAST_STUB_NAME)
99 };
101 #ifndef PRODUCT
102 // statistics
103 int Runtime1::_generic_arraycopy_cnt = 0;
104 int Runtime1::_primitive_arraycopy_cnt = 0;
105 int Runtime1::_oop_arraycopy_cnt = 0;
106 int Runtime1::_generic_arraycopystub_cnt = 0;
107 int Runtime1::_arraycopy_slowcase_cnt = 0;
108 int Runtime1::_arraycopy_checkcast_cnt = 0;
109 int Runtime1::_arraycopy_checkcast_attempt_cnt = 0;
110 int Runtime1::_new_type_array_slowcase_cnt = 0;
111 int Runtime1::_new_object_array_slowcase_cnt = 0;
112 int Runtime1::_new_instance_slowcase_cnt = 0;
113 int Runtime1::_new_multi_array_slowcase_cnt = 0;
114 int Runtime1::_monitorenter_slowcase_cnt = 0;
115 int Runtime1::_monitorexit_slowcase_cnt = 0;
116 int Runtime1::_patch_code_slowcase_cnt = 0;
117 int Runtime1::_throw_range_check_exception_count = 0;
118 int Runtime1::_throw_index_exception_count = 0;
119 int Runtime1::_throw_div0_exception_count = 0;
120 int Runtime1::_throw_null_pointer_exception_count = 0;
121 int Runtime1::_throw_class_cast_exception_count = 0;
122 int Runtime1::_throw_incompatible_class_change_error_count = 0;
123 int Runtime1::_throw_array_store_exception_count = 0;
124 int Runtime1::_throw_count = 0;
126 static int _byte_arraycopy_cnt = 0;
127 static int _short_arraycopy_cnt = 0;
128 static int _int_arraycopy_cnt = 0;
129 static int _long_arraycopy_cnt = 0;
130 static int _oop_arraycopy_cnt = 0;
132 address Runtime1::arraycopy_count_address(BasicType type) {
133 switch (type) {
134 case T_BOOLEAN:
135 case T_BYTE: return (address)&_byte_arraycopy_cnt;
136 case T_CHAR:
137 case T_SHORT: return (address)&_short_arraycopy_cnt;
138 case T_FLOAT:
139 case T_INT: return (address)&_int_arraycopy_cnt;
140 case T_DOUBLE:
141 case T_LONG: return (address)&_long_arraycopy_cnt;
142 case T_ARRAY:
143 case T_OBJECT: return (address)&_oop_arraycopy_cnt;
144 default:
145 ShouldNotReachHere();
146 return NULL;
147 }
148 }
151 #endif
153 // Simple helper to see if the caller of a runtime stub which
154 // entered the VM has been deoptimized
156 static bool caller_is_deopted() {
157 JavaThread* thread = JavaThread::current();
158 RegisterMap reg_map(thread, false);
159 frame runtime_frame = thread->last_frame();
160 frame caller_frame = runtime_frame.sender(®_map);
161 assert(caller_frame.is_compiled_frame(), "must be compiled");
162 return caller_frame.is_deoptimized_frame();
163 }
165 // Stress deoptimization
166 static void deopt_caller() {
167 if ( !caller_is_deopted()) {
168 JavaThread* thread = JavaThread::current();
169 RegisterMap reg_map(thread, false);
170 frame runtime_frame = thread->last_frame();
171 frame caller_frame = runtime_frame.sender(®_map);
172 Deoptimization::deoptimize_frame(thread, caller_frame.id());
173 assert(caller_is_deopted(), "Must be deoptimized");
174 }
175 }
178 void Runtime1::generate_blob_for(BufferBlob* buffer_blob, StubID id) {
179 assert(0 <= id && id < number_of_ids, "illegal stub id");
180 ResourceMark rm;
181 // create code buffer for code storage
182 CodeBuffer code(buffer_blob);
184 Compilation::setup_code_buffer(&code, 0);
186 // create assembler for code generation
187 StubAssembler* sasm = new StubAssembler(&code, name_for(id), id);
188 // generate code for runtime stub
189 OopMapSet* oop_maps;
190 oop_maps = generate_code_for(id, sasm);
191 assert(oop_maps == NULL || sasm->frame_size() != no_frame_size,
192 "if stub has an oop map it must have a valid frame size");
194 #ifdef ASSERT
195 // Make sure that stubs that need oopmaps have them
196 switch (id) {
197 // These stubs don't need to have an oopmap
198 case dtrace_object_alloc_id:
199 case g1_pre_barrier_slow_id:
200 case g1_post_barrier_slow_id:
201 case slow_subtype_check_id:
202 case fpu2long_stub_id:
203 case unwind_exception_id:
204 case counter_overflow_id:
205 #if defined(SPARC) || defined(PPC)
206 case handle_exception_nofpu_id: // Unused on sparc
207 #endif
208 break;
210 // All other stubs should have oopmaps
211 default:
212 assert(oop_maps != NULL, "must have an oopmap");
213 }
214 #endif
216 // align so printing shows nop's instead of random code at the end (SimpleStubs are aligned)
217 sasm->align(BytesPerWord);
218 // make sure all code is in code buffer
219 sasm->flush();
220 // create blob - distinguish a few special cases
221 CodeBlob* blob = RuntimeStub::new_runtime_stub(name_for(id),
222 &code,
223 CodeOffsets::frame_never_safe,
224 sasm->frame_size(),
225 oop_maps,
226 sasm->must_gc_arguments());
227 // install blob
228 assert(blob != NULL, "blob must exist");
229 _blobs[id] = blob;
230 }
233 void Runtime1::initialize(BufferBlob* blob) {
234 // platform-dependent initialization
235 initialize_pd();
236 // generate stubs
237 for (int id = 0; id < number_of_ids; id++) generate_blob_for(blob, (StubID)id);
238 // printing
239 #ifndef PRODUCT
240 if (PrintSimpleStubs) {
241 ResourceMark rm;
242 for (int id = 0; id < number_of_ids; id++) {
243 _blobs[id]->print();
244 if (_blobs[id]->oop_maps() != NULL) {
245 _blobs[id]->oop_maps()->print();
246 }
247 }
248 }
249 #endif
250 }
253 CodeBlob* Runtime1::blob_for(StubID id) {
254 assert(0 <= id && id < number_of_ids, "illegal stub id");
255 return _blobs[id];
256 }
259 const char* Runtime1::name_for(StubID id) {
260 assert(0 <= id && id < number_of_ids, "illegal stub id");
261 return _blob_names[id];
262 }
264 const char* Runtime1::name_for_address(address entry) {
265 for (int id = 0; id < number_of_ids; id++) {
266 if (entry == entry_for((StubID)id)) return name_for((StubID)id);
267 }
269 #define FUNCTION_CASE(a, f) \
270 if ((intptr_t)a == CAST_FROM_FN_PTR(intptr_t, f)) return #f
272 FUNCTION_CASE(entry, os::javaTimeMillis);
273 FUNCTION_CASE(entry, os::javaTimeNanos);
274 FUNCTION_CASE(entry, SharedRuntime::OSR_migration_end);
275 FUNCTION_CASE(entry, SharedRuntime::d2f);
276 FUNCTION_CASE(entry, SharedRuntime::d2i);
277 FUNCTION_CASE(entry, SharedRuntime::d2l);
278 FUNCTION_CASE(entry, SharedRuntime::dcos);
279 FUNCTION_CASE(entry, SharedRuntime::dexp);
280 FUNCTION_CASE(entry, SharedRuntime::dlog);
281 FUNCTION_CASE(entry, SharedRuntime::dlog10);
282 FUNCTION_CASE(entry, SharedRuntime::dpow);
283 FUNCTION_CASE(entry, SharedRuntime::drem);
284 FUNCTION_CASE(entry, SharedRuntime::dsin);
285 FUNCTION_CASE(entry, SharedRuntime::dtan);
286 FUNCTION_CASE(entry, SharedRuntime::f2i);
287 FUNCTION_CASE(entry, SharedRuntime::f2l);
288 FUNCTION_CASE(entry, SharedRuntime::frem);
289 FUNCTION_CASE(entry, SharedRuntime::l2d);
290 FUNCTION_CASE(entry, SharedRuntime::l2f);
291 FUNCTION_CASE(entry, SharedRuntime::ldiv);
292 FUNCTION_CASE(entry, SharedRuntime::lmul);
293 FUNCTION_CASE(entry, SharedRuntime::lrem);
294 FUNCTION_CASE(entry, SharedRuntime::lrem);
295 FUNCTION_CASE(entry, SharedRuntime::dtrace_method_entry);
296 FUNCTION_CASE(entry, SharedRuntime::dtrace_method_exit);
297 FUNCTION_CASE(entry, trace_block_entry);
299 #undef FUNCTION_CASE
301 // Soft float adds more runtime names.
302 return pd_name_for_address(entry);
303 }
306 JRT_ENTRY(void, Runtime1::new_instance(JavaThread* thread, klassOopDesc* klass))
307 NOT_PRODUCT(_new_instance_slowcase_cnt++;)
309 assert(oop(klass)->is_klass(), "not a class");
310 instanceKlassHandle h(thread, klass);
311 h->check_valid_for_instantiation(true, CHECK);
312 // make sure klass is initialized
313 h->initialize(CHECK);
314 // allocate instance and return via TLS
315 oop obj = h->allocate_instance(CHECK);
316 thread->set_vm_result(obj);
317 JRT_END
320 JRT_ENTRY(void, Runtime1::new_type_array(JavaThread* thread, klassOopDesc* klass, jint length))
321 NOT_PRODUCT(_new_type_array_slowcase_cnt++;)
322 // Note: no handle for klass needed since they are not used
323 // anymore after new_typeArray() and no GC can happen before.
324 // (This may have to change if this code changes!)
325 assert(oop(klass)->is_klass(), "not a class");
326 BasicType elt_type = typeArrayKlass::cast(klass)->element_type();
327 oop obj = oopFactory::new_typeArray(elt_type, length, CHECK);
328 thread->set_vm_result(obj);
329 // This is pretty rare but this runtime patch is stressful to deoptimization
330 // if we deoptimize here so force a deopt to stress the path.
331 if (DeoptimizeALot) {
332 deopt_caller();
333 }
335 JRT_END
338 JRT_ENTRY(void, Runtime1::new_object_array(JavaThread* thread, klassOopDesc* array_klass, jint length))
339 NOT_PRODUCT(_new_object_array_slowcase_cnt++;)
341 // Note: no handle for klass needed since they are not used
342 // anymore after new_objArray() and no GC can happen before.
343 // (This may have to change if this code changes!)
344 assert(oop(array_klass)->is_klass(), "not a class");
345 klassOop elem_klass = objArrayKlass::cast(array_klass)->element_klass();
346 objArrayOop obj = oopFactory::new_objArray(elem_klass, length, CHECK);
347 thread->set_vm_result(obj);
348 // This is pretty rare but this runtime patch is stressful to deoptimization
349 // if we deoptimize here so force a deopt to stress the path.
350 if (DeoptimizeALot) {
351 deopt_caller();
352 }
353 JRT_END
356 JRT_ENTRY(void, Runtime1::new_multi_array(JavaThread* thread, klassOopDesc* klass, int rank, jint* dims))
357 NOT_PRODUCT(_new_multi_array_slowcase_cnt++;)
359 assert(oop(klass)->is_klass(), "not a class");
360 assert(rank >= 1, "rank must be nonzero");
361 oop obj = arrayKlass::cast(klass)->multi_allocate(rank, dims, CHECK);
362 thread->set_vm_result(obj);
363 JRT_END
366 JRT_ENTRY(void, Runtime1::unimplemented_entry(JavaThread* thread, StubID id))
367 tty->print_cr("Runtime1::entry_for(%d) returned unimplemented entry point", id);
368 JRT_END
371 JRT_ENTRY(void, Runtime1::throw_array_store_exception(JavaThread* thread, oopDesc* obj))
372 ResourceMark rm(thread);
373 const char* klass_name = Klass::cast(obj->klass())->external_name();
374 SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_ArrayStoreException(), klass_name);
375 JRT_END
378 // counter_overflow() is called from within C1-compiled methods. The enclosing method is the method
379 // associated with the top activation record. The inlinee (that is possibly included in the enclosing
380 // method) method oop is passed as an argument. In order to do that it is embedded in the code as
381 // a constant.
382 static nmethod* counter_overflow_helper(JavaThread* THREAD, int branch_bci, methodOopDesc* m) {
383 nmethod* osr_nm = NULL;
384 methodHandle method(THREAD, m);
386 RegisterMap map(THREAD, false);
387 frame fr = THREAD->last_frame().sender(&map);
388 nmethod* nm = (nmethod*) fr.cb();
389 assert(nm!= NULL && nm->is_nmethod(), "Sanity check");
390 methodHandle enclosing_method(THREAD, nm->method());
392 CompLevel level = (CompLevel)nm->comp_level();
393 int bci = InvocationEntryBci;
394 if (branch_bci != InvocationEntryBci) {
395 // Compute desination bci
396 address pc = method()->code_base() + branch_bci;
397 Bytecodes::Code branch = Bytecodes::code_at(method(), pc);
398 int offset = 0;
399 switch (branch) {
400 case Bytecodes::_if_icmplt: case Bytecodes::_iflt:
401 case Bytecodes::_if_icmpgt: case Bytecodes::_ifgt:
402 case Bytecodes::_if_icmple: case Bytecodes::_ifle:
403 case Bytecodes::_if_icmpge: case Bytecodes::_ifge:
404 case Bytecodes::_if_icmpeq: case Bytecodes::_if_acmpeq: case Bytecodes::_ifeq:
405 case Bytecodes::_if_icmpne: case Bytecodes::_if_acmpne: case Bytecodes::_ifne:
406 case Bytecodes::_ifnull: case Bytecodes::_ifnonnull: case Bytecodes::_goto:
407 offset = (int16_t)Bytes::get_Java_u2(pc + 1);
408 break;
409 case Bytecodes::_goto_w:
410 offset = Bytes::get_Java_u4(pc + 1);
411 break;
412 default: ;
413 }
414 bci = branch_bci + offset;
415 }
417 osr_nm = CompilationPolicy::policy()->event(enclosing_method, method, branch_bci, bci, level, nm, THREAD);
418 return osr_nm;
419 }
421 JRT_BLOCK_ENTRY(address, Runtime1::counter_overflow(JavaThread* thread, int bci, methodOopDesc* method))
422 nmethod* osr_nm;
423 JRT_BLOCK
424 osr_nm = counter_overflow_helper(thread, bci, method);
425 if (osr_nm != NULL) {
426 RegisterMap map(thread, false);
427 frame fr = thread->last_frame().sender(&map);
428 Deoptimization::deoptimize_frame(thread, fr.id());
429 }
430 JRT_BLOCK_END
431 return NULL;
432 JRT_END
434 extern void vm_exit(int code);
436 // Enter this method from compiled code handler below. This is where we transition
437 // to VM mode. This is done as a helper routine so that the method called directly
438 // from compiled code does not have to transition to VM. This allows the entry
439 // method to see if the nmethod that we have just looked up a handler for has
440 // been deoptimized while we were in the vm. This simplifies the assembly code
441 // cpu directories.
442 //
443 // We are entering here from exception stub (via the entry method below)
444 // If there is a compiled exception handler in this method, we will continue there;
445 // otherwise we will unwind the stack and continue at the caller of top frame method
446 // Note: we enter in Java using a special JRT wrapper. This wrapper allows us to
447 // control the area where we can allow a safepoint. After we exit the safepoint area we can
448 // check to see if the handler we are going to return is now in a nmethod that has
449 // been deoptimized. If that is the case we return the deopt blob
450 // unpack_with_exception entry instead. This makes life for the exception blob easier
451 // because making that same check and diverting is painful from assembly language.
452 JRT_ENTRY_NO_ASYNC(static address, exception_handler_for_pc_helper(JavaThread* thread, oopDesc* ex, address pc, nmethod*& nm))
453 // Reset method handle flag.
454 thread->set_is_method_handle_return(false);
456 Handle exception(thread, ex);
457 nm = CodeCache::find_nmethod(pc);
458 assert(nm != NULL, "this is not an nmethod");
459 // Adjust the pc as needed/
460 if (nm->is_deopt_pc(pc)) {
461 RegisterMap map(thread, false);
462 frame exception_frame = thread->last_frame().sender(&map);
463 // if the frame isn't deopted then pc must not correspond to the caller of last_frame
464 assert(exception_frame.is_deoptimized_frame(), "must be deopted");
465 pc = exception_frame.pc();
466 }
467 #ifdef ASSERT
468 assert(exception.not_null(), "NULL exceptions should be handled by throw_exception");
469 assert(exception->is_oop(), "just checking");
470 // Check that exception is a subclass of Throwable, otherwise we have a VerifyError
471 if (!(exception->is_a(SystemDictionary::Throwable_klass()))) {
472 if (ExitVMOnVerifyError) vm_exit(-1);
473 ShouldNotReachHere();
474 }
475 #endif
477 // Check the stack guard pages and reenable them if necessary and there is
478 // enough space on the stack to do so. Use fast exceptions only if the guard
479 // pages are enabled.
480 bool guard_pages_enabled = thread->stack_yellow_zone_enabled();
481 if (!guard_pages_enabled) guard_pages_enabled = thread->reguard_stack();
483 if (JvmtiExport::can_post_on_exceptions()) {
484 // To ensure correct notification of exception catches and throws
485 // we have to deoptimize here. If we attempted to notify the
486 // catches and throws during this exception lookup it's possible
487 // we could deoptimize on the way out of the VM and end back in
488 // the interpreter at the throw site. This would result in double
489 // notifications since the interpreter would also notify about
490 // these same catches and throws as it unwound the frame.
492 RegisterMap reg_map(thread);
493 frame stub_frame = thread->last_frame();
494 frame caller_frame = stub_frame.sender(®_map);
496 // We don't really want to deoptimize the nmethod itself since we
497 // can actually continue in the exception handler ourselves but I
498 // don't see an easy way to have the desired effect.
499 Deoptimization::deoptimize_frame(thread, caller_frame.id());
500 assert(caller_is_deopted(), "Must be deoptimized");
502 return SharedRuntime::deopt_blob()->unpack_with_exception_in_tls();
503 }
505 // ExceptionCache is used only for exceptions at call sites and not for implicit exceptions
506 if (guard_pages_enabled) {
507 address fast_continuation = nm->handler_for_exception_and_pc(exception, pc);
508 if (fast_continuation != NULL) {
509 // Set flag if return address is a method handle call site.
510 thread->set_is_method_handle_return(nm->is_method_handle_return(pc));
511 return fast_continuation;
512 }
513 }
515 // If the stack guard pages are enabled, check whether there is a handler in
516 // the current method. Otherwise (guard pages disabled), force an unwind and
517 // skip the exception cache update (i.e., just leave continuation==NULL).
518 address continuation = NULL;
519 if (guard_pages_enabled) {
521 // New exception handling mechanism can support inlined methods
522 // with exception handlers since the mappings are from PC to PC
524 // debugging support
525 // tracing
526 if (TraceExceptions) {
527 ttyLocker ttyl;
528 ResourceMark rm;
529 tty->print_cr("Exception <%s> (0x%x) thrown in compiled method <%s> at PC " PTR_FORMAT " for thread 0x%x",
530 exception->print_value_string(), (address)exception(), nm->method()->print_value_string(), pc, thread);
531 }
532 // for AbortVMOnException flag
533 NOT_PRODUCT(Exceptions::debug_check_abort(exception));
535 // Clear out the exception oop and pc since looking up an
536 // exception handler can cause class loading, which might throw an
537 // exception and those fields are expected to be clear during
538 // normal bytecode execution.
539 thread->set_exception_oop(NULL);
540 thread->set_exception_pc(NULL);
542 continuation = SharedRuntime::compute_compiled_exc_handler(nm, pc, exception, false, false);
543 // If an exception was thrown during exception dispatch, the exception oop may have changed
544 thread->set_exception_oop(exception());
545 thread->set_exception_pc(pc);
547 // the exception cache is used only by non-implicit exceptions
548 if (continuation != NULL) {
549 nm->add_handler_for_exception_and_pc(exception, pc, continuation);
550 }
551 }
553 thread->set_vm_result(exception());
554 // Set flag if return address is a method handle call site.
555 thread->set_is_method_handle_return(nm->is_method_handle_return(pc));
557 if (TraceExceptions) {
558 ttyLocker ttyl;
559 ResourceMark rm;
560 tty->print_cr("Thread " PTR_FORMAT " continuing at PC " PTR_FORMAT " for exception thrown at PC " PTR_FORMAT,
561 thread, continuation, pc);
562 }
564 return continuation;
565 JRT_END
567 // Enter this method from compiled code only if there is a Java exception handler
568 // in the method handling the exception.
569 // We are entering here from exception stub. We don't do a normal VM transition here.
570 // We do it in a helper. This is so we can check to see if the nmethod we have just
571 // searched for an exception handler has been deoptimized in the meantime.
572 address Runtime1::exception_handler_for_pc(JavaThread* thread) {
573 oop exception = thread->exception_oop();
574 address pc = thread->exception_pc();
575 // Still in Java mode
576 DEBUG_ONLY(ResetNoHandleMark rnhm);
577 nmethod* nm = NULL;
578 address continuation = NULL;
579 {
580 // Enter VM mode by calling the helper
581 ResetNoHandleMark rnhm;
582 continuation = exception_handler_for_pc_helper(thread, exception, pc, nm);
583 }
584 // Back in JAVA, use no oops DON'T safepoint
586 // Now check to see if the nmethod we were called from is now deoptimized.
587 // If so we must return to the deopt blob and deoptimize the nmethod
588 if (nm != NULL && caller_is_deopted()) {
589 continuation = SharedRuntime::deopt_blob()->unpack_with_exception_in_tls();
590 }
592 assert(continuation != NULL, "no handler found");
593 return continuation;
594 }
597 JRT_ENTRY(void, Runtime1::throw_range_check_exception(JavaThread* thread, int index))
598 NOT_PRODUCT(_throw_range_check_exception_count++;)
599 Events::log("throw_range_check");
600 char message[jintAsStringSize];
601 sprintf(message, "%d", index);
602 SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_ArrayIndexOutOfBoundsException(), message);
603 JRT_END
606 JRT_ENTRY(void, Runtime1::throw_index_exception(JavaThread* thread, int index))
607 NOT_PRODUCT(_throw_index_exception_count++;)
608 Events::log("throw_index");
609 char message[16];
610 sprintf(message, "%d", index);
611 SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_IndexOutOfBoundsException(), message);
612 JRT_END
615 JRT_ENTRY(void, Runtime1::throw_div0_exception(JavaThread* thread))
616 NOT_PRODUCT(_throw_div0_exception_count++;)
617 SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_ArithmeticException(), "/ by zero");
618 JRT_END
621 JRT_ENTRY(void, Runtime1::throw_null_pointer_exception(JavaThread* thread))
622 NOT_PRODUCT(_throw_null_pointer_exception_count++;)
623 SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_NullPointerException());
624 JRT_END
627 JRT_ENTRY(void, Runtime1::throw_class_cast_exception(JavaThread* thread, oopDesc* object))
628 NOT_PRODUCT(_throw_class_cast_exception_count++;)
629 ResourceMark rm(thread);
630 char* message = SharedRuntime::generate_class_cast_message(
631 thread, Klass::cast(object->klass())->external_name());
632 SharedRuntime::throw_and_post_jvmti_exception(
633 thread, vmSymbols::java_lang_ClassCastException(), message);
634 JRT_END
637 JRT_ENTRY(void, Runtime1::throw_incompatible_class_change_error(JavaThread* thread))
638 NOT_PRODUCT(_throw_incompatible_class_change_error_count++;)
639 ResourceMark rm(thread);
640 SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_IncompatibleClassChangeError());
641 JRT_END
644 JRT_ENTRY_NO_ASYNC(void, Runtime1::monitorenter(JavaThread* thread, oopDesc* obj, BasicObjectLock* lock))
645 NOT_PRODUCT(_monitorenter_slowcase_cnt++;)
646 if (PrintBiasedLockingStatistics) {
647 Atomic::inc(BiasedLocking::slow_path_entry_count_addr());
648 }
649 Handle h_obj(thread, obj);
650 assert(h_obj()->is_oop(), "must be NULL or an object");
651 if (UseBiasedLocking) {
652 // Retry fast entry if bias is revoked to avoid unnecessary inflation
653 ObjectSynchronizer::fast_enter(h_obj, lock->lock(), true, CHECK);
654 } else {
655 if (UseFastLocking) {
656 // When using fast locking, the compiled code has already tried the fast case
657 assert(obj == lock->obj(), "must match");
658 ObjectSynchronizer::slow_enter(h_obj, lock->lock(), THREAD);
659 } else {
660 lock->set_obj(obj);
661 ObjectSynchronizer::fast_enter(h_obj, lock->lock(), false, THREAD);
662 }
663 }
664 JRT_END
667 JRT_LEAF(void, Runtime1::monitorexit(JavaThread* thread, BasicObjectLock* lock))
668 NOT_PRODUCT(_monitorexit_slowcase_cnt++;)
669 assert(thread == JavaThread::current(), "threads must correspond");
670 assert(thread->last_Java_sp(), "last_Java_sp must be set");
671 // monitorexit is non-blocking (leaf routine) => no exceptions can be thrown
672 EXCEPTION_MARK;
674 oop obj = lock->obj();
675 assert(obj->is_oop(), "must be NULL or an object");
676 if (UseFastLocking) {
677 // When using fast locking, the compiled code has already tried the fast case
678 ObjectSynchronizer::slow_exit(obj, lock->lock(), THREAD);
679 } else {
680 ObjectSynchronizer::fast_exit(obj, lock->lock(), THREAD);
681 }
682 JRT_END
685 static klassOop resolve_field_return_klass(methodHandle caller, int bci, TRAPS) {
686 Bytecode_field field_access(caller, bci);
687 // This can be static or non-static field access
688 Bytecodes::Code code = field_access.code();
690 // We must load class, initialize class and resolvethe field
691 FieldAccessInfo result; // initialize class if needed
692 constantPoolHandle constants(THREAD, caller->constants());
693 LinkResolver::resolve_field(result, constants, field_access.index(), Bytecodes::java_code(code), false, CHECK_NULL);
694 return result.klass()();
695 }
698 //
699 // This routine patches sites where a class wasn't loaded or
700 // initialized at the time the code was generated. It handles
701 // references to classes, fields and forcing of initialization. Most
702 // of the cases are straightforward and involving simply forcing
703 // resolution of a class, rewriting the instruction stream with the
704 // needed constant and replacing the call in this function with the
705 // patched code. The case for static field is more complicated since
706 // the thread which is in the process of initializing a class can
707 // access it's static fields but other threads can't so the code
708 // either has to deoptimize when this case is detected or execute a
709 // check that the current thread is the initializing thread. The
710 // current
711 //
712 // Patches basically look like this:
713 //
714 //
715 // patch_site: jmp patch stub ;; will be patched
716 // continue: ...
717 // ...
718 // ...
719 // ...
720 //
721 // They have a stub which looks like this:
722 //
723 // ;; patch body
724 // movl <const>, reg (for class constants)
725 // <or> movl [reg1 + <const>], reg (for field offsets)
726 // <or> movl reg, [reg1 + <const>] (for field offsets)
727 // <being_init offset> <bytes to copy> <bytes to skip>
728 // patch_stub: call Runtime1::patch_code (through a runtime stub)
729 // jmp patch_site
730 //
731 //
732 // A normal patch is done by rewriting the patch body, usually a move,
733 // and then copying it into place over top of the jmp instruction
734 // being careful to flush caches and doing it in an MP-safe way. The
735 // constants following the patch body are used to find various pieces
736 // of the patch relative to the call site for Runtime1::patch_code.
737 // The case for getstatic and putstatic is more complicated because
738 // getstatic and putstatic have special semantics when executing while
739 // the class is being initialized. getstatic/putstatic on a class
740 // which is being_initialized may be executed by the initializing
741 // thread but other threads have to block when they execute it. This
742 // is accomplished in compiled code by executing a test of the current
743 // thread against the initializing thread of the class. It's emitted
744 // as boilerplate in their stub which allows the patched code to be
745 // executed before it's copied back into the main body of the nmethod.
746 //
747 // being_init: get_thread(<tmp reg>
748 // cmpl [reg1 + <init_thread_offset>], <tmp reg>
749 // jne patch_stub
750 // movl [reg1 + <const>], reg (for field offsets) <or>
751 // movl reg, [reg1 + <const>] (for field offsets)
752 // jmp continue
753 // <being_init offset> <bytes to copy> <bytes to skip>
754 // patch_stub: jmp Runtim1::patch_code (through a runtime stub)
755 // jmp patch_site
756 //
757 // If the class is being initialized the patch body is rewritten and
758 // the patch site is rewritten to jump to being_init, instead of
759 // patch_stub. Whenever this code is executed it checks the current
760 // thread against the intializing thread so other threads will enter
761 // the runtime and end up blocked waiting the class to finish
762 // initializing inside the calls to resolve_field below. The
763 // initializing class will continue on it's way. Once the class is
764 // fully_initialized, the intializing_thread of the class becomes
765 // NULL, so the next thread to execute this code will fail the test,
766 // call into patch_code and complete the patching process by copying
767 // the patch body back into the main part of the nmethod and resume
768 // executing.
769 //
770 //
772 JRT_ENTRY(void, Runtime1::patch_code(JavaThread* thread, Runtime1::StubID stub_id ))
773 NOT_PRODUCT(_patch_code_slowcase_cnt++;)
775 ResourceMark rm(thread);
776 RegisterMap reg_map(thread, false);
777 frame runtime_frame = thread->last_frame();
778 frame caller_frame = runtime_frame.sender(®_map);
780 // last java frame on stack
781 vframeStream vfst(thread, true);
782 assert(!vfst.at_end(), "Java frame must exist");
784 methodHandle caller_method(THREAD, vfst.method());
785 // Note that caller_method->code() may not be same as caller_code because of OSR's
786 // Note also that in the presence of inlining it is not guaranteed
787 // that caller_method() == caller_code->method()
790 int bci = vfst.bci();
792 Events::log("patch_code @ " INTPTR_FORMAT , caller_frame.pc());
794 Bytecodes::Code code = caller_method()->java_code_at(bci);
796 #ifndef PRODUCT
797 // this is used by assertions in the access_field_patching_id
798 BasicType patch_field_type = T_ILLEGAL;
799 #endif // PRODUCT
800 bool deoptimize_for_volatile = false;
801 int patch_field_offset = -1;
802 KlassHandle init_klass(THREAD, klassOop(NULL)); // klass needed by access_field_patching code
803 Handle load_klass(THREAD, NULL); // oop needed by load_klass_patching code
804 if (stub_id == Runtime1::access_field_patching_id) {
806 Bytecode_field field_access(caller_method, bci);
807 FieldAccessInfo result; // initialize class if needed
808 Bytecodes::Code code = field_access.code();
809 constantPoolHandle constants(THREAD, caller_method->constants());
810 LinkResolver::resolve_field(result, constants, field_access.index(), Bytecodes::java_code(code), false, CHECK);
811 patch_field_offset = result.field_offset();
813 // If we're patching a field which is volatile then at compile it
814 // must not have been know to be volatile, so the generated code
815 // isn't correct for a volatile reference. The nmethod has to be
816 // deoptimized so that the code can be regenerated correctly.
817 // This check is only needed for access_field_patching since this
818 // is the path for patching field offsets. load_klass is only
819 // used for patching references to oops which don't need special
820 // handling in the volatile case.
821 deoptimize_for_volatile = result.access_flags().is_volatile();
823 #ifndef PRODUCT
824 patch_field_type = result.field_type();
825 #endif
826 } else if (stub_id == Runtime1::load_klass_patching_id) {
827 oop k;
828 switch (code) {
829 case Bytecodes::_putstatic:
830 case Bytecodes::_getstatic:
831 { klassOop klass = resolve_field_return_klass(caller_method, bci, CHECK);
832 // Save a reference to the class that has to be checked for initialization
833 init_klass = KlassHandle(THREAD, klass);
834 k = klass->java_mirror();
835 }
836 break;
837 case Bytecodes::_new:
838 { Bytecode_new bnew(caller_method(), caller_method->bcp_from(bci));
839 k = caller_method->constants()->klass_at(bnew.index(), CHECK);
840 }
841 break;
842 case Bytecodes::_multianewarray:
843 { Bytecode_multianewarray mna(caller_method(), caller_method->bcp_from(bci));
844 k = caller_method->constants()->klass_at(mna.index(), CHECK);
845 }
846 break;
847 case Bytecodes::_instanceof:
848 { Bytecode_instanceof io(caller_method(), caller_method->bcp_from(bci));
849 k = caller_method->constants()->klass_at(io.index(), CHECK);
850 }
851 break;
852 case Bytecodes::_checkcast:
853 { Bytecode_checkcast cc(caller_method(), caller_method->bcp_from(bci));
854 k = caller_method->constants()->klass_at(cc.index(), CHECK);
855 }
856 break;
857 case Bytecodes::_anewarray:
858 { Bytecode_anewarray anew(caller_method(), caller_method->bcp_from(bci));
859 klassOop ek = caller_method->constants()->klass_at(anew.index(), CHECK);
860 k = Klass::cast(ek)->array_klass(CHECK);
861 }
862 break;
863 case Bytecodes::_ldc:
864 case Bytecodes::_ldc_w:
865 {
866 Bytecode_loadconstant cc(caller_method, bci);
867 k = cc.resolve_constant(CHECK);
868 assert(k != NULL && !k->is_klass(), "must be class mirror or other Java constant");
869 }
870 break;
871 default: Unimplemented();
872 }
873 // convert to handle
874 load_klass = Handle(THREAD, k);
875 } else {
876 ShouldNotReachHere();
877 }
879 if (deoptimize_for_volatile) {
880 // At compile time we assumed the field wasn't volatile but after
881 // loading it turns out it was volatile so we have to throw the
882 // compiled code out and let it be regenerated.
883 if (TracePatching) {
884 tty->print_cr("Deoptimizing for patching volatile field reference");
885 }
886 // It's possible the nmethod was invalidated in the last
887 // safepoint, but if it's still alive then make it not_entrant.
888 nmethod* nm = CodeCache::find_nmethod(caller_frame.pc());
889 if (nm != NULL) {
890 nm->make_not_entrant();
891 }
893 Deoptimization::deoptimize_frame(thread, caller_frame.id());
895 // Return to the now deoptimized frame.
896 }
898 // If we are patching in a non-perm oop, make sure the nmethod
899 // is on the right list.
900 if (ScavengeRootsInCode && load_klass.not_null() && load_klass->is_scavengable()) {
901 MutexLockerEx ml_code (CodeCache_lock, Mutex::_no_safepoint_check_flag);
902 nmethod* nm = CodeCache::find_nmethod(caller_frame.pc());
903 guarantee(nm != NULL, "only nmethods can contain non-perm oops");
904 if (!nm->on_scavenge_root_list())
905 CodeCache::add_scavenge_root_nmethod(nm);
906 }
908 // Now copy code back
910 {
911 MutexLockerEx ml_patch (Patching_lock, Mutex::_no_safepoint_check_flag);
912 //
913 // Deoptimization may have happened while we waited for the lock.
914 // In that case we don't bother to do any patching we just return
915 // and let the deopt happen
916 if (!caller_is_deopted()) {
917 NativeGeneralJump* jump = nativeGeneralJump_at(caller_frame.pc());
918 address instr_pc = jump->jump_destination();
919 NativeInstruction* ni = nativeInstruction_at(instr_pc);
920 if (ni->is_jump() ) {
921 // the jump has not been patched yet
922 // The jump destination is slow case and therefore not part of the stubs
923 // (stubs are only for StaticCalls)
925 // format of buffer
926 // ....
927 // instr byte 0 <-- copy_buff
928 // instr byte 1
929 // ..
930 // instr byte n-1
931 // n
932 // .... <-- call destination
934 address stub_location = caller_frame.pc() + PatchingStub::patch_info_offset();
935 unsigned char* byte_count = (unsigned char*) (stub_location - 1);
936 unsigned char* byte_skip = (unsigned char*) (stub_location - 2);
937 unsigned char* being_initialized_entry_offset = (unsigned char*) (stub_location - 3);
938 address copy_buff = stub_location - *byte_skip - *byte_count;
939 address being_initialized_entry = stub_location - *being_initialized_entry_offset;
940 if (TracePatching) {
941 tty->print_cr(" Patching %s at bci %d at address 0x%x (%s)", Bytecodes::name(code), bci,
942 instr_pc, (stub_id == Runtime1::access_field_patching_id) ? "field" : "klass");
943 nmethod* caller_code = CodeCache::find_nmethod(caller_frame.pc());
944 assert(caller_code != NULL, "nmethod not found");
946 // NOTE we use pc() not original_pc() because we already know they are
947 // identical otherwise we'd have never entered this block of code
949 OopMap* map = caller_code->oop_map_for_return_address(caller_frame.pc());
950 assert(map != NULL, "null check");
951 map->print();
952 tty->cr();
954 Disassembler::decode(copy_buff, copy_buff + *byte_count, tty);
955 }
956 // depending on the code below, do_patch says whether to copy the patch body back into the nmethod
957 bool do_patch = true;
958 if (stub_id == Runtime1::access_field_patching_id) {
959 // The offset may not be correct if the class was not loaded at code generation time.
960 // Set it now.
961 NativeMovRegMem* n_move = nativeMovRegMem_at(copy_buff);
962 assert(n_move->offset() == 0 || (n_move->offset() == 4 && (patch_field_type == T_DOUBLE || patch_field_type == T_LONG)), "illegal offset for type");
963 assert(patch_field_offset >= 0, "illegal offset");
964 n_move->add_offset_in_bytes(patch_field_offset);
965 } else if (stub_id == Runtime1::load_klass_patching_id) {
966 // If a getstatic or putstatic is referencing a klass which
967 // isn't fully initialized, the patch body isn't copied into
968 // place until initialization is complete. In this case the
969 // patch site is setup so that any threads besides the
970 // initializing thread are forced to come into the VM and
971 // block.
972 do_patch = (code != Bytecodes::_getstatic && code != Bytecodes::_putstatic) ||
973 instanceKlass::cast(init_klass())->is_initialized();
974 NativeGeneralJump* jump = nativeGeneralJump_at(instr_pc);
975 if (jump->jump_destination() == being_initialized_entry) {
976 assert(do_patch == true, "initialization must be complete at this point");
977 } else {
978 // patch the instruction <move reg, klass>
979 NativeMovConstReg* n_copy = nativeMovConstReg_at(copy_buff);
981 assert(n_copy->data() == 0 ||
982 n_copy->data() == (intptr_t)Universe::non_oop_word(),
983 "illegal init value");
984 assert(load_klass() != NULL, "klass not set");
985 n_copy->set_data((intx) (load_klass()));
987 if (TracePatching) {
988 Disassembler::decode(copy_buff, copy_buff + *byte_count, tty);
989 }
991 #if defined(SPARC) || defined(PPC)
992 // Update the oop location in the nmethod with the proper
993 // oop. When the code was generated, a NULL was stuffed
994 // in the oop table and that table needs to be update to
995 // have the right value. On intel the value is kept
996 // directly in the instruction instead of in the oop
997 // table, so set_data above effectively updated the value.
998 nmethod* nm = CodeCache::find_nmethod(instr_pc);
999 assert(nm != NULL, "invalid nmethod_pc");
1000 RelocIterator oops(nm, copy_buff, copy_buff + 1);
1001 bool found = false;
1002 while (oops.next() && !found) {
1003 if (oops.type() == relocInfo::oop_type) {
1004 oop_Relocation* r = oops.oop_reloc();
1005 oop* oop_adr = r->oop_addr();
1006 *oop_adr = load_klass();
1007 r->fix_oop_relocation();
1008 found = true;
1009 }
1010 }
1011 assert(found, "the oop must exist!");
1012 #endif
1014 }
1015 } else {
1016 ShouldNotReachHere();
1017 }
1018 if (do_patch) {
1019 // replace instructions
1020 // first replace the tail, then the call
1021 #ifdef ARM
1022 if(stub_id == Runtime1::load_klass_patching_id && !VM_Version::supports_movw()) {
1023 nmethod* nm = CodeCache::find_nmethod(instr_pc);
1024 oop* oop_addr = NULL;
1025 assert(nm != NULL, "invalid nmethod_pc");
1026 RelocIterator oops(nm, copy_buff, copy_buff + 1);
1027 while (oops.next()) {
1028 if (oops.type() == relocInfo::oop_type) {
1029 oop_Relocation* r = oops.oop_reloc();
1030 oop_addr = r->oop_addr();
1031 break;
1032 }
1033 }
1034 assert(oop_addr != NULL, "oop relocation must exist");
1035 copy_buff -= *byte_count;
1036 NativeMovConstReg* n_copy2 = nativeMovConstReg_at(copy_buff);
1037 n_copy2->set_pc_relative_offset((address)oop_addr, instr_pc);
1038 }
1039 #endif
1041 for (int i = NativeCall::instruction_size; i < *byte_count; i++) {
1042 address ptr = copy_buff + i;
1043 int a_byte = (*ptr) & 0xFF;
1044 address dst = instr_pc + i;
1045 *(unsigned char*)dst = (unsigned char) a_byte;
1046 }
1047 ICache::invalidate_range(instr_pc, *byte_count);
1048 NativeGeneralJump::replace_mt_safe(instr_pc, copy_buff);
1050 if (stub_id == Runtime1::load_klass_patching_id) {
1051 // update relocInfo to oop
1052 nmethod* nm = CodeCache::find_nmethod(instr_pc);
1053 assert(nm != NULL, "invalid nmethod_pc");
1055 // The old patch site is now a move instruction so update
1056 // the reloc info so that it will get updated during
1057 // future GCs.
1058 RelocIterator iter(nm, (address)instr_pc, (address)(instr_pc + 1));
1059 relocInfo::change_reloc_info_for_address(&iter, (address) instr_pc,
1060 relocInfo::none, relocInfo::oop_type);
1061 #ifdef SPARC
1062 // Sparc takes two relocations for an oop so update the second one.
1063 address instr_pc2 = instr_pc + NativeMovConstReg::add_offset;
1064 RelocIterator iter2(nm, instr_pc2, instr_pc2 + 1);
1065 relocInfo::change_reloc_info_for_address(&iter2, (address) instr_pc2,
1066 relocInfo::none, relocInfo::oop_type);
1067 #endif
1068 #ifdef PPC
1069 { address instr_pc2 = instr_pc + NativeMovConstReg::lo_offset;
1070 RelocIterator iter2(nm, instr_pc2, instr_pc2 + 1);
1071 relocInfo::change_reloc_info_for_address(&iter2, (address) instr_pc2, relocInfo::none, relocInfo::oop_type);
1072 }
1073 #endif
1074 }
1076 } else {
1077 ICache::invalidate_range(copy_buff, *byte_count);
1078 NativeGeneralJump::insert_unconditional(instr_pc, being_initialized_entry);
1079 }
1080 }
1081 }
1082 }
1083 JRT_END
1085 //
1086 // Entry point for compiled code. We want to patch a nmethod.
1087 // We don't do a normal VM transition here because we want to
1088 // know after the patching is complete and any safepoint(s) are taken
1089 // if the calling nmethod was deoptimized. We do this by calling a
1090 // helper method which does the normal VM transition and when it
1091 // completes we can check for deoptimization. This simplifies the
1092 // assembly code in the cpu directories.
1093 //
1094 int Runtime1::move_klass_patching(JavaThread* thread) {
1095 //
1096 // NOTE: we are still in Java
1097 //
1098 Thread* THREAD = thread;
1099 debug_only(NoHandleMark nhm;)
1100 {
1101 // Enter VM mode
1103 ResetNoHandleMark rnhm;
1104 patch_code(thread, load_klass_patching_id);
1105 }
1106 // Back in JAVA, use no oops DON'T safepoint
1108 // Return true if calling code is deoptimized
1110 return caller_is_deopted();
1111 }
1113 //
1114 // Entry point for compiled code. We want to patch a nmethod.
1115 // We don't do a normal VM transition here because we want to
1116 // know after the patching is complete and any safepoint(s) are taken
1117 // if the calling nmethod was deoptimized. We do this by calling a
1118 // helper method which does the normal VM transition and when it
1119 // completes we can check for deoptimization. This simplifies the
1120 // assembly code in the cpu directories.
1121 //
1123 int Runtime1::access_field_patching(JavaThread* thread) {
1124 //
1125 // NOTE: we are still in Java
1126 //
1127 Thread* THREAD = thread;
1128 debug_only(NoHandleMark nhm;)
1129 {
1130 // Enter VM mode
1132 ResetNoHandleMark rnhm;
1133 patch_code(thread, access_field_patching_id);
1134 }
1135 // Back in JAVA, use no oops DON'T safepoint
1137 // Return true if calling code is deoptimized
1139 return caller_is_deopted();
1140 JRT_END
1143 JRT_LEAF(void, Runtime1::trace_block_entry(jint block_id))
1144 // for now we just print out the block id
1145 tty->print("%d ", block_id);
1146 JRT_END
1149 // Array copy return codes.
1150 enum {
1151 ac_failed = -1, // arraycopy failed
1152 ac_ok = 0 // arraycopy succeeded
1153 };
1156 // Below length is the # elements copied.
1157 template <class T> int obj_arraycopy_work(oopDesc* src, T* src_addr,
1158 oopDesc* dst, T* dst_addr,
1159 int length) {
1161 // For performance reasons, we assume we are using a card marking write
1162 // barrier. The assert will fail if this is not the case.
1163 // Note that we use the non-virtual inlineable variant of write_ref_array.
1164 BarrierSet* bs = Universe::heap()->barrier_set();
1165 assert(bs->has_write_ref_array_opt(), "Barrier set must have ref array opt");
1166 assert(bs->has_write_ref_array_pre_opt(), "For pre-barrier as well.");
1167 if (src == dst) {
1168 // same object, no check
1169 bs->write_ref_array_pre(dst_addr, length);
1170 Copy::conjoint_oops_atomic(src_addr, dst_addr, length);
1171 bs->write_ref_array((HeapWord*)dst_addr, length);
1172 return ac_ok;
1173 } else {
1174 klassOop bound = objArrayKlass::cast(dst->klass())->element_klass();
1175 klassOop stype = objArrayKlass::cast(src->klass())->element_klass();
1176 if (stype == bound || Klass::cast(stype)->is_subtype_of(bound)) {
1177 // Elements are guaranteed to be subtypes, so no check necessary
1178 bs->write_ref_array_pre(dst_addr, length);
1179 Copy::conjoint_oops_atomic(src_addr, dst_addr, length);
1180 bs->write_ref_array((HeapWord*)dst_addr, length);
1181 return ac_ok;
1182 }
1183 }
1184 return ac_failed;
1185 }
1187 // fast and direct copy of arrays; returning -1, means that an exception may be thrown
1188 // and we did not copy anything
1189 JRT_LEAF(int, Runtime1::arraycopy(oopDesc* src, int src_pos, oopDesc* dst, int dst_pos, int length))
1190 #ifndef PRODUCT
1191 _generic_arraycopy_cnt++; // Slow-path oop array copy
1192 #endif
1194 if (src == NULL || dst == NULL || src_pos < 0 || dst_pos < 0 || length < 0) return ac_failed;
1195 if (!dst->is_array() || !src->is_array()) return ac_failed;
1196 if ((unsigned int) arrayOop(src)->length() < (unsigned int)src_pos + (unsigned int)length) return ac_failed;
1197 if ((unsigned int) arrayOop(dst)->length() < (unsigned int)dst_pos + (unsigned int)length) return ac_failed;
1199 if (length == 0) return ac_ok;
1200 if (src->is_typeArray()) {
1201 const klassOop klass_oop = src->klass();
1202 if (klass_oop != dst->klass()) return ac_failed;
1203 typeArrayKlass* klass = typeArrayKlass::cast(klass_oop);
1204 const int l2es = klass->log2_element_size();
1205 const int ihs = klass->array_header_in_bytes() / wordSize;
1206 char* src_addr = (char*) ((oopDesc**)src + ihs) + (src_pos << l2es);
1207 char* dst_addr = (char*) ((oopDesc**)dst + ihs) + (dst_pos << l2es);
1208 // Potential problem: memmove is not guaranteed to be word atomic
1209 // Revisit in Merlin
1210 memmove(dst_addr, src_addr, length << l2es);
1211 return ac_ok;
1212 } else if (src->is_objArray() && dst->is_objArray()) {
1213 if (UseCompressedOops) {
1214 narrowOop *src_addr = objArrayOop(src)->obj_at_addr<narrowOop>(src_pos);
1215 narrowOop *dst_addr = objArrayOop(dst)->obj_at_addr<narrowOop>(dst_pos);
1216 return obj_arraycopy_work(src, src_addr, dst, dst_addr, length);
1217 } else {
1218 oop *src_addr = objArrayOop(src)->obj_at_addr<oop>(src_pos);
1219 oop *dst_addr = objArrayOop(dst)->obj_at_addr<oop>(dst_pos);
1220 return obj_arraycopy_work(src, src_addr, dst, dst_addr, length);
1221 }
1222 }
1223 return ac_failed;
1224 JRT_END
1227 JRT_LEAF(void, Runtime1::primitive_arraycopy(HeapWord* src, HeapWord* dst, int length))
1228 #ifndef PRODUCT
1229 _primitive_arraycopy_cnt++;
1230 #endif
1232 if (length == 0) return;
1233 // Not guaranteed to be word atomic, but that doesn't matter
1234 // for anything but an oop array, which is covered by oop_arraycopy.
1235 Copy::conjoint_jbytes(src, dst, length);
1236 JRT_END
1238 JRT_LEAF(void, Runtime1::oop_arraycopy(HeapWord* src, HeapWord* dst, int num))
1239 #ifndef PRODUCT
1240 _oop_arraycopy_cnt++;
1241 #endif
1243 if (num == 0) return;
1244 BarrierSet* bs = Universe::heap()->barrier_set();
1245 assert(bs->has_write_ref_array_opt(), "Barrier set must have ref array opt");
1246 assert(bs->has_write_ref_array_pre_opt(), "For pre-barrier as well.");
1247 if (UseCompressedOops) {
1248 bs->write_ref_array_pre((narrowOop*)dst, num);
1249 Copy::conjoint_oops_atomic((narrowOop*) src, (narrowOop*) dst, num);
1250 } else {
1251 bs->write_ref_array_pre((oop*)dst, num);
1252 Copy::conjoint_oops_atomic((oop*) src, (oop*) dst, num);
1253 }
1254 bs->write_ref_array(dst, num);
1255 JRT_END
1258 #ifndef PRODUCT
1259 void Runtime1::print_statistics() {
1260 tty->print_cr("C1 Runtime statistics:");
1261 tty->print_cr(" _resolve_invoke_virtual_cnt: %d", SharedRuntime::_resolve_virtual_ctr);
1262 tty->print_cr(" _resolve_invoke_opt_virtual_cnt: %d", SharedRuntime::_resolve_opt_virtual_ctr);
1263 tty->print_cr(" _resolve_invoke_static_cnt: %d", SharedRuntime::_resolve_static_ctr);
1264 tty->print_cr(" _handle_wrong_method_cnt: %d", SharedRuntime::_wrong_method_ctr);
1265 tty->print_cr(" _ic_miss_cnt: %d", SharedRuntime::_ic_miss_ctr);
1266 tty->print_cr(" _generic_arraycopy_cnt: %d", _generic_arraycopy_cnt);
1267 tty->print_cr(" _generic_arraycopystub_cnt: %d", _generic_arraycopystub_cnt);
1268 tty->print_cr(" _byte_arraycopy_cnt: %d", _byte_arraycopy_cnt);
1269 tty->print_cr(" _short_arraycopy_cnt: %d", _short_arraycopy_cnt);
1270 tty->print_cr(" _int_arraycopy_cnt: %d", _int_arraycopy_cnt);
1271 tty->print_cr(" _long_arraycopy_cnt: %d", _long_arraycopy_cnt);
1272 tty->print_cr(" _primitive_arraycopy_cnt: %d", _primitive_arraycopy_cnt);
1273 tty->print_cr(" _oop_arraycopy_cnt (C): %d", Runtime1::_oop_arraycopy_cnt);
1274 tty->print_cr(" _oop_arraycopy_cnt (stub): %d", _oop_arraycopy_cnt);
1275 tty->print_cr(" _arraycopy_slowcase_cnt: %d", _arraycopy_slowcase_cnt);
1276 tty->print_cr(" _arraycopy_checkcast_cnt: %d", _arraycopy_checkcast_cnt);
1277 tty->print_cr(" _arraycopy_checkcast_attempt_cnt:%d", _arraycopy_checkcast_attempt_cnt);
1279 tty->print_cr(" _new_type_array_slowcase_cnt: %d", _new_type_array_slowcase_cnt);
1280 tty->print_cr(" _new_object_array_slowcase_cnt: %d", _new_object_array_slowcase_cnt);
1281 tty->print_cr(" _new_instance_slowcase_cnt: %d", _new_instance_slowcase_cnt);
1282 tty->print_cr(" _new_multi_array_slowcase_cnt: %d", _new_multi_array_slowcase_cnt);
1283 tty->print_cr(" _monitorenter_slowcase_cnt: %d", _monitorenter_slowcase_cnt);
1284 tty->print_cr(" _monitorexit_slowcase_cnt: %d", _monitorexit_slowcase_cnt);
1285 tty->print_cr(" _patch_code_slowcase_cnt: %d", _patch_code_slowcase_cnt);
1287 tty->print_cr(" _throw_range_check_exception_count: %d:", _throw_range_check_exception_count);
1288 tty->print_cr(" _throw_index_exception_count: %d:", _throw_index_exception_count);
1289 tty->print_cr(" _throw_div0_exception_count: %d:", _throw_div0_exception_count);
1290 tty->print_cr(" _throw_null_pointer_exception_count: %d:", _throw_null_pointer_exception_count);
1291 tty->print_cr(" _throw_class_cast_exception_count: %d:", _throw_class_cast_exception_count);
1292 tty->print_cr(" _throw_incompatible_class_change_error_count: %d:", _throw_incompatible_class_change_error_count);
1293 tty->print_cr(" _throw_array_store_exception_count: %d:", _throw_array_store_exception_count);
1294 tty->print_cr(" _throw_count: %d:", _throw_count);
1296 SharedRuntime::print_ic_miss_histogram();
1297 tty->cr();
1298 }
1299 #endif // PRODUCT