Mercurial > hg > truffle
annotate src/share/vm/c1/c1_Runtime1.cpp @ 113:ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
Summary: Compressed oops in instances, arrays, and headers. Code contributors are coleenp, phh, never, swamyv
Reviewed-by: jmasa, kamg, acorn, tbell, kvn, rasbold
| author | coleenp |
|---|---|
| date | Sun, 13 Apr 2008 17:43:42 -0400 |
| parents | a61af66fc99e |
| children | d1605aabd0a1 37f87013dfd8 |
| rev | line source |
|---|---|
| 0 | 1 /* |
| 2 * Copyright 1999-2007 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
| 20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
| 21 * have any questions. | |
| 22 * | |
| 23 */ | |
| 24 | |
| 25 #include "incls/_precompiled.incl" | |
| 26 #include "incls/_c1_Runtime1.cpp.incl" | |
| 27 | |
| 28 | |
| 29 // Implementation of StubAssembler | |
| 30 | |
| 31 StubAssembler::StubAssembler(CodeBuffer* code, const char * name, int stub_id) : C1_MacroAssembler(code) { | |
| 32 _name = name; | |
| 33 _must_gc_arguments = false; | |
| 34 _frame_size = no_frame_size; | |
| 35 _num_rt_args = 0; | |
| 36 _stub_id = stub_id; | |
| 37 } | |
| 38 | |
| 39 | |
| 40 void StubAssembler::set_info(const char* name, bool must_gc_arguments) { | |
| 41 _name = name; | |
| 42 _must_gc_arguments = must_gc_arguments; | |
| 43 } | |
| 44 | |
| 45 | |
| 46 void StubAssembler::set_frame_size(int size) { | |
| 47 if (_frame_size == no_frame_size) { | |
| 48 _frame_size = size; | |
| 49 } | |
| 50 assert(_frame_size == size, "can't change the frame size"); | |
| 51 } | |
| 52 | |
| 53 | |
| 54 void StubAssembler::set_num_rt_args(int args) { | |
| 55 if (_num_rt_args == 0) { | |
| 56 _num_rt_args = args; | |
| 57 } | |
| 58 assert(_num_rt_args == args, "can't change the number of args"); | |
| 59 } | |
| 60 | |
| 61 // Implementation of Runtime1 | |
| 62 | |
| 63 bool Runtime1::_is_initialized = false; | |
| 64 CodeBlob* Runtime1::_blobs[Runtime1::number_of_ids]; | |
| 65 const char *Runtime1::_blob_names[] = { | |
| 66 RUNTIME1_STUBS(STUB_NAME, LAST_STUB_NAME) | |
| 67 }; | |
| 68 | |
| 69 #ifndef PRODUCT | |
| 70 // statistics | |
| 71 int Runtime1::_generic_arraycopy_cnt = 0; | |
| 72 int Runtime1::_primitive_arraycopy_cnt = 0; | |
| 73 int Runtime1::_oop_arraycopy_cnt = 0; | |
| 74 int Runtime1::_arraycopy_slowcase_cnt = 0; | |
| 75 int Runtime1::_new_type_array_slowcase_cnt = 0; | |
| 76 int Runtime1::_new_object_array_slowcase_cnt = 0; | |
| 77 int Runtime1::_new_instance_slowcase_cnt = 0; | |
| 78 int Runtime1::_new_multi_array_slowcase_cnt = 0; | |
| 79 int Runtime1::_monitorenter_slowcase_cnt = 0; | |
| 80 int Runtime1::_monitorexit_slowcase_cnt = 0; | |
| 81 int Runtime1::_patch_code_slowcase_cnt = 0; | |
| 82 int Runtime1::_throw_range_check_exception_count = 0; | |
| 83 int Runtime1::_throw_index_exception_count = 0; | |
| 84 int Runtime1::_throw_div0_exception_count = 0; | |
| 85 int Runtime1::_throw_null_pointer_exception_count = 0; | |
| 86 int Runtime1::_throw_class_cast_exception_count = 0; | |
| 87 int Runtime1::_throw_incompatible_class_change_error_count = 0; | |
| 88 int Runtime1::_throw_array_store_exception_count = 0; | |
| 89 int Runtime1::_throw_count = 0; | |
| 90 #endif | |
| 91 | |
| 92 BufferBlob* Runtime1::_buffer_blob = NULL; | |
| 93 | |
| 94 // Simple helper to see if the caller of a runtime stub which | |
| 95 // entered the VM has been deoptimized | |
| 96 | |
| 97 static bool caller_is_deopted() { | |
| 98 JavaThread* thread = JavaThread::current(); | |
| 99 RegisterMap reg_map(thread, false); | |
| 100 frame runtime_frame = thread->last_frame(); | |
| 101 frame caller_frame = runtime_frame.sender(®_map); | |
| 102 assert(caller_frame.is_compiled_frame(), "must be compiled"); | |
| 103 return caller_frame.is_deoptimized_frame(); | |
| 104 } | |
| 105 | |
| 106 // Stress deoptimization | |
| 107 static void deopt_caller() { | |
| 108 if ( !caller_is_deopted()) { | |
| 109 JavaThread* thread = JavaThread::current(); | |
| 110 RegisterMap reg_map(thread, false); | |
| 111 frame runtime_frame = thread->last_frame(); | |
| 112 frame caller_frame = runtime_frame.sender(®_map); | |
| 113 VM_DeoptimizeFrame deopt(thread, caller_frame.id()); | |
| 114 VMThread::execute(&deopt); | |
| 115 assert(caller_is_deopted(), "Must be deoptimized"); | |
| 116 } | |
| 117 } | |
| 118 | |
| 119 | |
| 120 BufferBlob* Runtime1::get_buffer_blob() { | |
| 121 // Allocate code buffer space only once | |
| 122 BufferBlob* blob = _buffer_blob; | |
| 123 if (blob == NULL) { | |
| 124 // setup CodeBuffer. Preallocate a BufferBlob of size | |
| 125 // NMethodSizeLimit plus some extra space for constants. | |
| 126 int code_buffer_size = desired_max_code_buffer_size() + desired_max_constant_size(); | |
| 127 blob = BufferBlob::create("Compiler1 temporary CodeBuffer", | |
| 128 code_buffer_size); | |
| 129 guarantee(blob != NULL, "must create initial code buffer"); | |
| 130 _buffer_blob = blob; | |
| 131 } | |
| 132 return _buffer_blob; | |
| 133 } | |
| 134 | |
| 135 void Runtime1::setup_code_buffer(CodeBuffer* code, int call_stub_estimate) { | |
| 136 // Preinitialize the consts section to some large size: | |
| 137 int locs_buffer_size = 20 * (relocInfo::length_limit + sizeof(relocInfo)); | |
| 138 char* locs_buffer = NEW_RESOURCE_ARRAY(char, locs_buffer_size); | |
| 139 code->insts()->initialize_shared_locs((relocInfo*)locs_buffer, | |
| 140 locs_buffer_size / sizeof(relocInfo)); | |
| 141 code->initialize_consts_size(desired_max_constant_size()); | |
| 142 // Call stubs + deopt/exception handler | |
| 143 code->initialize_stubs_size((call_stub_estimate * LIR_Assembler::call_stub_size) + | |
| 144 LIR_Assembler::exception_handler_size + | |
| 145 LIR_Assembler::deopt_handler_size); | |
| 146 } | |
| 147 | |
| 148 | |
| 149 void Runtime1::generate_blob_for(StubID id) { | |
| 150 assert(0 <= id && id < number_of_ids, "illegal stub id"); | |
| 151 ResourceMark rm; | |
| 152 // create code buffer for code storage | |
| 153 CodeBuffer code(get_buffer_blob()->instructions_begin(), | |
| 154 get_buffer_blob()->instructions_size()); | |
| 155 | |
| 156 setup_code_buffer(&code, 0); | |
| 157 | |
| 158 // create assembler for code generation | |
| 159 StubAssembler* sasm = new StubAssembler(&code, name_for(id), id); | |
| 160 // generate code for runtime stub | |
| 161 OopMapSet* oop_maps; | |
| 162 oop_maps = generate_code_for(id, sasm); | |
| 163 assert(oop_maps == NULL || sasm->frame_size() != no_frame_size, | |
| 164 "if stub has an oop map it must have a valid frame size"); | |
| 165 | |
| 166 #ifdef ASSERT | |
| 167 // Make sure that stubs that need oopmaps have them | |
| 168 switch (id) { | |
| 169 // These stubs don't need to have an oopmap | |
| 170 case dtrace_object_alloc_id: | |
| 171 case slow_subtype_check_id: | |
| 172 case fpu2long_stub_id: | |
| 173 case unwind_exception_id: | |
| 174 #ifndef TIERED | |
| 175 case counter_overflow_id: // Not generated outside the tiered world | |
| 176 #endif | |
| 177 #ifdef SPARC | |
| 178 case handle_exception_nofpu_id: // Unused on sparc | |
| 179 #endif | |
| 180 break; | |
| 181 | |
| 182 // All other stubs should have oopmaps | |
| 183 default: | |
| 184 assert(oop_maps != NULL, "must have an oopmap"); | |
| 185 } | |
| 186 #endif | |
| 187 | |
| 188 // align so printing shows nop's instead of random code at the end (SimpleStubs are aligned) | |
| 189 sasm->align(BytesPerWord); | |
| 190 // make sure all code is in code buffer | |
| 191 sasm->flush(); | |
| 192 // create blob - distinguish a few special cases | |
| 193 CodeBlob* blob = RuntimeStub::new_runtime_stub(name_for(id), | |
| 194 &code, | |
| 195 CodeOffsets::frame_never_safe, | |
| 196 sasm->frame_size(), | |
| 197 oop_maps, | |
| 198 sasm->must_gc_arguments()); | |
| 199 // install blob | |
| 200 assert(blob != NULL, "blob must exist"); | |
| 201 _blobs[id] = blob; | |
| 202 } | |
| 203 | |
| 204 | |
| 205 void Runtime1::initialize() { | |
| 206 // Warning: If we have more than one compilation running in parallel, we | |
| 207 // need a lock here with the current setup (lazy initialization). | |
| 208 if (!is_initialized()) { | |
| 209 _is_initialized = true; | |
| 210 | |
| 211 // platform-dependent initialization | |
| 212 initialize_pd(); | |
| 213 // generate stubs | |
| 214 for (int id = 0; id < number_of_ids; id++) generate_blob_for((StubID)id); | |
| 215 // printing | |
| 216 #ifndef PRODUCT | |
| 217 if (PrintSimpleStubs) { | |
| 218 ResourceMark rm; | |
| 219 for (int id = 0; id < number_of_ids; id++) { | |
| 220 _blobs[id]->print(); | |
| 221 if (_blobs[id]->oop_maps() != NULL) { | |
| 222 _blobs[id]->oop_maps()->print(); | |
| 223 } | |
| 224 } | |
| 225 } | |
| 226 #endif | |
| 227 } | |
| 228 } | |
| 229 | |
| 230 | |
| 231 CodeBlob* Runtime1::blob_for(StubID id) { | |
| 232 assert(0 <= id && id < number_of_ids, "illegal stub id"); | |
| 233 if (!is_initialized()) initialize(); | |
| 234 return _blobs[id]; | |
| 235 } | |
| 236 | |
| 237 | |
| 238 const char* Runtime1::name_for(StubID id) { | |
| 239 assert(0 <= id && id < number_of_ids, "illegal stub id"); | |
| 240 return _blob_names[id]; | |
| 241 } | |
| 242 | |
| 243 const char* Runtime1::name_for_address(address entry) { | |
| 244 for (int id = 0; id < number_of_ids; id++) { | |
| 245 if (entry == entry_for((StubID)id)) return name_for((StubID)id); | |
| 246 } | |
| 247 | |
| 248 #define FUNCTION_CASE(a, f) \ | |
| 249 if ((intptr_t)a == CAST_FROM_FN_PTR(intptr_t, f)) return #f | |
| 250 | |
| 251 FUNCTION_CASE(entry, os::javaTimeMillis); | |
| 252 FUNCTION_CASE(entry, os::javaTimeNanos); | |
| 253 FUNCTION_CASE(entry, SharedRuntime::OSR_migration_end); | |
| 254 FUNCTION_CASE(entry, SharedRuntime::d2f); | |
| 255 FUNCTION_CASE(entry, SharedRuntime::d2i); | |
| 256 FUNCTION_CASE(entry, SharedRuntime::d2l); | |
| 257 FUNCTION_CASE(entry, SharedRuntime::dcos); | |
| 258 FUNCTION_CASE(entry, SharedRuntime::dexp); | |
| 259 FUNCTION_CASE(entry, SharedRuntime::dlog); | |
| 260 FUNCTION_CASE(entry, SharedRuntime::dlog10); | |
| 261 FUNCTION_CASE(entry, SharedRuntime::dpow); | |
| 262 FUNCTION_CASE(entry, SharedRuntime::drem); | |
| 263 FUNCTION_CASE(entry, SharedRuntime::dsin); | |
| 264 FUNCTION_CASE(entry, SharedRuntime::dtan); | |
| 265 FUNCTION_CASE(entry, SharedRuntime::f2i); | |
| 266 FUNCTION_CASE(entry, SharedRuntime::f2l); | |
| 267 FUNCTION_CASE(entry, SharedRuntime::frem); | |
| 268 FUNCTION_CASE(entry, SharedRuntime::l2d); | |
| 269 FUNCTION_CASE(entry, SharedRuntime::l2f); | |
| 270 FUNCTION_CASE(entry, SharedRuntime::ldiv); | |
| 271 FUNCTION_CASE(entry, SharedRuntime::lmul); | |
| 272 FUNCTION_CASE(entry, SharedRuntime::lrem); | |
| 273 FUNCTION_CASE(entry, SharedRuntime::lrem); | |
| 274 FUNCTION_CASE(entry, SharedRuntime::dtrace_method_entry); | |
| 275 FUNCTION_CASE(entry, SharedRuntime::dtrace_method_exit); | |
| 276 FUNCTION_CASE(entry, trace_block_entry); | |
| 277 | |
| 278 #undef FUNCTION_CASE | |
| 279 | |
| 280 return "<unknown function>"; | |
| 281 } | |
| 282 | |
| 283 | |
| 284 JRT_ENTRY(void, Runtime1::new_instance(JavaThread* thread, klassOopDesc* klass)) | |
| 285 NOT_PRODUCT(_new_instance_slowcase_cnt++;) | |
| 286 | |
| 287 assert(oop(klass)->is_klass(), "not a class"); | |
| 288 instanceKlassHandle h(thread, klass); | |
| 289 h->check_valid_for_instantiation(true, CHECK); | |
| 290 // make sure klass is initialized | |
| 291 h->initialize(CHECK); | |
| 292 // allocate instance and return via TLS | |
| 293 oop obj = h->allocate_instance(CHECK); | |
| 294 thread->set_vm_result(obj); | |
| 295 JRT_END | |
| 296 | |
| 297 | |
| 298 JRT_ENTRY(void, Runtime1::new_type_array(JavaThread* thread, klassOopDesc* klass, jint length)) | |
| 299 NOT_PRODUCT(_new_type_array_slowcase_cnt++;) | |
| 300 // Note: no handle for klass needed since they are not used | |
| 301 // anymore after new_typeArray() and no GC can happen before. | |
| 302 // (This may have to change if this code changes!) | |
| 303 assert(oop(klass)->is_klass(), "not a class"); | |
| 304 BasicType elt_type = typeArrayKlass::cast(klass)->element_type(); | |
| 305 oop obj = oopFactory::new_typeArray(elt_type, length, CHECK); | |
| 306 thread->set_vm_result(obj); | |
| 307 // This is pretty rare but this runtime patch is stressful to deoptimization | |
| 308 // if we deoptimize here so force a deopt to stress the path. | |
| 309 if (DeoptimizeALot) { | |
| 310 deopt_caller(); | |
| 311 } | |
| 312 | |
| 313 JRT_END | |
| 314 | |
| 315 | |
| 316 JRT_ENTRY(void, Runtime1::new_object_array(JavaThread* thread, klassOopDesc* array_klass, jint length)) | |
| 317 NOT_PRODUCT(_new_object_array_slowcase_cnt++;) | |
| 318 | |
| 319 // Note: no handle for klass needed since they are not used | |
| 320 // anymore after new_objArray() and no GC can happen before. | |
| 321 // (This may have to change if this code changes!) | |
| 322 assert(oop(array_klass)->is_klass(), "not a class"); | |
| 323 klassOop elem_klass = objArrayKlass::cast(array_klass)->element_klass(); | |
| 324 objArrayOop obj = oopFactory::new_objArray(elem_klass, length, CHECK); | |
| 325 thread->set_vm_result(obj); | |
| 326 // This is pretty rare but this runtime patch is stressful to deoptimization | |
| 327 // if we deoptimize here so force a deopt to stress the path. | |
| 328 if (DeoptimizeALot) { | |
| 329 deopt_caller(); | |
| 330 } | |
| 331 JRT_END | |
| 332 | |
| 333 | |
| 334 JRT_ENTRY(void, Runtime1::new_multi_array(JavaThread* thread, klassOopDesc* klass, int rank, jint* dims)) | |
| 335 NOT_PRODUCT(_new_multi_array_slowcase_cnt++;) | |
| 336 | |
| 337 assert(oop(klass)->is_klass(), "not a class"); | |
| 338 assert(rank >= 1, "rank must be nonzero"); | |
| 339 #ifdef _LP64 | |
| 340 // In 64 bit mode, the sizes are stored in the top 32 bits | |
| 341 // of each 64 bit stack entry. | |
| 342 // dims is actually an intptr_t * because the arguments | |
| 343 // are pushed onto a 64 bit stack. | |
| 344 // We must create an array of jints to pass to multi_allocate. | |
| 345 // We reuse the current stack because it will be popped | |
| 346 // after this bytecode is completed. | |
| 347 if ( rank > 1 ) { | |
| 348 int index; | |
| 349 for ( index = 1; index < rank; index++ ) { // First size is ok | |
| 350 dims[index] = dims[index*2]; | |
| 351 } | |
| 352 } | |
| 353 #endif | |
| 354 oop obj = arrayKlass::cast(klass)->multi_allocate(rank, dims, CHECK); | |
| 355 thread->set_vm_result(obj); | |
| 356 JRT_END | |
| 357 | |
| 358 | |
| 359 JRT_ENTRY(void, Runtime1::unimplemented_entry(JavaThread* thread, StubID id)) | |
| 360 tty->print_cr("Runtime1::entry_for(%d) returned unimplemented entry point", id); | |
| 361 JRT_END | |
| 362 | |
| 363 | |
| 364 JRT_ENTRY(void, Runtime1::throw_array_store_exception(JavaThread* thread)) | |
| 365 THROW(vmSymbolHandles::java_lang_ArrayStoreException()); | |
| 366 JRT_END | |
| 367 | |
| 368 | |
| 369 JRT_ENTRY(void, Runtime1::post_jvmti_exception_throw(JavaThread* thread)) | |
| 370 if (JvmtiExport::can_post_exceptions()) { | |
| 371 vframeStream vfst(thread, true); | |
| 372 address bcp = vfst.method()->bcp_from(vfst.bci()); | |
| 373 JvmtiExport::post_exception_throw(thread, vfst.method(), bcp, thread->exception_oop()); | |
| 374 } | |
| 375 JRT_END | |
| 376 | |
| 377 #ifdef TIERED | |
| 378 JRT_ENTRY(void, Runtime1::counter_overflow(JavaThread* thread, int bci)) | |
| 379 RegisterMap map(thread, false); | |
| 380 frame fr = thread->last_frame().sender(&map); | |
| 381 nmethod* nm = (nmethod*) fr.cb(); | |
| 382 assert(nm!= NULL && nm->is_nmethod(), "what?"); | |
| 383 methodHandle method(thread, nm->method()); | |
| 384 if (bci == 0) { | |
| 385 // invocation counter overflow | |
| 386 if (!Tier1CountOnly) { | |
| 387 CompilationPolicy::policy()->method_invocation_event(method, CHECK); | |
| 388 } else { | |
| 389 method()->invocation_counter()->reset(); | |
| 390 } | |
| 391 } else { | |
| 392 if (!Tier1CountOnly) { | |
| 393 // Twe have a bci but not the destination bci and besides a backedge | |
| 394 // event is more for OSR which we don't want here. | |
| 395 CompilationPolicy::policy()->method_invocation_event(method, CHECK); | |
| 396 } else { | |
| 397 method()->backedge_counter()->reset(); | |
| 398 } | |
| 399 } | |
| 400 JRT_END | |
| 401 #endif // TIERED | |
| 402 | |
| 403 extern void vm_exit(int code); | |
| 404 | |
| 405 // Enter this method from compiled code handler below. This is where we transition | |
| 406 // to VM mode. This is done as a helper routine so that the method called directly | |
| 407 // from compiled code does not have to transition to VM. This allows the entry | |
| 408 // method to see if the nmethod that we have just looked up a handler for has | |
| 409 // been deoptimized while we were in the vm. This simplifies the assembly code | |
| 410 // cpu directories. | |
| 411 // | |
| 412 // We are entering here from exception stub (via the entry method below) | |
| 413 // If there is a compiled exception handler in this method, we will continue there; | |
| 414 // otherwise we will unwind the stack and continue at the caller of top frame method | |
| 415 // Note: we enter in Java using a special JRT wrapper. This wrapper allows us to | |
| 416 // control the area where we can allow a safepoint. After we exit the safepoint area we can | |
| 417 // check to see if the handler we are going to return is now in a nmethod that has | |
| 418 // been deoptimized. If that is the case we return the deopt blob | |
| 419 // unpack_with_exception entry instead. This makes life for the exception blob easier | |
| 420 // because making that same check and diverting is painful from assembly language. | |
| 421 // | |
| 422 | |
| 423 | |
| 424 JRT_ENTRY_NO_ASYNC(static address, exception_handler_for_pc_helper(JavaThread* thread, oopDesc* ex, address pc, nmethod*& nm)) | |
| 425 | |
| 426 Handle exception(thread, ex); | |
| 427 nm = CodeCache::find_nmethod(pc); | |
| 428 assert(nm != NULL, "this is not an nmethod"); | |
| 429 // Adjust the pc as needed/ | |
| 430 if (nm->is_deopt_pc(pc)) { | |
| 431 RegisterMap map(thread, false); | |
| 432 frame exception_frame = thread->last_frame().sender(&map); | |
| 433 // if the frame isn't deopted then pc must not correspond to the caller of last_frame | |
| 434 assert(exception_frame.is_deoptimized_frame(), "must be deopted"); | |
| 435 pc = exception_frame.pc(); | |
| 436 } | |
| 437 #ifdef ASSERT | |
| 438 assert(exception.not_null(), "NULL exceptions should be handled by throw_exception"); | |
| 439 assert(exception->is_oop(), "just checking"); | |
| 440 // Check that exception is a subclass of Throwable, otherwise we have a VerifyError | |
| 441 if (!(exception->is_a(SystemDictionary::throwable_klass()))) { | |
| 442 if (ExitVMOnVerifyError) vm_exit(-1); | |
| 443 ShouldNotReachHere(); | |
| 444 } | |
| 445 #endif | |
| 446 | |
| 447 // Check the stack guard pages and reenable them if necessary and there is | |
| 448 // enough space on the stack to do so. Use fast exceptions only if the guard | |
| 449 // pages are enabled. | |
| 450 bool guard_pages_enabled = thread->stack_yellow_zone_enabled(); | |
| 451 if (!guard_pages_enabled) guard_pages_enabled = thread->reguard_stack(); | |
| 452 | |
| 453 if (JvmtiExport::can_post_exceptions()) { | |
| 454 // To ensure correct notification of exception catches and throws | |
| 455 // we have to deoptimize here. If we attempted to notify the | |
| 456 // catches and throws during this exception lookup it's possible | |
| 457 // we could deoptimize on the way out of the VM and end back in | |
| 458 // the interpreter at the throw site. This would result in double | |
| 459 // notifications since the interpreter would also notify about | |
| 460 // these same catches and throws as it unwound the frame. | |
| 461 | |
| 462 RegisterMap reg_map(thread); | |
| 463 frame stub_frame = thread->last_frame(); | |
| 464 frame caller_frame = stub_frame.sender(®_map); | |
| 465 | |
| 466 // We don't really want to deoptimize the nmethod itself since we | |
| 467 // can actually continue in the exception handler ourselves but I | |
| 468 // don't see an easy way to have the desired effect. | |
| 469 VM_DeoptimizeFrame deopt(thread, caller_frame.id()); | |
| 470 VMThread::execute(&deopt); | |
| 471 | |
| 472 return SharedRuntime::deopt_blob()->unpack_with_exception_in_tls(); | |
| 473 } | |
| 474 | |
| 475 // ExceptionCache is used only for exceptions at call and not for implicit exceptions | |
| 476 if (guard_pages_enabled) { | |
| 477 address fast_continuation = nm->handler_for_exception_and_pc(exception, pc); | |
| 478 if (fast_continuation != NULL) { | |
| 479 if (fast_continuation == ExceptionCache::unwind_handler()) fast_continuation = NULL; | |
| 480 return fast_continuation; | |
| 481 } | |
| 482 } | |
| 483 | |
| 484 // If the stack guard pages are enabled, check whether there is a handler in | |
| 485 // the current method. Otherwise (guard pages disabled), force an unwind and | |
| 486 // skip the exception cache update (i.e., just leave continuation==NULL). | |
| 487 address continuation = NULL; | |
| 488 if (guard_pages_enabled) { | |
| 489 | |
| 490 // New exception handling mechanism can support inlined methods | |
| 491 // with exception handlers since the mappings are from PC to PC | |
| 492 | |
| 493 // debugging support | |
| 494 // tracing | |
| 495 if (TraceExceptions) { | |
| 496 ttyLocker ttyl; | |
| 497 ResourceMark rm; | |
| 498 tty->print_cr("Exception <%s> (0x%x) thrown in compiled method <%s> at PC " PTR_FORMAT " for thread 0x%x", | |
| 499 exception->print_value_string(), (address)exception(), nm->method()->print_value_string(), pc, thread); | |
| 500 } | |
| 501 // for AbortVMOnException flag | |
| 502 NOT_PRODUCT(Exceptions::debug_check_abort(exception)); | |
| 503 | |
| 504 // Clear out the exception oop and pc since looking up an | |
| 505 // exception handler can cause class loading, which might throw an | |
| 506 // exception and those fields are expected to be clear during | |
| 507 // normal bytecode execution. | |
| 508 thread->set_exception_oop(NULL); | |
| 509 thread->set_exception_pc(NULL); | |
| 510 | |
| 511 continuation = SharedRuntime::compute_compiled_exc_handler(nm, pc, exception, false, false); | |
| 512 // If an exception was thrown during exception dispatch, the exception oop may have changed | |
| 513 thread->set_exception_oop(exception()); | |
| 514 thread->set_exception_pc(pc); | |
| 515 | |
| 516 // the exception cache is used only by non-implicit exceptions | |
| 517 if (continuation == NULL) { | |
| 518 nm->add_handler_for_exception_and_pc(exception, pc, ExceptionCache::unwind_handler()); | |
| 519 } else { | |
| 520 nm->add_handler_for_exception_and_pc(exception, pc, continuation); | |
| 521 } | |
| 522 } | |
| 523 | |
| 524 thread->set_vm_result(exception()); | |
| 525 | |
| 526 if (TraceExceptions) { | |
| 527 ttyLocker ttyl; | |
| 528 ResourceMark rm; | |
| 529 tty->print_cr("Thread " PTR_FORMAT " continuing at PC " PTR_FORMAT " for exception thrown at PC " PTR_FORMAT, | |
| 530 thread, continuation, pc); | |
| 531 } | |
| 532 | |
| 533 return continuation; | |
| 534 JRT_END | |
| 535 | |
| 536 // Enter this method from compiled code only if there is a Java exception handler | |
| 537 // in the method handling the exception | |
| 538 // We are entering here from exception stub. We don't do a normal VM transition here. | |
| 539 // We do it in a helper. This is so we can check to see if the nmethod we have just | |
| 540 // searched for an exception handler has been deoptimized in the meantime. | |
| 541 address Runtime1::exception_handler_for_pc(JavaThread* thread) { | |
| 542 oop exception = thread->exception_oop(); | |
| 543 address pc = thread->exception_pc(); | |
| 544 // Still in Java mode | |
| 545 debug_only(ResetNoHandleMark rnhm); | |
| 546 nmethod* nm = NULL; | |
| 547 address continuation = NULL; | |
| 548 { | |
| 549 // Enter VM mode by calling the helper | |
| 550 | |
| 551 ResetNoHandleMark rnhm; | |
| 552 continuation = exception_handler_for_pc_helper(thread, exception, pc, nm); | |
| 553 } | |
| 554 // Back in JAVA, use no oops DON'T safepoint | |
| 555 | |
| 556 // Now check to see if the nmethod we were called from is now deoptimized. | |
| 557 // If so we must return to the deopt blob and deoptimize the nmethod | |
| 558 | |
| 559 if (nm != NULL && caller_is_deopted()) { | |
| 560 continuation = SharedRuntime::deopt_blob()->unpack_with_exception_in_tls(); | |
| 561 } | |
| 562 | |
| 563 return continuation; | |
| 564 } | |
| 565 | |
| 566 | |
| 567 JRT_ENTRY(void, Runtime1::throw_range_check_exception(JavaThread* thread, int index)) | |
| 568 NOT_PRODUCT(_throw_range_check_exception_count++;) | |
| 569 Events::log("throw_range_check"); | |
| 570 char message[jintAsStringSize]; | |
| 571 sprintf(message, "%d", index); | |
| 572 SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_ArrayIndexOutOfBoundsException(), message); | |
| 573 JRT_END | |
| 574 | |
| 575 | |
| 576 JRT_ENTRY(void, Runtime1::throw_index_exception(JavaThread* thread, int index)) | |
| 577 NOT_PRODUCT(_throw_index_exception_count++;) | |
| 578 Events::log("throw_index"); | |
| 579 char message[16]; | |
| 580 sprintf(message, "%d", index); | |
| 581 SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_IndexOutOfBoundsException(), message); | |
| 582 JRT_END | |
| 583 | |
| 584 | |
| 585 JRT_ENTRY(void, Runtime1::throw_div0_exception(JavaThread* thread)) | |
| 586 NOT_PRODUCT(_throw_div0_exception_count++;) | |
| 587 SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_ArithmeticException(), "/ by zero"); | |
| 588 JRT_END | |
| 589 | |
| 590 | |
| 591 JRT_ENTRY(void, Runtime1::throw_null_pointer_exception(JavaThread* thread)) | |
| 592 NOT_PRODUCT(_throw_null_pointer_exception_count++;) | |
| 593 SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_NullPointerException()); | |
| 594 JRT_END | |
| 595 | |
| 596 | |
| 597 JRT_ENTRY(void, Runtime1::throw_class_cast_exception(JavaThread* thread, oopDesc* object)) | |
| 598 NOT_PRODUCT(_throw_class_cast_exception_count++;) | |
| 599 ResourceMark rm(thread); | |
| 600 char* message = SharedRuntime::generate_class_cast_message( | |
| 601 thread, Klass::cast(object->klass())->external_name()); | |
| 602 SharedRuntime::throw_and_post_jvmti_exception( | |
| 603 thread, vmSymbols::java_lang_ClassCastException(), message); | |
| 604 JRT_END | |
| 605 | |
| 606 | |
| 607 JRT_ENTRY(void, Runtime1::throw_incompatible_class_change_error(JavaThread* thread)) | |
| 608 NOT_PRODUCT(_throw_incompatible_class_change_error_count++;) | |
| 609 ResourceMark rm(thread); | |
| 610 SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_IncompatibleClassChangeError()); | |
| 611 JRT_END | |
| 612 | |
| 613 | |
| 614 JRT_ENTRY_NO_ASYNC(void, Runtime1::monitorenter(JavaThread* thread, oopDesc* obj, BasicObjectLock* lock)) | |
| 615 NOT_PRODUCT(_monitorenter_slowcase_cnt++;) | |
| 616 if (PrintBiasedLockingStatistics) { | |
| 617 Atomic::inc(BiasedLocking::slow_path_entry_count_addr()); | |
| 618 } | |
| 619 Handle h_obj(thread, obj); | |
| 620 assert(h_obj()->is_oop(), "must be NULL or an object"); | |
| 621 if (UseBiasedLocking) { | |
| 622 // Retry fast entry if bias is revoked to avoid unnecessary inflation | |
| 623 ObjectSynchronizer::fast_enter(h_obj, lock->lock(), true, CHECK); | |
| 624 } else { | |
| 625 if (UseFastLocking) { | |
| 626 // When using fast locking, the compiled code has already tried the fast case | |
| 627 assert(obj == lock->obj(), "must match"); | |
| 628 ObjectSynchronizer::slow_enter(h_obj, lock->lock(), THREAD); | |
| 629 } else { | |
| 630 lock->set_obj(obj); | |
| 631 ObjectSynchronizer::fast_enter(h_obj, lock->lock(), false, THREAD); | |
| 632 } | |
| 633 } | |
| 634 JRT_END | |
| 635 | |
| 636 | |
| 637 JRT_LEAF(void, Runtime1::monitorexit(JavaThread* thread, BasicObjectLock* lock)) | |
| 638 NOT_PRODUCT(_monitorexit_slowcase_cnt++;) | |
| 639 assert(thread == JavaThread::current(), "threads must correspond"); | |
| 640 assert(thread->last_Java_sp(), "last_Java_sp must be set"); | |
| 641 // monitorexit is non-blocking (leaf routine) => no exceptions can be thrown | |
| 642 EXCEPTION_MARK; | |
| 643 | |
| 644 oop obj = lock->obj(); | |
| 645 assert(obj->is_oop(), "must be NULL or an object"); | |
| 646 if (UseFastLocking) { | |
| 647 // When using fast locking, the compiled code has already tried the fast case | |
| 648 ObjectSynchronizer::slow_exit(obj, lock->lock(), THREAD); | |
| 649 } else { | |
| 650 ObjectSynchronizer::fast_exit(obj, lock->lock(), THREAD); | |
| 651 } | |
| 652 JRT_END | |
| 653 | |
| 654 | |
| 655 static klassOop resolve_field_return_klass(methodHandle caller, int bci, TRAPS) { | |
| 656 Bytecode_field* field_access = Bytecode_field_at(caller(), caller->bcp_from(bci)); | |
| 657 // This can be static or non-static field access | |
| 658 Bytecodes::Code code = field_access->code(); | |
| 659 | |
| 660 // We must load class, initialize class and resolvethe field | |
| 661 FieldAccessInfo result; // initialize class if needed | |
| 662 constantPoolHandle constants(THREAD, caller->constants()); | |
| 663 LinkResolver::resolve_field(result, constants, field_access->index(), Bytecodes::java_code(code), false, CHECK_NULL); | |
| 664 return result.klass()(); | |
| 665 } | |
| 666 | |
| 667 | |
| 668 // | |
| 669 // This routine patches sites where a class wasn't loaded or | |
| 670 // initialized at the time the code was generated. It handles | |
| 671 // references to classes, fields and forcing of initialization. Most | |
| 672 // of the cases are straightforward and involving simply forcing | |
| 673 // resolution of a class, rewriting the instruction stream with the | |
| 674 // needed constant and replacing the call in this function with the | |
| 675 // patched code. The case for static field is more complicated since | |
| 676 // the thread which is in the process of initializing a class can | |
| 677 // access it's static fields but other threads can't so the code | |
| 678 // either has to deoptimize when this case is detected or execute a | |
| 679 // check that the current thread is the initializing thread. The | |
| 680 // current | |
| 681 // | |
| 682 // Patches basically look like this: | |
| 683 // | |
| 684 // | |
| 685 // patch_site: jmp patch stub ;; will be patched | |
| 686 // continue: ... | |
| 687 // ... | |
| 688 // ... | |
| 689 // ... | |
| 690 // | |
| 691 // They have a stub which looks like this: | |
| 692 // | |
| 693 // ;; patch body | |
| 694 // movl <const>, reg (for class constants) | |
| 695 // <or> movl [reg1 + <const>], reg (for field offsets) | |
| 696 // <or> movl reg, [reg1 + <const>] (for field offsets) | |
| 697 // <being_init offset> <bytes to copy> <bytes to skip> | |
| 698 // patch_stub: call Runtime1::patch_code (through a runtime stub) | |
| 699 // jmp patch_site | |
| 700 // | |
| 701 // | |
| 702 // A normal patch is done by rewriting the patch body, usually a move, | |
| 703 // and then copying it into place over top of the jmp instruction | |
| 704 // being careful to flush caches and doing it in an MP-safe way. The | |
| 705 // constants following the patch body are used to find various pieces | |
| 706 // of the patch relative to the call site for Runtime1::patch_code. | |
| 707 // The case for getstatic and putstatic is more complicated because | |
| 708 // getstatic and putstatic have special semantics when executing while | |
| 709 // the class is being initialized. getstatic/putstatic on a class | |
| 710 // which is being_initialized may be executed by the initializing | |
| 711 // thread but other threads have to block when they execute it. This | |
| 712 // is accomplished in compiled code by executing a test of the current | |
| 713 // thread against the initializing thread of the class. It's emitted | |
| 714 // as boilerplate in their stub which allows the patched code to be | |
| 715 // executed before it's copied back into the main body of the nmethod. | |
| 716 // | |
| 717 // being_init: get_thread(<tmp reg> | |
| 718 // cmpl [reg1 + <init_thread_offset>], <tmp reg> | |
| 719 // jne patch_stub | |
| 720 // movl [reg1 + <const>], reg (for field offsets) <or> | |
| 721 // movl reg, [reg1 + <const>] (for field offsets) | |
| 722 // jmp continue | |
| 723 // <being_init offset> <bytes to copy> <bytes to skip> | |
| 724 // patch_stub: jmp Runtim1::patch_code (through a runtime stub) | |
| 725 // jmp patch_site | |
| 726 // | |
| 727 // If the class is being initialized the patch body is rewritten and | |
| 728 // the patch site is rewritten to jump to being_init, instead of | |
| 729 // patch_stub. Whenever this code is executed it checks the current | |
| 730 // thread against the intializing thread so other threads will enter | |
| 731 // the runtime and end up blocked waiting the class to finish | |
| 732 // initializing inside the calls to resolve_field below. The | |
| 733 // initializing class will continue on it's way. Once the class is | |
| 734 // fully_initialized, the intializing_thread of the class becomes | |
| 735 // NULL, so the next thread to execute this code will fail the test, | |
| 736 // call into patch_code and complete the patching process by copying | |
| 737 // the patch body back into the main part of the nmethod and resume | |
| 738 // executing. | |
| 739 // | |
| 740 // | |
| 741 | |
| 742 JRT_ENTRY(void, Runtime1::patch_code(JavaThread* thread, Runtime1::StubID stub_id )) | |
| 743 NOT_PRODUCT(_patch_code_slowcase_cnt++;) | |
| 744 | |
| 745 ResourceMark rm(thread); | |
| 746 RegisterMap reg_map(thread, false); | |
| 747 frame runtime_frame = thread->last_frame(); | |
| 748 frame caller_frame = runtime_frame.sender(®_map); | |
| 749 | |
| 750 // last java frame on stack | |
| 751 vframeStream vfst(thread, true); | |
| 752 assert(!vfst.at_end(), "Java frame must exist"); | |
| 753 | |
| 754 methodHandle caller_method(THREAD, vfst.method()); | |
| 755 // Note that caller_method->code() may not be same as caller_code because of OSR's | |
| 756 // Note also that in the presence of inlining it is not guaranteed | |
| 757 // that caller_method() == caller_code->method() | |
| 758 | |
| 759 | |
| 760 int bci = vfst.bci(); | |
| 761 | |
| 762 Events::log("patch_code @ " INTPTR_FORMAT , caller_frame.pc()); | |
| 763 | |
| 764 Bytecodes::Code code = Bytecode_at(caller_method->bcp_from(bci))->java_code(); | |
| 765 | |
| 766 #ifndef PRODUCT | |
| 767 // this is used by assertions in the access_field_patching_id | |
| 768 BasicType patch_field_type = T_ILLEGAL; | |
| 769 #endif // PRODUCT | |
| 770 bool deoptimize_for_volatile = false; | |
| 771 int patch_field_offset = -1; | |
| 772 KlassHandle init_klass(THREAD, klassOop(NULL)); // klass needed by access_field_patching code | |
| 773 Handle load_klass(THREAD, NULL); // oop needed by load_klass_patching code | |
| 774 if (stub_id == Runtime1::access_field_patching_id) { | |
| 775 | |
| 776 Bytecode_field* field_access = Bytecode_field_at(caller_method(), caller_method->bcp_from(bci)); | |
| 777 FieldAccessInfo result; // initialize class if needed | |
| 778 Bytecodes::Code code = field_access->code(); | |
| 779 constantPoolHandle constants(THREAD, caller_method->constants()); | |
| 780 LinkResolver::resolve_field(result, constants, field_access->index(), Bytecodes::java_code(code), false, CHECK); | |
| 781 patch_field_offset = result.field_offset(); | |
| 782 | |
| 783 // If we're patching a field which is volatile then at compile it | |
| 784 // must not have been know to be volatile, so the generated code | |
| 785 // isn't correct for a volatile reference. The nmethod has to be | |
| 786 // deoptimized so that the code can be regenerated correctly. | |
| 787 // This check is only needed for access_field_patching since this | |
| 788 // is the path for patching field offsets. load_klass is only | |
| 789 // used for patching references to oops which don't need special | |
| 790 // handling in the volatile case. | |
| 791 deoptimize_for_volatile = result.access_flags().is_volatile(); | |
| 792 | |
| 793 #ifndef PRODUCT | |
| 794 patch_field_type = result.field_type(); | |
| 795 #endif | |
| 796 } else if (stub_id == Runtime1::load_klass_patching_id) { | |
| 797 oop k; | |
| 798 switch (code) { | |
| 799 case Bytecodes::_putstatic: | |
| 800 case Bytecodes::_getstatic: | |
| 801 { klassOop klass = resolve_field_return_klass(caller_method, bci, CHECK); | |
| 802 // Save a reference to the class that has to be checked for initialization | |
| 803 init_klass = KlassHandle(THREAD, klass); | |
| 804 k = klass; | |
| 805 } | |
| 806 break; | |
| 807 case Bytecodes::_new: | |
| 808 { Bytecode_new* bnew = Bytecode_new_at(caller_method->bcp_from(bci)); | |
| 809 k = caller_method->constants()->klass_at(bnew->index(), CHECK); | |
| 810 } | |
| 811 break; | |
| 812 case Bytecodes::_multianewarray: | |
| 813 { Bytecode_multianewarray* mna = Bytecode_multianewarray_at(caller_method->bcp_from(bci)); | |
| 814 k = caller_method->constants()->klass_at(mna->index(), CHECK); | |
| 815 } | |
| 816 break; | |
| 817 case Bytecodes::_instanceof: | |
| 818 { Bytecode_instanceof* io = Bytecode_instanceof_at(caller_method->bcp_from(bci)); | |
| 819 k = caller_method->constants()->klass_at(io->index(), CHECK); | |
| 820 } | |
| 821 break; | |
| 822 case Bytecodes::_checkcast: | |
| 823 { Bytecode_checkcast* cc = Bytecode_checkcast_at(caller_method->bcp_from(bci)); | |
| 824 k = caller_method->constants()->klass_at(cc->index(), CHECK); | |
| 825 } | |
| 826 break; | |
| 827 case Bytecodes::_anewarray: | |
| 828 { Bytecode_anewarray* anew = Bytecode_anewarray_at(caller_method->bcp_from(bci)); | |
| 829 klassOop ek = caller_method->constants()->klass_at(anew->index(), CHECK); | |
| 830 k = Klass::cast(ek)->array_klass(CHECK); | |
| 831 } | |
| 832 break; | |
| 833 case Bytecodes::_ldc: | |
| 834 case Bytecodes::_ldc_w: | |
| 835 { | |
| 836 Bytecode_loadconstant* cc = Bytecode_loadconstant_at(caller_method(), | |
| 837 caller_method->bcp_from(bci)); | |
| 838 klassOop resolved = caller_method->constants()->klass_at(cc->index(), CHECK); | |
| 839 // ldc wants the java mirror. | |
| 840 k = resolved->klass_part()->java_mirror(); | |
| 841 } | |
| 842 break; | |
| 843 default: Unimplemented(); | |
| 844 } | |
| 845 // convert to handle | |
| 846 load_klass = Handle(THREAD, k); | |
| 847 } else { | |
| 848 ShouldNotReachHere(); | |
| 849 } | |
| 850 | |
| 851 if (deoptimize_for_volatile) { | |
| 852 // At compile time we assumed the field wasn't volatile but after | |
| 853 // loading it turns out it was volatile so we have to throw the | |
| 854 // compiled code out and let it be regenerated. | |
| 855 if (TracePatching) { | |
| 856 tty->print_cr("Deoptimizing for patching volatile field reference"); | |
| 857 } | |
| 858 VM_DeoptimizeFrame deopt(thread, caller_frame.id()); | |
| 859 VMThread::execute(&deopt); | |
| 860 | |
| 861 // Return to the now deoptimized frame. | |
| 862 } | |
| 863 | |
| 864 | |
| 865 // Now copy code back | |
| 866 | |
| 867 { | |
| 868 MutexLockerEx ml_patch (Patching_lock, Mutex::_no_safepoint_check_flag); | |
| 869 // | |
| 870 // Deoptimization may have happened while we waited for the lock. | |
| 871 // In that case we don't bother to do any patching we just return | |
| 872 // and let the deopt happen | |
| 873 if (!caller_is_deopted()) { | |
| 874 NativeGeneralJump* jump = nativeGeneralJump_at(caller_frame.pc()); | |
| 875 address instr_pc = jump->jump_destination(); | |
| 876 NativeInstruction* ni = nativeInstruction_at(instr_pc); | |
| 877 if (ni->is_jump() ) { | |
| 878 // the jump has not been patched yet | |
| 879 // The jump destination is slow case and therefore not part of the stubs | |
| 880 // (stubs are only for StaticCalls) | |
| 881 | |
| 882 // format of buffer | |
| 883 // .... | |
| 884 // instr byte 0 <-- copy_buff | |
| 885 // instr byte 1 | |
| 886 // .. | |
| 887 // instr byte n-1 | |
| 888 // n | |
| 889 // .... <-- call destination | |
| 890 | |
| 891 address stub_location = caller_frame.pc() + PatchingStub::patch_info_offset(); | |
| 892 unsigned char* byte_count = (unsigned char*) (stub_location - 1); | |
| 893 unsigned char* byte_skip = (unsigned char*) (stub_location - 2); | |
| 894 unsigned char* being_initialized_entry_offset = (unsigned char*) (stub_location - 3); | |
| 895 address copy_buff = stub_location - *byte_skip - *byte_count; | |
| 896 address being_initialized_entry = stub_location - *being_initialized_entry_offset; | |
| 897 if (TracePatching) { | |
| 898 tty->print_cr(" Patching %s at bci %d at address 0x%x (%s)", Bytecodes::name(code), bci, | |
| 899 instr_pc, (stub_id == Runtime1::access_field_patching_id) ? "field" : "klass"); | |
| 900 nmethod* caller_code = CodeCache::find_nmethod(caller_frame.pc()); | |
| 901 assert(caller_code != NULL, "nmethod not found"); | |
| 902 | |
| 903 // NOTE we use pc() not original_pc() because we already know they are | |
| 904 // identical otherwise we'd have never entered this block of code | |
| 905 | |
| 906 OopMap* map = caller_code->oop_map_for_return_address(caller_frame.pc()); | |
| 907 assert(map != NULL, "null check"); | |
| 908 map->print(); | |
| 909 tty->cr(); | |
| 910 | |
| 911 Disassembler::decode(copy_buff, copy_buff + *byte_count, tty); | |
| 912 } | |
| 913 // depending on the code below, do_patch says whether to copy the patch body back into the nmethod | |
| 914 bool do_patch = true; | |
| 915 if (stub_id == Runtime1::access_field_patching_id) { | |
| 916 // The offset may not be correct if the class was not loaded at code generation time. | |
| 917 // Set it now. | |
| 918 NativeMovRegMem* n_move = nativeMovRegMem_at(copy_buff); | |
| 919 assert(n_move->offset() == 0 || (n_move->offset() == 4 && (patch_field_type == T_DOUBLE || patch_field_type == T_LONG)), "illegal offset for type"); | |
| 920 assert(patch_field_offset >= 0, "illegal offset"); | |
| 921 n_move->add_offset_in_bytes(patch_field_offset); | |
| 922 } else if (stub_id == Runtime1::load_klass_patching_id) { | |
| 923 // If a getstatic or putstatic is referencing a klass which | |
| 924 // isn't fully initialized, the patch body isn't copied into | |
| 925 // place until initialization is complete. In this case the | |
| 926 // patch site is setup so that any threads besides the | |
| 927 // initializing thread are forced to come into the VM and | |
| 928 // block. | |
| 929 do_patch = (code != Bytecodes::_getstatic && code != Bytecodes::_putstatic) || | |
| 930 instanceKlass::cast(init_klass())->is_initialized(); | |
| 931 NativeGeneralJump* jump = nativeGeneralJump_at(instr_pc); | |
| 932 if (jump->jump_destination() == being_initialized_entry) { | |
| 933 assert(do_patch == true, "initialization must be complete at this point"); | |
| 934 } else { | |
| 935 // patch the instruction <move reg, klass> | |
| 936 NativeMovConstReg* n_copy = nativeMovConstReg_at(copy_buff); | |
| 937 assert(n_copy->data() == 0, "illegal init value"); | |
| 938 assert(load_klass() != NULL, "klass not set"); | |
| 939 n_copy->set_data((intx) (load_klass())); | |
| 940 | |
| 941 if (TracePatching) { | |
| 942 Disassembler::decode(copy_buff, copy_buff + *byte_count, tty); | |
| 943 } | |
| 944 | |
| 945 #ifdef SPARC | |
| 946 // Update the oop location in the nmethod with the proper | |
| 947 // oop. When the code was generated, a NULL was stuffed | |
| 948 // in the oop table and that table needs to be update to | |
| 949 // have the right value. On intel the value is kept | |
| 950 // directly in the instruction instead of in the oop | |
| 951 // table, so set_data above effectively updated the value. | |
| 952 nmethod* nm = CodeCache::find_nmethod(instr_pc); | |
| 953 assert(nm != NULL, "invalid nmethod_pc"); | |
| 954 RelocIterator oops(nm, copy_buff, copy_buff + 1); | |
| 955 bool found = false; | |
| 956 while (oops.next() && !found) { | |
| 957 if (oops.type() == relocInfo::oop_type) { | |
| 958 oop_Relocation* r = oops.oop_reloc(); | |
| 959 oop* oop_adr = r->oop_addr(); | |
| 960 *oop_adr = load_klass(); | |
| 961 r->fix_oop_relocation(); | |
| 962 found = true; | |
| 963 } | |
| 964 } | |
| 965 assert(found, "the oop must exist!"); | |
| 966 #endif | |
| 967 | |
| 968 } | |
| 969 } else { | |
| 970 ShouldNotReachHere(); | |
| 971 } | |
| 972 if (do_patch) { | |
| 973 // replace instructions | |
| 974 // first replace the tail, then the call | |
| 975 for (int i = NativeCall::instruction_size; i < *byte_count; i++) { | |
| 976 address ptr = copy_buff + i; | |
| 977 int a_byte = (*ptr) & 0xFF; | |
| 978 address dst = instr_pc + i; | |
| 979 *(unsigned char*)dst = (unsigned char) a_byte; | |
| 980 } | |
| 981 ICache::invalidate_range(instr_pc, *byte_count); | |
| 982 NativeGeneralJump::replace_mt_safe(instr_pc, copy_buff); | |
| 983 | |
| 984 if (stub_id == Runtime1::load_klass_patching_id) { | |
| 985 // update relocInfo to oop | |
| 986 nmethod* nm = CodeCache::find_nmethod(instr_pc); | |
| 987 assert(nm != NULL, "invalid nmethod_pc"); | |
| 988 | |
| 989 // The old patch site is now a move instruction so update | |
| 990 // the reloc info so that it will get updated during | |
| 991 // future GCs. | |
| 992 RelocIterator iter(nm, (address)instr_pc, (address)(instr_pc + 1)); | |
| 993 relocInfo::change_reloc_info_for_address(&iter, (address) instr_pc, | |
| 994 relocInfo::none, relocInfo::oop_type); | |
| 995 #ifdef SPARC | |
| 996 // Sparc takes two relocations for an oop so update the second one. | |
| 997 address instr_pc2 = instr_pc + NativeMovConstReg::add_offset; | |
| 998 RelocIterator iter2(nm, instr_pc2, instr_pc2 + 1); | |
| 999 relocInfo::change_reloc_info_for_address(&iter2, (address) instr_pc2, | |
| 1000 relocInfo::none, relocInfo::oop_type); | |
| 1001 #endif | |
| 1002 } | |
| 1003 | |
| 1004 } else { | |
| 1005 ICache::invalidate_range(copy_buff, *byte_count); | |
| 1006 NativeGeneralJump::insert_unconditional(instr_pc, being_initialized_entry); | |
| 1007 } | |
| 1008 } | |
| 1009 } | |
| 1010 } | |
| 1011 JRT_END | |
| 1012 | |
| 1013 // | |
| 1014 // Entry point for compiled code. We want to patch a nmethod. | |
| 1015 // We don't do a normal VM transition here because we want to | |
| 1016 // know after the patching is complete and any safepoint(s) are taken | |
| 1017 // if the calling nmethod was deoptimized. We do this by calling a | |
| 1018 // helper method which does the normal VM transition and when it | |
| 1019 // completes we can check for deoptimization. This simplifies the | |
| 1020 // assembly code in the cpu directories. | |
| 1021 // | |
| 1022 int Runtime1::move_klass_patching(JavaThread* thread) { | |
| 1023 // | |
| 1024 // NOTE: we are still in Java | |
| 1025 // | |
| 1026 Thread* THREAD = thread; | |
| 1027 debug_only(NoHandleMark nhm;) | |
| 1028 { | |
| 1029 // Enter VM mode | |
| 1030 | |
| 1031 ResetNoHandleMark rnhm; | |
| 1032 patch_code(thread, load_klass_patching_id); | |
| 1033 } | |
| 1034 // Back in JAVA, use no oops DON'T safepoint | |
| 1035 | |
| 1036 // Return true if calling code is deoptimized | |
| 1037 | |
| 1038 return caller_is_deopted(); | |
| 1039 } | |
| 1040 | |
| 1041 // | |
| 1042 // Entry point for compiled code. We want to patch a nmethod. | |
| 1043 // We don't do a normal VM transition here because we want to | |
| 1044 // know after the patching is complete and any safepoint(s) are taken | |
| 1045 // if the calling nmethod was deoptimized. We do this by calling a | |
| 1046 // helper method which does the normal VM transition and when it | |
| 1047 // completes we can check for deoptimization. This simplifies the | |
| 1048 // assembly code in the cpu directories. | |
| 1049 // | |
| 1050 | |
| 1051 int Runtime1::access_field_patching(JavaThread* thread) { | |
| 1052 // | |
| 1053 // NOTE: we are still in Java | |
| 1054 // | |
| 1055 Thread* THREAD = thread; | |
| 1056 debug_only(NoHandleMark nhm;) | |
| 1057 { | |
| 1058 // Enter VM mode | |
| 1059 | |
| 1060 ResetNoHandleMark rnhm; | |
| 1061 patch_code(thread, access_field_patching_id); | |
| 1062 } | |
| 1063 // Back in JAVA, use no oops DON'T safepoint | |
| 1064 | |
| 1065 // Return true if calling code is deoptimized | |
| 1066 | |
| 1067 return caller_is_deopted(); | |
| 1068 JRT_END | |
| 1069 | |
| 1070 | |
| 1071 JRT_LEAF(void, Runtime1::trace_block_entry(jint block_id)) | |
| 1072 // for now we just print out the block id | |
| 1073 tty->print("%d ", block_id); | |
| 1074 JRT_END | |
| 1075 | |
| 1076 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1077 // Array copy return codes. |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1078 enum { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1079 ac_failed = -1, // arraycopy failed |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1080 ac_ok = 0 // arraycopy succeeded |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1081 }; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1082 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1083 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1084 template <class T> int obj_arraycopy_work(oopDesc* src, T* src_addr, |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1085 oopDesc* dst, T* dst_addr, |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1086 int length) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1087 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1088 // For performance reasons, we assume we are using a card marking write |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1089 // barrier. The assert will fail if this is not the case. |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1090 // Note that we use the non-virtual inlineable variant of write_ref_array. |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1091 BarrierSet* bs = Universe::heap()->barrier_set(); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1092 assert(bs->has_write_ref_array_opt(), |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1093 "Barrier set must have ref array opt"); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1094 if (src == dst) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1095 // same object, no check |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1096 Copy::conjoint_oops_atomic(src_addr, dst_addr, length); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1097 bs->write_ref_array(MemRegion((HeapWord*)dst_addr, |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1098 (HeapWord*)(dst_addr + length))); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1099 return ac_ok; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1100 } else { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1101 klassOop bound = objArrayKlass::cast(dst->klass())->element_klass(); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1102 klassOop stype = objArrayKlass::cast(src->klass())->element_klass(); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1103 if (stype == bound || Klass::cast(stype)->is_subtype_of(bound)) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1104 // Elements are guaranteed to be subtypes, so no check necessary |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1105 Copy::conjoint_oops_atomic(src_addr, dst_addr, length); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1106 bs->write_ref_array(MemRegion((HeapWord*)dst_addr, |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1107 (HeapWord*)(dst_addr + length))); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1108 return ac_ok; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1109 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1110 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1111 return ac_failed; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1112 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1113 |
| 0 | 1114 // fast and direct copy of arrays; returning -1, means that an exception may be thrown |
| 1115 // and we did not copy anything | |
| 1116 JRT_LEAF(int, Runtime1::arraycopy(oopDesc* src, int src_pos, oopDesc* dst, int dst_pos, int length)) | |
| 1117 #ifndef PRODUCT | |
| 1118 _generic_arraycopy_cnt++; // Slow-path oop array copy | |
| 1119 #endif | |
| 1120 | |
| 1121 if (src == NULL || dst == NULL || src_pos < 0 || dst_pos < 0 || length < 0) return ac_failed; | |
| 1122 if (!dst->is_array() || !src->is_array()) return ac_failed; | |
| 1123 if ((unsigned int) arrayOop(src)->length() < (unsigned int)src_pos + (unsigned int)length) return ac_failed; | |
| 1124 if ((unsigned int) arrayOop(dst)->length() < (unsigned int)dst_pos + (unsigned int)length) return ac_failed; | |
| 1125 | |
| 1126 if (length == 0) return ac_ok; | |
| 1127 if (src->is_typeArray()) { | |
| 1128 const klassOop klass_oop = src->klass(); | |
| 1129 if (klass_oop != dst->klass()) return ac_failed; | |
| 1130 typeArrayKlass* klass = typeArrayKlass::cast(klass_oop); | |
| 1131 const int l2es = klass->log2_element_size(); | |
| 1132 const int ihs = klass->array_header_in_bytes() / wordSize; | |
| 1133 char* src_addr = (char*) ((oopDesc**)src + ihs) + (src_pos << l2es); | |
| 1134 char* dst_addr = (char*) ((oopDesc**)dst + ihs) + (dst_pos << l2es); | |
| 1135 // Potential problem: memmove is not guaranteed to be word atomic | |
| 1136 // Revisit in Merlin | |
| 1137 memmove(dst_addr, src_addr, length << l2es); | |
| 1138 return ac_ok; | |
| 1139 } else if (src->is_objArray() && dst->is_objArray()) { | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1140 if (UseCompressedOops) { // will need for tiered |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1141 narrowOop *src_addr = objArrayOop(src)->obj_at_addr<narrowOop>(src_pos); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1142 narrowOop *dst_addr = objArrayOop(dst)->obj_at_addr<narrowOop>(dst_pos); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1143 return obj_arraycopy_work(src, src_addr, dst, dst_addr, length); |
| 0 | 1144 } else { |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1145 oop *src_addr = objArrayOop(src)->obj_at_addr<oop>(src_pos); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1146 oop *dst_addr = objArrayOop(dst)->obj_at_addr<oop>(dst_pos); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1147 return obj_arraycopy_work(src, src_addr, dst, dst_addr, length); |
| 0 | 1148 } |
| 1149 } | |
| 1150 return ac_failed; | |
| 1151 JRT_END | |
| 1152 | |
| 1153 | |
| 1154 JRT_LEAF(void, Runtime1::primitive_arraycopy(HeapWord* src, HeapWord* dst, int length)) | |
| 1155 #ifndef PRODUCT | |
| 1156 _primitive_arraycopy_cnt++; | |
| 1157 #endif | |
| 1158 | |
| 1159 if (length == 0) return; | |
| 1160 // Not guaranteed to be word atomic, but that doesn't matter | |
| 1161 // for anything but an oop array, which is covered by oop_arraycopy. | |
| 1162 Copy::conjoint_bytes(src, dst, length); | |
| 1163 JRT_END | |
| 1164 | |
| 1165 JRT_LEAF(void, Runtime1::oop_arraycopy(HeapWord* src, HeapWord* dst, int num)) | |
| 1166 #ifndef PRODUCT | |
| 1167 _oop_arraycopy_cnt++; | |
| 1168 #endif | |
| 1169 | |
| 1170 if (num == 0) return; | |
| 1171 Copy::conjoint_oops_atomic((oop*) src, (oop*) dst, num); | |
| 1172 BarrierSet* bs = Universe::heap()->barrier_set(); | |
| 1173 bs->write_ref_array(MemRegion(dst, dst + num)); | |
| 1174 JRT_END | |
| 1175 | |
| 1176 | |
| 1177 #ifndef PRODUCT | |
| 1178 void Runtime1::print_statistics() { | |
| 1179 tty->print_cr("C1 Runtime statistics:"); | |
| 1180 tty->print_cr(" _resolve_invoke_virtual_cnt: %d", SharedRuntime::_resolve_virtual_ctr); | |
| 1181 tty->print_cr(" _resolve_invoke_opt_virtual_cnt: %d", SharedRuntime::_resolve_opt_virtual_ctr); | |
| 1182 tty->print_cr(" _resolve_invoke_static_cnt: %d", SharedRuntime::_resolve_static_ctr); | |
| 1183 tty->print_cr(" _handle_wrong_method_cnt: %d", SharedRuntime::_wrong_method_ctr); | |
| 1184 tty->print_cr(" _ic_miss_cnt: %d", SharedRuntime::_ic_miss_ctr); | |
| 1185 tty->print_cr(" _generic_arraycopy_cnt: %d", _generic_arraycopy_cnt); | |
| 1186 tty->print_cr(" _primitive_arraycopy_cnt: %d", _primitive_arraycopy_cnt); | |
| 1187 tty->print_cr(" _oop_arraycopy_cnt: %d", _oop_arraycopy_cnt); | |
| 1188 tty->print_cr(" _arraycopy_slowcase_cnt: %d", _arraycopy_slowcase_cnt); | |
| 1189 | |
| 1190 tty->print_cr(" _new_type_array_slowcase_cnt: %d", _new_type_array_slowcase_cnt); | |
| 1191 tty->print_cr(" _new_object_array_slowcase_cnt: %d", _new_object_array_slowcase_cnt); | |
| 1192 tty->print_cr(" _new_instance_slowcase_cnt: %d", _new_instance_slowcase_cnt); | |
| 1193 tty->print_cr(" _new_multi_array_slowcase_cnt: %d", _new_multi_array_slowcase_cnt); | |
| 1194 tty->print_cr(" _monitorenter_slowcase_cnt: %d", _monitorenter_slowcase_cnt); | |
| 1195 tty->print_cr(" _monitorexit_slowcase_cnt: %d", _monitorexit_slowcase_cnt); | |
| 1196 tty->print_cr(" _patch_code_slowcase_cnt: %d", _patch_code_slowcase_cnt); | |
| 1197 | |
| 1198 tty->print_cr(" _throw_range_check_exception_count: %d:", _throw_range_check_exception_count); | |
| 1199 tty->print_cr(" _throw_index_exception_count: %d:", _throw_index_exception_count); | |
| 1200 tty->print_cr(" _throw_div0_exception_count: %d:", _throw_div0_exception_count); | |
| 1201 tty->print_cr(" _throw_null_pointer_exception_count: %d:", _throw_null_pointer_exception_count); | |
| 1202 tty->print_cr(" _throw_class_cast_exception_count: %d:", _throw_class_cast_exception_count); | |
| 1203 tty->print_cr(" _throw_incompatible_class_change_error_count: %d:", _throw_incompatible_class_change_error_count); | |
| 1204 tty->print_cr(" _throw_array_store_exception_count: %d:", _throw_array_store_exception_count); | |
| 1205 tty->print_cr(" _throw_count: %d:", _throw_count); | |
| 1206 | |
| 1207 SharedRuntime::print_ic_miss_histogram(); | |
| 1208 tty->cr(); | |
| 1209 } | |
| 1210 #endif // PRODUCT |