1 /* 2 * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. 3 * Copyright 2012, 2014 SAP AG. All rights reserved. 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This code is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 only, as 8 * published by the Free Software Foundation. 9 * 10 * This code is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 13 * version 2 for more details (a copy is included in the LICENSE file that 14 * accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License version 17 * 2 along with this work; if not, write to the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 21 * or visit www.oracle.com if you need additional information or have any 22 * questions. 23 * 24 */ 25 26 #include "precompiled.hpp" 27 #include "asm/assembler.hpp" 28 #include "asm/macroAssembler.inline.hpp" 29 #include "interpreter/bytecodeHistogram.hpp" 30 #include "interpreter/interpreter.hpp" 31 #include "interpreter/interpreterGenerator.hpp" 32 #include "interpreter/interpreterRuntime.hpp" 33 #include "interpreter/interp_masm.hpp" 34 #include "interpreter/templateTable.hpp" 35 #include "oops/arrayOop.hpp" 36 #include "oops/methodData.hpp" 37 #include "oops/method.hpp" 38 #include "oops/oop.inline.hpp" 39 #include "prims/jvmtiExport.hpp" 40 #include "prims/jvmtiThreadState.hpp" 41 #include "prims/methodHandles.hpp" 42 #include "runtime/arguments.hpp" 43 #include "runtime/deoptimization.hpp" 44 #include "runtime/frame.inline.hpp" 45 #include "runtime/sharedRuntime.hpp" 46 #include "runtime/stubRoutines.hpp" 47 #include "runtime/synchronizer.hpp" 48 #include "runtime/timer.hpp" 49 #include "runtime/vframeArray.hpp" 50 #include "utilities/debug.hpp" 51 #ifdef COMPILER1 52 #include "c1/c1_Runtime1.hpp" 53 #endif 54 55 #define __ _masm-> 56 57 #ifdef PRODUCT 58 #define BLOCK_COMMENT(str) // nothing 59 #else 60 #define BLOCK_COMMENT(str) __ block_comment(str) 61 #endif 62 63 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") 64 65 int AbstractInterpreter::BasicType_as_index(BasicType type) { 66 int i = 0; 67 switch (type) { 68 case T_BOOLEAN: i = 0; break; 69 case T_CHAR : i = 1; break; 70 case T_BYTE : i = 2; break; 71 case T_SHORT : i = 3; break; 72 case T_INT : i = 4; break; 73 case T_LONG : i = 5; break; 74 case T_VOID : i = 6; break; 75 case T_FLOAT : i = 7; break; 76 case T_DOUBLE : i = 8; break; 77 case T_OBJECT : i = 9; break; 78 case T_ARRAY : i = 9; break; 79 default : ShouldNotReachHere(); 80 } 81 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds"); 82 return i; 83 } 84 85 address AbstractInterpreterGenerator::generate_slow_signature_handler() { 86 // Slow_signature handler that respects the PPC C calling conventions. 87 // 88 // We get called by the native entry code with our output register 89 // area == 8. First we call InterpreterRuntime::get_result_handler 90 // to copy the pointer to the signature string temporarily to the 91 // first C-argument and to return the result_handler in 92 // R3_RET. Since native_entry will copy the jni-pointer to the 93 // first C-argument slot later on, it is OK to occupy this slot 94 // temporarilly. Then we copy the argument list on the java 95 // expression stack into native varargs format on the native stack 96 // and load arguments into argument registers. Integer arguments in 97 // the varargs vector will be sign-extended to 8 bytes. 98 // 99 // On entry: 100 // R3_ARG1 - intptr_t* Address of java argument list in memory. 101 // R15_prev_state - BytecodeInterpreter* Address of interpreter state for 102 // this method 103 // R19_method 104 // 105 // On exit (just before return instruction): 106 // R3_RET - contains the address of the result_handler. 107 // R4_ARG2 - is not updated for static methods and contains "this" otherwise. 108 // R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double, 109 // ARGi contains this argument. Otherwise, ARGi is not updated. 110 // F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double. 111 112 const int LogSizeOfTwoInstructions = 3; 113 114 // FIXME: use Argument:: GL: Argument names different numbers! 115 const int max_fp_register_arguments = 13; 116 const int max_int_register_arguments = 6; // first 2 are reserved 117 118 const Register arg_java = R21_tmp1; 119 const Register arg_c = R22_tmp2; 120 const Register signature = R23_tmp3; // is string 121 const Register sig_byte = R24_tmp4; 122 const Register fpcnt = R25_tmp5; 123 const Register argcnt = R26_tmp6; 124 const Register intSlot = R27_tmp7; 125 const Register target_sp = R28_tmp8; 126 const FloatRegister floatSlot = F0; 127 128 address entry = __ function_entry(); 129 130 __ save_LR_CR(R0); 131 __ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14)); 132 // We use target_sp for storing arguments in the C frame. 133 __ mr(target_sp, R1_SP); 134 __ push_frame_reg_args_nonvolatiles(0, R11_scratch1); 135 136 __ mr(arg_java, R3_ARG1); 137 138 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method); 139 140 // Signature is in R3_RET. Signature is callee saved. 141 __ mr(signature, R3_RET); 142 143 // Get the result handler. 144 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method); 145 146 { 147 Label L; 148 // test if static 149 // _access_flags._flags must be at offset 0. 150 // TODO PPC port: requires change in shared code. 151 //assert(in_bytes(AccessFlags::flags_offset()) == 0, 152 // "MethodDesc._access_flags == MethodDesc._access_flags._flags"); 153 // _access_flags must be a 32 bit value. 154 assert(sizeof(AccessFlags) == 4, "wrong size"); 155 __ lwa(R11_scratch1/*access_flags*/, method_(access_flags)); 156 // testbit with condition register. 157 __ testbitdi(CCR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT); 158 __ btrue(CCR0, L); 159 // For non-static functions, pass "this" in R4_ARG2 and copy it 160 // to 2nd C-arg slot. 161 // We need to box the Java object here, so we use arg_java 162 // (address of current Java stack slot) as argument and don't 163 // dereference it as in case of ints, floats, etc. 164 __ mr(R4_ARG2, arg_java); 165 __ addi(arg_java, arg_java, -BytesPerWord); 166 __ std(R4_ARG2, _abi(carg_2), target_sp); 167 __ bind(L); 168 } 169 170 // Will be incremented directly after loop_start. argcnt=0 171 // corresponds to 3rd C argument. 172 __ li(argcnt, -1); 173 // arg_c points to 3rd C argument 174 __ addi(arg_c, target_sp, _abi(carg_3)); 175 // no floating-point args parsed so far 176 __ li(fpcnt, 0); 177 178 Label move_intSlot_to_ARG, move_floatSlot_to_FARG; 179 Label loop_start, loop_end; 180 Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed; 181 182 // signature points to '(' at entry 183 #ifdef ASSERT 184 __ lbz(sig_byte, 0, signature); 185 __ cmplwi(CCR0, sig_byte, '('); 186 __ bne(CCR0, do_dontreachhere); 187 #endif 188 189 __ bind(loop_start); 190 191 __ addi(argcnt, argcnt, 1); 192 __ lbzu(sig_byte, 1, signature); 193 194 __ cmplwi(CCR0, sig_byte, ')'); // end of signature 195 __ beq(CCR0, loop_end); 196 197 __ cmplwi(CCR0, sig_byte, 'B'); // byte 198 __ beq(CCR0, do_int); 199 200 __ cmplwi(CCR0, sig_byte, 'C'); // char 201 __ beq(CCR0, do_int); 202 203 __ cmplwi(CCR0, sig_byte, 'D'); // double 204 __ beq(CCR0, do_double); 205 206 __ cmplwi(CCR0, sig_byte, 'F'); // float 207 __ beq(CCR0, do_float); 208 209 __ cmplwi(CCR0, sig_byte, 'I'); // int 210 __ beq(CCR0, do_int); 211 212 __ cmplwi(CCR0, sig_byte, 'J'); // long 213 __ beq(CCR0, do_long); 214 215 __ cmplwi(CCR0, sig_byte, 'S'); // short 216 __ beq(CCR0, do_int); 217 218 __ cmplwi(CCR0, sig_byte, 'Z'); // boolean 219 __ beq(CCR0, do_int); 220 221 __ cmplwi(CCR0, sig_byte, 'L'); // object 222 __ beq(CCR0, do_object); 223 224 __ cmplwi(CCR0, sig_byte, '['); // array 225 __ beq(CCR0, do_array); 226 227 // __ cmplwi(CCR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type 228 // __ beq(CCR0, do_void); 229 230 __ bind(do_dontreachhere); 231 232 __ unimplemented("ShouldNotReachHere in slow_signature_handler", 120); 233 234 __ bind(do_array); 235 236 { 237 Label start_skip, end_skip; 238 239 __ bind(start_skip); 240 __ lbzu(sig_byte, 1, signature); 241 __ cmplwi(CCR0, sig_byte, '['); 242 __ beq(CCR0, start_skip); // skip further brackets 243 __ cmplwi(CCR0, sig_byte, '9'); 244 __ bgt(CCR0, end_skip); // no optional size 245 __ cmplwi(CCR0, sig_byte, '0'); 246 __ bge(CCR0, start_skip); // skip optional size 247 __ bind(end_skip); 248 249 __ cmplwi(CCR0, sig_byte, 'L'); 250 __ beq(CCR0, do_object); // for arrays of objects, the name of the object must be skipped 251 __ b(do_boxed); // otherwise, go directly to do_boxed 252 } 253 254 __ bind(do_object); 255 { 256 Label L; 257 __ bind(L); 258 __ lbzu(sig_byte, 1, signature); 259 __ cmplwi(CCR0, sig_byte, ';'); 260 __ bne(CCR0, L); 261 } 262 // Need to box the Java object here, so we use arg_java (address of 263 // current Java stack slot) as argument and don't dereference it as 264 // in case of ints, floats, etc. 265 Label do_null; 266 __ bind(do_boxed); 267 __ ld(R0,0, arg_java); 268 __ cmpdi(CCR0, R0, 0); 269 __ li(intSlot,0); 270 __ beq(CCR0, do_null); 271 __ mr(intSlot, arg_java); 272 __ bind(do_null); 273 __ std(intSlot, 0, arg_c); 274 __ addi(arg_java, arg_java, -BytesPerWord); 275 __ addi(arg_c, arg_c, BytesPerWord); 276 __ cmplwi(CCR0, argcnt, max_int_register_arguments); 277 __ blt(CCR0, move_intSlot_to_ARG); 278 __ b(loop_start); 279 280 __ bind(do_int); 281 __ lwa(intSlot, 0, arg_java); 282 __ std(intSlot, 0, arg_c); 283 __ addi(arg_java, arg_java, -BytesPerWord); 284 __ addi(arg_c, arg_c, BytesPerWord); 285 __ cmplwi(CCR0, argcnt, max_int_register_arguments); 286 __ blt(CCR0, move_intSlot_to_ARG); 287 __ b(loop_start); 288 289 __ bind(do_long); 290 __ ld(intSlot, -BytesPerWord, arg_java); 291 __ std(intSlot, 0, arg_c); 292 __ addi(arg_java, arg_java, - 2 * BytesPerWord); 293 __ addi(arg_c, arg_c, BytesPerWord); 294 __ cmplwi(CCR0, argcnt, max_int_register_arguments); 295 __ blt(CCR0, move_intSlot_to_ARG); 296 __ b(loop_start); 297 298 __ bind(do_float); 299 __ lfs(floatSlot, 0, arg_java); 300 #if defined(LINUX) 301 __ stfs(floatSlot, 4, arg_c); 302 #elif defined(AIX) 303 __ stfs(floatSlot, 0, arg_c); 304 #else 305 #error "unknown OS" 306 #endif 307 __ addi(arg_java, arg_java, -BytesPerWord); 308 __ addi(arg_c, arg_c, BytesPerWord); 309 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments); 310 __ blt(CCR0, move_floatSlot_to_FARG); 311 __ b(loop_start); 312 313 __ bind(do_double); 314 __ lfd(floatSlot, - BytesPerWord, arg_java); 315 __ stfd(floatSlot, 0, arg_c); 316 __ addi(arg_java, arg_java, - 2 * BytesPerWord); 317 __ addi(arg_c, arg_c, BytesPerWord); 318 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments); 319 __ blt(CCR0, move_floatSlot_to_FARG); 320 __ b(loop_start); 321 322 __ bind(loop_end); 323 324 __ pop_frame(); 325 __ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14)); 326 __ restore_LR_CR(R0); 327 328 __ blr(); 329 330 Label move_int_arg, move_float_arg; 331 __ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions) 332 __ mr(R5_ARG3, intSlot); __ b(loop_start); 333 __ mr(R6_ARG4, intSlot); __ b(loop_start); 334 __ mr(R7_ARG5, intSlot); __ b(loop_start); 335 __ mr(R8_ARG6, intSlot); __ b(loop_start); 336 __ mr(R9_ARG7, intSlot); __ b(loop_start); 337 __ mr(R10_ARG8, intSlot); __ b(loop_start); 338 339 __ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions) 340 __ fmr(F1_ARG1, floatSlot); __ b(loop_start); 341 __ fmr(F2_ARG2, floatSlot); __ b(loop_start); 342 __ fmr(F3_ARG3, floatSlot); __ b(loop_start); 343 __ fmr(F4_ARG4, floatSlot); __ b(loop_start); 344 __ fmr(F5_ARG5, floatSlot); __ b(loop_start); 345 __ fmr(F6_ARG6, floatSlot); __ b(loop_start); 346 __ fmr(F7_ARG7, floatSlot); __ b(loop_start); 347 __ fmr(F8_ARG8, floatSlot); __ b(loop_start); 348 __ fmr(F9_ARG9, floatSlot); __ b(loop_start); 349 __ fmr(F10_ARG10, floatSlot); __ b(loop_start); 350 __ fmr(F11_ARG11, floatSlot); __ b(loop_start); 351 __ fmr(F12_ARG12, floatSlot); __ b(loop_start); 352 __ fmr(F13_ARG13, floatSlot); __ b(loop_start); 353 354 __ bind(move_intSlot_to_ARG); 355 __ sldi(R0, argcnt, LogSizeOfTwoInstructions); 356 __ load_const(R11_scratch1, move_int_arg); // Label must be bound here. 357 __ add(R11_scratch1, R0, R11_scratch1); 358 __ mtctr(R11_scratch1/*branch_target*/); 359 __ bctr(); 360 __ bind(move_floatSlot_to_FARG); 361 __ sldi(R0, fpcnt, LogSizeOfTwoInstructions); 362 __ addi(fpcnt, fpcnt, 1); 363 __ load_const(R11_scratch1, move_float_arg); // Label must be bound here. 364 __ add(R11_scratch1, R0, R11_scratch1); 365 __ mtctr(R11_scratch1/*branch_target*/); 366 __ bctr(); 367 368 return entry; 369 } 370 371 address AbstractInterpreterGenerator::generate_result_handler_for(BasicType type) { 372 // 373 // Registers alive 374 // R3_RET 375 // LR 376 // 377 // Registers updated 378 // R3_RET 379 // 380 381 Label done; 382 address entry = __ pc(); 383 384 switch (type) { 385 case T_BOOLEAN: 386 // convert !=0 to 1 387 __ neg(R0, R3_RET); 388 __ orr(R0, R3_RET, R0); 389 __ srwi(R3_RET, R0, 31); 390 break; 391 case T_BYTE: 392 // sign extend 8 bits 393 __ extsb(R3_RET, R3_RET); 394 break; 395 case T_CHAR: 396 // zero extend 16 bits 397 __ clrldi(R3_RET, R3_RET, 48); 398 break; 399 case T_SHORT: 400 // sign extend 16 bits 401 __ extsh(R3_RET, R3_RET); 402 break; 403 case T_INT: 404 // sign extend 32 bits 405 __ extsw(R3_RET, R3_RET); 406 break; 407 case T_LONG: 408 break; 409 case T_OBJECT: 410 // unbox result if not null 411 __ cmpdi(CCR0, R3_RET, 0); 412 __ beq(CCR0, done); 413 __ ld(R3_RET, 0, R3_RET); 414 __ verify_oop(R3_RET); 415 break; 416 case T_FLOAT: 417 break; 418 case T_DOUBLE: 419 break; 420 case T_VOID: 421 break; 422 default: ShouldNotReachHere(); 423 } 424 425 __ BIND(done); 426 __ blr(); 427 428 return entry; 429 } 430 431 // Abstract method entry. 432 // 433 address InterpreterGenerator::generate_abstract_entry(void) { 434 address entry = __ pc(); 435 436 // 437 // Registers alive 438 // R16_thread - JavaThread* 439 // R19_method - callee's method (method to be invoked) 440 // R1_SP - SP prepared such that caller's outgoing args are near top 441 // LR - return address to caller 442 // 443 // Stack layout at this point: 444 // 445 // 0 [TOP_IJAVA_FRAME_ABI] <-- R1_SP 446 // alignment (optional) 447 // [outgoing Java arguments] 448 // ... 449 // PARENT [PARENT_IJAVA_FRAME_ABI] 450 // ... 451 // 452 453 // Can't use call_VM here because we have not set up a new 454 // interpreter state. Make the call to the vm and make it look like 455 // our caller set up the JavaFrameAnchor. 456 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/); 457 458 // Push a new C frame and save LR. 459 __ save_LR_CR(R0); 460 __ push_frame_reg_args(0, R11_scratch1); 461 462 // This is not a leaf but we have a JavaFrameAnchor now and we will 463 // check (create) exceptions afterward so this is ok. 464 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError)); 465 466 // Pop the C frame and restore LR. 467 __ pop_frame(); 468 __ restore_LR_CR(R0); 469 470 // Reset JavaFrameAnchor from call_VM_leaf above. 471 __ reset_last_Java_frame(); 472 473 #ifdef CC_INTERP 474 // Return to frame manager, it will handle the pending exception. 475 __ blr(); 476 #else 477 // We don't know our caller, so jump to the general forward exception stub, 478 // which will also pop our full frame off. Satisfy the interface of 479 // SharedRuntime::generate_forward_exception() 480 __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0); 481 __ mtctr(R11_scratch1); 482 __ bctr(); 483 #endif 484 485 return entry; 486 } 487 488 // Call an accessor method (assuming it is resolved, otherwise drop into 489 // vanilla (slow path) entry. 490 address InterpreterGenerator::generate_accessor_entry(void) { 491 if (!UseFastAccessorMethods && (!FLAG_IS_ERGO(UseFastAccessorMethods))) { 492 return NULL; 493 } 494 495 Label Lslow_path, Lacquire; 496 497 const Register 498 Rclass_or_obj = R3_ARG1, 499 Rconst_method = R4_ARG2, 500 Rcodes = Rconst_method, 501 Rcpool_cache = R5_ARG3, 502 Rscratch = R11_scratch1, 503 Rjvmti_mode = Rscratch, 504 Roffset = R12_scratch2, 505 Rflags = R6_ARG4, 506 Rbtable = R7_ARG5; 507 508 static address branch_table[number_of_states]; 509 510 address entry = __ pc(); 511 512 // Check for safepoint: 513 // Ditch this, real man don't need safepoint checks. 514 515 // Also check for JVMTI mode 516 // Check for null obj, take slow path if so. 517 __ ld(Rclass_or_obj, Interpreter::stackElementSize, CC_INTERP_ONLY(R17_tos) NOT_CC_INTERP(R15_esp)); 518 __ lwz(Rjvmti_mode, thread_(interp_only_mode)); 519 __ cmpdi(CCR1, Rclass_or_obj, 0); 520 __ cmpwi(CCR0, Rjvmti_mode, 0); 521 __ crorc(/*CCR0 eq*/2, /*CCR1 eq*/4+2, /*CCR0 eq*/2); 522 __ beq(CCR0, Lslow_path); // this==null or jvmti_mode!=0 523 524 // Do 2 things in parallel: 525 // 1. Load the index out of the first instruction word, which looks like this: 526 // <0x2a><0xb4><index (2 byte, native endianess)>. 527 // 2. Load constant pool cache base. 528 __ ld(Rconst_method, in_bytes(Method::const_offset()), R19_method); 529 __ ld(Rcpool_cache, in_bytes(ConstMethod::constants_offset()), Rconst_method); 530 531 __ lhz(Rcodes, in_bytes(ConstMethod::codes_offset()) + 2, Rconst_method); // Lower half of 32 bit field. 532 __ ld(Rcpool_cache, ConstantPool::cache_offset_in_bytes(), Rcpool_cache); 533 534 // Get the const pool entry by means of <index>. 535 const int codes_shift = exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord); 536 __ slwi(Rscratch, Rcodes, codes_shift); // (codes&0xFFFF)<<codes_shift 537 __ add(Rcpool_cache, Rscratch, Rcpool_cache); 538 539 // Check if cpool cache entry is resolved. 540 // We are resolved if the indices offset contains the current bytecode. 541 ByteSize cp_base_offset = ConstantPoolCache::base_offset(); 542 // Big Endian: 543 __ lbz(Rscratch, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::indices_offset()) + 7 - 2, Rcpool_cache); 544 __ cmpwi(CCR0, Rscratch, Bytecodes::_getfield); 545 __ bne(CCR0, Lslow_path); 546 __ isync(); // Order succeeding loads wrt. load of _indices field from cpool_cache. 547 548 // Finally, start loading the value: Get cp cache entry into regs. 549 __ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcpool_cache); 550 __ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcpool_cache); 551 552 // Following code is from templateTable::getfield_or_static 553 // Load pointer to branch table 554 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch); 555 556 // Get volatile flag 557 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // extract volatile bit 558 // note: sync is needed before volatile load on PPC64 559 560 // Check field type 561 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits); 562 563 #ifdef ASSERT 564 Label LFlagInvalid; 565 __ cmpldi(CCR0, Rflags, number_of_states); 566 __ bge(CCR0, LFlagInvalid); 567 568 __ ld(R9_ARG7, 0, R1_SP); 569 __ ld(R10_ARG8, 0, R21_sender_SP); 570 __ cmpd(CCR0, R9_ARG7, R10_ARG8); 571 __ asm_assert_eq("backlink", 0x543); 572 #endif // ASSERT 573 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 574 575 // Load from branch table and dispatch (volatile case: one instruction ahead) 576 __ sldi(Rflags, Rflags, LogBytesPerWord); 577 __ cmpwi(CCR6, Rscratch, 1); // volatile? 578 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { 579 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // volatile ? size of 1 instruction : 0 580 } 581 __ ldx(Rbtable, Rbtable, Rflags); 582 583 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { 584 __ subf(Rbtable, Rscratch, Rbtable); // point to volatile/non-volatile entry point 585 } 586 __ mtctr(Rbtable); 587 __ bctr(); 588 589 #ifdef ASSERT 590 __ bind(LFlagInvalid); 591 __ stop("got invalid flag", 0x6541); 592 593 bool all_uninitialized = true, 594 all_initialized = true; 595 for (int i = 0; i<number_of_states; ++i) { 596 all_uninitialized = all_uninitialized && (branch_table[i] == NULL); 597 all_initialized = all_initialized && (branch_table[i] != NULL); 598 } 599 assert(all_uninitialized != all_initialized, "consistency"); // either or 600 601 __ fence(); // volatile entry point (one instruction before non-volatile_entry point) 602 if (branch_table[vtos] == 0) branch_table[vtos] = __ pc(); // non-volatile_entry point 603 if (branch_table[dtos] == 0) branch_table[dtos] = __ pc(); // non-volatile_entry point 604 if (branch_table[ftos] == 0) branch_table[ftos] = __ pc(); // non-volatile_entry point 605 __ stop("unexpected type", 0x6551); 606 #endif 607 608 if (branch_table[itos] == 0) { // generate only once 609 __ align(32, 28, 28); // align load 610 __ fence(); // volatile entry point (one instruction before non-volatile_entry point) 611 branch_table[itos] = __ pc(); // non-volatile_entry point 612 __ lwax(R3_RET, Rclass_or_obj, Roffset); 613 __ beq(CCR6, Lacquire); 614 __ blr(); 615 } 616 617 if (branch_table[ltos] == 0) { // generate only once 618 __ align(32, 28, 28); // align load 619 __ fence(); // volatile entry point (one instruction before non-volatile_entry point) 620 branch_table[ltos] = __ pc(); // non-volatile_entry point 621 __ ldx(R3_RET, Rclass_or_obj, Roffset); 622 __ beq(CCR6, Lacquire); 623 __ blr(); 624 } 625 626 if (branch_table[btos] == 0) { // generate only once 627 __ align(32, 28, 28); // align load 628 __ fence(); // volatile entry point (one instruction before non-volatile_entry point) 629 branch_table[btos] = __ pc(); // non-volatile_entry point 630 __ lbzx(R3_RET, Rclass_or_obj, Roffset); 631 __ extsb(R3_RET, R3_RET); 632 __ beq(CCR6, Lacquire); 633 __ blr(); 634 } 635 636 if (branch_table[ctos] == 0) { // generate only once 637 __ align(32, 28, 28); // align load 638 __ fence(); // volatile entry point (one instruction before non-volatile_entry point) 639 branch_table[ctos] = __ pc(); // non-volatile_entry point 640 __ lhzx(R3_RET, Rclass_or_obj, Roffset); 641 __ beq(CCR6, Lacquire); 642 __ blr(); 643 } 644 645 if (branch_table[stos] == 0) { // generate only once 646 __ align(32, 28, 28); // align load 647 __ fence(); // volatile entry point (one instruction before non-volatile_entry point) 648 branch_table[stos] = __ pc(); // non-volatile_entry point 649 __ lhax(R3_RET, Rclass_or_obj, Roffset); 650 __ beq(CCR6, Lacquire); 651 __ blr(); 652 } 653 654 if (branch_table[atos] == 0) { // generate only once 655 __ align(32, 28, 28); // align load 656 __ fence(); // volatile entry point (one instruction before non-volatile_entry point) 657 branch_table[atos] = __ pc(); // non-volatile_entry point 658 __ load_heap_oop(R3_RET, (RegisterOrConstant)Roffset, Rclass_or_obj); 659 __ verify_oop(R3_RET); 660 //__ dcbt(R3_RET); // prefetch 661 __ beq(CCR6, Lacquire); 662 __ blr(); 663 } 664 665 __ align(32, 12); 666 __ bind(Lacquire); 667 __ twi_0(R3_RET); 668 __ isync(); // acquire 669 __ blr(); 670 671 #ifdef ASSERT 672 for (int i = 0; i<number_of_states; ++i) { 673 assert(branch_table[i], "accessor_entry initialization"); 674 //tty->print_cr("accessor_entry: branch_table[%d] = 0x%llx (opcode 0x%llx)", i, branch_table[i], *((unsigned int*)branch_table[i])); 675 } 676 #endif 677 678 __ bind(Lslow_path); 679 __ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), Rscratch); 680 __ flush(); 681 682 return entry; 683 } 684 685 // Interpreter intrinsic for WeakReference.get(). 686 // 1. Don't push a full blown frame and go on dispatching, but fetch the value 687 // into R8 and return quickly 688 // 2. If G1 is active we *must* execute this intrinsic for corrrectness: 689 // It contains a GC barrier which puts the reference into the satb buffer 690 // to indicate that someone holds a strong reference to the object the 691 // weak ref points to! 692 address InterpreterGenerator::generate_Reference_get_entry(void) { 693 // Code: _aload_0, _getfield, _areturn 694 // parameter size = 1 695 // 696 // The code that gets generated by this routine is split into 2 parts: 697 // 1. the "intrinsified" code for G1 (or any SATB based GC), 698 // 2. the slow path - which is an expansion of the regular method entry. 699 // 700 // Notes: 701 // * In the G1 code we do not check whether we need to block for 702 // a safepoint. If G1 is enabled then we must execute the specialized 703 // code for Reference.get (except when the Reference object is null) 704 // so that we can log the value in the referent field with an SATB 705 // update buffer. 706 // If the code for the getfield template is modified so that the 707 // G1 pre-barrier code is executed when the current method is 708 // Reference.get() then going through the normal method entry 709 // will be fine. 710 // * The G1 code can, however, check the receiver object (the instance 711 // of java.lang.Reference) and jump to the slow path if null. If the 712 // Reference object is null then we obviously cannot fetch the referent 713 // and so we don't need to call the G1 pre-barrier. Thus we can use the 714 // regular method entry code to generate the NPE. 715 // 716 // This code is based on generate_accessor_enty. 717 718 address entry = __ pc(); 719 720 const int referent_offset = java_lang_ref_Reference::referent_offset; 721 guarantee(referent_offset > 0, "referent offset not initialized"); 722 723 if (UseG1GC) { 724 Label slow_path; 725 726 // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH); 727 728 // In the G1 code we don't check if we need to reach a safepoint. We 729 // continue and the thread will safepoint at the next bytecode dispatch. 730 731 // If the receiver is null then it is OK to jump to the slow path. 732 __ ld(R3_RET, Interpreter::stackElementSize, CC_INTERP_ONLY(R17_tos) NOT_CC_INTERP(R15_esp)); // get receiver 733 734 // Check if receiver == NULL and go the slow path. 735 __ cmpdi(CCR0, R3_RET, 0); 736 __ beq(CCR0, slow_path); 737 738 // Load the value of the referent field. 739 __ load_heap_oop(R3_RET, referent_offset, R3_RET); 740 741 // Generate the G1 pre-barrier code to log the value of 742 // the referent field in an SATB buffer. Note with 743 // these parameters the pre-barrier does not generate 744 // the load of the previous value. 745 746 // Restore caller sp for c2i case. 747 #ifdef ASSERT 748 __ ld(R9_ARG7, 0, R1_SP); 749 __ ld(R10_ARG8, 0, R21_sender_SP); 750 __ cmpd(CCR0, R9_ARG7, R10_ARG8); 751 __ asm_assert_eq("backlink", 0x544); 752 #endif // ASSERT 753 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 754 755 __ g1_write_barrier_pre(noreg, // obj 756 noreg, // offset 757 R3_RET, // pre_val 758 R11_scratch1, // tmp 759 R12_scratch2, // tmp 760 true); // needs_frame 761 762 __ blr(); 763 764 // Generate regular method entry. 765 __ bind(slow_path); 766 __ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1); 767 __ flush(); 768 769 return entry; 770 } else { 771 return generate_accessor_entry(); 772 } 773 } 774 775 void Deoptimization::unwind_callee_save_values(frame* f, vframeArray* vframe_array) { 776 // This code is sort of the equivalent of C2IAdapter::setup_stack_frame back in 777 // the days we had adapter frames. When we deoptimize a situation where a 778 // compiled caller calls a compiled caller will have registers it expects 779 // to survive the call to the callee. If we deoptimize the callee the only 780 // way we can restore these registers is to have the oldest interpreter 781 // frame that we create restore these values. That is what this routine 782 // will accomplish. 783 784 // At the moment we have modified c2 to not have any callee save registers 785 // so this problem does not exist and this routine is just a place holder. 786 787 assert(f->is_interpreted_frame(), "must be interpreted"); 788 }