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