1 /* 2 * Copyright (c) 2014, 2015, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2015 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/macroAssembler.inline.hpp" 28 #include "interpreter/bytecodeHistogram.hpp" 29 #include "interpreter/interpreter.hpp" 30 #include "interpreter/interpreterRuntime.hpp" 31 #include "interpreter/interp_masm.hpp" 32 #include "interpreter/templateInterpreterGenerator.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 "runtime/arguments.hpp" 41 #include "runtime/deoptimization.hpp" 42 #include "runtime/frame.inline.hpp" 43 #include "runtime/sharedRuntime.hpp" 44 #include "runtime/stubRoutines.hpp" 45 #include "runtime/synchronizer.hpp" 46 #include "runtime/timer.hpp" 47 #include "runtime/vframeArray.hpp" 48 #include "utilities/debug.hpp" 49 #include "utilities/macros.hpp" 50 51 #undef __ 52 #define __ _masm-> 53 54 // Size of interpreter code. Increase if too small. Interpreter will 55 // fail with a guarantee ("not enough space for interpreter generation"); 56 // if too small. 57 // Run with +PrintInterpreter to get the VM to print out the size. 58 // Max size with JVMTI 59 int TemplateInterpreter::InterpreterCodeSize = 230*K; 60 61 #ifdef PRODUCT 62 #define BLOCK_COMMENT(str) /* nothing */ 63 #else 64 #define BLOCK_COMMENT(str) __ block_comment(str) 65 #endif 66 67 #define BIND(label) __ bind(label); BLOCK_COMMENT(#label ":") 68 69 //----------------------------------------------------------------------------- 70 71 address TemplateInterpreterGenerator::generate_slow_signature_handler() { 72 // Slow_signature handler that respects the PPC C calling conventions. 73 // 74 // We get called by the native entry code with our output register 75 // area == 8. First we call InterpreterRuntime::get_result_handler 76 // to copy the pointer to the signature string temporarily to the 77 // first C-argument and to return the result_handler in 78 // R3_RET. Since native_entry will copy the jni-pointer to the 79 // first C-argument slot later on, it is OK to occupy this slot 80 // temporarilly. Then we copy the argument list on the java 81 // expression stack into native varargs format on the native stack 82 // and load arguments into argument registers. Integer arguments in 83 // the varargs vector will be sign-extended to 8 bytes. 84 // 85 // On entry: 86 // R3_ARG1 - intptr_t* Address of java argument list in memory. 87 // R15_prev_state - BytecodeInterpreter* Address of interpreter state for 88 // this method 89 // R19_method 90 // 91 // On exit (just before return instruction): 92 // R3_RET - contains the address of the result_handler. 93 // R4_ARG2 - is not updated for static methods and contains "this" otherwise. 94 // R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double, 95 // ARGi contains this argument. Otherwise, ARGi is not updated. 96 // F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double. 97 98 const int LogSizeOfTwoInstructions = 3; 99 100 // FIXME: use Argument:: GL: Argument names different numbers! 101 const int max_fp_register_arguments = 13; 102 const int max_int_register_arguments = 6; // first 2 are reserved 103 104 const Register arg_java = R21_tmp1; 105 const Register arg_c = R22_tmp2; 106 const Register signature = R23_tmp3; // is string 107 const Register sig_byte = R24_tmp4; 108 const Register fpcnt = R25_tmp5; 109 const Register argcnt = R26_tmp6; 110 const Register intSlot = R27_tmp7; 111 const Register target_sp = R28_tmp8; 112 const FloatRegister floatSlot = F0; 113 114 address entry = __ function_entry(); 115 116 __ save_LR_CR(R0); 117 __ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14)); 118 // We use target_sp for storing arguments in the C frame. 119 __ mr(target_sp, R1_SP); 120 __ push_frame_reg_args_nonvolatiles(0, R11_scratch1); 121 122 __ mr(arg_java, R3_ARG1); 123 124 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method); 125 126 // Signature is in R3_RET. Signature is callee saved. 127 __ mr(signature, R3_RET); 128 129 // Get the result handler. 130 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method); 131 132 { 133 Label L; 134 // test if static 135 // _access_flags._flags must be at offset 0. 136 // TODO PPC port: requires change in shared code. 137 //assert(in_bytes(AccessFlags::flags_offset()) == 0, 138 // "MethodDesc._access_flags == MethodDesc._access_flags._flags"); 139 // _access_flags must be a 32 bit value. 140 assert(sizeof(AccessFlags) == 4, "wrong size"); 141 __ lwa(R11_scratch1/*access_flags*/, method_(access_flags)); 142 // testbit with condition register. 143 __ testbitdi(CCR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT); 144 __ btrue(CCR0, L); 145 // For non-static functions, pass "this" in R4_ARG2 and copy it 146 // to 2nd C-arg slot. 147 // We need to box the Java object here, so we use arg_java 148 // (address of current Java stack slot) as argument and don't 149 // dereference it as in case of ints, floats, etc. 150 __ mr(R4_ARG2, arg_java); 151 __ addi(arg_java, arg_java, -BytesPerWord); 152 __ std(R4_ARG2, _abi(carg_2), target_sp); 153 __ bind(L); 154 } 155 156 // Will be incremented directly after loop_start. argcnt=0 157 // corresponds to 3rd C argument. 158 __ li(argcnt, -1); 159 // arg_c points to 3rd C argument 160 __ addi(arg_c, target_sp, _abi(carg_3)); 161 // no floating-point args parsed so far 162 __ li(fpcnt, 0); 163 164 Label move_intSlot_to_ARG, move_floatSlot_to_FARG; 165 Label loop_start, loop_end; 166 Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed; 167 168 // signature points to '(' at entry 169 #ifdef ASSERT 170 __ lbz(sig_byte, 0, signature); 171 __ cmplwi(CCR0, sig_byte, '('); 172 __ bne(CCR0, do_dontreachhere); 173 #endif 174 175 __ bind(loop_start); 176 177 __ addi(argcnt, argcnt, 1); 178 __ lbzu(sig_byte, 1, signature); 179 180 __ cmplwi(CCR0, sig_byte, ')'); // end of signature 181 __ beq(CCR0, loop_end); 182 183 __ cmplwi(CCR0, sig_byte, 'B'); // byte 184 __ beq(CCR0, do_int); 185 186 __ cmplwi(CCR0, sig_byte, 'C'); // char 187 __ beq(CCR0, do_int); 188 189 __ cmplwi(CCR0, sig_byte, 'D'); // double 190 __ beq(CCR0, do_double); 191 192 __ cmplwi(CCR0, sig_byte, 'F'); // float 193 __ beq(CCR0, do_float); 194 195 __ cmplwi(CCR0, sig_byte, 'I'); // int 196 __ beq(CCR0, do_int); 197 198 __ cmplwi(CCR0, sig_byte, 'J'); // long 199 __ beq(CCR0, do_long); 200 201 __ cmplwi(CCR0, sig_byte, 'S'); // short 202 __ beq(CCR0, do_int); 203 204 __ cmplwi(CCR0, sig_byte, 'Z'); // boolean 205 __ beq(CCR0, do_int); 206 207 __ cmplwi(CCR0, sig_byte, 'L'); // object 208 __ beq(CCR0, do_object); 209 210 __ cmplwi(CCR0, sig_byte, '['); // array 211 __ beq(CCR0, do_array); 212 213 // __ cmplwi(CCR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type 214 // __ beq(CCR0, do_void); 215 216 __ bind(do_dontreachhere); 217 218 __ unimplemented("ShouldNotReachHere in slow_signature_handler", 120); 219 220 __ bind(do_array); 221 222 { 223 Label start_skip, end_skip; 224 225 __ bind(start_skip); 226 __ lbzu(sig_byte, 1, signature); 227 __ cmplwi(CCR0, sig_byte, '['); 228 __ beq(CCR0, start_skip); // skip further brackets 229 __ cmplwi(CCR0, sig_byte, '9'); 230 __ bgt(CCR0, end_skip); // no optional size 231 __ cmplwi(CCR0, sig_byte, '0'); 232 __ bge(CCR0, start_skip); // skip optional size 233 __ bind(end_skip); 234 235 __ cmplwi(CCR0, sig_byte, 'L'); 236 __ beq(CCR0, do_object); // for arrays of objects, the name of the object must be skipped 237 __ b(do_boxed); // otherwise, go directly to do_boxed 238 } 239 240 __ bind(do_object); 241 { 242 Label L; 243 __ bind(L); 244 __ lbzu(sig_byte, 1, signature); 245 __ cmplwi(CCR0, sig_byte, ';'); 246 __ bne(CCR0, L); 247 } 248 // Need to box the Java object here, so we use arg_java (address of 249 // current Java stack slot) as argument and don't dereference it as 250 // in case of ints, floats, etc. 251 Label do_null; 252 __ bind(do_boxed); 253 __ ld(R0,0, arg_java); 254 __ cmpdi(CCR0, R0, 0); 255 __ li(intSlot,0); 256 __ beq(CCR0, do_null); 257 __ mr(intSlot, arg_java); 258 __ bind(do_null); 259 __ std(intSlot, 0, arg_c); 260 __ addi(arg_java, arg_java, -BytesPerWord); 261 __ addi(arg_c, arg_c, BytesPerWord); 262 __ cmplwi(CCR0, argcnt, max_int_register_arguments); 263 __ blt(CCR0, move_intSlot_to_ARG); 264 __ b(loop_start); 265 266 __ bind(do_int); 267 __ lwa(intSlot, 0, arg_java); 268 __ std(intSlot, 0, arg_c); 269 __ addi(arg_java, arg_java, -BytesPerWord); 270 __ addi(arg_c, arg_c, BytesPerWord); 271 __ cmplwi(CCR0, argcnt, max_int_register_arguments); 272 __ blt(CCR0, move_intSlot_to_ARG); 273 __ b(loop_start); 274 275 __ bind(do_long); 276 __ ld(intSlot, -BytesPerWord, arg_java); 277 __ std(intSlot, 0, arg_c); 278 __ addi(arg_java, arg_java, - 2 * BytesPerWord); 279 __ addi(arg_c, arg_c, BytesPerWord); 280 __ cmplwi(CCR0, argcnt, max_int_register_arguments); 281 __ blt(CCR0, move_intSlot_to_ARG); 282 __ b(loop_start); 283 284 __ bind(do_float); 285 __ lfs(floatSlot, 0, arg_java); 286 #if defined(LINUX) 287 // Linux uses ELF ABI. Both original ELF and ELFv2 ABIs have float 288 // in the least significant word of an argument slot. 289 #if defined(VM_LITTLE_ENDIAN) 290 __ stfs(floatSlot, 0, arg_c); 291 #else 292 __ stfs(floatSlot, 4, arg_c); 293 #endif 294 #elif defined(AIX) 295 // Although AIX runs on big endian CPU, float is in most significant 296 // word of an argument slot. 297 __ stfs(floatSlot, 0, arg_c); 298 #else 299 #error "unknown OS" 300 #endif 301 __ addi(arg_java, arg_java, -BytesPerWord); 302 __ addi(arg_c, arg_c, BytesPerWord); 303 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments); 304 __ blt(CCR0, move_floatSlot_to_FARG); 305 __ b(loop_start); 306 307 __ bind(do_double); 308 __ lfd(floatSlot, - BytesPerWord, arg_java); 309 __ stfd(floatSlot, 0, arg_c); 310 __ addi(arg_java, arg_java, - 2 * BytesPerWord); 311 __ addi(arg_c, arg_c, BytesPerWord); 312 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments); 313 __ blt(CCR0, move_floatSlot_to_FARG); 314 __ b(loop_start); 315 316 __ bind(loop_end); 317 318 __ pop_frame(); 319 __ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14)); 320 __ restore_LR_CR(R0); 321 322 __ blr(); 323 324 Label move_int_arg, move_float_arg; 325 __ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions) 326 __ mr(R5_ARG3, intSlot); __ b(loop_start); 327 __ mr(R6_ARG4, intSlot); __ b(loop_start); 328 __ mr(R7_ARG5, intSlot); __ b(loop_start); 329 __ mr(R8_ARG6, intSlot); __ b(loop_start); 330 __ mr(R9_ARG7, intSlot); __ b(loop_start); 331 __ mr(R10_ARG8, intSlot); __ b(loop_start); 332 333 __ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions) 334 __ fmr(F1_ARG1, floatSlot); __ b(loop_start); 335 __ fmr(F2_ARG2, floatSlot); __ b(loop_start); 336 __ fmr(F3_ARG3, floatSlot); __ b(loop_start); 337 __ fmr(F4_ARG4, floatSlot); __ b(loop_start); 338 __ fmr(F5_ARG5, floatSlot); __ b(loop_start); 339 __ fmr(F6_ARG6, floatSlot); __ b(loop_start); 340 __ fmr(F7_ARG7, floatSlot); __ b(loop_start); 341 __ fmr(F8_ARG8, floatSlot); __ b(loop_start); 342 __ fmr(F9_ARG9, floatSlot); __ b(loop_start); 343 __ fmr(F10_ARG10, floatSlot); __ b(loop_start); 344 __ fmr(F11_ARG11, floatSlot); __ b(loop_start); 345 __ fmr(F12_ARG12, floatSlot); __ b(loop_start); 346 __ fmr(F13_ARG13, floatSlot); __ b(loop_start); 347 348 __ bind(move_intSlot_to_ARG); 349 __ sldi(R0, argcnt, LogSizeOfTwoInstructions); 350 __ load_const(R11_scratch1, move_int_arg); // Label must be bound here. 351 __ add(R11_scratch1, R0, R11_scratch1); 352 __ mtctr(R11_scratch1/*branch_target*/); 353 __ bctr(); 354 __ bind(move_floatSlot_to_FARG); 355 __ sldi(R0, fpcnt, LogSizeOfTwoInstructions); 356 __ addi(fpcnt, fpcnt, 1); 357 __ load_const(R11_scratch1, move_float_arg); // Label must be bound here. 358 __ add(R11_scratch1, R0, R11_scratch1); 359 __ mtctr(R11_scratch1/*branch_target*/); 360 __ bctr(); 361 362 return entry; 363 } 364 365 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) { 366 // 367 // Registers alive 368 // R3_RET 369 // LR 370 // 371 // Registers updated 372 // R3_RET 373 // 374 375 Label done; 376 address entry = __ pc(); 377 378 switch (type) { 379 case T_BOOLEAN: 380 // convert !=0 to 1 381 __ neg(R0, R3_RET); 382 __ orr(R0, R3_RET, R0); 383 __ srwi(R3_RET, R0, 31); 384 break; 385 case T_BYTE: 386 // sign extend 8 bits 387 __ extsb(R3_RET, R3_RET); 388 break; 389 case T_CHAR: 390 // zero extend 16 bits 391 __ clrldi(R3_RET, R3_RET, 48); 392 break; 393 case T_SHORT: 394 // sign extend 16 bits 395 __ extsh(R3_RET, R3_RET); 396 break; 397 case T_INT: 398 // sign extend 32 bits 399 __ extsw(R3_RET, R3_RET); 400 break; 401 case T_LONG: 402 break; 403 case T_OBJECT: 404 // unbox result if not null 405 __ cmpdi(CCR0, R3_RET, 0); 406 __ beq(CCR0, done); 407 __ ld(R3_RET, 0, R3_RET); 408 __ verify_oop(R3_RET); 409 break; 410 case T_FLOAT: 411 break; 412 case T_DOUBLE: 413 break; 414 case T_VOID: 415 break; 416 default: ShouldNotReachHere(); 417 } 418 419 __ BIND(done); 420 __ blr(); 421 422 return entry; 423 } 424 425 // Abstract method entry. 426 // 427 address TemplateInterpreterGenerator::generate_abstract_entry(void) { 428 address entry = __ pc(); 429 430 // 431 // Registers alive 432 // R16_thread - JavaThread* 433 // R19_method - callee's method (method to be invoked) 434 // R1_SP - SP prepared such that caller's outgoing args are near top 435 // LR - return address to caller 436 // 437 // Stack layout at this point: 438 // 439 // 0 [TOP_IJAVA_FRAME_ABI] <-- R1_SP 440 // alignment (optional) 441 // [outgoing Java arguments] 442 // ... 443 // PARENT [PARENT_IJAVA_FRAME_ABI] 444 // ... 445 // 446 447 // Can't use call_VM here because we have not set up a new 448 // interpreter state. Make the call to the vm and make it look like 449 // our caller set up the JavaFrameAnchor. 450 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/); 451 452 // Push a new C frame and save LR. 453 __ save_LR_CR(R0); 454 __ push_frame_reg_args(0, R11_scratch1); 455 456 // This is not a leaf but we have a JavaFrameAnchor now and we will 457 // check (create) exceptions afterward so this is ok. 458 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError), 459 R16_thread); 460 461 // Pop the C frame and restore LR. 462 __ pop_frame(); 463 __ restore_LR_CR(R0); 464 465 // Reset JavaFrameAnchor from call_VM_leaf above. 466 __ reset_last_Java_frame(); 467 468 // We don't know our caller, so jump to the general forward exception stub, 469 // which will also pop our full frame off. Satisfy the interface of 470 // SharedRuntime::generate_forward_exception() 471 __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0); 472 __ mtctr(R11_scratch1); 473 __ bctr(); 474 475 return entry; 476 } 477 478 // Interpreter intrinsic for WeakReference.get(). 479 // 1. Don't push a full blown frame and go on dispatching, but fetch the value 480 // into R8 and return quickly 481 // 2. If G1 is active we *must* execute this intrinsic for corrrectness: 482 // It contains a GC barrier which puts the reference into the satb buffer 483 // to indicate that someone holds a strong reference to the object the 484 // weak ref points to! 485 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) { 486 // Code: _aload_0, _getfield, _areturn 487 // parameter size = 1 488 // 489 // The code that gets generated by this routine is split into 2 parts: 490 // 1. the "intrinsified" code for G1 (or any SATB based GC), 491 // 2. the slow path - which is an expansion of the regular method entry. 492 // 493 // Notes: 494 // * In the G1 code we do not check whether we need to block for 495 // a safepoint. If G1 is enabled then we must execute the specialized 496 // code for Reference.get (except when the Reference object is null) 497 // so that we can log the value in the referent field with an SATB 498 // update buffer. 499 // If the code for the getfield template is modified so that the 500 // G1 pre-barrier code is executed when the current method is 501 // Reference.get() then going through the normal method entry 502 // will be fine. 503 // * The G1 code can, however, check the receiver object (the instance 504 // of java.lang.Reference) and jump to the slow path if null. If the 505 // Reference object is null then we obviously cannot fetch the referent 506 // and so we don't need to call the G1 pre-barrier. Thus we can use the 507 // regular method entry code to generate the NPE. 508 // 509 510 if (UseG1GC) { 511 address entry = __ pc(); 512 513 const int referent_offset = java_lang_ref_Reference::referent_offset; 514 guarantee(referent_offset > 0, "referent offset not initialized"); 515 516 Label slow_path; 517 518 // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH); 519 520 // In the G1 code we don't check if we need to reach a safepoint. We 521 // continue and the thread will safepoint at the next bytecode dispatch. 522 523 // If the receiver is null then it is OK to jump to the slow path. 524 __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver 525 526 // Check if receiver == NULL and go the slow path. 527 __ cmpdi(CCR0, R3_RET, 0); 528 __ beq(CCR0, slow_path); 529 530 // Load the value of the referent field. 531 __ load_heap_oop(R3_RET, referent_offset, R3_RET); 532 533 // Generate the G1 pre-barrier code to log the value of 534 // the referent field in an SATB buffer. Note with 535 // these parameters the pre-barrier does not generate 536 // the load of the previous value. 537 538 // Restore caller sp for c2i case. 539 #ifdef ASSERT 540 __ ld(R9_ARG7, 0, R1_SP); 541 __ ld(R10_ARG8, 0, R21_sender_SP); 542 __ cmpd(CCR0, R9_ARG7, R10_ARG8); 543 __ asm_assert_eq("backlink", 0x544); 544 #endif // ASSERT 545 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 546 547 __ g1_write_barrier_pre(noreg, // obj 548 noreg, // offset 549 R3_RET, // pre_val 550 R11_scratch1, // tmp 551 R12_scratch2, // tmp 552 true); // needs_frame 553 554 __ blr(); 555 556 // Generate regular method entry. 557 __ bind(slow_path); 558 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1); 559 return entry; 560 } 561 562 return NULL; 563 } 564 565 // Actually we should never reach here since we do stack overflow checks before pushing any frame. 566 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { 567 address entry = __ pc(); 568 __ unimplemented("generate_StackOverflowError_handler"); 569 return entry; 570 } 571 572 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) { 573 address entry = __ pc(); 574 __ empty_expression_stack(); 575 __ load_const_optimized(R4_ARG2, (address) name); 576 // Index is in R17_tos. 577 __ mr(R5_ARG3, R17_tos); 578 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException)); 579 return entry; 580 } 581 582 #if 0 583 // Call special ClassCastException constructor taking object to cast 584 // and target class as arguments. 585 address TemplateInterpreterGenerator::generate_ClassCastException_verbose_handler() { 586 address entry = __ pc(); 587 588 // Expression stack must be empty before entering the VM if an 589 // exception happened. 590 __ empty_expression_stack(); 591 592 // Thread will be loaded to R3_ARG1. 593 // Target class oop is in register R5_ARG3 by convention! 594 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException_verbose), R17_tos, R5_ARG3); 595 // Above call must not return here since exception pending. 596 DEBUG_ONLY(__ should_not_reach_here();) 597 return entry; 598 } 599 #endif 600 601 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 602 address entry = __ pc(); 603 // Expression stack must be empty before entering the VM if an 604 // exception happened. 605 __ empty_expression_stack(); 606 607 // Load exception object. 608 // Thread will be loaded to R3_ARG1. 609 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos); 610 #ifdef ASSERT 611 // Above call must not return here since exception pending. 612 __ should_not_reach_here(); 613 #endif 614 return entry; 615 } 616 617 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) { 618 address entry = __ pc(); 619 //__ untested("generate_exception_handler_common"); 620 Register Rexception = R17_tos; 621 622 // Expression stack must be empty before entering the VM if an exception happened. 623 __ empty_expression_stack(); 624 625 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1); 626 if (pass_oop) { 627 __ mr(R5_ARG3, Rexception); 628 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), false); 629 } else { 630 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1); 631 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), false); 632 } 633 634 // Throw exception. 635 __ mr(R3_ARG1, Rexception); 636 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2); 637 __ mtctr(R11_scratch1); 638 __ bctr(); 639 640 return entry; 641 } 642 643 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) { 644 address entry = __ pc(); 645 __ unimplemented("generate_continuation_for"); 646 return entry; 647 } 648 649 // This entry is returned to when a call returns to the interpreter. 650 // When we arrive here, we expect that the callee stack frame is already popped. 651 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) { 652 address entry = __ pc(); 653 654 // Move the value out of the return register back to the TOS cache of current frame. 655 switch (state) { 656 case ltos: 657 case btos: 658 case ctos: 659 case stos: 660 case atos: 661 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache 662 case ftos: 663 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET 664 case vtos: break; // Nothing to do, this was a void return. 665 default : ShouldNotReachHere(); 666 } 667 668 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp. 669 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); 670 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); 671 672 // Compiled code destroys templateTableBase, reload. 673 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2); 674 675 if (state == atos) { 676 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2); 677 } 678 679 const Register cache = R11_scratch1; 680 const Register size = R12_scratch2; 681 __ get_cache_and_index_at_bcp(cache, 1, index_size); 682 683 // Get least significant byte of 64 bit value: 684 #if defined(VM_LITTLE_ENDIAN) 685 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache); 686 #else 687 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache); 688 #endif 689 __ sldi(size, size, Interpreter::logStackElementSize); 690 __ add(R15_esp, R15_esp, size); 691 __ dispatch_next(state, step); 692 return entry; 693 } 694 695 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) { 696 address entry = __ pc(); 697 // If state != vtos, we're returning from a native method, which put it's result 698 // into the result register. So move the value out of the return register back 699 // to the TOS cache of current frame. 700 701 switch (state) { 702 case ltos: 703 case btos: 704 case ctos: 705 case stos: 706 case atos: 707 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache 708 case ftos: 709 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET 710 case vtos: break; // Nothing to do, this was a void return. 711 default : ShouldNotReachHere(); 712 } 713 714 // Load LcpoolCache @@@ should be already set! 715 __ get_constant_pool_cache(R27_constPoolCache); 716 717 // Handle a pending exception, fall through if none. 718 __ check_and_forward_exception(R11_scratch1, R12_scratch2); 719 720 // Start executing bytecodes. 721 __ dispatch_next(state, step); 722 723 return entry; 724 } 725 726 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) { 727 address entry = __ pc(); 728 729 __ push(state); 730 __ call_VM(noreg, runtime_entry); 731 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); 732 733 return entry; 734 } 735 736 // Helpers for commoning out cases in the various type of method entries. 737 738 // Increment invocation count & check for overflow. 739 // 740 // Note: checking for negative value instead of overflow 741 // so we have a 'sticky' overflow test. 742 // 743 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) { 744 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not. 745 Register Rscratch1 = R11_scratch1; 746 Register Rscratch2 = R12_scratch2; 747 Register R3_counters = R3_ARG1; 748 Label done; 749 750 if (TieredCompilation) { 751 const int increment = InvocationCounter::count_increment; 752 const int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift; 753 Label no_mdo; 754 if (ProfileInterpreter) { 755 const Register Rmdo = Rscratch1; 756 // If no method data exists, go to profile_continue. 757 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method); 758 __ cmpdi(CCR0, Rmdo, 0); 759 __ beq(CCR0, no_mdo); 760 761 // Increment backedge counter in the MDO. 762 const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset()); 763 __ lwz(Rscratch2, mdo_bc_offs, Rmdo); 764 __ addi(Rscratch2, Rscratch2, increment); 765 __ stw(Rscratch2, mdo_bc_offs, Rmdo); 766 __ load_const_optimized(Rscratch1, mask, R0); 767 __ and_(Rscratch1, Rscratch2, Rscratch1); 768 __ bne(CCR0, done); 769 __ b(*overflow); 770 } 771 772 // Increment counter in MethodCounters*. 773 const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset()); 774 __ bind(no_mdo); 775 __ get_method_counters(R19_method, R3_counters, done); 776 __ lwz(Rscratch2, mo_bc_offs, R3_counters); 777 __ addi(Rscratch2, Rscratch2, increment); 778 __ stw(Rscratch2, mo_bc_offs, R3_counters); 779 __ load_const_optimized(Rscratch1, mask, R0); 780 __ and_(Rscratch1, Rscratch2, Rscratch1); 781 __ beq(CCR0, *overflow); 782 783 __ bind(done); 784 785 } else { 786 787 // Update standard invocation counters. 788 Register Rsum_ivc_bec = R4_ARG2; 789 __ get_method_counters(R19_method, R3_counters, done); 790 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2); 791 // Increment interpreter invocation counter. 792 if (ProfileInterpreter) { // %%% Merge this into methodDataOop. 793 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters); 794 __ addi(R12_scratch2, R12_scratch2, 1); 795 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters); 796 } 797 // Check if we must create a method data obj. 798 if (ProfileInterpreter && profile_method != NULL) { 799 const Register profile_limit = Rscratch1; 800 int pl_offs = __ load_const_optimized(profile_limit, &InvocationCounter::InterpreterProfileLimit, R0, true); 801 __ lwz(profile_limit, pl_offs, profile_limit); 802 // Test to see if we should create a method data oop. 803 __ cmpw(CCR0, Rsum_ivc_bec, profile_limit); 804 __ blt(CCR0, *profile_method_continue); 805 // If no method data exists, go to profile_method. 806 __ test_method_data_pointer(*profile_method); 807 } 808 // Finally check for counter overflow. 809 if (overflow) { 810 const Register invocation_limit = Rscratch1; 811 int il_offs = __ load_const_optimized(invocation_limit, &InvocationCounter::InterpreterInvocationLimit, R0, true); 812 __ lwz(invocation_limit, il_offs, invocation_limit); 813 assert(4 == sizeof(InvocationCounter::InterpreterInvocationLimit), "unexpected field size"); 814 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit); 815 __ bge(CCR0, *overflow); 816 } 817 818 __ bind(done); 819 } 820 } 821 822 // Generate code to initiate compilation on invocation counter overflow. 823 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) { 824 // Generate code to initiate compilation on the counter overflow. 825 826 // InterpreterRuntime::frequency_counter_overflow takes one arguments, 827 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp) 828 // We pass zero in. 829 // The call returns the address of the verified entry point for the method or NULL 830 // if the compilation did not complete (either went background or bailed out). 831 // 832 // Unlike the C++ interpreter above: Check exceptions! 833 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed 834 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur. 835 836 __ li(R4_ARG2, 0); 837 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true); 838 839 // Returns verified_entry_point or NULL. 840 // We ignore it in any case. 841 __ b(continue_entry); 842 } 843 844 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) { 845 assert_different_registers(Rmem_frame_size, Rscratch1); 846 __ generate_stack_overflow_check_with_compare_and_throw(Rmem_frame_size, Rscratch1); 847 } 848 849 void TemplateInterpreterGenerator::unlock_method(bool check_exceptions) { 850 __ unlock_object(R26_monitor, check_exceptions); 851 } 852 853 // Lock the current method, interpreter register window must be set up! 854 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) { 855 const Register Robj_to_lock = Rscratch2; 856 857 { 858 if (!flags_preloaded) { 859 __ lwz(Rflags, method_(access_flags)); 860 } 861 862 #ifdef ASSERT 863 // Check if methods needs synchronization. 864 { 865 Label Lok; 866 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT); 867 __ btrue(CCR0,Lok); 868 __ stop("method doesn't need synchronization"); 869 __ bind(Lok); 870 } 871 #endif // ASSERT 872 } 873 874 // Get synchronization object to Rscratch2. 875 { 876 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 877 Label Lstatic; 878 Label Ldone; 879 880 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT); 881 __ btrue(CCR0, Lstatic); 882 883 // Non-static case: load receiver obj from stack and we're done. 884 __ ld(Robj_to_lock, R18_locals); 885 __ b(Ldone); 886 887 __ bind(Lstatic); // Static case: Lock the java mirror 888 __ ld(Robj_to_lock, in_bytes(Method::const_offset()), R19_method); 889 __ ld(Robj_to_lock, in_bytes(ConstMethod::constants_offset()), Robj_to_lock); 890 __ ld(Robj_to_lock, ConstantPool::pool_holder_offset_in_bytes(), Robj_to_lock); 891 __ ld(Robj_to_lock, mirror_offset, Robj_to_lock); 892 893 __ bind(Ldone); 894 __ verify_oop(Robj_to_lock); 895 } 896 897 // Got the oop to lock => execute! 898 __ add_monitor_to_stack(true, Rscratch1, R0); 899 900 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor); 901 __ lock_object(R26_monitor, Robj_to_lock); 902 } 903 904 // Generate a fixed interpreter frame for pure interpreter 905 // and I2N native transition frames. 906 // 907 // Before (stack grows downwards): 908 // 909 // | ... | 910 // |------------- | 911 // | java arg0 | 912 // | ... | 913 // | java argn | 914 // | | <- R15_esp 915 // | | 916 // |--------------| 917 // | abi_112 | 918 // | | <- R1_SP 919 // |==============| 920 // 921 // 922 // After: 923 // 924 // | ... | 925 // | java arg0 |<- R18_locals 926 // | ... | 927 // | java argn | 928 // |--------------| 929 // | | 930 // | java locals | 931 // | | 932 // |--------------| 933 // | abi_48 | 934 // |==============| 935 // | | 936 // | istate | 937 // | | 938 // |--------------| 939 // | monitor |<- R26_monitor 940 // |--------------| 941 // | |<- R15_esp 942 // | expression | 943 // | stack | 944 // | | 945 // |--------------| 946 // | | 947 // | abi_112 |<- R1_SP 948 // |==============| 949 // 950 // The top most frame needs an abi space of 112 bytes. This space is needed, 951 // since we call to c. The c function may spill their arguments to the caller 952 // frame. When we call to java, we don't need these spill slots. In order to save 953 // space on the stack, we resize the caller. However, java local reside in 954 // the caller frame and the frame has to be increased. The frame_size for the 955 // current frame was calculated based on max_stack as size for the expression 956 // stack. At the call, just a part of the expression stack might be used. 957 // We don't want to waste this space and cut the frame back accordingly. 958 // The resulting amount for resizing is calculated as follows: 959 // resize = (number_of_locals - number_of_arguments) * slot_size 960 // + (R1_SP - R15_esp) + 48 961 // 962 // The size for the callee frame is calculated: 963 // framesize = 112 + max_stack + monitor + state_size 964 // 965 // maxstack: Max number of slots on the expression stack, loaded from the method. 966 // monitor: We statically reserve room for one monitor object. 967 // state_size: We save the current state of the interpreter to this area. 968 // 969 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) { 970 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes. 971 top_frame_size = R7_ARG5, 972 Rconst_method = R8_ARG6; 973 974 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size); 975 976 __ ld(Rconst_method, method_(const)); 977 __ lhz(Rsize_of_parameters /* number of params */, 978 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method); 979 if (native_call) { 980 // If we're calling a native method, we reserve space for the worst-case signature 981 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2). 982 // We add two slots to the parameter_count, one for the jni 983 // environment and one for a possible native mirror. 984 Label skip_native_calculate_max_stack; 985 __ addi(top_frame_size, Rsize_of_parameters, 2); 986 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters); 987 __ bge(CCR0, skip_native_calculate_max_stack); 988 __ li(top_frame_size, Argument::n_register_parameters); 989 __ bind(skip_native_calculate_max_stack); 990 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize); 991 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize); 992 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize! 993 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters. 994 } else { 995 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method); 996 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize); 997 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize); 998 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method); 999 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0 1000 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize! 1001 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize); 1002 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1); 1003 } 1004 1005 // Compute top frame size. 1006 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size); 1007 1008 // Cut back area between esp and max_stack. 1009 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize); 1010 1011 __ round_to(top_frame_size, frame::alignment_in_bytes); 1012 __ round_to(parent_frame_resize, frame::alignment_in_bytes); 1013 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size. 1014 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48. 1015 1016 { 1017 // -------------------------------------------------------------------------- 1018 // Stack overflow check 1019 1020 Label cont; 1021 __ add(R11_scratch1, parent_frame_resize, top_frame_size); 1022 generate_stack_overflow_check(R11_scratch1, R12_scratch2); 1023 } 1024 1025 // Set up interpreter state registers. 1026 1027 __ add(R18_locals, R15_esp, Rsize_of_parameters); 1028 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method); 1029 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache); 1030 1031 // Set method data pointer. 1032 if (ProfileInterpreter) { 1033 Label zero_continue; 1034 __ ld(R28_mdx, method_(method_data)); 1035 __ cmpdi(CCR0, R28_mdx, 0); 1036 __ beq(CCR0, zero_continue); 1037 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1038 __ bind(zero_continue); 1039 } 1040 1041 if (native_call) { 1042 __ li(R14_bcp, 0); // Must initialize. 1043 } else { 1044 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method); 1045 } 1046 1047 // Resize parent frame. 1048 __ mflr(R12_scratch2); 1049 __ neg(parent_frame_resize, parent_frame_resize); 1050 __ resize_frame(parent_frame_resize, R11_scratch1); 1051 __ std(R12_scratch2, _abi(lr), R1_SP); 1052 1053 __ addi(R26_monitor, R1_SP, - frame::ijava_state_size); 1054 __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); 1055 1056 // Store values. 1057 // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls 1058 // in InterpreterMacroAssembler::call_from_interpreter. 1059 __ std(R19_method, _ijava_state_neg(method), R1_SP); 1060 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP); 1061 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP); 1062 __ std(R18_locals, _ijava_state_neg(locals), R1_SP); 1063 1064 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only 1065 // be found in the frame after save_interpreter_state is done. This is always true 1066 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong, 1067 // because e.g. frame::interpreter_frame_bcp() will not access the correct value 1068 // (Enhanced Stack Trace). 1069 // The signal handler does not save the interpreter state into the frame. 1070 __ li(R0, 0); 1071 #ifdef ASSERT 1072 // Fill remaining slots with constants. 1073 __ load_const_optimized(R11_scratch1, 0x5afe); 1074 __ load_const_optimized(R12_scratch2, 0xdead); 1075 #endif 1076 // We have to initialize some frame slots for native calls (accessed by GC). 1077 if (native_call) { 1078 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP); 1079 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP); 1080 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); } 1081 } 1082 #ifdef ASSERT 1083 else { 1084 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP); 1085 __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP); 1086 __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP); 1087 } 1088 __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP); 1089 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP); 1090 __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP); 1091 __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP); 1092 #endif 1093 __ subf(R12_scratch2, top_frame_size, R1_SP); 1094 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); 1095 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP); 1096 1097 // Push top frame. 1098 __ push_frame(top_frame_size, R11_scratch1); 1099 } 1100 1101 // End of helpers 1102 1103 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) { 1104 if (!TemplateInterpreter::math_entry_available(kind)) { 1105 NOT_PRODUCT(__ should_not_reach_here();) 1106 return NULL; 1107 } 1108 1109 address entry = __ pc(); 1110 1111 __ lfd(F1_RET, Interpreter::stackElementSize, R15_esp); 1112 1113 // Pop c2i arguments (if any) off when we return. 1114 #ifdef ASSERT 1115 __ ld(R9_ARG7, 0, R1_SP); 1116 __ ld(R10_ARG8, 0, R21_sender_SP); 1117 __ cmpd(CCR0, R9_ARG7, R10_ARG8); 1118 __ asm_assert_eq("backlink", 0x545); 1119 #endif // ASSERT 1120 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 1121 1122 if (kind == Interpreter::java_lang_math_sqrt) { 1123 __ fsqrt(F1_RET, F1_RET); 1124 } else if (kind == Interpreter::java_lang_math_abs) { 1125 __ fabs(F1_RET, F1_RET); 1126 } else { 1127 ShouldNotReachHere(); 1128 } 1129 1130 // And we're done. 1131 __ blr(); 1132 1133 __ flush(); 1134 1135 return entry; 1136 } 1137 1138 // Interpreter stub for calling a native method. (asm interpreter) 1139 // This sets up a somewhat different looking stack for calling the 1140 // native method than the typical interpreter frame setup. 1141 // 1142 // On entry: 1143 // R19_method - method 1144 // R16_thread - JavaThread* 1145 // R15_esp - intptr_t* sender tos 1146 // 1147 // abstract stack (grows up) 1148 // [ IJava (caller of JNI callee) ] <-- ASP 1149 // ... 1150 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) { 1151 1152 address entry = __ pc(); 1153 1154 const bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; 1155 1156 // ----------------------------------------------------------------------------- 1157 // Allocate a new frame that represents the native callee (i2n frame). 1158 // This is not a full-blown interpreter frame, but in particular, the 1159 // following registers are valid after this: 1160 // - R19_method 1161 // - R18_local (points to start of argumuments to native function) 1162 // 1163 // abstract stack (grows up) 1164 // [ IJava (caller of JNI callee) ] <-- ASP 1165 // ... 1166 1167 const Register signature_handler_fd = R11_scratch1; 1168 const Register pending_exception = R0; 1169 const Register result_handler_addr = R31; 1170 const Register native_method_fd = R11_scratch1; 1171 const Register access_flags = R22_tmp2; 1172 const Register active_handles = R11_scratch1; // R26_monitor saved to state. 1173 const Register sync_state = R12_scratch2; 1174 const Register sync_state_addr = sync_state; // Address is dead after use. 1175 const Register suspend_flags = R11_scratch1; 1176 1177 //============================================================================= 1178 // Allocate new frame and initialize interpreter state. 1179 1180 Label exception_return; 1181 Label exception_return_sync_check; 1182 Label stack_overflow_return; 1183 1184 // Generate new interpreter state and jump to stack_overflow_return in case of 1185 // a stack overflow. 1186 //generate_compute_interpreter_state(stack_overflow_return); 1187 1188 Register size_of_parameters = R22_tmp2; 1189 1190 generate_fixed_frame(true, size_of_parameters, noreg /* unused */); 1191 1192 //============================================================================= 1193 // Increment invocation counter. On overflow, entry to JNI method 1194 // will be compiled. 1195 Label invocation_counter_overflow, continue_after_compile; 1196 if (inc_counter) { 1197 if (synchronized) { 1198 // Since at this point in the method invocation the exception handler 1199 // would try to exit the monitor of synchronized methods which hasn't 1200 // been entered yet, we set the thread local variable 1201 // _do_not_unlock_if_synchronized to true. If any exception was thrown by 1202 // runtime, exception handling i.e. unlock_if_synchronized_method will 1203 // check this thread local flag. 1204 // This flag has two effects, one is to force an unwind in the topmost 1205 // interpreter frame and not perform an unlock while doing so. 1206 __ li(R0, 1); 1207 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1208 } 1209 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); 1210 1211 BIND(continue_after_compile); 1212 // Reset the _do_not_unlock_if_synchronized flag. 1213 if (synchronized) { 1214 __ li(R0, 0); 1215 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1216 } 1217 } 1218 1219 // access_flags = method->access_flags(); 1220 // Load access flags. 1221 assert(access_flags->is_nonvolatile(), 1222 "access_flags must be in a non-volatile register"); 1223 // Type check. 1224 assert(4 == sizeof(AccessFlags), "unexpected field size"); 1225 __ lwz(access_flags, method_(access_flags)); 1226 1227 // We don't want to reload R19_method and access_flags after calls 1228 // to some helper functions. 1229 assert(R19_method->is_nonvolatile(), 1230 "R19_method must be a non-volatile register"); 1231 1232 // Check for synchronized methods. Must happen AFTER invocation counter 1233 // check, so method is not locked if counter overflows. 1234 1235 if (synchronized) { 1236 lock_method(access_flags, R11_scratch1, R12_scratch2, true); 1237 1238 // Update monitor in state. 1239 __ ld(R11_scratch1, 0, R1_SP); 1240 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1); 1241 } 1242 1243 // jvmti/jvmpi support 1244 __ notify_method_entry(); 1245 1246 //============================================================================= 1247 // Get and call the signature handler. 1248 1249 __ ld(signature_handler_fd, method_(signature_handler)); 1250 Label call_signature_handler; 1251 1252 __ cmpdi(CCR0, signature_handler_fd, 0); 1253 __ bne(CCR0, call_signature_handler); 1254 1255 // Method has never been called. Either generate a specialized 1256 // handler or point to the slow one. 1257 // 1258 // Pass parameter 'false' to avoid exception check in call_VM. 1259 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false); 1260 1261 // Check for an exception while looking up the target method. If we 1262 // incurred one, bail. 1263 __ ld(pending_exception, thread_(pending_exception)); 1264 __ cmpdi(CCR0, pending_exception, 0); 1265 __ bne(CCR0, exception_return_sync_check); // Has pending exception. 1266 1267 // Reload signature handler, it may have been created/assigned in the meanwhile. 1268 __ ld(signature_handler_fd, method_(signature_handler)); 1269 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below). 1270 1271 BIND(call_signature_handler); 1272 1273 // Before we call the signature handler we push a new frame to 1274 // protect the interpreter frame volatile registers when we return 1275 // from jni but before we can get back to Java. 1276 1277 // First set the frame anchor while the SP/FP registers are 1278 // convenient and the slow signature handler can use this same frame 1279 // anchor. 1280 1281 // We have a TOP_IJAVA_FRAME here, which belongs to us. 1282 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/); 1283 1284 // Now the interpreter frame (and its call chain) have been 1285 // invalidated and flushed. We are now protected against eager 1286 // being enabled in native code. Even if it goes eager the 1287 // registers will be reloaded as clean and we will invalidate after 1288 // the call so no spurious flush should be possible. 1289 1290 // Call signature handler and pass locals address. 1291 // 1292 // Our signature handlers copy required arguments to the C stack 1293 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13. 1294 __ mr(R3_ARG1, R18_locals); 1295 #if !defined(ABI_ELFv2) 1296 __ ld(signature_handler_fd, 0, signature_handler_fd); 1297 #endif 1298 1299 __ call_stub(signature_handler_fd); 1300 1301 // Remove the register parameter varargs slots we allocated in 1302 // compute_interpreter_state. SP+16 ends up pointing to the ABI 1303 // outgoing argument area. 1304 // 1305 // Not needed on PPC64. 1306 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord); 1307 1308 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register"); 1309 // Save across call to native method. 1310 __ mr(result_handler_addr, R3_RET); 1311 1312 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror. 1313 1314 // Set up fixed parameters and call the native method. 1315 // If the method is static, get mirror into R4_ARG2. 1316 { 1317 Label method_is_not_static; 1318 // Access_flags is non-volatile and still, no need to restore it. 1319 1320 // Restore access flags. 1321 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT); 1322 __ bfalse(CCR0, method_is_not_static); 1323 1324 // constants = method->constants(); 1325 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method); 1326 __ ld(R11_scratch1, in_bytes(ConstMethod::constants_offset()), R11_scratch1); 1327 // pool_holder = method->constants()->pool_holder(); 1328 __ ld(R11_scratch1/*pool_holder*/, ConstantPool::pool_holder_offset_in_bytes(), 1329 R11_scratch1/*constants*/); 1330 1331 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 1332 1333 // mirror = pool_holder->klass_part()->java_mirror(); 1334 __ ld(R0/*mirror*/, mirror_offset, R11_scratch1/*pool_holder*/); 1335 // state->_native_mirror = mirror; 1336 1337 __ ld(R11_scratch1, 0, R1_SP); 1338 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); 1339 // R4_ARG2 = &state->_oop_temp; 1340 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp)); 1341 BIND(method_is_not_static); 1342 } 1343 1344 // At this point, arguments have been copied off the stack into 1345 // their JNI positions. Oops are boxed in-place on the stack, with 1346 // handles copied to arguments. The result handler address is in a 1347 // register. 1348 1349 // Pass JNIEnv address as first parameter. 1350 __ addir(R3_ARG1, thread_(jni_environment)); 1351 1352 // Load the native_method entry before we change the thread state. 1353 __ ld(native_method_fd, method_(native_function)); 1354 1355 //============================================================================= 1356 // Transition from _thread_in_Java to _thread_in_native. As soon as 1357 // we make this change the safepoint code needs to be certain that 1358 // the last Java frame we established is good. The pc in that frame 1359 // just needs to be near here not an actual return address. 1360 1361 // We use release_store_fence to update values like the thread state, where 1362 // we don't want the current thread to continue until all our prior memory 1363 // accesses (including the new thread state) are visible to other threads. 1364 __ li(R0, _thread_in_native); 1365 __ release(); 1366 1367 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size"); 1368 __ stw(R0, thread_(thread_state)); 1369 1370 if (UseMembar) { 1371 __ fence(); 1372 } 1373 1374 //============================================================================= 1375 // Call the native method. Argument registers must not have been 1376 // overwritten since "__ call_stub(signature_handler);" (except for 1377 // ARG1 and ARG2 for static methods). 1378 __ call_c(native_method_fd); 1379 1380 __ li(R0, 0); 1381 __ ld(R11_scratch1, 0, R1_SP); 1382 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1); 1383 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1); 1384 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset 1385 1386 // Note: C++ interpreter needs the following here: 1387 // The frame_manager_lr field, which we use for setting the last 1388 // java frame, gets overwritten by the signature handler. Restore 1389 // it now. 1390 //__ get_PC_trash_LR(R11_scratch1); 1391 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 1392 1393 // Because of GC R19_method may no longer be valid. 1394 1395 // Block, if necessary, before resuming in _thread_in_Java state. 1396 // In order for GC to work, don't clear the last_Java_sp until after 1397 // blocking. 1398 1399 //============================================================================= 1400 // Switch thread to "native transition" state before reading the 1401 // synchronization state. This additional state is necessary 1402 // because reading and testing the synchronization state is not 1403 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A, 1404 // in _thread_in_native state, loads _not_synchronized and is 1405 // preempted. VM thread changes sync state to synchronizing and 1406 // suspends threads for GC. Thread A is resumed to finish this 1407 // native method, but doesn't block here since it didn't see any 1408 // synchronization in progress, and escapes. 1409 1410 // We use release_store_fence to update values like the thread state, where 1411 // we don't want the current thread to continue until all our prior memory 1412 // accesses (including the new thread state) are visible to other threads. 1413 __ li(R0/*thread_state*/, _thread_in_native_trans); 1414 __ release(); 1415 __ stw(R0/*thread_state*/, thread_(thread_state)); 1416 if (UseMembar) { 1417 __ fence(); 1418 } 1419 // Write serialization page so that the VM thread can do a pseudo remote 1420 // membar. We use the current thread pointer to calculate a thread 1421 // specific offset to write to within the page. This minimizes bus 1422 // traffic due to cache line collision. 1423 else { 1424 __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2); 1425 } 1426 1427 // Now before we return to java we must look for a current safepoint 1428 // (a new safepoint can not start since we entered native_trans). 1429 // We must check here because a current safepoint could be modifying 1430 // the callers registers right this moment. 1431 1432 // Acquire isn't strictly necessary here because of the fence, but 1433 // sync_state is declared to be volatile, so we do it anyway 1434 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path). 1435 int sync_state_offs = __ load_const_optimized(sync_state_addr, SafepointSynchronize::address_of_state(), /*temp*/R0, true); 1436 1437 // TODO PPC port assert(4 == SafepointSynchronize::sz_state(), "unexpected field size"); 1438 __ lwz(sync_state, sync_state_offs, sync_state_addr); 1439 1440 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size"); 1441 __ lwz(suspend_flags, thread_(suspend_flags)); 1442 1443 Label sync_check_done; 1444 Label do_safepoint; 1445 // No synchronization in progress nor yet synchronized. 1446 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized); 1447 // Not suspended. 1448 __ cmpwi(CCR1, suspend_flags, 0); 1449 1450 __ bne(CCR0, do_safepoint); 1451 __ beq(CCR1, sync_check_done); 1452 __ bind(do_safepoint); 1453 __ isync(); 1454 // Block. We do the call directly and leave the current 1455 // last_Java_frame setup undisturbed. We must save any possible 1456 // native result across the call. No oop is present. 1457 1458 __ mr(R3_ARG1, R16_thread); 1459 #if defined(ABI_ELFv2) 1460 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), 1461 relocInfo::none); 1462 #else 1463 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans), 1464 relocInfo::none); 1465 #endif 1466 1467 __ bind(sync_check_done); 1468 1469 //============================================================================= 1470 // <<<<<< Back in Interpreter Frame >>>>> 1471 1472 // We are in thread_in_native_trans here and back in the normal 1473 // interpreter frame. We don't have to do anything special about 1474 // safepoints and we can switch to Java mode anytime we are ready. 1475 1476 // Note: frame::interpreter_frame_result has a dependency on how the 1477 // method result is saved across the call to post_method_exit. For 1478 // native methods it assumes that the non-FPU/non-void result is 1479 // saved in _native_lresult and a FPU result in _native_fresult. If 1480 // this changes then the interpreter_frame_result implementation 1481 // will need to be updated too. 1482 1483 // On PPC64, we have stored the result directly after the native call. 1484 1485 //============================================================================= 1486 // Back in Java 1487 1488 // We use release_store_fence to update values like the thread state, where 1489 // we don't want the current thread to continue until all our prior memory 1490 // accesses (including the new thread state) are visible to other threads. 1491 __ li(R0/*thread_state*/, _thread_in_Java); 1492 __ release(); 1493 __ stw(R0/*thread_state*/, thread_(thread_state)); 1494 if (UseMembar) { 1495 __ fence(); 1496 } 1497 1498 __ reset_last_Java_frame(); 1499 1500 // Jvmdi/jvmpi support. Whether we've got an exception pending or 1501 // not, and whether unlocking throws an exception or not, we notify 1502 // on native method exit. If we do have an exception, we'll end up 1503 // in the caller's context to handle it, so if we don't do the 1504 // notify here, we'll drop it on the floor. 1505 __ notify_method_exit(true/*native method*/, 1506 ilgl /*illegal state (not used for native methods)*/, 1507 InterpreterMacroAssembler::NotifyJVMTI, 1508 false /*check_exceptions*/); 1509 1510 //============================================================================= 1511 // Handle exceptions 1512 1513 if (synchronized) { 1514 // Don't check for exceptions since we're still in the i2n frame. Do that 1515 // manually afterwards. 1516 unlock_method(false); 1517 } 1518 1519 // Reset active handles after returning from native. 1520 // thread->active_handles()->clear(); 1521 __ ld(active_handles, thread_(active_handles)); 1522 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size"); 1523 __ li(R0, 0); 1524 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles); 1525 1526 Label exception_return_sync_check_already_unlocked; 1527 __ ld(R0/*pending_exception*/, thread_(pending_exception)); 1528 __ cmpdi(CCR0, R0/*pending_exception*/, 0); 1529 __ bne(CCR0, exception_return_sync_check_already_unlocked); 1530 1531 //----------------------------------------------------------------------------- 1532 // No exception pending. 1533 1534 // Move native method result back into proper registers and return. 1535 // Invoke result handler (may unbox/promote). 1536 __ ld(R11_scratch1, 0, R1_SP); 1537 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1); 1538 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1); 1539 __ call_stub(result_handler_addr); 1540 1541 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1542 1543 // Must use the return pc which was loaded from the caller's frame 1544 // as the VM uses return-pc-patching for deoptimization. 1545 __ mtlr(R0); 1546 __ blr(); 1547 1548 //----------------------------------------------------------------------------- 1549 // An exception is pending. We call into the runtime only if the 1550 // caller was not interpreted. If it was interpreted the 1551 // interpreter will do the correct thing. If it isn't interpreted 1552 // (call stub/compiled code) we will change our return and continue. 1553 1554 BIND(exception_return_sync_check); 1555 1556 if (synchronized) { 1557 // Don't check for exceptions since we're still in the i2n frame. Do that 1558 // manually afterwards. 1559 unlock_method(false); 1560 } 1561 BIND(exception_return_sync_check_already_unlocked); 1562 1563 const Register return_pc = R31; 1564 1565 __ ld(return_pc, 0, R1_SP); 1566 __ ld(return_pc, _abi(lr), return_pc); 1567 1568 // Get the address of the exception handler. 1569 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), 1570 R16_thread, 1571 return_pc /* return pc */); 1572 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2); 1573 1574 // Load the PC of the the exception handler into LR. 1575 __ mtlr(R3_RET); 1576 1577 // Load exception into R3_ARG1 and clear pending exception in thread. 1578 __ ld(R3_ARG1/*exception*/, thread_(pending_exception)); 1579 __ li(R4_ARG2, 0); 1580 __ std(R4_ARG2, thread_(pending_exception)); 1581 1582 // Load the original return pc into R4_ARG2. 1583 __ mr(R4_ARG2/*issuing_pc*/, return_pc); 1584 1585 // Return to exception handler. 1586 __ blr(); 1587 1588 //============================================================================= 1589 // Counter overflow. 1590 1591 if (inc_counter) { 1592 // Handle invocation counter overflow. 1593 __ bind(invocation_counter_overflow); 1594 1595 generate_counter_overflow(continue_after_compile); 1596 } 1597 1598 return entry; 1599 } 1600 1601 // Generic interpreted method entry to (asm) interpreter. 1602 // 1603 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) { 1604 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; 1605 address entry = __ pc(); 1606 // Generate the code to allocate the interpreter stack frame. 1607 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame. 1608 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame. 1609 1610 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals); 1611 1612 // -------------------------------------------------------------------------- 1613 // Zero out non-parameter locals. 1614 // Note: *Always* zero out non-parameter locals as Sparc does. It's not 1615 // worth to ask the flag, just do it. 1616 Register Rslot_addr = R6_ARG4, 1617 Rnum = R7_ARG5; 1618 Label Lno_locals, Lzero_loop; 1619 1620 // Set up the zeroing loop. 1621 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals); 1622 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals); 1623 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize); 1624 __ beq(CCR0, Lno_locals); 1625 __ li(R0, 0); 1626 __ mtctr(Rnum); 1627 1628 // The zero locals loop. 1629 __ bind(Lzero_loop); 1630 __ std(R0, 0, Rslot_addr); 1631 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize); 1632 __ bdnz(Lzero_loop); 1633 1634 __ bind(Lno_locals); 1635 1636 // -------------------------------------------------------------------------- 1637 // Counter increment and overflow check. 1638 Label invocation_counter_overflow, 1639 profile_method, 1640 profile_method_continue; 1641 if (inc_counter || ProfileInterpreter) { 1642 1643 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1; 1644 if (synchronized) { 1645 // Since at this point in the method invocation the exception handler 1646 // would try to exit the monitor of synchronized methods which hasn't 1647 // been entered yet, we set the thread local variable 1648 // _do_not_unlock_if_synchronized to true. If any exception was thrown by 1649 // runtime, exception handling i.e. unlock_if_synchronized_method will 1650 // check this thread local flag. 1651 // This flag has two effects, one is to force an unwind in the topmost 1652 // interpreter frame and not perform an unlock while doing so. 1653 __ li(R0, 1); 1654 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1655 } 1656 1657 // Argument and return type profiling. 1658 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4); 1659 1660 // Increment invocation counter and check for overflow. 1661 if (inc_counter) { 1662 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue); 1663 } 1664 1665 __ bind(profile_method_continue); 1666 1667 // Reset the _do_not_unlock_if_synchronized flag. 1668 if (synchronized) { 1669 __ li(R0, 0); 1670 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1671 } 1672 } 1673 1674 // -------------------------------------------------------------------------- 1675 // Locking of synchronized methods. Must happen AFTER invocation_counter 1676 // check and stack overflow check, so method is not locked if overflows. 1677 if (synchronized) { 1678 lock_method(R3_ARG1, R4_ARG2, R5_ARG3); 1679 } 1680 #ifdef ASSERT 1681 else { 1682 Label Lok; 1683 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method); 1684 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED); 1685 __ asm_assert_eq("method needs synchronization", 0x8521); 1686 __ bind(Lok); 1687 } 1688 #endif // ASSERT 1689 1690 __ verify_thread(); 1691 1692 // -------------------------------------------------------------------------- 1693 // JVMTI support 1694 __ notify_method_entry(); 1695 1696 // -------------------------------------------------------------------------- 1697 // Start executing instructions. 1698 __ dispatch_next(vtos); 1699 1700 // -------------------------------------------------------------------------- 1701 // Out of line counter overflow and MDO creation code. 1702 if (ProfileInterpreter) { 1703 // We have decided to profile this method in the interpreter. 1704 __ bind(profile_method); 1705 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1706 __ set_method_data_pointer_for_bcp(); 1707 __ b(profile_method_continue); 1708 } 1709 1710 if (inc_counter) { 1711 // Handle invocation counter overflow. 1712 __ bind(invocation_counter_overflow); 1713 generate_counter_overflow(profile_method_continue); 1714 } 1715 return entry; 1716 } 1717 1718 // CRC32 Intrinsics. 1719 // 1720 // Contract on scratch and work registers. 1721 // ======================================= 1722 // 1723 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers. 1724 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set. 1725 // You can't rely on these registers across calls. 1726 // 1727 // The generators for CRC32_update and for CRC32_updateBytes use the 1728 // scratch/work register set internally, passing the work registers 1729 // as arguments to the MacroAssembler emitters as required. 1730 // 1731 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments. 1732 // Their contents is not constant but may change according to the requirements 1733 // of the emitted code. 1734 // 1735 // All other registers from the scratch/work register set are used "internally" 1736 // and contain garbage (i.e. unpredictable values) once blr() is reached. 1737 // Basically, only R3_RET contains a defined value which is the function result. 1738 // 1739 /** 1740 * Method entry for static native methods: 1741 * int java.util.zip.CRC32.update(int crc, int b) 1742 */ 1743 address TemplateInterpreterGenerator::generate_CRC32_update_entry() { 1744 if (UseCRC32Intrinsics) { 1745 address start = __ pc(); // Remember stub start address (is rtn value). 1746 Label slow_path; 1747 1748 // Safepoint check 1749 const Register sync_state = R11_scratch1; 1750 int sync_state_offs = __ load_const_optimized(sync_state, SafepointSynchronize::address_of_state(), /*temp*/R0, true); 1751 __ lwz(sync_state, sync_state_offs, sync_state); 1752 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized); 1753 __ bne(CCR0, slow_path); 1754 1755 // We don't generate local frame and don't align stack because 1756 // we not even call stub code (we generate the code inline) 1757 // and there is no safepoint on this path. 1758 1759 // Load java parameters. 1760 // R15_esp is callers operand stack pointer, i.e. it points to the parameters. 1761 const Register argP = R15_esp; 1762 const Register crc = R3_ARG1; // crc value 1763 const Register data = R4_ARG2; // address of java byte value (kernel_crc32 needs address) 1764 const Register dataLen = R5_ARG3; // source data len (1 byte). Not used because calling the single-byte emitter. 1765 const Register table = R6_ARG4; // address of crc32 table 1766 const Register tmp = dataLen; // Reuse unused len register to show we don't actually need a separate tmp here. 1767 1768 BLOCK_COMMENT("CRC32_update {"); 1769 1770 // Arguments are reversed on java expression stack 1771 #ifdef VM_LITTLE_ENDIAN 1772 __ addi(data, argP, 0+1*wordSize); // (stack) address of byte value. Emitter expects address, not value. 1773 // Being passed as an int, the single byte is at offset +0. 1774 #else 1775 __ addi(data, argP, 3+1*wordSize); // (stack) address of byte value. Emitter expects address, not value. 1776 // Being passed from java as an int, the single byte is at offset +3. 1777 #endif 1778 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register. 1779 1780 StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table); 1781 __ kernel_crc32_singleByte(crc, data, dataLen, table, tmp); 1782 1783 // Restore caller sp for c2i case and return. 1784 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 1785 __ blr(); 1786 1787 // Generate a vanilla native entry as the slow path. 1788 BLOCK_COMMENT("} CRC32_update"); 1789 BIND(slow_path); 1790 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1); 1791 return start; 1792 } 1793 1794 return NULL; 1795 } 1796 1797 // CRC32 Intrinsics. 1798 /** 1799 * Method entry for static native methods: 1800 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len) 1801 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len) 1802 */ 1803 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 1804 if (UseCRC32Intrinsics) { 1805 address start = __ pc(); // Remember stub start address (is rtn value). 1806 Label slow_path; 1807 1808 // Safepoint check 1809 const Register sync_state = R11_scratch1; 1810 int sync_state_offs = __ load_const_optimized(sync_state, SafepointSynchronize::address_of_state(), /*temp*/R0, true); 1811 __ lwz(sync_state, sync_state_offs, sync_state); 1812 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized); 1813 __ bne(CCR0, slow_path); 1814 1815 // We don't generate local frame and don't align stack because 1816 // we not even call stub code (we generate the code inline) 1817 // and there is no safepoint on this path. 1818 1819 // Load parameters. 1820 // Z_esp is callers operand stack pointer, i.e. it points to the parameters. 1821 const Register argP = R15_esp; 1822 const Register crc = R3_ARG1; // crc value 1823 const Register data = R4_ARG2; // address of java byte array 1824 const Register dataLen = R5_ARG3; // source data len 1825 const Register table = R6_ARG4; // address of crc32 table 1826 1827 const Register t0 = R9; // scratch registers for crc calculation 1828 const Register t1 = R10; 1829 const Register t2 = R11; 1830 const Register t3 = R12; 1831 1832 const Register tc0 = R2; // registers to hold pre-calculated column addresses 1833 const Register tc1 = R7; 1834 const Register tc2 = R8; 1835 const Register tc3 = table; // table address is reconstructed at the end of kernel_crc32_* emitters 1836 1837 const Register tmp = t0; // Only used very locally to calculate byte buffer address. 1838 1839 // Arguments are reversed on java expression stack. 1840 // Calculate address of start element. 1841 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct". 1842 BLOCK_COMMENT("CRC32_updateByteBuffer {"); 1843 // crc @ (SP + 5W) (32bit) 1844 // buf @ (SP + 3W) (64bit ptr to long array) 1845 // off @ (SP + 2W) (32bit) 1846 // dataLen @ (SP + 1W) (32bit) 1847 // data = buf + off 1848 __ ld( data, 3*wordSize, argP); // start of byte buffer 1849 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 1850 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 1851 __ lwz( crc, 5*wordSize, argP); // current crc state 1852 __ add( data, data, tmp); // Add byte buffer offset. 1853 } else { // Used for "updateBytes update". 1854 BLOCK_COMMENT("CRC32_updateBytes {"); 1855 // crc @ (SP + 4W) (32bit) 1856 // buf @ (SP + 3W) (64bit ptr to byte array) 1857 // off @ (SP + 2W) (32bit) 1858 // dataLen @ (SP + 1W) (32bit) 1859 // data = buf + off + base_offset 1860 __ ld( data, 3*wordSize, argP); // start of byte buffer 1861 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 1862 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 1863 __ add( data, data, tmp); // add byte buffer offset 1864 __ lwz( crc, 4*wordSize, argP); // current crc state 1865 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1866 } 1867 1868 StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table); 1869 1870 // Performance measurements show the 1word and 2word variants to be almost equivalent, 1871 // with very light advantages for the 1word variant. We chose the 1word variant for 1872 // code compactness. 1873 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3); 1874 1875 // Restore caller sp for c2i case and return. 1876 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 1877 __ blr(); 1878 1879 // Generate a vanilla native entry as the slow path. 1880 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)"); 1881 BIND(slow_path); 1882 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1); 1883 return start; 1884 } 1885 1886 return NULL; 1887 } 1888 1889 // Not supported 1890 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 1891 return NULL; 1892 } 1893 1894 // ============================================================================= 1895 // Exceptions 1896 1897 void TemplateInterpreterGenerator::generate_throw_exception() { 1898 Register Rexception = R17_tos, 1899 Rcontinuation = R3_RET; 1900 1901 // -------------------------------------------------------------------------- 1902 // Entry point if an method returns with a pending exception (rethrow). 1903 Interpreter::_rethrow_exception_entry = __ pc(); 1904 { 1905 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp. 1906 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); 1907 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); 1908 1909 // Compiled code destroys templateTableBase, reload. 1910 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1); 1911 } 1912 1913 // Entry point if a interpreted method throws an exception (throw). 1914 Interpreter::_throw_exception_entry = __ pc(); 1915 { 1916 __ mr(Rexception, R3_RET); 1917 1918 __ verify_thread(); 1919 __ verify_oop(Rexception); 1920 1921 // Expression stack must be empty before entering the VM in case of an exception. 1922 __ empty_expression_stack(); 1923 // Find exception handler address and preserve exception oop. 1924 // Call C routine to find handler and jump to it. 1925 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception); 1926 __ mtctr(Rcontinuation); 1927 // Push exception for exception handler bytecodes. 1928 __ push_ptr(Rexception); 1929 1930 // Jump to exception handler (may be remove activation entry!). 1931 __ bctr(); 1932 } 1933 1934 // If the exception is not handled in the current frame the frame is 1935 // removed and the exception is rethrown (i.e. exception 1936 // continuation is _rethrow_exception). 1937 // 1938 // Note: At this point the bci is still the bxi for the instruction 1939 // which caused the exception and the expression stack is 1940 // empty. Thus, for any VM calls at this point, GC will find a legal 1941 // oop map (with empty expression stack). 1942 1943 // In current activation 1944 // tos: exception 1945 // bcp: exception bcp 1946 1947 // -------------------------------------------------------------------------- 1948 // JVMTI PopFrame support 1949 1950 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 1951 { 1952 // Set the popframe_processing bit in popframe_condition indicating that we are 1953 // currently handling popframe, so that call_VMs that may happen later do not 1954 // trigger new popframe handling cycles. 1955 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 1956 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit); 1957 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 1958 1959 // Empty the expression stack, as in normal exception handling. 1960 __ empty_expression_stack(); 1961 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false); 1962 1963 // Check to see whether we are returning to a deoptimized frame. 1964 // (The PopFrame call ensures that the caller of the popped frame is 1965 // either interpreted or compiled and deoptimizes it if compiled.) 1966 // Note that we don't compare the return PC against the 1967 // deoptimization blob's unpack entry because of the presence of 1968 // adapter frames in C2. 1969 Label Lcaller_not_deoptimized; 1970 Register return_pc = R3_ARG1; 1971 __ ld(return_pc, 0, R1_SP); 1972 __ ld(return_pc, _abi(lr), return_pc); 1973 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc); 1974 __ cmpdi(CCR0, R3_RET, 0); 1975 __ bne(CCR0, Lcaller_not_deoptimized); 1976 1977 // The deoptimized case. 1978 // In this case, we can't call dispatch_next() after the frame is 1979 // popped, but instead must save the incoming arguments and restore 1980 // them after deoptimization has occurred. 1981 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method); 1982 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2); 1983 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize); 1984 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize); 1985 __ subf(R5_ARG3, R4_ARG2, R5_ARG3); 1986 // Save these arguments. 1987 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3); 1988 1989 // Inform deoptimization that it is responsible for restoring these arguments. 1990 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit); 1991 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 1992 1993 // Return from the current method into the deoptimization blob. Will eventually 1994 // end up in the deopt interpeter entry, deoptimization prepared everything that 1995 // we will reexecute the call that called us. 1996 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2); 1997 __ mtlr(return_pc); 1998 __ blr(); 1999 2000 // The non-deoptimized case. 2001 __ bind(Lcaller_not_deoptimized); 2002 2003 // Clear the popframe condition flag. 2004 __ li(R0, 0); 2005 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 2006 2007 // Get out of the current method and re-execute the call that called us. 2008 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2); 2009 __ restore_interpreter_state(R11_scratch1); 2010 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); 2011 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); 2012 if (ProfileInterpreter) { 2013 __ set_method_data_pointer_for_bcp(); 2014 __ ld(R11_scratch1, 0, R1_SP); 2015 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1); 2016 } 2017 #if INCLUDE_JVMTI 2018 Label L_done; 2019 2020 __ lbz(R11_scratch1, 0, R14_bcp); 2021 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic); 2022 __ bne(CCR0, L_done); 2023 2024 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. 2025 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL. 2026 __ ld(R4_ARG2, 0, R18_locals); 2027 __ MacroAssembler::call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp, false); 2028 __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); 2029 __ cmpdi(CCR0, R4_ARG2, 0); 2030 __ beq(CCR0, L_done); 2031 __ std(R4_ARG2, wordSize, R15_esp); 2032 __ bind(L_done); 2033 #endif // INCLUDE_JVMTI 2034 __ dispatch_next(vtos); 2035 } 2036 // end of JVMTI PopFrame support 2037 2038 // -------------------------------------------------------------------------- 2039 // Remove activation exception entry. 2040 // This is jumped to if an interpreted method can't handle an exception itself 2041 // (we come from the throw/rethrow exception entry above). We're going to call 2042 // into the VM to find the exception handler in the caller, pop the current 2043 // frame and return the handler we calculated. 2044 Interpreter::_remove_activation_entry = __ pc(); 2045 { 2046 __ pop_ptr(Rexception); 2047 __ verify_thread(); 2048 __ verify_oop(Rexception); 2049 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread); 2050 2051 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true); 2052 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false); 2053 2054 __ get_vm_result(Rexception); 2055 2056 // We are done with this activation frame; find out where to go next. 2057 // The continuation point will be an exception handler, which expects 2058 // the following registers set up: 2059 // 2060 // RET: exception oop 2061 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled. 2062 2063 Register return_pc = R31; // Needs to survive the runtime call. 2064 __ ld(return_pc, 0, R1_SP); 2065 __ ld(return_pc, _abi(lr), return_pc); 2066 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc); 2067 2068 // Remove the current activation. 2069 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2); 2070 2071 __ mr(R4_ARG2, return_pc); 2072 __ mtlr(R3_RET); 2073 __ mr(R3_RET, Rexception); 2074 __ blr(); 2075 } 2076 } 2077 2078 // JVMTI ForceEarlyReturn support. 2079 // Returns "in the middle" of a method with a "fake" return value. 2080 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 2081 2082 Register Rscratch1 = R11_scratch1, 2083 Rscratch2 = R12_scratch2; 2084 2085 address entry = __ pc(); 2086 __ empty_expression_stack(); 2087 2088 __ load_earlyret_value(state, Rscratch1); 2089 2090 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 2091 // Clear the earlyret state. 2092 __ li(R0, 0); 2093 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1); 2094 2095 __ remove_activation(state, false, false); 2096 // Copied from TemplateTable::_return. 2097 // Restoration of lr done by remove_activation. 2098 switch (state) { 2099 case ltos: 2100 case btos: 2101 case ctos: 2102 case stos: 2103 case atos: 2104 case itos: __ mr(R3_RET, R17_tos); break; 2105 case ftos: 2106 case dtos: __ fmr(F1_RET, F15_ftos); break; 2107 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need 2108 // to get visible before the reference to the object gets stored anywhere. 2109 __ membar(Assembler::StoreStore); break; 2110 default : ShouldNotReachHere(); 2111 } 2112 __ blr(); 2113 2114 return entry; 2115 } // end of ForceEarlyReturn support 2116 2117 //----------------------------------------------------------------------------- 2118 // Helper for vtos entry point generation 2119 2120 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, 2121 address& bep, 2122 address& cep, 2123 address& sep, 2124 address& aep, 2125 address& iep, 2126 address& lep, 2127 address& fep, 2128 address& dep, 2129 address& vep) { 2130 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 2131 Label L; 2132 2133 aep = __ pc(); __ push_ptr(); __ b(L); 2134 fep = __ pc(); __ push_f(); __ b(L); 2135 dep = __ pc(); __ push_d(); __ b(L); 2136 lep = __ pc(); __ push_l(); __ b(L); 2137 __ align(32, 12, 24); // align L 2138 bep = cep = sep = 2139 iep = __ pc(); __ push_i(); 2140 vep = __ pc(); 2141 __ bind(L); 2142 generate_and_dispatch(t); 2143 } 2144 2145 //----------------------------------------------------------------------------- 2146 2147 // Non-product code 2148 #ifndef PRODUCT 2149 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 2150 //__ flush_bundle(); 2151 address entry = __ pc(); 2152 2153 const char *bname = NULL; 2154 uint tsize = 0; 2155 switch(state) { 2156 case ftos: 2157 bname = "trace_code_ftos {"; 2158 tsize = 2; 2159 break; 2160 case btos: 2161 bname = "trace_code_btos {"; 2162 tsize = 2; 2163 break; 2164 case ctos: 2165 bname = "trace_code_ctos {"; 2166 tsize = 2; 2167 break; 2168 case stos: 2169 bname = "trace_code_stos {"; 2170 tsize = 2; 2171 break; 2172 case itos: 2173 bname = "trace_code_itos {"; 2174 tsize = 2; 2175 break; 2176 case ltos: 2177 bname = "trace_code_ltos {"; 2178 tsize = 3; 2179 break; 2180 case atos: 2181 bname = "trace_code_atos {"; 2182 tsize = 2; 2183 break; 2184 case vtos: 2185 // Note: In case of vtos, the topmost of stack value could be a int or doubl 2186 // In case of a double (2 slots) we won't see the 2nd stack value. 2187 // Maybe we simply should print the topmost 3 stack slots to cope with the problem. 2188 bname = "trace_code_vtos {"; 2189 tsize = 2; 2190 2191 break; 2192 case dtos: 2193 bname = "trace_code_dtos {"; 2194 tsize = 3; 2195 break; 2196 default: 2197 ShouldNotReachHere(); 2198 } 2199 BLOCK_COMMENT(bname); 2200 2201 // Support short-cut for TraceBytecodesAt. 2202 // Don't call into the VM if we don't want to trace to speed up things. 2203 Label Lskip_vm_call; 2204 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) { 2205 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true); 2206 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true); 2207 __ ld(R11_scratch1, offs1, R11_scratch1); 2208 __ lwa(R12_scratch2, offs2, R12_scratch2); 2209 __ cmpd(CCR0, R12_scratch2, R11_scratch1); 2210 __ blt(CCR0, Lskip_vm_call); 2211 } 2212 2213 __ push(state); 2214 // Load 2 topmost expression stack values. 2215 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp); 2216 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp); 2217 __ mflr(R31); 2218 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false); 2219 __ mtlr(R31); 2220 __ pop(state); 2221 2222 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) { 2223 __ bind(Lskip_vm_call); 2224 } 2225 __ blr(); 2226 BLOCK_COMMENT("} trace_code"); 2227 return entry; 2228 } 2229 2230 void TemplateInterpreterGenerator::count_bytecode() { 2231 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true); 2232 __ lwz(R12_scratch2, offs, R11_scratch1); 2233 __ addi(R12_scratch2, R12_scratch2, 1); 2234 __ stw(R12_scratch2, offs, R11_scratch1); 2235 } 2236 2237 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { 2238 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true); 2239 __ lwz(R12_scratch2, offs, R11_scratch1); 2240 __ addi(R12_scratch2, R12_scratch2, 1); 2241 __ stw(R12_scratch2, offs, R11_scratch1); 2242 } 2243 2244 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { 2245 const Register addr = R11_scratch1, 2246 tmp = R12_scratch2; 2247 // Get index, shift out old bytecode, bring in new bytecode, and store it. 2248 // _index = (_index >> log2_number_of_codes) | 2249 // (bytecode << log2_number_of_codes); 2250 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true); 2251 __ lwz(tmp, offs1, addr); 2252 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes); 2253 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes); 2254 __ stw(tmp, offs1, addr); 2255 2256 // Bump bucket contents. 2257 // _counters[_index] ++; 2258 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true); 2259 __ sldi(tmp, tmp, LogBytesPerInt); 2260 __ add(addr, tmp, addr); 2261 __ lwz(tmp, offs2, addr); 2262 __ addi(tmp, tmp, 1); 2263 __ stw(tmp, offs2, addr); 2264 } 2265 2266 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 2267 // Call a little run-time stub to avoid blow-up for each bytecode. 2268 // The run-time runtime saves the right registers, depending on 2269 // the tosca in-state for the given template. 2270 2271 assert(Interpreter::trace_code(t->tos_in()) != NULL, 2272 "entry must have been generated"); 2273 2274 // Note: we destroy LR here. 2275 __ bl(Interpreter::trace_code(t->tos_in())); 2276 } 2277 2278 void TemplateInterpreterGenerator::stop_interpreter_at() { 2279 Label L; 2280 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true); 2281 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true); 2282 __ ld(R11_scratch1, offs1, R11_scratch1); 2283 __ lwa(R12_scratch2, offs2, R12_scratch2); 2284 __ cmpd(CCR0, R12_scratch2, R11_scratch1); 2285 __ bne(CCR0, L); 2286 __ illtrap(); 2287 __ bind(L); 2288 } 2289 2290 #endif // !PRODUCT