1 /* 2 * Copyright (c) 2014, 2017, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2015, 2017, SAP SE. 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 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { 566 address entry = __ pc(); 567 568 // Expression stack must be empty before entering the VM if an 569 // exception happened. 570 __ empty_expression_stack(); 571 // Throw exception. 572 __ call_VM(noreg, 573 CAST_FROM_FN_PTR(address, 574 InterpreterRuntime::throw_StackOverflowError)); 575 return entry; 576 } 577 578 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) { 579 address entry = __ pc(); 580 __ empty_expression_stack(); 581 __ load_const_optimized(R4_ARG2, (address) name); 582 // Index is in R17_tos. 583 __ mr(R5_ARG3, R17_tos); 584 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException)); 585 return entry; 586 } 587 588 #if 0 589 // Call special ClassCastException constructor taking object to cast 590 // and target class as arguments. 591 address TemplateInterpreterGenerator::generate_ClassCastException_verbose_handler() { 592 address entry = __ pc(); 593 594 // Expression stack must be empty before entering the VM if an 595 // exception happened. 596 __ empty_expression_stack(); 597 598 // Thread will be loaded to R3_ARG1. 599 // Target class oop is in register R5_ARG3 by convention! 600 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException_verbose), R17_tos, R5_ARG3); 601 // Above call must not return here since exception pending. 602 DEBUG_ONLY(__ should_not_reach_here();) 603 return entry; 604 } 605 #endif 606 607 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 608 address entry = __ pc(); 609 // Expression stack must be empty before entering the VM if an 610 // exception happened. 611 __ empty_expression_stack(); 612 613 // Load exception object. 614 // Thread will be loaded to R3_ARG1. 615 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos); 616 #ifdef ASSERT 617 // Above call must not return here since exception pending. 618 __ should_not_reach_here(); 619 #endif 620 return entry; 621 } 622 623 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) { 624 address entry = __ pc(); 625 //__ untested("generate_exception_handler_common"); 626 Register Rexception = R17_tos; 627 628 // Expression stack must be empty before entering the VM if an exception happened. 629 __ empty_expression_stack(); 630 631 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1); 632 if (pass_oop) { 633 __ mr(R5_ARG3, Rexception); 634 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), false); 635 } else { 636 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1); 637 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), false); 638 } 639 640 // Throw exception. 641 __ mr(R3_ARG1, Rexception); 642 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2); 643 __ mtctr(R11_scratch1); 644 __ bctr(); 645 646 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 ztos: 659 case ctos: 660 case stos: 661 case atos: 662 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache 663 case ftos: 664 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET 665 case vtos: break; // Nothing to do, this was a void return. 666 default : ShouldNotReachHere(); 667 } 668 669 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp. 670 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); 671 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); 672 673 // Compiled code destroys templateTableBase, reload. 674 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2); 675 676 if (state == atos) { 677 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2); 678 } 679 680 const Register cache = R11_scratch1; 681 const Register size = R12_scratch2; 682 __ get_cache_and_index_at_bcp(cache, 1, index_size); 683 684 // Get least significant byte of 64 bit value: 685 #if defined(VM_LITTLE_ENDIAN) 686 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache); 687 #else 688 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache); 689 #endif 690 __ sldi(size, size, Interpreter::logStackElementSize); 691 __ add(R15_esp, R15_esp, size); 692 __ dispatch_next(state, step); 693 return entry; 694 } 695 696 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) { 697 address entry = __ pc(); 698 // If state != vtos, we're returning from a native method, which put it's result 699 // into the result register. So move the value out of the return register back 700 // to the TOS cache of current frame. 701 702 switch (state) { 703 case ltos: 704 case btos: 705 case ztos: 706 case ctos: 707 case stos: 708 case atos: 709 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache 710 case ftos: 711 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET 712 case vtos: break; // Nothing to do, this was a void return. 713 default : ShouldNotReachHere(); 714 } 715 716 // Load LcpoolCache @@@ should be already set! 717 __ get_constant_pool_cache(R27_constPoolCache); 718 719 // Handle a pending exception, fall through if none. 720 __ check_and_forward_exception(R11_scratch1, R12_scratch2); 721 722 // Start executing bytecodes. 723 __ dispatch_next(state, step); 724 725 return entry; 726 } 727 728 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) { 729 address entry = __ pc(); 730 731 __ push(state); 732 __ call_VM(noreg, runtime_entry); 733 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); 734 735 return entry; 736 } 737 738 // Helpers for commoning out cases in the various type of method entries. 739 740 // Increment invocation count & check for overflow. 741 // 742 // Note: checking for negative value instead of overflow 743 // so we have a 'sticky' overflow test. 744 // 745 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) { 746 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not. 747 Register Rscratch1 = R11_scratch1; 748 Register Rscratch2 = R12_scratch2; 749 Register R3_counters = R3_ARG1; 750 Label done; 751 752 if (TieredCompilation) { 753 const int increment = InvocationCounter::count_increment; 754 Label no_mdo; 755 if (ProfileInterpreter) { 756 const Register Rmdo = R3_counters; 757 // If no method data exists, go to profile_continue. 758 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method); 759 __ cmpdi(CCR0, Rmdo, 0); 760 __ beq(CCR0, no_mdo); 761 762 // Increment invocation counter in the MDO. 763 const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset()); 764 __ lwz(Rscratch2, mdo_ic_offs, Rmdo); 765 __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo); 766 __ addi(Rscratch2, Rscratch2, increment); 767 __ stw(Rscratch2, mdo_ic_offs, Rmdo); 768 __ and_(Rscratch1, Rscratch2, Rscratch1); 769 __ bne(CCR0, done); 770 __ b(*overflow); 771 } 772 773 // Increment counter in MethodCounters*. 774 const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset()); 775 __ bind(no_mdo); 776 __ get_method_counters(R19_method, R3_counters, done); 777 __ lwz(Rscratch2, mo_ic_offs, R3_counters); 778 __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters); 779 __ addi(Rscratch2, Rscratch2, increment); 780 __ stw(Rscratch2, mo_ic_offs, R3_counters); 781 __ and_(Rscratch1, Rscratch2, Rscratch1); 782 __ beq(CCR0, *overflow); 783 784 __ bind(done); 785 786 } else { 787 788 // Update standard invocation counters. 789 Register Rsum_ivc_bec = R4_ARG2; 790 __ get_method_counters(R19_method, R3_counters, done); 791 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2); 792 // Increment interpreter invocation counter. 793 if (ProfileInterpreter) { // %%% Merge this into methodDataOop. 794 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters); 795 __ addi(R12_scratch2, R12_scratch2, 1); 796 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters); 797 } 798 // Check if we must create a method data obj. 799 if (ProfileInterpreter && profile_method != NULL) { 800 const Register profile_limit = Rscratch1; 801 __ lwz(profile_limit, in_bytes(MethodCounters::interpreter_profile_limit_offset()), R3_counters); 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 __ lwz(invocation_limit, in_bytes(MethodCounters::interpreter_invocation_limit_offset()), R3_counters); 812 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit); 813 __ bge(CCR0, *overflow); 814 } 815 816 __ bind(done); 817 } 818 } 819 820 // Generate code to initiate compilation on invocation counter overflow. 821 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) { 822 // Generate code to initiate compilation on the counter overflow. 823 824 // InterpreterRuntime::frequency_counter_overflow takes one arguments, 825 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp) 826 // We pass zero in. 827 // The call returns the address of the verified entry point for the method or NULL 828 // if the compilation did not complete (either went background or bailed out). 829 // 830 // Unlike the C++ interpreter above: Check exceptions! 831 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed 832 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur. 833 834 __ li(R4_ARG2, 0); 835 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true); 836 837 // Returns verified_entry_point or NULL. 838 // We ignore it in any case. 839 __ b(continue_entry); 840 } 841 842 // See if we've got enough room on the stack for locals plus overhead below 843 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError 844 // without going through the signal handler, i.e., reserved and yellow zones 845 // will not be made usable. The shadow zone must suffice to handle the 846 // overflow. 847 // 848 // Kills Rmem_frame_size, Rscratch1. 849 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) { 850 Label done; 851 assert_different_registers(Rmem_frame_size, Rscratch1); 852 853 BLOCK_COMMENT("stack_overflow_check_with_compare {"); 854 __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size); 855 __ ld(Rscratch1, thread_(stack_overflow_limit)); 856 __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1); 857 __ bgt(CCR0/*is_stack_overflow*/, done); 858 859 // The stack overflows. Load target address of the runtime stub and call it. 860 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order"); 861 __ load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0); 862 __ mtctr(Rscratch1); 863 // Restore caller_sp. 864 #ifdef ASSERT 865 __ ld(Rscratch1, 0, R1_SP); 866 __ ld(R0, 0, R21_sender_SP); 867 __ cmpd(CCR0, R0, Rscratch1); 868 __ asm_assert_eq("backlink", 0x547); 869 #endif // ASSERT 870 __ mr(R1_SP, R21_sender_SP); 871 __ bctr(); 872 873 __ align(32, 12); 874 __ bind(done); 875 BLOCK_COMMENT("} stack_overflow_check_with_compare"); 876 } 877 878 // Lock the current method, interpreter register window must be set up! 879 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) { 880 const Register Robj_to_lock = Rscratch2; 881 882 { 883 if (!flags_preloaded) { 884 __ lwz(Rflags, method_(access_flags)); 885 } 886 887 #ifdef ASSERT 888 // Check if methods needs synchronization. 889 { 890 Label Lok; 891 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT); 892 __ btrue(CCR0,Lok); 893 __ stop("method doesn't need synchronization"); 894 __ bind(Lok); 895 } 896 #endif // ASSERT 897 } 898 899 // Get synchronization object to Rscratch2. 900 { 901 Label Lstatic; 902 Label Ldone; 903 904 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT); 905 __ btrue(CCR0, Lstatic); 906 907 // Non-static case: load receiver obj from stack and we're done. 908 __ ld(Robj_to_lock, R18_locals); 909 __ b(Ldone); 910 911 __ bind(Lstatic); // Static case: Lock the java mirror 912 // Load mirror from interpreter frame. 913 __ ld(Robj_to_lock, _abi(callers_sp), R1_SP); 914 __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock); 915 916 __ bind(Ldone); 917 __ verify_oop(Robj_to_lock); 918 } 919 920 // Got the oop to lock => execute! 921 __ add_monitor_to_stack(true, Rscratch1, R0); 922 923 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor); 924 __ lock_object(R26_monitor, Robj_to_lock); 925 } 926 927 // Generate a fixed interpreter frame for pure interpreter 928 // and I2N native transition frames. 929 // 930 // Before (stack grows downwards): 931 // 932 // | ... | 933 // |------------- | 934 // | java arg0 | 935 // | ... | 936 // | java argn | 937 // | | <- R15_esp 938 // | | 939 // |--------------| 940 // | abi_112 | 941 // | | <- R1_SP 942 // |==============| 943 // 944 // 945 // After: 946 // 947 // | ... | 948 // | java arg0 |<- R18_locals 949 // | ... | 950 // | java argn | 951 // |--------------| 952 // | | 953 // | java locals | 954 // | | 955 // |--------------| 956 // | abi_48 | 957 // |==============| 958 // | | 959 // | istate | 960 // | | 961 // |--------------| 962 // | monitor |<- R26_monitor 963 // |--------------| 964 // | |<- R15_esp 965 // | expression | 966 // | stack | 967 // | | 968 // |--------------| 969 // | | 970 // | abi_112 |<- R1_SP 971 // |==============| 972 // 973 // The top most frame needs an abi space of 112 bytes. This space is needed, 974 // since we call to c. The c function may spill their arguments to the caller 975 // frame. When we call to java, we don't need these spill slots. In order to save 976 // space on the stack, we resize the caller. However, java locals reside in 977 // the caller frame and the frame has to be increased. The frame_size for the 978 // current frame was calculated based on max_stack as size for the expression 979 // stack. At the call, just a part of the expression stack might be used. 980 // We don't want to waste this space and cut the frame back accordingly. 981 // The resulting amount for resizing is calculated as follows: 982 // resize = (number_of_locals - number_of_arguments) * slot_size 983 // + (R1_SP - R15_esp) + 48 984 // 985 // The size for the callee frame is calculated: 986 // framesize = 112 + max_stack + monitor + state_size 987 // 988 // maxstack: Max number of slots on the expression stack, loaded from the method. 989 // monitor: We statically reserve room for one monitor object. 990 // state_size: We save the current state of the interpreter to this area. 991 // 992 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) { 993 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes. 994 top_frame_size = R7_ARG5, 995 Rconst_method = R8_ARG6; 996 997 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size); 998 999 __ ld(Rconst_method, method_(const)); 1000 __ lhz(Rsize_of_parameters /* number of params */, 1001 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method); 1002 if (native_call) { 1003 // If we're calling a native method, we reserve space for the worst-case signature 1004 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2). 1005 // We add two slots to the parameter_count, one for the jni 1006 // environment and one for a possible native mirror. 1007 Label skip_native_calculate_max_stack; 1008 __ addi(top_frame_size, Rsize_of_parameters, 2); 1009 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters); 1010 __ bge(CCR0, skip_native_calculate_max_stack); 1011 __ li(top_frame_size, Argument::n_register_parameters); 1012 __ bind(skip_native_calculate_max_stack); 1013 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize); 1014 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize); 1015 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize! 1016 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters. 1017 } else { 1018 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method); 1019 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize); 1020 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize); 1021 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method); 1022 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0 1023 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize! 1024 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize); 1025 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1); 1026 } 1027 1028 // Compute top frame size. 1029 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size); 1030 1031 // Cut back area between esp and max_stack. 1032 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize); 1033 1034 __ round_to(top_frame_size, frame::alignment_in_bytes); 1035 __ round_to(parent_frame_resize, frame::alignment_in_bytes); 1036 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size. 1037 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48. 1038 1039 if (!native_call) { 1040 // Stack overflow check. 1041 // Native calls don't need the stack size check since they have no 1042 // expression stack and the arguments are already on the stack and 1043 // we only add a handful of words to the stack. 1044 __ add(R11_scratch1, parent_frame_resize, top_frame_size); 1045 generate_stack_overflow_check(R11_scratch1, R12_scratch2); 1046 } 1047 1048 // Set up interpreter state registers. 1049 1050 __ add(R18_locals, R15_esp, Rsize_of_parameters); 1051 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method); 1052 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache); 1053 1054 // Set method data pointer. 1055 if (ProfileInterpreter) { 1056 Label zero_continue; 1057 __ ld(R28_mdx, method_(method_data)); 1058 __ cmpdi(CCR0, R28_mdx, 0); 1059 __ beq(CCR0, zero_continue); 1060 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1061 __ bind(zero_continue); 1062 } 1063 1064 if (native_call) { 1065 __ li(R14_bcp, 0); // Must initialize. 1066 } else { 1067 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method); 1068 } 1069 1070 // Resize parent frame. 1071 __ mflr(R12_scratch2); 1072 __ neg(parent_frame_resize, parent_frame_resize); 1073 __ resize_frame(parent_frame_resize, R11_scratch1); 1074 __ std(R12_scratch2, _abi(lr), R1_SP); 1075 1076 // Get mirror and store it in the frame as GC root for this Method*. 1077 __ load_mirror_from_const_method(R12_scratch2, Rconst_method); 1078 1079 __ addi(R26_monitor, R1_SP, - frame::ijava_state_size); 1080 __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); 1081 1082 // Store values. 1083 // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls 1084 // in InterpreterMacroAssembler::call_from_interpreter. 1085 __ std(R19_method, _ijava_state_neg(method), R1_SP); 1086 __ std(R12_scratch2, _ijava_state_neg(mirror), R1_SP); 1087 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP); 1088 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP); 1089 __ std(R18_locals, _ijava_state_neg(locals), R1_SP); 1090 1091 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only 1092 // be found in the frame after save_interpreter_state is done. This is always true 1093 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong, 1094 // because e.g. frame::interpreter_frame_bcp() will not access the correct value 1095 // (Enhanced Stack Trace). 1096 // The signal handler does not save the interpreter state into the frame. 1097 __ li(R0, 0); 1098 #ifdef ASSERT 1099 // Fill remaining slots with constants. 1100 __ load_const_optimized(R11_scratch1, 0x5afe); 1101 __ load_const_optimized(R12_scratch2, 0xdead); 1102 #endif 1103 // We have to initialize some frame slots for native calls (accessed by GC). 1104 if (native_call) { 1105 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP); 1106 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP); 1107 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); } 1108 } 1109 #ifdef ASSERT 1110 else { 1111 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP); 1112 __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP); 1113 __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP); 1114 } 1115 __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP); 1116 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP); 1117 __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP); 1118 __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP); 1119 #endif 1120 __ subf(R12_scratch2, top_frame_size, R1_SP); 1121 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); 1122 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP); 1123 1124 // Push top frame. 1125 __ push_frame(top_frame_size, R11_scratch1); 1126 } 1127 1128 // End of helpers 1129 1130 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) { 1131 1132 // Decide what to do: Use same platform specific instructions and runtime calls as compilers. 1133 bool use_instruction = false; 1134 address runtime_entry = NULL; 1135 int num_args = 1; 1136 bool double_precision = true; 1137 1138 // PPC64 specific: 1139 switch (kind) { 1140 case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break; 1141 case Interpreter::java_lang_math_abs: use_instruction = true; break; 1142 case Interpreter::java_lang_math_fmaF: 1143 case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break; 1144 default: break; // Fall back to runtime call. 1145 } 1146 1147 switch (kind) { 1148 case Interpreter::java_lang_math_sin : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); break; 1149 case Interpreter::java_lang_math_cos : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); break; 1150 case Interpreter::java_lang_math_tan : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); break; 1151 case Interpreter::java_lang_math_abs : /* run interpreted */ break; 1152 case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); break; 1153 case Interpreter::java_lang_math_log : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); break; 1154 case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break; 1155 case Interpreter::java_lang_math_pow : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break; 1156 case Interpreter::java_lang_math_exp : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); break; 1157 case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break; 1158 case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break; 1159 default: ShouldNotReachHere(); 1160 } 1161 1162 // Use normal entry if neither instruction nor runtime call is used. 1163 if (!use_instruction && runtime_entry == NULL) return NULL; 1164 1165 address entry = __ pc(); 1166 1167 // Load arguments 1168 assert(num_args <= 13, "passed in registers"); 1169 if (double_precision) { 1170 int offset = (2 * num_args - 1) * Interpreter::stackElementSize; 1171 for (int i = 0; i < num_args; ++i) { 1172 __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp); 1173 offset -= 2 * Interpreter::stackElementSize; 1174 } 1175 } else { 1176 int offset = num_args * Interpreter::stackElementSize; 1177 for (int i = 0; i < num_args; ++i) { 1178 __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp); 1179 offset -= Interpreter::stackElementSize; 1180 } 1181 } 1182 1183 // Pop c2i arguments (if any) off when we return. 1184 #ifdef ASSERT 1185 __ ld(R9_ARG7, 0, R1_SP); 1186 __ ld(R10_ARG8, 0, R21_sender_SP); 1187 __ cmpd(CCR0, R9_ARG7, R10_ARG8); 1188 __ asm_assert_eq("backlink", 0x545); 1189 #endif // ASSERT 1190 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 1191 1192 if (use_instruction) { 1193 switch (kind) { 1194 case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1); break; 1195 case Interpreter::java_lang_math_abs: __ fabs(F1_RET, F1); break; 1196 case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break; 1197 case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3); break; 1198 default: ShouldNotReachHere(); 1199 } 1200 } else { 1201 // Comment: Can use tail call if the unextended frame is always C ABI compliant: 1202 //__ load_const_optimized(R12_scratch2, runtime_entry, R0); 1203 //__ call_c_and_return_to_caller(R12_scratch2); 1204 1205 // Push a new C frame and save LR. 1206 __ save_LR_CR(R0); 1207 __ push_frame_reg_args(0, R11_scratch1); 1208 1209 __ call_VM_leaf(runtime_entry); 1210 1211 // Pop the C frame and restore LR. 1212 __ pop_frame(); 1213 __ restore_LR_CR(R0); 1214 } 1215 1216 __ blr(); 1217 1218 __ flush(); 1219 1220 return entry; 1221 } 1222 1223 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) { 1224 // Quick & dirty stack overflow checking: bang the stack & handle trap. 1225 // Note that we do the banging after the frame is setup, since the exception 1226 // handling code expects to find a valid interpreter frame on the stack. 1227 // Doing the banging earlier fails if the caller frame is not an interpreter 1228 // frame. 1229 // (Also, the exception throwing code expects to unlock any synchronized 1230 // method receiever, so do the banging after locking the receiver.) 1231 1232 // Bang each page in the shadow zone. We can't assume it's been done for 1233 // an interpreter frame with greater than a page of locals, so each page 1234 // needs to be checked. Only true for non-native. 1235 if (UseStackBanging) { 1236 const int page_size = os::vm_page_size(); 1237 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size; 1238 const int start_page = native_call ? n_shadow_pages : 1; 1239 BLOCK_COMMENT("bang_stack_shadow_pages:"); 1240 for (int pages = start_page; pages <= n_shadow_pages; pages++) { 1241 __ bang_stack_with_offset(pages*page_size); 1242 } 1243 } 1244 } 1245 1246 // Interpreter stub for calling a native method. (asm interpreter) 1247 // This sets up a somewhat different looking stack for calling the 1248 // native method than the typical interpreter frame setup. 1249 // 1250 // On entry: 1251 // R19_method - method 1252 // R16_thread - JavaThread* 1253 // R15_esp - intptr_t* sender tos 1254 // 1255 // abstract stack (grows up) 1256 // [ IJava (caller of JNI callee) ] <-- ASP 1257 // ... 1258 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) { 1259 1260 address entry = __ pc(); 1261 1262 const bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; 1263 1264 // ----------------------------------------------------------------------------- 1265 // Allocate a new frame that represents the native callee (i2n frame). 1266 // This is not a full-blown interpreter frame, but in particular, the 1267 // following registers are valid after this: 1268 // - R19_method 1269 // - R18_local (points to start of arguments to native function) 1270 // 1271 // abstract stack (grows up) 1272 // [ IJava (caller of JNI callee) ] <-- ASP 1273 // ... 1274 1275 const Register signature_handler_fd = R11_scratch1; 1276 const Register pending_exception = R0; 1277 const Register result_handler_addr = R31; 1278 const Register native_method_fd = R11_scratch1; 1279 const Register access_flags = R22_tmp2; 1280 const Register active_handles = R11_scratch1; // R26_monitor saved to state. 1281 const Register sync_state = R12_scratch2; 1282 const Register sync_state_addr = sync_state; // Address is dead after use. 1283 const Register suspend_flags = R11_scratch1; 1284 1285 //============================================================================= 1286 // Allocate new frame and initialize interpreter state. 1287 1288 Label exception_return; 1289 Label exception_return_sync_check; 1290 Label stack_overflow_return; 1291 1292 // Generate new interpreter state and jump to stack_overflow_return in case of 1293 // a stack overflow. 1294 //generate_compute_interpreter_state(stack_overflow_return); 1295 1296 Register size_of_parameters = R22_tmp2; 1297 1298 generate_fixed_frame(true, size_of_parameters, noreg /* unused */); 1299 1300 //============================================================================= 1301 // Increment invocation counter. On overflow, entry to JNI method 1302 // will be compiled. 1303 Label invocation_counter_overflow, continue_after_compile; 1304 if (inc_counter) { 1305 if (synchronized) { 1306 // Since at this point in the method invocation the exception handler 1307 // would try to exit the monitor of synchronized methods which hasn't 1308 // been entered yet, we set the thread local variable 1309 // _do_not_unlock_if_synchronized to true. If any exception was thrown by 1310 // runtime, exception handling i.e. unlock_if_synchronized_method will 1311 // check this thread local flag. 1312 // This flag has two effects, one is to force an unwind in the topmost 1313 // interpreter frame and not perform an unlock while doing so. 1314 __ li(R0, 1); 1315 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1316 } 1317 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); 1318 1319 BIND(continue_after_compile); 1320 } 1321 1322 bang_stack_shadow_pages(true); 1323 1324 if (inc_counter) { 1325 // Reset the _do_not_unlock_if_synchronized flag. 1326 if (synchronized) { 1327 __ li(R0, 0); 1328 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1329 } 1330 } 1331 1332 // access_flags = method->access_flags(); 1333 // Load access flags. 1334 assert(access_flags->is_nonvolatile(), 1335 "access_flags must be in a non-volatile register"); 1336 // Type check. 1337 assert(4 == sizeof(AccessFlags), "unexpected field size"); 1338 __ lwz(access_flags, method_(access_flags)); 1339 1340 // We don't want to reload R19_method and access_flags after calls 1341 // to some helper functions. 1342 assert(R19_method->is_nonvolatile(), 1343 "R19_method must be a non-volatile register"); 1344 1345 // Check for synchronized methods. Must happen AFTER invocation counter 1346 // check, so method is not locked if counter overflows. 1347 1348 if (synchronized) { 1349 lock_method(access_flags, R11_scratch1, R12_scratch2, true); 1350 1351 // Update monitor in state. 1352 __ ld(R11_scratch1, 0, R1_SP); 1353 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1); 1354 } 1355 1356 // jvmti/jvmpi support 1357 __ notify_method_entry(); 1358 1359 //============================================================================= 1360 // Get and call the signature handler. 1361 1362 __ ld(signature_handler_fd, method_(signature_handler)); 1363 Label call_signature_handler; 1364 1365 __ cmpdi(CCR0, signature_handler_fd, 0); 1366 __ bne(CCR0, call_signature_handler); 1367 1368 // Method has never been called. Either generate a specialized 1369 // handler or point to the slow one. 1370 // 1371 // Pass parameter 'false' to avoid exception check in call_VM. 1372 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false); 1373 1374 // Check for an exception while looking up the target method. If we 1375 // incurred one, bail. 1376 __ ld(pending_exception, thread_(pending_exception)); 1377 __ cmpdi(CCR0, pending_exception, 0); 1378 __ bne(CCR0, exception_return_sync_check); // Has pending exception. 1379 1380 // Reload signature handler, it may have been created/assigned in the meanwhile. 1381 __ ld(signature_handler_fd, method_(signature_handler)); 1382 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below). 1383 1384 BIND(call_signature_handler); 1385 1386 // Before we call the signature handler we push a new frame to 1387 // protect the interpreter frame volatile registers when we return 1388 // from jni but before we can get back to Java. 1389 1390 // First set the frame anchor while the SP/FP registers are 1391 // convenient and the slow signature handler can use this same frame 1392 // anchor. 1393 1394 // We have a TOP_IJAVA_FRAME here, which belongs to us. 1395 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/); 1396 1397 // Now the interpreter frame (and its call chain) have been 1398 // invalidated and flushed. We are now protected against eager 1399 // being enabled in native code. Even if it goes eager the 1400 // registers will be reloaded as clean and we will invalidate after 1401 // the call so no spurious flush should be possible. 1402 1403 // Call signature handler and pass locals address. 1404 // 1405 // Our signature handlers copy required arguments to the C stack 1406 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13. 1407 __ mr(R3_ARG1, R18_locals); 1408 #if !defined(ABI_ELFv2) 1409 __ ld(signature_handler_fd, 0, signature_handler_fd); 1410 #endif 1411 1412 __ call_stub(signature_handler_fd); 1413 1414 // Remove the register parameter varargs slots we allocated in 1415 // compute_interpreter_state. SP+16 ends up pointing to the ABI 1416 // outgoing argument area. 1417 // 1418 // Not needed on PPC64. 1419 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord); 1420 1421 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register"); 1422 // Save across call to native method. 1423 __ mr(result_handler_addr, R3_RET); 1424 1425 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror. 1426 1427 // Set up fixed parameters and call the native method. 1428 // If the method is static, get mirror into R4_ARG2. 1429 { 1430 Label method_is_not_static; 1431 // Access_flags is non-volatile and still, no need to restore it. 1432 1433 // Restore access flags. 1434 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT); 1435 __ bfalse(CCR0, method_is_not_static); 1436 1437 __ ld(R11_scratch1, _abi(callers_sp), R1_SP); 1438 // Load mirror from interpreter frame. 1439 __ ld(R12_scratch2, _ijava_state_neg(mirror), R11_scratch1); 1440 // R4_ARG2 = &state->_oop_temp; 1441 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp)); 1442 __ std(R12_scratch2/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); 1443 BIND(method_is_not_static); 1444 } 1445 1446 // At this point, arguments have been copied off the stack into 1447 // their JNI positions. Oops are boxed in-place on the stack, with 1448 // handles copied to arguments. The result handler address is in a 1449 // register. 1450 1451 // Pass JNIEnv address as first parameter. 1452 __ addir(R3_ARG1, thread_(jni_environment)); 1453 1454 // Load the native_method entry before we change the thread state. 1455 __ ld(native_method_fd, method_(native_function)); 1456 1457 //============================================================================= 1458 // Transition from _thread_in_Java to _thread_in_native. As soon as 1459 // we make this change the safepoint code needs to be certain that 1460 // the last Java frame we established is good. The pc in that frame 1461 // just needs to be near here not an actual return address. 1462 1463 // We use release_store_fence to update values like the thread state, where 1464 // we don't want the current thread to continue until all our prior memory 1465 // accesses (including the new thread state) are visible to other threads. 1466 __ li(R0, _thread_in_native); 1467 __ release(); 1468 1469 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size"); 1470 __ stw(R0, thread_(thread_state)); 1471 1472 if (UseMembar) { 1473 __ fence(); 1474 } 1475 1476 //============================================================================= 1477 // Call the native method. Argument registers must not have been 1478 // overwritten since "__ call_stub(signature_handler);" (except for 1479 // ARG1 and ARG2 for static methods). 1480 __ call_c(native_method_fd); 1481 1482 __ li(R0, 0); 1483 __ ld(R11_scratch1, 0, R1_SP); 1484 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1); 1485 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1); 1486 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset 1487 1488 // Note: C++ interpreter needs the following here: 1489 // The frame_manager_lr field, which we use for setting the last 1490 // java frame, gets overwritten by the signature handler. Restore 1491 // it now. 1492 //__ get_PC_trash_LR(R11_scratch1); 1493 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 1494 1495 // Because of GC R19_method may no longer be valid. 1496 1497 // Block, if necessary, before resuming in _thread_in_Java state. 1498 // In order for GC to work, don't clear the last_Java_sp until after 1499 // blocking. 1500 1501 //============================================================================= 1502 // Switch thread to "native transition" state before reading the 1503 // synchronization state. This additional state is necessary 1504 // because reading and testing the synchronization state is not 1505 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A, 1506 // in _thread_in_native state, loads _not_synchronized and is 1507 // preempted. VM thread changes sync state to synchronizing and 1508 // suspends threads for GC. Thread A is resumed to finish this 1509 // native method, but doesn't block here since it didn't see any 1510 // synchronization in progress, and escapes. 1511 1512 // We use release_store_fence to update values like the thread state, where 1513 // we don't want the current thread to continue until all our prior memory 1514 // accesses (including the new thread state) are visible to other threads. 1515 __ li(R0/*thread_state*/, _thread_in_native_trans); 1516 __ release(); 1517 __ stw(R0/*thread_state*/, thread_(thread_state)); 1518 if (UseMembar) { 1519 __ fence(); 1520 } 1521 // Write serialization page so that the VM thread can do a pseudo remote 1522 // membar. We use the current thread pointer to calculate a thread 1523 // specific offset to write to within the page. This minimizes bus 1524 // traffic due to cache line collision. 1525 else { 1526 __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2); 1527 } 1528 1529 // Now before we return to java we must look for a current safepoint 1530 // (a new safepoint can not start since we entered native_trans). 1531 // We must check here because a current safepoint could be modifying 1532 // the callers registers right this moment. 1533 1534 // Acquire isn't strictly necessary here because of the fence, but 1535 // sync_state is declared to be volatile, so we do it anyway 1536 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path). 1537 int sync_state_offs = __ load_const_optimized(sync_state_addr, SafepointSynchronize::address_of_state(), /*temp*/R0, true); 1538 1539 // TODO PPC port assert(4 == SafepointSynchronize::sz_state(), "unexpected field size"); 1540 __ lwz(sync_state, sync_state_offs, sync_state_addr); 1541 1542 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size"); 1543 __ lwz(suspend_flags, thread_(suspend_flags)); 1544 1545 Label sync_check_done; 1546 Label do_safepoint; 1547 // No synchronization in progress nor yet synchronized. 1548 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized); 1549 // Not suspended. 1550 __ cmpwi(CCR1, suspend_flags, 0); 1551 1552 __ bne(CCR0, do_safepoint); 1553 __ beq(CCR1, sync_check_done); 1554 __ bind(do_safepoint); 1555 __ isync(); 1556 // Block. We do the call directly and leave the current 1557 // last_Java_frame setup undisturbed. We must save any possible 1558 // native result across the call. No oop is present. 1559 1560 __ mr(R3_ARG1, R16_thread); 1561 #if defined(ABI_ELFv2) 1562 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), 1563 relocInfo::none); 1564 #else 1565 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans), 1566 relocInfo::none); 1567 #endif 1568 1569 __ bind(sync_check_done); 1570 1571 //============================================================================= 1572 // <<<<<< Back in Interpreter Frame >>>>> 1573 1574 // We are in thread_in_native_trans here and back in the normal 1575 // interpreter frame. We don't have to do anything special about 1576 // safepoints and we can switch to Java mode anytime we are ready. 1577 1578 // Note: frame::interpreter_frame_result has a dependency on how the 1579 // method result is saved across the call to post_method_exit. For 1580 // native methods it assumes that the non-FPU/non-void result is 1581 // saved in _native_lresult and a FPU result in _native_fresult. If 1582 // this changes then the interpreter_frame_result implementation 1583 // will need to be updated too. 1584 1585 // On PPC64, we have stored the result directly after the native call. 1586 1587 //============================================================================= 1588 // Back in Java 1589 1590 // We use release_store_fence to update values like the thread state, where 1591 // we don't want the current thread to continue until all our prior memory 1592 // accesses (including the new thread state) are visible to other threads. 1593 __ li(R0/*thread_state*/, _thread_in_Java); 1594 __ release(); 1595 __ stw(R0/*thread_state*/, thread_(thread_state)); 1596 if (UseMembar) { 1597 __ fence(); 1598 } 1599 1600 if (CheckJNICalls) { 1601 // clear_pending_jni_exception_check 1602 __ load_const_optimized(R0, 0L); 1603 __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread); 1604 } 1605 1606 __ reset_last_Java_frame(); 1607 1608 // Jvmdi/jvmpi support. Whether we've got an exception pending or 1609 // not, and whether unlocking throws an exception or not, we notify 1610 // on native method exit. If we do have an exception, we'll end up 1611 // in the caller's context to handle it, so if we don't do the 1612 // notify here, we'll drop it on the floor. 1613 __ notify_method_exit(true/*native method*/, 1614 ilgl /*illegal state (not used for native methods)*/, 1615 InterpreterMacroAssembler::NotifyJVMTI, 1616 false /*check_exceptions*/); 1617 1618 //============================================================================= 1619 // Handle exceptions 1620 1621 if (synchronized) { 1622 // Don't check for exceptions since we're still in the i2n frame. Do that 1623 // manually afterwards. 1624 __ unlock_object(R26_monitor, false); // Can also unlock methods. 1625 } 1626 1627 // Reset active handles after returning from native. 1628 // thread->active_handles()->clear(); 1629 __ ld(active_handles, thread_(active_handles)); 1630 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size"); 1631 __ li(R0, 0); 1632 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles); 1633 1634 Label exception_return_sync_check_already_unlocked; 1635 __ ld(R0/*pending_exception*/, thread_(pending_exception)); 1636 __ cmpdi(CCR0, R0/*pending_exception*/, 0); 1637 __ bne(CCR0, exception_return_sync_check_already_unlocked); 1638 1639 //----------------------------------------------------------------------------- 1640 // No exception pending. 1641 1642 // Move native method result back into proper registers and return. 1643 // Invoke result handler (may unbox/promote). 1644 __ ld(R11_scratch1, 0, R1_SP); 1645 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1); 1646 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1); 1647 __ call_stub(result_handler_addr); 1648 1649 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1650 1651 // Must use the return pc which was loaded from the caller's frame 1652 // as the VM uses return-pc-patching for deoptimization. 1653 __ mtlr(R0); 1654 __ blr(); 1655 1656 //----------------------------------------------------------------------------- 1657 // An exception is pending. We call into the runtime only if the 1658 // caller was not interpreted. If it was interpreted the 1659 // interpreter will do the correct thing. If it isn't interpreted 1660 // (call stub/compiled code) we will change our return and continue. 1661 1662 BIND(exception_return_sync_check); 1663 1664 if (synchronized) { 1665 // Don't check for exceptions since we're still in the i2n frame. Do that 1666 // manually afterwards. 1667 __ unlock_object(R26_monitor, false); // Can also unlock methods. 1668 } 1669 BIND(exception_return_sync_check_already_unlocked); 1670 1671 const Register return_pc = R31; 1672 1673 __ ld(return_pc, 0, R1_SP); 1674 __ ld(return_pc, _abi(lr), return_pc); 1675 1676 // Get the address of the exception handler. 1677 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), 1678 R16_thread, 1679 return_pc /* return pc */); 1680 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2); 1681 1682 // Load the PC of the the exception handler into LR. 1683 __ mtlr(R3_RET); 1684 1685 // Load exception into R3_ARG1 and clear pending exception in thread. 1686 __ ld(R3_ARG1/*exception*/, thread_(pending_exception)); 1687 __ li(R4_ARG2, 0); 1688 __ std(R4_ARG2, thread_(pending_exception)); 1689 1690 // Load the original return pc into R4_ARG2. 1691 __ mr(R4_ARG2/*issuing_pc*/, return_pc); 1692 1693 // Return to exception handler. 1694 __ blr(); 1695 1696 //============================================================================= 1697 // Counter overflow. 1698 1699 if (inc_counter) { 1700 // Handle invocation counter overflow. 1701 __ bind(invocation_counter_overflow); 1702 1703 generate_counter_overflow(continue_after_compile); 1704 } 1705 1706 return entry; 1707 } 1708 1709 // Generic interpreted method entry to (asm) interpreter. 1710 // 1711 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) { 1712 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; 1713 address entry = __ pc(); 1714 // Generate the code to allocate the interpreter stack frame. 1715 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame. 1716 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame. 1717 1718 // Does also a stack check to assure this frame fits on the stack. 1719 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals); 1720 1721 // -------------------------------------------------------------------------- 1722 // Zero out non-parameter locals. 1723 // Note: *Always* zero out non-parameter locals as Sparc does. It's not 1724 // worth to ask the flag, just do it. 1725 Register Rslot_addr = R6_ARG4, 1726 Rnum = R7_ARG5; 1727 Label Lno_locals, Lzero_loop; 1728 1729 // Set up the zeroing loop. 1730 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals); 1731 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals); 1732 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize); 1733 __ beq(CCR0, Lno_locals); 1734 __ li(R0, 0); 1735 __ mtctr(Rnum); 1736 1737 // The zero locals loop. 1738 __ bind(Lzero_loop); 1739 __ std(R0, 0, Rslot_addr); 1740 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize); 1741 __ bdnz(Lzero_loop); 1742 1743 __ bind(Lno_locals); 1744 1745 // -------------------------------------------------------------------------- 1746 // Counter increment and overflow check. 1747 Label invocation_counter_overflow, 1748 profile_method, 1749 profile_method_continue; 1750 if (inc_counter || ProfileInterpreter) { 1751 1752 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1; 1753 if (synchronized) { 1754 // Since at this point in the method invocation the exception handler 1755 // would try to exit the monitor of synchronized methods which hasn't 1756 // been entered yet, we set the thread local variable 1757 // _do_not_unlock_if_synchronized to true. If any exception was thrown by 1758 // runtime, exception handling i.e. unlock_if_synchronized_method will 1759 // check this thread local flag. 1760 // This flag has two effects, one is to force an unwind in the topmost 1761 // interpreter frame and not perform an unlock while doing so. 1762 __ li(R0, 1); 1763 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1764 } 1765 1766 // Argument and return type profiling. 1767 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4); 1768 1769 // Increment invocation counter and check for overflow. 1770 if (inc_counter) { 1771 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue); 1772 } 1773 1774 __ bind(profile_method_continue); 1775 } 1776 1777 bang_stack_shadow_pages(false); 1778 1779 if (inc_counter || ProfileInterpreter) { 1780 // Reset the _do_not_unlock_if_synchronized flag. 1781 if (synchronized) { 1782 __ li(R0, 0); 1783 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1784 } 1785 } 1786 1787 // -------------------------------------------------------------------------- 1788 // Locking of synchronized methods. Must happen AFTER invocation_counter 1789 // check and stack overflow check, so method is not locked if overflows. 1790 if (synchronized) { 1791 lock_method(R3_ARG1, R4_ARG2, R5_ARG3); 1792 } 1793 #ifdef ASSERT 1794 else { 1795 Label Lok; 1796 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method); 1797 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED); 1798 __ asm_assert_eq("method needs synchronization", 0x8521); 1799 __ bind(Lok); 1800 } 1801 #endif // ASSERT 1802 1803 __ verify_thread(); 1804 1805 // -------------------------------------------------------------------------- 1806 // JVMTI support 1807 __ notify_method_entry(); 1808 1809 // -------------------------------------------------------------------------- 1810 // Start executing instructions. 1811 __ dispatch_next(vtos); 1812 1813 // -------------------------------------------------------------------------- 1814 // Out of line counter overflow and MDO creation code. 1815 if (ProfileInterpreter) { 1816 // We have decided to profile this method in the interpreter. 1817 __ bind(profile_method); 1818 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1819 __ set_method_data_pointer_for_bcp(); 1820 __ b(profile_method_continue); 1821 } 1822 1823 if (inc_counter) { 1824 // Handle invocation counter overflow. 1825 __ bind(invocation_counter_overflow); 1826 generate_counter_overflow(profile_method_continue); 1827 } 1828 return entry; 1829 } 1830 1831 // CRC32 Intrinsics. 1832 // 1833 // Contract on scratch and work registers. 1834 // ======================================= 1835 // 1836 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers. 1837 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set. 1838 // You can't rely on these registers across calls. 1839 // 1840 // The generators for CRC32_update and for CRC32_updateBytes use the 1841 // scratch/work register set internally, passing the work registers 1842 // as arguments to the MacroAssembler emitters as required. 1843 // 1844 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments. 1845 // Their contents is not constant but may change according to the requirements 1846 // of the emitted code. 1847 // 1848 // All other registers from the scratch/work register set are used "internally" 1849 // and contain garbage (i.e. unpredictable values) once blr() is reached. 1850 // Basically, only R3_RET contains a defined value which is the function result. 1851 // 1852 /** 1853 * Method entry for static native methods: 1854 * int java.util.zip.CRC32.update(int crc, int b) 1855 */ 1856 address TemplateInterpreterGenerator::generate_CRC32_update_entry() { 1857 if (UseCRC32Intrinsics) { 1858 address start = __ pc(); // Remember stub start address (is rtn value). 1859 Label slow_path; 1860 1861 // Safepoint check 1862 const Register sync_state = R11_scratch1; 1863 int sync_state_offs = __ load_const_optimized(sync_state, SafepointSynchronize::address_of_state(), /*temp*/R0, true); 1864 __ lwz(sync_state, sync_state_offs, sync_state); 1865 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized); 1866 __ bne(CCR0, slow_path); 1867 1868 // We don't generate local frame and don't align stack because 1869 // we not even call stub code (we generate the code inline) 1870 // and there is no safepoint on this path. 1871 1872 // Load java parameters. 1873 // R15_esp is callers operand stack pointer, i.e. it points to the parameters. 1874 const Register argP = R15_esp; 1875 const Register crc = R3_ARG1; // crc value 1876 const Register data = R4_ARG2; // address of java byte value (kernel_crc32 needs address) 1877 const Register dataLen = R5_ARG3; // source data len (1 byte). Not used because calling the single-byte emitter. 1878 const Register table = R6_ARG4; // address of crc32 table 1879 const Register tmp = dataLen; // Reuse unused len register to show we don't actually need a separate tmp here. 1880 1881 BLOCK_COMMENT("CRC32_update {"); 1882 1883 // Arguments are reversed on java expression stack 1884 #ifdef VM_LITTLE_ENDIAN 1885 __ addi(data, argP, 0+1*wordSize); // (stack) address of byte value. Emitter expects address, not value. 1886 // Being passed as an int, the single byte is at offset +0. 1887 #else 1888 __ addi(data, argP, 3+1*wordSize); // (stack) address of byte value. Emitter expects address, not value. 1889 // Being passed from java as an int, the single byte is at offset +3. 1890 #endif 1891 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register. 1892 1893 StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table); 1894 __ kernel_crc32_singleByte(crc, data, dataLen, table, tmp, true); 1895 1896 // Restore caller sp for c2i case and return. 1897 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 1898 __ blr(); 1899 1900 // Generate a vanilla native entry as the slow path. 1901 BLOCK_COMMENT("} CRC32_update"); 1902 BIND(slow_path); 1903 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1); 1904 return start; 1905 } 1906 1907 return NULL; 1908 } 1909 1910 // TODO: generate_CRC32_updateBytes_entry and generate_CRC32C_updateBytes_entry are identical 1911 // except for using different crc tables and some block comment strings. 1912 // We should provide a common implementation. 1913 1914 // CRC32 Intrinsics. 1915 /** 1916 * Method entry for static native methods: 1917 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len) 1918 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len) 1919 */ 1920 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 1921 if (UseCRC32Intrinsics) { 1922 address start = __ pc(); // Remember stub start address (is rtn value). 1923 Label slow_path; 1924 1925 // Safepoint check 1926 const Register sync_state = R11_scratch1; 1927 int sync_state_offs = __ load_const_optimized(sync_state, SafepointSynchronize::address_of_state(), /*temp*/R0, true); 1928 __ lwz(sync_state, sync_state_offs, sync_state); 1929 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized); 1930 __ bne(CCR0, slow_path); 1931 1932 // We don't generate local frame and don't align stack because 1933 // we not even call stub code (we generate the code inline) 1934 // and there is no safepoint on this path. 1935 1936 // Load parameters. 1937 // Z_esp is callers operand stack pointer, i.e. it points to the parameters. 1938 const Register argP = R15_esp; 1939 const Register crc = R3_ARG1; // crc value 1940 const Register data = R4_ARG2; // address of java byte array 1941 const Register dataLen = R5_ARG3; // source data len 1942 const Register table = R6_ARG4; // address of crc32 table 1943 1944 const Register t0 = R9; // scratch registers for crc calculation 1945 const Register t1 = R10; 1946 const Register t2 = R11; 1947 const Register t3 = R12; 1948 1949 const Register tc0 = R2; // registers to hold pre-calculated column addresses 1950 const Register tc1 = R7; 1951 const Register tc2 = R8; 1952 const Register tc3 = table; // table address is reconstructed at the end of kernel_crc32_* emitters 1953 1954 const Register tmp = t0; // Only used very locally to calculate byte buffer address. 1955 1956 // Arguments are reversed on java expression stack. 1957 // Calculate address of start element. 1958 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct". 1959 BLOCK_COMMENT("CRC32_updateByteBuffer {"); 1960 // crc @ (SP + 5W) (32bit) 1961 // buf @ (SP + 3W) (64bit ptr to long array) 1962 // off @ (SP + 2W) (32bit) 1963 // dataLen @ (SP + 1W) (32bit) 1964 // data = buf + off 1965 __ ld( data, 3*wordSize, argP); // start of byte buffer 1966 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 1967 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 1968 __ lwz( crc, 5*wordSize, argP); // current crc state 1969 __ add( data, data, tmp); // Add byte buffer offset. 1970 } else { // Used for "updateBytes update". 1971 BLOCK_COMMENT("CRC32_updateBytes {"); 1972 // crc @ (SP + 4W) (32bit) 1973 // buf @ (SP + 3W) (64bit ptr to byte array) 1974 // off @ (SP + 2W) (32bit) 1975 // dataLen @ (SP + 1W) (32bit) 1976 // data = buf + off + base_offset 1977 __ ld( data, 3*wordSize, argP); // start of byte buffer 1978 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 1979 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 1980 __ add( data, data, tmp); // add byte buffer offset 1981 __ lwz( crc, 4*wordSize, argP); // current crc state 1982 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1983 } 1984 1985 StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table); 1986 1987 // Performance measurements show the 1word and 2word variants to be almost equivalent, 1988 // with very light advantages for the 1word variant. We chose the 1word variant for 1989 // code compactness. 1990 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3, true); 1991 1992 // Restore caller sp for c2i case and return. 1993 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 1994 __ blr(); 1995 1996 // Generate a vanilla native entry as the slow path. 1997 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)"); 1998 BIND(slow_path); 1999 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1); 2000 return start; 2001 } 2002 2003 return NULL; 2004 } 2005 2006 // CRC32C Intrinsics. 2007 /** 2008 * Method entry for static native methods: 2009 * int java.util.zip.CRC32C.updateBytes( int crc, byte[] b, int off, int len) 2010 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int len) 2011 **/ 2012 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 2013 if (UseCRC32CIntrinsics) { 2014 address start = __ pc(); // Remember stub start address (is rtn value). 2015 2016 // We don't generate local frame and don't align stack because 2017 // we not even call stub code (we generate the code inline) 2018 // and there is no safepoint on this path. 2019 2020 // Load parameters. 2021 // Z_esp is callers operand stack pointer, i.e. it points to the parameters. 2022 const Register argP = R15_esp; 2023 const Register crc = R3_ARG1; // crc value 2024 const Register data = R4_ARG2; // address of java byte array 2025 const Register dataLen = R5_ARG3; // source data len 2026 const Register table = R6_ARG4; // address of crc32c table 2027 2028 const Register t0 = R9; // scratch registers for crc calculation 2029 const Register t1 = R10; 2030 const Register t2 = R11; 2031 const Register t3 = R12; 2032 2033 const Register tc0 = R2; // registers to hold pre-calculated column addresses 2034 const Register tc1 = R7; 2035 const Register tc2 = R8; 2036 const Register tc3 = table; // table address is reconstructed at the end of kernel_crc32_* emitters 2037 2038 const Register tmp = t0; // Only used very locally to calculate byte buffer address. 2039 2040 // Arguments are reversed on java expression stack. 2041 // Calculate address of start element. 2042 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer". 2043 BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {"); 2044 // crc @ (SP + 5W) (32bit) 2045 // buf @ (SP + 3W) (64bit ptr to long array) 2046 // off @ (SP + 2W) (32bit) 2047 // dataLen @ (SP + 1W) (32bit) 2048 // data = buf + off 2049 __ ld( data, 3*wordSize, argP); // start of byte buffer 2050 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 2051 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 2052 __ lwz( crc, 5*wordSize, argP); // current crc state 2053 __ add( data, data, tmp); // Add byte buffer offset. 2054 } else { // Used for "updateBytes update". 2055 BLOCK_COMMENT("CRC32C_updateBytes {"); 2056 // crc @ (SP + 4W) (32bit) 2057 // buf @ (SP + 3W) (64bit ptr to byte array) 2058 // off @ (SP + 2W) (32bit) 2059 // dataLen @ (SP + 1W) (32bit) 2060 // data = buf + off + base_offset 2061 __ ld( data, 3*wordSize, argP); // start of byte buffer 2062 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 2063 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 2064 __ add( data, data, tmp); // add byte buffer offset 2065 __ lwz( crc, 4*wordSize, argP); // current crc state 2066 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 2067 } 2068 2069 StubRoutines::ppc64::generate_load_crc32c_table_addr(_masm, table); 2070 2071 // Performance measurements show the 1word and 2word variants to be almost equivalent, 2072 // with very light advantages for the 1word variant. We chose the 1word variant for 2073 // code compactness. 2074 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3, false); 2075 2076 // Restore caller sp for c2i case and return. 2077 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 2078 __ blr(); 2079 2080 BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}"); 2081 return start; 2082 } 2083 2084 return NULL; 2085 } 2086 2087 // ============================================================================= 2088 // Exceptions 2089 2090 void TemplateInterpreterGenerator::generate_throw_exception() { 2091 Register Rexception = R17_tos, 2092 Rcontinuation = R3_RET; 2093 2094 // -------------------------------------------------------------------------- 2095 // Entry point if an method returns with a pending exception (rethrow). 2096 Interpreter::_rethrow_exception_entry = __ pc(); 2097 { 2098 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp. 2099 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); 2100 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); 2101 2102 // Compiled code destroys templateTableBase, reload. 2103 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1); 2104 } 2105 2106 // Entry point if a interpreted method throws an exception (throw). 2107 Interpreter::_throw_exception_entry = __ pc(); 2108 { 2109 __ mr(Rexception, R3_RET); 2110 2111 __ verify_thread(); 2112 __ verify_oop(Rexception); 2113 2114 // Expression stack must be empty before entering the VM in case of an exception. 2115 __ empty_expression_stack(); 2116 // Find exception handler address and preserve exception oop. 2117 // Call C routine to find handler and jump to it. 2118 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception); 2119 __ mtctr(Rcontinuation); 2120 // Push exception for exception handler bytecodes. 2121 __ push_ptr(Rexception); 2122 2123 // Jump to exception handler (may be remove activation entry!). 2124 __ bctr(); 2125 } 2126 2127 // If the exception is not handled in the current frame the frame is 2128 // removed and the exception is rethrown (i.e. exception 2129 // continuation is _rethrow_exception). 2130 // 2131 // Note: At this point the bci is still the bxi for the instruction 2132 // which caused the exception and the expression stack is 2133 // empty. Thus, for any VM calls at this point, GC will find a legal 2134 // oop map (with empty expression stack). 2135 2136 // In current activation 2137 // tos: exception 2138 // bcp: exception bcp 2139 2140 // -------------------------------------------------------------------------- 2141 // JVMTI PopFrame support 2142 2143 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 2144 { 2145 // Set the popframe_processing bit in popframe_condition indicating that we are 2146 // currently handling popframe, so that call_VMs that may happen later do not 2147 // trigger new popframe handling cycles. 2148 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 2149 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit); 2150 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 2151 2152 // Empty the expression stack, as in normal exception handling. 2153 __ empty_expression_stack(); 2154 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false); 2155 2156 // Check to see whether we are returning to a deoptimized frame. 2157 // (The PopFrame call ensures that the caller of the popped frame is 2158 // either interpreted or compiled and deoptimizes it if compiled.) 2159 // Note that we don't compare the return PC against the 2160 // deoptimization blob's unpack entry because of the presence of 2161 // adapter frames in C2. 2162 Label Lcaller_not_deoptimized; 2163 Register return_pc = R3_ARG1; 2164 __ ld(return_pc, 0, R1_SP); 2165 __ ld(return_pc, _abi(lr), return_pc); 2166 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc); 2167 __ cmpdi(CCR0, R3_RET, 0); 2168 __ bne(CCR0, Lcaller_not_deoptimized); 2169 2170 // The deoptimized case. 2171 // In this case, we can't call dispatch_next() after the frame is 2172 // popped, but instead must save the incoming arguments and restore 2173 // them after deoptimization has occurred. 2174 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method); 2175 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2); 2176 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize); 2177 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize); 2178 __ subf(R5_ARG3, R4_ARG2, R5_ARG3); 2179 // Save these arguments. 2180 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3); 2181 2182 // Inform deoptimization that it is responsible for restoring these arguments. 2183 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit); 2184 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 2185 2186 // Return from the current method into the deoptimization blob. Will eventually 2187 // end up in the deopt interpeter entry, deoptimization prepared everything that 2188 // we will reexecute the call that called us. 2189 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2); 2190 __ mtlr(return_pc); 2191 __ blr(); 2192 2193 // The non-deoptimized case. 2194 __ bind(Lcaller_not_deoptimized); 2195 2196 // Clear the popframe condition flag. 2197 __ li(R0, 0); 2198 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 2199 2200 // Get out of the current method and re-execute the call that called us. 2201 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2); 2202 __ restore_interpreter_state(R11_scratch1); 2203 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); 2204 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); 2205 if (ProfileInterpreter) { 2206 __ set_method_data_pointer_for_bcp(); 2207 __ ld(R11_scratch1, 0, R1_SP); 2208 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1); 2209 } 2210 #if INCLUDE_JVMTI 2211 Label L_done; 2212 2213 __ lbz(R11_scratch1, 0, R14_bcp); 2214 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic); 2215 __ bne(CCR0, L_done); 2216 2217 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. 2218 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL. 2219 __ ld(R4_ARG2, 0, R18_locals); 2220 __ MacroAssembler::call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp, false); 2221 __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); 2222 __ cmpdi(CCR0, R4_ARG2, 0); 2223 __ beq(CCR0, L_done); 2224 __ std(R4_ARG2, wordSize, R15_esp); 2225 __ bind(L_done); 2226 #endif // INCLUDE_JVMTI 2227 __ dispatch_next(vtos); 2228 } 2229 // end of JVMTI PopFrame support 2230 2231 // -------------------------------------------------------------------------- 2232 // Remove activation exception entry. 2233 // This is jumped to if an interpreted method can't handle an exception itself 2234 // (we come from the throw/rethrow exception entry above). We're going to call 2235 // into the VM to find the exception handler in the caller, pop the current 2236 // frame and return the handler we calculated. 2237 Interpreter::_remove_activation_entry = __ pc(); 2238 { 2239 __ pop_ptr(Rexception); 2240 __ verify_thread(); 2241 __ verify_oop(Rexception); 2242 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread); 2243 2244 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true); 2245 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false); 2246 2247 __ get_vm_result(Rexception); 2248 2249 // We are done with this activation frame; find out where to go next. 2250 // The continuation point will be an exception handler, which expects 2251 // the following registers set up: 2252 // 2253 // RET: exception oop 2254 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled. 2255 2256 Register return_pc = R31; // Needs to survive the runtime call. 2257 __ ld(return_pc, 0, R1_SP); 2258 __ ld(return_pc, _abi(lr), return_pc); 2259 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc); 2260 2261 // Remove the current activation. 2262 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2); 2263 2264 __ mr(R4_ARG2, return_pc); 2265 __ mtlr(R3_RET); 2266 __ mr(R3_RET, Rexception); 2267 __ blr(); 2268 } 2269 } 2270 2271 // JVMTI ForceEarlyReturn support. 2272 // Returns "in the middle" of a method with a "fake" return value. 2273 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 2274 2275 Register Rscratch1 = R11_scratch1, 2276 Rscratch2 = R12_scratch2; 2277 2278 address entry = __ pc(); 2279 __ empty_expression_stack(); 2280 2281 __ load_earlyret_value(state, Rscratch1); 2282 2283 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 2284 // Clear the earlyret state. 2285 __ li(R0, 0); 2286 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1); 2287 2288 __ remove_activation(state, false, false); 2289 // Copied from TemplateTable::_return. 2290 // Restoration of lr done by remove_activation. 2291 switch (state) { 2292 // Narrow result if state is itos but result type is smaller. 2293 case btos: 2294 case ztos: 2295 case ctos: 2296 case stos: 2297 case itos: __ narrow(R17_tos); /* fall through */ 2298 case ltos: 2299 case atos: __ mr(R3_RET, R17_tos); break; 2300 case ftos: 2301 case dtos: __ fmr(F1_RET, F15_ftos); break; 2302 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need 2303 // to get visible before the reference to the object gets stored anywhere. 2304 __ membar(Assembler::StoreStore); break; 2305 default : ShouldNotReachHere(); 2306 } 2307 __ blr(); 2308 2309 return entry; 2310 } // end of ForceEarlyReturn support 2311 2312 //----------------------------------------------------------------------------- 2313 // Helper for vtos entry point generation 2314 2315 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, 2316 address& bep, 2317 address& cep, 2318 address& sep, 2319 address& aep, 2320 address& iep, 2321 address& lep, 2322 address& fep, 2323 address& dep, 2324 address& vep) { 2325 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 2326 Label L; 2327 2328 aep = __ pc(); __ push_ptr(); __ b(L); 2329 fep = __ pc(); __ push_f(); __ b(L); 2330 dep = __ pc(); __ push_d(); __ b(L); 2331 lep = __ pc(); __ push_l(); __ b(L); 2332 __ align(32, 12, 24); // align L 2333 bep = cep = sep = 2334 iep = __ pc(); __ push_i(); 2335 vep = __ pc(); 2336 __ bind(L); 2337 generate_and_dispatch(t); 2338 } 2339 2340 //----------------------------------------------------------------------------- 2341 2342 // Non-product code 2343 #ifndef PRODUCT 2344 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 2345 //__ flush_bundle(); 2346 address entry = __ pc(); 2347 2348 const char *bname = NULL; 2349 uint tsize = 0; 2350 switch(state) { 2351 case ftos: 2352 bname = "trace_code_ftos {"; 2353 tsize = 2; 2354 break; 2355 case btos: 2356 bname = "trace_code_btos {"; 2357 tsize = 2; 2358 break; 2359 case ztos: 2360 bname = "trace_code_ztos {"; 2361 tsize = 2; 2362 break; 2363 case ctos: 2364 bname = "trace_code_ctos {"; 2365 tsize = 2; 2366 break; 2367 case stos: 2368 bname = "trace_code_stos {"; 2369 tsize = 2; 2370 break; 2371 case itos: 2372 bname = "trace_code_itos {"; 2373 tsize = 2; 2374 break; 2375 case ltos: 2376 bname = "trace_code_ltos {"; 2377 tsize = 3; 2378 break; 2379 case atos: 2380 bname = "trace_code_atos {"; 2381 tsize = 2; 2382 break; 2383 case vtos: 2384 // Note: In case of vtos, the topmost of stack value could be a int or doubl 2385 // In case of a double (2 slots) we won't see the 2nd stack value. 2386 // Maybe we simply should print the topmost 3 stack slots to cope with the problem. 2387 bname = "trace_code_vtos {"; 2388 tsize = 2; 2389 2390 break; 2391 case dtos: 2392 bname = "trace_code_dtos {"; 2393 tsize = 3; 2394 break; 2395 default: 2396 ShouldNotReachHere(); 2397 } 2398 BLOCK_COMMENT(bname); 2399 2400 // Support short-cut for TraceBytecodesAt. 2401 // Don't call into the VM if we don't want to trace to speed up things. 2402 Label Lskip_vm_call; 2403 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) { 2404 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true); 2405 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true); 2406 __ ld(R11_scratch1, offs1, R11_scratch1); 2407 __ lwa(R12_scratch2, offs2, R12_scratch2); 2408 __ cmpd(CCR0, R12_scratch2, R11_scratch1); 2409 __ blt(CCR0, Lskip_vm_call); 2410 } 2411 2412 __ push(state); 2413 // Load 2 topmost expression stack values. 2414 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp); 2415 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp); 2416 __ mflr(R31); 2417 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false); 2418 __ mtlr(R31); 2419 __ pop(state); 2420 2421 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) { 2422 __ bind(Lskip_vm_call); 2423 } 2424 __ blr(); 2425 BLOCK_COMMENT("} trace_code"); 2426 return entry; 2427 } 2428 2429 void TemplateInterpreterGenerator::count_bytecode() { 2430 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true); 2431 __ lwz(R12_scratch2, offs, R11_scratch1); 2432 __ addi(R12_scratch2, R12_scratch2, 1); 2433 __ stw(R12_scratch2, offs, R11_scratch1); 2434 } 2435 2436 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { 2437 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true); 2438 __ lwz(R12_scratch2, offs, R11_scratch1); 2439 __ addi(R12_scratch2, R12_scratch2, 1); 2440 __ stw(R12_scratch2, offs, R11_scratch1); 2441 } 2442 2443 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { 2444 const Register addr = R11_scratch1, 2445 tmp = R12_scratch2; 2446 // Get index, shift out old bytecode, bring in new bytecode, and store it. 2447 // _index = (_index >> log2_number_of_codes) | 2448 // (bytecode << log2_number_of_codes); 2449 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true); 2450 __ lwz(tmp, offs1, addr); 2451 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes); 2452 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes); 2453 __ stw(tmp, offs1, addr); 2454 2455 // Bump bucket contents. 2456 // _counters[_index] ++; 2457 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true); 2458 __ sldi(tmp, tmp, LogBytesPerInt); 2459 __ add(addr, tmp, addr); 2460 __ lwz(tmp, offs2, addr); 2461 __ addi(tmp, tmp, 1); 2462 __ stw(tmp, offs2, addr); 2463 } 2464 2465 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 2466 // Call a little run-time stub to avoid blow-up for each bytecode. 2467 // The run-time runtime saves the right registers, depending on 2468 // the tosca in-state for the given template. 2469 2470 assert(Interpreter::trace_code(t->tos_in()) != NULL, 2471 "entry must have been generated"); 2472 2473 // Note: we destroy LR here. 2474 __ bl(Interpreter::trace_code(t->tos_in())); 2475 } 2476 2477 void TemplateInterpreterGenerator::stop_interpreter_at() { 2478 Label L; 2479 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true); 2480 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true); 2481 __ ld(R11_scratch1, offs1, R11_scratch1); 2482 __ lwa(R12_scratch2, offs2, R12_scratch2); 2483 __ cmpd(CCR0, R12_scratch2, R11_scratch1); 2484 __ bne(CCR0, L); 2485 __ illtrap(); 2486 __ bind(L); 2487 } 2488 2489 #endif // !PRODUCT