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 = 256*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 // JNIHandles::resolve result. 405 __ resolve_jobject(R3_RET, R11_scratch1, R12_scratch2, /* needs_frame */ true); // kills R31 406 break; 407 case T_FLOAT: 408 break; 409 case T_DOUBLE: 410 break; 411 case T_VOID: 412 break; 413 default: ShouldNotReachHere(); 414 } 415 416 BIND(done); 417 __ blr(); 418 419 return entry; 420 } 421 422 // Abstract method entry. 423 // 424 address TemplateInterpreterGenerator::generate_abstract_entry(void) { 425 address entry = __ pc(); 426 427 // 428 // Registers alive 429 // R16_thread - JavaThread* 430 // R19_method - callee's method (method to be invoked) 431 // R1_SP - SP prepared such that caller's outgoing args are near top 432 // LR - return address to caller 433 // 434 // Stack layout at this point: 435 // 436 // 0 [TOP_IJAVA_FRAME_ABI] <-- R1_SP 437 // alignment (optional) 438 // [outgoing Java arguments] 439 // ... 440 // PARENT [PARENT_IJAVA_FRAME_ABI] 441 // ... 442 // 443 444 // Can't use call_VM here because we have not set up a new 445 // interpreter state. Make the call to the vm and make it look like 446 // our caller set up the JavaFrameAnchor. 447 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/); 448 449 // Push a new C frame and save LR. 450 __ save_LR_CR(R0); 451 __ push_frame_reg_args(0, R11_scratch1); 452 453 // This is not a leaf but we have a JavaFrameAnchor now and we will 454 // check (create) exceptions afterward so this is ok. 455 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError), 456 R16_thread); 457 458 // Pop the C frame and restore LR. 459 __ pop_frame(); 460 __ restore_LR_CR(R0); 461 462 // Reset JavaFrameAnchor from call_VM_leaf above. 463 __ reset_last_Java_frame(); 464 465 // We don't know our caller, so jump to the general forward exception stub, 466 // which will also pop our full frame off. Satisfy the interface of 467 // SharedRuntime::generate_forward_exception() 468 __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0); 469 __ mtctr(R11_scratch1); 470 __ bctr(); 471 472 return entry; 473 } 474 475 // Interpreter intrinsic for WeakReference.get(). 476 // 1. Don't push a full blown frame and go on dispatching, but fetch the value 477 // into R8 and return quickly 478 // 2. If G1 is active we *must* execute this intrinsic for corrrectness: 479 // It contains a GC barrier which puts the reference into the satb buffer 480 // to indicate that someone holds a strong reference to the object the 481 // weak ref points to! 482 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) { 483 // Code: _aload_0, _getfield, _areturn 484 // parameter size = 1 485 // 486 // The code that gets generated by this routine is split into 2 parts: 487 // 1. the "intrinsified" code for G1 (or any SATB based GC), 488 // 2. the slow path - which is an expansion of the regular method entry. 489 // 490 // Notes: 491 // * In the G1 code we do not check whether we need to block for 492 // a safepoint. If G1 is enabled then we must execute the specialized 493 // code for Reference.get (except when the Reference object is null) 494 // so that we can log the value in the referent field with an SATB 495 // update buffer. 496 // If the code for the getfield template is modified so that the 497 // G1 pre-barrier code is executed when the current method is 498 // Reference.get() then going through the normal method entry 499 // will be fine. 500 // * The G1 code can, however, check the receiver object (the instance 501 // of java.lang.Reference) and jump to the slow path if null. If the 502 // Reference object is null then we obviously cannot fetch the referent 503 // and so we don't need to call the G1 pre-barrier. Thus we can use the 504 // regular method entry code to generate the NPE. 505 // 506 507 if (UseG1GC) { 508 address entry = __ pc(); 509 510 const int referent_offset = java_lang_ref_Reference::referent_offset; 511 guarantee(referent_offset > 0, "referent offset not initialized"); 512 513 Label slow_path; 514 515 // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH); 516 517 // In the G1 code we don't check if we need to reach a safepoint. We 518 // continue and the thread will safepoint at the next bytecode dispatch. 519 520 // If the receiver is null then it is OK to jump to the slow path. 521 __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver 522 523 // Check if receiver == NULL and go the slow path. 524 __ cmpdi(CCR0, R3_RET, 0); 525 __ beq(CCR0, slow_path); 526 527 // Load the value of the referent field. 528 __ load_heap_oop(R3_RET, referent_offset, R3_RET); 529 530 // Generate the G1 pre-barrier code to log the value of 531 // the referent field in an SATB buffer. Note with 532 // these parameters the pre-barrier does not generate 533 // the load of the previous value. 534 535 // Restore caller sp for c2i case. 536 #ifdef ASSERT 537 __ ld(R9_ARG7, 0, R1_SP); 538 __ ld(R10_ARG8, 0, R21_sender_SP); 539 __ cmpd(CCR0, R9_ARG7, R10_ARG8); 540 __ asm_assert_eq("backlink", 0x544); 541 #endif // ASSERT 542 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 543 544 __ g1_write_barrier_pre(noreg, // obj 545 noreg, // offset 546 R3_RET, // pre_val 547 R11_scratch1, // tmp 548 R12_scratch2, // tmp 549 true); // needs_frame 550 551 __ blr(); 552 553 // Generate regular method entry. 554 __ bind(slow_path); 555 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1); 556 return entry; 557 } 558 559 return NULL; 560 } 561 562 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { 563 address entry = __ pc(); 564 565 // Expression stack must be empty before entering the VM if an 566 // exception happened. 567 __ empty_expression_stack(); 568 // Throw exception. 569 __ call_VM(noreg, 570 CAST_FROM_FN_PTR(address, 571 InterpreterRuntime::throw_StackOverflowError)); 572 return entry; 573 } 574 575 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) { 576 address entry = __ pc(); 577 __ empty_expression_stack(); 578 __ load_const_optimized(R4_ARG2, (address) name); 579 // Index is in R17_tos. 580 __ mr(R5_ARG3, R17_tos); 581 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException)); 582 return entry; 583 } 584 585 #if 0 586 // Call special ClassCastException constructor taking object to cast 587 // and target class as arguments. 588 address TemplateInterpreterGenerator::generate_ClassCastException_verbose_handler() { 589 address entry = __ pc(); 590 591 // Expression stack must be empty before entering the VM if an 592 // exception happened. 593 __ empty_expression_stack(); 594 595 // Thread will be loaded to R3_ARG1. 596 // Target class oop is in register R5_ARG3 by convention! 597 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException_verbose), R17_tos, R5_ARG3); 598 // Above call must not return here since exception pending. 599 DEBUG_ONLY(__ should_not_reach_here();) 600 return entry; 601 } 602 #endif 603 604 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 605 address entry = __ pc(); 606 // Expression stack must be empty before entering the VM if an 607 // exception happened. 608 __ empty_expression_stack(); 609 610 // Load exception object. 611 // Thread will be loaded to R3_ARG1. 612 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos); 613 #ifdef ASSERT 614 // Above call must not return here since exception pending. 615 __ should_not_reach_here(); 616 #endif 617 return entry; 618 } 619 620 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) { 621 address entry = __ pc(); 622 //__ untested("generate_exception_handler_common"); 623 Register Rexception = R17_tos; 624 625 // Expression stack must be empty before entering the VM if an exception happened. 626 __ empty_expression_stack(); 627 628 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1); 629 if (pass_oop) { 630 __ mr(R5_ARG3, Rexception); 631 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), false); 632 } else { 633 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1); 634 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), false); 635 } 636 637 // Throw exception. 638 __ mr(R3_ARG1, Rexception); 639 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2); 640 __ mtctr(R11_scratch1); 641 __ bctr(); 642 643 return entry; 644 } 645 646 // This entry is returned to when a call returns to the interpreter. 647 // When we arrive here, we expect that the callee stack frame is already popped. 648 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) { 649 address entry = __ pc(); 650 651 // Move the value out of the return register back to the TOS cache of current frame. 652 switch (state) { 653 case ltos: 654 case btos: 655 case ztos: 656 case ctos: 657 case stos: 658 case atos: 659 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache 660 case ftos: 661 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET 662 case vtos: break; // Nothing to do, this was a void return. 663 default : ShouldNotReachHere(); 664 } 665 666 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp. 667 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); 668 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); 669 670 // Compiled code destroys templateTableBase, reload. 671 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2); 672 673 if (state == atos) { 674 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2); 675 } 676 677 const Register cache = R11_scratch1; 678 const Register size = R12_scratch2; 679 __ get_cache_and_index_at_bcp(cache, 1, index_size); 680 681 // Get least significant byte of 64 bit value: 682 #if defined(VM_LITTLE_ENDIAN) 683 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache); 684 #else 685 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache); 686 #endif 687 __ sldi(size, size, Interpreter::logStackElementSize); 688 __ add(R15_esp, R15_esp, size); 689 690 __ check_and_handle_popframe(R11_scratch1); 691 __ check_and_handle_earlyret(R11_scratch1); 692 693 __ dispatch_next(state, step); 694 return entry; 695 } 696 697 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) { 698 address entry = __ pc(); 699 // If state != vtos, we're returning from a native method, which put it's result 700 // into the result register. So move the value out of the return register back 701 // to the TOS cache of current frame. 702 703 switch (state) { 704 case ltos: 705 case btos: 706 case ztos: 707 case ctos: 708 case stos: 709 case atos: 710 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache 711 case ftos: 712 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET 713 case vtos: break; // Nothing to do, this was a void return. 714 default : ShouldNotReachHere(); 715 } 716 717 // Load LcpoolCache @@@ should be already set! 718 __ get_constant_pool_cache(R27_constPoolCache); 719 720 // Handle a pending exception, fall through if none. 721 __ check_and_forward_exception(R11_scratch1, R12_scratch2); 722 723 // Start executing bytecodes. 724 if (continuation == NULL) { 725 __ dispatch_next(state, step); 726 } else { 727 __ jump_to_entry(continuation, R11_scratch1); 728 } 729 730 return entry; 731 } 732 733 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) { 734 address entry = __ pc(); 735 736 __ push(state); 737 __ call_VM(noreg, runtime_entry); 738 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); 739 740 return entry; 741 } 742 743 // Helpers for commoning out cases in the various type of method entries. 744 745 // Increment invocation count & check for overflow. 746 // 747 // Note: checking for negative value instead of overflow 748 // so we have a 'sticky' overflow test. 749 // 750 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) { 751 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not. 752 Register Rscratch1 = R11_scratch1; 753 Register Rscratch2 = R12_scratch2; 754 Register R3_counters = R3_ARG1; 755 Label done; 756 757 if (TieredCompilation) { 758 const int increment = InvocationCounter::count_increment; 759 Label no_mdo; 760 if (ProfileInterpreter) { 761 const Register Rmdo = R3_counters; 762 // If no method data exists, go to profile_continue. 763 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method); 764 __ cmpdi(CCR0, Rmdo, 0); 765 __ beq(CCR0, no_mdo); 766 767 // Increment invocation counter in the MDO. 768 const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset()); 769 __ lwz(Rscratch2, mdo_ic_offs, Rmdo); 770 __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo); 771 __ addi(Rscratch2, Rscratch2, increment); 772 __ stw(Rscratch2, mdo_ic_offs, Rmdo); 773 __ and_(Rscratch1, Rscratch2, Rscratch1); 774 __ bne(CCR0, done); 775 __ b(*overflow); 776 } 777 778 // Increment counter in MethodCounters*. 779 const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset()); 780 __ bind(no_mdo); 781 __ get_method_counters(R19_method, R3_counters, done); 782 __ lwz(Rscratch2, mo_ic_offs, R3_counters); 783 __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters); 784 __ addi(Rscratch2, Rscratch2, increment); 785 __ stw(Rscratch2, mo_ic_offs, R3_counters); 786 __ and_(Rscratch1, Rscratch2, Rscratch1); 787 __ beq(CCR0, *overflow); 788 789 __ bind(done); 790 791 } else { 792 793 // Update standard invocation counters. 794 Register Rsum_ivc_bec = R4_ARG2; 795 __ get_method_counters(R19_method, R3_counters, done); 796 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2); 797 // Increment interpreter invocation counter. 798 if (ProfileInterpreter) { // %%% Merge this into methodDataOop. 799 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters); 800 __ addi(R12_scratch2, R12_scratch2, 1); 801 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters); 802 } 803 // Check if we must create a method data obj. 804 if (ProfileInterpreter && profile_method != NULL) { 805 const Register profile_limit = Rscratch1; 806 __ lwz(profile_limit, in_bytes(MethodCounters::interpreter_profile_limit_offset()), R3_counters); 807 // Test to see if we should create a method data oop. 808 __ cmpw(CCR0, Rsum_ivc_bec, profile_limit); 809 __ blt(CCR0, *profile_method_continue); 810 // If no method data exists, go to profile_method. 811 __ test_method_data_pointer(*profile_method); 812 } 813 // Finally check for counter overflow. 814 if (overflow) { 815 const Register invocation_limit = Rscratch1; 816 __ lwz(invocation_limit, in_bytes(MethodCounters::interpreter_invocation_limit_offset()), R3_counters); 817 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit); 818 __ bge(CCR0, *overflow); 819 } 820 821 __ bind(done); 822 } 823 } 824 825 // Generate code to initiate compilation on invocation counter overflow. 826 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) { 827 // Generate code to initiate compilation on the counter overflow. 828 829 // InterpreterRuntime::frequency_counter_overflow takes one arguments, 830 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp) 831 // We pass zero in. 832 // The call returns the address of the verified entry point for the method or NULL 833 // if the compilation did not complete (either went background or bailed out). 834 // 835 // Unlike the C++ interpreter above: Check exceptions! 836 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed 837 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur. 838 839 __ li(R4_ARG2, 0); 840 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true); 841 842 // Returns verified_entry_point or NULL. 843 // We ignore it in any case. 844 __ b(continue_entry); 845 } 846 847 // See if we've got enough room on the stack for locals plus overhead below 848 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError 849 // without going through the signal handler, i.e., reserved and yellow zones 850 // will not be made usable. The shadow zone must suffice to handle the 851 // overflow. 852 // 853 // Kills Rmem_frame_size, Rscratch1. 854 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) { 855 Label done; 856 assert_different_registers(Rmem_frame_size, Rscratch1); 857 858 BLOCK_COMMENT("stack_overflow_check_with_compare {"); 859 __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size); 860 __ ld(Rscratch1, thread_(stack_overflow_limit)); 861 __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1); 862 __ bgt(CCR0/*is_stack_overflow*/, done); 863 864 // The stack overflows. Load target address of the runtime stub and call it. 865 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order"); 866 __ load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0); 867 __ mtctr(Rscratch1); 868 // Restore caller_sp. 869 #ifdef ASSERT 870 __ ld(Rscratch1, 0, R1_SP); 871 __ ld(R0, 0, R21_sender_SP); 872 __ cmpd(CCR0, R0, Rscratch1); 873 __ asm_assert_eq("backlink", 0x547); 874 #endif // ASSERT 875 __ mr(R1_SP, R21_sender_SP); 876 __ bctr(); 877 878 __ align(32, 12); 879 __ bind(done); 880 BLOCK_COMMENT("} stack_overflow_check_with_compare"); 881 } 882 883 // Lock the current method, interpreter register window must be set up! 884 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) { 885 const Register Robj_to_lock = Rscratch2; 886 887 { 888 if (!flags_preloaded) { 889 __ lwz(Rflags, method_(access_flags)); 890 } 891 892 #ifdef ASSERT 893 // Check if methods needs synchronization. 894 { 895 Label Lok; 896 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT); 897 __ btrue(CCR0,Lok); 898 __ stop("method doesn't need synchronization"); 899 __ bind(Lok); 900 } 901 #endif // ASSERT 902 } 903 904 // Get synchronization object to Rscratch2. 905 { 906 Label Lstatic; 907 Label Ldone; 908 909 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT); 910 __ btrue(CCR0, Lstatic); 911 912 // Non-static case: load receiver obj from stack and we're done. 913 __ ld(Robj_to_lock, R18_locals); 914 __ b(Ldone); 915 916 __ bind(Lstatic); // Static case: Lock the java mirror 917 // Load mirror from interpreter frame. 918 __ ld(Robj_to_lock, _abi(callers_sp), R1_SP); 919 __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock); 920 921 __ bind(Ldone); 922 __ verify_oop(Robj_to_lock); 923 } 924 925 // Got the oop to lock => execute! 926 __ add_monitor_to_stack(true, Rscratch1, R0); 927 928 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor); 929 __ lock_object(R26_monitor, Robj_to_lock); 930 } 931 932 // Generate a fixed interpreter frame for pure interpreter 933 // and I2N native transition frames. 934 // 935 // Before (stack grows downwards): 936 // 937 // | ... | 938 // |------------- | 939 // | java arg0 | 940 // | ... | 941 // | java argn | 942 // | | <- R15_esp 943 // | | 944 // |--------------| 945 // | abi_112 | 946 // | | <- R1_SP 947 // |==============| 948 // 949 // 950 // After: 951 // 952 // | ... | 953 // | java arg0 |<- R18_locals 954 // | ... | 955 // | java argn | 956 // |--------------| 957 // | | 958 // | java locals | 959 // | | 960 // |--------------| 961 // | abi_48 | 962 // |==============| 963 // | | 964 // | istate | 965 // | | 966 // |--------------| 967 // | monitor |<- R26_monitor 968 // |--------------| 969 // | |<- R15_esp 970 // | expression | 971 // | stack | 972 // | | 973 // |--------------| 974 // | | 975 // | abi_112 |<- R1_SP 976 // |==============| 977 // 978 // The top most frame needs an abi space of 112 bytes. This space is needed, 979 // since we call to c. The c function may spill their arguments to the caller 980 // frame. When we call to java, we don't need these spill slots. In order to save 981 // space on the stack, we resize the caller. However, java locals reside in 982 // the caller frame and the frame has to be increased. The frame_size for the 983 // current frame was calculated based on max_stack as size for the expression 984 // stack. At the call, just a part of the expression stack might be used. 985 // We don't want to waste this space and cut the frame back accordingly. 986 // The resulting amount for resizing is calculated as follows: 987 // resize = (number_of_locals - number_of_arguments) * slot_size 988 // + (R1_SP - R15_esp) + 48 989 // 990 // The size for the callee frame is calculated: 991 // framesize = 112 + max_stack + monitor + state_size 992 // 993 // maxstack: Max number of slots on the expression stack, loaded from the method. 994 // monitor: We statically reserve room for one monitor object. 995 // state_size: We save the current state of the interpreter to this area. 996 // 997 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) { 998 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes. 999 top_frame_size = R7_ARG5, 1000 Rconst_method = R8_ARG6; 1001 1002 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size); 1003 1004 __ ld(Rconst_method, method_(const)); 1005 __ lhz(Rsize_of_parameters /* number of params */, 1006 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method); 1007 if (native_call) { 1008 // If we're calling a native method, we reserve space for the worst-case signature 1009 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2). 1010 // We add two slots to the parameter_count, one for the jni 1011 // environment and one for a possible native mirror. 1012 Label skip_native_calculate_max_stack; 1013 __ addi(top_frame_size, Rsize_of_parameters, 2); 1014 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters); 1015 __ bge(CCR0, skip_native_calculate_max_stack); 1016 __ li(top_frame_size, Argument::n_register_parameters); 1017 __ bind(skip_native_calculate_max_stack); 1018 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize); 1019 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize); 1020 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize! 1021 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters. 1022 } else { 1023 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method); 1024 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize); 1025 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize); 1026 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method); 1027 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0 1028 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize! 1029 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize); 1030 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1); 1031 } 1032 1033 // Compute top frame size. 1034 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size); 1035 1036 // Cut back area between esp and max_stack. 1037 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize); 1038 1039 __ round_to(top_frame_size, frame::alignment_in_bytes); 1040 __ round_to(parent_frame_resize, frame::alignment_in_bytes); 1041 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size. 1042 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48. 1043 1044 if (!native_call) { 1045 // Stack overflow check. 1046 // Native calls don't need the stack size check since they have no 1047 // expression stack and the arguments are already on the stack and 1048 // we only add a handful of words to the stack. 1049 __ add(R11_scratch1, parent_frame_resize, top_frame_size); 1050 generate_stack_overflow_check(R11_scratch1, R12_scratch2); 1051 } 1052 1053 // Set up interpreter state registers. 1054 1055 __ add(R18_locals, R15_esp, Rsize_of_parameters); 1056 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method); 1057 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache); 1058 1059 // Set method data pointer. 1060 if (ProfileInterpreter) { 1061 Label zero_continue; 1062 __ ld(R28_mdx, method_(method_data)); 1063 __ cmpdi(CCR0, R28_mdx, 0); 1064 __ beq(CCR0, zero_continue); 1065 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1066 __ bind(zero_continue); 1067 } 1068 1069 if (native_call) { 1070 __ li(R14_bcp, 0); // Must initialize. 1071 } else { 1072 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method); 1073 } 1074 1075 // Resize parent frame. 1076 __ mflr(R12_scratch2); 1077 __ neg(parent_frame_resize, parent_frame_resize); 1078 __ resize_frame(parent_frame_resize, R11_scratch1); 1079 __ std(R12_scratch2, _abi(lr), R1_SP); 1080 1081 // Get mirror and store it in the frame as GC root for this Method*. 1082 __ load_mirror_from_const_method(R12_scratch2, Rconst_method); 1083 1084 __ addi(R26_monitor, R1_SP, - frame::ijava_state_size); 1085 __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); 1086 1087 // Store values. 1088 // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls 1089 // in InterpreterMacroAssembler::call_from_interpreter. 1090 __ std(R19_method, _ijava_state_neg(method), R1_SP); 1091 __ std(R12_scratch2, _ijava_state_neg(mirror), R1_SP); 1092 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP); 1093 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP); 1094 __ std(R18_locals, _ijava_state_neg(locals), R1_SP); 1095 1096 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only 1097 // be found in the frame after save_interpreter_state is done. This is always true 1098 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong, 1099 // because e.g. frame::interpreter_frame_bcp() will not access the correct value 1100 // (Enhanced Stack Trace). 1101 // The signal handler does not save the interpreter state into the frame. 1102 __ li(R0, 0); 1103 #ifdef ASSERT 1104 // Fill remaining slots with constants. 1105 __ load_const_optimized(R11_scratch1, 0x5afe); 1106 __ load_const_optimized(R12_scratch2, 0xdead); 1107 #endif 1108 // We have to initialize some frame slots for native calls (accessed by GC). 1109 if (native_call) { 1110 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP); 1111 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP); 1112 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); } 1113 } 1114 #ifdef ASSERT 1115 else { 1116 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP); 1117 __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP); 1118 __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP); 1119 } 1120 __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP); 1121 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP); 1122 __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP); 1123 __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP); 1124 #endif 1125 __ subf(R12_scratch2, top_frame_size, R1_SP); 1126 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); 1127 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP); 1128 1129 // Push top frame. 1130 __ push_frame(top_frame_size, R11_scratch1); 1131 } 1132 1133 // End of helpers 1134 1135 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) { 1136 1137 // Decide what to do: Use same platform specific instructions and runtime calls as compilers. 1138 bool use_instruction = false; 1139 address runtime_entry = NULL; 1140 int num_args = 1; 1141 bool double_precision = true; 1142 1143 // PPC64 specific: 1144 switch (kind) { 1145 case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break; 1146 case Interpreter::java_lang_math_abs: use_instruction = true; break; 1147 case Interpreter::java_lang_math_fmaF: 1148 case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break; 1149 default: break; // Fall back to runtime call. 1150 } 1151 1152 switch (kind) { 1153 case Interpreter::java_lang_math_sin : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); break; 1154 case Interpreter::java_lang_math_cos : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); break; 1155 case Interpreter::java_lang_math_tan : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); break; 1156 case Interpreter::java_lang_math_abs : /* run interpreted */ break; 1157 case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); break; 1158 case Interpreter::java_lang_math_log : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); break; 1159 case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break; 1160 case Interpreter::java_lang_math_pow : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break; 1161 case Interpreter::java_lang_math_exp : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); break; 1162 case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break; 1163 case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break; 1164 default: ShouldNotReachHere(); 1165 } 1166 1167 // Use normal entry if neither instruction nor runtime call is used. 1168 if (!use_instruction && runtime_entry == NULL) return NULL; 1169 1170 address entry = __ pc(); 1171 1172 // Load arguments 1173 assert(num_args <= 13, "passed in registers"); 1174 if (double_precision) { 1175 int offset = (2 * num_args - 1) * Interpreter::stackElementSize; 1176 for (int i = 0; i < num_args; ++i) { 1177 __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp); 1178 offset -= 2 * Interpreter::stackElementSize; 1179 } 1180 } else { 1181 int offset = num_args * Interpreter::stackElementSize; 1182 for (int i = 0; i < num_args; ++i) { 1183 __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp); 1184 offset -= Interpreter::stackElementSize; 1185 } 1186 } 1187 1188 // Pop c2i arguments (if any) off when we return. 1189 #ifdef ASSERT 1190 __ ld(R9_ARG7, 0, R1_SP); 1191 __ ld(R10_ARG8, 0, R21_sender_SP); 1192 __ cmpd(CCR0, R9_ARG7, R10_ARG8); 1193 __ asm_assert_eq("backlink", 0x545); 1194 #endif // ASSERT 1195 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 1196 1197 if (use_instruction) { 1198 switch (kind) { 1199 case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1); break; 1200 case Interpreter::java_lang_math_abs: __ fabs(F1_RET, F1); break; 1201 case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break; 1202 case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3); break; 1203 default: ShouldNotReachHere(); 1204 } 1205 } else { 1206 // Comment: Can use tail call if the unextended frame is always C ABI compliant: 1207 //__ load_const_optimized(R12_scratch2, runtime_entry, R0); 1208 //__ call_c_and_return_to_caller(R12_scratch2); 1209 1210 // Push a new C frame and save LR. 1211 __ save_LR_CR(R0); 1212 __ push_frame_reg_args(0, R11_scratch1); 1213 1214 __ call_VM_leaf(runtime_entry); 1215 1216 // Pop the C frame and restore LR. 1217 __ pop_frame(); 1218 __ restore_LR_CR(R0); 1219 } 1220 1221 __ blr(); 1222 1223 __ flush(); 1224 1225 return entry; 1226 } 1227 1228 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) { 1229 // Quick & dirty stack overflow checking: bang the stack & handle trap. 1230 // Note that we do the banging after the frame is setup, since the exception 1231 // handling code expects to find a valid interpreter frame on the stack. 1232 // Doing the banging earlier fails if the caller frame is not an interpreter 1233 // frame. 1234 // (Also, the exception throwing code expects to unlock any synchronized 1235 // method receiever, so do the banging after locking the receiver.) 1236 1237 // Bang each page in the shadow zone. We can't assume it's been done for 1238 // an interpreter frame with greater than a page of locals, so each page 1239 // needs to be checked. Only true for non-native. 1240 if (UseStackBanging) { 1241 const int page_size = os::vm_page_size(); 1242 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size; 1243 const int start_page = native_call ? n_shadow_pages : 1; 1244 BLOCK_COMMENT("bang_stack_shadow_pages:"); 1245 for (int pages = start_page; pages <= n_shadow_pages; pages++) { 1246 __ bang_stack_with_offset(pages*page_size); 1247 } 1248 } 1249 } 1250 1251 // Interpreter stub for calling a native method. (asm interpreter) 1252 // This sets up a somewhat different looking stack for calling the 1253 // native method than the typical interpreter frame setup. 1254 // 1255 // On entry: 1256 // R19_method - method 1257 // R16_thread - JavaThread* 1258 // R15_esp - intptr_t* sender tos 1259 // 1260 // abstract stack (grows up) 1261 // [ IJava (caller of JNI callee) ] <-- ASP 1262 // ... 1263 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) { 1264 1265 address entry = __ pc(); 1266 1267 const bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; 1268 1269 // ----------------------------------------------------------------------------- 1270 // Allocate a new frame that represents the native callee (i2n frame). 1271 // This is not a full-blown interpreter frame, but in particular, the 1272 // following registers are valid after this: 1273 // - R19_method 1274 // - R18_local (points to start of arguments to native function) 1275 // 1276 // abstract stack (grows up) 1277 // [ IJava (caller of JNI callee) ] <-- ASP 1278 // ... 1279 1280 const Register signature_handler_fd = R11_scratch1; 1281 const Register pending_exception = R0; 1282 const Register result_handler_addr = R31; 1283 const Register native_method_fd = R11_scratch1; 1284 const Register access_flags = R22_tmp2; 1285 const Register active_handles = R11_scratch1; // R26_monitor saved to state. 1286 const Register sync_state = R12_scratch2; 1287 const Register sync_state_addr = sync_state; // Address is dead after use. 1288 const Register suspend_flags = R11_scratch1; 1289 1290 //============================================================================= 1291 // Allocate new frame and initialize interpreter state. 1292 1293 Label exception_return; 1294 Label exception_return_sync_check; 1295 Label stack_overflow_return; 1296 1297 // Generate new interpreter state and jump to stack_overflow_return in case of 1298 // a stack overflow. 1299 //generate_compute_interpreter_state(stack_overflow_return); 1300 1301 Register size_of_parameters = R22_tmp2; 1302 1303 generate_fixed_frame(true, size_of_parameters, noreg /* unused */); 1304 1305 //============================================================================= 1306 // Increment invocation counter. On overflow, entry to JNI method 1307 // will be compiled. 1308 Label invocation_counter_overflow, continue_after_compile; 1309 if (inc_counter) { 1310 if (synchronized) { 1311 // Since at this point in the method invocation the exception handler 1312 // would try to exit the monitor of synchronized methods which hasn't 1313 // been entered yet, we set the thread local variable 1314 // _do_not_unlock_if_synchronized to true. If any exception was thrown by 1315 // runtime, exception handling i.e. unlock_if_synchronized_method will 1316 // check this thread local flag. 1317 // This flag has two effects, one is to force an unwind in the topmost 1318 // interpreter frame and not perform an unlock while doing so. 1319 __ li(R0, 1); 1320 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1321 } 1322 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); 1323 1324 BIND(continue_after_compile); 1325 } 1326 1327 bang_stack_shadow_pages(true); 1328 1329 if (inc_counter) { 1330 // Reset the _do_not_unlock_if_synchronized flag. 1331 if (synchronized) { 1332 __ li(R0, 0); 1333 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1334 } 1335 } 1336 1337 // access_flags = method->access_flags(); 1338 // Load access flags. 1339 assert(access_flags->is_nonvolatile(), 1340 "access_flags must be in a non-volatile register"); 1341 // Type check. 1342 assert(4 == sizeof(AccessFlags), "unexpected field size"); 1343 __ lwz(access_flags, method_(access_flags)); 1344 1345 // We don't want to reload R19_method and access_flags after calls 1346 // to some helper functions. 1347 assert(R19_method->is_nonvolatile(), 1348 "R19_method must be a non-volatile register"); 1349 1350 // Check for synchronized methods. Must happen AFTER invocation counter 1351 // check, so method is not locked if counter overflows. 1352 1353 if (synchronized) { 1354 lock_method(access_flags, R11_scratch1, R12_scratch2, true); 1355 1356 // Update monitor in state. 1357 __ ld(R11_scratch1, 0, R1_SP); 1358 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1); 1359 } 1360 1361 // jvmti/jvmpi support 1362 __ notify_method_entry(); 1363 1364 //============================================================================= 1365 // Get and call the signature handler. 1366 1367 __ ld(signature_handler_fd, method_(signature_handler)); 1368 Label call_signature_handler; 1369 1370 __ cmpdi(CCR0, signature_handler_fd, 0); 1371 __ bne(CCR0, call_signature_handler); 1372 1373 // Method has never been called. Either generate a specialized 1374 // handler or point to the slow one. 1375 // 1376 // Pass parameter 'false' to avoid exception check in call_VM. 1377 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false); 1378 1379 // Check for an exception while looking up the target method. If we 1380 // incurred one, bail. 1381 __ ld(pending_exception, thread_(pending_exception)); 1382 __ cmpdi(CCR0, pending_exception, 0); 1383 __ bne(CCR0, exception_return_sync_check); // Has pending exception. 1384 1385 // Reload signature handler, it may have been created/assigned in the meanwhile. 1386 __ ld(signature_handler_fd, method_(signature_handler)); 1387 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below). 1388 1389 BIND(call_signature_handler); 1390 1391 // Before we call the signature handler we push a new frame to 1392 // protect the interpreter frame volatile registers when we return 1393 // from jni but before we can get back to Java. 1394 1395 // First set the frame anchor while the SP/FP registers are 1396 // convenient and the slow signature handler can use this same frame 1397 // anchor. 1398 1399 // We have a TOP_IJAVA_FRAME here, which belongs to us. 1400 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/); 1401 1402 // Now the interpreter frame (and its call chain) have been 1403 // invalidated and flushed. We are now protected against eager 1404 // being enabled in native code. Even if it goes eager the 1405 // registers will be reloaded as clean and we will invalidate after 1406 // the call so no spurious flush should be possible. 1407 1408 // Call signature handler and pass locals address. 1409 // 1410 // Our signature handlers copy required arguments to the C stack 1411 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13. 1412 __ mr(R3_ARG1, R18_locals); 1413 #if !defined(ABI_ELFv2) 1414 __ ld(signature_handler_fd, 0, signature_handler_fd); 1415 #endif 1416 1417 __ call_stub(signature_handler_fd); 1418 1419 // Remove the register parameter varargs slots we allocated in 1420 // compute_interpreter_state. SP+16 ends up pointing to the ABI 1421 // outgoing argument area. 1422 // 1423 // Not needed on PPC64. 1424 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord); 1425 1426 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register"); 1427 // Save across call to native method. 1428 __ mr(result_handler_addr, R3_RET); 1429 1430 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror. 1431 1432 // Set up fixed parameters and call the native method. 1433 // If the method is static, get mirror into R4_ARG2. 1434 { 1435 Label method_is_not_static; 1436 // Access_flags is non-volatile and still, no need to restore it. 1437 1438 // Restore access flags. 1439 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT); 1440 __ bfalse(CCR0, method_is_not_static); 1441 1442 __ ld(R11_scratch1, _abi(callers_sp), R1_SP); 1443 // Load mirror from interpreter frame. 1444 __ ld(R12_scratch2, _ijava_state_neg(mirror), R11_scratch1); 1445 // R4_ARG2 = &state->_oop_temp; 1446 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp)); 1447 __ std(R12_scratch2/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); 1448 BIND(method_is_not_static); 1449 } 1450 1451 // At this point, arguments have been copied off the stack into 1452 // their JNI positions. Oops are boxed in-place on the stack, with 1453 // handles copied to arguments. The result handler address is in a 1454 // register. 1455 1456 // Pass JNIEnv address as first parameter. 1457 __ addir(R3_ARG1, thread_(jni_environment)); 1458 1459 // Load the native_method entry before we change the thread state. 1460 __ ld(native_method_fd, method_(native_function)); 1461 1462 //============================================================================= 1463 // Transition from _thread_in_Java to _thread_in_native. As soon as 1464 // we make this change the safepoint code needs to be certain that 1465 // the last Java frame we established is good. The pc in that frame 1466 // just needs to be near here not an actual return address. 1467 1468 // We use release_store_fence to update values like the thread state, where 1469 // we don't want the current thread to continue until all our prior memory 1470 // accesses (including the new thread state) are visible to other threads. 1471 __ li(R0, _thread_in_native); 1472 __ release(); 1473 1474 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size"); 1475 __ stw(R0, thread_(thread_state)); 1476 1477 //============================================================================= 1478 // Call the native method. Argument registers must not have been 1479 // overwritten since "__ call_stub(signature_handler);" (except for 1480 // ARG1 and ARG2 for static methods). 1481 __ call_c(native_method_fd); 1482 1483 __ li(R0, 0); 1484 __ ld(R11_scratch1, 0, R1_SP); 1485 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1); 1486 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1); 1487 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset 1488 1489 // Note: C++ interpreter needs the following here: 1490 // The frame_manager_lr field, which we use for setting the last 1491 // java frame, gets overwritten by the signature handler. Restore 1492 // it now. 1493 //__ get_PC_trash_LR(R11_scratch1); 1494 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 1495 1496 // Because of GC R19_method may no longer be valid. 1497 1498 // Block, if necessary, before resuming in _thread_in_Java state. 1499 // In order for GC to work, don't clear the last_Java_sp until after 1500 // blocking. 1501 1502 //============================================================================= 1503 // Switch thread to "native transition" state before reading the 1504 // synchronization state. This additional state is necessary 1505 // because reading and testing the synchronization state is not 1506 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A, 1507 // in _thread_in_native state, loads _not_synchronized and is 1508 // preempted. VM thread changes sync state to synchronizing and 1509 // suspends threads for GC. Thread A is resumed to finish this 1510 // native method, but doesn't block here since it didn't see any 1511 // synchronization in progress, and escapes. 1512 1513 // We use release_store_fence to update values like the thread state, where 1514 // we don't want the current thread to continue until all our prior memory 1515 // accesses (including the new thread state) are visible to other threads. 1516 __ li(R0/*thread_state*/, _thread_in_native_trans); 1517 __ release(); 1518 __ stw(R0/*thread_state*/, thread_(thread_state)); 1519 if (UseMembar) { 1520 __ fence(); 1521 } 1522 // Write serialization page so that the VM thread can do a pseudo remote 1523 // membar. We use the current thread pointer to calculate a thread 1524 // specific offset to write to within the page. This minimizes bus 1525 // traffic due to cache line collision. 1526 else { 1527 __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2); 1528 } 1529 1530 // Now before we return to java we must look for a current safepoint 1531 // (a new safepoint can not start since we entered native_trans). 1532 // We must check here because a current safepoint could be modifying 1533 // the callers registers right this moment. 1534 1535 // Acquire isn't strictly necessary here because of the fence, but 1536 // sync_state is declared to be volatile, so we do it anyway 1537 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path). 1538 1539 Label do_safepoint, sync_check_done; 1540 // No synchronization in progress nor yet synchronized. 1541 __ safepoint_poll(do_safepoint, sync_state); 1542 1543 // Not suspended. 1544 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size"); 1545 __ lwz(suspend_flags, thread_(suspend_flags)); 1546 __ cmpwi(CCR1, suspend_flags, 0); 1547 __ beq(CCR1, sync_check_done); 1548 1549 __ bind(do_safepoint); 1550 __ isync(); 1551 // Block. We do the call directly and leave the current 1552 // last_Java_frame setup undisturbed. We must save any possible 1553 // native result across the call. No oop is present. 1554 1555 __ mr(R3_ARG1, R16_thread); 1556 #if defined(ABI_ELFv2) 1557 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), 1558 relocInfo::none); 1559 #else 1560 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans), 1561 relocInfo::none); 1562 #endif 1563 1564 __ bind(sync_check_done); 1565 1566 //============================================================================= 1567 // <<<<<< Back in Interpreter Frame >>>>> 1568 1569 // We are in thread_in_native_trans here and back in the normal 1570 // interpreter frame. We don't have to do anything special about 1571 // safepoints and we can switch to Java mode anytime we are ready. 1572 1573 // Note: frame::interpreter_frame_result has a dependency on how the 1574 // method result is saved across the call to post_method_exit. For 1575 // native methods it assumes that the non-FPU/non-void result is 1576 // saved in _native_lresult and a FPU result in _native_fresult. If 1577 // this changes then the interpreter_frame_result implementation 1578 // will need to be updated too. 1579 1580 // On PPC64, we have stored the result directly after the native call. 1581 1582 //============================================================================= 1583 // Back in Java 1584 1585 // We use release_store_fence to update values like the thread state, where 1586 // we don't want the current thread to continue until all our prior memory 1587 // accesses (including the new thread state) are visible to other threads. 1588 __ li(R0/*thread_state*/, _thread_in_Java); 1589 __ lwsync(); // Acquire safepoint and suspend state, release thread state. 1590 __ stw(R0/*thread_state*/, thread_(thread_state)); 1591 1592 if (CheckJNICalls) { 1593 // clear_pending_jni_exception_check 1594 __ load_const_optimized(R0, 0L); 1595 __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread); 1596 } 1597 1598 __ reset_last_Java_frame(); 1599 1600 // Jvmdi/jvmpi support. Whether we've got an exception pending or 1601 // not, and whether unlocking throws an exception or not, we notify 1602 // on native method exit. If we do have an exception, we'll end up 1603 // in the caller's context to handle it, so if we don't do the 1604 // notify here, we'll drop it on the floor. 1605 __ notify_method_exit(true/*native method*/, 1606 ilgl /*illegal state (not used for native methods)*/, 1607 InterpreterMacroAssembler::NotifyJVMTI, 1608 false /*check_exceptions*/); 1609 1610 //============================================================================= 1611 // Handle exceptions 1612 1613 if (synchronized) { 1614 // Don't check for exceptions since we're still in the i2n frame. Do that 1615 // manually afterwards. 1616 __ unlock_object(R26_monitor, false); // Can also unlock methods. 1617 } 1618 1619 // Reset active handles after returning from native. 1620 // thread->active_handles()->clear(); 1621 __ ld(active_handles, thread_(active_handles)); 1622 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size"); 1623 __ li(R0, 0); 1624 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles); 1625 1626 Label exception_return_sync_check_already_unlocked; 1627 __ ld(R0/*pending_exception*/, thread_(pending_exception)); 1628 __ cmpdi(CCR0, R0/*pending_exception*/, 0); 1629 __ bne(CCR0, exception_return_sync_check_already_unlocked); 1630 1631 //----------------------------------------------------------------------------- 1632 // No exception pending. 1633 1634 // Move native method result back into proper registers and return. 1635 // Invoke result handler (may unbox/promote). 1636 __ ld(R11_scratch1, 0, R1_SP); 1637 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1); 1638 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1); 1639 __ call_stub(result_handler_addr); 1640 1641 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1642 1643 // Must use the return pc which was loaded from the caller's frame 1644 // as the VM uses return-pc-patching for deoptimization. 1645 __ mtlr(R0); 1646 __ blr(); 1647 1648 //----------------------------------------------------------------------------- 1649 // An exception is pending. We call into the runtime only if the 1650 // caller was not interpreted. If it was interpreted the 1651 // interpreter will do the correct thing. If it isn't interpreted 1652 // (call stub/compiled code) we will change our return and continue. 1653 1654 BIND(exception_return_sync_check); 1655 1656 if (synchronized) { 1657 // Don't check for exceptions since we're still in the i2n frame. Do that 1658 // manually afterwards. 1659 __ unlock_object(R26_monitor, false); // Can also unlock methods. 1660 } 1661 BIND(exception_return_sync_check_already_unlocked); 1662 1663 const Register return_pc = R31; 1664 1665 __ ld(return_pc, 0, R1_SP); 1666 __ ld(return_pc, _abi(lr), return_pc); 1667 1668 // Get the address of the exception handler. 1669 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), 1670 R16_thread, 1671 return_pc /* return pc */); 1672 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2); 1673 1674 // Load the PC of the the exception handler into LR. 1675 __ mtlr(R3_RET); 1676 1677 // Load exception into R3_ARG1 and clear pending exception in thread. 1678 __ ld(R3_ARG1/*exception*/, thread_(pending_exception)); 1679 __ li(R4_ARG2, 0); 1680 __ std(R4_ARG2, thread_(pending_exception)); 1681 1682 // Load the original return pc into R4_ARG2. 1683 __ mr(R4_ARG2/*issuing_pc*/, return_pc); 1684 1685 // Return to exception handler. 1686 __ blr(); 1687 1688 //============================================================================= 1689 // Counter overflow. 1690 1691 if (inc_counter) { 1692 // Handle invocation counter overflow. 1693 __ bind(invocation_counter_overflow); 1694 1695 generate_counter_overflow(continue_after_compile); 1696 } 1697 1698 return entry; 1699 } 1700 1701 // Generic interpreted method entry to (asm) interpreter. 1702 // 1703 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) { 1704 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; 1705 address entry = __ pc(); 1706 // Generate the code to allocate the interpreter stack frame. 1707 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame. 1708 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame. 1709 1710 // Does also a stack check to assure this frame fits on the stack. 1711 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals); 1712 1713 // -------------------------------------------------------------------------- 1714 // Zero out non-parameter locals. 1715 // Note: *Always* zero out non-parameter locals as Sparc does. It's not 1716 // worth to ask the flag, just do it. 1717 Register Rslot_addr = R6_ARG4, 1718 Rnum = R7_ARG5; 1719 Label Lno_locals, Lzero_loop; 1720 1721 // Set up the zeroing loop. 1722 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals); 1723 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals); 1724 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize); 1725 __ beq(CCR0, Lno_locals); 1726 __ li(R0, 0); 1727 __ mtctr(Rnum); 1728 1729 // The zero locals loop. 1730 __ bind(Lzero_loop); 1731 __ std(R0, 0, Rslot_addr); 1732 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize); 1733 __ bdnz(Lzero_loop); 1734 1735 __ bind(Lno_locals); 1736 1737 // -------------------------------------------------------------------------- 1738 // Counter increment and overflow check. 1739 Label invocation_counter_overflow, 1740 profile_method, 1741 profile_method_continue; 1742 if (inc_counter || ProfileInterpreter) { 1743 1744 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1; 1745 if (synchronized) { 1746 // Since at this point in the method invocation the exception handler 1747 // would try to exit the monitor of synchronized methods which hasn't 1748 // been entered yet, we set the thread local variable 1749 // _do_not_unlock_if_synchronized to true. If any exception was thrown by 1750 // runtime, exception handling i.e. unlock_if_synchronized_method will 1751 // check this thread local flag. 1752 // This flag has two effects, one is to force an unwind in the topmost 1753 // interpreter frame and not perform an unlock while doing so. 1754 __ li(R0, 1); 1755 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1756 } 1757 1758 // Argument and return type profiling. 1759 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4); 1760 1761 // Increment invocation counter and check for overflow. 1762 if (inc_counter) { 1763 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue); 1764 } 1765 1766 __ bind(profile_method_continue); 1767 } 1768 1769 bang_stack_shadow_pages(false); 1770 1771 if (inc_counter || ProfileInterpreter) { 1772 // Reset the _do_not_unlock_if_synchronized flag. 1773 if (synchronized) { 1774 __ li(R0, 0); 1775 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1776 } 1777 } 1778 1779 // -------------------------------------------------------------------------- 1780 // Locking of synchronized methods. Must happen AFTER invocation_counter 1781 // check and stack overflow check, so method is not locked if overflows. 1782 if (synchronized) { 1783 lock_method(R3_ARG1, R4_ARG2, R5_ARG3); 1784 } 1785 #ifdef ASSERT 1786 else { 1787 Label Lok; 1788 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method); 1789 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED); 1790 __ asm_assert_eq("method needs synchronization", 0x8521); 1791 __ bind(Lok); 1792 } 1793 #endif // ASSERT 1794 1795 __ verify_thread(); 1796 1797 // -------------------------------------------------------------------------- 1798 // JVMTI support 1799 __ notify_method_entry(); 1800 1801 // -------------------------------------------------------------------------- 1802 // Start executing instructions. 1803 __ dispatch_next(vtos); 1804 1805 // -------------------------------------------------------------------------- 1806 // Out of line counter overflow and MDO creation code. 1807 if (ProfileInterpreter) { 1808 // We have decided to profile this method in the interpreter. 1809 __ bind(profile_method); 1810 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1811 __ set_method_data_pointer_for_bcp(); 1812 __ b(profile_method_continue); 1813 } 1814 1815 if (inc_counter) { 1816 // Handle invocation counter overflow. 1817 __ bind(invocation_counter_overflow); 1818 generate_counter_overflow(profile_method_continue); 1819 } 1820 return entry; 1821 } 1822 1823 // CRC32 Intrinsics. 1824 // 1825 // Contract on scratch and work registers. 1826 // ======================================= 1827 // 1828 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers. 1829 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set. 1830 // You can't rely on these registers across calls. 1831 // 1832 // The generators for CRC32_update and for CRC32_updateBytes use the 1833 // scratch/work register set internally, passing the work registers 1834 // as arguments to the MacroAssembler emitters as required. 1835 // 1836 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments. 1837 // Their contents is not constant but may change according to the requirements 1838 // of the emitted code. 1839 // 1840 // All other registers from the scratch/work register set are used "internally" 1841 // and contain garbage (i.e. unpredictable values) once blr() is reached. 1842 // Basically, only R3_RET contains a defined value which is the function result. 1843 // 1844 /** 1845 * Method entry for static native methods: 1846 * int java.util.zip.CRC32.update(int crc, int b) 1847 */ 1848 address TemplateInterpreterGenerator::generate_CRC32_update_entry() { 1849 if (UseCRC32Intrinsics) { 1850 address start = __ pc(); // Remember stub start address (is rtn value). 1851 Label slow_path; 1852 1853 // Safepoint check 1854 const Register sync_state = R11_scratch1; 1855 __ safepoint_poll(slow_path, sync_state); 1856 1857 // We don't generate local frame and don't align stack because 1858 // we not even call stub code (we generate the code inline) 1859 // and there is no safepoint on this path. 1860 1861 // Load java parameters. 1862 // R15_esp is callers operand stack pointer, i.e. it points to the parameters. 1863 const Register argP = R15_esp; 1864 const Register crc = R3_ARG1; // crc value 1865 const Register data = R4_ARG2; // address of java byte value (kernel_crc32 needs address) 1866 const Register dataLen = R5_ARG3; // source data len (1 byte). Not used because calling the single-byte emitter. 1867 const Register table = R6_ARG4; // address of crc32 table 1868 const Register tmp = dataLen; // Reuse unused len register to show we don't actually need a separate tmp here. 1869 1870 BLOCK_COMMENT("CRC32_update {"); 1871 1872 // Arguments are reversed on java expression stack 1873 #ifdef VM_LITTLE_ENDIAN 1874 __ addi(data, argP, 0+1*wordSize); // (stack) address of byte value. Emitter expects address, not value. 1875 // Being passed as an int, the single byte is at offset +0. 1876 #else 1877 __ addi(data, argP, 3+1*wordSize); // (stack) address of byte value. Emitter expects address, not value. 1878 // Being passed from java as an int, the single byte is at offset +3. 1879 #endif 1880 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register. 1881 1882 StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table); 1883 __ kernel_crc32_singleByte(crc, data, dataLen, table, tmp, true); 1884 1885 // Restore caller sp for c2i case and return. 1886 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 1887 __ blr(); 1888 1889 // Generate a vanilla native entry as the slow path. 1890 BLOCK_COMMENT("} CRC32_update"); 1891 BIND(slow_path); 1892 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1); 1893 return start; 1894 } 1895 1896 return NULL; 1897 } 1898 1899 1900 /** 1901 * Method entry for static native methods: 1902 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len) 1903 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len) 1904 */ 1905 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 1906 if (UseCRC32Intrinsics) { 1907 address start = __ pc(); // Remember stub start address (is rtn value). 1908 Label slow_path; 1909 1910 // Safepoint check 1911 const Register sync_state = R11_scratch1; 1912 __ safepoint_poll(slow_path, sync_state); 1913 1914 // We don't generate local frame and don't align stack because 1915 // we not even call stub code (we generate the code inline) 1916 // and there is no safepoint on this path. 1917 1918 // Load parameters. 1919 // Z_esp is callers operand stack pointer, i.e. it points to the parameters. 1920 const Register argP = R15_esp; 1921 const Register crc = R3_ARG1; // crc value 1922 const Register data = R4_ARG2; // address of java byte array 1923 const Register dataLen = R5_ARG3; // source data len 1924 const Register table = R6_ARG4; // address of crc32 table 1925 1926 const Register t0 = R9; // scratch registers for crc calculation 1927 const Register t1 = R10; 1928 const Register t2 = R11; 1929 const Register t3 = R12; 1930 1931 const Register tc0 = R2; // registers to hold pre-calculated column addresses 1932 const Register tc1 = R7; 1933 const Register tc2 = R8; 1934 const Register tc3 = table; // table address is reconstructed at the end of kernel_crc32_* emitters 1935 1936 const Register tmp = t0; // Only used very locally to calculate byte buffer address. 1937 1938 // Arguments are reversed on java expression stack. 1939 // Calculate address of start element. 1940 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct". 1941 BLOCK_COMMENT("CRC32_updateByteBuffer {"); 1942 // crc @ (SP + 5W) (32bit) 1943 // buf @ (SP + 3W) (64bit ptr to long array) 1944 // off @ (SP + 2W) (32bit) 1945 // dataLen @ (SP + 1W) (32bit) 1946 // data = buf + off 1947 __ ld( data, 3*wordSize, argP); // start of byte buffer 1948 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 1949 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 1950 __ lwz( crc, 5*wordSize, argP); // current crc state 1951 __ add( data, data, tmp); // Add byte buffer offset. 1952 } else { // Used for "updateBytes update". 1953 BLOCK_COMMENT("CRC32_updateBytes {"); 1954 // crc @ (SP + 4W) (32bit) 1955 // buf @ (SP + 3W) (64bit ptr to byte array) 1956 // off @ (SP + 2W) (32bit) 1957 // dataLen @ (SP + 1W) (32bit) 1958 // data = buf + off + base_offset 1959 __ ld( data, 3*wordSize, argP); // start of byte buffer 1960 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 1961 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 1962 __ add( data, data, tmp); // add byte buffer offset 1963 __ lwz( crc, 4*wordSize, argP); // current crc state 1964 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1965 } 1966 1967 StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table); 1968 1969 // Performance measurements show the 1word and 2word variants to be almost equivalent, 1970 // with very light advantages for the 1word variant. We chose the 1word variant for 1971 // code compactness. 1972 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3, true); 1973 1974 // Restore caller sp for c2i case and return. 1975 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 1976 __ blr(); 1977 1978 // Generate a vanilla native entry as the slow path. 1979 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)"); 1980 BIND(slow_path); 1981 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1); 1982 return start; 1983 } 1984 1985 return NULL; 1986 } 1987 1988 1989 /** 1990 * Method entry for intrinsic-candidate (non-native) methods: 1991 * int java.util.zip.CRC32C.updateBytes( int crc, byte[] b, int off, int end) 1992 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end) 1993 * Unlike CRC32, CRC32C does not have any methods marked as native 1994 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses 1995 **/ 1996 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 1997 if (UseCRC32CIntrinsics) { 1998 address start = __ pc(); // Remember stub start address (is rtn value). 1999 2000 // We don't generate local frame and don't align stack because 2001 // we not even call stub code (we generate the code inline) 2002 // and there is no safepoint on this path. 2003 2004 // Load parameters. 2005 // Z_esp is callers operand stack pointer, i.e. it points to the parameters. 2006 const Register argP = R15_esp; 2007 const Register crc = R3_ARG1; // crc value 2008 const Register data = R4_ARG2; // address of java byte array 2009 const Register dataLen = R5_ARG3; // source data len 2010 const Register table = R6_ARG4; // address of crc32c table 2011 2012 const Register t0 = R9; // scratch registers for crc calculation 2013 const Register t1 = R10; 2014 const Register t2 = R11; 2015 const Register t3 = R12; 2016 2017 const Register tc0 = R2; // registers to hold pre-calculated column addresses 2018 const Register tc1 = R7; 2019 const Register tc2 = R8; 2020 const Register tc3 = table; // table address is reconstructed at the end of kernel_crc32_* emitters 2021 2022 const Register tmp = t0; // Only used very locally to calculate byte buffer address. 2023 2024 // Arguments are reversed on java expression stack. 2025 // Calculate address of start element. 2026 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer". 2027 BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {"); 2028 // crc @ (SP + 5W) (32bit) 2029 // buf @ (SP + 3W) (64bit ptr to long array) 2030 // off @ (SP + 2W) (32bit) 2031 // dataLen @ (SP + 1W) (32bit) 2032 // data = buf + off 2033 __ ld( data, 3*wordSize, argP); // start of byte buffer 2034 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 2035 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 2036 __ lwz( crc, 5*wordSize, argP); // current crc state 2037 __ add( data, data, tmp); // Add byte buffer offset. 2038 __ sub( dataLen, dataLen, tmp); // (end_index - offset) 2039 } else { // Used for "updateBytes update". 2040 BLOCK_COMMENT("CRC32C_updateBytes {"); 2041 // crc @ (SP + 4W) (32bit) 2042 // buf @ (SP + 3W) (64bit ptr to byte array) 2043 // off @ (SP + 2W) (32bit) 2044 // dataLen @ (SP + 1W) (32bit) 2045 // data = buf + off + base_offset 2046 __ ld( data, 3*wordSize, argP); // start of byte buffer 2047 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 2048 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 2049 __ add( data, data, tmp); // add byte buffer offset 2050 __ sub( dataLen, dataLen, tmp); // (end_index - offset) 2051 __ lwz( crc, 4*wordSize, argP); // current crc state 2052 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 2053 } 2054 2055 StubRoutines::ppc64::generate_load_crc32c_table_addr(_masm, table); 2056 2057 // Performance measurements show the 1word and 2word variants to be almost equivalent, 2058 // with very light advantages for the 1word variant. We chose the 1word variant for 2059 // code compactness. 2060 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3, false); 2061 2062 // Restore caller sp for c2i case and return. 2063 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 2064 __ blr(); 2065 2066 BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}"); 2067 return start; 2068 } 2069 2070 return NULL; 2071 } 2072 2073 // ============================================================================= 2074 // Exceptions 2075 2076 void TemplateInterpreterGenerator::generate_throw_exception() { 2077 Register Rexception = R17_tos, 2078 Rcontinuation = R3_RET; 2079 2080 // -------------------------------------------------------------------------- 2081 // Entry point if an method returns with a pending exception (rethrow). 2082 Interpreter::_rethrow_exception_entry = __ pc(); 2083 { 2084 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp. 2085 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); 2086 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); 2087 2088 // Compiled code destroys templateTableBase, reload. 2089 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1); 2090 } 2091 2092 // Entry point if a interpreted method throws an exception (throw). 2093 Interpreter::_throw_exception_entry = __ pc(); 2094 { 2095 __ mr(Rexception, R3_RET); 2096 2097 __ verify_thread(); 2098 __ verify_oop(Rexception); 2099 2100 // Expression stack must be empty before entering the VM in case of an exception. 2101 __ empty_expression_stack(); 2102 // Find exception handler address and preserve exception oop. 2103 // Call C routine to find handler and jump to it. 2104 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception); 2105 __ mtctr(Rcontinuation); 2106 // Push exception for exception handler bytecodes. 2107 __ push_ptr(Rexception); 2108 2109 // Jump to exception handler (may be remove activation entry!). 2110 __ bctr(); 2111 } 2112 2113 // If the exception is not handled in the current frame the frame is 2114 // removed and the exception is rethrown (i.e. exception 2115 // continuation is _rethrow_exception). 2116 // 2117 // Note: At this point the bci is still the bxi for the instruction 2118 // which caused the exception and the expression stack is 2119 // empty. Thus, for any VM calls at this point, GC will find a legal 2120 // oop map (with empty expression stack). 2121 2122 // In current activation 2123 // tos: exception 2124 // bcp: exception bcp 2125 2126 // -------------------------------------------------------------------------- 2127 // JVMTI PopFrame support 2128 2129 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 2130 { 2131 // Set the popframe_processing bit in popframe_condition indicating that we are 2132 // currently handling popframe, so that call_VMs that may happen later do not 2133 // trigger new popframe handling cycles. 2134 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 2135 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit); 2136 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 2137 2138 // Empty the expression stack, as in normal exception handling. 2139 __ empty_expression_stack(); 2140 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false); 2141 2142 // Check to see whether we are returning to a deoptimized frame. 2143 // (The PopFrame call ensures that the caller of the popped frame is 2144 // either interpreted or compiled and deoptimizes it if compiled.) 2145 // Note that we don't compare the return PC against the 2146 // deoptimization blob's unpack entry because of the presence of 2147 // adapter frames in C2. 2148 Label Lcaller_not_deoptimized; 2149 Register return_pc = R3_ARG1; 2150 __ ld(return_pc, 0, R1_SP); 2151 __ ld(return_pc, _abi(lr), return_pc); 2152 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc); 2153 __ cmpdi(CCR0, R3_RET, 0); 2154 __ bne(CCR0, Lcaller_not_deoptimized); 2155 2156 // The deoptimized case. 2157 // In this case, we can't call dispatch_next() after the frame is 2158 // popped, but instead must save the incoming arguments and restore 2159 // them after deoptimization has occurred. 2160 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method); 2161 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2); 2162 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize); 2163 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize); 2164 __ subf(R5_ARG3, R4_ARG2, R5_ARG3); 2165 // Save these arguments. 2166 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3); 2167 2168 // Inform deoptimization that it is responsible for restoring these arguments. 2169 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit); 2170 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 2171 2172 // Return from the current method into the deoptimization blob. Will eventually 2173 // end up in the deopt interpeter entry, deoptimization prepared everything that 2174 // we will reexecute the call that called us. 2175 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2); 2176 __ mtlr(return_pc); 2177 __ blr(); 2178 2179 // The non-deoptimized case. 2180 __ bind(Lcaller_not_deoptimized); 2181 2182 // Clear the popframe condition flag. 2183 __ li(R0, 0); 2184 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 2185 2186 // Get out of the current method and re-execute the call that called us. 2187 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2); 2188 __ restore_interpreter_state(R11_scratch1); 2189 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); 2190 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); 2191 if (ProfileInterpreter) { 2192 __ set_method_data_pointer_for_bcp(); 2193 __ ld(R11_scratch1, 0, R1_SP); 2194 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1); 2195 } 2196 #if INCLUDE_JVMTI 2197 Label L_done; 2198 2199 __ lbz(R11_scratch1, 0, R14_bcp); 2200 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic); 2201 __ bne(CCR0, L_done); 2202 2203 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. 2204 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL. 2205 __ ld(R4_ARG2, 0, R18_locals); 2206 __ MacroAssembler::call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp, false); 2207 __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); 2208 __ cmpdi(CCR0, R4_ARG2, 0); 2209 __ beq(CCR0, L_done); 2210 __ std(R4_ARG2, wordSize, R15_esp); 2211 __ bind(L_done); 2212 #endif // INCLUDE_JVMTI 2213 __ dispatch_next(vtos); 2214 } 2215 // end of JVMTI PopFrame support 2216 2217 // -------------------------------------------------------------------------- 2218 // Remove activation exception entry. 2219 // This is jumped to if an interpreted method can't handle an exception itself 2220 // (we come from the throw/rethrow exception entry above). We're going to call 2221 // into the VM to find the exception handler in the caller, pop the current 2222 // frame and return the handler we calculated. 2223 Interpreter::_remove_activation_entry = __ pc(); 2224 { 2225 __ pop_ptr(Rexception); 2226 __ verify_thread(); 2227 __ verify_oop(Rexception); 2228 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread); 2229 2230 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true); 2231 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false); 2232 2233 __ get_vm_result(Rexception); 2234 2235 // We are done with this activation frame; find out where to go next. 2236 // The continuation point will be an exception handler, which expects 2237 // the following registers set up: 2238 // 2239 // RET: exception oop 2240 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled. 2241 2242 Register return_pc = R31; // Needs to survive the runtime call. 2243 __ ld(return_pc, 0, R1_SP); 2244 __ ld(return_pc, _abi(lr), return_pc); 2245 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc); 2246 2247 // Remove the current activation. 2248 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2); 2249 2250 __ mr(R4_ARG2, return_pc); 2251 __ mtlr(R3_RET); 2252 __ mr(R3_RET, Rexception); 2253 __ blr(); 2254 } 2255 } 2256 2257 // JVMTI ForceEarlyReturn support. 2258 // Returns "in the middle" of a method with a "fake" return value. 2259 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 2260 2261 Register Rscratch1 = R11_scratch1, 2262 Rscratch2 = R12_scratch2; 2263 2264 address entry = __ pc(); 2265 __ empty_expression_stack(); 2266 2267 __ load_earlyret_value(state, Rscratch1); 2268 2269 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 2270 // Clear the earlyret state. 2271 __ li(R0, 0); 2272 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1); 2273 2274 __ remove_activation(state, false, false); 2275 // Copied from TemplateTable::_return. 2276 // Restoration of lr done by remove_activation. 2277 switch (state) { 2278 // Narrow result if state is itos but result type is smaller. 2279 case btos: 2280 case ztos: 2281 case ctos: 2282 case stos: 2283 case itos: __ narrow(R17_tos); /* fall through */ 2284 case ltos: 2285 case atos: __ mr(R3_RET, R17_tos); break; 2286 case ftos: 2287 case dtos: __ fmr(F1_RET, F15_ftos); break; 2288 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need 2289 // to get visible before the reference to the object gets stored anywhere. 2290 __ membar(Assembler::StoreStore); break; 2291 default : ShouldNotReachHere(); 2292 } 2293 __ blr(); 2294 2295 return entry; 2296 } // end of ForceEarlyReturn support 2297 2298 //----------------------------------------------------------------------------- 2299 // Helper for vtos entry point generation 2300 2301 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, 2302 address& bep, 2303 address& cep, 2304 address& sep, 2305 address& aep, 2306 address& iep, 2307 address& lep, 2308 address& fep, 2309 address& dep, 2310 address& vep) { 2311 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 2312 Label L; 2313 2314 aep = __ pc(); __ push_ptr(); __ b(L); 2315 fep = __ pc(); __ push_f(); __ b(L); 2316 dep = __ pc(); __ push_d(); __ b(L); 2317 lep = __ pc(); __ push_l(); __ b(L); 2318 __ align(32, 12, 24); // align L 2319 bep = cep = sep = 2320 iep = __ pc(); __ push_i(); 2321 vep = __ pc(); 2322 __ bind(L); 2323 generate_and_dispatch(t); 2324 } 2325 2326 //----------------------------------------------------------------------------- 2327 2328 // Non-product code 2329 #ifndef PRODUCT 2330 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 2331 //__ flush_bundle(); 2332 address entry = __ pc(); 2333 2334 const char *bname = NULL; 2335 uint tsize = 0; 2336 switch(state) { 2337 case ftos: 2338 bname = "trace_code_ftos {"; 2339 tsize = 2; 2340 break; 2341 case btos: 2342 bname = "trace_code_btos {"; 2343 tsize = 2; 2344 break; 2345 case ztos: 2346 bname = "trace_code_ztos {"; 2347 tsize = 2; 2348 break; 2349 case ctos: 2350 bname = "trace_code_ctos {"; 2351 tsize = 2; 2352 break; 2353 case stos: 2354 bname = "trace_code_stos {"; 2355 tsize = 2; 2356 break; 2357 case itos: 2358 bname = "trace_code_itos {"; 2359 tsize = 2; 2360 break; 2361 case ltos: 2362 bname = "trace_code_ltos {"; 2363 tsize = 3; 2364 break; 2365 case atos: 2366 bname = "trace_code_atos {"; 2367 tsize = 2; 2368 break; 2369 case vtos: 2370 // Note: In case of vtos, the topmost of stack value could be a int or doubl 2371 // In case of a double (2 slots) we won't see the 2nd stack value. 2372 // Maybe we simply should print the topmost 3 stack slots to cope with the problem. 2373 bname = "trace_code_vtos {"; 2374 tsize = 2; 2375 2376 break; 2377 case dtos: 2378 bname = "trace_code_dtos {"; 2379 tsize = 3; 2380 break; 2381 default: 2382 ShouldNotReachHere(); 2383 } 2384 BLOCK_COMMENT(bname); 2385 2386 // Support short-cut for TraceBytecodesAt. 2387 // Don't call into the VM if we don't want to trace to speed up things. 2388 Label Lskip_vm_call; 2389 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) { 2390 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true); 2391 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true); 2392 __ ld(R11_scratch1, offs1, R11_scratch1); 2393 __ lwa(R12_scratch2, offs2, R12_scratch2); 2394 __ cmpd(CCR0, R12_scratch2, R11_scratch1); 2395 __ blt(CCR0, Lskip_vm_call); 2396 } 2397 2398 __ push(state); 2399 // Load 2 topmost expression stack values. 2400 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp); 2401 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp); 2402 __ mflr(R31); 2403 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false); 2404 __ mtlr(R31); 2405 __ pop(state); 2406 2407 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) { 2408 __ bind(Lskip_vm_call); 2409 } 2410 __ blr(); 2411 BLOCK_COMMENT("} trace_code"); 2412 return entry; 2413 } 2414 2415 void TemplateInterpreterGenerator::count_bytecode() { 2416 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true); 2417 __ lwz(R12_scratch2, offs, R11_scratch1); 2418 __ addi(R12_scratch2, R12_scratch2, 1); 2419 __ stw(R12_scratch2, offs, R11_scratch1); 2420 } 2421 2422 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { 2423 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true); 2424 __ lwz(R12_scratch2, offs, R11_scratch1); 2425 __ addi(R12_scratch2, R12_scratch2, 1); 2426 __ stw(R12_scratch2, offs, R11_scratch1); 2427 } 2428 2429 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { 2430 const Register addr = R11_scratch1, 2431 tmp = R12_scratch2; 2432 // Get index, shift out old bytecode, bring in new bytecode, and store it. 2433 // _index = (_index >> log2_number_of_codes) | 2434 // (bytecode << log2_number_of_codes); 2435 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true); 2436 __ lwz(tmp, offs1, addr); 2437 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes); 2438 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes); 2439 __ stw(tmp, offs1, addr); 2440 2441 // Bump bucket contents. 2442 // _counters[_index] ++; 2443 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true); 2444 __ sldi(tmp, tmp, LogBytesPerInt); 2445 __ add(addr, tmp, addr); 2446 __ lwz(tmp, offs2, addr); 2447 __ addi(tmp, tmp, 1); 2448 __ stw(tmp, offs2, addr); 2449 } 2450 2451 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 2452 // Call a little run-time stub to avoid blow-up for each bytecode. 2453 // The run-time runtime saves the right registers, depending on 2454 // the tosca in-state for the given template. 2455 2456 assert(Interpreter::trace_code(t->tos_in()) != NULL, 2457 "entry must have been generated"); 2458 2459 // Note: we destroy LR here. 2460 __ bl(Interpreter::trace_code(t->tos_in())); 2461 } 2462 2463 void TemplateInterpreterGenerator::stop_interpreter_at() { 2464 Label L; 2465 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true); 2466 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true); 2467 __ ld(R11_scratch1, offs1, R11_scratch1); 2468 __ lwa(R12_scratch2, offs2, R12_scratch2); 2469 __ cmpd(CCR0, R12_scratch2, R11_scratch1); 2470 __ bne(CCR0, L); 2471 __ illtrap(); 2472 __ bind(L); 2473 } 2474 2475 #endif // !PRODUCT