1 /* 2 * Copyright (c) 2014, 2023, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2015, 2023 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 "classfile/javaClasses.hpp" 29 #include "compiler/disassembler.hpp" 30 #include "gc/shared/barrierSetAssembler.hpp" 31 #include "interpreter/bytecodeHistogram.hpp" 32 #include "interpreter/interpreter.hpp" 33 #include "interpreter/interpreterRuntime.hpp" 34 #include "interpreter/interp_masm.hpp" 35 #include "interpreter/templateInterpreterGenerator.hpp" 36 #include "interpreter/templateTable.hpp" 37 #include "oops/arrayOop.hpp" 38 #include "oops/method.hpp" 39 #include "oops/methodCounters.hpp" 40 #include "oops/methodData.hpp" 41 #include "oops/oop.inline.hpp" 42 #include "oops/resolvedIndyEntry.hpp" 43 #include "oops/resolvedMethodEntry.hpp" 44 #include "prims/jvmtiExport.hpp" 45 #include "prims/jvmtiThreadState.hpp" 46 #include "runtime/arguments.hpp" 47 #include "runtime/deoptimization.hpp" 48 #include "runtime/frame.inline.hpp" 49 #include "runtime/jniHandles.hpp" 50 #include "runtime/sharedRuntime.hpp" 51 #include "runtime/stubRoutines.hpp" 52 #include "runtime/synchronizer.hpp" 53 #include "runtime/timer.hpp" 54 #include "runtime/vframeArray.hpp" 55 #include "runtime/vm_version.hpp" 56 #include "utilities/debug.hpp" 57 #include "utilities/macros.hpp" 58 59 #undef __ 60 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)-> 61 62 // Size of interpreter code. Increase if too small. Interpreter will 63 // fail with a guarantee ("not enough space for interpreter generation"); 64 // if too small. 65 // Run with +PrintInterpreter to get the VM to print out the size. 66 // Max size with JVMTI 67 int TemplateInterpreter::InterpreterCodeSize = 256*K; 68 69 #ifdef PRODUCT 70 #define BLOCK_COMMENT(str) /* nothing */ 71 #else 72 #define BLOCK_COMMENT(str) __ block_comment(str) 73 #endif 74 75 #define BIND(label) __ bind(label); BLOCK_COMMENT(#label ":") 76 77 //----------------------------------------------------------------------------- 78 79 address TemplateInterpreterGenerator::generate_slow_signature_handler() { 80 // Slow_signature handler that respects the PPC C calling conventions. 81 // 82 // We get called by the native entry code with our output register 83 // area == 8. First we call InterpreterRuntime::get_result_handler 84 // to copy the pointer to the signature string temporarily to the 85 // first C-argument and to return the result_handler in 86 // R3_RET. Since native_entry will copy the jni-pointer to the 87 // first C-argument slot later on, it is OK to occupy this slot 88 // temporarily. Then we copy the argument list on the java 89 // expression stack into native varargs format on the native stack 90 // and load arguments into argument registers. Integer arguments in 91 // the varargs vector will be sign-extended to 8 bytes. 92 // 93 // On entry: 94 // R3_ARG1 - intptr_t* Address of java argument list in memory. 95 // R15_prev_state - BytecodeInterpreter* Address of interpreter state for 96 // this method 97 // R19_method 98 // 99 // On exit (just before return instruction): 100 // R3_RET - contains the address of the result_handler. 101 // R4_ARG2 - is not updated for static methods and contains "this" otherwise. 102 // R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double, 103 // ARGi contains this argument. Otherwise, ARGi is not updated. 104 // F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double. 105 106 const int LogSizeOfTwoInstructions = 3; 107 108 // FIXME: use Argument:: GL: Argument names different numbers! 109 const int max_fp_register_arguments = 13; 110 const int max_int_register_arguments = 6; // first 2 are reserved 111 112 const Register arg_java = R21_tmp1; 113 const Register arg_c = R22_tmp2; 114 const Register signature = R23_tmp3; // is string 115 const Register sig_byte = R24_tmp4; 116 const Register fpcnt = R25_tmp5; 117 const Register argcnt = R26_tmp6; 118 const Register intSlot = R27_tmp7; 119 const Register target_sp = R28_tmp8; 120 const FloatRegister floatSlot = F0; 121 122 address entry = __ function_entry(); 123 124 __ save_LR_CR(R0); 125 __ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14)); 126 // We use target_sp for storing arguments in the C frame. 127 __ mr(target_sp, R1_SP); 128 __ push_frame_reg_args_nonvolatiles(0, R11_scratch1); 129 130 __ mr(arg_java, R3_ARG1); 131 132 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method); 133 134 // Signature is in R3_RET. Signature is callee saved. 135 __ mr(signature, R3_RET); 136 137 // Get the result handler. 138 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method); 139 140 { 141 Label L; 142 // test if static 143 // _access_flags._flags must be at offset 0. 144 // TODO PPC port: requires change in shared code. 145 //assert(in_bytes(AccessFlags::flags_offset()) == 0, 146 // "MethodDesc._access_flags == MethodDesc._access_flags._flags"); 147 // _access_flags must be a 32 bit value. 148 assert(sizeof(AccessFlags) == 4, "wrong size"); 149 __ lwa(R11_scratch1/*access_flags*/, method_(access_flags)); 150 // testbit with condition register. 151 __ testbitdi(CCR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT); 152 __ btrue(CCR0, L); 153 // For non-static functions, pass "this" in R4_ARG2 and copy it 154 // to 2nd C-arg slot. 155 // We need to box the Java object here, so we use arg_java 156 // (address of current Java stack slot) as argument and don't 157 // dereference it as in case of ints, floats, etc. 158 __ mr(R4_ARG2, arg_java); 159 __ addi(arg_java, arg_java, -BytesPerWord); 160 __ std(R4_ARG2, _abi0(carg_2), target_sp); 161 __ bind(L); 162 } 163 164 // Will be incremented directly after loop_start. argcnt=0 165 // corresponds to 3rd C argument. 166 __ li(argcnt, -1); 167 // arg_c points to 3rd C argument 168 __ addi(arg_c, target_sp, _abi0(carg_3)); 169 // no floating-point args parsed so far 170 __ li(fpcnt, 0); 171 172 Label move_intSlot_to_ARG, move_floatSlot_to_FARG; 173 Label loop_start, loop_end; 174 Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed; 175 176 // signature points to '(' at entry 177 #ifdef ASSERT 178 __ lbz(sig_byte, 0, signature); 179 __ cmplwi(CCR0, sig_byte, '('); 180 __ bne(CCR0, do_dontreachhere); 181 #endif 182 183 __ bind(loop_start); 184 185 __ addi(argcnt, argcnt, 1); 186 __ lbzu(sig_byte, 1, signature); 187 188 __ cmplwi(CCR0, sig_byte, ')'); // end of signature 189 __ beq(CCR0, loop_end); 190 191 __ cmplwi(CCR0, sig_byte, 'B'); // byte 192 __ beq(CCR0, do_int); 193 194 __ cmplwi(CCR0, sig_byte, 'C'); // char 195 __ beq(CCR0, do_int); 196 197 __ cmplwi(CCR0, sig_byte, 'D'); // double 198 __ beq(CCR0, do_double); 199 200 __ cmplwi(CCR0, sig_byte, 'F'); // float 201 __ beq(CCR0, do_float); 202 203 __ cmplwi(CCR0, sig_byte, 'I'); // int 204 __ beq(CCR0, do_int); 205 206 __ cmplwi(CCR0, sig_byte, 'J'); // long 207 __ beq(CCR0, do_long); 208 209 __ cmplwi(CCR0, sig_byte, 'S'); // short 210 __ beq(CCR0, do_int); 211 212 __ cmplwi(CCR0, sig_byte, 'Z'); // boolean 213 __ beq(CCR0, do_int); 214 215 __ cmplwi(CCR0, sig_byte, 'L'); // object 216 __ beq(CCR0, do_object); 217 218 __ cmplwi(CCR0, sig_byte, '['); // array 219 __ beq(CCR0, do_array); 220 221 // __ cmplwi(CCR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type 222 // __ beq(CCR0, do_void); 223 224 __ bind(do_dontreachhere); 225 226 __ unimplemented("ShouldNotReachHere in slow_signature_handler"); 227 228 __ bind(do_array); 229 230 { 231 Label start_skip, end_skip; 232 233 __ bind(start_skip); 234 __ lbzu(sig_byte, 1, signature); 235 __ cmplwi(CCR0, sig_byte, '['); 236 __ beq(CCR0, start_skip); // skip further brackets 237 __ cmplwi(CCR0, sig_byte, '9'); 238 __ bgt(CCR0, end_skip); // no optional size 239 __ cmplwi(CCR0, sig_byte, '0'); 240 __ bge(CCR0, start_skip); // skip optional size 241 __ bind(end_skip); 242 243 __ cmplwi(CCR0, sig_byte, 'L'); 244 __ beq(CCR0, do_object); // for arrays of objects, the name of the object must be skipped 245 __ b(do_boxed); // otherwise, go directly to do_boxed 246 } 247 248 __ bind(do_object); 249 { 250 Label L; 251 __ bind(L); 252 __ lbzu(sig_byte, 1, signature); 253 __ cmplwi(CCR0, sig_byte, ';'); 254 __ bne(CCR0, L); 255 } 256 // Need to box the Java object here, so we use arg_java (address of 257 // current Java stack slot) as argument and don't dereference it as 258 // in case of ints, floats, etc. 259 Label do_null; 260 __ bind(do_boxed); 261 __ ld(R0,0, arg_java); 262 __ cmpdi(CCR0, R0, 0); 263 __ li(intSlot,0); 264 __ beq(CCR0, do_null); 265 __ mr(intSlot, arg_java); 266 __ bind(do_null); 267 __ std(intSlot, 0, arg_c); 268 __ addi(arg_java, arg_java, -BytesPerWord); 269 __ addi(arg_c, arg_c, BytesPerWord); 270 __ cmplwi(CCR0, argcnt, max_int_register_arguments); 271 __ blt(CCR0, move_intSlot_to_ARG); 272 __ b(loop_start); 273 274 __ bind(do_int); 275 __ lwa(intSlot, 0, arg_java); 276 __ std(intSlot, 0, arg_c); 277 __ addi(arg_java, arg_java, -BytesPerWord); 278 __ addi(arg_c, arg_c, BytesPerWord); 279 __ cmplwi(CCR0, argcnt, max_int_register_arguments); 280 __ blt(CCR0, move_intSlot_to_ARG); 281 __ b(loop_start); 282 283 __ bind(do_long); 284 __ ld(intSlot, -BytesPerWord, arg_java); 285 __ std(intSlot, 0, arg_c); 286 __ addi(arg_java, arg_java, - 2 * BytesPerWord); 287 __ addi(arg_c, arg_c, BytesPerWord); 288 __ cmplwi(CCR0, argcnt, max_int_register_arguments); 289 __ blt(CCR0, move_intSlot_to_ARG); 290 __ b(loop_start); 291 292 __ bind(do_float); 293 __ lfs(floatSlot, 0, arg_java); 294 __ stfs(floatSlot, Argument::float_on_stack_offset_in_bytes_c, arg_c); 295 __ addi(arg_java, arg_java, -BytesPerWord); 296 __ addi(arg_c, arg_c, BytesPerWord); 297 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments); 298 __ blt(CCR0, move_floatSlot_to_FARG); 299 __ b(loop_start); 300 301 __ bind(do_double); 302 __ lfd(floatSlot, - BytesPerWord, arg_java); 303 __ stfd(floatSlot, 0, arg_c); 304 __ addi(arg_java, arg_java, - 2 * BytesPerWord); 305 __ addi(arg_c, arg_c, BytesPerWord); 306 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments); 307 __ blt(CCR0, move_floatSlot_to_FARG); 308 __ b(loop_start); 309 310 __ bind(loop_end); 311 312 __ pop_frame(); 313 __ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14)); 314 __ restore_LR_CR(R0); 315 316 __ blr(); 317 318 Label move_int_arg, move_float_arg; 319 __ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions) 320 __ mr(R5_ARG3, intSlot); __ b(loop_start); 321 __ mr(R6_ARG4, intSlot); __ b(loop_start); 322 __ mr(R7_ARG5, intSlot); __ b(loop_start); 323 __ mr(R8_ARG6, intSlot); __ b(loop_start); 324 __ mr(R9_ARG7, intSlot); __ b(loop_start); 325 __ mr(R10_ARG8, intSlot); __ b(loop_start); 326 327 __ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions) 328 __ fmr(F1_ARG1, floatSlot); __ b(loop_start); 329 __ fmr(F2_ARG2, floatSlot); __ b(loop_start); 330 __ fmr(F3_ARG3, floatSlot); __ b(loop_start); 331 __ fmr(F4_ARG4, floatSlot); __ b(loop_start); 332 __ fmr(F5_ARG5, floatSlot); __ b(loop_start); 333 __ fmr(F6_ARG6, floatSlot); __ b(loop_start); 334 __ fmr(F7_ARG7, floatSlot); __ b(loop_start); 335 __ fmr(F8_ARG8, floatSlot); __ b(loop_start); 336 __ fmr(F9_ARG9, floatSlot); __ b(loop_start); 337 __ fmr(F10_ARG10, floatSlot); __ b(loop_start); 338 __ fmr(F11_ARG11, floatSlot); __ b(loop_start); 339 __ fmr(F12_ARG12, floatSlot); __ b(loop_start); 340 __ fmr(F13_ARG13, floatSlot); __ b(loop_start); 341 342 __ bind(move_intSlot_to_ARG); 343 __ sldi(R0, argcnt, LogSizeOfTwoInstructions); 344 __ load_const(R11_scratch1, move_int_arg); // Label must be bound here. 345 __ add(R11_scratch1, R0, R11_scratch1); 346 __ mtctr(R11_scratch1/*branch_target*/); 347 __ bctr(); 348 __ bind(move_floatSlot_to_FARG); 349 __ sldi(R0, fpcnt, LogSizeOfTwoInstructions); 350 __ addi(fpcnt, fpcnt, 1); 351 __ load_const(R11_scratch1, move_float_arg); // Label must be bound here. 352 __ add(R11_scratch1, R0, R11_scratch1); 353 __ mtctr(R11_scratch1/*branch_target*/); 354 __ bctr(); 355 356 return entry; 357 } 358 359 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) { 360 // 361 // Registers alive 362 // R3_RET 363 // LR 364 // 365 // Registers updated 366 // R3_RET 367 // 368 369 Label done; 370 address entry = __ pc(); 371 372 switch (type) { 373 case T_BOOLEAN: 374 // convert !=0 to 1 375 __ neg(R0, R3_RET); 376 __ orr(R0, R3_RET, R0); 377 __ srwi(R3_RET, R0, 31); 378 break; 379 case T_BYTE: 380 // sign extend 8 bits 381 __ extsb(R3_RET, R3_RET); 382 break; 383 case T_CHAR: 384 // zero extend 16 bits 385 __ clrldi(R3_RET, R3_RET, 48); 386 break; 387 case T_SHORT: 388 // sign extend 16 bits 389 __ extsh(R3_RET, R3_RET); 390 break; 391 case T_INT: 392 // sign extend 32 bits 393 __ extsw(R3_RET, R3_RET); 394 break; 395 case T_LONG: 396 break; 397 case T_OBJECT: 398 // JNIHandles::resolve result. 399 __ resolve_jobject(R3_RET, R11_scratch1, R31, MacroAssembler::PRESERVATION_FRAME_LR); // kills R31 400 break; 401 case T_FLOAT: 402 break; 403 case T_DOUBLE: 404 break; 405 case T_VOID: 406 break; 407 default: ShouldNotReachHere(); 408 } 409 410 BIND(done); 411 __ blr(); 412 413 return entry; 414 } 415 416 // Abstract method entry. 417 // 418 address TemplateInterpreterGenerator::generate_abstract_entry(void) { 419 address entry = __ pc(); 420 421 // 422 // Registers alive 423 // R16_thread - JavaThread* 424 // R19_method - callee's method (method to be invoked) 425 // R1_SP - SP prepared such that caller's outgoing args are near top 426 // LR - return address to caller 427 // 428 // Stack layout at this point: 429 // 430 // 0 [TOP_IJAVA_FRAME_ABI] <-- R1_SP 431 // alignment (optional) 432 // [outgoing Java arguments] 433 // ... 434 // PARENT [PARENT_IJAVA_FRAME_ABI] 435 // ... 436 // 437 438 // Can't use call_VM here because we have not set up a new 439 // interpreter state. Make the call to the vm and make it look like 440 // our caller set up the JavaFrameAnchor. 441 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/); 442 443 // Push a new C frame and save LR. 444 __ save_LR_CR(R0); 445 __ push_frame_reg_args(0, R11_scratch1); 446 447 // This is not a leaf but we have a JavaFrameAnchor now and we will 448 // check (create) exceptions afterward so this is ok. 449 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod), 450 R16_thread, R19_method); 451 452 // Pop the C frame and restore LR. 453 __ pop_frame(); 454 __ restore_LR_CR(R0); 455 456 // Reset JavaFrameAnchor from call_VM_leaf above. 457 __ reset_last_Java_frame(); 458 459 // We don't know our caller, so jump to the general forward exception stub, 460 // which will also pop our full frame off. Satisfy the interface of 461 // SharedRuntime::generate_forward_exception() 462 __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0); 463 __ mtctr(R11_scratch1); 464 __ bctr(); 465 466 return entry; 467 } 468 469 // Interpreter intrinsic for WeakReference.get(). 470 // 1. Don't push a full blown frame and go on dispatching, but fetch the value 471 // into R8 and return quickly 472 // 2. If G1 is active we *must* execute this intrinsic for corrrectness: 473 // It contains a GC barrier which puts the reference into the satb buffer 474 // to indicate that someone holds a strong reference to the object the 475 // weak ref points to! 476 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) { 477 // Code: _aload_0, _getfield, _areturn 478 // parameter size = 1 479 // 480 // The code that gets generated by this routine is split into 2 parts: 481 // 1. the "intrinsified" code for G1 (or any SATB based GC), 482 // 2. the slow path - which is an expansion of the regular method entry. 483 // 484 // Notes: 485 // * In the G1 code we do not check whether we need to block for 486 // a safepoint. If G1 is enabled then we must execute the specialized 487 // code for Reference.get (except when the Reference object is null) 488 // so that we can log the value in the referent field with an SATB 489 // update buffer. 490 // If the code for the getfield template is modified so that the 491 // G1 pre-barrier code is executed when the current method is 492 // Reference.get() then going through the normal method entry 493 // will be fine. 494 // * The G1 code can, however, check the receiver object (the instance 495 // of java.lang.Reference) and jump to the slow path if null. If the 496 // Reference object is null then we obviously cannot fetch the referent 497 // and so we don't need to call the G1 pre-barrier. Thus we can use the 498 // regular method entry code to generate the NPE. 499 // 500 501 address entry = __ pc(); 502 503 const int referent_offset = java_lang_ref_Reference::referent_offset(); 504 505 Label slow_path; 506 507 // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH); 508 509 // In the G1 code we don't check if we need to reach a safepoint. We 510 // continue and the thread will safepoint at the next bytecode dispatch. 511 512 // If the receiver is null then it is OK to jump to the slow path. 513 __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver 514 515 // Check if receiver == nullptr and go the slow path. 516 __ cmpdi(CCR0, R3_RET, 0); 517 __ beq(CCR0, slow_path); 518 519 __ load_heap_oop(R3_RET, referent_offset, R3_RET, 520 /* non-volatile temp */ R31, R11_scratch1, 521 MacroAssembler::PRESERVATION_FRAME_LR, 522 ON_WEAK_OOP_REF); 523 524 // Generate the G1 pre-barrier code to log the value of 525 // the referent field in an SATB buffer. Note with 526 // these parameters the pre-barrier does not generate 527 // the load of the previous value. 528 529 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted). 530 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0); 531 532 __ blr(); 533 534 __ bind(slow_path); 535 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1); 536 return entry; 537 } 538 539 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { 540 address entry = __ pc(); 541 542 // Expression stack must be empty before entering the VM if an 543 // exception happened. 544 __ empty_expression_stack(); 545 // Throw exception. 546 __ call_VM(noreg, 547 CAST_FROM_FN_PTR(address, 548 InterpreterRuntime::throw_StackOverflowError)); 549 return entry; 550 } 551 552 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() { 553 address entry = __ pc(); 554 __ empty_expression_stack(); 555 // R4_ARG2 already contains the array. 556 // Index is in R17_tos. 557 __ mr(R5_ARG3, R17_tos); 558 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), R4_ARG2, R5_ARG3); 559 return entry; 560 } 561 562 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 563 address entry = __ pc(); 564 // Expression stack must be empty before entering the VM if an 565 // exception happened. 566 __ empty_expression_stack(); 567 568 // Load exception object. 569 // Thread will be loaded to R3_ARG1. 570 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos); 571 #ifdef ASSERT 572 // Above call must not return here since exception pending. 573 __ should_not_reach_here(); 574 #endif 575 return entry; 576 } 577 578 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) { 579 address entry = __ pc(); 580 //__ untested("generate_exception_handler_common"); 581 Register Rexception = R17_tos; 582 583 // Expression stack must be empty before entering the VM if an exception happened. 584 __ empty_expression_stack(); 585 586 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1); 587 if (pass_oop) { 588 __ mr(R5_ARG3, Rexception); 589 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception)); 590 } else { 591 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1); 592 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception)); 593 } 594 595 // Throw exception. 596 __ mr(R3_ARG1, Rexception); 597 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2); 598 __ mtctr(R11_scratch1); 599 __ bctr(); 600 601 return entry; 602 } 603 604 // This entry is returned to when a call returns to the interpreter. 605 // When we arrive here, we expect that the callee stack frame is already popped. 606 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) { 607 address entry = __ pc(); 608 609 // Move the value out of the return register back to the TOS cache of current frame. 610 switch (state) { 611 case ltos: 612 case btos: 613 case ztos: 614 case ctos: 615 case stos: 616 case atos: 617 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache 618 case ftos: 619 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET 620 case vtos: break; // Nothing to do, this was a void return. 621 default : ShouldNotReachHere(); 622 } 623 624 __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/); 625 626 // Compiled code destroys templateTableBase, reload. 627 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2); 628 629 if (state == atos) { 630 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2); 631 } 632 633 const Register cache = R11_scratch1; 634 const Register size = R12_scratch2; 635 if (index_size == sizeof(u4)) { 636 __ load_resolved_indy_entry(cache, size /* tmp */); 637 __ lhz(size, in_bytes(ResolvedIndyEntry::num_parameters_offset()), cache); 638 } else { 639 assert(index_size == sizeof(u2), "Can only be u2"); 640 __ load_method_entry(cache, size /* tmp */); 641 __ lhz(size, in_bytes(ResolvedMethodEntry::num_parameters_offset()), cache); 642 } 643 __ sldi(size, size, Interpreter::logStackElementSize); 644 __ add(R15_esp, R15_esp, size); 645 646 __ check_and_handle_popframe(R11_scratch1); 647 __ check_and_handle_earlyret(R11_scratch1); 648 649 __ dispatch_next(state, step); 650 return entry; 651 } 652 653 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) { 654 address entry = __ pc(); 655 // If state != vtos, we're returning from a native method, which put it's result 656 // into the result register. So move the value out of the return register back 657 // to the TOS cache of current frame. 658 659 switch (state) { 660 case ltos: 661 case btos: 662 case ztos: 663 case ctos: 664 case stos: 665 case atos: 666 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache 667 case ftos: 668 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET 669 case vtos: break; // Nothing to do, this was a void return. 670 default : ShouldNotReachHere(); 671 } 672 673 // Load LcpoolCache @@@ should be already set! 674 __ get_constant_pool_cache(R27_constPoolCache); 675 676 // Handle a pending exception, fall through if none. 677 __ check_and_forward_exception(R11_scratch1, R12_scratch2); 678 679 // Start executing bytecodes. 680 if (continuation == nullptr) { 681 __ dispatch_next(state, step); 682 } else { 683 __ jump_to_entry(continuation, R11_scratch1); 684 } 685 686 return entry; 687 } 688 689 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) { 690 address entry = __ pc(); 691 692 __ push(state); 693 __ push_cont_fastpath(); 694 __ call_VM(noreg, runtime_entry); 695 __ pop_cont_fastpath(); 696 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); 697 698 return entry; 699 } 700 701 address TemplateInterpreterGenerator::generate_cont_preempt_rerun_interpreter_adapter() { 702 return nullptr; 703 } 704 705 706 // Helpers for commoning out cases in the various type of method entries. 707 708 // Increment invocation count & check for overflow. 709 // 710 // Note: checking for negative value instead of overflow 711 // so we have a 'sticky' overflow test. 712 // 713 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) { 714 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not. 715 Register Rscratch1 = R11_scratch1; 716 Register Rscratch2 = R12_scratch2; 717 Register R3_counters = R3_ARG1; 718 Label done; 719 720 const int increment = InvocationCounter::count_increment; 721 Label no_mdo; 722 if (ProfileInterpreter) { 723 const Register Rmdo = R3_counters; 724 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method); 725 __ cmpdi(CCR0, Rmdo, 0); 726 __ beq(CCR0, no_mdo); 727 728 // Increment invocation counter in the MDO. 729 const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset()); 730 __ lwz(Rscratch2, mdo_ic_offs, Rmdo); 731 __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo); 732 __ addi(Rscratch2, Rscratch2, increment); 733 __ stw(Rscratch2, mdo_ic_offs, Rmdo); 734 __ and_(Rscratch1, Rscratch2, Rscratch1); 735 __ bne(CCR0, done); 736 __ b(*overflow); 737 } 738 739 // Increment counter in MethodCounters*. 740 const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset()); 741 __ bind(no_mdo); 742 __ get_method_counters(R19_method, R3_counters, done); 743 __ lwz(Rscratch2, mo_ic_offs, R3_counters); 744 __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters); 745 __ addi(Rscratch2, Rscratch2, increment); 746 __ stw(Rscratch2, mo_ic_offs, R3_counters); 747 __ and_(Rscratch1, Rscratch2, Rscratch1); 748 __ beq(CCR0, *overflow); 749 750 __ bind(done); 751 } 752 753 // Generate code to initiate compilation on invocation counter overflow. 754 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) { 755 // Generate code to initiate compilation on the counter overflow. 756 757 // InterpreterRuntime::frequency_counter_overflow takes one arguments, 758 // which indicates if the counter overflow occurs at a backwards branch (null bcp) 759 // We pass zero in. 760 // The call returns the address of the verified entry point for the method or null 761 // if the compilation did not complete (either went background or bailed out). 762 // 763 // Unlike the C++ interpreter above: Check exceptions! 764 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed 765 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur. 766 767 __ li(R4_ARG2, 0); 768 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true); 769 770 // Returns verified_entry_point or null. 771 // We ignore it in any case. 772 __ b(continue_entry); 773 } 774 775 // See if we've got enough room on the stack for locals plus overhead below 776 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError 777 // without going through the signal handler, i.e., reserved and yellow zones 778 // will not be made usable. The shadow zone must suffice to handle the 779 // overflow. 780 // 781 // Kills Rmem_frame_size, Rscratch1. 782 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) { 783 Label done; 784 assert_different_registers(Rmem_frame_size, Rscratch1); 785 786 BLOCK_COMMENT("stack_overflow_check_with_compare {"); 787 __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size); 788 __ ld(Rscratch1, thread_(stack_overflow_limit)); 789 __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1); 790 __ bgt(CCR0/*is_stack_overflow*/, done); 791 792 // The stack overflows. Load target address of the runtime stub and call it. 793 assert(StubRoutines::throw_StackOverflowError_entry() != nullptr, "generated in wrong order"); 794 __ load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0); 795 __ mtctr(Rscratch1); 796 // Restore caller_sp (c2i adapter may exist, but no shrinking of interpreted caller frame). 797 #ifdef ASSERT 798 Label frame_not_shrunk; 799 __ cmpld(CCR0, R1_SP, R21_sender_SP); 800 __ ble(CCR0, frame_not_shrunk); 801 __ stop("frame shrunk"); 802 __ bind(frame_not_shrunk); 803 __ ld(Rscratch1, 0, R1_SP); 804 __ ld(R0, 0, R21_sender_SP); 805 __ cmpd(CCR0, R0, Rscratch1); 806 __ asm_assert_eq("backlink"); 807 #endif // ASSERT 808 __ mr(R1_SP, R21_sender_SP); 809 __ bctr(); 810 811 __ align(32, 12); 812 __ bind(done); 813 BLOCK_COMMENT("} stack_overflow_check_with_compare"); 814 } 815 816 // Lock the current method, interpreter register window must be set up! 817 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) { 818 const Register Robj_to_lock = Rscratch2; 819 820 { 821 if (!flags_preloaded) { 822 __ lwz(Rflags, method_(access_flags)); 823 } 824 825 #ifdef ASSERT 826 // Check if methods needs synchronization. 827 { 828 Label Lok; 829 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT); 830 __ btrue(CCR0,Lok); 831 __ stop("method doesn't need synchronization"); 832 __ bind(Lok); 833 } 834 #endif // ASSERT 835 } 836 837 // Get synchronization object to Rscratch2. 838 { 839 Label Lstatic; 840 Label Ldone; 841 842 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT); 843 __ btrue(CCR0, Lstatic); 844 845 // Non-static case: load receiver obj from stack and we're done. 846 __ ld(Robj_to_lock, R18_locals); 847 __ b(Ldone); 848 849 __ bind(Lstatic); // Static case: Lock the java mirror 850 // Load mirror from interpreter frame. 851 __ ld(Robj_to_lock, _abi0(callers_sp), R1_SP); 852 __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock); 853 854 __ bind(Ldone); 855 __ verify_oop(Robj_to_lock); 856 } 857 858 // Got the oop to lock => execute! 859 __ add_monitor_to_stack(true, Rscratch1, R0); 860 861 __ std(Robj_to_lock, in_bytes(BasicObjectLock::obj_offset()), R26_monitor); 862 __ lock_object(R26_monitor, Robj_to_lock); 863 } 864 865 // Generate a fixed interpreter frame for pure interpreter 866 // and I2N native transition frames. 867 // 868 // Before (stack grows downwards): 869 // 870 // | ... | 871 // |------------- | 872 // | java arg0 | 873 // | ... | 874 // | java argn | 875 // | | <- R15_esp 876 // | | 877 // |--------------| 878 // | abi_112 | 879 // | | <- R1_SP 880 // |==============| 881 // 882 // 883 // After: 884 // 885 // | ... | 886 // | java arg0 |<- R18_locals 887 // | ... | 888 // | java argn | 889 // |--------------| 890 // | | 891 // | java locals | 892 // | | 893 // |--------------| 894 // | abi_48 | 895 // |==============| 896 // | | 897 // | istate | 898 // | | 899 // |--------------| 900 // | monitor |<- R26_monitor 901 // |--------------| 902 // | |<- R15_esp 903 // | expression | 904 // | stack | 905 // | | 906 // |--------------| 907 // | | 908 // | abi_112 |<- R1_SP 909 // |==============| 910 // 911 // The top most frame needs an abi space of 112 bytes. This space is needed, 912 // since we call to c. The c function may spill their arguments to the caller 913 // frame. When we call to java, we don't need these spill slots. In order to save 914 // space on the stack, we resize the caller. However, java locals reside in 915 // the caller frame and the frame has to be increased. The frame_size for the 916 // current frame was calculated based on max_stack as size for the expression 917 // stack. At the call, just a part of the expression stack might be used. 918 // We don't want to waste this space and cut the frame back accordingly. 919 // The resulting amount for resizing is calculated as follows: 920 // resize = (number_of_locals - number_of_arguments) * slot_size 921 // + (R1_SP - R15_esp) + 48 922 // 923 // The size for the callee frame is calculated: 924 // framesize = 112 + max_stack + monitor + state_size 925 // 926 // maxstack: Max number of slots on the expression stack, loaded from the method. 927 // monitor: We statically reserve room for one monitor object. 928 // state_size: We save the current state of the interpreter to this area. 929 // 930 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) { 931 Register Rparent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes. 932 Rtop_frame_size = R7_ARG5, 933 Rconst_method = R8_ARG6, 934 Rconst_pool = R9_ARG7, 935 Rmirror = R10_ARG8; 936 937 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, Rparent_frame_resize, Rtop_frame_size, 938 Rconst_method, Rconst_pool); 939 940 __ ld(Rconst_method, method_(const)); 941 __ lhz(Rsize_of_parameters /* number of params */, 942 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method); 943 if (native_call) { 944 // If we're calling a native method, we reserve space for the worst-case signature 945 // handler varargs vector, which is max(Argument::n_int_register_parameters_c, parameter_count+2). 946 // We add two slots to the parameter_count, one for the jni 947 // environment and one for a possible native mirror. 948 Label skip_native_calculate_max_stack; 949 __ addi(Rtop_frame_size, Rsize_of_parameters, 2); 950 __ cmpwi(CCR0, Rtop_frame_size, Argument::n_int_register_parameters_c); 951 __ bge(CCR0, skip_native_calculate_max_stack); 952 __ li(Rtop_frame_size, Argument::n_int_register_parameters_c); 953 __ bind(skip_native_calculate_max_stack); 954 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize); 955 __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize); 956 __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize! 957 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters. 958 } else { 959 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method); 960 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize); 961 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize); 962 __ lhz(Rtop_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method); 963 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0 964 __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize! 965 __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize); 966 __ add(Rparent_frame_resize, Rparent_frame_resize, R11_scratch1); 967 } 968 969 // Compute top frame size. 970 __ addi(Rtop_frame_size, Rtop_frame_size, frame::top_ijava_frame_abi_size + frame::ijava_state_size); 971 972 // Cut back area between esp and max_stack. 973 __ addi(Rparent_frame_resize, Rparent_frame_resize, frame::parent_ijava_frame_abi_size - Interpreter::stackElementSize); 974 975 __ round_to(Rtop_frame_size, frame::alignment_in_bytes); 976 __ round_to(Rparent_frame_resize, frame::alignment_in_bytes); 977 // Rparent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size. 978 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48. 979 980 if (!native_call) { 981 // Stack overflow check. 982 // Native calls don't need the stack size check since they have no 983 // expression stack and the arguments are already on the stack and 984 // we only add a handful of words to the stack. 985 __ add(R11_scratch1, Rparent_frame_resize, Rtop_frame_size); 986 generate_stack_overflow_check(R11_scratch1, R12_scratch2); 987 } 988 989 // Set up interpreter state registers. 990 991 __ add(R18_locals, R15_esp, Rsize_of_parameters); 992 __ ld(Rconst_pool, in_bytes(ConstMethod::constants_offset()), Rconst_method); 993 __ ld(R27_constPoolCache, ConstantPool::cache_offset(), Rconst_pool); 994 995 // Set method data pointer. 996 if (ProfileInterpreter) { 997 Label zero_continue; 998 __ ld(R28_mdx, method_(method_data)); 999 __ cmpdi(CCR0, R28_mdx, 0); 1000 __ beq(CCR0, zero_continue); 1001 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1002 __ bind(zero_continue); 1003 } 1004 1005 if (native_call) { 1006 __ li(R14_bcp, 0); // Must initialize. 1007 } else { 1008 __ addi(R14_bcp, Rconst_method, in_bytes(ConstMethod::codes_offset())); 1009 } 1010 1011 // Resize parent frame. 1012 __ mflr(R12_scratch2); 1013 __ neg(Rparent_frame_resize, Rparent_frame_resize); 1014 __ resize_frame(Rparent_frame_resize, R11_scratch1); 1015 __ std(R12_scratch2, _abi0(lr), R1_SP); 1016 1017 // Get mirror and store it in the frame as GC root for this Method*. 1018 __ ld(Rmirror, ConstantPool::pool_holder_offset(), Rconst_pool); 1019 __ ld(Rmirror, in_bytes(Klass::java_mirror_offset()), Rmirror); 1020 __ resolve_oop_handle(Rmirror, R11_scratch1, R12_scratch2, MacroAssembler::PRESERVATION_FRAME_LR_GP_REGS); 1021 1022 __ addi(R26_monitor, R1_SP, -frame::ijava_state_size); 1023 __ addi(R15_esp, R26_monitor, -Interpreter::stackElementSize); 1024 1025 // Store values. 1026 __ std(R19_method, _ijava_state_neg(method), R1_SP); 1027 __ std(Rmirror, _ijava_state_neg(mirror), R1_SP); 1028 __ sub(R12_scratch2, R18_locals, R1_SP); 1029 __ srdi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize); 1030 // Store relativized R18_locals, see frame::interpreter_frame_locals(). 1031 __ std(R12_scratch2, _ijava_state_neg(locals), R1_SP); 1032 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP); 1033 1034 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only 1035 // be found in the frame after save_interpreter_state is done. This is always true 1036 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong, 1037 // because e.g. frame::interpreter_frame_bcp() will not access the correct value 1038 // (Enhanced Stack Trace). 1039 // The signal handler does not save the interpreter state into the frame. 1040 1041 // We have to initialize some of these frame slots for native calls (accessed by GC). 1042 // Also initialize them for non-native calls for better tool support (even though 1043 // you may not get the most recent version as described above). 1044 __ li(R0, 0); 1045 __ li(R12_scratch2, -(frame::ijava_state_size / wordSize)); 1046 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP); 1047 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP); 1048 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); } 1049 __ sub(R12_scratch2, R15_esp, R1_SP); 1050 __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize); 1051 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP); 1052 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); // only used for native_call 1053 1054 // Store sender's SP and this frame's top SP. 1055 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP); 1056 __ neg(R12_scratch2, Rtop_frame_size); 1057 __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize); 1058 // Store relativized top_frame_sp 1059 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP); 1060 1061 // Push top frame. 1062 __ push_frame(Rtop_frame_size, R11_scratch1); 1063 } 1064 1065 // End of helpers 1066 1067 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) { 1068 1069 // Decide what to do: Use same platform specific instructions and runtime calls as compilers. 1070 bool use_instruction = false; 1071 address runtime_entry = nullptr; 1072 int num_args = 1; 1073 bool double_precision = true; 1074 1075 // PPC64 specific: 1076 switch (kind) { 1077 case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break; 1078 case Interpreter::java_lang_math_abs: use_instruction = true; break; 1079 case Interpreter::java_lang_math_fmaF: 1080 case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break; 1081 default: break; // Fall back to runtime call. 1082 } 1083 1084 switch (kind) { 1085 case Interpreter::java_lang_math_sin : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); break; 1086 case Interpreter::java_lang_math_cos : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); break; 1087 case Interpreter::java_lang_math_tan : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); break; 1088 case Interpreter::java_lang_math_abs : /* run interpreted */ break; 1089 case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); break; 1090 case Interpreter::java_lang_math_log : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); break; 1091 case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break; 1092 case Interpreter::java_lang_math_pow : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break; 1093 case Interpreter::java_lang_math_exp : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); break; 1094 case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break; 1095 case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break; 1096 default: ShouldNotReachHere(); 1097 } 1098 1099 // Use normal entry if neither instruction nor runtime call is used. 1100 if (!use_instruction && runtime_entry == nullptr) return nullptr; 1101 1102 address entry = __ pc(); 1103 1104 // Load arguments 1105 assert(num_args <= 13, "passed in registers"); 1106 if (double_precision) { 1107 int offset = (2 * num_args - 1) * Interpreter::stackElementSize; 1108 for (int i = 0; i < num_args; ++i) { 1109 __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp); 1110 offset -= 2 * Interpreter::stackElementSize; 1111 } 1112 } else { 1113 int offset = num_args * Interpreter::stackElementSize; 1114 for (int i = 0; i < num_args; ++i) { 1115 __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp); 1116 offset -= Interpreter::stackElementSize; 1117 } 1118 } 1119 1120 if (use_instruction) { 1121 switch (kind) { 1122 case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1); break; 1123 case Interpreter::java_lang_math_abs: __ fabs(F1_RET, F1); break; 1124 case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break; 1125 case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3); break; 1126 default: ShouldNotReachHere(); 1127 } 1128 } else { 1129 // Comment: Can use tail call if the unextended frame is always C ABI compliant: 1130 //__ load_const_optimized(R12_scratch2, runtime_entry, R0); 1131 //__ call_c_and_return_to_caller(R12_scratch2); 1132 1133 // Push a new C frame and save LR. 1134 __ save_LR_CR(R0); 1135 __ push_frame_reg_args(0, R11_scratch1); 1136 1137 __ call_VM_leaf(runtime_entry); 1138 1139 // Pop the C frame and restore LR. 1140 __ pop_frame(); 1141 __ restore_LR_CR(R0); 1142 } 1143 1144 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted). 1145 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0); 1146 __ blr(); 1147 1148 __ flush(); 1149 1150 return entry; 1151 } 1152 1153 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) { 1154 // Quick & dirty stack overflow checking: bang the stack & handle trap. 1155 // Note that we do the banging after the frame is setup, since the exception 1156 // handling code expects to find a valid interpreter frame on the stack. 1157 // Doing the banging earlier fails if the caller frame is not an interpreter 1158 // frame. 1159 // (Also, the exception throwing code expects to unlock any synchronized 1160 // method receiever, so do the banging after locking the receiver.) 1161 1162 // Bang each page in the shadow zone. We can't assume it's been done for 1163 // an interpreter frame with greater than a page of locals, so each page 1164 // needs to be checked. Only true for non-native. 1165 const size_t page_size = os::vm_page_size(); 1166 const int n_shadow_pages = StackOverflow::stack_shadow_zone_size() / page_size; 1167 const int start_page = native_call ? n_shadow_pages : 1; 1168 BLOCK_COMMENT("bang_stack_shadow_pages:"); 1169 for (int pages = start_page; pages <= n_shadow_pages; pages++) { 1170 __ bang_stack_with_offset(pages*page_size); 1171 } 1172 } 1173 1174 // Interpreter stub for calling a native method. (asm interpreter) 1175 // This sets up a somewhat different looking stack for calling the 1176 // native method than the typical interpreter frame setup. 1177 // 1178 // On entry: 1179 // R19_method - method 1180 // R16_thread - JavaThread* 1181 // R15_esp - intptr_t* sender tos 1182 // 1183 // abstract stack (grows up) 1184 // [ IJava (caller of JNI callee) ] <-- ASP 1185 // ... 1186 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) { 1187 1188 address entry = __ pc(); 1189 1190 const bool inc_counter = UseCompiler || CountCompiledCalls; 1191 1192 // ----------------------------------------------------------------------------- 1193 // Allocate a new frame that represents the native callee (i2n frame). 1194 // This is not a full-blown interpreter frame, but in particular, the 1195 // following registers are valid after this: 1196 // - R19_method 1197 // - R18_local (points to start of arguments to native function) 1198 // 1199 // abstract stack (grows up) 1200 // [ IJava (caller of JNI callee) ] <-- ASP 1201 // ... 1202 1203 const Register signature_handler_fd = R11_scratch1; 1204 const Register pending_exception = R0; 1205 const Register result_handler_addr = R31; 1206 const Register native_method_fd = R11_scratch1; 1207 const Register access_flags = R22_tmp2; 1208 const Register active_handles = R11_scratch1; // R26_monitor saved to state. 1209 const Register sync_state = R12_scratch2; 1210 const Register sync_state_addr = sync_state; // Address is dead after use. 1211 const Register suspend_flags = R11_scratch1; 1212 1213 //============================================================================= 1214 // Allocate new frame and initialize interpreter state. 1215 1216 Label exception_return; 1217 Label exception_return_sync_check; 1218 Label stack_overflow_return; 1219 1220 // Generate new interpreter state and jump to stack_overflow_return in case of 1221 // a stack overflow. 1222 //generate_compute_interpreter_state(stack_overflow_return); 1223 1224 Register size_of_parameters = R22_tmp2; 1225 1226 generate_fixed_frame(true, size_of_parameters, noreg /* unused */); 1227 1228 //============================================================================= 1229 // Increment invocation counter. On overflow, entry to JNI method 1230 // will be compiled. 1231 Label invocation_counter_overflow, continue_after_compile; 1232 if (inc_counter) { 1233 if (synchronized) { 1234 // Since at this point in the method invocation the exception handler 1235 // would try to exit the monitor of synchronized methods which hasn't 1236 // been entered yet, we set the thread local variable 1237 // _do_not_unlock_if_synchronized to true. If any exception was thrown by 1238 // runtime, exception handling i.e. unlock_if_synchronized_method will 1239 // check this thread local flag. 1240 // This flag has two effects, one is to force an unwind in the topmost 1241 // interpreter frame and not perform an unlock while doing so. 1242 __ li(R0, 1); 1243 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1244 } 1245 generate_counter_incr(&invocation_counter_overflow); 1246 1247 BIND(continue_after_compile); 1248 } 1249 1250 bang_stack_shadow_pages(true); 1251 1252 if (inc_counter) { 1253 // Reset the _do_not_unlock_if_synchronized flag. 1254 if (synchronized) { 1255 __ li(R0, 0); 1256 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1257 } 1258 } 1259 1260 // access_flags = method->access_flags(); 1261 // Load access flags. 1262 assert(access_flags->is_nonvolatile(), 1263 "access_flags must be in a non-volatile register"); 1264 // Type check. 1265 assert(4 == sizeof(AccessFlags), "unexpected field size"); 1266 __ lwz(access_flags, method_(access_flags)); 1267 1268 // We don't want to reload R19_method and access_flags after calls 1269 // to some helper functions. 1270 assert(R19_method->is_nonvolatile(), 1271 "R19_method must be a non-volatile register"); 1272 1273 // Check for synchronized methods. Must happen AFTER invocation counter 1274 // check, so method is not locked if counter overflows. 1275 1276 if (synchronized) { 1277 lock_method(access_flags, R11_scratch1, R12_scratch2, true); 1278 1279 // Update monitor in state. 1280 __ ld(R11_scratch1, 0, R1_SP); 1281 __ sub(R12_scratch2, R26_monitor, R11_scratch1); 1282 __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize); 1283 __ std(R12_scratch2, _ijava_state_neg(monitors), R11_scratch1); 1284 } 1285 1286 // jvmti/jvmpi support 1287 __ notify_method_entry(); 1288 1289 //============================================================================= 1290 // Get and call the signature handler. 1291 1292 __ ld(signature_handler_fd, method_(signature_handler)); 1293 Label call_signature_handler; 1294 1295 __ cmpdi(CCR0, signature_handler_fd, 0); 1296 __ bne(CCR0, call_signature_handler); 1297 1298 // Method has never been called. Either generate a specialized 1299 // handler or point to the slow one. 1300 // 1301 // Pass parameter 'false' to avoid exception check in call_VM. 1302 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false); 1303 1304 // Check for an exception while looking up the target method. If we 1305 // incurred one, bail. 1306 __ ld(pending_exception, thread_(pending_exception)); 1307 __ cmpdi(CCR0, pending_exception, 0); 1308 __ bne(CCR0, exception_return_sync_check); // Has pending exception. 1309 1310 // Reload signature handler, it may have been created/assigned in the meanwhile. 1311 __ ld(signature_handler_fd, method_(signature_handler)); 1312 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below). 1313 1314 BIND(call_signature_handler); 1315 1316 // Before we call the signature handler we push a new frame to 1317 // protect the interpreter frame volatile registers when we return 1318 // from jni but before we can get back to Java. 1319 1320 // First set the frame anchor while the SP/FP registers are 1321 // convenient and the slow signature handler can use this same frame 1322 // anchor. 1323 1324 // We have a TOP_IJAVA_FRAME here, which belongs to us. 1325 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/); 1326 1327 // Now the interpreter frame (and its call chain) have been 1328 // invalidated and flushed. We are now protected against eager 1329 // being enabled in native code. Even if it goes eager the 1330 // registers will be reloaded as clean and we will invalidate after 1331 // the call so no spurious flush should be possible. 1332 1333 // Call signature handler and pass locals address. 1334 // 1335 // Our signature handlers copy required arguments to the C stack 1336 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13. 1337 __ mr(R3_ARG1, R18_locals); 1338 #if !defined(ABI_ELFv2) 1339 __ ld(signature_handler_fd, 0, signature_handler_fd); 1340 #endif 1341 1342 __ call_stub(signature_handler_fd); 1343 1344 // Remove the register parameter varargs slots we allocated in 1345 // compute_interpreter_state. SP+16 ends up pointing to the ABI 1346 // outgoing argument area. 1347 // 1348 // Not needed on PPC64. 1349 //__ add(SP, SP, Argument::n_int_register_parameters_c*BytesPerWord); 1350 1351 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register"); 1352 // Save across call to native method. 1353 __ mr(result_handler_addr, R3_RET); 1354 1355 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror. 1356 1357 // Set up fixed parameters and call the native method. 1358 // If the method is static, get mirror into R4_ARG2. 1359 { 1360 Label method_is_not_static; 1361 // Access_flags is non-volatile and still, no need to restore it. 1362 1363 // Restore access flags. 1364 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT); 1365 __ bfalse(CCR0, method_is_not_static); 1366 1367 __ ld(R11_scratch1, _abi0(callers_sp), R1_SP); 1368 // Load mirror from interpreter frame. 1369 __ ld(R12_scratch2, _ijava_state_neg(mirror), R11_scratch1); 1370 // R4_ARG2 = &state->_oop_temp; 1371 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp)); 1372 __ std(R12_scratch2/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); 1373 BIND(method_is_not_static); 1374 } 1375 1376 // At this point, arguments have been copied off the stack into 1377 // their JNI positions. Oops are boxed in-place on the stack, with 1378 // handles copied to arguments. The result handler address is in a 1379 // register. 1380 1381 // Pass JNIEnv address as first parameter. 1382 __ addir(R3_ARG1, thread_(jni_environment)); 1383 1384 // Load the native_method entry before we change the thread state. 1385 __ ld(native_method_fd, method_(native_function)); 1386 1387 //============================================================================= 1388 // Transition from _thread_in_Java to _thread_in_native. As soon as 1389 // we make this change the safepoint code needs to be certain that 1390 // the last Java frame we established is good. The pc in that frame 1391 // just needs to be near here not an actual return address. 1392 1393 // We use release_store_fence to update values like the thread state, where 1394 // we don't want the current thread to continue until all our prior memory 1395 // accesses (including the new thread state) are visible to other threads. 1396 __ li(R0, _thread_in_native); 1397 __ release(); 1398 1399 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size"); 1400 __ stw(R0, thread_(thread_state)); 1401 1402 //============================================================================= 1403 // Call the native method. Argument registers must not have been 1404 // overwritten since "__ call_stub(signature_handler);" (except for 1405 // ARG1 and ARG2 for static methods). 1406 __ call_c(native_method_fd); 1407 1408 __ li(R0, 0); 1409 __ ld(R11_scratch1, 0, R1_SP); 1410 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1); 1411 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1); 1412 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset 1413 1414 // Note: C++ interpreter needs the following here: 1415 // The frame_manager_lr field, which we use for setting the last 1416 // java frame, gets overwritten by the signature handler. Restore 1417 // it now. 1418 //__ get_PC_trash_LR(R11_scratch1); 1419 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 1420 1421 // Because of GC R19_method may no longer be valid. 1422 1423 // Block, if necessary, before resuming in _thread_in_Java state. 1424 // In order for GC to work, don't clear the last_Java_sp until after 1425 // blocking. 1426 1427 //============================================================================= 1428 // Switch thread to "native transition" state before reading the 1429 // synchronization state. This additional state is necessary 1430 // because reading and testing the synchronization state is not 1431 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A, 1432 // in _thread_in_native state, loads _not_synchronized and is 1433 // preempted. VM thread changes sync state to synchronizing and 1434 // suspends threads for GC. Thread A is resumed to finish this 1435 // native method, but doesn't block here since it didn't see any 1436 // synchronization in progress, and escapes. 1437 1438 // We use release_store_fence to update values like the thread state, where 1439 // we don't want the current thread to continue until all our prior memory 1440 // accesses (including the new thread state) are visible to other threads. 1441 __ li(R0/*thread_state*/, _thread_in_native_trans); 1442 __ release(); 1443 __ stw(R0/*thread_state*/, thread_(thread_state)); 1444 if (!UseSystemMemoryBarrier) { 1445 __ fence(); 1446 } 1447 1448 // Now before we return to java we must look for a current safepoint 1449 // (a new safepoint can not start since we entered native_trans). 1450 // We must check here because a current safepoint could be modifying 1451 // the callers registers right this moment. 1452 1453 // Acquire isn't strictly necessary here because of the fence, but 1454 // sync_state is declared to be volatile, so we do it anyway 1455 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path). 1456 1457 Label do_safepoint, sync_check_done; 1458 // No synchronization in progress nor yet synchronized. 1459 __ safepoint_poll(do_safepoint, sync_state, true /* at_return */, false /* in_nmethod */); 1460 1461 // Not suspended. 1462 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size"); 1463 __ lwz(suspend_flags, thread_(suspend_flags)); 1464 __ cmpwi(CCR1, suspend_flags, 0); 1465 __ beq(CCR1, sync_check_done); 1466 1467 __ bind(do_safepoint); 1468 __ isync(); 1469 // Block. We do the call directly and leave the current 1470 // last_Java_frame setup undisturbed. We must save any possible 1471 // native result across the call. No oop is present. 1472 1473 __ mr(R3_ARG1, R16_thread); 1474 #if defined(ABI_ELFv2) 1475 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), 1476 relocInfo::none); 1477 #else 1478 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans), 1479 relocInfo::none); 1480 #endif 1481 1482 __ bind(sync_check_done); 1483 1484 //============================================================================= 1485 // <<<<<< Back in Interpreter Frame >>>>> 1486 1487 // We are in thread_in_native_trans here and back in the normal 1488 // interpreter frame. We don't have to do anything special about 1489 // safepoints and we can switch to Java mode anytime we are ready. 1490 1491 // Note: frame::interpreter_frame_result has a dependency on how the 1492 // method result is saved across the call to post_method_exit. For 1493 // native methods it assumes that the non-FPU/non-void result is 1494 // saved in _native_lresult and a FPU result in _native_fresult. If 1495 // this changes then the interpreter_frame_result implementation 1496 // will need to be updated too. 1497 1498 // On PPC64, we have stored the result directly after the native call. 1499 1500 //============================================================================= 1501 // Back in Java 1502 1503 // We use release_store_fence to update values like the thread state, where 1504 // we don't want the current thread to continue until all our prior memory 1505 // accesses (including the new thread state) are visible to other threads. 1506 __ li(R0/*thread_state*/, _thread_in_Java); 1507 __ lwsync(); // Acquire safepoint and suspend state, release thread state. 1508 __ stw(R0/*thread_state*/, thread_(thread_state)); 1509 1510 if (CheckJNICalls) { 1511 // clear_pending_jni_exception_check 1512 __ load_const_optimized(R0, 0L); 1513 __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread); 1514 } 1515 1516 __ reset_last_Java_frame(); 1517 1518 // Jvmdi/jvmpi support. Whether we've got an exception pending or 1519 // not, and whether unlocking throws an exception or not, we notify 1520 // on native method exit. If we do have an exception, we'll end up 1521 // in the caller's context to handle it, so if we don't do the 1522 // notify here, we'll drop it on the floor. 1523 __ notify_method_exit(true/*native method*/, 1524 ilgl /*illegal state (not used for native methods)*/, 1525 InterpreterMacroAssembler::NotifyJVMTI, 1526 false /*check_exceptions*/); 1527 1528 //============================================================================= 1529 // Handle exceptions 1530 1531 if (synchronized) { 1532 __ unlock_object(R26_monitor); // Can also unlock methods. 1533 } 1534 1535 // Reset active handles after returning from native. 1536 // thread->active_handles()->clear(); 1537 __ ld(active_handles, thread_(active_handles)); 1538 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size"); 1539 __ li(R0, 0); 1540 __ stw(R0, in_bytes(JNIHandleBlock::top_offset()), active_handles); 1541 1542 Label exception_return_sync_check_already_unlocked; 1543 __ ld(R0/*pending_exception*/, thread_(pending_exception)); 1544 __ cmpdi(CCR0, R0/*pending_exception*/, 0); 1545 __ bne(CCR0, exception_return_sync_check_already_unlocked); 1546 1547 //----------------------------------------------------------------------------- 1548 // No exception pending. 1549 1550 // Move native method result back into proper registers and return. 1551 // Invoke result handler (may unbox/promote). 1552 __ ld(R11_scratch1, 0, R1_SP); 1553 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1); 1554 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1); 1555 __ call_stub(result_handler_addr); 1556 1557 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1558 1559 // Must use the return pc which was loaded from the caller's frame 1560 // as the VM uses return-pc-patching for deoptimization. 1561 __ mtlr(R0); 1562 __ blr(); 1563 1564 //----------------------------------------------------------------------------- 1565 // An exception is pending. We call into the runtime only if the 1566 // caller was not interpreted. If it was interpreted the 1567 // interpreter will do the correct thing. If it isn't interpreted 1568 // (call stub/compiled code) we will change our return and continue. 1569 1570 BIND(exception_return_sync_check); 1571 1572 if (synchronized) { 1573 __ unlock_object(R26_monitor); // Can also unlock methods. 1574 } 1575 BIND(exception_return_sync_check_already_unlocked); 1576 1577 const Register return_pc = R31; 1578 1579 __ ld(return_pc, 0, R1_SP); 1580 __ ld(return_pc, _abi0(lr), return_pc); 1581 1582 // Get the address of the exception handler. 1583 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), 1584 R16_thread, 1585 return_pc /* return pc */); 1586 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2); 1587 1588 // Load the PC of the exception handler into LR. 1589 __ mtlr(R3_RET); 1590 1591 // Load exception into R3_ARG1 and clear pending exception in thread. 1592 __ ld(R3_ARG1/*exception*/, thread_(pending_exception)); 1593 __ li(R4_ARG2, 0); 1594 __ std(R4_ARG2, thread_(pending_exception)); 1595 1596 // Load the original return pc into R4_ARG2. 1597 __ mr(R4_ARG2/*issuing_pc*/, return_pc); 1598 1599 // Return to exception handler. 1600 __ blr(); 1601 1602 //============================================================================= 1603 // Counter overflow. 1604 1605 if (inc_counter) { 1606 // Handle invocation counter overflow. 1607 __ bind(invocation_counter_overflow); 1608 1609 generate_counter_overflow(continue_after_compile); 1610 } 1611 1612 return entry; 1613 } 1614 1615 // Generic interpreted method entry to (asm) interpreter. 1616 // 1617 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) { 1618 bool inc_counter = UseCompiler || CountCompiledCalls; 1619 address entry = __ pc(); 1620 // Generate the code to allocate the interpreter stack frame. 1621 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame. 1622 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame. 1623 1624 // Does also a stack check to assure this frame fits on the stack. 1625 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals); 1626 1627 // -------------------------------------------------------------------------- 1628 // Zero out non-parameter locals. 1629 // Note: *Always* zero out non-parameter locals as Sparc does. It's not 1630 // worth to ask the flag, just do it. 1631 Register Rslot_addr = R6_ARG4, 1632 Rnum = R7_ARG5; 1633 Label Lno_locals, Lzero_loop; 1634 1635 // Set up the zeroing loop. 1636 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals); 1637 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals); 1638 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize); 1639 __ beq(CCR0, Lno_locals); 1640 __ li(R0, 0); 1641 __ mtctr(Rnum); 1642 1643 // The zero locals loop. 1644 __ bind(Lzero_loop); 1645 __ std(R0, 0, Rslot_addr); 1646 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize); 1647 __ bdnz(Lzero_loop); 1648 1649 __ bind(Lno_locals); 1650 1651 // -------------------------------------------------------------------------- 1652 // Counter increment and overflow check. 1653 Label invocation_counter_overflow; 1654 Label continue_after_compile; 1655 if (inc_counter || ProfileInterpreter) { 1656 1657 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1; 1658 if (synchronized) { 1659 // Since at this point in the method invocation the exception handler 1660 // would try to exit the monitor of synchronized methods which hasn't 1661 // been entered yet, we set the thread local variable 1662 // _do_not_unlock_if_synchronized to true. If any exception was thrown by 1663 // runtime, exception handling i.e. unlock_if_synchronized_method will 1664 // check this thread local flag. 1665 // This flag has two effects, one is to force an unwind in the topmost 1666 // interpreter frame and not perform an unlock while doing so. 1667 __ li(R0, 1); 1668 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1669 } 1670 1671 // Argument and return type profiling. 1672 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4); 1673 1674 // Increment invocation counter and check for overflow. 1675 if (inc_counter) { 1676 generate_counter_incr(&invocation_counter_overflow); 1677 } 1678 1679 __ bind(continue_after_compile); 1680 } 1681 1682 bang_stack_shadow_pages(false); 1683 1684 if (inc_counter || ProfileInterpreter) { 1685 // Reset the _do_not_unlock_if_synchronized flag. 1686 if (synchronized) { 1687 __ li(R0, 0); 1688 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1689 } 1690 } 1691 1692 // -------------------------------------------------------------------------- 1693 // Locking of synchronized methods. Must happen AFTER invocation_counter 1694 // check and stack overflow check, so method is not locked if overflows. 1695 if (synchronized) { 1696 lock_method(R3_ARG1, R4_ARG2, R5_ARG3); 1697 } 1698 #ifdef ASSERT 1699 else { 1700 Label Lok; 1701 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method); 1702 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED); 1703 __ asm_assert_eq("method needs synchronization"); 1704 __ bind(Lok); 1705 } 1706 #endif // ASSERT 1707 1708 // -------------------------------------------------------------------------- 1709 // JVMTI support 1710 __ notify_method_entry(); 1711 1712 // -------------------------------------------------------------------------- 1713 // Start executing instructions. 1714 __ dispatch_next(vtos); 1715 1716 // -------------------------------------------------------------------------- 1717 if (inc_counter) { 1718 // Handle invocation counter overflow. 1719 __ bind(invocation_counter_overflow); 1720 generate_counter_overflow(continue_after_compile); 1721 } 1722 return entry; 1723 } 1724 1725 // CRC32 Intrinsics. 1726 // 1727 // Contract on scratch and work registers. 1728 // ======================================= 1729 // 1730 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers. 1731 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set. 1732 // You can't rely on these registers across calls. 1733 // 1734 // The generators for CRC32_update and for CRC32_updateBytes use the 1735 // scratch/work register set internally, passing the work registers 1736 // as arguments to the MacroAssembler emitters as required. 1737 // 1738 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments. 1739 // Their contents is not constant but may change according to the requirements 1740 // of the emitted code. 1741 // 1742 // All other registers from the scratch/work register set are used "internally" 1743 // and contain garbage (i.e. unpredictable values) once blr() is reached. 1744 // Basically, only R3_RET contains a defined value which is the function result. 1745 // 1746 /** 1747 * Method entry for static native methods: 1748 * int java.util.zip.CRC32.update(int crc, int b) 1749 */ 1750 address TemplateInterpreterGenerator::generate_CRC32_update_entry() { 1751 assert(UseCRC32Intrinsics, "this intrinsic is not supported"); 1752 address start = __ pc(); // Remember stub start address (is rtn value). 1753 Label slow_path; 1754 1755 // Safepoint check 1756 const Register sync_state = R11_scratch1; 1757 __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */); 1758 1759 // We don't generate local frame and don't align stack because 1760 // we not even call stub code (we generate the code inline) 1761 // and there is no safepoint on this path. 1762 1763 // Load java parameters. 1764 // R15_esp is callers operand stack pointer, i.e. it points to the parameters. 1765 const Register argP = R15_esp; 1766 const Register crc = R3_ARG1; // crc value 1767 const Register data = R4_ARG2; 1768 const Register table = R5_ARG3; // address of crc32 table 1769 1770 BLOCK_COMMENT("CRC32_update {"); 1771 1772 // Arguments are reversed on java expression stack 1773 #ifdef VM_LITTLE_ENDIAN 1774 int data_offs = 0+1*wordSize; // (stack) address of byte value. Emitter expects address, not value. 1775 // Being passed as an int, the single byte is at offset +0. 1776 #else 1777 int data_offs = 3+1*wordSize; // (stack) address of byte value. Emitter expects address, not value. 1778 // Being passed from java as an int, the single byte is at offset +3. 1779 #endif 1780 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register. 1781 __ lbz(data, data_offs, argP); // Byte from buffer, zero-extended. 1782 __ load_const_optimized(table, StubRoutines::crc_table_addr(), R0); 1783 __ kernel_crc32_singleByteReg(crc, data, table, true); 1784 1785 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted). 1786 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0); 1787 __ blr(); 1788 1789 // Generate a vanilla native entry as the slow path. 1790 BLOCK_COMMENT("} CRC32_update"); 1791 BIND(slow_path); 1792 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1); 1793 return start; 1794 } 1795 1796 /** 1797 * Method entry for static native methods: 1798 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len) 1799 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len) 1800 */ 1801 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 1802 assert(UseCRC32Intrinsics, "this intrinsic is not supported"); 1803 address start = __ pc(); // Remember stub start address (is rtn value). 1804 Label slow_path; 1805 1806 // Safepoint check 1807 const Register sync_state = R11_scratch1; 1808 __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */); 1809 1810 // We don't generate local frame and don't align stack because 1811 // we not even call stub code (we generate the code inline) 1812 // and there is no safepoint on this path. 1813 1814 // Load parameters. 1815 // Z_esp is callers operand stack pointer, i.e. it points to the parameters. 1816 const Register argP = R15_esp; 1817 const Register crc = R3_ARG1; // crc value 1818 const Register data = R4_ARG2; // address of java byte array 1819 const Register dataLen = R5_ARG3; // source data len 1820 const Register tmp = R11_scratch1; 1821 1822 // Arguments are reversed on java expression stack. 1823 // Calculate address of start element. 1824 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct". 1825 BLOCK_COMMENT("CRC32_updateByteBuffer {"); 1826 // crc @ (SP + 5W) (32bit) 1827 // buf @ (SP + 3W) (64bit ptr to long array) 1828 // off @ (SP + 2W) (32bit) 1829 // dataLen @ (SP + 1W) (32bit) 1830 // data = buf + off 1831 __ ld( data, 3*wordSize, argP); // start of byte buffer 1832 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 1833 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 1834 __ lwz( crc, 5*wordSize, argP); // current crc state 1835 __ add( data, data, tmp); // Add byte buffer offset. 1836 } else { // Used for "updateBytes update". 1837 BLOCK_COMMENT("CRC32_updateBytes {"); 1838 // crc @ (SP + 4W) (32bit) 1839 // buf @ (SP + 3W) (64bit ptr to byte array) 1840 // off @ (SP + 2W) (32bit) 1841 // dataLen @ (SP + 1W) (32bit) 1842 // data = buf + off + base_offset 1843 __ ld( data, 3*wordSize, argP); // start of byte buffer 1844 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 1845 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 1846 __ add( data, data, tmp); // add byte buffer offset 1847 __ lwz( crc, 4*wordSize, argP); // current crc state 1848 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1849 } 1850 1851 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, false); 1852 1853 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted). 1854 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0); 1855 __ blr(); 1856 1857 // Generate a vanilla native entry as the slow path. 1858 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)"); 1859 BIND(slow_path); 1860 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1); 1861 return start; 1862 } 1863 1864 1865 /** 1866 * Method entry for intrinsic-candidate (non-native) methods: 1867 * int java.util.zip.CRC32C.updateBytes( int crc, byte[] b, int off, int end) 1868 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end) 1869 * Unlike CRC32, CRC32C does not have any methods marked as native 1870 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses 1871 **/ 1872 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 1873 assert(UseCRC32CIntrinsics, "this intrinsic is not supported"); 1874 address start = __ pc(); // Remember stub start address (is rtn value). 1875 1876 // We don't generate local frame and don't align stack because 1877 // we not even call stub code (we generate the code inline) 1878 // and there is no safepoint on this path. 1879 1880 // Load parameters. 1881 // Z_esp is callers operand stack pointer, i.e. it points to the parameters. 1882 const Register argP = R15_esp; 1883 const Register crc = R3_ARG1; // crc value 1884 const Register data = R4_ARG2; // address of java byte array 1885 const Register dataLen = R5_ARG3; // source data len 1886 const Register tmp = R11_scratch1; 1887 1888 // Arguments are reversed on java expression stack. 1889 // Calculate address of start element. 1890 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer". 1891 BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {"); 1892 // crc @ (SP + 5W) (32bit) 1893 // buf @ (SP + 3W) (64bit ptr to long array) 1894 // off @ (SP + 2W) (32bit) 1895 // dataLen @ (SP + 1W) (32bit) 1896 // data = buf + off 1897 __ ld( data, 3*wordSize, argP); // start of byte buffer 1898 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 1899 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 1900 __ lwz( crc, 5*wordSize, argP); // current crc state 1901 __ add( data, data, tmp); // Add byte buffer offset. 1902 __ sub( dataLen, dataLen, tmp); // (end_index - offset) 1903 } else { // Used for "updateBytes update". 1904 BLOCK_COMMENT("CRC32C_updateBytes {"); 1905 // crc @ (SP + 4W) (32bit) 1906 // buf @ (SP + 3W) (64bit ptr to byte array) 1907 // off @ (SP + 2W) (32bit) 1908 // dataLen @ (SP + 1W) (32bit) 1909 // data = buf + off + base_offset 1910 __ ld( data, 3*wordSize, argP); // start of byte buffer 1911 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset 1912 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process 1913 __ add( data, data, tmp); // add byte buffer offset 1914 __ sub( dataLen, dataLen, tmp); // (end_index - offset) 1915 __ lwz( crc, 4*wordSize, argP); // current crc state 1916 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1917 } 1918 1919 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, true); 1920 1921 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted). 1922 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0); 1923 __ blr(); 1924 1925 BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}"); 1926 return start; 1927 } 1928 1929 // Not supported 1930 address TemplateInterpreterGenerator::generate_currentThread() { return nullptr; } 1931 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; } 1932 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; } 1933 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; } 1934 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; } 1935 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() { return nullptr; } 1936 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() { return nullptr; } 1937 1938 // ============================================================================= 1939 // Exceptions 1940 1941 void TemplateInterpreterGenerator::generate_throw_exception() { 1942 Register Rexception = R17_tos, 1943 Rcontinuation = R3_RET; 1944 1945 // -------------------------------------------------------------------------- 1946 // Entry point if an method returns with a pending exception (rethrow). 1947 Interpreter::_rethrow_exception_entry = __ pc(); 1948 { 1949 __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/); 1950 1951 // Compiled code destroys templateTableBase, reload. 1952 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1); 1953 } 1954 1955 // Entry point if a interpreted method throws an exception (throw). 1956 Interpreter::_throw_exception_entry = __ pc(); 1957 { 1958 __ mr(Rexception, R3_RET); 1959 1960 __ verify_oop(Rexception); 1961 1962 // Expression stack must be empty before entering the VM in case of an exception. 1963 __ empty_expression_stack(); 1964 // Find exception handler address and preserve exception oop. 1965 // Call C routine to find handler and jump to it. 1966 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception); 1967 __ mtctr(Rcontinuation); 1968 // Push exception for exception handler bytecodes. 1969 __ push_ptr(Rexception); 1970 1971 // Jump to exception handler (may be remove activation entry!). 1972 __ bctr(); 1973 } 1974 1975 // If the exception is not handled in the current frame the frame is 1976 // removed and the exception is rethrown (i.e. exception 1977 // continuation is _rethrow_exception). 1978 // 1979 // Note: At this point the bci is still the bxi for the instruction 1980 // which caused the exception and the expression stack is 1981 // empty. Thus, for any VM calls at this point, GC will find a legal 1982 // oop map (with empty expression stack). 1983 1984 // In current activation 1985 // tos: exception 1986 // bcp: exception bcp 1987 1988 // -------------------------------------------------------------------------- 1989 // JVMTI PopFrame support 1990 1991 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 1992 { 1993 // Set the popframe_processing bit in popframe_condition indicating that we are 1994 // currently handling popframe, so that call_VMs that may happen later do not 1995 // trigger new popframe handling cycles. 1996 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 1997 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit); 1998 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 1999 2000 // Empty the expression stack, as in normal exception handling. 2001 __ empty_expression_stack(); 2002 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false); 2003 2004 // Check to see whether we are returning to a deoptimized frame. 2005 // (The PopFrame call ensures that the caller of the popped frame is 2006 // either interpreted or compiled and deoptimizes it if compiled.) 2007 // Note that we don't compare the return PC against the 2008 // deoptimization blob's unpack entry because of the presence of 2009 // adapter frames in C2. 2010 Label Lcaller_not_deoptimized; 2011 Register return_pc = R3_ARG1; 2012 __ ld(return_pc, 0, R1_SP); 2013 __ ld(return_pc, _abi0(lr), return_pc); 2014 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc); 2015 __ cmpdi(CCR0, R3_RET, 0); 2016 __ bne(CCR0, Lcaller_not_deoptimized); 2017 2018 // The deoptimized case. 2019 // In this case, we can't call dispatch_next() after the frame is 2020 // popped, but instead must save the incoming arguments and restore 2021 // them after deoptimization has occurred. 2022 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method); 2023 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2); 2024 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize); 2025 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize); 2026 __ subf(R5_ARG3, R4_ARG2, R5_ARG3); 2027 // Save these arguments. 2028 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3); 2029 2030 // Inform deoptimization that it is responsible for restoring these arguments. 2031 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit); 2032 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 2033 2034 // Return from the current method into the deoptimization blob. Will eventually 2035 // end up in the deopt interpreter entry, deoptimization prepared everything that 2036 // we will reexecute the call that called us. 2037 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2); 2038 __ mtlr(return_pc); 2039 __ pop_cont_fastpath(); 2040 __ blr(); 2041 2042 // The non-deoptimized case. 2043 __ bind(Lcaller_not_deoptimized); 2044 2045 // Clear the popframe condition flag. 2046 __ li(R0, 0); 2047 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 2048 2049 // Get out of the current method and re-execute the call that called us. 2050 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2); 2051 __ pop_cont_fastpath(); 2052 __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/); 2053 if (ProfileInterpreter) { 2054 __ set_method_data_pointer_for_bcp(); 2055 __ ld(R11_scratch1, 0, R1_SP); 2056 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1); 2057 } 2058 #if INCLUDE_JVMTI 2059 Label L_done; 2060 2061 __ lbz(R11_scratch1, 0, R14_bcp); 2062 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic); 2063 __ bne(CCR0, L_done); 2064 2065 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. 2066 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null. 2067 __ ld(R4_ARG2, 0, R18_locals); 2068 __ call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp); 2069 2070 __ cmpdi(CCR0, R4_ARG2, 0); 2071 __ beq(CCR0, L_done); 2072 __ std(R4_ARG2, wordSize, R15_esp); 2073 __ bind(L_done); 2074 #endif // INCLUDE_JVMTI 2075 __ dispatch_next(vtos); 2076 } 2077 // end of JVMTI PopFrame support 2078 2079 // -------------------------------------------------------------------------- 2080 // Remove activation exception entry. 2081 // This is jumped to if an interpreted method can't handle an exception itself 2082 // (we come from the throw/rethrow exception entry above). We're going to call 2083 // into the VM to find the exception handler in the caller, pop the current 2084 // frame and return the handler we calculated. 2085 Interpreter::_remove_activation_entry = __ pc(); 2086 { 2087 __ pop_ptr(Rexception); 2088 __ verify_oop(Rexception); 2089 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread); 2090 2091 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true); 2092 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false); 2093 2094 __ get_vm_result(Rexception); 2095 2096 // We are done with this activation frame; find out where to go next. 2097 // The continuation point will be an exception handler, which expects 2098 // the following registers set up: 2099 // 2100 // RET: exception oop 2101 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled. 2102 2103 Register return_pc = R31; // Needs to survive the runtime call. 2104 __ ld(return_pc, 0, R1_SP); 2105 __ ld(return_pc, _abi0(lr), return_pc); 2106 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc); 2107 2108 // Remove the current activation. 2109 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2); 2110 __ pop_cont_fastpath(); 2111 2112 __ mr(R4_ARG2, return_pc); 2113 __ mtlr(R3_RET); 2114 __ mr(R3_RET, Rexception); 2115 __ blr(); 2116 } 2117 } 2118 2119 // JVMTI ForceEarlyReturn support. 2120 // Returns "in the middle" of a method with a "fake" return value. 2121 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 2122 2123 Register Rscratch1 = R11_scratch1, 2124 Rscratch2 = R12_scratch2; 2125 2126 address entry = __ pc(); 2127 __ empty_expression_stack(); 2128 2129 __ load_earlyret_value(state, Rscratch1); 2130 2131 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 2132 // Clear the earlyret state. 2133 __ li(R0, 0); 2134 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1); 2135 2136 __ remove_activation(state, false, false); 2137 // Copied from TemplateTable::_return. 2138 // Restoration of lr done by remove_activation. 2139 switch (state) { 2140 // Narrow result if state is itos but result type is smaller. 2141 case btos: 2142 case ztos: 2143 case ctos: 2144 case stos: 2145 case itos: __ narrow(R17_tos); /* fall through */ 2146 case ltos: 2147 case atos: __ mr(R3_RET, R17_tos); break; 2148 case ftos: 2149 case dtos: __ fmr(F1_RET, F15_ftos); break; 2150 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need 2151 // to get visible before the reference to the object gets stored anywhere. 2152 __ membar(Assembler::StoreStore); break; 2153 default : ShouldNotReachHere(); 2154 } 2155 __ blr(); 2156 2157 return entry; 2158 } // end of ForceEarlyReturn support 2159 2160 //----------------------------------------------------------------------------- 2161 // Helper for vtos entry point generation 2162 2163 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, 2164 address& bep, 2165 address& cep, 2166 address& sep, 2167 address& aep, 2168 address& iep, 2169 address& lep, 2170 address& fep, 2171 address& dep, 2172 address& vep) { 2173 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 2174 Label L; 2175 2176 aep = __ pc(); __ push_ptr(); __ b(L); 2177 fep = __ pc(); __ push_f(); __ b(L); 2178 dep = __ pc(); __ push_d(); __ b(L); 2179 lep = __ pc(); __ push_l(); __ b(L); 2180 __ align(32, 12, 24); // align L 2181 bep = cep = sep = 2182 iep = __ pc(); __ push_i(); 2183 vep = __ pc(); 2184 __ bind(L); 2185 generate_and_dispatch(t); 2186 } 2187 2188 //----------------------------------------------------------------------------- 2189 2190 // Non-product code 2191 #ifndef PRODUCT 2192 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 2193 //__ flush_bundle(); 2194 address entry = __ pc(); 2195 2196 const char *bname = nullptr; 2197 uint tsize = 0; 2198 switch(state) { 2199 case ftos: 2200 bname = "trace_code_ftos {"; 2201 tsize = 2; 2202 break; 2203 case btos: 2204 bname = "trace_code_btos {"; 2205 tsize = 2; 2206 break; 2207 case ztos: 2208 bname = "trace_code_ztos {"; 2209 tsize = 2; 2210 break; 2211 case ctos: 2212 bname = "trace_code_ctos {"; 2213 tsize = 2; 2214 break; 2215 case stos: 2216 bname = "trace_code_stos {"; 2217 tsize = 2; 2218 break; 2219 case itos: 2220 bname = "trace_code_itos {"; 2221 tsize = 2; 2222 break; 2223 case ltos: 2224 bname = "trace_code_ltos {"; 2225 tsize = 3; 2226 break; 2227 case atos: 2228 bname = "trace_code_atos {"; 2229 tsize = 2; 2230 break; 2231 case vtos: 2232 // Note: In case of vtos, the topmost of stack value could be a int or doubl 2233 // In case of a double (2 slots) we won't see the 2nd stack value. 2234 // Maybe we simply should print the topmost 3 stack slots to cope with the problem. 2235 bname = "trace_code_vtos {"; 2236 tsize = 2; 2237 2238 break; 2239 case dtos: 2240 bname = "trace_code_dtos {"; 2241 tsize = 3; 2242 break; 2243 default: 2244 ShouldNotReachHere(); 2245 } 2246 BLOCK_COMMENT(bname); 2247 2248 // Support short-cut for TraceBytecodesAt. 2249 // Don't call into the VM if we don't want to trace to speed up things. 2250 Label Lskip_vm_call; 2251 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) { 2252 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true); 2253 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true); 2254 __ ld(R11_scratch1, offs1, R11_scratch1); 2255 __ lwa(R12_scratch2, offs2, R12_scratch2); 2256 __ cmpd(CCR0, R12_scratch2, R11_scratch1); 2257 __ blt(CCR0, Lskip_vm_call); 2258 } 2259 2260 __ push(state); 2261 // Load 2 topmost expression stack values. 2262 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp); 2263 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp); 2264 __ mflr(R31); 2265 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false); 2266 __ mtlr(R31); 2267 __ pop(state); 2268 2269 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) { 2270 __ bind(Lskip_vm_call); 2271 } 2272 __ blr(); 2273 BLOCK_COMMENT("} trace_code"); 2274 return entry; 2275 } 2276 2277 void TemplateInterpreterGenerator::count_bytecode() { 2278 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true); 2279 __ lwz(R12_scratch2, offs, R11_scratch1); 2280 __ addi(R12_scratch2, R12_scratch2, 1); 2281 __ stw(R12_scratch2, offs, R11_scratch1); 2282 } 2283 2284 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { 2285 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true); 2286 __ lwz(R12_scratch2, offs, R11_scratch1); 2287 __ addi(R12_scratch2, R12_scratch2, 1); 2288 __ stw(R12_scratch2, offs, R11_scratch1); 2289 } 2290 2291 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { 2292 const Register addr = R11_scratch1, 2293 tmp = R12_scratch2; 2294 // Get index, shift out old bytecode, bring in new bytecode, and store it. 2295 // _index = (_index >> log2_number_of_codes) | 2296 // (bytecode << log2_number_of_codes); 2297 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true); 2298 __ lwz(tmp, offs1, addr); 2299 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes); 2300 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes); 2301 __ stw(tmp, offs1, addr); 2302 2303 // Bump bucket contents. 2304 // _counters[_index] ++; 2305 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true); 2306 __ sldi(tmp, tmp, LogBytesPerInt); 2307 __ add(addr, tmp, addr); 2308 __ lwz(tmp, offs2, addr); 2309 __ addi(tmp, tmp, 1); 2310 __ stw(tmp, offs2, addr); 2311 } 2312 2313 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 2314 // Call a little run-time stub to avoid blow-up for each bytecode. 2315 // The run-time runtime saves the right registers, depending on 2316 // the tosca in-state for the given template. 2317 2318 assert(Interpreter::trace_code(t->tos_in()) != nullptr, 2319 "entry must have been generated"); 2320 2321 // Note: we destroy LR here. 2322 __ bl(Interpreter::trace_code(t->tos_in())); 2323 } 2324 2325 void TemplateInterpreterGenerator::stop_interpreter_at() { 2326 Label L; 2327 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true); 2328 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true); 2329 __ ld(R11_scratch1, offs1, R11_scratch1); 2330 __ lwa(R12_scratch2, offs2, R12_scratch2); 2331 __ cmpd(CCR0, R12_scratch2, R11_scratch1); 2332 __ bne(CCR0, L); 2333 __ illtrap(); 2334 __ bind(L); 2335 } 2336 2337 #endif // !PRODUCT