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