1 /* 2 * Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "asm/macroAssembler.inline.hpp" 27 #include "c1/c1_LIRAssembler.hpp" 28 #include "c1/c1_MacroAssembler.hpp" 29 #include "gc/g1/c1G1BSCodeGen.hpp" 30 #include "gc/g1/g1CardTable.hpp" 31 #include "gc/g1/g1BarrierSet.hpp" 32 #include "gc/g1/g1BSCodeGen.hpp" 33 #include "gc/g1/heapRegion.hpp" 34 #include "gc/shared/collectedHeap.hpp" 35 #include "runtime/thread.hpp" 36 #include "interpreter/interp_masm.hpp" 37 38 #define __ masm-> 39 40 void G1BSCodeGen::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators, Register addr, Register count) { 41 bool dest_uninitialized = (decorators & DEST_NOT_INITIALIZED) != 0; 42 43 if (!dest_uninitialized) { 44 __ pusha(); // push registers 45 if (count == c_rarg0) { 46 if (addr == c_rarg1) { 47 // exactly backwards!! 48 __ xchgptr(c_rarg1, c_rarg0); 49 } else { 50 __ movptr(c_rarg1, count); 51 __ movptr(c_rarg0, addr); 52 } 53 } else { 54 __ movptr(c_rarg0, addr); 55 __ movptr(c_rarg1, count); 56 } 57 __ call_VM_leaf(CAST_FROM_FN_PTR(address, ModRefBarrierSet::static_write_ref_array_pre), 2); 58 __ popa(); 59 } 60 } 61 62 void G1BSCodeGen::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type, 63 Register dst, Address src, Register tmp1, Register tmp2) { 64 bool on_oop = type == T_OBJECT || type == T_ARRAY; 65 bool on_weak = (decorators & GC_ACCESS_ON_WEAK) != 0; 66 bool on_phantom = (decorators & GC_ACCESS_ON_PHANTOM) != 0; 67 bool on_reference = on_weak || on_phantom; 68 ModRefBSCodeGen::load_at(masm, decorators, type, dst, src, tmp1, tmp2); 69 if (on_oop && on_reference) { 70 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread); 71 NOT_LP64(__ get_thread(thread)); 72 73 // Generate the G1 pre-barrier code to log the value of 74 // the referent field or jniHandle in an SATB buffer. 75 g1_write_barrier_pre(masm /* masm */, 76 noreg /* obj */, 77 dst /* pre_val */, 78 thread /* thread */, 79 tmp1 /* tmp */, 80 true /* tosca_live */, 81 true /* expand_call */); 82 } 83 } 84 85 void G1BSCodeGen::g1_write_barrier_pre(MacroAssembler* masm, 86 Register obj, 87 Register pre_val, 88 Register thread, 89 Register tmp, 90 bool tosca_live, 91 bool expand_call) { 92 // If expand_call is true then we expand the call_VM_leaf macro 93 // directly to skip generating the check by 94 // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp. 95 96 #ifdef _LP64 97 assert(thread == r15_thread, "must be"); 98 #endif // _LP64 99 100 Label done; 101 Label runtime; 102 103 assert(pre_val != noreg, "check this code"); 104 105 if (obj != noreg) { 106 assert_different_registers(obj, pre_val, tmp); 107 assert(pre_val != rax, "check this code"); 108 } 109 110 Address in_progress(thread, in_bytes(JavaThread::satb_mark_queue_offset() + 111 SATBMarkQueue::byte_offset_of_active())); 112 Address index(thread, in_bytes(JavaThread::satb_mark_queue_offset() + 113 SATBMarkQueue::byte_offset_of_index())); 114 Address buffer(thread, in_bytes(JavaThread::satb_mark_queue_offset() + 115 SATBMarkQueue::byte_offset_of_buf())); 116 117 118 // Is marking active? 119 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) { 120 __ cmpl(in_progress, 0); 121 } else { 122 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption"); 123 __ cmpb(in_progress, 0); 124 } 125 __ jcc(Assembler::equal, done); 126 127 // Do we need to load the previous value? 128 if (obj != noreg) { 129 __ load_heap_oop(pre_val, Address(obj, 0)); 130 } 131 132 // Is the previous value null? 133 __ cmpptr(pre_val, (int32_t) NULL_WORD); 134 __ jcc(Assembler::equal, done); 135 136 // Can we store original value in the thread's buffer? 137 // Is index == 0? 138 // (The index field is typed as size_t.) 139 140 __ movptr(tmp, index); // tmp := *index_adr 141 __ cmpptr(tmp, 0); // tmp == 0? 142 __ jcc(Assembler::equal, runtime); // If yes, goto runtime 143 144 __ subptr(tmp, wordSize); // tmp := tmp - wordSize 145 __ movptr(index, tmp); // *index_adr := tmp 146 __ addptr(tmp, buffer); // tmp := tmp + *buffer_adr 147 148 // Record the previous value 149 __ movptr(Address(tmp, 0), pre_val); 150 __ jmp(done); 151 152 __ bind(runtime); 153 // save the live input values 154 if(tosca_live) __ push(rax); 155 156 if (obj != noreg && obj != rax) 157 __ push(obj); 158 159 if (pre_val != rax) 160 __ push(pre_val); 161 162 // Calling the runtime using the regular call_VM_leaf mechanism generates 163 // code (generated by InterpreterMacroAssember::call_VM_leaf_base) 164 // that checks that the *(ebp+frame::interpreter_frame_last_sp) == NULL. 165 // 166 // If we care generating the pre-barrier without a frame (e.g. in the 167 // intrinsified Reference.get() routine) then ebp might be pointing to 168 // the caller frame and so this check will most likely fail at runtime. 169 // 170 // Expanding the call directly bypasses the generation of the check. 171 // So when we do not have have a full interpreter frame on the stack 172 // expand_call should be passed true. 173 174 NOT_LP64( __ push(thread); ) 175 176 if (expand_call) { 177 LP64_ONLY( assert(pre_val != c_rarg1, "smashed arg"); ) 178 if (c_rarg1 != thread) { 179 __ mov(c_rarg1, thread); 180 } 181 if (c_rarg0 != pre_val) { 182 __ mov(c_rarg0, pre_val); 183 } 184 __ MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, G1BarrierSet::g1_wb_pre), 2); 185 } else { 186 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSet::g1_wb_pre), pre_val, thread); 187 } 188 189 NOT_LP64( __ pop(thread); ) 190 191 // save the live input values 192 if (pre_val != rax) 193 __ pop(pre_val); 194 195 if (obj != noreg && obj != rax) 196 __ pop(obj); 197 198 if(tosca_live) __ pop(rax); 199 200 __ bind(done); 201 } 202 203 void G1BSCodeGen::g1_write_barrier_post(MacroAssembler* masm, 204 Register store_addr, 205 Register new_val, 206 Register thread, 207 Register tmp, 208 Register tmp2) { 209 #ifdef _LP64 210 assert(thread == r15_thread, "must be"); 211 #endif // _LP64 212 213 Address queue_index(thread, in_bytes(JavaThread::dirty_card_queue_offset() + 214 DirtyCardQueue::byte_offset_of_index())); 215 Address buffer(thread, in_bytes(JavaThread::dirty_card_queue_offset() + 216 DirtyCardQueue::byte_offset_of_buf())); 217 218 CardTableModRefBS* ct = 219 barrier_set_cast<CardTableModRefBS>(Universe::heap()->barrier_set()); 220 assert(sizeof(*ct->card_table()->byte_map_base()) == sizeof(jbyte), "adjust this code"); 221 222 Label done; 223 Label runtime; 224 225 // Does store cross heap regions? 226 227 __ movptr(tmp, store_addr); 228 __ xorptr(tmp, new_val); 229 __ shrptr(tmp, HeapRegion::LogOfHRGrainBytes); 230 __ jcc(Assembler::equal, done); 231 232 // crosses regions, storing NULL? 233 234 __ cmpptr(new_val, (int32_t) NULL_WORD); 235 __ jcc(Assembler::equal, done); 236 237 // storing region crossing non-NULL, is card already dirty? 238 239 const Register card_addr = tmp; 240 const Register cardtable = tmp2; 241 242 __ movptr(card_addr, store_addr); 243 __ shrptr(card_addr, CardTable::card_shift); 244 // Do not use ExternalAddress to load 'byte_map_base', since 'byte_map_base' is NOT 245 // a valid address and therefore is not properly handled by the relocation code. 246 __ movptr(cardtable, (intptr_t)ct->card_table()->byte_map_base()); 247 __ addptr(card_addr, cardtable); 248 249 __ cmpb(Address(card_addr, 0), (int)G1CardTable::g1_young_card_val()); 250 __ jcc(Assembler::equal, done); 251 252 __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad)); 253 __ cmpb(Address(card_addr, 0), (int)G1CardTable::dirty_card_val()); 254 __ jcc(Assembler::equal, done); 255 256 257 // storing a region crossing, non-NULL oop, card is clean. 258 // dirty card and log. 259 260 __ movb(Address(card_addr, 0), (int)G1CardTable::dirty_card_val()); 261 262 __ cmpl(queue_index, 0); 263 __ jcc(Assembler::equal, runtime); 264 __ subl(queue_index, wordSize); 265 __ movptr(tmp2, buffer); 266 #ifdef _LP64 267 __ movslq(rscratch1, queue_index); 268 __ addq(tmp2, rscratch1); 269 __ movq(Address(tmp2, 0), card_addr); 270 #else 271 __ addl(tmp2, queue_index); 272 __ movl(Address(tmp2, 0), card_addr); 273 #endif 274 __ jmp(done); 275 276 __ bind(runtime); 277 // save the live input values 278 __ push(store_addr); 279 __ push(new_val); 280 #ifdef _LP64 281 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSet::g1_wb_post), card_addr, r15_thread); 282 #else 283 __ push(thread); 284 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSet::g1_wb_post), card_addr, thread); 285 __ pop(thread); 286 #endif 287 __ pop(new_val); 288 __ pop(store_addr); 289 290 __ bind(done); 291 } 292 293 void G1BSCodeGen::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type, 294 Address dst, Register val, Register tmp1, Register tmp2) { 295 Register tmp3 = LP64_ONLY(r8) NOT_LP64(rsi); 296 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 297 // flatten object address if needed 298 // We do it regardless of precise because we need the registers 299 if (dst.index() == noreg && dst.disp() == 0) { 300 if (dst.base() != tmp1) { 301 __ movptr(tmp1, dst.base()); 302 } 303 } else { 304 __ lea(tmp1, dst); 305 } 306 307 NOT_LP64(__ get_thread(rcx)); 308 NOT_LP64(__ save_bcp()); 309 310 g1_write_barrier_pre(masm /*masm*/, 311 tmp1 /* obj */, 312 tmp2 /* pre_val */, 313 rthread /* thread */, 314 tmp3 /* tmp */, 315 val != noreg /* tosca_live */, 316 false /* expand_call */); 317 if (val == noreg) { 318 __ store_heap_oop_null(Address(tmp1, 0)); 319 } else { 320 // G1 barrier needs uncompressed oop for region cross check. 321 Register new_val = val; 322 if (UseCompressedOops) { 323 new_val = tmp2; 324 __ movptr(new_val, val); 325 } 326 __ store_heap_oop(Address(tmp1, 0), val); 327 g1_write_barrier_post(masm /*masm*/, 328 tmp1 /* store_adr */, 329 new_val /* new_val */, 330 rthread /* thread */, 331 tmp3 /* tmp */, 332 tmp2 /* tmp2 */); 333 } 334 NOT_LP64( __ restore_bcp()); 335 } 336 337 void G1BSCodeGen::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators, Register addr, Register count, Register tmp) { 338 __ pusha(); // push registers (overkill) 339 if (c_rarg0 == count) { // On win64 c_rarg0 == rcx 340 assert_different_registers(c_rarg1, addr); 341 __ mov(c_rarg1, count); 342 __ mov(c_rarg0, addr); 343 } else { 344 assert_different_registers(c_rarg0, count); 345 __ mov(c_rarg0, addr); 346 __ mov(c_rarg1, count); 347 } 348 __ call_VM_leaf(CAST_FROM_FN_PTR(address, ModRefBarrierSet::static_write_ref_array_post), 2); 349 __ popa(); 350 } 351 352 #undef __ 353 #define __ ce->masm()-> 354 355 void G1BSCodeGen::gen_g1_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) { 356 C1G1BSCodeGen* code_gen = (C1G1BSCodeGen*)Universe::heap()->barrier_set()->c1_code_gen(); 357 // At this point we know that marking is in progress. 358 // If do_load() is true then we have to emit the 359 // load of the previous value; otherwise it has already 360 // been loaded into _pre_val. 361 362 __ bind(stub->_entry); 363 assert(stub->pre_val()->is_register(), "Precondition."); 364 365 Register pre_val_reg = stub->pre_val()->as_register(); 366 367 if (stub->do_load()) { 368 ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/, false /*unaligned*/); 369 } 370 371 __ cmpptr(pre_val_reg, (int32_t)NULL_WORD); 372 __ jcc(Assembler::equal, stub->_continuation); 373 ce->store_parameter(stub->pre_val()->as_register(), 0); 374 __ call(RuntimeAddress(code_gen->pre_barrier_c1_runtime_code_blob()->code_begin())); 375 __ jmp(stub->_continuation); 376 377 } 378 379 void G1BSCodeGen::gen_g1_post_barrier_stub(LIR_Assembler* ce, G1PostBarrierStub* stub) { 380 C1G1BSCodeGen* code_gen = (C1G1BSCodeGen*)Universe::heap()->barrier_set()->c1_code_gen(); 381 __ bind(stub->_entry); 382 assert(stub->addr()->is_register(), "Precondition."); 383 assert(stub->new_val()->is_register(), "Precondition."); 384 Register new_val_reg = stub->new_val()->as_register(); 385 __ cmpptr(new_val_reg, (int32_t) NULL_WORD); 386 __ jcc(Assembler::equal, stub->_continuation); 387 ce->store_parameter(stub->addr()->as_pointer_register(), 0); 388 __ call(RuntimeAddress(code_gen->post_barrier_c1_runtime_code_blob()->code_begin())); 389 __ jmp(stub->_continuation); 390 } 391 392 #undef __ 393 394 #define __ sasm-> 395 396 void G1BSCodeGen::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) { 397 __ prologue("g1_pre_barrier", false); 398 // arg0 : previous value of memory 399 400 BarrierSet* bs = Universe::heap()->barrier_set(); 401 __ push(rax); 402 __ push(rdx); 403 404 const Register pre_val = rax; 405 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread); 406 const Register tmp = rdx; 407 408 NOT_LP64(__ get_thread(thread);) 409 410 Address queue_active(thread, in_bytes(JavaThread::satb_mark_queue_offset() + 411 SATBMarkQueue::byte_offset_of_active())); 412 Address queue_index(thread, in_bytes(JavaThread::satb_mark_queue_offset() + 413 SATBMarkQueue::byte_offset_of_index())); 414 Address buffer(thread, in_bytes(JavaThread::satb_mark_queue_offset() + 415 SATBMarkQueue::byte_offset_of_buf())); 416 417 Label done; 418 Label runtime; 419 420 // Is marking still active? 421 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) { 422 __ cmpl(queue_active, 0); 423 } else { 424 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption"); 425 __ cmpb(queue_active, 0); 426 } 427 __ jcc(Assembler::equal, done); 428 429 // Can we store original value in the thread's buffer? 430 431 __ movptr(tmp, queue_index); 432 __ testptr(tmp, tmp); 433 __ jcc(Assembler::zero, runtime); 434 __ subptr(tmp, wordSize); 435 __ movptr(queue_index, tmp); 436 __ addptr(tmp, buffer); 437 438 // prev_val (rax) 439 __ load_parameter(0, pre_val); 440 __ movptr(Address(tmp, 0), pre_val); 441 __ jmp(done); 442 443 __ bind(runtime); 444 445 __ save_live_registers_no_oop_map(3, true); 446 447 // load the pre-value 448 __ load_parameter(0, rcx); 449 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSet::g1_wb_pre), rcx, thread); 450 451 __ restore_live_registers(true); 452 453 __ bind(done); 454 455 __ pop(rdx); 456 __ pop(rax); 457 458 __ epilogue(); 459 } 460 461 void G1BSCodeGen::generate_c1_post_barrier_runtime_stub(StubAssembler* sasm) { 462 __ prologue("g1_post_barrier", false); 463 464 // arg0: store_address 465 Address store_addr(rbp, 2*BytesPerWord); 466 467 CardTableModRefBS* ct = 468 barrier_set_cast<CardTableModRefBS>(Universe::heap()->barrier_set()); 469 assert(sizeof(*ct->card_table()->byte_map_base()) == sizeof(jbyte), "adjust this code"); 470 471 Label done; 472 Label enqueued; 473 Label runtime; 474 475 // At this point we know new_value is non-NULL and the new_value crosses regions. 476 // Must check to see if card is already dirty 477 478 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread); 479 480 Address queue_index(thread, in_bytes(JavaThread::dirty_card_queue_offset() + 481 DirtyCardQueue::byte_offset_of_index())); 482 Address buffer(thread, in_bytes(JavaThread::dirty_card_queue_offset() + 483 DirtyCardQueue::byte_offset_of_buf())); 484 485 __ push(rax); 486 __ push(rcx); 487 488 const Register cardtable = rax; 489 const Register card_addr = rcx; 490 491 __ load_parameter(0, card_addr); 492 __ shrptr(card_addr, CardTable::card_shift); 493 // Do not use ExternalAddress to load 'byte_map_base', since 'byte_map_base' is NOT 494 // a valid address and therefore is not properly handled by the relocation code. 495 __ movptr(cardtable, (intptr_t)ct->card_table()->byte_map_base()); 496 __ addptr(card_addr, cardtable); 497 498 NOT_LP64(__ get_thread(thread);) 499 500 __ cmpb(Address(card_addr, 0), (int)G1CardTable::g1_young_card_val()); 501 __ jcc(Assembler::equal, done); 502 503 __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad)); 504 __ cmpb(Address(card_addr, 0), (int)G1CardTable::dirty_card_val()); 505 __ jcc(Assembler::equal, done); 506 507 // storing region crossing non-NULL, card is clean. 508 // dirty card and log. 509 510 __ movb(Address(card_addr, 0), (int)G1CardTable::dirty_card_val()); 511 512 const Register tmp = rdx; 513 __ push(rdx); 514 515 __ movptr(tmp, queue_index); 516 __ testptr(tmp, tmp); 517 __ jcc(Assembler::zero, runtime); 518 __ subptr(tmp, wordSize); 519 __ movptr(queue_index, tmp); 520 __ addptr(tmp, buffer); 521 __ movptr(Address(tmp, 0), card_addr); 522 __ jmp(enqueued); 523 524 __ bind(runtime); 525 526 __ save_live_registers_no_oop_map(3, true); 527 528 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSet::g1_wb_post), card_addr, thread); 529 530 __ restore_live_registers(true); 531 532 __ bind(enqueued); 533 __ pop(rdx); 534 535 __ bind(done); 536 __ pop(rcx); 537 __ pop(rax); 538 539 __ epilogue(); 540 } 541 542 #undef __