1 /*
   2  * Copyright (c) 2018, 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 "gc/g1/g1BarrierSet.hpp"
  28 #include "gc/g1/g1BarrierSetAssembler.hpp"
  29 #include "gc/g1/g1BarrierSetRuntime.hpp"
  30 #include "gc/g1/g1CardTable.hpp"
  31 #include "gc/g1/g1ThreadLocalData.hpp"
  32 #include "gc/g1/heapRegion.hpp"
  33 #include "gc/shared/collectedHeap.hpp"
  34 #include "runtime/sharedRuntime.hpp"
  35 #include "runtime/thread.hpp"
  36 #include "interpreter/interp_masm.hpp"
  37 #include "runtime/sharedRuntime.hpp"
  38 #ifdef COMPILER1
  39 #include "c1/c1_LIRAssembler.hpp"
  40 #include "c1/c1_MacroAssembler.hpp"
  41 #include "gc/g1/c1/g1BarrierSetC1.hpp"
  42 #endif
  43 
  44 #define __ masm->
  45 
  46 void G1BarrierSetAssembler::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators,
  47                                                             Register addr, Register count, RegSet saved_regs) {
  48   bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;
  49   if (!dest_uninitialized) {
  50     __ push(saved_regs, sp);
  51     if (count == c_rarg0) {
  52       if (addr == c_rarg1) {
  53         // exactly backwards!!
  54         __ mov(rscratch1, c_rarg0);
  55         __ mov(c_rarg0, c_rarg1);
  56         __ mov(c_rarg1, rscratch1);
  57       } else {
  58         __ mov(c_rarg1, count);
  59         __ mov(c_rarg0, addr);
  60       }
  61     } else {
  62       __ mov(c_rarg0, addr);
  63       __ mov(c_rarg1, count);
  64     }
  65     if (UseCompressedOops) {
  66       __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_narrow_oop_entry), 2);
  67     } else {
  68       __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry), 2);
  69     }
  70     __ pop(saved_regs, sp);
  71   }
  72 }
  73 
  74 void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
  75                                                              Register start, Register count, Register scratch, RegSet saved_regs) {
  76   __ push(saved_regs, sp);
  77   assert_different_registers(start, count, scratch);
  78   assert_different_registers(c_rarg0, count);
  79   __ mov(c_rarg0, start);
  80   __ mov(c_rarg1, count);
  81   __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_post_entry), 2);
  82   __ pop(saved_regs, sp);
  83 }
  84 
  85 void G1BarrierSetAssembler::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   assert(thread == rthread, "must be");
  97 
  98   Label done;
  99   Label runtime;
 100 
 101   assert_different_registers(obj, pre_val, tmp, rscratch1);
 102   assert(pre_val != noreg &&  tmp != noreg, "expecting a register");
 103 
 104   Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
 105   Address index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
 106   Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
 107 
 108   // Is marking active?
 109   if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
 110     __ ldrw(tmp, in_progress);
 111   } else {
 112     assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
 113     __ ldrb(tmp, in_progress);
 114   }
 115   __ cbzw(tmp, done);
 116 
 117   // Do we need to load the previous value?
 118   if (obj != noreg) {
 119     __ load_heap_oop(pre_val, Address(obj, 0), noreg, noreg, AS_RAW);
 120   }
 121 
 122   // Is the previous value null?
 123   __ cbz(pre_val, done);
 124 
 125   // Can we store original value in the thread's buffer?
 126   // Is index == 0?
 127   // (The index field is typed as size_t.)
 128 
 129   __ ldr(tmp, index);                      // tmp := *index_adr
 130   __ cbz(tmp, runtime);                    // tmp == 0?
 131                                         // If yes, goto runtime
 132 
 133   __ sub(tmp, tmp, wordSize);              // tmp := tmp - wordSize
 134   __ str(tmp, index);                      // *index_adr := tmp
 135   __ ldr(rscratch1, buffer);
 136   __ add(tmp, tmp, rscratch1);             // tmp := tmp + *buffer_adr
 137 
 138   // Record the previous value
 139   __ str(pre_val, Address(tmp, 0));
 140   __ b(done);
 141 
 142   __ bind(runtime);
 143   // save the live input values
 144   RegSet saved = RegSet::of(pre_val);
 145   if (tosca_live) saved += RegSet::of(r0);
 146   if (obj != noreg) saved += RegSet::of(obj);
 147 
 148   __ push(saved, sp);
 149 
 150   // Calling the runtime using the regular call_VM_leaf mechanism generates
 151   // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
 152   // that checks that the *(rfp+frame::interpreter_frame_last_sp) == NULL.
 153   //
 154   // If we care generating the pre-barrier without a frame (e.g. in the
 155   // intrinsified Reference.get() routine) then ebp might be pointing to
 156   // the caller frame and so this check will most likely fail at runtime.
 157   //
 158   // Expanding the call directly bypasses the generation of the check.
 159   // So when we do not have have a full interpreter frame on the stack
 160   // expand_call should be passed true.
 161 
 162   if (expand_call) {
 163     assert(pre_val != c_rarg1, "smashed arg");
 164     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
 165   } else {
 166     __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
 167   }
 168 
 169   __ pop(saved, sp);
 170 
 171   __ bind(done);
 172 
 173 }
 174 
 175 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
 176                                                   Register store_addr,
 177                                                   Register new_val,
 178                                                   Register thread,
 179                                                   Register tmp,
 180                                                   Register tmp2) {
 181   assert(thread == rthread, "must be");
 182   assert_different_registers(store_addr, new_val, thread, tmp, tmp2,
 183                              rscratch1);
 184   assert(store_addr != noreg && new_val != noreg && tmp != noreg
 185          && tmp2 != noreg, "expecting a register");
 186 
 187   Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
 188   Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
 189 
 190   BarrierSet* bs = BarrierSet::barrier_set();
 191   CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs);
 192   CardTable* ct = ctbs->card_table();
 193 
 194   Label done;
 195   Label runtime;
 196 
 197   // Does store cross heap regions?
 198 
 199   __ eor(tmp, store_addr, new_val);
 200   __ lsr(tmp, tmp, HeapRegion::LogOfHRGrainBytes);
 201   __ cbz(tmp, done);
 202 
 203   // crosses regions, storing NULL?
 204 
 205   __ cbz(new_val, done);
 206 
 207   // storing region crossing non-NULL, is card already dirty?
 208 
 209   ExternalAddress cardtable((address) ct->byte_map_base());
 210   const Register card_addr = tmp;
 211 
 212   __ lsr(card_addr, store_addr, CardTable::card_shift);
 213 
 214   // get the address of the card
 215   __ load_byte_map_base(tmp2);
 216   __ add(card_addr, card_addr, tmp2);
 217   __ ldrb(tmp2, Address(card_addr));
 218   __ cmpw(tmp2, (int)G1CardTable::g1_young_card_val());
 219   __ br(Assembler::EQ, done);
 220 
 221   assert((int)CardTable::dirty_card_val() == 0, "must be 0");
 222 
 223   __ membar(Assembler::StoreLoad);
 224 
 225   __ ldrb(tmp2, Address(card_addr));
 226   __ cbzw(tmp2, done);
 227 
 228   // storing a region crossing, non-NULL oop, card is clean.
 229   // dirty card and log.
 230 
 231   __ strb(zr, Address(card_addr));
 232 
 233   __ ldr(rscratch1, queue_index);
 234   __ cbz(rscratch1, runtime);
 235   __ sub(rscratch1, rscratch1, wordSize);
 236   __ str(rscratch1, queue_index);
 237 
 238   __ ldr(tmp2, buffer);
 239   __ str(card_addr, Address(tmp2, rscratch1));
 240   __ b(done);
 241 
 242   __ bind(runtime);
 243   // save the live input values
 244   RegSet saved = RegSet::of(store_addr, new_val);
 245   __ push(saved, sp);
 246   __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
 247   __ pop(saved, sp);
 248 
 249   __ bind(done);
 250 }
 251 
 252 void G1BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
 253                                     Register dst, Address src, Register tmp1, Register tmp_thread) {
 254   bool on_oop = type == T_OBJECT || type == T_ARRAY;
 255   bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
 256   bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
 257   bool on_reference = on_weak || on_phantom;
 258   ModRefBarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1, tmp_thread);
 259   if (on_oop && on_reference) {
 260     // LR is live.  It must be saved around calls.
 261     __ enter(); // barrier may call runtime
 262     // Generate the G1 pre-barrier code to log the value of
 263     // the referent field in an SATB buffer.
 264     g1_write_barrier_pre(masm /* masm */,
 265                          noreg /* obj */,
 266                          dst /* pre_val */,
 267                          rthread /* thread */,
 268                          tmp1 /* tmp */,
 269                          true /* tosca_live */,
 270                          true /* expand_call */);
 271     __ leave();
 272   }
 273 }
 274 
 275 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
 276                                          Address dst, Register val, Register tmp1, Register tmp2, Register tmp3) {
 277 
 278    if (tmp3 == noreg) {
 279        tmp3 = r8;  
 280    }
 281 
 282   // flatten object address if needed
 283   if (dst.index() == noreg && dst.offset() == 0) {
 284     if (dst.base() != r3) {
 285       __ mov(r3, dst.base());
 286     }
 287   } else {
 288     __ lea(r3, dst);
 289   }
 290 
 291   g1_write_barrier_pre(masm,
 292                        r3 /* obj */,
 293                        tmp2 /* pre_val */,
 294                        rthread /* thread */,
 295                        tmp1  /* tmp */,
 296                        val != noreg /* tosca_live */,
 297                        false /* expand_call */);
 298 
 299   if (val == noreg) {
 300     BarrierSetAssembler::store_at(masm, decorators, type, Address(r3, 0), noreg, noreg, noreg, noreg);
 301   } else {
 302     // G1 barrier needs uncompressed oop for region cross check.
 303     Register new_val = val;
 304     if (UseCompressedOops) {
 305       new_val = rscratch2;
 306       __ mov(new_val, val);
 307     }
 308     BarrierSetAssembler::store_at(masm, decorators, type, Address(r3, 0), val, noreg, noreg, noreg);
 309     g1_write_barrier_post(masm,
 310                           r3 /* store_adr */,
 311                           new_val /* new_val */,
 312                           rthread /* thread */,
 313                           tmp1 /* tmp */,
 314                           tmp2 /* tmp2 */);
 315   }
 316 
 317 }
 318 
 319 #ifdef COMPILER1
 320 
 321 #undef __
 322 #define __ ce->masm()->
 323 
 324 void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
 325   G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
 326   // At this point we know that marking is in progress.
 327   // If do_load() is true then we have to emit the
 328   // load of the previous value; otherwise it has already
 329   // been loaded into _pre_val.
 330 
 331   __ bind(*stub->entry());
 332 
 333   assert(stub->pre_val()->is_register(), "Precondition.");
 334 
 335   Register pre_val_reg = stub->pre_val()->as_register();
 336 
 337   if (stub->do_load()) {
 338     ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/, false /*unaligned*/);
 339   }
 340   __ cbz(pre_val_reg, *stub->continuation());
 341   ce->store_parameter(stub->pre_val()->as_register(), 0);
 342   __ far_call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
 343   __ b(*stub->continuation());
 344 }
 345 
 346 void G1BarrierSetAssembler::gen_post_barrier_stub(LIR_Assembler* ce, G1PostBarrierStub* stub) {
 347   G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
 348   __ bind(*stub->entry());
 349   assert(stub->addr()->is_register(), "Precondition.");
 350   assert(stub->new_val()->is_register(), "Precondition.");
 351   Register new_val_reg = stub->new_val()->as_register();
 352   __ cbz(new_val_reg, *stub->continuation());
 353   ce->store_parameter(stub->addr()->as_pointer_register(), 0);
 354   __ far_call(RuntimeAddress(bs->post_barrier_c1_runtime_code_blob()->code_begin()));
 355   __ b(*stub->continuation());
 356 }
 357 
 358 #undef __
 359 
 360 #define __ sasm->
 361 
 362 void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
 363   __ prologue("g1_pre_barrier", false);
 364 
 365   // arg0 : previous value of memory
 366 
 367   BarrierSet* bs = BarrierSet::barrier_set();
 368 
 369   const Register pre_val = r0;
 370   const Register thread = rthread;
 371   const Register tmp = rscratch1;
 372 
 373   Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
 374   Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
 375   Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
 376 
 377   Label done;
 378   Label runtime;
 379 
 380   // Is marking still active?
 381   if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
 382     __ ldrw(tmp, in_progress);
 383   } else {
 384     assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
 385     __ ldrb(tmp, in_progress);
 386   }
 387   __ cbzw(tmp, done);
 388 
 389   // Can we store original value in the thread's buffer?
 390   __ ldr(tmp, queue_index);
 391   __ cbz(tmp, runtime);
 392 
 393   __ sub(tmp, tmp, wordSize);
 394   __ str(tmp, queue_index);
 395   __ ldr(rscratch2, buffer);
 396   __ add(tmp, tmp, rscratch2);
 397   __ load_parameter(0, rscratch2);
 398   __ str(rscratch2, Address(tmp, 0));
 399   __ b(done);
 400 
 401   __ bind(runtime);
 402   __ push_call_clobbered_registers();
 403   __ load_parameter(0, pre_val);
 404   __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
 405   __ pop_call_clobbered_registers();
 406   __ bind(done);
 407 
 408   __ epilogue();
 409 }
 410 
 411 void G1BarrierSetAssembler::generate_c1_post_barrier_runtime_stub(StubAssembler* sasm) {
 412   __ prologue("g1_post_barrier", false);
 413 
 414   // arg0: store_address
 415   Address store_addr(rfp, 2*BytesPerWord);
 416 
 417   BarrierSet* bs = BarrierSet::barrier_set();
 418   CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs);
 419   CardTable* ct = ctbs->card_table();
 420 
 421   Label done;
 422   Label runtime;
 423 
 424   // At this point we know new_value is non-NULL and the new_value crosses regions.
 425   // Must check to see if card is already dirty
 426 
 427   const Register thread = rthread;
 428 
 429   Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
 430   Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
 431 
 432   const Register card_offset = rscratch2;
 433   // LR is free here, so we can use it to hold the byte_map_base.
 434   const Register byte_map_base = lr;
 435 
 436   assert_different_registers(card_offset, byte_map_base, rscratch1);
 437 
 438   __ load_parameter(0, card_offset);
 439   __ lsr(card_offset, card_offset, CardTable::card_shift);
 440   __ load_byte_map_base(byte_map_base);
 441   __ ldrb(rscratch1, Address(byte_map_base, card_offset));
 442   __ cmpw(rscratch1, (int)G1CardTable::g1_young_card_val());
 443   __ br(Assembler::EQ, done);
 444 
 445   assert((int)CardTable::dirty_card_val() == 0, "must be 0");
 446 
 447   __ membar(Assembler::StoreLoad);
 448   __ ldrb(rscratch1, Address(byte_map_base, card_offset));
 449   __ cbzw(rscratch1, done);
 450 
 451   // storing region crossing non-NULL, card is clean.
 452   // dirty card and log.
 453   __ strb(zr, Address(byte_map_base, card_offset));
 454 
 455   // Convert card offset into an address in card_addr
 456   Register card_addr = card_offset;
 457   __ add(card_addr, byte_map_base, card_addr);
 458 
 459   __ ldr(rscratch1, queue_index);
 460   __ cbz(rscratch1, runtime);
 461   __ sub(rscratch1, rscratch1, wordSize);
 462   __ str(rscratch1, queue_index);
 463 
 464   // Reuse LR to hold buffer_addr
 465   const Register buffer_addr = lr;
 466 
 467   __ ldr(buffer_addr, buffer);
 468   __ str(card_addr, Address(buffer_addr, rscratch1));
 469   __ b(done);
 470 
 471   __ bind(runtime);
 472   __ push_call_clobbered_registers();
 473   __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
 474   __ pop_call_clobbered_registers();
 475   __ bind(done);
 476   __ epilogue();
 477 }
 478 
 479 #undef __
 480 
 481 #endif // COMPILER1