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/g1CardTable.hpp" 30 #include "gc/g1/g1ThreadLocalData.hpp" 31 #include "gc/g1/heapRegion.hpp" 32 #include "interpreter/interp_masm.hpp" 33 #include "runtime/sharedRuntime.hpp" 34 35 #define __ masm-> 36 37 void G1BarrierSetAssembler::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators, 38 Register addr, Register count, RegSet saved_regs) { 39 bool dest_uninitialized = (decorators & AS_DEST_NOT_INITIALIZED) != 0; 40 if (!dest_uninitialized) { 41 __ push(saved_regs, sp); 42 if (count == c_rarg0) { 43 if (addr == c_rarg1) { 44 // exactly backwards!! 45 __ mov(rscratch1, c_rarg0); 46 __ mov(c_rarg0, c_rarg1); 47 __ mov(c_rarg1, rscratch1); 48 } else { 49 __ mov(c_rarg1, count); 50 __ mov(c_rarg0, addr); 51 } 52 } else { 53 __ mov(c_rarg0, addr); 54 __ mov(c_rarg1, count); 55 } 56 if (UseCompressedOops) { 57 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSet::write_ref_array_pre_narrow_oop_entry), 2); 58 } else { 59 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSet::write_ref_array_pre_oop_entry), 2); 60 } 61 __ pop(saved_regs, sp); 62 } 63 } 64 65 void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators, 66 Register start, Register end, Register scratch, RegSet saved_regs) { 67 __ push(saved_regs, sp); 68 // must compute element count unless barrier set interface is changed (other platforms supply count) 69 assert_different_registers(start, end, scratch); 70 __ lea(scratch, Address(end, BytesPerHeapOop)); 71 __ sub(scratch, scratch, start); // subtract start to get #bytes 72 __ lsr(scratch, scratch, LogBytesPerHeapOop); // convert to element count 73 __ mov(c_rarg0, start); 74 __ mov(c_rarg1, scratch); 75 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSet::write_ref_array_post_entry), 2); 76 __ pop(saved_regs, sp); 77 } 78 79 void G1BarrierSetAssembler::g1_write_barrier_pre(MacroAssembler* masm, 80 Register obj, 81 Register pre_val, 82 Register thread, 83 Register tmp, 84 bool tosca_live, 85 bool expand_call) { 86 // If expand_call is true then we expand the call_VM_leaf macro 87 // directly to skip generating the check by 88 // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp. 89 90 assert(thread == rthread, "must be"); 91 92 Label done; 93 Label runtime; 94 95 assert_different_registers(obj, pre_val, tmp, rscratch1); 96 assert(pre_val != noreg && tmp != noreg, "expecting a register"); 97 98 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset())); 99 Address index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset())); 100 Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset())); 101 102 // Is marking active? 103 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) { 104 __ ldrw(tmp, in_progress); 105 } else { 106 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption"); 107 __ ldrb(tmp, in_progress); 108 } 109 __ cbzw(tmp, done); 110 111 // Do we need to load the previous value? 112 if (obj != noreg) { 113 __ load_heap_oop(pre_val, Address(obj, 0)); 114 } 115 116 // Is the previous value null? 117 __ cbz(pre_val, done); 118 119 // Can we store original value in the thread's buffer? 120 // Is index == 0? 121 // (The index field is typed as size_t.) 122 123 __ ldr(tmp, index); // tmp := *index_adr 124 __ cbz(tmp, runtime); // tmp == 0? 125 // If yes, goto runtime 126 127 __ sub(tmp, tmp, wordSize); // tmp := tmp - wordSize 128 __ str(tmp, index); // *index_adr := tmp 129 __ ldr(rscratch1, buffer); 130 __ add(tmp, tmp, rscratch1); // tmp := tmp + *buffer_adr 131 132 // Record the previous value 133 __ str(pre_val, Address(tmp, 0)); 134 __ b(done); 135 136 __ bind(runtime); 137 // save the live input values 138 RegSet saved = RegSet::of(pre_val); 139 if (tosca_live) saved += RegSet::of(r0); 140 if (obj != noreg) saved += RegSet::of(obj); 141 142 __ push(saved, sp); 143 144 // Calling the runtime using the regular call_VM_leaf mechanism generates 145 // code (generated by InterpreterMacroAssember::call_VM_leaf_base) 146 // that checks that the *(rfp+frame::interpreter_frame_last_sp) == NULL. 147 // 148 // If we care generating the pre-barrier without a frame (e.g. in the 149 // intrinsified Reference.get() routine) then ebp might be pointing to 150 // the caller frame and so this check will most likely fail at runtime. 151 // 152 // Expanding the call directly bypasses the generation of the check. 153 // So when we do not have have a full interpreter frame on the stack 154 // expand_call should be passed true. 155 156 if (expand_call) { 157 assert(pre_val != c_rarg1, "smashed arg"); 158 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), pre_val, thread); 159 } else { 160 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), pre_val, thread); 161 } 162 163 __ pop(saved, sp); 164 165 __ bind(done); 166 167 } 168 169 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm, 170 Register store_addr, 171 Register new_val, 172 Register thread, 173 Register tmp, 174 Register tmp2) { 175 assert(thread == rthread, "must be"); 176 assert_different_registers(store_addr, new_val, thread, tmp, tmp2, 177 rscratch1); 178 assert(store_addr != noreg && new_val != noreg && tmp != noreg 179 && tmp2 != noreg, "expecting a register"); 180 181 Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset())); 182 Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset())); 183 184 BarrierSet* bs = BarrierSet::barrier_set(); 185 CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs); 186 CardTable* ct = ctbs->card_table(); 187 assert(sizeof(*ct->byte_map_base()) == sizeof(jbyte), "adjust this code"); 188 189 Label done; 190 Label runtime; 191 192 // Does store cross heap regions? 193 194 __ eor(tmp, store_addr, new_val); 195 __ lsr(tmp, tmp, HeapRegion::LogOfHRGrainBytes); 196 __ cbz(tmp, done); 197 198 // crosses regions, storing NULL? 199 200 __ cbz(new_val, done); 201 202 // storing region crossing non-NULL, is card already dirty? 203 204 ExternalAddress cardtable((address) ct->byte_map_base()); 205 assert(sizeof(*ct->byte_map_base()) == sizeof(jbyte), "adjust this code"); 206 const Register card_addr = tmp; 207 208 __ lsr(card_addr, store_addr, CardTable::card_shift); 209 210 // get the address of the card 211 __ load_byte_map_base(tmp2); 212 __ add(card_addr, card_addr, tmp2); 213 __ ldrb(tmp2, Address(card_addr)); 214 __ cmpw(tmp2, (int)G1CardTable::g1_young_card_val()); 215 __ br(Assembler::EQ, done); 216 217 assert((int)CardTable::dirty_card_val() == 0, "must be 0"); 218 219 __ membar(Assembler::StoreLoad); 220 221 __ ldrb(tmp2, Address(card_addr)); 222 __ cbzw(tmp2, done); 223 224 // storing a region crossing, non-NULL oop, card is clean. 225 // dirty card and log. 226 227 __ strb(zr, Address(card_addr)); 228 229 __ ldr(rscratch1, queue_index); 230 __ cbz(rscratch1, runtime); 231 __ sub(rscratch1, rscratch1, wordSize); 232 __ str(rscratch1, queue_index); 233 234 __ ldr(tmp2, buffer); 235 __ str(card_addr, Address(tmp2, rscratch1)); 236 __ b(done); 237 238 __ bind(runtime); 239 // save the live input values 240 RegSet saved = RegSet::of(store_addr, new_val); 241 __ push(saved, sp); 242 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, thread); 243 __ pop(saved, sp); 244 245 __ bind(done); 246 } 247 248 void G1BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type, 249 Register dst, Address src, Register tmp1, Register tmp_thread) { 250 bool on_oop = type == T_OBJECT || type == T_ARRAY; 251 bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0; 252 bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0; 253 bool on_reference = on_weak || on_phantom; 254 ModRefBarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1, tmp_thread); 255 if (on_oop && on_reference) { 256 // Generate the G1 pre-barrier code to log the value of 257 // the referent field in an SATB buffer. 258 g1_write_barrier_pre(masm /* masm */, 259 noreg /* obj */, 260 dst /* pre_val */, 261 rthread /* thread */, 262 tmp1 /* tmp */, 263 true /* tosca_live */, 264 true /* expand_call */); 265 } 266 } 267 268 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type, 269 Address dst, Register val, Register tmp1, Register tmp2) { 270 // flatten object address if needed 271 if (dst.index() == noreg && dst.offset() == 0) { 272 if (dst.base() != r3) { 273 __ mov(r3, dst.base()); 274 } 275 } else { 276 __ lea(r3, dst); 277 } 278 279 g1_write_barrier_pre(masm, 280 r3 /* obj */, 281 tmp2 /* pre_val */, 282 rthread /* thread */, 283 tmp1 /* tmp */, 284 val != noreg /* tosca_live */, 285 false /* expand_call */); 286 287 if (val == noreg) { 288 __ store_heap_oop_null(Address(r3, 0)); 289 } else { 290 // G1 barrier needs uncompressed oop for region cross check. 291 Register new_val = val; 292 if (UseCompressedOops) { 293 new_val = rscratch2; 294 __ mov(new_val, val); 295 } 296 __ store_heap_oop(Address(r3, 0), val); 297 g1_write_barrier_post(masm, 298 r3 /* store_adr */, 299 new_val /* new_val */, 300 rthread /* thread */, 301 tmp1 /* tmp */, 302 tmp2 /* tmp2 */); 303 } 304 305 } 306 307 #undef __