1 /*
2 * Copyright (c) 2018, 2023, 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/g1HeapRegion.hpp"
32 #include "gc/g1/g1ThreadLocalData.hpp"
33 #include "gc/shared/collectedHeap.hpp"
34 #include "interpreter/interp_masm.hpp"
35 #include "runtime/javaThread.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #ifdef COMPILER1
38 #include "c1/c1_LIRAssembler.hpp"
39 #include "c1/c1_MacroAssembler.hpp"
40 #include "gc/g1/c1/g1BarrierSetC1.hpp"
41 #endif // COMPILER1
42 #ifdef COMPILER2
43 #include "gc/g1/c2/g1BarrierSetC2.hpp"
44 #endif // COMPILER2
45
46 #define __ masm->
47
48 void G1BarrierSetAssembler::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators,
49 Register addr, Register count, RegSet saved_regs) {
50 bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;
51 if (!dest_uninitialized) {
52 Label done;
53 Address in_progress(rthread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
54
55 // Is marking active?
56 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
57 __ ldrw(rscratch1, in_progress);
58 } else {
59 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
60 __ ldrb(rscratch1, in_progress);
61 }
62 __ cbzw(rscratch1, done);
63
64 __ push(saved_regs, sp);
65 if (count == c_rarg0) {
66 if (addr == c_rarg1) {
67 // exactly backwards!!
68 __ mov(rscratch1, c_rarg0);
69 __ mov(c_rarg0, c_rarg1);
70 __ mov(c_rarg1, rscratch1);
71 } else {
72 __ mov(c_rarg1, count);
73 __ mov(c_rarg0, addr);
74 }
75 } else {
76 __ mov(c_rarg0, addr);
77 __ mov(c_rarg1, count);
78 }
79 if (UseCompressedOops) {
80 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_narrow_oop_entry), 2);
81 } else {
82 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry), 2);
83 }
84 __ pop(saved_regs, sp);
85
86 __ bind(done);
87 }
88 }
89
90 void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
91 Register start, Register count, Register scratch, RegSet saved_regs) {
92 __ push(saved_regs, sp);
93 assert_different_registers(start, count, scratch);
94 assert_different_registers(c_rarg0, count);
95 __ mov(c_rarg0, start);
96 __ mov(c_rarg1, count);
97 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_post_entry), 2);
98 __ pop(saved_regs, sp);
99 }
100
101 static void generate_queue_test_and_insertion(MacroAssembler* masm, ByteSize index_offset, ByteSize buffer_offset, Label& runtime,
102 const Register thread, const Register value, const Register temp1, const Register temp2) {
103 // Can we store a value in the given thread's buffer?
104 // (The index field is typed as size_t.)
105 __ ldr(temp1, Address(thread, in_bytes(index_offset))); // temp1 := *(index address)
106 __ cbz(temp1, runtime); // jump to runtime if index == 0 (full buffer)
107 // The buffer is not full, store value into it.
108 __ sub(temp1, temp1, wordSize); // temp1 := next index
109 __ str(temp1, Address(thread, in_bytes(index_offset))); // *(index address) := next index
110 __ ldr(temp2, Address(thread, in_bytes(buffer_offset))); // temp2 := buffer address
111 __ str(value, Address(temp2, temp1)); // *(buffer address + next index) := value
112 }
113
114 static void generate_pre_barrier_fast_path(MacroAssembler* masm,
115 const Register thread,
116 const Register tmp1) {
117 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
118 // Is marking active?
119 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
120 __ ldrw(tmp1, in_progress);
121 } else {
122 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
123 __ ldrb(tmp1, in_progress);
124 }
125 }
126
127 static void generate_pre_barrier_slow_path(MacroAssembler* masm,
128 const Register obj,
129 const Register pre_val,
130 const Register thread,
131 const Register tmp1,
132 const Register tmp2,
133 Label& done,
134 Label& runtime) {
135 // Do we need to load the previous value?
136 if (obj != noreg) {
137 __ load_heap_oop(pre_val, Address(obj, 0), noreg, noreg, AS_RAW);
138 }
139 // Is the previous value null?
140 __ cbz(pre_val, done);
141 generate_queue_test_and_insertion(masm,
142 G1ThreadLocalData::satb_mark_queue_index_offset(),
143 G1ThreadLocalData::satb_mark_queue_buffer_offset(),
144 runtime,
145 thread, pre_val, tmp1, tmp2);
146 __ b(done);
147 }
148
149 void G1BarrierSetAssembler::g1_write_barrier_pre(MacroAssembler* masm,
150 Register obj,
151 Register pre_val,
152 Register thread,
153 Register tmp1,
154 Register tmp2,
155 bool tosca_live,
156 bool expand_call) {
157 // If expand_call is true then we expand the call_VM_leaf macro
158 // directly to skip generating the check by
159 // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
160
161 assert(thread == rthread, "must be");
162
163 Label done;
164 Label runtime;
165
166 assert_different_registers(obj, pre_val, tmp1, tmp2);
167 assert(pre_val != noreg && tmp1 != noreg && tmp2 != noreg, "expecting a register");
168
169 generate_pre_barrier_fast_path(masm, thread, tmp1);
170 // If marking is not active (*(mark queue active address) == 0), jump to done
171 __ cbzw(tmp1, done);
172 generate_pre_barrier_slow_path(masm, obj, pre_val, thread, tmp1, tmp2, done, runtime);
173
174 __ bind(runtime);
175
176 __ push_call_clobbered_registers();
177
178 // Calling the runtime using the regular call_VM_leaf mechanism generates
179 // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
180 // that checks that the *(rfp+frame::interpreter_frame_last_sp) == nullptr.
181 //
182 // If we care generating the pre-barrier without a frame (e.g. in the
183 // intrinsified Reference.get() routine) then rfp might be pointing to
184 // the caller frame and so this check will most likely fail at runtime.
185 //
186 // Expanding the call directly bypasses the generation of the check.
187 // So when we do not have have a full interpreter frame on the stack
188 // expand_call should be passed true.
189
190 if (expand_call) {
191 assert(pre_val != c_rarg1, "smashed arg");
192 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
193 } else {
194 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
195 }
196
197 __ pop_call_clobbered_registers();
198
199 __ bind(done);
200
201 }
202
203 static void generate_post_barrier_fast_path(MacroAssembler* masm,
204 const Register store_addr,
205 const Register new_val,
206 const Register tmp1,
207 const Register tmp2,
208 Label& done,
209 bool new_val_may_be_null) {
210 // Does store cross heap regions?
211 __ eor(tmp1, store_addr, new_val); // tmp1 := store address ^ new value
212 __ lsr(tmp1, tmp1, G1HeapRegion::LogOfHRGrainBytes); // tmp1 := ((store address ^ new value) >> LogOfHRGrainBytes)
213 __ cbz(tmp1, done);
214 // Crosses regions, storing null?
215 if (new_val_may_be_null) {
216 __ cbz(new_val, done);
217 }
218 // Storing region crossing non-null, is card young?
219 __ lsr(tmp1, store_addr, CardTable::card_shift()); // tmp1 := card address relative to card table base
220 __ load_byte_map_base(tmp2); // tmp2 := card table base address
221 __ add(tmp1, tmp1, tmp2); // tmp1 := card address
222 __ ldrb(tmp2, Address(tmp1)); // tmp2 := card
223 __ cmpw(tmp2, (int)G1CardTable::g1_young_card_val()); // tmp2 := card == young_card_val?
224 }
225
226 static void generate_post_barrier_slow_path(MacroAssembler* masm,
227 const Register thread,
228 const Register tmp1,
229 const Register tmp2,
230 Label& done,
231 Label& runtime) {
232 __ membar(Assembler::StoreLoad); // StoreLoad membar
233 __ ldrb(tmp2, Address(tmp1)); // tmp2 := card
234 __ cbzw(tmp2, done);
235 // Storing a region crossing, non-null oop, card is clean.
236 // Dirty card and log.
237 STATIC_ASSERT(CardTable::dirty_card_val() == 0);
238 __ strb(zr, Address(tmp1)); // *(card address) := dirty_card_val
239 generate_queue_test_and_insertion(masm,
240 G1ThreadLocalData::dirty_card_queue_index_offset(),
241 G1ThreadLocalData::dirty_card_queue_buffer_offset(),
242 runtime,
243 thread, tmp1, tmp2, rscratch1);
244 __ b(done);
245 }
246
247 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
248 Register store_addr,
249 Register new_val,
250 Register thread,
251 Register tmp1,
252 Register tmp2) {
253 assert(thread == rthread, "must be");
254 assert_different_registers(store_addr, new_val, thread, tmp1, tmp2,
255 rscratch1);
256 assert(store_addr != noreg && new_val != noreg && tmp1 != noreg
257 && tmp2 != noreg, "expecting a register");
258
259 Label done;
260 Label runtime;
261
262 generate_post_barrier_fast_path(masm, store_addr, new_val, tmp1, tmp2, done, true /* new_val_may_be_null */);
263 // If card is young, jump to done
264 __ br(Assembler::EQ, done);
265 generate_post_barrier_slow_path(masm, thread, tmp1, tmp2, done, runtime);
266
267 __ bind(runtime);
268 // save the live input values
269 RegSet saved = RegSet::of(store_addr);
270 __ push(saved, sp);
271 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), tmp1, thread);
272 __ pop(saved, sp);
273
274 __ bind(done);
275 }
276
277 #if defined(COMPILER2)
278
279 static void generate_c2_barrier_runtime_call(MacroAssembler* masm, G1BarrierStubC2* stub, const Register arg, const address runtime_path) {
280 SaveLiveRegisters save_registers(masm, stub);
281 if (c_rarg0 != arg) {
282 __ mov(c_rarg0, arg);
283 }
284 __ mov(c_rarg1, rthread);
285 __ mov(rscratch1, runtime_path);
286 __ blr(rscratch1);
287 }
288
289 void G1BarrierSetAssembler::g1_write_barrier_pre_c2(MacroAssembler* masm,
290 Register obj,
291 Register pre_val,
292 Register thread,
293 Register tmp1,
294 Register tmp2,
295 G1PreBarrierStubC2* stub) {
296 assert(thread == rthread, "must be");
297 assert_different_registers(obj, pre_val, tmp1, tmp2);
298 assert(pre_val != noreg && tmp1 != noreg && tmp2 != noreg, "expecting a register");
299
300 stub->initialize_registers(obj, pre_val, thread, tmp1, tmp2);
301
302 generate_pre_barrier_fast_path(masm, thread, tmp1);
303 // If marking is active (*(mark queue active address) != 0), jump to stub (slow path)
304 __ cbnzw(tmp1, *stub->entry());
305
306 __ bind(*stub->continuation());
307 }
308
309 void G1BarrierSetAssembler::generate_c2_pre_barrier_stub(MacroAssembler* masm,
310 G1PreBarrierStubC2* stub) const {
311 Assembler::InlineSkippedInstructionsCounter skip_counter(masm);
312 Label runtime;
313 Register obj = stub->obj();
314 Register pre_val = stub->pre_val();
315 Register thread = stub->thread();
316 Register tmp1 = stub->tmp1();
317 Register tmp2 = stub->tmp2();
318
319 __ bind(*stub->entry());
320 generate_pre_barrier_slow_path(masm, obj, pre_val, thread, tmp1, tmp2, *stub->continuation(), runtime);
321
322 __ bind(runtime);
323 generate_c2_barrier_runtime_call(masm, stub, pre_val, CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry));
324 __ b(*stub->continuation());
325 }
326
327 void G1BarrierSetAssembler::g1_write_barrier_post_c2(MacroAssembler* masm,
328 Register store_addr,
329 Register new_val,
330 Register thread,
331 Register tmp1,
332 Register tmp2,
333 G1PostBarrierStubC2* stub) {
334 assert(thread == rthread, "must be");
335 assert_different_registers(store_addr, new_val, thread, tmp1, tmp2,
336 rscratch1);
337 assert(store_addr != noreg && new_val != noreg && tmp1 != noreg
338 && tmp2 != noreg, "expecting a register");
339
340 stub->initialize_registers(thread, tmp1, tmp2);
341
342 bool new_val_may_be_null = (stub->barrier_data() & G1C2BarrierPostNotNull) == 0;
343 generate_post_barrier_fast_path(masm, store_addr, new_val, tmp1, tmp2, *stub->continuation(), new_val_may_be_null);
344 // If card is not young, jump to stub (slow path)
345 __ br(Assembler::NE, *stub->entry());
346
347 __ bind(*stub->continuation());
348 }
349
350 void G1BarrierSetAssembler::generate_c2_post_barrier_stub(MacroAssembler* masm,
351 G1PostBarrierStubC2* stub) const {
352 Assembler::InlineSkippedInstructionsCounter skip_counter(masm);
353 Label runtime;
354 Register thread = stub->thread();
355 Register tmp1 = stub->tmp1(); // tmp1 holds the card address.
356 Register tmp2 = stub->tmp2();
357 assert(stub->tmp3() == noreg, "not needed in this platform");
358
359 __ bind(*stub->entry());
360 generate_post_barrier_slow_path(masm, thread, tmp1, tmp2, *stub->continuation(), runtime);
361
362 __ bind(runtime);
363 generate_c2_barrier_runtime_call(masm, stub, tmp1, CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry));
364 __ b(*stub->continuation());
365 }
366
367 #endif // COMPILER2
368
369 void G1BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
370 Register dst, Address src, Register tmp1, Register tmp2) {
371 bool on_oop = is_reference_type(type);
372 bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
373 bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
374 bool on_reference = on_weak || on_phantom;
375 ModRefBarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1, tmp2);
376 if (on_oop && on_reference) {
377 // LR is live. It must be saved around calls.
378 __ enter(/*strip_ret_addr*/true); // barrier may call runtime
379 // Generate the G1 pre-barrier code to log the value of
380 // the referent field in an SATB buffer.
381 g1_write_barrier_pre(masm /* masm */,
382 noreg /* obj */,
383 dst /* pre_val */,
384 rthread /* thread */,
385 tmp1 /* tmp1 */,
386 tmp2 /* tmp2 */,
387 true /* tosca_live */,
388 true /* expand_call */);
389 __ leave();
390 }
391 }
392
393 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
394 Address dst, Register val, Register tmp1, Register tmp2, Register tmp3) {
395 // flatten object address if needed
396 if (dst.index() == noreg && dst.offset() == 0) {
397 if (dst.base() != tmp3) {
398 __ mov(tmp3, dst.base());
399 }
400 } else {
401 __ lea(tmp3, dst);
402 }
403
404 g1_write_barrier_pre(masm,
405 tmp3 /* obj */,
406 tmp2 /* pre_val */,
407 rthread /* thread */,
408 tmp1 /* tmp1 */,
409 rscratch2 /* tmp2 */,
410 val != noreg /* tosca_live */,
411 false /* expand_call */);
412
413 if (val == noreg) {
414 BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp3, 0), noreg, noreg, noreg, noreg);
415 } else {
416 // G1 barrier needs uncompressed oop for region cross check.
417 Register new_val = val;
418 if (UseCompressedOops) {
419 new_val = rscratch2;
420 __ mov(new_val, val);
421 }
422 BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp3, 0), val, noreg, noreg, noreg);
423 g1_write_barrier_post(masm,
424 tmp3 /* store_adr */,
425 new_val /* new_val */,
426 rthread /* thread */,
427 tmp1 /* tmp1 */,
428 tmp2 /* tmp2 */);
429 }
430
431 }
432
433 #ifdef COMPILER1
434
435 #undef __
436 #define __ ce->masm()->
437
438 void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
439 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
440 // At this point we know that marking is in progress.
441 // If do_load() is true then we have to emit the
442 // load of the previous value; otherwise it has already
443 // been loaded into _pre_val.
444
445 __ bind(*stub->entry());
446
447 assert(stub->pre_val()->is_register(), "Precondition.");
448
449 Register pre_val_reg = stub->pre_val()->as_register();
450
451 if (stub->do_load()) {
452 ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/);
453 }
454 __ cbz(pre_val_reg, *stub->continuation());
455 ce->store_parameter(stub->pre_val()->as_register(), 0);
456 __ far_call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
457 __ b(*stub->continuation());
458 }
459
460 void G1BarrierSetAssembler::gen_post_barrier_stub(LIR_Assembler* ce, G1PostBarrierStub* stub) {
461 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
462 __ bind(*stub->entry());
463 assert(stub->addr()->is_register(), "Precondition.");
464 assert(stub->new_val()->is_register(), "Precondition.");
465 Register new_val_reg = stub->new_val()->as_register();
466 __ cbz(new_val_reg, *stub->continuation());
467 ce->store_parameter(stub->addr()->as_pointer_register(), 0);
468 __ far_call(RuntimeAddress(bs->post_barrier_c1_runtime_code_blob()->code_begin()));
469 __ b(*stub->continuation());
470 }
471
472 #undef __
473
474 #define __ sasm->
475
476 void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
477 __ prologue("g1_pre_barrier", false);
478
479 // arg0 : previous value of memory
480
481 BarrierSet* bs = BarrierSet::barrier_set();
482
483 const Register pre_val = r0;
484 const Register thread = rthread;
485 const Register tmp = rscratch1;
486
487 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
488 Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
489 Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
490
491 Label done;
492 Label runtime;
493
494 // Is marking still active?
495 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
496 __ ldrw(tmp, in_progress);
497 } else {
498 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
499 __ ldrb(tmp, in_progress);
500 }
501 __ cbzw(tmp, done);
502
503 // Can we store original value in the thread's buffer?
504 __ ldr(tmp, queue_index);
505 __ cbz(tmp, runtime);
506
507 __ sub(tmp, tmp, wordSize);
508 __ str(tmp, queue_index);
509 __ ldr(rscratch2, buffer);
510 __ add(tmp, tmp, rscratch2);
511 __ load_parameter(0, rscratch2);
512 __ str(rscratch2, Address(tmp, 0));
513 __ b(done);
514
515 __ bind(runtime);
516 __ push_call_clobbered_registers();
517 __ load_parameter(0, pre_val);
518 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
519 __ pop_call_clobbered_registers();
520 __ bind(done);
521
522 __ epilogue();
523 }
524
525 void G1BarrierSetAssembler::generate_c1_post_barrier_runtime_stub(StubAssembler* sasm) {
526 __ prologue("g1_post_barrier", false);
527
528 // arg0: store_address
529 Address store_addr(rfp, 2*BytesPerWord);
530
531 BarrierSet* bs = BarrierSet::barrier_set();
532 CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs);
533 CardTable* ct = ctbs->card_table();
534
535 Label done;
536 Label runtime;
537
538 // At this point we know new_value is non-null and the new_value crosses regions.
539 // Must check to see if card is already dirty
540
541 const Register thread = rthread;
542
543 Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
544 Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
545
546 const Register card_offset = rscratch2;
547 // LR is free here, so we can use it to hold the byte_map_base.
548 const Register byte_map_base = lr;
549
550 assert_different_registers(card_offset, byte_map_base, rscratch1);
551
552 __ load_parameter(0, card_offset);
553 __ lsr(card_offset, card_offset, CardTable::card_shift());
554 __ load_byte_map_base(byte_map_base);
555 __ ldrb(rscratch1, Address(byte_map_base, card_offset));
556 __ cmpw(rscratch1, (int)G1CardTable::g1_young_card_val());
557 __ br(Assembler::EQ, done);
558
559 assert((int)CardTable::dirty_card_val() == 0, "must be 0");
560
561 __ membar(Assembler::StoreLoad);
562 __ ldrb(rscratch1, Address(byte_map_base, card_offset));
563 __ cbzw(rscratch1, done);
564
565 // storing region crossing non-null, card is clean.
566 // dirty card and log.
567 __ strb(zr, Address(byte_map_base, card_offset));
568
569 // Convert card offset into an address in card_addr
570 Register card_addr = card_offset;
571 __ add(card_addr, byte_map_base, card_addr);
572
573 __ ldr(rscratch1, queue_index);
574 __ cbz(rscratch1, runtime);
575 __ sub(rscratch1, rscratch1, wordSize);
576 __ str(rscratch1, queue_index);
577
578 // Reuse LR to hold buffer_addr
579 const Register buffer_addr = lr;
580
581 __ ldr(buffer_addr, buffer);
582 __ str(card_addr, Address(buffer_addr, rscratch1));
583 __ b(done);
584
585 __ bind(runtime);
586 __ push_call_clobbered_registers();
587 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
588 __ pop_call_clobbered_registers();
589 __ bind(done);
590 __ epilogue();
591 }
592
593 #undef __
594
595 #endif // COMPILER1