1 /*
2 * Copyright (c) 2018, 2024, 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 #if INCLUDE_CDS
28 #include "code/SCCache.hpp"
29 #endif
30 #include "gc/g1/g1BarrierSet.hpp"
31 #include "gc/g1/g1BarrierSetAssembler.hpp"
32 #include "gc/g1/g1BarrierSetRuntime.hpp"
33 #include "gc/g1/g1CardTable.hpp"
34 #include "gc/g1/g1HeapRegion.hpp"
35 #include "gc/g1/g1ThreadLocalData.hpp"
36 #include "gc/shared/collectedHeap.hpp"
37 #include "interpreter/interp_masm.hpp"
38 #include "runtime/javaThread.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #ifdef COMPILER1
41 #include "c1/c1_LIRAssembler.hpp"
42 #include "c1/c1_MacroAssembler.hpp"
43 #include "gc/g1/c1/g1BarrierSetC1.hpp"
44 #endif // COMPILER1
45 #ifdef COMPILER2
46 #include "gc/g1/c2/g1BarrierSetC2.hpp"
47 #endif // COMPILER2
48
49 #define __ masm->
50
51 void G1BarrierSetAssembler::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators,
52 Register addr, Register count, RegSet saved_regs) {
53 bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;
54 if (!dest_uninitialized) {
55 Label done;
56 Address in_progress(rthread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
57
58 // Is marking active?
59 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
60 __ ldrw(rscratch1, in_progress);
61 } else {
62 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
63 __ ldrb(rscratch1, in_progress);
64 }
65 __ cbzw(rscratch1, done);
66
67 __ push(saved_regs, sp);
68 if (count == c_rarg0) {
69 if (addr == c_rarg1) {
70 // exactly backwards!!
71 __ mov(rscratch1, c_rarg0);
72 __ mov(c_rarg0, c_rarg1);
73 __ mov(c_rarg1, rscratch1);
74 } else {
75 __ mov(c_rarg1, count);
76 __ mov(c_rarg0, addr);
77 }
78 } else {
79 __ mov(c_rarg0, addr);
80 __ mov(c_rarg1, count);
81 }
82 if (UseCompressedOops) {
83 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_narrow_oop_entry), 2);
84 } else {
85 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry), 2);
86 }
87 __ pop(saved_regs, sp);
88
89 __ bind(done);
90 }
91 }
92
93 void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
94 Register start, Register count, Register scratch, RegSet saved_regs) {
95 __ push(saved_regs, sp);
96 assert_different_registers(start, count, scratch);
97 assert_different_registers(c_rarg0, count);
98 __ mov(c_rarg0, start);
99 __ mov(c_rarg1, count);
100 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_post_entry), 2);
101 __ pop(saved_regs, sp);
102 }
103
104 static void generate_queue_test_and_insertion(MacroAssembler* masm, ByteSize index_offset, ByteSize buffer_offset, Label& runtime,
105 const Register thread, const Register value, const Register temp1, const Register temp2) {
106 // Can we store a value in the given thread's buffer?
107 // (The index field is typed as size_t.)
108 __ ldr(temp1, Address(thread, in_bytes(index_offset))); // temp1 := *(index address)
109 __ cbz(temp1, runtime); // jump to runtime if index == 0 (full buffer)
110 // The buffer is not full, store value into it.
111 __ sub(temp1, temp1, wordSize); // temp1 := next index
112 __ str(temp1, Address(thread, in_bytes(index_offset))); // *(index address) := next index
113 __ ldr(temp2, Address(thread, in_bytes(buffer_offset))); // temp2 := buffer address
114 __ str(value, Address(temp2, temp1)); // *(buffer address + next index) := value
115 }
116
117 static void generate_pre_barrier_fast_path(MacroAssembler* masm,
118 const Register thread,
119 const Register tmp1) {
120 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
121 // Is marking active?
122 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
123 __ ldrw(tmp1, in_progress);
124 } else {
125 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
126 __ ldrb(tmp1, in_progress);
127 }
128 }
129
130 static void generate_pre_barrier_slow_path(MacroAssembler* masm,
131 const Register obj,
132 const Register pre_val,
133 const Register thread,
134 const Register tmp1,
135 const Register tmp2,
136 Label& done,
137 Label& runtime) {
138 // Do we need to load the previous value?
139 if (obj != noreg) {
140 __ load_heap_oop(pre_val, Address(obj, 0), noreg, noreg, AS_RAW);
141 }
142 // Is the previous value null?
143 __ cbz(pre_val, done);
144 generate_queue_test_and_insertion(masm,
145 G1ThreadLocalData::satb_mark_queue_index_offset(),
146 G1ThreadLocalData::satb_mark_queue_buffer_offset(),
147 runtime,
148 thread, pre_val, tmp1, tmp2);
149 __ b(done);
150 }
151
152 void G1BarrierSetAssembler::g1_write_barrier_pre(MacroAssembler* masm,
153 Register obj,
154 Register pre_val,
155 Register thread,
156 Register tmp1,
157 Register tmp2,
158 bool tosca_live,
159 bool expand_call) {
160 // If expand_call is true then we expand the call_VM_leaf macro
161 // directly to skip generating the check by
162 // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
163
164 assert(thread == rthread, "must be");
165
166 Label done;
167 Label runtime;
168
169 assert_different_registers(obj, pre_val, tmp1, tmp2);
170 assert(pre_val != noreg && tmp1 != noreg && tmp2 != noreg, "expecting a register");
171
172 generate_pre_barrier_fast_path(masm, thread, tmp1);
173 // If marking is not active (*(mark queue active address) == 0), jump to done
174 __ cbzw(tmp1, done);
175 generate_pre_barrier_slow_path(masm, obj, pre_val, thread, tmp1, tmp2, done, runtime);
176
177 __ bind(runtime);
178
179 __ push_call_clobbered_registers();
180
181 // Calling the runtime using the regular call_VM_leaf mechanism generates
182 // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
183 // that checks that the *(rfp+frame::interpreter_frame_last_sp) == nullptr.
184 //
185 // If we care generating the pre-barrier without a frame (e.g. in the
186 // intrinsified Reference.get() routine) then rfp might be pointing to
187 // the caller frame and so this check will most likely fail at runtime.
188 //
189 // Expanding the call directly bypasses the generation of the check.
190 // So when we do not have have a full interpreter frame on the stack
191 // expand_call should be passed true.
192
193 if (expand_call) {
194 assert(pre_val != c_rarg1, "smashed arg");
195 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
196 } else {
197 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
198 }
199
200 __ pop_call_clobbered_registers();
201
202 __ bind(done);
203
204 }
205
206 static void generate_post_barrier_fast_path(MacroAssembler* masm,
207 const Register store_addr,
208 const Register new_val,
209 const Register tmp1,
210 const Register tmp2,
211 Label& done,
212 bool new_val_may_be_null) {
213 // Does store cross heap regions?
214 #if INCLUDE_CDS
215 // AOT code needs to load the barrier grain shift from the aot
216 // runtime constants area in the code cache otherwise we can compile
217 // it as an immediate operand
218 if (SCCache::is_on_for_write()) {
219 address grain_shift_address = (address)AOTRuntimeConstants::grain_shift_address();
220 __ eor(tmp1, store_addr, new_val);
221 __ lea(tmp2, ExternalAddress(grain_shift_address));
222 __ ldrb(tmp2, tmp2);
223 __ lsrv(tmp1, tmp1, tmp2);
224 __ cbz(tmp1, done);
225 } else
226 #endif
227 {
228 __ eor(tmp1, store_addr, new_val); // tmp1 := store address ^ new value
229 __ lsr(tmp1, tmp1, G1HeapRegion::LogOfHRGrainBytes); // tmp1 := ((store address ^ new value) >> LogOfHRGrainBytes)
230 __ cbz(tmp1, done);
231 }
232
233 // Crosses regions, storing null?
234 if (new_val_may_be_null) {
235 __ cbz(new_val, done);
236 }
237 // Storing region crossing non-null, is card young?
238
239 #if INCLUDE_CDS
240 // AOT code needs to load the barrier card shift from the aot
241 // runtime constants area in the code cache otherwise we can compile
242 // it as an immediate operand
243 if (SCCache::is_on_for_write()) {
244 address card_shift_address = (address)AOTRuntimeConstants::card_shift_address();
245 __ lea(tmp2, ExternalAddress(card_shift_address));
246 __ ldrb(tmp2, tmp2);
247 __ lsrv(tmp1, store_addr, tmp2); // tmp1 := card address relative to card table base
248 } else
249 #endif
250 {
251 __ lsr(tmp1, store_addr, CardTable::card_shift()); // tmp1 := card address relative to card table base
252 }
253
254 __ load_byte_map_base(tmp2); // tmp2 := card table base address
255 __ add(tmp1, tmp1, tmp2); // tmp1 := card address
256 __ ldrb(tmp2, Address(tmp1)); // tmp2 := card
257 __ cmpw(tmp2, (int)G1CardTable::g1_young_card_val()); // tmp2 := card == young_card_val?
258 }
259
260 static void generate_post_barrier_slow_path(MacroAssembler* masm,
261 const Register thread,
262 const Register tmp1,
263 const Register tmp2,
264 Label& done,
265 Label& runtime) {
266 __ membar(Assembler::StoreLoad); // StoreLoad membar
267 __ ldrb(tmp2, Address(tmp1)); // tmp2 := card
268 __ cbzw(tmp2, done);
269 // Storing a region crossing, non-null oop, card is clean.
270 // Dirty card and log.
271 STATIC_ASSERT(CardTable::dirty_card_val() == 0);
272 __ strb(zr, Address(tmp1)); // *(card address) := dirty_card_val
273 generate_queue_test_and_insertion(masm,
274 G1ThreadLocalData::dirty_card_queue_index_offset(),
275 G1ThreadLocalData::dirty_card_queue_buffer_offset(),
276 runtime,
277 thread, tmp1, tmp2, rscratch1);
278 __ b(done);
279 }
280
281 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
282 Register store_addr,
283 Register new_val,
284 Register thread,
285 Register tmp1,
286 Register tmp2) {
287 assert(thread == rthread, "must be");
288 assert_different_registers(store_addr, new_val, thread, tmp1, tmp2,
289 rscratch1);
290 assert(store_addr != noreg && new_val != noreg && tmp1 != noreg
291 && tmp2 != noreg, "expecting a register");
292
293 Label done;
294 Label runtime;
295
296 generate_post_barrier_fast_path(masm, store_addr, new_val, tmp1, tmp2, done, true /* new_val_may_be_null */);
297 // If card is young, jump to done
298 __ br(Assembler::EQ, done);
299 generate_post_barrier_slow_path(masm, thread, tmp1, tmp2, done, runtime);
300
301 __ bind(runtime);
302 // save the live input values
303 RegSet saved = RegSet::of(store_addr);
304 __ push(saved, sp);
305 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), tmp1, thread);
306 __ pop(saved, sp);
307
308 __ bind(done);
309 }
310
311 #if defined(COMPILER2)
312
313 static void generate_c2_barrier_runtime_call(MacroAssembler* masm, G1BarrierStubC2* stub, const Register arg, const address runtime_path) {
314 SaveLiveRegisters save_registers(masm, stub);
315 if (c_rarg0 != arg) {
316 __ mov(c_rarg0, arg);
317 }
318 __ mov(c_rarg1, rthread);
319 __ lea(rscratch1, RuntimeAddress(runtime_path));
320 __ blr(rscratch1);
321 }
322
323 void G1BarrierSetAssembler::g1_write_barrier_pre_c2(MacroAssembler* masm,
324 Register obj,
325 Register pre_val,
326 Register thread,
327 Register tmp1,
328 Register tmp2,
329 G1PreBarrierStubC2* stub) {
330 assert(thread == rthread, "must be");
331 assert_different_registers(obj, pre_val, tmp1, tmp2);
332 assert(pre_val != noreg && tmp1 != noreg && tmp2 != noreg, "expecting a register");
333
334 stub->initialize_registers(obj, pre_val, thread, tmp1, tmp2);
335
336 generate_pre_barrier_fast_path(masm, thread, tmp1);
337 // If marking is active (*(mark queue active address) != 0), jump to stub (slow path)
338 __ cbnzw(tmp1, *stub->entry());
339
340 __ bind(*stub->continuation());
341 }
342
343 void G1BarrierSetAssembler::generate_c2_pre_barrier_stub(MacroAssembler* masm,
344 G1PreBarrierStubC2* stub) const {
345 Assembler::InlineSkippedInstructionsCounter skip_counter(masm);
346 Label runtime;
347 Register obj = stub->obj();
348 Register pre_val = stub->pre_val();
349 Register thread = stub->thread();
350 Register tmp1 = stub->tmp1();
351 Register tmp2 = stub->tmp2();
352
353 __ bind(*stub->entry());
354 generate_pre_barrier_slow_path(masm, obj, pre_val, thread, tmp1, tmp2, *stub->continuation(), runtime);
355
356 __ bind(runtime);
357 generate_c2_barrier_runtime_call(masm, stub, pre_val, CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry));
358 __ b(*stub->continuation());
359 }
360
361 void G1BarrierSetAssembler::g1_write_barrier_post_c2(MacroAssembler* masm,
362 Register store_addr,
363 Register new_val,
364 Register thread,
365 Register tmp1,
366 Register tmp2,
367 G1PostBarrierStubC2* stub) {
368 assert(thread == rthread, "must be");
369 assert_different_registers(store_addr, new_val, thread, tmp1, tmp2,
370 rscratch1);
371 assert(store_addr != noreg && new_val != noreg && tmp1 != noreg
372 && tmp2 != noreg, "expecting a register");
373
374 stub->initialize_registers(thread, tmp1, tmp2);
375
376 bool new_val_may_be_null = (stub->barrier_data() & G1C2BarrierPostNotNull) == 0;
377 generate_post_barrier_fast_path(masm, store_addr, new_val, tmp1, tmp2, *stub->continuation(), new_val_may_be_null);
378 // If card is not young, jump to stub (slow path)
379 __ br(Assembler::NE, *stub->entry());
380
381 __ bind(*stub->continuation());
382 }
383
384 void G1BarrierSetAssembler::generate_c2_post_barrier_stub(MacroAssembler* masm,
385 G1PostBarrierStubC2* stub) const {
386 Assembler::InlineSkippedInstructionsCounter skip_counter(masm);
387 Label runtime;
388 Register thread = stub->thread();
389 Register tmp1 = stub->tmp1(); // tmp1 holds the card address.
390 Register tmp2 = stub->tmp2();
391 assert(stub->tmp3() == noreg, "not needed in this platform");
392
393 __ bind(*stub->entry());
394 generate_post_barrier_slow_path(masm, thread, tmp1, tmp2, *stub->continuation(), runtime);
395
396 __ bind(runtime);
397 generate_c2_barrier_runtime_call(masm, stub, tmp1, CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry));
398 __ b(*stub->continuation());
399 }
400
401 #endif // COMPILER2
402
403 void G1BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
404 Register dst, Address src, Register tmp1, Register tmp2) {
405 bool on_oop = is_reference_type(type);
406 bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
407 bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
408 bool on_reference = on_weak || on_phantom;
409 ModRefBarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1, tmp2);
410 if (on_oop && on_reference) {
411 // LR is live. It must be saved around calls.
412 __ enter(/*strip_ret_addr*/true); // barrier may call runtime
413 // Generate the G1 pre-barrier code to log the value of
414 // the referent field in an SATB buffer.
415 g1_write_barrier_pre(masm /* masm */,
416 noreg /* obj */,
417 dst /* pre_val */,
418 rthread /* thread */,
419 tmp1 /* tmp1 */,
420 tmp2 /* tmp2 */,
421 true /* tosca_live */,
422 true /* expand_call */);
423 __ leave();
424 }
425 }
426
427 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
428 Address dst, Register val, Register tmp1, Register tmp2, Register tmp3) {
429 // flatten object address if needed
430 if (dst.index() == noreg && dst.offset() == 0) {
431 if (dst.base() != tmp3) {
432 __ mov(tmp3, dst.base());
433 }
434 } else {
435 __ lea(tmp3, dst);
436 }
437
438 g1_write_barrier_pre(masm,
439 tmp3 /* obj */,
440 tmp2 /* pre_val */,
441 rthread /* thread */,
442 tmp1 /* tmp1 */,
443 rscratch2 /* tmp2 */,
444 val != noreg /* tosca_live */,
445 false /* expand_call */);
446
447 if (val == noreg) {
448 BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp3, 0), noreg, noreg, noreg, noreg);
449 } else {
450 // G1 barrier needs uncompressed oop for region cross check.
451 Register new_val = val;
452 if (UseCompressedOops) {
453 new_val = rscratch2;
454 __ mov(new_val, val);
455 }
456 BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp3, 0), val, noreg, noreg, noreg);
457 g1_write_barrier_post(masm,
458 tmp3 /* store_adr */,
459 new_val /* new_val */,
460 rthread /* thread */,
461 tmp1 /* tmp1 */,
462 tmp2 /* tmp2 */);
463 }
464
465 }
466
467 #ifdef COMPILER1
468
469 #undef __
470 #define __ ce->masm()->
471
472 void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
473 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
474 // At this point we know that marking is in progress.
475 // If do_load() is true then we have to emit the
476 // load of the previous value; otherwise it has already
477 // been loaded into _pre_val.
478
479 __ bind(*stub->entry());
480
481 assert(stub->pre_val()->is_register(), "Precondition.");
482
483 Register pre_val_reg = stub->pre_val()->as_register();
484
485 if (stub->do_load()) {
486 ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/);
487 }
488 __ cbz(pre_val_reg, *stub->continuation());
489 ce->store_parameter(stub->pre_val()->as_register(), 0);
490 __ far_call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
491 __ b(*stub->continuation());
492 }
493
494 void G1BarrierSetAssembler::gen_post_barrier_stub(LIR_Assembler* ce, G1PostBarrierStub* stub) {
495 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
496 __ bind(*stub->entry());
497 assert(stub->addr()->is_register(), "Precondition.");
498 assert(stub->new_val()->is_register(), "Precondition.");
499 Register new_val_reg = stub->new_val()->as_register();
500 __ cbz(new_val_reg, *stub->continuation());
501 ce->store_parameter(stub->addr()->as_pointer_register(), 0);
502 __ far_call(RuntimeAddress(bs->post_barrier_c1_runtime_code_blob()->code_begin()));
503 __ b(*stub->continuation());
504 }
505
506 #undef __
507
508 #define __ sasm->
509
510 void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
511 __ prologue("g1_pre_barrier", false);
512
513 // arg0 : previous value of memory
514
515 BarrierSet* bs = BarrierSet::barrier_set();
516
517 const Register pre_val = r0;
518 const Register thread = rthread;
519 const Register tmp = rscratch1;
520
521 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
522 Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
523 Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
524
525 Label done;
526 Label runtime;
527
528 // Is marking still active?
529 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
530 __ ldrw(tmp, in_progress);
531 } else {
532 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
533 __ ldrb(tmp, in_progress);
534 }
535 __ cbzw(tmp, done);
536
537 // Can we store original value in the thread's buffer?
538 __ ldr(tmp, queue_index);
539 __ cbz(tmp, runtime);
540
541 __ sub(tmp, tmp, wordSize);
542 __ str(tmp, queue_index);
543 __ ldr(rscratch2, buffer);
544 __ add(tmp, tmp, rscratch2);
545 __ load_parameter(0, rscratch2);
546 __ str(rscratch2, Address(tmp, 0));
547 __ b(done);
548
549 __ bind(runtime);
550 __ push_call_clobbered_registers();
551 __ load_parameter(0, pre_val);
552 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
553 __ pop_call_clobbered_registers();
554 __ bind(done);
555
556 __ epilogue();
557 }
558
559 void G1BarrierSetAssembler::generate_c1_post_barrier_runtime_stub(StubAssembler* sasm) {
560 __ prologue("g1_post_barrier", false);
561
562 // arg0: store_address
563 Address store_addr(rfp, 2*BytesPerWord);
564
565 BarrierSet* bs = BarrierSet::barrier_set();
566 CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs);
567 CardTable* ct = ctbs->card_table();
568
569 Label done;
570 Label runtime;
571
572 // At this point we know new_value is non-null and the new_value crosses regions.
573 // Must check to see if card is already dirty
574
575 const Register thread = rthread;
576
577 Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
578 Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
579
580 const Register card_offset = rscratch2;
581 // LR is free here, so we can use it to hold the byte_map_base.
582 const Register byte_map_base = lr;
583
584 assert_different_registers(card_offset, byte_map_base, rscratch1);
585
586 __ load_parameter(0, card_offset);
587 __ lsr(card_offset, card_offset, CardTable::card_shift());
588 __ load_byte_map_base(byte_map_base);
589 __ ldrb(rscratch1, Address(byte_map_base, card_offset));
590 __ cmpw(rscratch1, (int)G1CardTable::g1_young_card_val());
591 __ br(Assembler::EQ, done);
592
593 assert((int)CardTable::dirty_card_val() == 0, "must be 0");
594
595 __ membar(Assembler::StoreLoad);
596 __ ldrb(rscratch1, Address(byte_map_base, card_offset));
597 __ cbzw(rscratch1, done);
598
599 // storing region crossing non-null, card is clean.
600 // dirty card and log.
601 __ strb(zr, Address(byte_map_base, card_offset));
602
603 // Convert card offset into an address in card_addr
604 Register card_addr = card_offset;
605 __ add(card_addr, byte_map_base, card_addr);
606
607 __ ldr(rscratch1, queue_index);
608 __ cbz(rscratch1, runtime);
609 __ sub(rscratch1, rscratch1, wordSize);
610 __ str(rscratch1, queue_index);
611
612 // Reuse LR to hold buffer_addr
613 const Register buffer_addr = lr;
614
615 __ ldr(buffer_addr, buffer);
616 __ str(card_addr, Address(buffer_addr, rscratch1));
617 __ b(done);
618
619 __ bind(runtime);
620 __ push_call_clobbered_registers();
621 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
622 __ pop_call_clobbered_registers();
623 __ bind(done);
624 __ epilogue();
625 }
626
627 #undef __
628
629 #endif // COMPILER1