1 /*
2 * Copyright (c) 2018, 2025, 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 "asm/macroAssembler.inline.hpp"
26 #include "gc/g1/g1BarrierSet.hpp"
27 #include "gc/g1/g1BarrierSetAssembler.hpp"
28 #include "gc/g1/g1BarrierSetRuntime.hpp"
29 #include "gc/g1/g1CardTable.hpp"
30 #include "gc/g1/g1HeapRegion.hpp"
31 #include "gc/g1/g1ThreadLocalData.hpp"
32 #include "interpreter/interp_masm.hpp"
33 #include "runtime/sharedRuntime.hpp"
34 #include "utilities/debug.hpp"
35 #include "utilities/macros.hpp"
36 #ifdef COMPILER1
37 #include "c1/c1_LIRAssembler.hpp"
38 #include "c1/c1_MacroAssembler.hpp"
39 #include "gc/g1/c1/g1BarrierSetC1.hpp"
40 #endif // COMPILER1
41 #ifdef COMPILER2
42 #include "gc/g1/c2/g1BarrierSetC2.hpp"
43 #endif // COMPILER2
44
45 #define __ masm->
46
47 void G1BarrierSetAssembler::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators,
48 Register addr, Register count) {
49 bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;
50
51 if (!dest_uninitialized) {
52 Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
53 #ifndef _LP64
54 __ push(thread);
55 __ get_thread(thread);
56 #endif
57
58 Label filtered;
59 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
60 // Is marking active?
61 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
62 __ cmpl(in_progress, 0);
63 } else {
64 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
65 __ cmpb(in_progress, 0);
66 }
67
68 NOT_LP64(__ pop(thread);)
69
70 __ jcc(Assembler::equal, filtered);
71
72 __ push_call_clobbered_registers(false /* save_fpu */);
73 #ifdef _LP64
74 if (count == c_rarg0) {
75 if (addr == c_rarg1) {
76 // exactly backwards!!
77 __ xchgptr(c_rarg1, c_rarg0);
78 } else {
79 __ movptr(c_rarg1, count);
80 __ movptr(c_rarg0, addr);
81 }
82 } else {
83 __ movptr(c_rarg0, addr);
84 __ movptr(c_rarg1, count);
85 }
86 if (UseCompressedOops) {
87 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_narrow_oop_entry), 2);
88 } else {
89 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry), 2);
90 }
91 #else
92 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry),
93 addr, count);
94 #endif
95 __ pop_call_clobbered_registers(false /* save_fpu */);
96
97 __ bind(filtered);
98 }
99 }
100
101 void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
102 Register addr, Register count, Register tmp) {
103 __ push_call_clobbered_registers(false /* save_fpu */);
104 #ifdef _LP64
105 if (c_rarg0 == count) { // On win64 c_rarg0 == rcx
106 assert_different_registers(c_rarg1, addr);
107 __ mov(c_rarg1, count);
108 __ mov(c_rarg0, addr);
109 } else {
110 assert_different_registers(c_rarg0, count);
111 __ mov(c_rarg0, addr);
112 __ mov(c_rarg1, count);
113 }
114 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_post_entry), 2);
115 #else
116 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_post_entry),
117 addr, count);
118 #endif
119 __ pop_call_clobbered_registers(false /* save_fpu */);
120 }
121
122 void G1BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
123 Register dst, Address src, Register tmp1, Register tmp_thread) {
124 bool on_oop = is_reference_type(type);
125 bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
126 bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
127 bool on_reference = on_weak || on_phantom;
128 ModRefBarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1, tmp_thread);
129 if (on_oop && on_reference) {
130 Register thread = NOT_LP64(tmp_thread) LP64_ONLY(r15_thread);
131
132 #ifndef _LP64
133 // Work around the x86_32 bug that only manifests with Loom for some reason.
134 // MacroAssembler::resolve_weak_handle calls this barrier with tmp_thread == noreg.
135 if (thread == noreg) {
136 if (dst != rcx && tmp1 != rcx) {
137 thread = rcx;
138 } else if (dst != rdx && tmp1 != rdx) {
139 thread = rdx;
140 } else if (dst != rdi && tmp1 != rdi) {
141 thread = rdi;
142 }
143 }
144 assert_different_registers(dst, tmp1, thread);
145 __ push(thread);
146 __ get_thread(thread);
147 #endif
148
149 // Generate the G1 pre-barrier code to log the value of
150 // the referent field in an SATB buffer.
151 g1_write_barrier_pre(masm /* masm */,
152 noreg /* obj */,
153 dst /* pre_val */,
154 thread /* thread */,
155 tmp1 /* tmp */,
156 true /* tosca_live */,
157 true /* expand_call */);
158
159 #ifndef _LP64
160 __ pop(thread);
161 #endif
162 }
163 }
164
165 static void generate_queue_insertion(MacroAssembler* masm, ByteSize index_offset, ByteSize buffer_offset, Label& runtime,
166 const Register thread, const Register value, const Register temp) {
167 // This code assumes that buffer index is pointer sized.
168 STATIC_ASSERT(in_bytes(SATBMarkQueue::byte_width_of_index()) == sizeof(intptr_t));
169 // Can we store a value in the given thread's buffer?
170 // (The index field is typed as size_t.)
171 __ movptr(temp, Address(thread, in_bytes(index_offset))); // temp := *(index address)
172 __ testptr(temp, temp); // index == 0?
173 __ jcc(Assembler::zero, runtime); // jump to runtime if index == 0 (full buffer)
174 // The buffer is not full, store value into it.
175 __ subptr(temp, wordSize); // temp := next index
176 __ movptr(Address(thread, in_bytes(index_offset)), temp); // *(index address) := next index
177 __ addptr(temp, Address(thread, in_bytes(buffer_offset))); // temp := buffer address + next index
178 __ movptr(Address(temp, 0), value); // *(buffer address + next index) := value
179 }
180
181 static void generate_pre_barrier_fast_path(MacroAssembler* masm,
182 const Register thread) {
183 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
184 // Is marking active?
185 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
186 __ cmpl(in_progress, 0);
187 } else {
188 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
189 __ cmpb(in_progress, 0);
190 }
191 }
192
193 static void generate_pre_barrier_slow_path(MacroAssembler* masm,
194 const Register obj,
195 const Register pre_val,
196 const Register thread,
197 const Register tmp,
198 Label& done,
199 Label& runtime) {
200 // Do we need to load the previous value?
201 if (obj != noreg) {
202 __ load_heap_oop(pre_val, Address(obj, 0), noreg, noreg, AS_RAW);
203 }
204 // Is the previous value null?
205 __ cmpptr(pre_val, NULL_WORD);
206 __ jcc(Assembler::equal, done);
207 generate_queue_insertion(masm,
208 G1ThreadLocalData::satb_mark_queue_index_offset(),
209 G1ThreadLocalData::satb_mark_queue_buffer_offset(),
210 runtime,
211 thread, pre_val, tmp);
212 __ jmp(done);
213 }
214
215 void G1BarrierSetAssembler::g1_write_barrier_pre(MacroAssembler* masm,
216 Register obj,
217 Register pre_val,
218 Register thread,
219 Register tmp,
220 bool tosca_live,
221 bool expand_call) {
222 // If expand_call is true then we expand the call_VM_leaf macro
223 // directly to skip generating the check by
224 // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
225
226 #ifdef _LP64
227 assert(thread == r15_thread, "must be");
228 #endif // _LP64
229
230 Label done;
231 Label runtime;
232
233 assert(pre_val != noreg, "check this code");
234
235 if (obj != noreg) {
236 assert_different_registers(obj, pre_val, tmp);
237 assert(pre_val != rax, "check this code");
238 }
239
240 generate_pre_barrier_fast_path(masm, thread);
241 // If marking is not active (*(mark queue active address) == 0), jump to done
242 __ jcc(Assembler::equal, done);
243 generate_pre_barrier_slow_path(masm, obj, pre_val, thread, tmp, done, runtime);
244
245 __ bind(runtime);
246
247 // Determine and save the live input values
248 __ push_call_clobbered_registers();
249
250 // Calling the runtime using the regular call_VM_leaf mechanism generates
251 // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
252 // that checks that the *(ebp+frame::interpreter_frame_last_sp) == nullptr.
253 //
254 // If we care generating the pre-barrier without a frame (e.g. in the
255 // intrinsified Reference.get() routine) then ebp might be pointing to
256 // the caller frame and so this check will most likely fail at runtime.
257 //
258 // Expanding the call directly bypasses the generation of the check.
259 // So when we do not have have a full interpreter frame on the stack
260 // expand_call should be passed true.
261
262 if (expand_call) {
263 LP64_ONLY( assert(pre_val != c_rarg1, "smashed arg"); )
264 #ifdef _LP64
265 if (c_rarg1 != thread) {
266 __ mov(c_rarg1, thread);
267 }
268 if (c_rarg0 != pre_val) {
269 __ mov(c_rarg0, pre_val);
270 }
271 #else
272 __ push(thread);
273 __ push(pre_val);
274 #endif
275 __ MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), 2);
276 } else {
277 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
278 }
279
280 __ pop_call_clobbered_registers();
281
282 __ bind(done);
283 }
284
285 static void generate_post_barrier_fast_path(MacroAssembler* masm,
286 const Register store_addr,
287 const Register new_val,
288 const Register tmp,
289 const Register tmp2,
290 Label& done,
291 bool new_val_may_be_null) {
292 CardTableBarrierSet* ct = barrier_set_cast<CardTableBarrierSet>(BarrierSet::barrier_set());
293 // Does store cross heap regions?
294 __ movptr(tmp, store_addr); // tmp := store address
295 __ xorptr(tmp, new_val); // tmp := store address ^ new value
296 __ shrptr(tmp, G1HeapRegion::LogOfHRGrainBytes); // ((store address ^ new value) >> LogOfHRGrainBytes) == 0?
297 __ jcc(Assembler::equal, done);
298 // Crosses regions, storing null?
299 if (new_val_may_be_null) {
300 __ cmpptr(new_val, NULL_WORD); // new value == null?
301 __ jcc(Assembler::equal, done);
302 }
303 // Storing region crossing non-null, is card young?
304 __ movptr(tmp, store_addr); // tmp := store address
305 __ shrptr(tmp, CardTable::card_shift()); // tmp := card address relative to card table base
306 // Do not use ExternalAddress to load 'byte_map_base', since 'byte_map_base' is NOT
307 // a valid address and therefore is not properly handled by the relocation code.
308 __ movptr(tmp2, (intptr_t)ct->card_table()->byte_map_base()); // tmp2 := card table base address
309 __ addptr(tmp, tmp2); // tmp := card address
310 __ cmpb(Address(tmp, 0), G1CardTable::g1_young_card_val()); // *(card address) == young_card_val?
311 }
312
313 static void generate_post_barrier_slow_path(MacroAssembler* masm,
314 const Register thread,
315 const Register tmp,
316 const Register tmp2,
317 Label& done,
318 Label& runtime) {
319 __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad)); // StoreLoad membar
320 __ cmpb(Address(tmp, 0), G1CardTable::dirty_card_val()); // *(card address) == dirty_card_val?
321 __ jcc(Assembler::equal, done);
322 // Storing a region crossing, non-null oop, card is clean.
323 // Dirty card and log.
324 __ movb(Address(tmp, 0), G1CardTable::dirty_card_val()); // *(card address) := dirty_card_val
325 generate_queue_insertion(masm,
326 G1ThreadLocalData::dirty_card_queue_index_offset(),
327 G1ThreadLocalData::dirty_card_queue_buffer_offset(),
328 runtime,
329 thread, tmp, tmp2);
330 __ jmp(done);
331 }
332
333 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
334 Register store_addr,
335 Register new_val,
336 Register thread,
337 Register tmp,
338 Register tmp2) {
339 #ifdef _LP64
340 assert(thread == r15_thread, "must be");
341 #endif // _LP64
342
343 Label done;
344 Label runtime;
345
346 generate_post_barrier_fast_path(masm, store_addr, new_val, tmp, tmp2, done, true /* new_val_may_be_null */);
347 // If card is young, jump to done
348 __ jcc(Assembler::equal, done);
349 generate_post_barrier_slow_path(masm, thread, tmp, tmp2, done, runtime);
350
351 __ bind(runtime);
352 // save the live input values
353 RegSet saved = RegSet::of(store_addr NOT_LP64(COMMA thread));
354 __ push_set(saved);
355 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), tmp, thread);
356 __ pop_set(saved);
357
358 __ bind(done);
359 }
360
361 #if defined(COMPILER2)
362
363 static void generate_c2_barrier_runtime_call(MacroAssembler* masm, G1BarrierStubC2* stub, const Register arg, const address runtime_path) {
364 #ifdef _LP64
365 SaveLiveRegisters save_registers(masm, stub);
366 if (c_rarg0 != arg) {
367 __ mov(c_rarg0, arg);
368 }
369 __ mov(c_rarg1, r15_thread);
370 // rax is a caller-saved, non-argument-passing register, so it does not
371 // interfere with c_rarg0 or c_rarg1. If it contained any live value before
372 // entering this stub, it is saved at this point, and restored after the
373 // call. If it did not contain any live value, it is free to be used. In
374 // either case, it is safe to use it here as a call scratch register.
375 __ call(RuntimeAddress(runtime_path), rax);
376 #else
377 Unimplemented();
378 #endif // _LP64
379 }
380
381 void G1BarrierSetAssembler::g1_write_barrier_pre_c2(MacroAssembler* masm,
382 Register obj,
383 Register pre_val,
384 Register thread,
385 Register tmp,
386 G1PreBarrierStubC2* stub) {
387 #ifdef _LP64
388 assert(thread == r15_thread, "must be");
389 #endif // _LP64
390 assert(pre_val != noreg, "check this code");
391 if (obj != noreg) {
392 assert_different_registers(obj, pre_val, tmp);
393 }
394
395 stub->initialize_registers(obj, pre_val, thread, tmp);
396
397 generate_pre_barrier_fast_path(masm, thread);
398 // If marking is active (*(mark queue active address) != 0), jump to stub (slow path)
399 __ jcc(Assembler::notEqual, *stub->entry());
400
401 __ bind(*stub->continuation());
402 }
403
404 void G1BarrierSetAssembler::generate_c2_pre_barrier_stub(MacroAssembler* masm,
405 G1PreBarrierStubC2* stub) const {
406 Assembler::InlineSkippedInstructionsCounter skip_counter(masm);
407 Label runtime;
408 Register obj = stub->obj();
409 Register pre_val = stub->pre_val();
410 Register thread = stub->thread();
411 Register tmp = stub->tmp1();
412 assert(stub->tmp2() == noreg, "not needed in this platform");
413
414 __ bind(*stub->entry());
415 generate_pre_barrier_slow_path(masm, obj, pre_val, thread, tmp, *stub->continuation(), runtime);
416
417 __ bind(runtime);
418 generate_c2_barrier_runtime_call(masm, stub, pre_val, CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry));
419 __ jmp(*stub->continuation());
420 }
421
422 void G1BarrierSetAssembler::g1_write_barrier_post_c2(MacroAssembler* masm,
423 Register store_addr,
424 Register new_val,
425 Register thread,
426 Register tmp,
427 Register tmp2,
428 G1PostBarrierStubC2* stub) {
429 #ifdef _LP64
430 assert(thread == r15_thread, "must be");
431 #endif // _LP64
432
433 stub->initialize_registers(thread, tmp, tmp2);
434
435 bool new_val_may_be_null = (stub->barrier_data() & G1C2BarrierPostNotNull) == 0;
436 generate_post_barrier_fast_path(masm, store_addr, new_val, tmp, tmp2, *stub->continuation(), new_val_may_be_null);
437 // If card is not young, jump to stub (slow path)
438 __ jcc(Assembler::notEqual, *stub->entry());
439
440 __ bind(*stub->continuation());
441 }
442
443 void G1BarrierSetAssembler::generate_c2_post_barrier_stub(MacroAssembler* masm,
444 G1PostBarrierStubC2* stub) const {
445 Assembler::InlineSkippedInstructionsCounter skip_counter(masm);
446 Label runtime;
447 Register thread = stub->thread();
448 Register tmp = stub->tmp1(); // tmp holds the card address.
449 Register tmp2 = stub->tmp2();
450 assert(stub->tmp3() == noreg, "not needed in this platform");
451
452 __ bind(*stub->entry());
453 generate_post_barrier_slow_path(masm, thread, tmp, tmp2, *stub->continuation(), runtime);
454
455 __ bind(runtime);
456 generate_c2_barrier_runtime_call(masm, stub, tmp, CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry));
457 __ jmp(*stub->continuation());
458 }
459
460 #endif // COMPILER2
461
462 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
463 Address dst, Register val, Register tmp1, Register tmp2, Register tmp3) {
464 bool in_heap = (decorators & IN_HEAP) != 0;
465 bool as_normal = (decorators & AS_NORMAL) != 0;
466
467 bool needs_pre_barrier = as_normal;
468 bool needs_post_barrier = val != noreg && in_heap;
469
470 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
471 // flatten object address if needed
472 // We do it regardless of precise because we need the registers
473 if (dst.index() == noreg && dst.disp() == 0) {
474 if (dst.base() != tmp1) {
475 __ movptr(tmp1, dst.base());
476 }
477 } else {
478 __ lea(tmp1, dst);
479 }
480
481 #ifndef _LP64
482 InterpreterMacroAssembler *imasm = static_cast<InterpreterMacroAssembler*>(masm);
483 #endif
484
485 NOT_LP64(__ get_thread(rcx));
486 NOT_LP64(imasm->save_bcp());
487
488 if (needs_pre_barrier) {
489 g1_write_barrier_pre(masm /*masm*/,
490 tmp1 /* obj */,
491 tmp2 /* pre_val */,
492 rthread /* thread */,
493 tmp3 /* tmp */,
494 val != noreg /* tosca_live */,
495 false /* expand_call */);
496 }
497 if (val == noreg) {
498 BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg, noreg);
499 } else {
500 Register new_val = val;
501 if (needs_post_barrier) {
502 // G1 barrier needs uncompressed oop for region cross check.
503 if (UseCompressedOops) {
504 new_val = tmp2;
505 __ movptr(new_val, val);
506 }
507 }
508 BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg, noreg);
509 if (needs_post_barrier) {
510 g1_write_barrier_post(masm /*masm*/,
511 tmp1 /* store_adr */,
512 new_val /* new_val */,
513 rthread /* thread */,
514 tmp3 /* tmp */,
515 tmp2 /* tmp2 */);
516 }
517 }
518 NOT_LP64(imasm->restore_bcp());
519 }
520
521 #ifdef COMPILER1
522
523 #undef __
524 #define __ ce->masm()->
525
526 void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
527 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
528 // At this point we know that marking is in progress.
529 // If do_load() is true then we have to emit the
530 // load of the previous value; otherwise it has already
531 // been loaded into _pre_val.
532
533 __ bind(*stub->entry());
534 assert(stub->pre_val()->is_register(), "Precondition.");
535
536 Register pre_val_reg = stub->pre_val()->as_register();
537
538 if (stub->do_load()) {
539 ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/);
540 }
541
542 __ cmpptr(pre_val_reg, NULL_WORD);
543 __ jcc(Assembler::equal, *stub->continuation());
544 ce->store_parameter(stub->pre_val()->as_register(), 0);
545 __ call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
546 __ jmp(*stub->continuation());
547
548 }
549
550 void G1BarrierSetAssembler::gen_post_barrier_stub(LIR_Assembler* ce, G1PostBarrierStub* stub) {
551 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
552 __ bind(*stub->entry());
553 assert(stub->addr()->is_register(), "Precondition.");
554 assert(stub->new_val()->is_register(), "Precondition.");
555 Register new_val_reg = stub->new_val()->as_register();
556 __ cmpptr(new_val_reg, NULL_WORD);
557 __ jcc(Assembler::equal, *stub->continuation());
558 ce->store_parameter(stub->addr()->as_pointer_register(), 0);
559 __ call(RuntimeAddress(bs->post_barrier_c1_runtime_code_blob()->code_begin()));
560 __ jmp(*stub->continuation());
561 }
562
563 #undef __
564
565 #define __ sasm->
566
567 void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
568 // Generated code assumes that buffer index is pointer sized.
569 STATIC_ASSERT(in_bytes(SATBMarkQueue::byte_width_of_index()) == sizeof(intptr_t));
570
571 __ prologue("g1_pre_barrier", false);
572 // arg0 : previous value of memory
573
574 __ push(rax);
575 __ push(rdx);
576
577 const Register pre_val = rax;
578 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
579 const Register tmp = rdx;
580
581 NOT_LP64(__ get_thread(thread);)
582
583 Address queue_active(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
584 Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
585 Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
586
587 Label done;
588 Label runtime;
589
590 // Is marking still active?
591 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
592 __ cmpl(queue_active, 0);
593 } else {
594 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
595 __ cmpb(queue_active, 0);
596 }
597 __ jcc(Assembler::equal, done);
598
599 // Can we store original value in the thread's buffer?
600
601 __ movptr(tmp, queue_index);
602 __ testptr(tmp, tmp);
603 __ jcc(Assembler::zero, runtime);
604 __ subptr(tmp, wordSize);
605 __ movptr(queue_index, tmp);
606 __ addptr(tmp, buffer);
607
608 // prev_val (rax)
609 __ load_parameter(0, pre_val);
610 __ movptr(Address(tmp, 0), pre_val);
611 __ jmp(done);
612
613 __ bind(runtime);
614
615 __ push_call_clobbered_registers();
616
617 // load the pre-value
618 __ load_parameter(0, rcx);
619 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), rcx, thread);
620
621 __ pop_call_clobbered_registers();
622
623 __ bind(done);
624
625 __ pop(rdx);
626 __ pop(rax);
627
628 __ epilogue();
629 }
630
631 void G1BarrierSetAssembler::generate_c1_post_barrier_runtime_stub(StubAssembler* sasm) {
632 __ prologue("g1_post_barrier", false);
633
634 CardTableBarrierSet* ct =
635 barrier_set_cast<CardTableBarrierSet>(BarrierSet::barrier_set());
636
637 Label done;
638 Label enqueued;
639 Label runtime;
640
641 // At this point we know new_value is non-null and the new_value crosses regions.
642 // Must check to see if card is already dirty
643
644 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
645
646 Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
647 Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
648
649 __ push(rax);
650 __ push(rcx);
651
652 const Register cardtable = rax;
653 const Register card_addr = rcx;
654
655 __ load_parameter(0, card_addr);
656 __ shrptr(card_addr, CardTable::card_shift());
657 // Do not use ExternalAddress to load 'byte_map_base', since 'byte_map_base' is NOT
658 // a valid address and therefore is not properly handled by the relocation code.
659 __ movptr(cardtable, (intptr_t)ct->card_table()->byte_map_base());
660 __ addptr(card_addr, cardtable);
661
662 NOT_LP64(__ get_thread(thread);)
663
664 __ cmpb(Address(card_addr, 0), G1CardTable::g1_young_card_val());
665 __ jcc(Assembler::equal, done);
666
667 __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
668 __ cmpb(Address(card_addr, 0), CardTable::dirty_card_val());
669 __ jcc(Assembler::equal, done);
670
671 // storing region crossing non-null, card is clean.
672 // dirty card and log.
673
674 __ movb(Address(card_addr, 0), CardTable::dirty_card_val());
675
676 const Register tmp = rdx;
677 __ push(rdx);
678
679 __ movptr(tmp, queue_index);
680 __ testptr(tmp, tmp);
681 __ jcc(Assembler::zero, runtime);
682 __ subptr(tmp, wordSize);
683 __ movptr(queue_index, tmp);
684 __ addptr(tmp, buffer);
685 __ movptr(Address(tmp, 0), card_addr);
686 __ jmp(enqueued);
687
688 __ bind(runtime);
689 __ push_call_clobbered_registers();
690
691 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
692
693 __ pop_call_clobbered_registers();
694
695 __ bind(enqueued);
696 __ pop(rdx);
697
698 __ bind(done);
699 __ pop(rcx);
700 __ pop(rax);
701
702 __ epilogue();
703 }
704
705 #undef __
706
707 #endif // COMPILER1