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