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 "code/SCCache.hpp"
28 #include "gc/g1/g1BarrierSet.hpp"
29 #include "gc/g1/g1BarrierSetAssembler.hpp"
30 #include "gc/g1/g1BarrierSetRuntime.hpp"
31 #include "gc/g1/g1CardTable.hpp"
32 #include "gc/g1/g1HeapRegion.hpp"
33 #include "gc/g1/g1ThreadLocalData.hpp"
34 #include "interpreter/interp_masm.hpp"
35 #include "runtime/sharedRuntime.hpp"
36 #include "utilities/debug.hpp"
37 #include "utilities/macros.hpp"
38 #ifdef COMPILER1
39 #include "c1/c1_LIRAssembler.hpp"
40 #include "c1/c1_MacroAssembler.hpp"
41 #include "gc/g1/c1/g1BarrierSetC1.hpp"
42 #endif
43
44 #define __ masm->
45
46 void G1BarrierSetAssembler::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators,
47 Register addr, Register count) {
48 bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;
49
50 if (!dest_uninitialized) {
51 Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
52 #ifndef _LP64
53 __ push(thread);
54 __ get_thread(thread);
55 #endif
56
57 Label filtered;
58 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
59 // Is marking active?
60 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
61 __ cmpl(in_progress, 0);
62 } else {
63 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
64 __ cmpb(in_progress, 0);
65 }
66
67 NOT_LP64(__ pop(thread);)
68
69 __ jcc(Assembler::equal, filtered);
70
71 __ push_call_clobbered_registers(false /* save_fpu */);
72 #ifdef _LP64
73 if (count == c_rarg0) {
74 if (addr == c_rarg1) {
75 // exactly backwards!!
76 __ xchgptr(c_rarg1, c_rarg0);
77 } else {
78 __ movptr(c_rarg1, count);
79 __ movptr(c_rarg0, addr);
80 }
81 } else {
82 __ movptr(c_rarg0, addr);
83 __ movptr(c_rarg1, count);
84 }
85 if (UseCompressedOops) {
86 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_narrow_oop_entry), 2);
87 } else {
88 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry), 2);
89 }
90 #else
91 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry),
92 addr, count);
93 #endif
94 __ pop_call_clobbered_registers(false /* save_fpu */);
95
96 __ bind(filtered);
97 }
98 }
99
100 void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
101 Register addr, Register count, Register tmp) {
102 __ push_call_clobbered_registers(false /* save_fpu */);
103 #ifdef _LP64
104 if (c_rarg0 == count) { // On win64 c_rarg0 == rcx
105 assert_different_registers(c_rarg1, addr);
106 __ mov(c_rarg1, count);
107 __ mov(c_rarg0, addr);
108 } else {
109 assert_different_registers(c_rarg0, count);
110 __ mov(c_rarg0, addr);
111 __ mov(c_rarg1, count);
112 }
113 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_post_entry), 2);
114 #else
115 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_post_entry),
116 addr, count);
117 #endif
118 __ pop_call_clobbered_registers(false /* save_fpu */);
119 }
120
121 void G1BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
122 Register dst, Address src, Register tmp1, Register tmp_thread) {
123 bool on_oop = is_reference_type(type);
124 bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
125 bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
126 bool on_reference = on_weak || on_phantom;
127 ModRefBarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1, tmp_thread);
128 if (on_oop && on_reference) {
129 Register thread = NOT_LP64(tmp_thread) LP64_ONLY(r15_thread);
130
131 #ifndef _LP64
132 // Work around the x86_32 bug that only manifests with Loom for some reason.
133 // MacroAssembler::resolve_weak_handle calls this barrier with tmp_thread == noreg.
134 if (thread == noreg) {
135 if (dst != rcx && tmp1 != rcx) {
136 thread = rcx;
137 } else if (dst != rdx && tmp1 != rdx) {
138 thread = rdx;
139 } else if (dst != rdi && tmp1 != rdi) {
140 thread = rdi;
141 }
142 }
143 assert_different_registers(dst, tmp1, thread);
144 __ push(thread);
145 __ get_thread(thread);
146 #endif
147
148 // Generate the G1 pre-barrier code to log the value of
149 // the referent field in an SATB buffer.
150 g1_write_barrier_pre(masm /* masm */,
151 noreg /* obj */,
152 dst /* pre_val */,
153 thread /* thread */,
154 tmp1 /* tmp */,
155 true /* tosca_live */,
156 true /* expand_call */);
157
158 #ifndef _LP64
159 __ pop(thread);
160 #endif
161 }
162 }
163
164 void G1BarrierSetAssembler::g1_write_barrier_pre(MacroAssembler* masm,
165 Register obj,
166 Register pre_val,
167 Register thread,
168 Register tmp,
169 bool tosca_live,
170 bool expand_call) {
171 // If expand_call is true then we expand the call_VM_leaf macro
172 // directly to skip generating the check by
173 // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
174
175 #ifdef _LP64
176 assert(thread == r15_thread, "must be");
177 #endif // _LP64
178
179 Label done;
180 Label runtime;
181
182 assert(pre_val != noreg, "check this code");
183
184 if (obj != noreg) {
185 assert_different_registers(obj, pre_val, tmp);
186 assert(pre_val != rax, "check this code");
187 }
188
189 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
190 Address index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
191 Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
192
193 // Is marking active?
194 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
195 __ cmpl(in_progress, 0);
196 } else {
197 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
198 __ cmpb(in_progress, 0);
199 }
200 __ jcc(Assembler::equal, done);
201
202 // Do we need to load the previous value?
203 if (obj != noreg) {
204 __ load_heap_oop(pre_val, Address(obj, 0), noreg, noreg, AS_RAW);
205 }
206
207 // Is the previous value null?
208 __ cmpptr(pre_val, NULL_WORD);
209 __ jcc(Assembler::equal, done);
210
211 // Can we store original value in the thread's buffer?
212 // Is index == 0?
213 // (The index field is typed as size_t.)
214
215 __ movptr(tmp, index); // tmp := *index_adr
216 __ cmpptr(tmp, 0); // tmp == 0?
217 __ jcc(Assembler::equal, runtime); // If yes, goto runtime
218
219 __ subptr(tmp, wordSize); // tmp := tmp - wordSize
220 __ movptr(index, tmp); // *index_adr := tmp
221 __ addptr(tmp, buffer); // tmp := tmp + *buffer_adr
222
223 // Record the previous value
224 __ movptr(Address(tmp, 0), pre_val);
225 __ jmp(done);
226
227 __ bind(runtime);
228
229 // Determine and save the live input values
230 __ push_call_clobbered_registers();
231
232 // Calling the runtime using the regular call_VM_leaf mechanism generates
233 // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
234 // that checks that the *(ebp+frame::interpreter_frame_last_sp) == nullptr.
235 //
236 // If we care generating the pre-barrier without a frame (e.g. in the
237 // intrinsified Reference.get() routine) then ebp might be pointing to
238 // the caller frame and so this check will most likely fail at runtime.
239 //
240 // Expanding the call directly bypasses the generation of the check.
241 // So when we do not have have a full interpreter frame on the stack
242 // expand_call should be passed true.
243
244 if (expand_call) {
245 LP64_ONLY( assert(pre_val != c_rarg1, "smashed arg"); )
246 #ifdef _LP64
247 if (c_rarg1 != thread) {
248 __ mov(c_rarg1, thread);
249 }
250 if (c_rarg0 != pre_val) {
251 __ mov(c_rarg0, pre_val);
252 }
253 #else
254 __ push(thread);
255 __ push(pre_val);
256 #endif
257 __ MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), 2);
258 } else {
259 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
260 }
261
262 __ pop_call_clobbered_registers();
263
264 __ bind(done);
265 }
266
267 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
268 Register store_addr,
269 Register new_val,
270 Register thread,
271 Register tmp,
272 Register tmp2) {
273 #ifdef _LP64
274 assert(thread == r15_thread, "must be");
275 #endif // _LP64
276
277 Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
278 Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
279
280 CardTableBarrierSet* ct =
281 barrier_set_cast<CardTableBarrierSet>(BarrierSet::barrier_set());
282
283 Label done;
284 Label runtime;
285
286 // Does store cross heap regions?
287
288 __ movptr(tmp, store_addr);
289 __ xorptr(tmp, new_val);
290 __ shrptr(tmp, HeapRegion::LogOfHRGrainBytes);
291 __ jcc(Assembler::equal, done);
292
293 // crosses regions, storing null?
294
295 __ cmpptr(new_val, NULL_WORD);
296 __ jcc(Assembler::equal, done);
297
298 // storing region crossing non-null, is card already dirty?
299
300 const Register card_addr = tmp;
301 const Register cardtable = tmp2;
302
303 __ movptr(card_addr, store_addr);
304 __ shrptr(card_addr, CardTable::card_shift());
305 // Do not use ExternalAddress to load 'byte_map_base', since 'byte_map_base' is NOT
306 // a valid address and therefore is not properly handled by the relocation code.
307 if (SCCache::is_on_for_write()) {
308 // SCA needs relocation info for this address
309 __ lea(cardtable, ExternalAddress((address)ct->card_table()->byte_map_base()));
310 } else {
311 __ movptr(cardtable, (intptr_t)ct->card_table()->byte_map_base());
312 }
313 __ addptr(card_addr, cardtable);
314
315 __ cmpb(Address(card_addr, 0), G1CardTable::g1_young_card_val());
316 __ jcc(Assembler::equal, done);
317
318 __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
319 __ cmpb(Address(card_addr, 0), G1CardTable::dirty_card_val());
320 __ jcc(Assembler::equal, done);
321
322
323 // storing a region crossing, non-null oop, card is clean.
324 // dirty card and log.
325
326 __ movb(Address(card_addr, 0), G1CardTable::dirty_card_val());
327
328 // The code below assumes that buffer index is pointer sized.
329 STATIC_ASSERT(in_bytes(G1DirtyCardQueue::byte_width_of_index()) == sizeof(intptr_t));
330
331 __ movptr(tmp2, queue_index);
332 __ testptr(tmp2, tmp2);
333 __ jcc(Assembler::zero, runtime);
334 __ subptr(tmp2, wordSize);
335 __ movptr(queue_index, tmp2);
336 __ addptr(tmp2, buffer);
337 __ movptr(Address(tmp2, 0), card_addr);
338 __ jmp(done);
339
340 __ bind(runtime);
341 // save the live input values
342 RegSet saved = RegSet::of(store_addr NOT_LP64(COMMA thread));
343 __ push_set(saved);
344 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
345 __ pop_set(saved);
346
347 __ bind(done);
348 }
349
350 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
351 Address dst, Register val, Register tmp1, Register tmp2, Register tmp3) {
352 bool in_heap = (decorators & IN_HEAP) != 0;
353 bool as_normal = (decorators & AS_NORMAL) != 0;
354
355 bool needs_pre_barrier = as_normal;
356 bool needs_post_barrier = val != noreg && in_heap;
357
358 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
359 // flatten object address if needed
360 // We do it regardless of precise because we need the registers
361 if (dst.index() == noreg && dst.disp() == 0) {
362 if (dst.base() != tmp1) {
363 __ movptr(tmp1, dst.base());
364 }
365 } else {
366 __ lea(tmp1, dst);
367 }
368
369 #ifndef _LP64
370 InterpreterMacroAssembler *imasm = static_cast<InterpreterMacroAssembler*>(masm);
371 #endif
372
373 NOT_LP64(__ get_thread(rcx));
374 NOT_LP64(imasm->save_bcp());
375
376 if (needs_pre_barrier) {
377 g1_write_barrier_pre(masm /*masm*/,
378 tmp1 /* obj */,
379 tmp2 /* pre_val */,
380 rthread /* thread */,
381 tmp3 /* tmp */,
382 val != noreg /* tosca_live */,
383 false /* expand_call */);
384 }
385 if (val == noreg) {
386 BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg, noreg);
387 } else {
388 Register new_val = val;
389 if (needs_post_barrier) {
390 // G1 barrier needs uncompressed oop for region cross check.
391 if (UseCompressedOops) {
392 new_val = tmp2;
393 __ movptr(new_val, val);
394 }
395 }
396 BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg, noreg);
397 if (needs_post_barrier) {
398 g1_write_barrier_post(masm /*masm*/,
399 tmp1 /* store_adr */,
400 new_val /* new_val */,
401 rthread /* thread */,
402 tmp3 /* tmp */,
403 tmp2 /* tmp2 */);
404 }
405 }
406 NOT_LP64(imasm->restore_bcp());
407 }
408
409 #ifdef COMPILER1
410
411 #undef __
412 #define __ ce->masm()->
413
414 void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
415 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
416 // At this point we know that marking is in progress.
417 // If do_load() is true then we have to emit the
418 // load of the previous value; otherwise it has already
419 // been loaded into _pre_val.
420
421 __ bind(*stub->entry());
422 assert(stub->pre_val()->is_register(), "Precondition.");
423
424 Register pre_val_reg = stub->pre_val()->as_register();
425
426 if (stub->do_load()) {
427 ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/);
428 }
429
430 __ cmpptr(pre_val_reg, NULL_WORD);
431 __ jcc(Assembler::equal, *stub->continuation());
432 ce->store_parameter(stub->pre_val()->as_register(), 0);
433 __ call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
434 __ jmp(*stub->continuation());
435
436 }
437
438 void G1BarrierSetAssembler::gen_post_barrier_stub(LIR_Assembler* ce, G1PostBarrierStub* stub) {
439 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
440 __ bind(*stub->entry());
441 assert(stub->addr()->is_register(), "Precondition.");
442 assert(stub->new_val()->is_register(), "Precondition.");
443 Register new_val_reg = stub->new_val()->as_register();
444 __ cmpptr(new_val_reg, NULL_WORD);
445 __ jcc(Assembler::equal, *stub->continuation());
446 ce->store_parameter(stub->addr()->as_pointer_register(), 0);
447 __ call(RuntimeAddress(bs->post_barrier_c1_runtime_code_blob()->code_begin()));
448 __ jmp(*stub->continuation());
449 }
450
451 #undef __
452
453 #define __ sasm->
454
455 void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
456 // Generated code assumes that buffer index is pointer sized.
457 STATIC_ASSERT(in_bytes(SATBMarkQueue::byte_width_of_index()) == sizeof(intptr_t));
458
459 __ prologue("g1_pre_barrier", false);
460 // arg0 : previous value of memory
461
462 __ push(rax);
463 __ push(rdx);
464
465 const Register pre_val = rax;
466 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
467 const Register tmp = rdx;
468
469 NOT_LP64(__ get_thread(thread);)
470
471 Address queue_active(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
472 Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
473 Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
474
475 Label done;
476 Label runtime;
477
478 // Is marking still active?
479 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
480 __ cmpl(queue_active, 0);
481 } else {
482 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
483 __ cmpb(queue_active, 0);
484 }
485 __ jcc(Assembler::equal, done);
486
487 // Can we store original value in the thread's buffer?
488
489 __ movptr(tmp, queue_index);
490 __ testptr(tmp, tmp);
491 __ jcc(Assembler::zero, runtime);
492 __ subptr(tmp, wordSize);
493 __ movptr(queue_index, tmp);
494 __ addptr(tmp, buffer);
495
496 // prev_val (rax)
497 __ load_parameter(0, pre_val);
498 __ movptr(Address(tmp, 0), pre_val);
499 __ jmp(done);
500
501 __ bind(runtime);
502
503 __ push_call_clobbered_registers();
504
505 // load the pre-value
506 __ load_parameter(0, rcx);
507 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), rcx, thread);
508
509 __ pop_call_clobbered_registers();
510
511 __ bind(done);
512
513 __ pop(rdx);
514 __ pop(rax);
515
516 __ epilogue();
517 }
518
519 void G1BarrierSetAssembler::generate_c1_post_barrier_runtime_stub(StubAssembler* sasm) {
520 __ prologue("g1_post_barrier", false);
521
522 CardTableBarrierSet* ct =
523 barrier_set_cast<CardTableBarrierSet>(BarrierSet::barrier_set());
524
525 Label done;
526 Label enqueued;
527 Label runtime;
528
529 // At this point we know new_value is non-null and the new_value crosses regions.
530 // Must check to see if card is already dirty
531
532 const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
533
534 Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
535 Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
536
537 __ push(rax);
538 __ push(rcx);
539
540 const Register cardtable = rax;
541 const Register card_addr = rcx;
542
543 __ load_parameter(0, card_addr);
544 __ shrptr(card_addr, CardTable::card_shift());
545 // Do not use ExternalAddress to load 'byte_map_base', since 'byte_map_base' is NOT
546 // a valid address and therefore is not properly handled by the relocation code.
547 if (SCCache::is_on()) {
548 // SCA needs relocation info for this address
549 __ lea(cardtable, ExternalAddress((address)ct->card_table()->byte_map_base()));
550 } else {
551 __ movptr(cardtable, (intptr_t)ct->card_table()->byte_map_base());
552 }
553 __ addptr(card_addr, cardtable);
554
555 NOT_LP64(__ get_thread(thread);)
556
557 __ cmpb(Address(card_addr, 0), G1CardTable::g1_young_card_val());
558 __ jcc(Assembler::equal, done);
559
560 __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
561 __ cmpb(Address(card_addr, 0), CardTable::dirty_card_val());
562 __ jcc(Assembler::equal, done);
563
564 // storing region crossing non-null, card is clean.
565 // dirty card and log.
566
567 __ movb(Address(card_addr, 0), CardTable::dirty_card_val());
568
569 const Register tmp = rdx;
570 __ push(rdx);
571
572 __ movptr(tmp, queue_index);
573 __ testptr(tmp, tmp);
574 __ jcc(Assembler::zero, runtime);
575 __ subptr(tmp, wordSize);
576 __ movptr(queue_index, tmp);
577 __ addptr(tmp, buffer);
578 __ movptr(Address(tmp, 0), card_addr);
579 __ jmp(enqueued);
580
581 __ bind(runtime);
582 __ push_call_clobbered_registers();
583
584 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
585
586 __ pop_call_clobbered_registers();
587
588 __ bind(enqueued);
589 __ pop(rdx);
590
591 __ bind(done);
592 __ pop(rcx);
593 __ pop(rax);
594
595 __ epilogue();
596 }
597
598 #undef __
599
600 #endif // COMPILER1