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