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 "gc/shared/collectedHeap.hpp"
33 #include "interpreter/interp_masm.hpp"
34 #include "runtime/javaThread.hpp"
35 #include "runtime/sharedRuntime.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, RegSet saved_regs) {
49 bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;
50 if (!dest_uninitialized) {
51 Label done;
52 Address in_progress(rthread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
53
54 // Is marking active?
55 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
56 __ ldrw(rscratch1, in_progress);
57 } else {
58 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
59 __ ldrb(rscratch1, in_progress);
60 }
61 __ cbzw(rscratch1, done);
62
63 __ push(saved_regs, sp);
64 if (count == c_rarg0) {
65 if (addr == c_rarg1) {
66 // exactly backwards!!
67 __ mov(rscratch1, c_rarg0);
68 __ mov(c_rarg0, c_rarg1);
69 __ mov(c_rarg1, rscratch1);
70 } else {
71 __ mov(c_rarg1, count);
72 __ mov(c_rarg0, addr);
73 }
74 } else {
75 __ mov(c_rarg0, addr);
76 __ mov(c_rarg1, count);
77 }
78 if (UseCompressedOops) {
79 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_narrow_oop_entry), 2);
80 } else {
81 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry), 2);
82 }
83 __ pop(saved_regs, sp);
84
85 __ bind(done);
86 }
87 }
88
89 void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm,
90 DecoratorSet decorators,
91 Register start,
92 Register count,
93 Register scratch) {
94
95 Label done;
96 Label loop;
97 Label next;
98
99 __ cbz(count, done);
100
101 // Calculate the number of card marks to set. Since the object might start and
102 // end within a card, we need to calculate this via the card table indexes of
103 // the actual start and last addresses covered by the object.
104 // Temporarily use the count register for the last element address.
105 __ lea(count, Address(start, count, Address::lsl(LogBytesPerHeapOop))); // end = start + count << LogBytesPerHeapOop
106 __ sub(count, count, BytesPerHeapOop); // Use last element address for end.
107
108 __ lsr(start, start, CardTable::card_shift());
109 __ lsr(count, count, CardTable::card_shift());
110 __ sub(count, count, start); // Number of bytes to mark - 1.
111
112 // Add card table base offset to start.
113 __ ldr(scratch, Address(rthread, in_bytes(G1ThreadLocalData::card_table_base_offset())));
114 __ add(start, start, scratch);
115
116 __ bind(loop);
117 if (UseCondCardMark) {
118 __ ldrb(scratch, Address(start, count));
119 // Instead of loading clean_card_val and comparing, we exploit the fact that
120 // the LSB of non-clean cards is always 0, and the LSB of clean cards 1.
121 __ tbz(scratch, 0, next);
122 }
123 static_assert(G1CardTable::dirty_card_val() == 0, "must be to use zr");
124 __ strb(zr, Address(start, count));
125 __ bind(next);
126 __ subs(count, count, 1);
127 __ br(Assembler::GE, loop);
128
129 __ bind(done);
130 }
131
132 static void generate_queue_test_and_insertion(MacroAssembler* masm, ByteSize index_offset, ByteSize buffer_offset, Label& runtime,
133 const Register thread, const Register value, const Register temp1, const Register temp2) {
134 // Can we store a value in the given thread's buffer?
135 // (The index field is typed as size_t.)
136 __ ldr(temp1, Address(thread, in_bytes(index_offset))); // temp1 := *(index address)
137 __ cbz(temp1, runtime); // jump to runtime if index == 0 (full buffer)
138 // The buffer is not full, store value into it.
139 __ sub(temp1, temp1, wordSize); // temp1 := next index
140 __ str(temp1, Address(thread, in_bytes(index_offset))); // *(index address) := next index
141 __ ldr(temp2, Address(thread, in_bytes(buffer_offset))); // temp2 := buffer address
142 __ str(value, Address(temp2, temp1)); // *(buffer address + next index) := value
143 }
144
145 static void generate_pre_barrier_fast_path(MacroAssembler* masm,
146 const Register thread,
147 const Register tmp1) {
148 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
149 // Is marking active?
150 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
151 __ ldrw(tmp1, in_progress);
152 } else {
153 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
154 __ ldrb(tmp1, in_progress);
155 }
156 }
157
158 static void generate_pre_barrier_slow_path(MacroAssembler* masm,
159 const Register obj,
160 const Register pre_val,
161 const Register thread,
162 const Register tmp1,
163 const Register tmp2,
164 Label& done,
165 Label& runtime) {
166 // Do we need to load the previous value?
167 if (obj != noreg) {
168 __ load_heap_oop(pre_val, Address(obj, 0), noreg, noreg, AS_RAW);
169 }
170 // Is the previous value null?
171 __ cbz(pre_val, done);
172 generate_queue_test_and_insertion(masm,
173 G1ThreadLocalData::satb_mark_queue_index_offset(),
174 G1ThreadLocalData::satb_mark_queue_buffer_offset(),
175 runtime,
176 thread, pre_val, tmp1, tmp2);
177 __ b(done);
178 }
179
180 void G1BarrierSetAssembler::g1_write_barrier_pre(MacroAssembler* masm,
181 Register obj,
182 Register pre_val,
183 Register thread,
184 Register tmp1,
185 Register tmp2,
186 bool tosca_live,
187 bool expand_call) {
188 // If expand_call is true then we expand the call_VM_leaf macro
189 // directly to skip generating the check by
190 // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
191
192 assert(thread == rthread, "must be");
193
194 Label done;
195 Label runtime;
196
197 assert_different_registers(obj, pre_val, tmp1, tmp2);
198 assert(pre_val != noreg && tmp1 != noreg && tmp2 != noreg, "expecting a register");
199
200 generate_pre_barrier_fast_path(masm, thread, tmp1);
201 // If marking is not active (*(mark queue active address) == 0), jump to done
202 __ cbzw(tmp1, done);
203 generate_pre_barrier_slow_path(masm, obj, pre_val, thread, tmp1, tmp2, done, runtime);
204
205 __ bind(runtime);
206
207 __ push_call_clobbered_registers();
208
209 // Calling the runtime using the regular call_VM_leaf mechanism generates
210 // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
211 // that checks that the *(rfp+frame::interpreter_frame_last_sp) == nullptr.
212 //
213 // If we care generating the pre-barrier without a frame (e.g. in the
214 // intrinsified Reference.get() routine) then rfp might be pointing to
215 // the caller frame and so this check will most likely fail at runtime.
216 //
217 // Expanding the call directly bypasses the generation of the check.
218 // So when we do not have have a full interpreter frame on the stack
219 // expand_call should be passed true.
220
221 if (expand_call) {
222 assert(pre_val != c_rarg1, "smashed arg");
223 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
224 } else {
225 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
226 }
227
228 __ pop_call_clobbered_registers();
229
230 __ bind(done);
231
232 }
233
234 static void generate_post_barrier(MacroAssembler* masm,
235 const Register store_addr,
236 const Register new_val,
237 const Register thread,
238 const Register tmp1,
239 const Register tmp2,
240 Label& done,
241 bool new_val_may_be_null) {
242 assert(thread == rthread, "must be");
243 assert_different_registers(store_addr, new_val, thread, tmp1, tmp2, noreg, rscratch1);
244
245 // Does store cross heap regions?
246 __ eor(tmp1, store_addr, new_val); // tmp1 := store address ^ new value
247 __ lsr(tmp1, tmp1, G1HeapRegion::LogOfHRGrainBytes); // tmp1 := ((store address ^ new value) >> LogOfHRGrainBytes)
248 __ cbz(tmp1, done);
249 // Crosses regions, storing null?
250 if (new_val_may_be_null) {
251 __ cbz(new_val, done);
252 }
253 // Storing region crossing non-null.
254 __ lsr(tmp1, store_addr, CardTable::card_shift()); // tmp1 := card address relative to card table base
255
256 Address card_table_addr(thread, in_bytes(G1ThreadLocalData::card_table_base_offset()));
257 __ ldr(tmp2, card_table_addr); // tmp2 := card table base address
258 if (UseCondCardMark) {
259 __ ldrb(rscratch1, Address(tmp1, tmp2)); // rscratch1 := card
260 // Instead of loading clean_card_val and comparing, we exploit the fact that
261 // the LSB of non-clean cards is always 0, and the LSB of clean cards 1.
262 __ tbz(rscratch1, 0, done);
263 }
264 static_assert(G1CardTable::dirty_card_val() == 0, "must be to use zr");
265 __ strb(zr, Address(tmp1, tmp2)); // *(card address) := dirty_card_val
266 }
267
268 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
269 Register store_addr,
270 Register new_val,
271 Register thread,
272 Register tmp1,
273 Register tmp2) {
274 Label done;
275 generate_post_barrier(masm, store_addr, new_val, thread, tmp1, tmp2, done, false /* new_val_may_be_null */);
276 __ bind(done);
277 }
278
279 #if defined(COMPILER2)
280
281 static void generate_c2_barrier_runtime_call(MacroAssembler* masm, G1BarrierStubC2* stub, const Register arg, const address runtime_path) {
282 SaveLiveRegisters save_registers(masm, stub);
283 if (c_rarg0 != arg) {
284 __ mov(c_rarg0, arg);
285 }
286 __ mov(c_rarg1, rthread);
287 __ mov(rscratch1, runtime_path);
288 __ blr(rscratch1);
289 }
290
291 void G1BarrierSetAssembler::g1_write_barrier_pre_c2(MacroAssembler* masm,
292 Register obj,
293 Register pre_val,
294 Register thread,
295 Register tmp1,
296 Register tmp2,
297 G1PreBarrierStubC2* stub) {
298 assert(thread == rthread, "must be");
299 assert_different_registers(obj, pre_val, tmp1, tmp2);
300 assert(pre_val != noreg && tmp1 != noreg && tmp2 != noreg, "expecting a register");
301
302 stub->initialize_registers(obj, pre_val, thread, tmp1, tmp2);
303
304 generate_pre_barrier_fast_path(masm, thread, tmp1);
305 // If marking is active (*(mark queue active address) != 0), jump to stub (slow path)
306 __ cbnzw(tmp1, *stub->entry());
307
308 __ bind(*stub->continuation());
309 }
310
311 void G1BarrierSetAssembler::generate_c2_pre_barrier_stub(MacroAssembler* masm,
312 G1PreBarrierStubC2* stub) const {
313 Assembler::InlineSkippedInstructionsCounter skip_counter(masm);
314 Label runtime;
315 Register obj = stub->obj();
316 Register pre_val = stub->pre_val();
317 Register thread = stub->thread();
318 Register tmp1 = stub->tmp1();
319 Register tmp2 = stub->tmp2();
320
321 __ bind(*stub->entry());
322 generate_pre_barrier_slow_path(masm, obj, pre_val, thread, tmp1, tmp2, *stub->continuation(), runtime);
323
324 __ bind(runtime);
325 generate_c2_barrier_runtime_call(masm, stub, pre_val, CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry));
326 __ b(*stub->continuation());
327 }
328
329 void G1BarrierSetAssembler::g1_write_barrier_post_c2(MacroAssembler* masm,
330 Register store_addr,
331 Register new_val,
332 Register thread,
333 Register tmp1,
334 Register tmp2,
335 bool new_val_may_be_null) {
336 Label done;
337 generate_post_barrier(masm, store_addr, new_val, thread, tmp1, tmp2, done, new_val_may_be_null);
338 __ bind(done);
339 }
340
341 #endif // COMPILER2
342
343 void G1BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
344 Register dst, Address src, Register tmp1, Register tmp2) {
345 bool on_oop = is_reference_type(type);
346 bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
347 bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
348 bool on_reference = on_weak || on_phantom;
349 CardTableBarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1, tmp2);
350 if (on_oop && on_reference) {
351 // LR is live. It must be saved around calls.
352 __ enter(/*strip_ret_addr*/true); // barrier may call runtime
353 // Generate the G1 pre-barrier code to log the value of
354 // the referent field in an SATB buffer.
355 g1_write_barrier_pre(masm /* masm */,
356 noreg /* obj */,
357 dst /* pre_val */,
358 rthread /* thread */,
359 tmp1 /* tmp1 */,
360 tmp2 /* tmp2 */,
361 true /* tosca_live */,
362 true /* expand_call */);
363 __ leave();
364 }
365 }
366
367 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
368 Address dst, Register val, Register tmp1, Register tmp2, Register tmp3) {
369 // flatten object address if needed
370 if (dst.index() == noreg && dst.offset() == 0) {
371 if (dst.base() != tmp3) {
372 __ mov(tmp3, dst.base());
373 }
374 } else {
375 __ lea(tmp3, dst);
376 }
377
378 g1_write_barrier_pre(masm,
379 tmp3 /* obj */,
380 tmp2 /* pre_val */,
381 rthread /* thread */,
382 tmp1 /* tmp1 */,
383 rscratch2 /* tmp2 */,
384 val != noreg /* tosca_live */,
385 false /* expand_call */);
386
387 if (val == noreg) {
388 BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp3, 0), noreg, noreg, noreg, noreg);
389 } else {
390 // G1 barrier needs uncompressed oop for region cross check.
391 Register new_val = val;
392 if (UseCompressedOops) {
393 new_val = rscratch2;
394 __ mov(new_val, val);
395 }
396 BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp3, 0), val, noreg, noreg, noreg);
397 g1_write_barrier_post(masm,
398 tmp3 /* store_adr */,
399 new_val /* new_val */,
400 rthread /* thread */,
401 tmp1 /* tmp1 */,
402 tmp2 /* tmp2 */);
403 }
404
405 }
406
407 #ifdef COMPILER1
408
409 #undef __
410 #define __ ce->masm()->
411
412 void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
413 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
414 // At this point we know that marking is in progress.
415 // If do_load() is true then we have to emit the
416 // load of the previous value; otherwise it has already
417 // been loaded into _pre_val.
418
419 __ bind(*stub->entry());
420
421 assert(stub->pre_val()->is_register(), "Precondition.");
422
423 Register pre_val_reg = stub->pre_val()->as_register();
424
425 if (stub->do_load()) {
426 ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/);
427 }
428 __ cbz(pre_val_reg, *stub->continuation());
429 ce->store_parameter(stub->pre_val()->as_register(), 0);
430 __ far_call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
431 __ b(*stub->continuation());
432 }
433
434 #undef __
435
436 void G1BarrierSetAssembler::g1_write_barrier_post_c1(MacroAssembler* masm,
437 Register store_addr,
438 Register new_val,
439 Register thread,
440 Register tmp1,
441 Register tmp2) {
442 Label done;
443 generate_post_barrier(masm, store_addr, new_val, thread, tmp1, tmp2, done, true /* new_val_may_be_null */);
444 masm->bind(done);
445 }
446
447 #define __ sasm->
448
449 void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
450 __ prologue("g1_pre_barrier", false);
451
452 // arg0 : previous value of memory
453
454 BarrierSet* bs = BarrierSet::barrier_set();
455
456 const Register pre_val = r0;
457 const Register thread = rthread;
458 const Register tmp = rscratch1;
459
460 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
461 Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
462 Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
463
464 Label done;
465 Label runtime;
466
467 // Is marking still active?
468 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
469 __ ldrw(tmp, in_progress);
470 } else {
471 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
472 __ ldrb(tmp, in_progress);
473 }
474 __ cbzw(tmp, done);
475
476 // Can we store original value in the thread's buffer?
477 __ ldr(tmp, queue_index);
478 __ cbz(tmp, runtime);
479
480 __ sub(tmp, tmp, wordSize);
481 __ str(tmp, queue_index);
482 __ ldr(rscratch2, buffer);
483 __ add(tmp, tmp, rscratch2);
484 __ load_parameter(0, rscratch2);
485 __ str(rscratch2, Address(tmp, 0));
486 __ b(done);
487
488 __ bind(runtime);
489 __ push_call_clobbered_registers();
490 __ load_parameter(0, pre_val);
491 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
492 __ pop_call_clobbered_registers();
493 __ bind(done);
494
495 __ epilogue();
496 }
497
498 #undef __
499
500 #endif // COMPILER1