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 #if INCLUDE_CDS
28 #include "code/SCCache.hpp"
29 #endif
30 #include "gc/g1/g1BarrierSet.hpp"
31 #include "gc/g1/g1BarrierSetAssembler.hpp"
32 #include "gc/g1/g1BarrierSetRuntime.hpp"
33 #include "gc/g1/g1CardTable.hpp"
34 #include "gc/g1/g1HeapRegion.hpp"
35 #include "gc/g1/g1ThreadLocalData.hpp"
36 #include "gc/shared/collectedHeap.hpp"
37 #include "interpreter/interp_masm.hpp"
38 #include "runtime/javaThread.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #ifdef COMPILER1
41 #include "c1/c1_LIRAssembler.hpp"
42 #include "c1/c1_MacroAssembler.hpp"
43 #include "gc/g1/c1/g1BarrierSetC1.hpp"
44 #endif
45
46 #define __ masm->
47
48 void G1BarrierSetAssembler::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators,
49 Register addr, Register count, RegSet saved_regs) {
50 bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;
51 if (!dest_uninitialized) {
52 Label done;
53 Address in_progress(rthread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
54
55 // Is marking active?
56 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
57 __ ldrw(rscratch1, in_progress);
58 } else {
59 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
60 __ ldrb(rscratch1, in_progress);
61 }
62 __ cbzw(rscratch1, done);
63
64 __ push(saved_regs, sp);
65 if (count == c_rarg0) {
66 if (addr == c_rarg1) {
67 // exactly backwards!!
68 __ mov(rscratch1, c_rarg0);
69 __ mov(c_rarg0, c_rarg1);
70 __ mov(c_rarg1, rscratch1);
71 } else {
72 __ mov(c_rarg1, count);
73 __ mov(c_rarg0, addr);
74 }
75 } else {
76 __ mov(c_rarg0, addr);
77 __ mov(c_rarg1, count);
78 }
79 if (UseCompressedOops) {
80 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_narrow_oop_entry), 2);
81 } else {
82 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry), 2);
83 }
84 __ pop(saved_regs, sp);
85
86 __ bind(done);
87 }
88 }
89
90 void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
91 Register start, Register count, Register scratch, RegSet saved_regs) {
92 __ push(saved_regs, sp);
93 assert_different_registers(start, count, scratch);
94 assert_different_registers(c_rarg0, count);
95 __ mov(c_rarg0, start);
96 __ mov(c_rarg1, count);
97 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_post_entry), 2);
98 __ pop(saved_regs, sp);
99 }
100
101 void G1BarrierSetAssembler::g1_write_barrier_pre(MacroAssembler* masm,
102 Register obj,
103 Register pre_val,
104 Register thread,
105 Register tmp1,
106 Register tmp2,
107 bool tosca_live,
108 bool expand_call) {
109 // If expand_call is true then we expand the call_VM_leaf macro
110 // directly to skip generating the check by
111 // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
112
113 assert(thread == rthread, "must be");
114
115 Label done;
116 Label runtime;
117
118 assert_different_registers(obj, pre_val, tmp1, tmp2);
119 assert(pre_val != noreg && tmp1 != noreg && tmp2 != noreg, "expecting a register");
120
121 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
122 Address index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
123 Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
124
125 // Is marking active?
126 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
127 __ ldrw(tmp1, in_progress);
128 } else {
129 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
130 __ ldrb(tmp1, in_progress);
131 }
132 __ cbzw(tmp1, done);
133
134 // Do we need to load the previous value?
135 if (obj != noreg) {
136 __ load_heap_oop(pre_val, Address(obj, 0), noreg, noreg, AS_RAW);
137 }
138
139 // Is the previous value null?
140 __ cbz(pre_val, done);
141
142 // Can we store original value in the thread's buffer?
143 // Is index == 0?
144 // (The index field is typed as size_t.)
145
146 __ ldr(tmp1, index); // tmp := *index_adr
147 __ cbz(tmp1, runtime); // tmp == 0?
148 // If yes, goto runtime
149
150 __ sub(tmp1, tmp1, wordSize); // tmp := tmp - wordSize
151 __ str(tmp1, index); // *index_adr := tmp
152 __ ldr(tmp2, buffer);
153 __ add(tmp1, tmp1, tmp2); // tmp := tmp + *buffer_adr
154
155 // Record the previous value
156 __ str(pre_val, Address(tmp1, 0));
157 __ b(done);
158
159 __ bind(runtime);
160
161 __ push_call_clobbered_registers();
162
163 // Calling the runtime using the regular call_VM_leaf mechanism generates
164 // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
165 // that checks that the *(rfp+frame::interpreter_frame_last_sp) == nullptr.
166 //
167 // If we care generating the pre-barrier without a frame (e.g. in the
168 // intrinsified Reference.get() routine) then rfp might be pointing to
169 // the caller frame and so this check will most likely fail at runtime.
170 //
171 // Expanding the call directly bypasses the generation of the check.
172 // So when we do not have have a full interpreter frame on the stack
173 // expand_call should be passed true.
174
175 if (expand_call) {
176 assert(pre_val != c_rarg1, "smashed arg");
177 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
178 } else {
179 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
180 }
181
182 __ pop_call_clobbered_registers();
183
184 __ bind(done);
185
186 }
187
188 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
189 Register store_addr,
190 Register new_val,
191 Register thread,
192 Register tmp1,
193 Register tmp2) {
194 assert(thread == rthread, "must be");
195 assert_different_registers(store_addr, new_val, thread, tmp1, tmp2,
196 rscratch1);
197 assert(store_addr != noreg && new_val != noreg && tmp1 != noreg
198 && tmp2 != noreg, "expecting a register");
199
200 Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
201 Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
202
203 BarrierSet* bs = BarrierSet::barrier_set();
204 CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs);
205 CardTable* ct = ctbs->card_table();
206
207 Label done;
208 Label runtime;
209
210 // Does store cross heap regions?
211
212 #if INCLUDE_CDS
213 // AOT code needs to load the barrier grain shift from the aot
214 // runtime constants area in the code cache otherwise we can compile
215 // it as an immediate operand
216 if (SCCache::is_on_for_write()) {
217 address grain_shift_address = (address)AOTRuntimeConstants::grain_shift_address();
218 __ eor(tmp1, store_addr, new_val);
219 __ lea(tmp2, ExternalAddress(grain_shift_address));
220 __ ldrb(tmp2, tmp2);
221 __ lsrv(tmp1, tmp1, tmp2);
222 __ cbz(tmp1, done);
223 } else
224 #endif
225 {
226 __ eor(tmp1, store_addr, new_val);
227 __ lsr(tmp1, tmp1, G1HeapRegion::LogOfHRGrainBytes);
228 __ cbz(tmp1, done);
229 }
230 // crosses regions, storing null?
231
232 __ cbz(new_val, done);
233
234 // storing region crossing non-null, is card already dirty?
235
236 const Register card_addr = tmp1;
237
238 #if INCLUDE_CDS
239 // AOT code needs to load the barrier card shift from the aot
240 // runtime constants area in the code cache otherwise we can compile
241 // it as an immediate operand
242 if (SCCache::is_on_for_write()) {
243 address card_shift_address = (address)AOTRuntimeConstants::card_shift_address();
244 __ lea(tmp2, ExternalAddress(card_shift_address));
245 __ ldrb(tmp2, tmp2);
246 __ lsrv(card_addr, store_addr, tmp2);
247 } else
248 #endif
249 {
250 __ lsr(card_addr, store_addr, CardTable::card_shift());
251 }
252
253 // get the address of the card
254 __ load_byte_map_base(tmp2);
255 __ add(card_addr, card_addr, tmp2);
256 __ ldrb(tmp2, Address(card_addr));
257 __ cmpw(tmp2, (int)G1CardTable::g1_young_card_val());
258 __ br(Assembler::EQ, done);
259
260 assert((int)CardTable::dirty_card_val() == 0, "must be 0");
261
262 __ membar(Assembler::StoreLoad);
263
264 __ ldrb(tmp2, Address(card_addr));
265 __ cbzw(tmp2, done);
266
267 // storing a region crossing, non-null oop, card is clean.
268 // dirty card and log.
269
270 __ strb(zr, Address(card_addr));
271
272 __ ldr(rscratch1, queue_index);
273 __ cbz(rscratch1, runtime);
274 __ sub(rscratch1, rscratch1, wordSize);
275 __ str(rscratch1, queue_index);
276
277 __ ldr(tmp2, buffer);
278 __ str(card_addr, Address(tmp2, rscratch1));
279 __ b(done);
280
281 __ bind(runtime);
282 // save the live input values
283 RegSet saved = RegSet::of(store_addr);
284 __ push(saved, sp);
285 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
286 __ pop(saved, sp);
287
288 __ bind(done);
289 }
290
291 void G1BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
292 Register dst, Address src, Register tmp1, Register tmp2) {
293 bool on_oop = is_reference_type(type);
294 bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
295 bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
296 bool on_reference = on_weak || on_phantom;
297 ModRefBarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1, tmp2);
298 if (on_oop && on_reference) {
299 // LR is live. It must be saved around calls.
300 __ enter(/*strip_ret_addr*/true); // barrier may call runtime
301 // Generate the G1 pre-barrier code to log the value of
302 // the referent field in an SATB buffer.
303 g1_write_barrier_pre(masm /* masm */,
304 noreg /* obj */,
305 dst /* pre_val */,
306 rthread /* thread */,
307 tmp1 /* tmp1 */,
308 tmp2 /* tmp2 */,
309 true /* tosca_live */,
310 true /* expand_call */);
311 __ leave();
312 }
313 }
314
315 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
316 Address dst, Register val, Register tmp1, Register tmp2, Register tmp3) {
317 // flatten object address if needed
318 if (dst.index() == noreg && dst.offset() == 0) {
319 if (dst.base() != tmp3) {
320 __ mov(tmp3, dst.base());
321 }
322 } else {
323 __ lea(tmp3, dst);
324 }
325
326 g1_write_barrier_pre(masm,
327 tmp3 /* obj */,
328 tmp2 /* pre_val */,
329 rthread /* thread */,
330 tmp1 /* tmp1 */,
331 rscratch2 /* tmp2 */,
332 val != noreg /* tosca_live */,
333 false /* expand_call */);
334
335 if (val == noreg) {
336 BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp3, 0), noreg, noreg, noreg, noreg);
337 } else {
338 // G1 barrier needs uncompressed oop for region cross check.
339 Register new_val = val;
340 if (UseCompressedOops) {
341 new_val = rscratch2;
342 __ mov(new_val, val);
343 }
344 BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp3, 0), val, noreg, noreg, noreg);
345 g1_write_barrier_post(masm,
346 tmp3 /* store_adr */,
347 new_val /* new_val */,
348 rthread /* thread */,
349 tmp1 /* tmp1 */,
350 tmp2 /* tmp2 */);
351 }
352
353 }
354
355 #ifdef COMPILER1
356
357 #undef __
358 #define __ ce->masm()->
359
360 void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
361 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
362 // At this point we know that marking is in progress.
363 // If do_load() is true then we have to emit the
364 // load of the previous value; otherwise it has already
365 // been loaded into _pre_val.
366
367 __ bind(*stub->entry());
368
369 assert(stub->pre_val()->is_register(), "Precondition.");
370
371 Register pre_val_reg = stub->pre_val()->as_register();
372
373 if (stub->do_load()) {
374 ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/);
375 }
376 __ cbz(pre_val_reg, *stub->continuation());
377 ce->store_parameter(stub->pre_val()->as_register(), 0);
378 __ far_call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
379 __ b(*stub->continuation());
380 }
381
382 void G1BarrierSetAssembler::gen_post_barrier_stub(LIR_Assembler* ce, G1PostBarrierStub* stub) {
383 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
384 __ bind(*stub->entry());
385 assert(stub->addr()->is_register(), "Precondition.");
386 assert(stub->new_val()->is_register(), "Precondition.");
387 Register new_val_reg = stub->new_val()->as_register();
388 __ cbz(new_val_reg, *stub->continuation());
389 ce->store_parameter(stub->addr()->as_pointer_register(), 0);
390 __ far_call(RuntimeAddress(bs->post_barrier_c1_runtime_code_blob()->code_begin()));
391 __ b(*stub->continuation());
392 }
393
394 #undef __
395
396 #define __ sasm->
397
398 void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
399 __ prologue("g1_pre_barrier", false);
400
401 // arg0 : previous value of memory
402
403 BarrierSet* bs = BarrierSet::barrier_set();
404
405 const Register pre_val = r0;
406 const Register thread = rthread;
407 const Register tmp = rscratch1;
408
409 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
410 Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
411 Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
412
413 Label done;
414 Label runtime;
415
416 // Is marking still active?
417 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
418 __ ldrw(tmp, in_progress);
419 } else {
420 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
421 __ ldrb(tmp, in_progress);
422 }
423 __ cbzw(tmp, done);
424
425 // Can we store original value in the thread's buffer?
426 __ ldr(tmp, queue_index);
427 __ cbz(tmp, runtime);
428
429 __ sub(tmp, tmp, wordSize);
430 __ str(tmp, queue_index);
431 __ ldr(rscratch2, buffer);
432 __ add(tmp, tmp, rscratch2);
433 __ load_parameter(0, rscratch2);
434 __ str(rscratch2, Address(tmp, 0));
435 __ b(done);
436
437 __ bind(runtime);
438 __ push_call_clobbered_registers();
439 __ load_parameter(0, pre_val);
440 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
441 __ pop_call_clobbered_registers();
442 __ bind(done);
443
444 __ epilogue();
445 }
446
447 void G1BarrierSetAssembler::generate_c1_post_barrier_runtime_stub(StubAssembler* sasm) {
448 __ prologue("g1_post_barrier", false);
449
450 // arg0: store_address
451 Address store_addr(rfp, 2*BytesPerWord);
452
453 BarrierSet* bs = BarrierSet::barrier_set();
454 CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs);
455 CardTable* ct = ctbs->card_table();
456
457 Label done;
458 Label runtime;
459
460 // At this point we know new_value is non-null and the new_value crosses regions.
461 // Must check to see if card is already dirty
462
463 const Register thread = rthread;
464
465 Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
466 Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
467
468 const Register card_offset = rscratch2;
469 // LR is free here, so we can use it to hold the byte_map_base.
470 const Register byte_map_base = lr;
471
472 assert_different_registers(card_offset, byte_map_base, rscratch1);
473
474 __ load_parameter(0, card_offset);
475 __ lsr(card_offset, card_offset, CardTable::card_shift());
476 __ load_byte_map_base(byte_map_base);
477 __ ldrb(rscratch1, Address(byte_map_base, card_offset));
478 __ cmpw(rscratch1, (int)G1CardTable::g1_young_card_val());
479 __ br(Assembler::EQ, done);
480
481 assert((int)CardTable::dirty_card_val() == 0, "must be 0");
482
483 __ membar(Assembler::StoreLoad);
484 __ ldrb(rscratch1, Address(byte_map_base, card_offset));
485 __ cbzw(rscratch1, done);
486
487 // storing region crossing non-null, card is clean.
488 // dirty card and log.
489 __ strb(zr, Address(byte_map_base, card_offset));
490
491 // Convert card offset into an address in card_addr
492 Register card_addr = card_offset;
493 __ add(card_addr, byte_map_base, card_addr);
494
495 __ ldr(rscratch1, queue_index);
496 __ cbz(rscratch1, runtime);
497 __ sub(rscratch1, rscratch1, wordSize);
498 __ str(rscratch1, queue_index);
499
500 // Reuse LR to hold buffer_addr
501 const Register buffer_addr = lr;
502
503 __ ldr(buffer_addr, buffer);
504 __ str(card_addr, Address(buffer_addr, rscratch1));
505 __ b(done);
506
507 __ bind(runtime);
508 __ push_call_clobbered_registers();
509 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
510 __ pop_call_clobbered_registers();
511 __ bind(done);
512 __ epilogue();
513 }
514
515 #undef __
516
517 #endif // COMPILER1