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