1 /*
2 * Copyright (c) 2018, 2020, 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 "runtime/sharedRuntime.hpp"
35 #include "runtime/thread.hpp"
36 #include "interpreter/interp_masm.hpp"
37 #include "runtime/sharedRuntime.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
gen_write_ref_array_pre_barrier(MacroAssembler * masm,DecoratorSet decorators,Register addr,Register count,RegSet saved_regs)46 void G1BarrierSetAssembler::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators,
47 Register addr, Register count, RegSet saved_regs) {
48 bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;
49 if (!dest_uninitialized) {
50 Label done;
51 Address in_progress(rthread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
52
53 // Is marking active?
54 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
55 __ ldrw(rscratch1, in_progress);
56 } else {
57 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
58 __ ldrb(rscratch1, in_progress);
59 }
60 __ cbzw(rscratch1, done);
61
62 __ push(saved_regs, sp);
63 if (count == c_rarg0) {
64 if (addr == c_rarg1) {
65 // exactly backwards!!
66 __ mov(rscratch1, c_rarg0);
67 __ mov(c_rarg0, c_rarg1);
68 __ mov(c_rarg1, rscratch1);
69 } else {
70 __ mov(c_rarg1, count);
71 __ mov(c_rarg0, addr);
72 }
73 } else {
74 __ mov(c_rarg0, addr);
75 __ mov(c_rarg1, count);
76 }
77 if (UseCompressedOops) {
78 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_narrow_oop_entry), 2);
79 } else {
80 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry), 2);
81 }
82 __ pop(saved_regs, sp);
83
84 __ bind(done);
85 }
86 }
87
gen_write_ref_array_post_barrier(MacroAssembler * masm,DecoratorSet decorators,Register start,Register count,Register scratch,RegSet saved_regs)88 void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
89 Register start, Register count, Register scratch, RegSet saved_regs) {
90 __ push(saved_regs, sp);
91 assert_different_registers(start, count, scratch);
92 assert_different_registers(c_rarg0, count);
93 __ mov(c_rarg0, start);
94 __ mov(c_rarg1, count);
95 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_post_entry), 2);
96 __ pop(saved_regs, sp);
97 }
98
g1_write_barrier_pre(MacroAssembler * masm,Register obj,Register pre_val,Register thread,Register tmp,bool tosca_live,bool expand_call)99 void G1BarrierSetAssembler::g1_write_barrier_pre(MacroAssembler* masm,
100 Register obj,
101 Register pre_val,
102 Register thread,
103 Register tmp,
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, tmp, rscratch1);
116 assert(pre_val != noreg && tmp != 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(tmp, in_progress);
125 } else {
126 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
127 __ ldrb(tmp, in_progress);
128 }
129 __ cbzw(tmp, 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(tmp, index); // tmp := *index_adr
144 __ cbz(tmp, runtime); // tmp == 0?
145 // If yes, goto runtime
146
147 __ sub(tmp, tmp, wordSize); // tmp := tmp - wordSize
148 __ str(tmp, index); // *index_adr := tmp
149 __ ldr(rscratch1, buffer);
150 __ add(tmp, tmp, rscratch1); // tmp := tmp + *buffer_adr
151
152 // Record the previous value
153 __ str(pre_val, Address(tmp, 0));
154 __ b(done);
155
156 __ bind(runtime);
157 // save the live input values
158 RegSet saved = RegSet::of(pre_val);
159 if (tosca_live) saved += RegSet::of(r0);
160 if (obj != noreg) saved += RegSet::of(obj);
161
162 __ push(saved, sp);
163
164 // Calling the runtime using the regular call_VM_leaf mechanism generates
165 // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
166 // that checks that the *(rfp+frame::interpreter_frame_last_sp) == NULL.
167 //
168 // If we care generating the pre-barrier without a frame (e.g. in the
169 // intrinsified Reference.get() routine) then ebp might be pointing to
170 // the caller frame and so this check will most likely fail at runtime.
171 //
172 // Expanding the call directly bypasses the generation of the check.
173 // So when we do not have have a full interpreter frame on the stack
174 // expand_call should be passed true.
175
176 if (expand_call) {
177 assert(pre_val != c_rarg1, "smashed arg");
178 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
179 } else {
180 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
181 }
182
183 __ pop(saved, sp);
184
185 __ bind(done);
186
187 }
188
g1_write_barrier_post(MacroAssembler * masm,Register store_addr,Register new_val,Register thread,Register tmp,Register tmp2)189 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
190 Register store_addr,
191 Register new_val,
192 Register thread,
193 Register tmp,
194 Register tmp2) {
195 assert(thread == rthread, "must be");
196 assert_different_registers(store_addr, new_val, thread, tmp, tmp2,
197 rscratch1);
198 assert(store_addr != noreg && new_val != noreg && tmp != noreg
199 && tmp2 != noreg, "expecting a register");
200
201 Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
202 Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
203
204 BarrierSet* bs = BarrierSet::barrier_set();
205 CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs);
206 CardTable* ct = ctbs->card_table();
207 assert(sizeof(*ct->byte_map_base()) == sizeof(jbyte), "adjust this code");
208
209 Label done;
210 Label runtime;
211
212 // Does store cross heap regions?
213
214 __ eor(tmp, store_addr, new_val);
215 __ lsr(tmp, tmp, HeapRegion::LogOfHRGrainBytes);
216 __ cbz(tmp, done);
217
218 // crosses regions, storing NULL?
219
220 __ cbz(new_val, done);
221
222 // storing region crossing non-NULL, is card already dirty?
223
224 assert(sizeof(*ct->byte_map_base()) == sizeof(jbyte), "adjust this code");
225 const Register card_addr = tmp;
226
227 __ lsr(card_addr, store_addr, CardTable::card_shift);
228
229 // get the address of the card
230 __ load_byte_map_base(tmp2);
231 __ add(card_addr, card_addr, tmp2);
232 __ ldrb(tmp2, Address(card_addr));
233 __ cmpw(tmp2, (int)G1CardTable::g1_young_card_val());
234 __ br(Assembler::EQ, done);
235
236 assert((int)CardTable::dirty_card_val() == 0, "must be 0");
237
238 __ membar(Assembler::StoreLoad);
239
240 __ ldrb(tmp2, Address(card_addr));
241 __ cbzw(tmp2, done);
242
243 // storing a region crossing, non-NULL oop, card is clean.
244 // dirty card and log.
245
246 __ strb(zr, Address(card_addr));
247
248 __ ldr(rscratch1, queue_index);
249 __ cbz(rscratch1, runtime);
250 __ sub(rscratch1, rscratch1, wordSize);
251 __ str(rscratch1, queue_index);
252
253 __ ldr(tmp2, buffer);
254 __ str(card_addr, Address(tmp2, rscratch1));
255 __ b(done);
256
257 __ bind(runtime);
258 // save the live input values
259 RegSet saved = RegSet::of(store_addr, new_val);
260 __ push(saved, sp);
261 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
262 __ pop(saved, sp);
263
264 __ bind(done);
265 }
266
load_at(MacroAssembler * masm,DecoratorSet decorators,BasicType type,Register dst,Address src,Register tmp1,Register tmp_thread)267 void G1BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
268 Register dst, Address src, Register tmp1, Register tmp_thread) {
269 bool on_oop = type == T_OBJECT || type == T_ARRAY;
270 bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
271 bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
272 bool on_reference = on_weak || on_phantom;
273 ModRefBarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1, tmp_thread);
274 if (on_oop && on_reference) {
275 // LR is live. It must be saved around calls.
276 __ enter(); // barrier may call runtime
277 // Generate the G1 pre-barrier code to log the value of
278 // the referent field in an SATB buffer.
279 g1_write_barrier_pre(masm /* masm */,
280 noreg /* obj */,
281 dst /* pre_val */,
282 rthread /* thread */,
283 tmp1 /* tmp */,
284 true /* tosca_live */,
285 true /* expand_call */);
286 __ leave();
287 }
288 }
289
oop_store_at(MacroAssembler * masm,DecoratorSet decorators,BasicType type,Address dst,Register val,Register tmp1,Register tmp2)290 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
291 Address dst, Register val, Register tmp1, Register tmp2) {
292 // flatten object address if needed
293 if (dst.index() == noreg && dst.offset() == 0) {
294 if (dst.base() != r3) {
295 __ mov(r3, dst.base());
296 }
297 } else {
298 __ lea(r3, dst);
299 }
300
301 g1_write_barrier_pre(masm,
302 r3 /* obj */,
303 tmp2 /* pre_val */,
304 rthread /* thread */,
305 tmp1 /* tmp */,
306 val != noreg /* tosca_live */,
307 false /* expand_call */);
308
309 if (val == noreg) {
310 BarrierSetAssembler::store_at(masm, decorators, type, Address(r3, 0), noreg, noreg, noreg);
311 } else {
312 // G1 barrier needs uncompressed oop for region cross check.
313 Register new_val = val;
314 if (UseCompressedOops) {
315 new_val = rscratch2;
316 __ mov(new_val, val);
317 }
318 BarrierSetAssembler::store_at(masm, decorators, type, Address(r3, 0), val, noreg, noreg);
319 g1_write_barrier_post(masm,
320 r3 /* store_adr */,
321 new_val /* new_val */,
322 rthread /* thread */,
323 tmp1 /* tmp */,
324 tmp2 /* tmp2 */);
325 }
326
327 }
328
329 #ifdef COMPILER1
330
331 #undef __
332 #define __ ce->masm()->
333
gen_pre_barrier_stub(LIR_Assembler * ce,G1PreBarrierStub * stub)334 void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
335 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
336 // At this point we know that marking is in progress.
337 // If do_load() is true then we have to emit the
338 // load of the previous value; otherwise it has already
339 // been loaded into _pre_val.
340
341 __ bind(*stub->entry());
342
343 assert(stub->pre_val()->is_register(), "Precondition.");
344
345 Register pre_val_reg = stub->pre_val()->as_register();
346
347 if (stub->do_load()) {
348 ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/, false /*unaligned*/);
349 }
350 __ cbz(pre_val_reg, *stub->continuation());
351 ce->store_parameter(stub->pre_val()->as_register(), 0);
352 __ far_call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
353 __ b(*stub->continuation());
354 }
355
gen_post_barrier_stub(LIR_Assembler * ce,G1PostBarrierStub * stub)356 void G1BarrierSetAssembler::gen_post_barrier_stub(LIR_Assembler* ce, G1PostBarrierStub* stub) {
357 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
358 __ bind(*stub->entry());
359 assert(stub->addr()->is_register(), "Precondition.");
360 assert(stub->new_val()->is_register(), "Precondition.");
361 Register new_val_reg = stub->new_val()->as_register();
362 __ cbz(new_val_reg, *stub->continuation());
363 ce->store_parameter(stub->addr()->as_pointer_register(), 0);
364 __ far_call(RuntimeAddress(bs->post_barrier_c1_runtime_code_blob()->code_begin()));
365 __ b(*stub->continuation());
366 }
367
368 #undef __
369
370 #define __ sasm->
371
generate_c1_pre_barrier_runtime_stub(StubAssembler * sasm)372 void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
373 __ prologue("g1_pre_barrier", false);
374
375 // arg0 : previous value of memory
376
377 BarrierSet* bs = BarrierSet::barrier_set();
378
379 const Register pre_val = r0;
380 const Register thread = rthread;
381 const Register tmp = rscratch1;
382
383 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
384 Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
385 Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
386
387 Label done;
388 Label runtime;
389
390 // Is marking still active?
391 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
392 __ ldrw(tmp, in_progress);
393 } else {
394 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
395 __ ldrb(tmp, in_progress);
396 }
397 __ cbzw(tmp, done);
398
399 // Can we store original value in the thread's buffer?
400 __ ldr(tmp, queue_index);
401 __ cbz(tmp, runtime);
402
403 __ sub(tmp, tmp, wordSize);
404 __ str(tmp, queue_index);
405 __ ldr(rscratch2, buffer);
406 __ add(tmp, tmp, rscratch2);
407 __ load_parameter(0, rscratch2);
408 __ str(rscratch2, Address(tmp, 0));
409 __ b(done);
410
411 __ bind(runtime);
412 __ push_call_clobbered_registers();
413 __ load_parameter(0, pre_val);
414 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
415 __ pop_call_clobbered_registers();
416 __ bind(done);
417
418 __ epilogue();
419 }
420
generate_c1_post_barrier_runtime_stub(StubAssembler * sasm)421 void G1BarrierSetAssembler::generate_c1_post_barrier_runtime_stub(StubAssembler* sasm) {
422 __ prologue("g1_post_barrier", false);
423
424 // arg0: store_address
425 Address store_addr(rfp, 2*BytesPerWord);
426
427 BarrierSet* bs = BarrierSet::barrier_set();
428 CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs);
429 CardTable* ct = ctbs->card_table();
430 assert(sizeof(*ct->byte_map_base()) == sizeof(jbyte), "adjust this code");
431
432 Label done;
433 Label runtime;
434
435 // At this point we know new_value is non-NULL and the new_value crosses regions.
436 // Must check to see if card is already dirty
437
438 const Register thread = rthread;
439
440 Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
441 Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
442
443 const Register card_offset = rscratch2;
444 // LR is free here, so we can use it to hold the byte_map_base.
445 const Register byte_map_base = lr;
446
447 assert_different_registers(card_offset, byte_map_base, rscratch1);
448
449 __ load_parameter(0, card_offset);
450 __ lsr(card_offset, card_offset, CardTable::card_shift);
451 __ load_byte_map_base(byte_map_base);
452 __ ldrb(rscratch1, Address(byte_map_base, card_offset));
453 __ cmpw(rscratch1, (int)G1CardTable::g1_young_card_val());
454 __ br(Assembler::EQ, done);
455
456 assert((int)CardTable::dirty_card_val() == 0, "must be 0");
457
458 __ membar(Assembler::StoreLoad);
459 __ ldrb(rscratch1, Address(byte_map_base, card_offset));
460 __ cbzw(rscratch1, done);
461
462 // storing region crossing non-NULL, card is clean.
463 // dirty card and log.
464 __ strb(zr, Address(byte_map_base, card_offset));
465
466 // Convert card offset into an address in card_addr
467 Register card_addr = card_offset;
468 __ add(card_addr, byte_map_base, card_addr);
469
470 __ ldr(rscratch1, queue_index);
471 __ cbz(rscratch1, runtime);
472 __ sub(rscratch1, rscratch1, wordSize);
473 __ str(rscratch1, queue_index);
474
475 // Reuse LR to hold buffer_addr
476 const Register buffer_addr = lr;
477
478 __ ldr(buffer_addr, buffer);
479 __ str(card_addr, Address(buffer_addr, rscratch1));
480 __ b(done);
481
482 __ bind(runtime);
483 __ push_call_clobbered_registers();
484 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
485 __ pop_call_clobbered_registers();
486 __ bind(done);
487 __ epilogue();
488 }
489
490 #undef __
491
492 #endif // COMPILER1
493