1 /*
2 * Copyright (c) 2018, 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 "c1/c1_LIRAssembler.hpp"
28 #include "c1/c1_MacroAssembler.hpp"
29 #include "gc/g1/c1/g1BarrierSetC1.hpp"
30 #include "gc/g1/g1BarrierSet.hpp"
31 #include "gc/g1/g1BarrierSetAssembler.hpp"
32 #include "gc/g1/g1CardTable.hpp"
33 #include "gc/g1/g1ThreadLocalData.hpp"
34 #include "gc/g1/heapRegion.hpp"
35 #include "gc/shared/collectedHeap.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/thread.hpp"
38 #include "interpreter/interp_masm.hpp"
39 #include "runtime/sharedRuntime.hpp"
40
41 #define __ masm->
42
43 void G1BarrierSetAssembler::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators,
44 Register addr, Register count, RegSet saved_regs) {
45 bool dest_uninitialized = (decorators & AS_DEST_NOT_INITIALIZED) != 0;
46 if (!dest_uninitialized) {
47 __ push(saved_regs, sp);
48 if (count == c_rarg0) {
49 if (addr == c_rarg1) {
50 // exactly backwards!!
51 __ mov(rscratch1, c_rarg0);
52 __ mov(c_rarg0, c_rarg1);
53 __ mov(c_rarg1, rscratch1);
54 } else {
55 __ mov(c_rarg1, count);
56 __ mov(c_rarg0, addr);
57 }
58 } else {
59 __ mov(c_rarg0, addr);
60 __ mov(c_rarg1, count);
61 }
62 if (UseCompressedOops) {
63 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSet::write_ref_array_pre_narrow_oop_entry), 2);
64 } else {
65 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSet::write_ref_array_pre_oop_entry), 2);
66 }
67 __ pop(saved_regs, sp);
68 }
69 }
70
71 void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
72 Register start, Register end, Register scratch, RegSet saved_regs) {
73 __ push(saved_regs, sp);
74 // must compute element count unless barrier set interface is changed (other platforms supply count)
75 assert_different_registers(start, end, scratch);
76 __ lea(scratch, Address(end, BytesPerHeapOop));
77 __ sub(scratch, scratch, start); // subtract start to get #bytes
78 __ lsr(scratch, scratch, LogBytesPerHeapOop); // convert to element count
79 __ mov(c_rarg0, start);
80 __ mov(c_rarg1, scratch);
81 __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSet::write_ref_array_post_entry), 2);
82 __ pop(saved_regs, sp);
83 }
84
85 void G1BarrierSetAssembler::g1_write_barrier_pre(MacroAssembler* masm,
86 Register obj,
87 Register pre_val,
88 Register thread,
89 Register tmp,
90 bool tosca_live,
91 bool expand_call) {
92 // If expand_call is true then we expand the call_VM_leaf macro
93 // directly to skip generating the check by
94 // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
95
96 assert(thread == rthread, "must be");
97
98 Label done;
99 Label runtime;
100
101 assert_different_registers(obj, pre_val, tmp, rscratch1);
102 assert(pre_val != noreg && tmp != noreg, "expecting a register");
103
104 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
105 Address index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
106 Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
107
108 // Is marking active?
109 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
110 __ ldrw(tmp, in_progress);
111 } else {
112 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
113 __ ldrb(tmp, in_progress);
114 }
115 __ cbzw(tmp, done);
116
117 // Do we need to load the previous value?
118 if (obj != noreg) {
119 __ load_heap_oop(pre_val, Address(obj, 0), noreg, noreg, AS_RAW);
120 }
121
122 // Is the previous value null?
123 __ cbz(pre_val, done);
124
125 // Can we store original value in the thread's buffer?
126 // Is index == 0?
127 // (The index field is typed as size_t.)
128
129 __ ldr(tmp, index); // tmp := *index_adr
130 __ cbz(tmp, runtime); // tmp == 0?
131 // If yes, goto runtime
132
133 __ sub(tmp, tmp, wordSize); // tmp := tmp - wordSize
134 __ str(tmp, index); // *index_adr := tmp
135 __ ldr(rscratch1, buffer);
136 __ add(tmp, tmp, rscratch1); // tmp := tmp + *buffer_adr
137
138 // Record the previous value
139 __ str(pre_val, Address(tmp, 0));
140 __ b(done);
141
142 __ bind(runtime);
143 // save the live input values
144 RegSet saved = RegSet::of(pre_val);
145 if (tosca_live) saved += RegSet::of(r0);
146 if (obj != noreg) saved += RegSet::of(obj);
147
148 __ push(saved, sp);
149
150 // Calling the runtime using the regular call_VM_leaf mechanism generates
151 // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
152 // that checks that the *(rfp+frame::interpreter_frame_last_sp) == NULL.
153 //
154 // If we care generating the pre-barrier without a frame (e.g. in the
155 // intrinsified Reference.get() routine) then ebp might be pointing to
156 // the caller frame and so this check will most likely fail at runtime.
157 //
158 // Expanding the call directly bypasses the generation of the check.
159 // So when we do not have have a full interpreter frame on the stack
160 // expand_call should be passed true.
161
162 if (expand_call) {
163 assert(pre_val != c_rarg1, "smashed arg");
164 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), pre_val, thread);
165 } else {
166 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), pre_val, thread);
167 }
168
169 __ pop(saved, sp);
170
171 __ bind(done);
172
173 }
174
175 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
176 Register store_addr,
177 Register new_val,
178 Register thread,
179 Register tmp,
180 Register tmp2) {
181 assert(thread == rthread, "must be");
182 assert_different_registers(store_addr, new_val, thread, tmp, tmp2,
183 rscratch1);
184 assert(store_addr != noreg && new_val != noreg && tmp != noreg
185 && tmp2 != noreg, "expecting a register");
186
187 Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
188 Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
189
190 BarrierSet* bs = BarrierSet::barrier_set();
191 CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs);
192 CardTable* ct = ctbs->card_table();
193 assert(sizeof(*ct->byte_map_base()) == sizeof(jbyte), "adjust this code");
194
195 Label done;
196 Label runtime;
197
198 // Does store cross heap regions?
199
200 __ eor(tmp, store_addr, new_val);
201 __ lsr(tmp, tmp, HeapRegion::LogOfHRGrainBytes);
202 __ cbz(tmp, done);
203
204 // crosses regions, storing NULL?
205
206 __ cbz(new_val, done);
207
208 // storing region crossing non-NULL, is card already dirty?
209
210 ExternalAddress cardtable((address) ct->byte_map_base());
211 assert(sizeof(*ct->byte_map_base()) == sizeof(jbyte), "adjust this code");
212 const Register card_addr = tmp;
213
214 __ lsr(card_addr, store_addr, CardTable::card_shift);
215
216 // get the address of the card
217 __ load_byte_map_base(tmp2);
218 __ add(card_addr, card_addr, tmp2);
219 __ ldrb(tmp2, Address(card_addr));
220 __ cmpw(tmp2, (int)G1CardTable::g1_young_card_val());
221 __ br(Assembler::EQ, done);
222
223 assert((int)CardTable::dirty_card_val() == 0, "must be 0");
224
225 __ membar(Assembler::StoreLoad);
226
227 __ ldrb(tmp2, Address(card_addr));
228 __ cbzw(tmp2, done);
229
230 // storing a region crossing, non-NULL oop, card is clean.
231 // dirty card and log.
232
233 __ strb(zr, Address(card_addr));
234
235 __ ldr(rscratch1, queue_index);
236 __ cbz(rscratch1, runtime);
237 __ sub(rscratch1, rscratch1, wordSize);
238 __ str(rscratch1, queue_index);
239
240 __ ldr(tmp2, buffer);
241 __ str(card_addr, Address(tmp2, rscratch1));
242 __ b(done);
243
244 __ bind(runtime);
245 // save the live input values
246 RegSet saved = RegSet::of(store_addr, new_val);
247 __ push(saved, sp);
248 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, thread);
249 __ pop(saved, sp);
250
251 __ bind(done);
252 }
253
254 void G1BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
255 Register dst, Address src, Register tmp1, Register tmp_thread) {
256 bool on_oop = type == T_OBJECT || type == T_ARRAY;
257 bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
258 bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
259 bool on_reference = on_weak || on_phantom;
260 ModRefBarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1, tmp_thread);
261 if (on_oop && on_reference) {
262 // LR is live. It must be saved around calls.
263 __ enter(); // barrier may call runtime
264 // Generate the G1 pre-barrier code to log the value of
265 // the referent field in an SATB buffer.
266 g1_write_barrier_pre(masm /* masm */,
267 noreg /* obj */,
268 dst /* pre_val */,
269 rthread /* thread */,
270 tmp1 /* tmp */,
271 true /* tosca_live */,
272 true /* expand_call */);
273 __ leave();
274 }
275 }
276
277 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
278 Address dst, Register val, Register tmp1, Register tmp2) {
279 // flatten object address if needed
280 if (dst.index() == noreg && dst.offset() == 0) {
281 if (dst.base() != r3) {
282 __ mov(r3, dst.base());
283 }
284 } else {
285 __ lea(r3, dst);
286 }
287
288 g1_write_barrier_pre(masm,
289 r3 /* obj */,
290 tmp2 /* pre_val */,
291 rthread /* thread */,
292 tmp1 /* tmp */,
293 val != noreg /* tosca_live */,
294 false /* expand_call */);
295
296 if (val == noreg) {
297 BarrierSetAssembler::store_at(masm, decorators, type, Address(r3, 0), noreg, noreg, noreg);
298 } else {
299 // G1 barrier needs uncompressed oop for region cross check.
300 Register new_val = val;
301 if (UseCompressedOops) {
302 new_val = rscratch2;
303 __ mov(new_val, val);
304 }
305 BarrierSetAssembler::store_at(masm, decorators, type, Address(r3, 0), val, noreg, noreg);
306 g1_write_barrier_post(masm,
307 r3 /* store_adr */,
308 new_val /* new_val */,
309 rthread /* thread */,
310 tmp1 /* tmp */,
311 tmp2 /* tmp2 */);
312 }
313
314 }
315
316 #ifdef COMPILER1
317
318 #undef __
319 #define __ ce->masm()->
320
321 void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
322 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
323 // At this point we know that marking is in progress.
324 // If do_load() is true then we have to emit the
325 // load of the previous value; otherwise it has already
326 // been loaded into _pre_val.
327
328 __ bind(*stub->entry());
329
330 assert(stub->pre_val()->is_register(), "Precondition.");
331
332 Register pre_val_reg = stub->pre_val()->as_register();
333
334 if (stub->do_load()) {
335 ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/, false /*unaligned*/);
336 }
337 __ cbz(pre_val_reg, *stub->continuation());
338 ce->store_parameter(stub->pre_val()->as_register(), 0);
339 __ far_call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
340 __ b(*stub->continuation());
341 }
342
343 void G1BarrierSetAssembler::gen_post_barrier_stub(LIR_Assembler* ce, G1PostBarrierStub* stub) {
344 G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
345 __ bind(*stub->entry());
346 assert(stub->addr()->is_register(), "Precondition.");
347 assert(stub->new_val()->is_register(), "Precondition.");
348 Register new_val_reg = stub->new_val()->as_register();
349 __ cbz(new_val_reg, *stub->continuation());
350 ce->store_parameter(stub->addr()->as_pointer_register(), 0);
351 __ far_call(RuntimeAddress(bs->post_barrier_c1_runtime_code_blob()->code_begin()));
352 __ b(*stub->continuation());
353 }
354
355 #undef __
356
357 #define __ sasm->
358
359 void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
360 __ prologue("g1_pre_barrier", false);
361
362 // arg0 : previous value of memory
363
364 BarrierSet* bs = BarrierSet::barrier_set();
365
366 const Register pre_val = r0;
367 const Register thread = rthread;
368 const Register tmp = rscratch1;
369
370 Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
371 Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
372 Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
373
374 Label done;
375 Label runtime;
376
377 // Is marking still active?
378 if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
379 __ ldrw(tmp, in_progress);
380 } else {
381 assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
382 __ ldrb(tmp, in_progress);
383 }
384 __ cbzw(tmp, done);
385
386 // Can we store original value in the thread's buffer?
387 __ ldr(tmp, queue_index);
388 __ cbz(tmp, runtime);
389
390 __ sub(tmp, tmp, wordSize);
391 __ str(tmp, queue_index);
392 __ ldr(rscratch2, buffer);
393 __ add(tmp, tmp, rscratch2);
394 __ load_parameter(0, rscratch2);
395 __ str(rscratch2, Address(tmp, 0));
396 __ b(done);
397
398 __ bind(runtime);
399 __ push_call_clobbered_registers();
400 __ load_parameter(0, pre_val);
401 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), pre_val, thread);
402 __ pop_call_clobbered_registers();
403 __ bind(done);
404
405 __ epilogue();
406 }
407
408 void G1BarrierSetAssembler::generate_c1_post_barrier_runtime_stub(StubAssembler* sasm) {
409 __ prologue("g1_post_barrier", false);
410
411 // arg0: store_address
412 Address store_addr(rfp, 2*BytesPerWord);
413
414 BarrierSet* bs = BarrierSet::barrier_set();
415 CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs);
416 CardTable* ct = ctbs->card_table();
417 assert(sizeof(*ct->byte_map_base()) == sizeof(jbyte), "adjust this code");
418
419 Label done;
420 Label runtime;
421
422 // At this point we know new_value is non-NULL and the new_value crosses regions.
423 // Must check to see if card is already dirty
424
425 const Register thread = rthread;
426
427 Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
428 Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
429
430 const Register card_offset = rscratch2;
431 // LR is free here, so we can use it to hold the byte_map_base.
432 const Register byte_map_base = lr;
433
434 assert_different_registers(card_offset, byte_map_base, rscratch1);
435
436 __ load_parameter(0, card_offset);
437 __ lsr(card_offset, card_offset, CardTable::card_shift);
438 __ load_byte_map_base(byte_map_base);
439 __ ldrb(rscratch1, Address(byte_map_base, card_offset));
440 __ cmpw(rscratch1, (int)G1CardTable::g1_young_card_val());
441 __ br(Assembler::EQ, done);
442
443 assert((int)CardTable::dirty_card_val() == 0, "must be 0");
444
445 __ membar(Assembler::StoreLoad);
446 __ ldrb(rscratch1, Address(byte_map_base, card_offset));
447 __ cbzw(rscratch1, done);
448
449 // storing region crossing non-NULL, card is clean.
450 // dirty card and log.
451 __ strb(zr, Address(byte_map_base, card_offset));
452
453 // Convert card offset into an address in card_addr
454 Register card_addr = card_offset;
455 __ add(card_addr, byte_map_base, card_addr);
456
457 __ ldr(rscratch1, queue_index);
458 __ cbz(rscratch1, runtime);
459 __ sub(rscratch1, rscratch1, wordSize);
460 __ str(rscratch1, queue_index);
461
462 // Reuse LR to hold buffer_addr
463 const Register buffer_addr = lr;
464
465 __ ldr(buffer_addr, buffer);
466 __ str(card_addr, Address(buffer_addr, rscratch1));
467 __ b(done);
468
469 __ bind(runtime);
470 __ push_call_clobbered_registers();
471 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, thread);
472 __ pop_call_clobbered_registers();
473 __ bind(done);
474 __ epilogue();
475 }
476
477 #undef __
478
479 #endif // COMPILER1