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 "ci/ciUtilities.hpp"
26 #include "gc/shared/c2/cardTableBarrierSetC2.hpp"
27 #include "gc/shared/cardTable.hpp"
28 #include "gc/shared/cardTableBarrierSet.hpp"
29 #include "gc/shared/gc_globals.hpp"
30 #include "opto/arraycopynode.hpp"
31 #include "opto/graphKit.hpp"
32 #include "opto/idealKit.hpp"
33 #include "opto/macro.hpp"
34 #include "utilities/macros.hpp"
35
36 #define __ ideal.
37
38 Node* CardTableBarrierSetC2::store_at_resolved(C2Access& access, C2AccessValue& val) const {
39 DecoratorSet decorators = access.decorators();
40
41 Node* adr = access.addr().node();
42
43 bool is_array = (decorators & IS_ARRAY) != 0;
44 bool anonymous = (decorators & ON_UNKNOWN_OOP_REF) != 0;
45 bool in_heap = (decorators & IN_HEAP) != 0;
46 bool use_precise = is_array || anonymous;
47 bool tightly_coupled_alloc = (decorators & C2_TIGHTLY_COUPLED_ALLOC) != 0;
48
49 if (!access.is_oop() || tightly_coupled_alloc || (!in_heap && !anonymous)) {
50 return BarrierSetC2::store_at_resolved(access, val);
51 }
52
53 assert(access.is_parse_access(), "entry not supported at optimization time");
54 C2ParseAccess& parse_access = static_cast<C2ParseAccess&>(access);
55
56 Node* store = BarrierSetC2::store_at_resolved(access, val);
57 post_barrier(parse_access.kit(), access.base(), adr, val.node(), use_precise);
58
59 return store;
60 }
61
62 Node* CardTableBarrierSetC2::atomic_cmpxchg_val_at_resolved(C2AtomicParseAccess& access, Node* expected_val,
63 Node* new_val, const Type* value_type) const {
64 if (!access.is_oop()) {
65 return BarrierSetC2::atomic_cmpxchg_val_at_resolved(access, expected_val, new_val, value_type);
66 }
67
68 Node* result = BarrierSetC2::atomic_cmpxchg_val_at_resolved(access, expected_val, new_val, value_type);
69
70 post_barrier(access.kit(), access.base(), access.addr().node(), new_val, true);
71
72 return result;
73 }
74
75 Node* CardTableBarrierSetC2::atomic_cmpxchg_bool_at_resolved(C2AtomicParseAccess& access, Node* expected_val,
76 Node* new_val, const Type* value_type) const {
77 GraphKit* kit = access.kit();
78
79 if (!access.is_oop()) {
80 return BarrierSetC2::atomic_cmpxchg_bool_at_resolved(access, expected_val, new_val, value_type);
81 }
82
83 Node* load_store = BarrierSetC2::atomic_cmpxchg_bool_at_resolved(access, expected_val, new_val, value_type);
84
85 // Emit the post barrier only when the actual store happened. This makes sense
86 // to check only for LS_cmp_* that can fail to set the value.
87 // LS_cmp_exchange does not produce any branches by default, so there is no
88 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
89 // CompareAndExchange and move branches here, it would make sense to conditionalize
90 // post_barriers for LS_cmp_exchange as well.
91 //
92 // CAS success path is marked more likely since we anticipate this is a performance
93 // critical path, while CAS failure path can use the penalty for going through unlikely
94 // path as backoff. Which is still better than doing a store barrier there.
95 IdealKit ideal(kit);
96 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
97 kit->sync_kit(ideal);
98 post_barrier(kit, access.base(), access.addr().node(), new_val, true);
99 ideal.sync_kit(kit);
100 } ideal.end_if();
101 kit->final_sync(ideal);
102
103 return load_store;
104 }
105
106 Node* CardTableBarrierSetC2::atomic_xchg_at_resolved(C2AtomicParseAccess& access, Node* new_val, const Type* value_type) const {
107 Node* result = BarrierSetC2::atomic_xchg_at_resolved(access, new_val, value_type);
108 if (!access.is_oop()) {
109 return result;
110 }
111
112 post_barrier(access.kit(), access.base(), access.addr().node(), new_val, true);
113
114 return result;
115 }
116
117 Node* CardTableBarrierSetC2::byte_map_base_node(GraphKit* kit) const {
118 // Get base of card map
119 CardTable::CardValue* card_table_base = ci_card_table_address();
120 if (card_table_base != nullptr) {
121 return kit->makecon(TypeRawPtr::make((address)card_table_base));
122 } else {
123 return kit->null();
124 }
125 }
126
127 // vanilla post barrier
128 // Insert a write-barrier store. This is to let generational GC work; we have
129 // to flag all oop-stores before the next GC point.
130 void CardTableBarrierSetC2::post_barrier(GraphKit* kit,
131 Node* obj,
132 Node* adr,
133 Node* val,
134 bool use_precise) const {
135 // No store check needed if we're storing a null.
136 if (val != nullptr && val->is_Con()) {
137 const Type* t = val->bottom_type();
138 if (t == TypePtr::NULL_PTR || t == Type::TOP) {
139 return;
140 }
141 }
142
143 if (use_ReduceInitialCardMarks()
144 && obj == kit->just_allocated_object(kit->control())) {
145 // We can skip marks on a freshly-allocated object in Eden.
146 // Keep this code in sync with CardTableBarrierSet::on_slowpath_allocation_exit.
147 // That routine informs GC to take appropriate compensating steps,
148 // upon a slow-path allocation, so as to make this card-mark
149 // elision safe.
150 return;
151 }
152
153 if (!use_precise) {
154 // All card marks for a (non-array) instance are in one place:
155 adr = obj;
156 } else {
157 // Else it's an array (or unknown), and we want more precise card marks.
158 }
159
160 assert(adr != nullptr, "");
161
162 IdealKit ideal(kit, true);
163
164 // Convert the pointer to an int prior to doing math on it
165 Node* cast = __ CastPX(__ ctrl(), adr);
166
167 // Divide by card size
168 Node* card_offset = __ URShiftX(cast, __ ConI(CardTable::card_shift()));
169
170 // Combine card table base and card offset
171 Node* card_adr = __ AddP(__ top(), byte_map_base_node(kit), card_offset);
172
173 // Get the alias_index for raw card-mark memory
174 int adr_type = Compile::AliasIdxRaw;
175
176 // Dirty card value to store
177 Node* dirty = __ ConI(CardTable::dirty_card_val());
178
179 if (UseCondCardMark) {
180 // The classic GC reference write barrier is typically implemented
181 // as a store into the global card mark table. Unfortunately
182 // unconditional stores can result in false sharing and excessive
183 // coherence traffic as well as false transactional aborts.
184 // UseCondCardMark enables MP "polite" conditional card mark
185 // stores. In theory we could relax the load from ctrl() to
186 // no_ctrl, but that doesn't buy much latitude.
187 Node* card_val = __ load( __ ctrl(), card_adr, TypeInt::BYTE, T_BYTE, adr_type);
188 __ if_then(card_val, BoolTest::ne, dirty);
189 }
190
191 // Smash dirty value into card
192 __ store(__ ctrl(), card_adr, dirty, T_BYTE, adr_type, MemNode::unordered);
193
194 if (UseCondCardMark) {
195 __ end_if();
196 }
197
198 // Final sync IdealKit and GraphKit.
199 kit->final_sync(ideal);
200 }
201
202 bool CardTableBarrierSetC2::use_ReduceInitialCardMarks() {
203 return ReduceInitialCardMarks;
204 }
205
206 void CardTableBarrierSetC2::eliminate_gc_barrier(PhaseMacroExpand* macro, Node* node) const {
207 assert(node->Opcode() == Op_CastP2X, "ConvP2XNode required");
208 Node *shift = node->unique_out();
209 Node *addp = shift->unique_out();
210 for (DUIterator_Last jmin, j = addp->last_outs(jmin); j >= jmin; --j) {
211 Node *mem = addp->last_out(j);
212 if (UseCondCardMark && mem->is_Load()) {
213 assert(mem->Opcode() == Op_LoadB, "unexpected code shape");
214 // The load is checking if the card has been written so
215 // replace it with zero to fold the test.
216 macro->replace_node(mem, macro->intcon(0));
217 continue;
218 }
219 assert(mem->is_Store(), "store required");
220 macro->replace_node(mem, mem->in(MemNode::Memory));
221 }
222 }
223
224 bool CardTableBarrierSetC2::array_copy_requires_gc_barriers(bool tightly_coupled_alloc, BasicType type, bool is_clone, bool is_clone_instance, ArrayCopyPhase phase) const {
225 bool is_oop = is_reference_type(type);
226 return is_oop && (!tightly_coupled_alloc || !use_ReduceInitialCardMarks());
227 }