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 "code/aotCodeCache.hpp"
27 #include "gc/shared/c2/cardTableBarrierSetC2.hpp"
28 #include "gc/shared/cardTable.hpp"
29 #include "gc/shared/cardTableBarrierSet.hpp"
30 #include "gc/shared/gc_globals.hpp"
31 #include "opto/arraycopynode.hpp"
32 #include "opto/graphKit.hpp"
33 #include "opto/idealKit.hpp"
34 #include "opto/macro.hpp"
35 #include "utilities/macros.hpp"
36
37 #define __ ideal.
38
39 Node* CardTableBarrierSetC2::store_at_resolved(C2Access& access, C2AccessValue& val) const {
40 DecoratorSet decorators = access.decorators();
41
42 Node* adr = access.addr().node();
43
44 bool is_array = (decorators & IS_ARRAY) != 0;
45 bool anonymous = (decorators & ON_UNKNOWN_OOP_REF) != 0;
46 bool in_heap = (decorators & IN_HEAP) != 0;
47 bool use_precise = is_array || anonymous;
48 bool tightly_coupled_alloc = (decorators & C2_TIGHTLY_COUPLED_ALLOC) != 0;
49
50 if (!access.is_oop() || tightly_coupled_alloc || (!in_heap && !anonymous)) {
51 return BarrierSetC2::store_at_resolved(access, val);
52 }
53
54 assert(access.is_parse_access(), "entry not supported at optimization time");
55 C2ParseAccess& parse_access = static_cast<C2ParseAccess&>(access);
56
57 Node* store = BarrierSetC2::store_at_resolved(access, val);
58 post_barrier(parse_access.kit(), access.base(), adr, val.node(), use_precise);
59
60 return store;
61 }
62
63 Node* CardTableBarrierSetC2::atomic_cmpxchg_val_at_resolved(C2AtomicParseAccess& access, Node* expected_val,
64 Node* new_val, const Type* value_type) const {
65 if (!access.is_oop()) {
66 return BarrierSetC2::atomic_cmpxchg_val_at_resolved(access, expected_val, new_val, value_type);
67 }
68
69 Node* result = BarrierSetC2::atomic_cmpxchg_val_at_resolved(access, expected_val, new_val, value_type);
70
71 post_barrier(access.kit(), access.base(), access.addr().node(), new_val, true);
72
73 return result;
74 }
75
76 Node* CardTableBarrierSetC2::atomic_cmpxchg_bool_at_resolved(C2AtomicParseAccess& access, Node* expected_val,
77 Node* new_val, const Type* value_type) const {
78 GraphKit* kit = access.kit();
79
80 if (!access.is_oop()) {
81 return BarrierSetC2::atomic_cmpxchg_bool_at_resolved(access, expected_val, new_val, value_type);
82 }
83
84 Node* load_store = BarrierSetC2::atomic_cmpxchg_bool_at_resolved(access, expected_val, new_val, value_type);
85
86 // Emit the post barrier only when the actual store happened. This makes sense
87 // to check only for LS_cmp_* that can fail to set the value.
88 // LS_cmp_exchange does not produce any branches by default, so there is no
89 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
90 // CompareAndExchange and move branches here, it would make sense to conditionalize
91 // post_barriers for LS_cmp_exchange as well.
92 //
93 // CAS success path is marked more likely since we anticipate this is a performance
94 // critical path, while CAS failure path can use the penalty for going through unlikely
95 // path as backoff. Which is still better than doing a store barrier there.
96 IdealKit ideal(kit);
97 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
98 kit->sync_kit(ideal);
99 post_barrier(kit, access.base(), access.addr().node(), new_val, true);
100 ideal.sync_kit(kit);
101 } ideal.end_if();
102 kit->final_sync(ideal);
103
104 return load_store;
105 }
106
107 Node* CardTableBarrierSetC2::atomic_xchg_at_resolved(C2AtomicParseAccess& access, Node* new_val, const Type* value_type) const {
108 Node* result = BarrierSetC2::atomic_xchg_at_resolved(access, new_val, value_type);
109 if (!access.is_oop()) {
110 return result;
111 }
112
113 post_barrier(access.kit(), access.base(), access.addr().node(), new_val, true);
114
115 return result;
116 }
117
118 Node* CardTableBarrierSetC2::byte_map_base_node(IdealKit* kit) const {
119 // Get base of card map
120 #if INCLUDE_CDS
121 if (AOTCodeCache::is_on_for_dump()) {
122 // load the card table address from the AOT Runtime Constants area
123 Node* byte_map_base_adr = kit->makecon(TypeRawPtr::make(AOTRuntimeConstants::card_table_address()));
124 return kit->load_aot_const(byte_map_base_adr, TypeRawPtr::NOTNULL);
125 }
126 #endif
127 CardTable::CardValue* card_table_base = ci_card_table_address();
128 if (card_table_base != nullptr) {
129 return kit->makecon(TypeRawPtr::make((address)card_table_base));
130 } else {
131 return kit->makecon(Type::get_zero_type(T_ADDRESS));
132 }
133 }
134
135 // vanilla post barrier
136 // Insert a write-barrier store. This is to let generational GC work; we have
137 // to flag all oop-stores before the next GC point.
138 void CardTableBarrierSetC2::post_barrier(GraphKit* kit,
139 Node* obj,
140 Node* adr,
141 Node* val,
142 bool use_precise) const {
143 // No store check needed if we're storing a null.
144 if (val != nullptr && val->is_Con()) {
145 const Type* t = val->bottom_type();
146 if (t == TypePtr::NULL_PTR || t == Type::TOP) {
147 return;
148 }
149 }
150
151 if (use_ReduceInitialCardMarks()
152 && obj == kit->just_allocated_object(kit->control())) {
153 // We can skip marks on a freshly-allocated object in Eden.
154 // Keep this code in sync with CardTableBarrierSet::on_slowpath_allocation_exit.
155 // That routine informs GC to take appropriate compensating steps,
156 // upon a slow-path allocation, so as to make this card-mark
157 // elision safe.
158 return;
159 }
160
161 if (!use_precise) {
162 // All card marks for a (non-array) instance are in one place:
163 adr = obj;
164 } else {
165 // Else it's an array (or unknown), and we want more precise card marks.
166 }
167
168 assert(adr != nullptr, "");
169
170 IdealKit ideal(kit, true);
171
172 // Convert the pointer to an int prior to doing math on it
173 Node* cast = __ CastPX(__ ctrl(), adr);
174
175 // Divide by card size
176 Node* card_offset = __ URShiftX(cast, __ ConI(CardTable::card_shift()));
177
178 // Combine card table base and card offset
179 Node* card_adr = __ AddP(__ top(), byte_map_base_node(&ideal), card_offset);
180
181 // Get the alias_index for raw card-mark memory
182 int adr_type = Compile::AliasIdxRaw;
183
184 // Dirty card value to store
185 Node* dirty = __ ConI(CardTable::dirty_card_val());
186
187 if (UseCondCardMark) {
188 // The classic GC reference write barrier is typically implemented
189 // as a store into the global card mark table. Unfortunately
190 // unconditional stores can result in false sharing and excessive
191 // coherence traffic as well as false transactional aborts.
192 // UseCondCardMark enables MP "polite" conditional card mark
193 // stores. In theory we could relax the load from ctrl() to
194 // no_ctrl, but that doesn't buy much latitude.
195 Node* card_val = __ load( __ ctrl(), card_adr, TypeInt::BYTE, T_BYTE, adr_type);
196 __ if_then(card_val, BoolTest::ne, dirty);
197 }
198
199 // Smash dirty value into card
200 __ store(__ ctrl(), card_adr, dirty, T_BYTE, adr_type, MemNode::unordered);
201
202 if (UseCondCardMark) {
203 __ end_if();
204 }
205
206 // Final sync IdealKit and GraphKit.
207 kit->final_sync(ideal);
208 }
209
210 bool CardTableBarrierSetC2::use_ReduceInitialCardMarks() {
211 return ReduceInitialCardMarks;
212 }
213
214 void CardTableBarrierSetC2::eliminate_gc_barrier(PhaseMacroExpand* macro, Node* node) const {
215 assert(node->Opcode() == Op_CastP2X, "ConvP2XNode required");
216 Node *shift = node->unique_out();
217 Node *addp = shift->unique_out();
218 for (DUIterator_Last jmin, j = addp->last_outs(jmin); j >= jmin; --j) {
219 Node *mem = addp->last_out(j);
220 if (UseCondCardMark && mem->is_Load()) {
221 assert(mem->Opcode() == Op_LoadB, "unexpected code shape");
222 // The load is checking if the card has been written so
223 // replace it with zero to fold the test.
224 macro->replace_node(mem, macro->intcon(0));
225 continue;
226 }
227 assert(mem->is_Store(), "store required");
228 macro->replace_node(mem, mem->in(MemNode::Memory));
229 }
230 }
231
232 bool CardTableBarrierSetC2::array_copy_requires_gc_barriers(bool tightly_coupled_alloc, BasicType type, bool is_clone, bool is_clone_instance, ArrayCopyPhase phase) const {
233 bool is_oop = is_reference_type(type);
234 return is_oop && (!tightly_coupled_alloc || !use_ReduceInitialCardMarks());
235 }