1 /*
  2  * Copyright (c) 2018, 2026, 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   const InlineTypeNode* vt = nullptr;
 51   if (access.is_parse_access() && static_cast<C2ParseAccess&>(access).vt() != nullptr) {
 52     vt = static_cast<C2ParseAccess&>(access).vt();
 53   }
 54 
 55   if (vt == nullptr && (!access.is_oop() || tightly_coupled_alloc || (!in_heap && !anonymous))) {
 56     return BarrierSetC2::store_at_resolved(access, val);
 57   }
 58 
 59   assert(access.is_parse_access(), "entry not supported at optimization time");
 60   C2ParseAccess& parse_access = static_cast<C2ParseAccess&>(access);
 61 
 62   Node* store = BarrierSetC2::store_at_resolved(access, val);
 63   // TODO 8350865
 64   // - We actually only need the post barrier once for non-arrays (same for C1, right)?
 65   // - Value is only needed to determine if we are storing null. Maybe we can go with a simple boolean?
 66   GraphKit* kit = parse_access.kit();
 67   if (vt != nullptr) {
 68     for (uint i = 0; i < vt->field_count(); ++i) {
 69       ciField* field = vt->field(i);
 70       ciType* type = field->type();
 71       if (!type->is_primitive_type()) {
 72         ciInlineKlass* vk = vt->bottom_type()->inline_klass();
 73         int field_offset = field->offset_in_bytes() - vk->payload_offset();
 74         Node* value = vt->field_value(i);
 75         Node* field_adr = kit->basic_plus_adr(access.base(), adr, field_offset);
 76         post_barrier(kit, access.base(), field_adr, value, use_precise);
 77       }
 78     }
 79   } else {
 80     post_barrier(kit, access.base(), adr, val.node(), use_precise);
 81   }
 82 
 83   return store;
 84 }
 85 
 86 Node* CardTableBarrierSetC2::atomic_cmpxchg_val_at_resolved(C2AtomicParseAccess& access, Node* expected_val,
 87                                                             Node* new_val, const Type* value_type) const {
 88   if (!access.is_oop()) {
 89     return BarrierSetC2::atomic_cmpxchg_val_at_resolved(access, expected_val, new_val, value_type);
 90   }
 91 
 92   Node* result = BarrierSetC2::atomic_cmpxchg_val_at_resolved(access, expected_val, new_val, value_type);
 93 
 94   post_barrier(access.kit(), access.base(), access.addr().node(), new_val, true);
 95 
 96   return result;
 97 }
 98 
 99 Node* CardTableBarrierSetC2::atomic_cmpxchg_bool_at_resolved(C2AtomicParseAccess& access, Node* expected_val,
100                                                              Node* new_val, const Type* value_type) const {
101   GraphKit* kit = access.kit();
102 
103   if (!access.is_oop()) {
104     return BarrierSetC2::atomic_cmpxchg_bool_at_resolved(access, expected_val, new_val, value_type);
105   }
106 
107   Node* load_store = BarrierSetC2::atomic_cmpxchg_bool_at_resolved(access, expected_val, new_val, value_type);
108 
109   // Emit the post barrier only when the actual store happened. This makes sense
110   // to check only for LS_cmp_* that can fail to set the value.
111   // LS_cmp_exchange does not produce any branches by default, so there is no
112   // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
113   // CompareAndExchange and move branches here, it would make sense to conditionalize
114   // post_barriers for LS_cmp_exchange as well.
115   //
116   // CAS success path is marked more likely since we anticipate this is a performance
117   // critical path, while CAS failure path can use the penalty for going through unlikely
118   // path as backoff. Which is still better than doing a store barrier there.
119   IdealKit ideal(kit);
120   ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
121     kit->sync_kit(ideal);
122     post_barrier(kit, access.base(), access.addr().node(), new_val, true);
123     ideal.sync_kit(kit);
124   } ideal.end_if();
125   kit->final_sync(ideal);
126 
127   return load_store;
128 }
129 
130 Node* CardTableBarrierSetC2::atomic_xchg_at_resolved(C2AtomicParseAccess& access, Node* new_val, const Type* value_type) const {
131   Node* result = BarrierSetC2::atomic_xchg_at_resolved(access, new_val, value_type);
132   if (!access.is_oop()) {
133     return result;
134   }
135 
136   post_barrier(access.kit(), access.base(), access.addr().node(), new_val, true);
137 
138   return result;
139 }
140 
141 Node* CardTableBarrierSetC2::byte_map_base_node(IdealKit* kit) const {
142   // Get base of card map
143 #if INCLUDE_CDS
144   if (AOTCodeCache::is_on_for_dump()) {
145     // load the card table address from the AOT Runtime Constants area
146     Node* byte_map_base_adr = kit->makecon(TypeRawPtr::make(AOTRuntimeConstants::card_table_base_address()));
147     return kit->load_aot_const(byte_map_base_adr, TypeRawPtr::NOTNULL);
148   }
149 #endif
150   CardTable::CardValue* card_table_base = ci_card_table_address_const();
151    if (card_table_base != nullptr) {
152      return kit->makecon(TypeRawPtr::make((address)card_table_base));
153    } else {
154      return kit->makecon(Type::get_zero_type(T_ADDRESS));
155    }
156 }
157 
158 // vanilla post barrier
159 // Insert a write-barrier store.  This is to let generational GC work; we have
160 // to flag all oop-stores before the next GC point.
161 void CardTableBarrierSetC2::post_barrier(GraphKit* kit,
162                                          Node* obj,
163                                          Node* adr,
164                                          Node* val,
165                                          bool use_precise) const {
166   // No store check needed if we're storing a null.
167   if (val != nullptr && val->is_Con()) {
168     const Type* t = val->bottom_type();
169     if (t == TypePtr::NULL_PTR || t == Type::TOP) {
170       return;
171     }
172   }
173 
174   if (use_ReduceInitialCardMarks()
175       && obj == kit->just_allocated_object(kit->control())) {
176     // We can skip marks on a freshly-allocated object in Eden.
177     // Keep this code in sync with CardTableBarrierSet::on_slowpath_allocation_exit.
178     // That routine informs GC to take appropriate compensating steps,
179     // upon a slow-path allocation, so as to make this card-mark
180     // elision safe.
181     return;
182   }
183 
184   if (!use_precise) {
185     // All card marks for a (non-array) instance are in one place:
186     adr = obj;
187   } else {
188     // Else it's an array (or unknown), and we want more precise card marks.
189   }
190 
191   assert(adr != nullptr, "");
192 
193   IdealKit ideal(kit, true);
194 
195   // Convert the pointer to an int prior to doing math on it
196   Node* cast = __ CastPX(__ ctrl(), adr);
197 
198   // Divide by card size
199   Node* card_offset = __ URShiftX(cast, __ ConI(CardTable::card_shift()));
200 
201   // Combine card table base and card offset
202   Node* card_adr = __ AddP(__ top(), byte_map_base_node(&ideal), card_offset);
203 
204   // Get the alias_index for raw card-mark memory
205   int adr_type = Compile::AliasIdxRaw;
206 
207   // Dirty card value to store
208   Node* dirty = __ ConI(CardTable::dirty_card_val());
209 
210   if (UseCondCardMark) {
211     // The classic GC reference write barrier is typically implemented
212     // as a store into the global card mark table.  Unfortunately
213     // unconditional stores can result in false sharing and excessive
214     // coherence traffic as well as false transactional aborts.
215     // UseCondCardMark enables MP "polite" conditional card mark
216     // stores.  In theory we could relax the load from ctrl() to
217     // no_ctrl, but that doesn't buy much latitude.
218     Node* card_val = __ load( __ ctrl(), card_adr, TypeInt::BYTE, T_BYTE, adr_type);
219     __ if_then(card_val, BoolTest::ne, dirty);
220   }
221 
222   // Smash dirty value into card
223   __ store(__ ctrl(), card_adr, dirty, T_BYTE, adr_type, MemNode::unordered);
224 
225   if (UseCondCardMark) {
226     __ end_if();
227   }
228 
229   // Final sync IdealKit and GraphKit.
230   kit->final_sync(ideal);
231 }
232 
233 bool CardTableBarrierSetC2::use_ReduceInitialCardMarks() {
234   return ReduceInitialCardMarks;
235 }
236 
237 void CardTableBarrierSetC2::eliminate_gc_barrier(PhaseIterGVN* igvn, Node* node) const {
238   assert(node->Opcode() == Op_CastP2X, "ConvP2XNode required");
239   for (DUIterator_Last imin, i = node->last_outs(imin); i >= imin; --i) {
240     Node* shift = node->last_out(i);
241     for (DUIterator_Last jmin, j = shift->last_outs(jmin); j >= jmin; --j) {
242       Node* addp = shift->last_out(j);
243       for (DUIterator_Last kmin, k = addp->last_outs(kmin); k >= kmin; --k) {
244         Node* mem = addp->last_out(k);
245         if (UseCondCardMark && mem->is_Load()) {
246           assert(mem->Opcode() == Op_LoadB, "unexpected code shape");
247           // The load is checking if the card has been written so
248           // replace it with zero to fold the test.
249           igvn->replace_node(mem, igvn->intcon(0));
250           continue;
251         }
252         assert(mem->is_Store(), "store required");
253         igvn->replace_node(mem, mem->in(MemNode::Memory));
254       }
255     }
256   }
257 }
258 
259 bool CardTableBarrierSetC2::array_copy_requires_gc_barriers(bool tightly_coupled_alloc, BasicType type, bool is_clone, bool is_clone_instance, ArrayCopyPhase phase) const {
260   bool is_oop = type == T_OBJECT || type == T_ARRAY;
261   return is_oop && (!tightly_coupled_alloc || !use_ReduceInitialCardMarks());
262 }