1 /*
2 * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "asm/assembler.inline.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "code/aotCodeCache.hpp"
29 #include "code/compiledIC.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "interp_masm_aarch64.hpp"
32 #include "memory/resourceArea.hpp"
33 #include "oops/instanceKlass.hpp"
34 #include "oops/klassVtable.hpp"
35 #include "runtime/sharedRuntime.hpp"
36 #include "vmreg_aarch64.inline.hpp"
37 #ifdef COMPILER2
38 #include "opto/runtime.hpp"
39 #endif
40
41 // machine-dependent part of VtableStubs: create VtableStub of correct size and
42 // initialize its code
43
44 #define __ masm->
45
46 #ifndef PRODUCT
47 extern "C" void bad_compiled_vtable_index(JavaThread* thread, oop receiver, int index);
48 #endif
49
50 VtableStub* VtableStubs::create_vtable_stub(int vtable_index, bool caller_is_c1) {
51 // Read "A word on VtableStub sizing" in share/code/vtableStubs.hpp for details on stub sizing.
52 const int stub_code_length = code_size_limit(true);
53 VtableStub* s = new(stub_code_length) VtableStub(true, vtable_index, caller_is_c1);
54 // Can be null if there is no free space in the code cache.
55 if (s == nullptr) {
56 return nullptr;
57 }
58
59 // Count unused bytes in instruction sequences of variable size.
60 // We add them to the computed buffer size in order to avoid
61 // overflow in subsequently generated stubs.
62 address start_pc;
63 int slop_bytes = 0;
64 int slop_delta = 0;
65
66 // No variance was detected in vtable stub sizes. Setting index_dependent_slop == 0 will unveil any deviation from this observation.
67 const int index_dependent_slop = 0;
68 ByteSize entry_offset = caller_is_c1 ? Method::from_compiled_inline_offset() : Method::from_compiled_inline_ro_offset();
69
70 ResourceMark rm;
71 CodeBuffer cb(s->entry_point(), stub_code_length);
72 MacroAssembler* masm = new MacroAssembler(&cb);
73
74 #if (!defined(PRODUCT) && defined(COMPILER2))
75 if (CountCompiledCalls) {
76 __ lea(r16, ExternalAddress((address) SharedRuntime::nof_megamorphic_calls_addr()));
77 __ increment(Address(r16));
78 }
79 #endif
80
81 // get receiver (need to skip return address on top of stack)
82 assert(VtableStub::receiver_location() == j_rarg0->as_VMReg(), "receiver expected in j_rarg0");
83
84 // get receiver klass
85 address npe_addr = __ pc();
86 __ load_klass(r16, j_rarg0);
87
88 #ifndef PRODUCT
89 if (DebugVtables) {
90 Label L;
91 // TODO: find upper bound for this debug code.
92 start_pc = __ pc();
93
94 // check offset vs vtable length
95 __ ldrw(rscratch1, Address(r16, Klass::vtable_length_offset()));
96 __ cmpw(rscratch1, vtable_index * vtableEntry::size());
97 __ br(Assembler::GT, L);
98 __ enter();
99 __ mov(r2, vtable_index);
100
101 __ call_VM(noreg, CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), j_rarg0, r2);
102 const ptrdiff_t estimate = 256;
103 const ptrdiff_t codesize = __ pc() - start_pc;
104 slop_delta = estimate - codesize; // call_VM varies in length, depending on data
105 slop_bytes += slop_delta;
106 assert(slop_delta >= 0, "vtable #%d: Code size estimate (%d) for DebugVtables too small, required: %d", vtable_index, (int)estimate, (int)codesize);
107
108 __ leave();
109 __ bind(L);
110 }
111 #endif // PRODUCT
112
113 start_pc = __ pc();
114 __ lookup_virtual_method(r16, vtable_index, rmethod);
115 slop_delta = 8 - (int)(__ pc() - start_pc);
116 slop_bytes += slop_delta;
117 assert(slop_delta >= 0, "negative slop(%d) encountered, adjust code size estimate!", slop_delta);
118
119 #ifndef PRODUCT
120 if (DebugVtables) {
121 Label L;
122 __ cbz(rmethod, L);
123 __ ldr(rscratch1, Address(rmethod, entry_offset));
124 __ cbnz(rscratch1, L);
125 __ stop("Vtable entry is null");
126 __ bind(L);
127 }
128 #endif // PRODUCT
129
130 // r0: receiver klass
131 // rmethod: Method*
132 // r2: receiver
133 address ame_addr = __ pc();
134 __ ldr(rscratch1, Address(rmethod, entry_offset));
135 __ br(rscratch1);
136
137 masm->flush();
138 slop_bytes += index_dependent_slop; // add'l slop for size variance due to large itable offsets
139 bookkeeping(masm, tty, s, npe_addr, ame_addr, true, vtable_index, slop_bytes, index_dependent_slop);
140
141 return s;
142 }
143
144
145 VtableStub* VtableStubs::create_itable_stub(int itable_index, bool caller_is_c1) {
146 // Read "A word on VtableStub sizing" in share/code/vtableStubs.hpp for details on stub sizing.
147 const int stub_code_length = code_size_limit(false);
148 VtableStub* s = new(stub_code_length) VtableStub(false, itable_index, caller_is_c1);
149 // Can be null if there is no free space in the code cache.
150 if (s == nullptr) {
151 return nullptr;
152 }
153
154 // Count unused bytes in instruction sequences of variable size.
155 // We add them to the computed buffer size in order to avoid
156 // overflow in subsequently generated stubs.
157 address start_pc;
158 int slop_bytes = 0;
159 int slop_delta = 0;
160
161 const int index_dependent_slop = (itable_index == 0) ? 4 : // code size change with transition from 8-bit to 32-bit constant (@index == 16).
162 (itable_index < 16) ? 3 : 0; // index == 0 generates even shorter code.
163 ByteSize entry_offset = caller_is_c1 ? Method::from_compiled_inline_offset() : Method::from_compiled_inline_ro_offset();
164
165 ResourceMark rm;
166 CodeBuffer cb(s->entry_point(), stub_code_length);
167 MacroAssembler* masm = new MacroAssembler(&cb);
168
169 #if (!defined(PRODUCT) && defined(COMPILER2))
170 if (CountCompiledCalls) {
171 __ lea(r10, ExternalAddress((address) SharedRuntime::nof_megamorphic_calls_addr()));
172 __ increment(Address(r10));
173 }
174 #endif
175
176 // get receiver (need to skip return address on top of stack)
177 assert(VtableStub::receiver_location() == j_rarg0->as_VMReg(), "receiver expected in j_rarg0");
178
179 // Entry arguments:
180 // rscratch2: CompiledICData
181 // j_rarg0: Receiver
182
183 // This stub is called from compiled code which has no callee-saved registers,
184 // so all registers except arguments are free at this point.
185 const Register recv_klass_reg = r10;
186 const Register holder_klass_reg = r16; // declaring interface klass (DEFC)
187 const Register resolved_klass_reg = r17; // resolved interface klass (REFC)
188 const Register temp_reg = r11;
189 const Register temp_reg2 = r15;
190 const Register icdata_reg = rscratch2;
191
192 Label L_no_such_interface;
193
194 __ ldr(resolved_klass_reg, Address(icdata_reg, CompiledICData::itable_refc_klass_offset()));
195 __ ldr(holder_klass_reg, Address(icdata_reg, CompiledICData::itable_defc_klass_offset()));
196
197 start_pc = __ pc();
198
199 // get receiver klass (also an implicit null-check)
200 address npe_addr = __ pc();
201 __ load_klass(recv_klass_reg, j_rarg0);
202
203 // Receiver subtype check against REFC.
204 // Get selected method from declaring class and itable index
205 __ lookup_interface_method_stub(recv_klass_reg, holder_klass_reg, resolved_klass_reg, rmethod,
206 temp_reg, temp_reg2, itable_index, L_no_such_interface);
207
208 // Reduce "estimate" such that "padding" does not drop below 8.
209 const ptrdiff_t estimate = AOTCodeCache::is_on_for_dump() ? 148 : 144;
210 const ptrdiff_t codesize = __ pc() - start_pc;
211 slop_delta = (int)(estimate - codesize);
212 slop_bytes += slop_delta;
213 assert(slop_delta >= 0, "itable #%d: Code size estimate (%d) for lookup_interface_method too small, required: %d", itable_index, (int)estimate, (int)codesize);
214
215 #ifdef ASSERT
216 if (DebugVtables) {
217 Label L2;
218 __ cbz(rmethod, L2);
219 __ ldr(rscratch1, Address(rmethod, entry_offset));
220 __ cbnz(rscratch1, L2);
221 __ stop("compiler entrypoint is null");
222 __ bind(L2);
223 }
224 #endif // ASSERT
225
226 // rmethod: Method*
227 // j_rarg0: receiver
228 address ame_addr = __ pc();
229 __ ldr(rscratch1, Address(rmethod, entry_offset));
230 __ br(rscratch1);
231
232 __ bind(L_no_such_interface);
233 // Handle IncompatibleClassChangeError in itable stubs.
234 // More detailed error message.
235 // We force resolving of the call site by jumping to the "handle
236 // wrong method" stub, and so let the interpreter runtime do all the
237 // dirty work.
238 assert(SharedRuntime::get_handle_wrong_method_stub() != nullptr, "check initialization order");
239 __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
240
241 masm->flush();
242 slop_bytes += index_dependent_slop; // add'l slop for size variance due to large itable offsets
243 bookkeeping(masm, tty, s, npe_addr, ame_addr, false, itable_index, slop_bytes, index_dependent_slop);
244
245 return s;
246 }
247
248 int VtableStub::pd_code_alignment() {
249 // aarch64 cache line size is not an architected constant. We just align on 4 bytes (instruction size).
250 const unsigned int icache_line_size = 4;
251 return icache_line_size;
252 }