1 /*
  2  * Copyright (c) 2003, 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 "asm/macroAssembler.hpp"
 26 #include "code/compiledIC.hpp"
 27 #include "code/vtableStubs.hpp"
 28 #include "interp_masm_x86.hpp"
 29 #include "memory/resourceArea.hpp"
 30 #include "oops/instanceKlass.hpp"
 31 #include "oops/klassVtable.hpp"
 32 #include "runtime/sharedRuntime.hpp"
 33 #include "vmreg_x86.inline.hpp"
 34 #ifdef COMPILER2
 35 #include "opto/runtime.hpp"
 36 #endif
 37 
 38 // machine-dependent part of VtableStubs: create VtableStub of correct size and
 39 // initialize its code
 40 
 41 #define __ masm->
 42 
 43 #ifndef PRODUCT
 44 extern "C" void bad_compiled_vtable_index(JavaThread* thread, oop receiver, int index);
 45 #endif
 46 
 47 VtableStub* VtableStubs::create_vtable_stub(int vtable_index, bool caller_is_c1) {
 48   // Read "A word on VtableStub sizing" in share/code/vtableStubs.hpp for details on stub sizing.
 49   const int stub_code_length = code_size_limit(true);
 50   VtableStub* s = new(stub_code_length) VtableStub(true, vtable_index, caller_is_c1);
 51   // Can be nullptr if there is no free space in the code cache.
 52   if (s == nullptr) {
 53     return nullptr;
 54   }
 55 
 56   // Count unused bytes in instruction sequences of variable size.
 57   // We add them to the computed buffer size in order to avoid
 58   // overflow in subsequently generated stubs.
 59   address   start_pc;
 60   int       slop_bytes = 0;
 61   int       slop_delta = 0;
 62   // No variance was detected in vtable stub sizes. Setting index_dependent_slop == 0 will unveil any deviation from this observation.
 63   const int index_dependent_slop     = 0;
 64   ByteSize  entry_offset = caller_is_c1 ? Method::from_compiled_inline_offset() :  Method::from_compiled_inline_ro_offset();
 65 
 66   ResourceMark    rm;
 67   CodeBuffer      cb(s->entry_point(), stub_code_length);
 68   MacroAssembler* masm = new MacroAssembler(&cb);
 69 
 70 #if (!defined(PRODUCT) && defined(COMPILER2))
 71   if (CountCompiledCalls) {
 72     __ incrementq(ExternalAddress(SharedRuntime::nof_megamorphic_calls_addr()), rscratch1);
 73   }
 74 #endif
 75 
 76   // get receiver (need to skip return address on top of stack)
 77   assert(VtableStub::receiver_location() == j_rarg0->as_VMReg(), "receiver expected in j_rarg0");
 78 
 79   // Free registers (non-args) are rax, rbx
 80 
 81   // get receiver klass
 82   address npe_addr = __ pc();
 83   __ load_klass(rax, j_rarg0, rscratch1);
 84 
 85 #ifndef PRODUCT
 86   if (DebugVtables) {
 87     Label L;
 88     start_pc = __ pc();
 89     // check offset vs vtable length
 90     __ cmpl(Address(rax, Klass::vtable_length_offset()), vtable_index*vtableEntry::size());
 91     slop_delta  = 12 - (__ pc() - start_pc);  // cmpl varies in length, depending on data
 92     slop_bytes += slop_delta;
 93     assert(slop_delta >= 0, "negative slop(%d) encountered, adjust code size estimate!", slop_delta);
 94 
 95     __ jcc(Assembler::greater, L);
 96     __ movl(rbx, vtable_index);
 97     // VTABLE TODO: find upper bound for call_VM length.
 98     start_pc = __ pc();
 99     __ call_VM(noreg, CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), j_rarg0, rbx);
100     slop_delta  = 550 - (__ pc() - start_pc);
101     slop_bytes += slop_delta;
102     assert(slop_delta >= 0, "negative slop(%d) encountered, adjust code size estimate!", slop_delta);
103     __ bind(L);
104   }
105 #endif // PRODUCT
106 
107   const Register method = rbx;
108 
109   // load Method* and target address
110   start_pc = __ pc();
111   __ lookup_virtual_method(rax, vtable_index, method);
112   slop_delta  = 8 - (int)(__ pc() - start_pc);
113   slop_bytes += slop_delta;
114   assert(slop_delta >= 0, "negative slop(%d) encountered, adjust code size estimate!", slop_delta);
115 
116 #ifndef PRODUCT
117   if (DebugVtables) {
118     Label L;
119     __ cmpptr(method, NULL_WORD);
120     __ jcc(Assembler::equal, L);
121     __ cmpptr(Address(method, entry_offset), NULL_WORD);
122     __ jcc(Assembler::notZero, L);
123     __ stop("Vtable entry is null");
124     __ bind(L);
125   }
126 #endif // PRODUCT
127 
128   // rax: receiver klass
129   // method (rbx): Method*
130   // rcx: receiver
131   address ame_addr = __ pc();
132   __ jmp( Address(rbx, entry_offset));
133 
134   masm->flush();
135   slop_bytes += index_dependent_slop; // add'l slop for size variance due to large itable offsets
136   bookkeeping(masm, tty, s, npe_addr, ame_addr, true, vtable_index, slop_bytes, index_dependent_slop);
137 
138   return s;
139 }
140 
141 
142 VtableStub* VtableStubs::create_itable_stub(int itable_index, bool caller_is_c1) {
143   // Read "A word on VtableStub sizing" in share/code/vtableStubs.hpp for details on stub sizing.
144   const int stub_code_length = code_size_limit(false);
145   ByteSize  entry_offset = caller_is_c1 ? Method::from_compiled_inline_offset() :  Method::from_compiled_inline_ro_offset();
146   VtableStub* s = new(stub_code_length) VtableStub(false, itable_index, caller_is_c1);
147   // Can be nullptr if there is no free space in the code cache.
148   if (s == nullptr) {
149     return nullptr;
150   }
151 
152   // Count unused bytes in instruction sequences of variable size.
153   // We add them to the computed buffer size in order to avoid
154   // overflow in subsequently generated stubs.
155   address   start_pc;
156   int       slop_bytes = 0;
157   int       slop_delta = 0;
158   const int index_dependent_slop = (itable_index == 0) ? 4 :     // code size change with transition from 8-bit to 32-bit constant (@index == 16).
159                                    (itable_index < 16) ? 3 : 0;  // index == 0 generates even shorter code.
160 
161   ResourceMark    rm;
162   CodeBuffer      cb(s->entry_point(), stub_code_length);
163   MacroAssembler *masm = new MacroAssembler(&cb);
164 
165 #if (!defined(PRODUCT) && defined(COMPILER2))
166   if (CountCompiledCalls) {
167     __ incrementq(ExternalAddress(SharedRuntime::nof_megamorphic_calls_addr()), rscratch1);
168   }
169 #endif // PRODUCT
170 
171   // Entry arguments:
172   //  rax: CompiledICData
173   //  j_rarg0: Receiver
174 
175   // Most registers are in use; we'll use rax, rbx, r10, r11
176   // (various calling sequences use r[cd]x, r[sd]i, r[89]; stay away from them)
177   const Register recv_klass_reg     = r10;
178   const Register holder_klass_reg   = rax; // declaring interface klass (DEFC)
179   const Register resolved_klass_reg = r14; // resolved interface klass (REFC)
180   const Register temp_reg           = r11;
181   const Register temp_reg2          = r13;
182   const Register method             = rbx;
183   const Register icdata_reg         = rax;
184 
185   __ movptr(resolved_klass_reg, Address(icdata_reg, CompiledICData::itable_refc_klass_offset()));
186   __ movptr(holder_klass_reg,   Address(icdata_reg, CompiledICData::itable_defc_klass_offset()));
187 
188   Label L_no_such_interface;
189 
190   // get receiver klass (also an implicit null-check)
191   assert(VtableStub::receiver_location() == j_rarg0->as_VMReg(), "receiver expected in j_rarg0");
192   address npe_addr = __ pc();
193   __ load_klass(recv_klass_reg, j_rarg0, temp_reg);
194 
195   start_pc = __ pc();
196 
197   // Receiver subtype check against REFC.
198   // Get selected method from declaring class and itable index
199   __ lookup_interface_method_stub(recv_klass_reg, // input
200                                   holder_klass_reg, // input
201                                   resolved_klass_reg, // input
202                                   method, // output
203                                   temp_reg,
204                                   temp_reg2,
205                                   noreg,
206                                   itable_index,
207                                   L_no_such_interface);
208 
209   const ptrdiff_t  lookupSize = __ pc() - start_pc;
210 
211   // We expect we need index_dependent_slop extra bytes. Reason:
212   // The emitted code in lookup_interface_method changes when itable_index exceeds 15.
213   // For linux, a very narrow estimate would be 112, but Solaris requires some more space (130).
214   const ptrdiff_t estimate = 144;
215   const ptrdiff_t codesize = lookupSize + index_dependent_slop;
216   slop_delta  = (int)(estimate - codesize);
217   slop_bytes += slop_delta;
218   assert(slop_delta >= 0, "itable #%d: Code size estimate (%d) for lookup_interface_method too small, required: %d", itable_index, (int)estimate, (int)codesize);
219 
220   // If we take a trap while this arg is on the stack we will not
221   // be able to walk the stack properly. This is not an issue except
222   // when there are mistakes in this assembly code that could generate
223   // a spurious fault. Ask me how I know...
224 
225   // method (rbx): Method*
226   // j_rarg0: receiver
227 
228 #ifdef ASSERT
229   if (DebugVtables) {
230     Label L2;
231     __ cmpptr(method, NULL_WORD);
232     __ jcc(Assembler::equal, L2);
233     __ cmpptr(Address(method, entry_offset), NULL_WORD);
234     __ jcc(Assembler::notZero, L2);
235     __ stop("compiler entrypoint is null");
236     __ bind(L2);
237   }
238 #endif // ASSERT
239 
240   address ame_addr = __ pc();
241   __ jmp(Address(method, entry_offset));
242 
243   __ bind(L_no_such_interface);
244   // Handle IncompatibleClassChangeError in itable stubs.
245   // More detailed error message.
246   // We force resolving of the call site by jumping to the "handle
247   // wrong method" stub, and so let the interpreter runtime do all the
248   // dirty work.
249   __ jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
250 
251   masm->flush();
252   slop_bytes += index_dependent_slop; // add'l slop for size variance due to large itable offsets
253   bookkeeping(masm, tty, s, npe_addr, ame_addr, false, itable_index, slop_bytes, index_dependent_slop);
254 
255   return s;
256 }
257 
258 int VtableStub::pd_code_alignment() {
259   // x86 cache line size is 64 bytes, but we want to limit alignment loss.
260   const unsigned int icache_line_size = wordSize;
261   return icache_line_size;
262 }