1 /*
  2  * Copyright (c) 2003, 2023, 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 "precompiled.hpp"
 26 #include "asm/macroAssembler.hpp"
 27 #include "code/compiledIC.hpp"
 28 #include "code/vtableStubs.hpp"
 29 #include "interp_masm_x86.hpp"
 30 #include "memory/resourceArea.hpp"
 31 #include "oops/instanceKlass.hpp"
 32 #include "oops/klassVtable.hpp"
 33 #include "runtime/sharedRuntime.hpp"
 34 #include "vmreg_x86.inline.hpp"
 35 #ifdef COMPILER2
 36 #include "opto/runtime.hpp"
 37 #endif
 38 
 39 // machine-dependent part of VtableStubs: create VtableStub of correct size and
 40 // initialize its code
 41 
 42 #define __ masm->
 43 
 44 #ifndef PRODUCT
 45 extern "C" void bad_compiled_vtable_index(JavaThread* thread, oop receiver, int index);
 46 #endif
 47 
 48 VtableStub* VtableStubs::create_vtable_stub(int vtable_index, bool caller_is_c1) {
 49   // Read "A word on VtableStub sizing" in share/code/vtableStubs.hpp for details on stub sizing.
 50   const int stub_code_length = code_size_limit(true);
 51   VtableStub* s = new(stub_code_length) VtableStub(true, vtable_index, caller_is_c1);
 52   // Can be nullptr if there is no free space in the code cache.
 53   if (s == nullptr) {
 54     return nullptr;
 55   }
 56 
 57   // Count unused bytes in instruction sequences of variable size.
 58   // We add them to the computed buffer size in order to avoid
 59   // overflow in subsequently generated stubs.
 60   address   start_pc;
 61   int       slop_bytes = 0;
 62   int       slop_delta = 0;
 63   // No variance was detected in vtable stub sizes. Setting index_dependent_slop == 0 will unveil any deviation from this observation.
 64   const int index_dependent_slop     = 0;
 65   ByteSize  entry_offset = caller_is_c1 ? Method::from_compiled_inline_offset() :  Method::from_compiled_inline_ro_offset();
 66 
 67   ResourceMark    rm;
 68   CodeBuffer      cb(s->entry_point(), stub_code_length);
 69   MacroAssembler* masm = new MacroAssembler(&cb);
 70 
 71 #if (!defined(PRODUCT) && defined(COMPILER2))
 72   if (CountCompiledCalls) {
 73     __ incrementq(ExternalAddress(SharedRuntime::nof_megamorphic_calls_addr()), rscratch1);
 74   }
 75 #endif
 76 
 77   // get receiver (need to skip return address on top of stack)
 78   assert(VtableStub::receiver_location() == j_rarg0->as_VMReg(), "receiver expected in j_rarg0");
 79 
 80   // Free registers (non-args) are rax, rbx
 81 
 82   // get receiver klass
 83   address npe_addr = __ pc();
 84   __ load_klass(rax, j_rarg0, rscratch1);
 85 
 86 #ifndef PRODUCT
 87   if (DebugVtables) {
 88     Label L;
 89     start_pc = __ pc();
 90     // check offset vs vtable length
 91     __ cmpl(Address(rax, Klass::vtable_length_offset()), vtable_index*vtableEntry::size());
 92     slop_delta  = 12 - (__ pc() - start_pc);  // cmpl varies in length, depending on data
 93     slop_bytes += slop_delta;
 94     assert(slop_delta >= 0, "negative slop(%d) encountered, adjust code size estimate!", slop_delta);
 95 
 96     __ jcc(Assembler::greater, L);
 97     __ movl(rbx, vtable_index);
 98     // VTABLE TODO: find upper bound for call_VM length.
 99     start_pc = __ pc();
100     __ call_VM(noreg, CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), j_rarg0, rbx);
101     slop_delta  = 550 - (__ pc() - start_pc);
102     slop_bytes += slop_delta;
103     assert(slop_delta >= 0, "negative slop(%d) encountered, adjust code size estimate!", slop_delta);
104     __ bind(L);
105   }
106 #endif // PRODUCT
107 
108   const Register method = rbx;
109 
110   // load Method* and target address
111   start_pc = __ pc();
112   __ lookup_virtual_method(rax, vtable_index, method);
113   slop_delta  = 8 - (int)(__ pc() - start_pc);
114   slop_bytes += slop_delta;
115   assert(slop_delta >= 0, "negative slop(%d) encountered, adjust code size estimate!", slop_delta);
116 
117 #ifndef PRODUCT
118   if (DebugVtables) {
119     Label L;
120     __ cmpptr(method, NULL_WORD);
121     __ jcc(Assembler::equal, L);
122     __ cmpptr(Address(method, entry_offset), NULL_WORD);
123     __ jcc(Assembler::notZero, L);
124     __ stop("Vtable entry is null");
125     __ bind(L);
126   }
127 #endif // PRODUCT
128 
129   // rax: receiver klass
130   // method (rbx): Method*
131   // rcx: receiver
132   address ame_addr = __ pc();
133   __ jmp( Address(rbx, entry_offset));
134 
135   masm->flush();
136   slop_bytes += index_dependent_slop; // add'l slop for size variance due to large itable offsets
137   bookkeeping(masm, tty, s, npe_addr, ame_addr, true, vtable_index, slop_bytes, index_dependent_slop);
138 
139   return s;
140 }
141 
142 
143 VtableStub* VtableStubs::create_itable_stub(int itable_index, bool caller_is_c1) {
144   // Read "A word on VtableStub sizing" in share/code/vtableStubs.hpp for details on stub sizing.
145   const int stub_code_length = code_size_limit(false);
146   ByteSize  entry_offset = caller_is_c1 ? Method::from_compiled_inline_offset() :  Method::from_compiled_inline_ro_offset();
147   VtableStub* s = new(stub_code_length) VtableStub(false, itable_index, caller_is_c1);
148   // Can be nullptr if there is no free space in the code cache.
149   if (s == nullptr) {
150     return nullptr;
151   }
152 
153   // Count unused bytes in instruction sequences of variable size.
154   // We add them to the computed buffer size in order to avoid
155   // overflow in subsequently generated stubs.
156   address   start_pc;
157   int       slop_bytes = 0;
158   int       slop_delta = 0;
159   const int index_dependent_slop = (itable_index == 0) ? 4 :     // code size change with transition from 8-bit to 32-bit constant (@index == 16).
160                                    (itable_index < 16) ? 3 : 0;  // index == 0 generates even shorter code.
161 
162   ResourceMark    rm;
163   CodeBuffer      cb(s->entry_point(), stub_code_length);
164   MacroAssembler *masm = new MacroAssembler(&cb);
165 
166 #if (!defined(PRODUCT) && defined(COMPILER2))
167   if (CountCompiledCalls) {
168     __ incrementq(ExternalAddress(SharedRuntime::nof_megamorphic_calls_addr()), rscratch1);
169   }
170 #endif // PRODUCT
171 
172   // Entry arguments:
173   //  rax: CompiledICData
174   //  j_rarg0: Receiver
175 
176   // Most registers are in use; we'll use rax, rbx, r10, r11
177   // (various calling sequences use r[cd]x, r[sd]i, r[89]; stay away from them)
178   const Register recv_klass_reg     = r10;
179   const Register holder_klass_reg   = rax; // declaring interface klass (DEFC)
180   const Register resolved_klass_reg = r14; // resolved interface klass (REFC)
181   const Register temp_reg           = r11;
182   const Register temp_reg2          = r13;
183   const Register method             = rbx;
184   const Register icdata_reg         = rax;
185 
186   __ movptr(resolved_klass_reg, Address(icdata_reg, CompiledICData::itable_refc_klass_offset()));
187   __ movptr(holder_klass_reg,   Address(icdata_reg, CompiledICData::itable_defc_klass_offset()));
188 
189   Label L_no_such_interface;
190 
191   // get receiver klass (also an implicit null-check)
192   assert(VtableStub::receiver_location() == j_rarg0->as_VMReg(), "receiver expected in j_rarg0");
193   address npe_addr = __ pc();
194   __ load_klass(recv_klass_reg, j_rarg0, temp_reg);
195 
196   start_pc = __ pc();
197 
198   // Receiver subtype check against REFC.
199   // Get selected method from declaring class and itable index
200   __ lookup_interface_method_stub(recv_klass_reg, // input
201                                   holder_klass_reg, // input
202                                   resolved_klass_reg, // input
203                                   method, // output
204                                   temp_reg,
205                                   temp_reg2,
206                                   noreg,
207                                   itable_index,
208                                   L_no_such_interface);
209 
210   const ptrdiff_t  lookupSize = __ pc() - start_pc;
211 
212   // We expect we need index_dependent_slop extra bytes. Reason:
213   // The emitted code in lookup_interface_method changes when itable_index exceeds 15.
214   // For linux, a very narrow estimate would be 112, but Solaris requires some more space (130).
215   const ptrdiff_t estimate = 144;
216   const ptrdiff_t codesize = lookupSize + index_dependent_slop;
217   slop_delta  = (int)(estimate - codesize);
218   slop_bytes += slop_delta;
219   assert(slop_delta >= 0, "itable #%d: Code size estimate (%d) for lookup_interface_method too small, required: %d", itable_index, (int)estimate, (int)codesize);
220 
221   // If we take a trap while this arg is on the stack we will not
222   // be able to walk the stack properly. This is not an issue except
223   // when there are mistakes in this assembly code that could generate
224   // a spurious fault. Ask me how I know...
225 
226   // method (rbx): Method*
227   // j_rarg0: receiver
228 
229 #ifdef ASSERT
230   if (DebugVtables) {
231     Label L2;
232     __ cmpptr(method, NULL_WORD);
233     __ jcc(Assembler::equal, L2);
234     __ cmpptr(Address(method, entry_offset), NULL_WORD);
235     __ jcc(Assembler::notZero, L2);
236     __ stop("compiler entrypoint is null");
237     __ bind(L2);
238   }
239 #endif // ASSERT
240 
241   address ame_addr = __ pc();
242   __ jmp(Address(method, entry_offset));
243 
244   __ bind(L_no_such_interface);
245   // Handle IncompatibleClassChangeError in itable stubs.
246   // More detailed error message.
247   // We force resolving of the call site by jumping to the "handle
248   // wrong method" stub, and so let the interpreter runtime do all the
249   // dirty work.
250   __ jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
251 
252   masm->flush();
253   slop_bytes += index_dependent_slop; // add'l slop for size variance due to large itable offsets
254   bookkeeping(masm, tty, s, npe_addr, ame_addr, false, itable_index, slop_bytes, index_dependent_slop);
255 
256   return s;
257 }
258 
259 int VtableStub::pd_code_alignment() {
260   // x86 cache line size is 64 bytes, but we want to limit alignment loss.
261   const unsigned int icache_line_size = wordSize;
262   return icache_line_size;
263 }