1 /* 2 * Copyright (c) 1999, 2024, 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 "c1/c1_MacroAssembler.hpp" 27 #include "c1/c1_Runtime1.hpp" 28 #include "code/compiledIC.hpp" 29 #include "compiler/compilerDefinitions.inline.hpp" 30 #include "gc/shared/barrierSet.hpp" 31 #include "gc/shared/barrierSetAssembler.hpp" 32 #include "gc/shared/collectedHeap.hpp" 33 #include "gc/shared/tlab_globals.hpp" 34 #include "interpreter/interpreter.hpp" 35 #include "oops/arrayOop.hpp" 36 #include "oops/markWord.hpp" 37 #include "runtime/basicLock.hpp" 38 #include "runtime/globals.hpp" 39 #include "runtime/os.hpp" 40 #include "runtime/sharedRuntime.hpp" 41 #include "runtime/stubRoutines.hpp" 42 #include "utilities/checkedCast.hpp" 43 #include "utilities/globalDefinitions.hpp" 44 45 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) { 46 const int aligned_mask = BytesPerWord -1; 47 const int hdr_offset = oopDesc::mark_offset_in_bytes(); 48 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction"); 49 assert_different_registers(hdr, obj, disp_hdr, tmp); 50 int null_check_offset = -1; 51 52 verify_oop(obj); 53 54 // save object being locked into the BasicObjectLock 55 movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj); 56 57 null_check_offset = offset(); 58 59 if (DiagnoseSyncOnValueBasedClasses != 0) { 60 load_klass(hdr, obj, rscratch1); 61 testb(Address(hdr, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class); 62 jcc(Assembler::notZero, slow_case); 63 } 64 65 if (LockingMode == LM_LIGHTWEIGHT) { 66 #ifdef _LP64 67 const Register thread = r15_thread; 68 lightweight_lock(disp_hdr, obj, hdr, thread, tmp, slow_case); 69 #else 70 // Implicit null check. 71 movptr(hdr, Address(obj, oopDesc::mark_offset_in_bytes())); 72 // Lacking registers and thread on x86_32. Always take slow path. 73 jmp(slow_case); 74 #endif 75 } else if (LockingMode == LM_LEGACY) { 76 Label done; 77 // Load object header 78 movptr(hdr, Address(obj, hdr_offset)); 79 // and mark it as unlocked 80 orptr(hdr, markWord::unlocked_value); 81 // save unlocked object header into the displaced header location on the stack 82 movptr(Address(disp_hdr, 0), hdr); 83 // test if object header is still the same (i.e. unlocked), and if so, store the 84 // displaced header address in the object header - if it is not the same, get the 85 // object header instead 86 MacroAssembler::lock(); // must be immediately before cmpxchg! 87 cmpxchgptr(disp_hdr, Address(obj, hdr_offset)); 88 // if the object header was the same, we're done 89 jcc(Assembler::equal, done); 90 // if the object header was not the same, it is now in the hdr register 91 // => test if it is a stack pointer into the same stack (recursive locking), i.e.: 92 // 93 // 1) (hdr & aligned_mask) == 0 94 // 2) rsp <= hdr 95 // 3) hdr <= rsp + page_size 96 // 97 // these 3 tests can be done by evaluating the following expression: 98 // 99 // (hdr - rsp) & (aligned_mask - page_size) 100 // 101 // assuming both the stack pointer and page_size have their least 102 // significant 2 bits cleared and page_size is a power of 2 103 subptr(hdr, rsp); 104 andptr(hdr, aligned_mask - (int)os::vm_page_size()); 105 // for recursive locking, the result is zero => save it in the displaced header 106 // location (null in the displaced hdr location indicates recursive locking) 107 movptr(Address(disp_hdr, 0), hdr); 108 // otherwise we don't care about the result and handle locking via runtime call 109 jcc(Assembler::notZero, slow_case); 110 // done 111 bind(done); 112 inc_held_monitor_count(); 113 } 114 115 return null_check_offset; 116 } 117 118 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) { 119 const int aligned_mask = BytesPerWord -1; 120 const int hdr_offset = oopDesc::mark_offset_in_bytes(); 121 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction"); 122 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different"); 123 Label done; 124 125 if (LockingMode != LM_LIGHTWEIGHT) { 126 // load displaced header 127 movptr(hdr, Address(disp_hdr, 0)); 128 // if the loaded hdr is null we had recursive locking 129 testptr(hdr, hdr); 130 // if we had recursive locking, we are done 131 jcc(Assembler::zero, done); 132 } 133 134 // load object 135 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset())); 136 verify_oop(obj); 137 138 if (LockingMode == LM_LIGHTWEIGHT) { 139 #ifdef _LP64 140 lightweight_unlock(obj, disp_hdr, r15_thread, hdr, slow_case); 141 #else 142 // Lacking registers and thread on x86_32. Always take slow path. 143 jmp(slow_case); 144 #endif 145 } else if (LockingMode == LM_LEGACY) { 146 // test if object header is pointing to the displaced header, and if so, restore 147 // the displaced header in the object - if the object header is not pointing to 148 // the displaced header, get the object header instead 149 MacroAssembler::lock(); // must be immediately before cmpxchg! 150 cmpxchgptr(hdr, Address(obj, hdr_offset)); 151 // if the object header was not pointing to the displaced header, 152 // we do unlocking via runtime call 153 jcc(Assembler::notEqual, slow_case); 154 // done 155 bind(done); 156 dec_held_monitor_count(); 157 } 158 } 159 160 161 // Defines obj, preserves var_size_in_bytes 162 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) { 163 if (UseTLAB) { 164 tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case); 165 } else { 166 jmp(slow_case); 167 } 168 } 169 170 171 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) { 172 assert_different_registers(obj, klass, len); 173 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value())); 174 #ifdef _LP64 175 if (UseCompressedClassPointers) { // Take care not to kill klass 176 movptr(t1, klass); 177 encode_klass_not_null(t1, rscratch1); 178 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1); 179 } else 180 #endif 181 { 182 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass); 183 } 184 185 if (len->is_valid()) { 186 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len); 187 #ifdef _LP64 188 int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt; 189 if (!is_aligned(base_offset, BytesPerWord)) { 190 assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned"); 191 // Clear gap/first 4 bytes following the length field. 192 xorl(t1, t1); 193 movl(Address(obj, base_offset), t1); 194 } 195 #endif 196 } 197 #ifdef _LP64 198 else if (UseCompressedClassPointers) { 199 xorptr(t1, t1); 200 store_klass_gap(obj, t1); 201 } 202 #endif 203 } 204 205 206 // preserves obj, destroys len_in_bytes 207 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) { 208 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0"); 209 Label done; 210 211 // len_in_bytes is positive and ptr sized 212 subptr(len_in_bytes, hdr_size_in_bytes); 213 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1); 214 bind(done); 215 } 216 217 218 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) { 219 assert(obj == rax, "obj must be in rax, for cmpxchg"); 220 assert_different_registers(obj, t1, t2); // XXX really? 221 assert(header_size >= 0 && object_size >= header_size, "illegal sizes"); 222 223 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case); 224 225 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB); 226 } 227 228 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, bool is_tlab_allocated) { 229 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, 230 "con_size_in_bytes is not multiple of alignment"); 231 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize; 232 233 initialize_header(obj, klass, noreg, t1, t2); 234 235 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) { 236 // clear rest of allocated space 237 const Register t1_zero = t1; 238 const Register index = t2; 239 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below) 240 if (var_size_in_bytes != noreg) { 241 mov(index, var_size_in_bytes); 242 initialize_body(obj, index, hdr_size_in_bytes, t1_zero); 243 } else if (con_size_in_bytes <= threshold) { 244 // use explicit null stores 245 // code size = 2 + 3*n bytes (n = number of fields to clear) 246 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code) 247 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord) 248 movptr(Address(obj, i), t1_zero); 249 } else if (con_size_in_bytes > hdr_size_in_bytes) { 250 // use loop to null out the fields 251 // code size = 16 bytes for even n (n = number of fields to clear) 252 // initialize last object field first if odd number of fields 253 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code) 254 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3); 255 // initialize last object field if constant size is odd 256 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0) 257 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero); 258 // initialize remaining object fields: rdx is a multiple of 2 259 { Label loop; 260 bind(loop); 261 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)), 262 t1_zero); 263 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)), 264 t1_zero);) 265 decrement(index); 266 jcc(Assembler::notZero, loop); 267 } 268 } 269 } 270 271 if (CURRENT_ENV->dtrace_alloc_probes()) { 272 assert(obj == rax, "must be"); 273 call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id))); 274 } 275 276 verify_oop(obj); 277 } 278 279 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int base_offset_in_bytes, Address::ScaleFactor f, Register klass, Label& slow_case, bool zero_array) { 280 assert(obj == rax, "obj must be in rax, for cmpxchg"); 281 assert_different_registers(obj, len, t1, t2, klass); 282 283 // determine alignment mask 284 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work"); 285 286 // check for negative or excessive length 287 cmpptr(len, checked_cast<int32_t>(max_array_allocation_length)); 288 jcc(Assembler::above, slow_case); 289 290 const Register arr_size = t2; // okay to be the same 291 // align object end 292 movptr(arr_size, base_offset_in_bytes + MinObjAlignmentInBytesMask); 293 lea(arr_size, Address(arr_size, len, f)); 294 andptr(arr_size, ~MinObjAlignmentInBytesMask); 295 296 try_allocate(obj, arr_size, 0, t1, t2, slow_case); 297 298 initialize_header(obj, klass, len, t1, t2); 299 300 // clear rest of allocated space 301 if (zero_array) { 302 const Register len_zero = len; 303 // Align-up to word boundary, because we clear the 4 bytes potentially 304 // following the length field in initialize_header(). 305 int base_offset = align_up(base_offset_in_bytes, BytesPerWord); 306 initialize_body(obj, arr_size, base_offset, len_zero); 307 } 308 309 if (CURRENT_ENV->dtrace_alloc_probes()) { 310 assert(obj == rax, "must be"); 311 call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id))); 312 } 313 314 verify_oop(obj); 315 } 316 317 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) { 318 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect"); 319 // Make sure there is enough stack space for this method's activation. 320 // Note that we do this before doing an enter(). This matches the 321 // ordering of C2's stack overflow check / rsp decrement and allows 322 // the SharedRuntime stack overflow handling to be consistent 323 // between the two compilers. 324 generate_stack_overflow_check(bang_size_in_bytes); 325 326 push(rbp); 327 if (PreserveFramePointer) { 328 mov(rbp, rsp); 329 } 330 #if !defined(_LP64) && defined(COMPILER2) 331 if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) { 332 // c2 leaves fpu stack dirty. Clean it on entry 333 empty_FPU_stack(); 334 } 335 #endif // !_LP64 && COMPILER2 336 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0 337 338 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler(); 339 // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub 340 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */); 341 } 342 343 344 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) { 345 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0 346 pop(rbp); 347 } 348 349 350 void C1_MacroAssembler::verified_entry(bool breakAtEntry) { 351 if (breakAtEntry || VerifyFPU) { 352 // Verified Entry first instruction should be 5 bytes long for correct 353 // patching by patch_verified_entry(). 354 // 355 // Breakpoint and VerifyFPU have one byte first instruction. 356 // Also first instruction will be one byte "push(rbp)" if stack banging 357 // code is not generated (see build_frame() above). 358 // For all these cases generate long instruction first. 359 fat_nop(); 360 } 361 if (breakAtEntry) int3(); 362 // build frame 363 IA32_ONLY( verify_FPU(0, "method_entry"); ) 364 } 365 366 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) { 367 // rbp, + 0: link 368 // + 1: return address 369 // + 2: argument with offset 0 370 // + 3: argument with offset 1 371 // + 4: ... 372 373 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord)); 374 } 375 376 #ifndef PRODUCT 377 378 void C1_MacroAssembler::verify_stack_oop(int stack_offset) { 379 if (!VerifyOops) return; 380 verify_oop_addr(Address(rsp, stack_offset)); 381 } 382 383 void C1_MacroAssembler::verify_not_null_oop(Register r) { 384 if (!VerifyOops) return; 385 Label not_null; 386 testptr(r, r); 387 jcc(Assembler::notZero, not_null); 388 stop("non-null oop required"); 389 bind(not_null); 390 verify_oop(r); 391 } 392 393 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) { 394 #ifdef ASSERT 395 if (inv_rax) movptr(rax, 0xDEAD); 396 if (inv_rbx) movptr(rbx, 0xDEAD); 397 if (inv_rcx) movptr(rcx, 0xDEAD); 398 if (inv_rdx) movptr(rdx, 0xDEAD); 399 if (inv_rsi) movptr(rsi, 0xDEAD); 400 if (inv_rdi) movptr(rdi, 0xDEAD); 401 #endif 402 } 403 404 #endif // ifndef PRODUCT