1 /* 2 * Copyright (c) 1999, 2015, 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 "classfile/systemDictionary.hpp" 29 #include "gc/shared/collectedHeap.hpp" 30 #include "interpreter/interpreter.hpp" 31 #include "oops/arrayOop.hpp" 32 #include "oops/markOop.hpp" 33 #include "runtime/basicLock.hpp" 34 #include "runtime/biasedLocking.hpp" 35 #include "runtime/os.hpp" 36 #include "runtime/sharedRuntime.hpp" 37 #include "runtime/stubRoutines.hpp" 38 39 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) { 40 const int aligned_mask = BytesPerWord -1; 41 const int hdr_offset = oopDesc::mark_offset_in_bytes(); 42 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction"); 43 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different"); 44 Label done; 45 int null_check_offset = -1; 46 47 verify_oop(obj); 48 49 shenandoah_store_addr_check(obj); 50 51 // save object being locked into the BasicObjectLock 52 movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj); 53 54 if (UseBiasedLocking) { 55 assert(scratch != noreg, "should have scratch register at this point"); 56 null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case); 57 } else { 58 null_check_offset = offset(); 59 } 60 61 // Load object header 62 movptr(hdr, Address(obj, hdr_offset)); 63 // and mark it as unlocked 64 orptr(hdr, markOopDesc::unlocked_value); 65 // save unlocked object header into the displaced header location on the stack 66 movptr(Address(disp_hdr, 0), hdr); 67 // test if object header is still the same (i.e. unlocked), and if so, store the 68 // displaced header address in the object header - if it is not the same, get the 69 // object header instead 70 if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg! 71 cmpxchgptr(disp_hdr, Address(obj, hdr_offset)); 72 // if the object header was the same, we're done 73 if (PrintBiasedLockingStatistics) { 74 cond_inc32(Assembler::equal, 75 ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr())); 76 } 77 jcc(Assembler::equal, done); 78 // if the object header was not the same, it is now in the hdr register 79 // => test if it is a stack pointer into the same stack (recursive locking), i.e.: 80 // 81 // 1) (hdr & aligned_mask) == 0 82 // 2) rsp <= hdr 83 // 3) hdr <= rsp + page_size 84 // 85 // these 3 tests can be done by evaluating the following expression: 86 // 87 // (hdr - rsp) & (aligned_mask - page_size) 88 // 89 // assuming both the stack pointer and page_size have their least 90 // significant 2 bits cleared and page_size is a power of 2 91 subptr(hdr, rsp); 92 andptr(hdr, aligned_mask - os::vm_page_size()); 93 // for recursive locking, the result is zero => save it in the displaced header 94 // location (NULL in the displaced hdr location indicates recursive locking) 95 movptr(Address(disp_hdr, 0), hdr); 96 // otherwise we don't care about the result and handle locking via runtime call 97 jcc(Assembler::notZero, slow_case); 98 // done 99 bind(done); 100 return null_check_offset; 101 } 102 103 104 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) { 105 const int aligned_mask = BytesPerWord -1; 106 const int hdr_offset = oopDesc::mark_offset_in_bytes(); 107 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction"); 108 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different"); 109 Label done; 110 111 if (UseBiasedLocking) { 112 // load object 113 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); 114 biased_locking_exit(obj, hdr, done); 115 } 116 117 // load displaced header 118 movptr(hdr, Address(disp_hdr, 0)); 119 // if the loaded hdr is NULL we had recursive locking 120 testptr(hdr, hdr); 121 // if we had recursive locking, we are done 122 jcc(Assembler::zero, done); 123 if (!UseBiasedLocking) { 124 // load object 125 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); 126 } 127 verify_oop(obj); 128 129 shenandoah_store_addr_check(obj); 130 131 // test if object header is pointing to the displaced header, and if so, restore 132 // the displaced header in the object - if the object header is not pointing to 133 // the displaced header, get the object header instead 134 if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg! 135 cmpxchgptr(hdr, Address(obj, hdr_offset)); 136 // if the object header was not pointing to the displaced header, 137 // we do unlocking via runtime call 138 jcc(Assembler::notEqual, slow_case); 139 // done 140 bind(done); 141 } 142 143 144 // Defines obj, preserves var_size_in_bytes 145 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) { 146 if (UseTLAB) { 147 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case); 148 } else { 149 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case); 150 incr_allocated_bytes(noreg, var_size_in_bytes, con_size_in_bytes, t1); 151 } 152 } 153 154 155 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) { 156 assert_different_registers(obj, klass, len); 157 if (UseBiasedLocking && !len->is_valid()) { 158 assert_different_registers(obj, klass, len, t1, t2); 159 movptr(t1, Address(klass, Klass::prototype_header_offset())); 160 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1); 161 } else { 162 // This assumes that all prototype bits fit in an int32_t 163 movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markOopDesc::prototype()); 164 } 165 #ifdef _LP64 166 if (UseCompressedClassPointers) { // Take care not to kill klass 167 movptr(t1, klass); 168 encode_klass_not_null(t1); 169 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1); 170 } else 171 #endif 172 { 173 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass); 174 } 175 176 if (len->is_valid()) { 177 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len); 178 } 179 #ifdef _LP64 180 else if (UseCompressedClassPointers) { 181 xorptr(t1, t1); 182 store_klass_gap(obj, t1); 183 } 184 #endif 185 } 186 187 188 // preserves obj, destroys len_in_bytes 189 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) { 190 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0"); 191 Label done; 192 193 // len_in_bytes is positive and ptr sized 194 subptr(len_in_bytes, hdr_size_in_bytes); 195 jcc(Assembler::zero, done); 196 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1); 197 bind(done); 198 } 199 200 201 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) { 202 assert(obj == rax, "obj must be in rax, for cmpxchg"); 203 assert_different_registers(obj, t1, t2); // XXX really? 204 assert(header_size >= 0 && object_size >= header_size, "illegal sizes"); 205 206 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case); 207 208 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB); 209 } 210 211 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) { 212 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, 213 "con_size_in_bytes is not multiple of alignment"); 214 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize; 215 216 initialize_header(obj, klass, noreg, t1, t2); 217 218 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) { 219 // clear rest of allocated space 220 const Register t1_zero = t1; 221 const Register index = t2; 222 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below) 223 if (var_size_in_bytes != noreg) { 224 mov(index, var_size_in_bytes); 225 initialize_body(obj, index, hdr_size_in_bytes, t1_zero); 226 } else if (con_size_in_bytes <= threshold) { 227 // use explicit null stores 228 // code size = 2 + 3*n bytes (n = number of fields to clear) 229 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code) 230 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord) 231 movptr(Address(obj, i), t1_zero); 232 } else if (con_size_in_bytes > hdr_size_in_bytes) { 233 // use loop to null out the fields 234 // code size = 16 bytes for even n (n = number of fields to clear) 235 // initialize last object field first if odd number of fields 236 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code) 237 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3); 238 // initialize last object field if constant size is odd 239 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0) 240 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero); 241 // initialize remaining object fields: rdx is a multiple of 2 242 { Label loop; 243 bind(loop); 244 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)), 245 t1_zero); 246 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)), 247 t1_zero);) 248 decrement(index); 249 jcc(Assembler::notZero, loop); 250 } 251 } 252 } 253 254 if (CURRENT_ENV->dtrace_alloc_probes()) { 255 assert(obj == rax, "must be"); 256 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id))); 257 } 258 259 verify_oop(obj); 260 } 261 262 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) { 263 assert(obj == rax, "obj must be in rax, for cmpxchg"); 264 assert_different_registers(obj, len, t1, t2, klass); 265 266 // determine alignment mask 267 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work"); 268 269 // check for negative or excessive length 270 cmpptr(len, (int32_t)max_array_allocation_length); 271 jcc(Assembler::above, slow_case); 272 273 const Register arr_size = t2; // okay to be the same 274 // align object end 275 movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask); 276 lea(arr_size, Address(arr_size, len, f)); 277 andptr(arr_size, ~MinObjAlignmentInBytesMask); 278 279 try_allocate(obj, arr_size, 0, t1, t2, slow_case); 280 281 initialize_header(obj, klass, len, t1, t2); 282 283 // clear rest of allocated space 284 const Register len_zero = len; 285 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero); 286 287 if (CURRENT_ENV->dtrace_alloc_probes()) { 288 assert(obj == rax, "must be"); 289 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id))); 290 } 291 292 verify_oop(obj); 293 } 294 295 296 297 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) { 298 verify_oop(receiver); 299 // explicit NULL check not needed since load from [klass_offset] causes a trap 300 // check against inline cache 301 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check"); 302 int start_offset = offset(); 303 304 if (UseCompressedClassPointers) { 305 load_klass(rscratch1, receiver); 306 cmpptr(rscratch1, iCache); 307 } else { 308 cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes())); 309 } 310 // if icache check fails, then jump to runtime routine 311 // Note: RECEIVER must still contain the receiver! 312 jump_cc(Assembler::notEqual, 313 RuntimeAddress(SharedRuntime::get_ic_miss_stub())); 314 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9); 315 assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry"); 316 } 317 318 319 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) { 320 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect"); 321 // Make sure there is enough stack space for this method's activation. 322 // Note that we do this before doing an enter(). This matches the 323 // ordering of C2's stack overflow check / rsp decrement and allows 324 // the SharedRuntime stack overflow handling to be consistent 325 // between the two compilers. 326 generate_stack_overflow_check(bang_size_in_bytes); 327 328 push(rbp); 329 if (PreserveFramePointer) { 330 mov(rbp, rsp); 331 } 332 #ifdef TIERED 333 // c2 leaves fpu stack dirty. Clean it on entry 334 if (UseSSE < 2 ) { 335 empty_FPU_stack(); 336 } 337 #endif // TIERED 338 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0 339 } 340 341 342 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) { 343 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0 344 pop(rbp); 345 } 346 347 348 void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) { 349 if (C1Breakpoint) int3(); 350 inline_cache_check(receiver, ic_klass); 351 } 352 353 354 void C1_MacroAssembler::verified_entry() { 355 if (C1Breakpoint || VerifyFPU || !UseStackBanging) { 356 // Verified Entry first instruction should be 5 bytes long for correct 357 // patching by patch_verified_entry(). 358 // 359 // C1Breakpoint and VerifyFPU have one byte first instruction. 360 // Also first instruction will be one byte "push(rbp)" if stack banging 361 // code is not generated (see build_frame() above). 362 // For all these cases generate long instruction first. 363 fat_nop(); 364 } 365 if (C1Breakpoint)int3(); 366 // build frame 367 verify_FPU(0, "method_entry"); 368 } 369 370 371 #ifndef PRODUCT 372 373 void C1_MacroAssembler::verify_stack_oop(int stack_offset) { 374 if (!VerifyOops) return; 375 verify_oop_addr(Address(rsp, stack_offset)); 376 } 377 378 void C1_MacroAssembler::verify_not_null_oop(Register r) { 379 if (!VerifyOops) return; 380 Label not_null; 381 testptr(r, r); 382 jcc(Assembler::notZero, not_null); 383 stop("non-null oop required"); 384 bind(not_null); 385 verify_oop(r); 386 } 387 388 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) { 389 #ifdef ASSERT 390 if (inv_rax) movptr(rax, 0xDEAD); 391 if (inv_rbx) movptr(rbx, 0xDEAD); 392 if (inv_rcx) movptr(rcx, 0xDEAD); 393 if (inv_rdx) movptr(rdx, 0xDEAD); 394 if (inv_rsi) movptr(rsi, 0xDEAD); 395 if (inv_rdi) movptr(rdi, 0xDEAD); 396 #endif 397 } 398 399 #endif // ifndef PRODUCT