1 /* 2 * Copyright (c) 1999, 2010, 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_interface/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/stubRoutines.hpp" 37 38 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) { 39 const int aligned_mask = BytesPerWord -1; 40 const int hdr_offset = oopDesc::mark_offset_in_bytes(); 41 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction"); 42 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different"); 43 Label done; 44 int null_check_offset = -1; 45 46 verify_oop(obj); 47 48 // save object being locked into the BasicObjectLock 49 movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj); 50 51 if (UseBiasedLocking) { 52 assert(scratch != noreg, "should have scratch register at this point"); 53 null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case); 54 } else { 55 null_check_offset = offset(); 56 } 57 58 // Load object header 59 movptr(hdr, Address(obj, hdr_offset)); 60 // and mark it as unlocked 61 orptr(hdr, markOopDesc::unlocked_value); 62 // save unlocked object header into the displaced header location on the stack 63 movptr(Address(disp_hdr, 0), hdr); 64 // test if object header is still the same (i.e. unlocked), and if so, store the 65 // displaced header address in the object header - if it is not the same, get the 66 // object header instead 67 if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg! 68 cmpxchgptr(disp_hdr, Address(obj, hdr_offset)); 69 // if the object header was the same, we're done 70 if (PrintBiasedLockingStatistics) { 71 cond_inc32(Assembler::equal, 72 ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr())); 73 } 74 jcc(Assembler::equal, done); 75 // if the object header was not the same, it is now in the hdr register 76 // => test if it is a stack pointer into the same stack (recursive locking), i.e.: 77 // 78 // 1) (hdr & aligned_mask) == 0 79 // 2) rsp <= hdr 80 // 3) hdr <= rsp + page_size 81 // 82 // these 3 tests can be done by evaluating the following expression: 83 // 84 // (hdr - rsp) & (aligned_mask - page_size) 85 // 86 // assuming both the stack pointer and page_size have their least 87 // significant 2 bits cleared and page_size is a power of 2 88 subptr(hdr, rsp); 89 andptr(hdr, aligned_mask - os::vm_page_size()); 90 // for recursive locking, the result is zero => save it in the displaced header 91 // location (NULL in the displaced hdr location indicates recursive locking) 92 movptr(Address(disp_hdr, 0), hdr); 93 // otherwise we don't care about the result and handle locking via runtime call 94 jcc(Assembler::notZero, slow_case); 95 // done 96 bind(done); 97 return null_check_offset; 98 } 99 100 101 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) { 102 const int aligned_mask = BytesPerWord -1; 103 const int hdr_offset = oopDesc::mark_offset_in_bytes(); 104 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction"); 105 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different"); 106 Label done; 107 108 if (UseBiasedLocking) { 109 // load object 110 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); 111 biased_locking_exit(obj, hdr, done); 112 } 113 114 // load displaced header 115 movptr(hdr, Address(disp_hdr, 0)); 116 // if the loaded hdr is NULL we had recursive locking 117 testptr(hdr, hdr); 118 // if we had recursive locking, we are done 119 jcc(Assembler::zero, done); 120 if (!UseBiasedLocking) { 121 // load object 122 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); 123 } 124 verify_oop(obj); 125 // test if object header is pointing to the displaced header, and if so, restore 126 // the displaced header in the object - if the object header is not pointing to 127 // the displaced header, get the object header instead 128 if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg! 129 cmpxchgptr(hdr, Address(obj, hdr_offset)); 130 // if the object header was not pointing to the displaced header, 131 // we do unlocking via runtime call 132 jcc(Assembler::notEqual, slow_case); 133 // done 134 bind(done); 135 } 136 137 138 // Defines obj, preserves var_size_in_bytes 139 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) { 140 if (UseTLAB) { 141 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case); 142 } else { 143 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case); 144 } 145 } 146 147 148 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) { 149 assert_different_registers(obj, klass, len); 150 if (UseBiasedLocking && !len->is_valid()) { 151 assert_different_registers(obj, klass, len, t1, t2); 152 movptr(t1, Address(klass, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 153 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1); 154 } else { 155 // This assumes that all prototype bits fit in an int32_t 156 movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markOopDesc::prototype()); 157 } 158 159 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass); 160 if (len->is_valid()) { 161 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len); 162 } 163 } 164 165 166 // preserves obj, destroys len_in_bytes 167 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) { 168 Label done; 169 assert(obj != len_in_bytes && obj != t1 && t1 != len_in_bytes, "registers must be different"); 170 assert((hdr_size_in_bytes & (BytesPerWord - 1)) == 0, "header size is not a multiple of BytesPerWord"); 171 Register index = len_in_bytes; 172 // index is positive and ptr sized 173 subptr(index, hdr_size_in_bytes); 174 jcc(Assembler::zero, done); 175 // initialize topmost word, divide index by 2, check if odd and test if zero 176 // note: for the remaining code to work, index must be a multiple of BytesPerWord 177 #ifdef ASSERT 178 { Label L; 179 testptr(index, BytesPerWord - 1); 180 jcc(Assembler::zero, L); 181 stop("index is not a multiple of BytesPerWord"); 182 bind(L); 183 } 184 #endif 185 xorptr(t1, t1); // use _zero reg to clear memory (shorter code) 186 if (UseIncDec) { 187 shrptr(index, 3); // divide by 8/16 and set carry flag if bit 2 was set 188 } else { 189 shrptr(index, 2); // use 2 instructions to avoid partial flag stall 190 shrptr(index, 1); 191 } 192 #ifndef _LP64 193 // index could have been not a multiple of 8 (i.e., bit 2 was set) 194 { Label even; 195 // note: if index was a multiple of 8, than it cannot 196 // be 0 now otherwise it must have been 0 before 197 // => if it is even, we don't need to check for 0 again 198 jcc(Assembler::carryClear, even); 199 // clear topmost word (no jump needed if conditional assignment would work here) 200 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 0*BytesPerWord), t1); 201 // index could be 0 now, need to check again 202 jcc(Assembler::zero, done); 203 bind(even); 204 } 205 #endif // !_LP64 206 // initialize remaining object fields: rdx is a multiple of 2 now 207 { Label loop; 208 bind(loop); 209 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 1*BytesPerWord), t1); 210 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 2*BytesPerWord), t1);) 211 decrement(index); 212 jcc(Assembler::notZero, loop); 213 } 214 215 // done 216 bind(done); 217 } 218 219 220 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) { 221 assert(obj == rax, "obj must be in rax, for cmpxchg"); 222 assert(obj != t1 && obj != t2 && t1 != t2, "registers must be different"); // XXX really? 223 assert(header_size >= 0 && object_size >= header_size, "illegal sizes"); 224 225 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case); 226 227 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2); 228 } 229 230 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2) { 231 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, 232 "con_size_in_bytes is not multiple of alignment"); 233 const int hdr_size_in_bytes = instanceOopDesc::base_offset_in_bytes(); 234 235 initialize_header(obj, klass, noreg, t1, t2); 236 237 // clear rest of allocated space 238 const Register t1_zero = t1; 239 const Register index = t2; 240 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below) 241 if (var_size_in_bytes != noreg) { 242 mov(index, var_size_in_bytes); 243 initialize_body(obj, index, hdr_size_in_bytes, t1_zero); 244 } else if (con_size_in_bytes <= threshold) { 245 // use explicit null stores 246 // code size = 2 + 3*n bytes (n = number of fields to clear) 247 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code) 248 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord) 249 movptr(Address(obj, i), t1_zero); 250 } else if (con_size_in_bytes > hdr_size_in_bytes) { 251 // use loop to null out the fields 252 // code size = 16 bytes for even n (n = number of fields to clear) 253 // initialize last object field first if odd number of fields 254 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code) 255 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3); 256 // initialize last object field if constant size is odd 257 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0) 258 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero); 259 // initialize remaining object fields: rdx is a multiple of 2 260 { Label loop; 261 bind(loop); 262 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)), 263 t1_zero); 264 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)), 265 t1_zero);) 266 decrement(index); 267 jcc(Assembler::notZero, loop); 268 } 269 } 270 271 if (CURRENT_ENV->dtrace_alloc_probes()) { 272 assert(obj == rax, "must be"); 273 call(RuntimeAddress(Runtime1::entry_for(Runtime1::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 header_size, Address::ScaleFactor f, Register klass, Label& slow_case) { 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, (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, (int32_t)header_size * BytesPerWord + 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 const Register len_zero = len; 302 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero); 303 304 if (CURRENT_ENV->dtrace_alloc_probes()) { 305 assert(obj == rax, "must be"); 306 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id))); 307 } 308 309 verify_oop(obj); 310 } 311 312 313 314 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) { 315 verify_oop(receiver); 316 // explicit NULL check not needed since load from [klass_offset] causes a trap 317 // check against inline cache 318 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check"); 319 int start_offset = offset(); 320 cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes())); 321 // if icache check fails, then jump to runtime routine 322 // Note: RECEIVER must still contain the receiver! 323 jump_cc(Assembler::notEqual, 324 RuntimeAddress(SharedRuntime::get_ic_miss_stub())); 325 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9); 326 assert(offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry"); 327 } 328 329 330 void C1_MacroAssembler::build_frame(int frame_size_in_bytes) { 331 // Make sure there is enough stack space for this method's activation. 332 // Note that we do this before doing an enter(). This matches the 333 // ordering of C2's stack overflow check / rsp decrement and allows 334 // the SharedRuntime stack overflow handling to be consistent 335 // between the two compilers. 336 generate_stack_overflow_check(frame_size_in_bytes); 337 338 push(rbp); 339 #ifdef TIERED 340 // c2 leaves fpu stack dirty. Clean it on entry 341 if (UseSSE < 2 ) { 342 empty_FPU_stack(); 343 } 344 #endif // TIERED 345 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0 346 } 347 348 349 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) { 350 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0 351 pop(rbp); 352 } 353 354 355 void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) { 356 if (C1Breakpoint) int3(); 357 inline_cache_check(receiver, ic_klass); 358 } 359 360 361 void C1_MacroAssembler::verified_entry() { 362 if (C1Breakpoint)int3(); 363 // build frame 364 verify_FPU(0, "method_entry"); 365 } 366 367 368 #ifndef PRODUCT 369 370 void C1_MacroAssembler::verify_stack_oop(int stack_offset) { 371 if (!VerifyOops) return; 372 verify_oop_addr(Address(rsp, stack_offset)); 373 } 374 375 void C1_MacroAssembler::verify_not_null_oop(Register r) { 376 if (!VerifyOops) return; 377 Label not_null; 378 testptr(r, r); 379 jcc(Assembler::notZero, not_null); 380 stop("non-null oop required"); 381 bind(not_null); 382 verify_oop(r); 383 } 384 385 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) { 386 #ifdef ASSERT 387 if (inv_rax) movptr(rax, 0xDEAD); 388 if (inv_rbx) movptr(rbx, 0xDEAD); 389 if (inv_rcx) movptr(rcx, 0xDEAD); 390 if (inv_rdx) movptr(rdx, 0xDEAD); 391 if (inv_rsi) movptr(rsi, 0xDEAD); 392 if (inv_rdi) movptr(rdi, 0xDEAD); 393 #endif 394 } 395 396 #endif // ifndef PRODUCT