c1_MacroAssembler_x86.cpp revision 1472:c18cbe5936b8
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 "incls/_precompiled.incl" 26#include "incls/_c1_MacroAssembler_x86.cpp.incl" 27 28int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) { 29 const int aligned_mask = BytesPerWord -1; 30 const int hdr_offset = oopDesc::mark_offset_in_bytes(); 31 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction"); 32 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different"); 33 Label done; 34 int null_check_offset = -1; 35 36 verify_oop(obj); 37 38 // save object being locked into the BasicObjectLock 39 movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj); 40 41 if (UseBiasedLocking) { 42 assert(scratch != noreg, "should have scratch register at this point"); 43 null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case); 44 } else { 45 null_check_offset = offset(); 46 } 47 48 // Load object header 49 movptr(hdr, Address(obj, hdr_offset)); 50 // and mark it as unlocked 51 orptr(hdr, markOopDesc::unlocked_value); 52 // save unlocked object header into the displaced header location on the stack 53 movptr(Address(disp_hdr, 0), hdr); 54 // test if object header is still the same (i.e. unlocked), and if so, store the 55 // displaced header address in the object header - if it is not the same, get the 56 // object header instead 57 if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg! 58 cmpxchgptr(disp_hdr, Address(obj, hdr_offset)); 59 // if the object header was the same, we're done 60 if (PrintBiasedLockingStatistics) { 61 cond_inc32(Assembler::equal, 62 ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr())); 63 } 64 jcc(Assembler::equal, done); 65 // if the object header was not the same, it is now in the hdr register 66 // => test if it is a stack pointer into the same stack (recursive locking), i.e.: 67 // 68 // 1) (hdr & aligned_mask) == 0 69 // 2) rsp <= hdr 70 // 3) hdr <= rsp + page_size 71 // 72 // these 3 tests can be done by evaluating the following expression: 73 // 74 // (hdr - rsp) & (aligned_mask - page_size) 75 // 76 // assuming both the stack pointer and page_size have their least 77 // significant 2 bits cleared and page_size is a power of 2 78 subptr(hdr, rsp); 79 andptr(hdr, aligned_mask - os::vm_page_size()); 80 // for recursive locking, the result is zero => save it in the displaced header 81 // location (NULL in the displaced hdr location indicates recursive locking) 82 movptr(Address(disp_hdr, 0), hdr); 83 // otherwise we don't care about the result and handle locking via runtime call 84 jcc(Assembler::notZero, slow_case); 85 // done 86 bind(done); 87 return null_check_offset; 88} 89 90 91void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) { 92 const int aligned_mask = BytesPerWord -1; 93 const int hdr_offset = oopDesc::mark_offset_in_bytes(); 94 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction"); 95 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different"); 96 Label done; 97 98 if (UseBiasedLocking) { 99 // load object 100 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); 101 biased_locking_exit(obj, hdr, done); 102 } 103 104 // load displaced header 105 movptr(hdr, Address(disp_hdr, 0)); 106 // if the loaded hdr is NULL we had recursive locking 107 testptr(hdr, hdr); 108 // if we had recursive locking, we are done 109 jcc(Assembler::zero, done); 110 if (!UseBiasedLocking) { 111 // load object 112 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); 113 } 114 verify_oop(obj); 115 // test if object header is pointing to the displaced header, and if so, restore 116 // the displaced header in the object - if the object header is not pointing to 117 // the displaced header, get the object header instead 118 if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg! 119 cmpxchgptr(hdr, Address(obj, hdr_offset)); 120 // if the object header was not pointing to the displaced header, 121 // we do unlocking via runtime call 122 jcc(Assembler::notEqual, slow_case); 123 // done 124 bind(done); 125} 126 127 128// Defines obj, preserves var_size_in_bytes 129void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) { 130 if (UseTLAB) { 131 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case); 132 } else { 133 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case); 134 } 135} 136 137 138void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) { 139 assert_different_registers(obj, klass, len); 140 if (UseBiasedLocking && !len->is_valid()) { 141 assert_different_registers(obj, klass, len, t1, t2); 142 movptr(t1, Address(klass, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 143 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1); 144 } else { 145 // This assumes that all prototype bits fit in an int32_t 146 movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markOopDesc::prototype()); 147 } 148 149 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass); 150 if (len->is_valid()) { 151 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len); 152 } 153} 154 155 156// preserves obj, destroys len_in_bytes 157void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) { 158 Label done; 159 assert(obj != len_in_bytes && obj != t1 && t1 != len_in_bytes, "registers must be different"); 160 assert((hdr_size_in_bytes & (BytesPerWord - 1)) == 0, "header size is not a multiple of BytesPerWord"); 161 Register index = len_in_bytes; 162 // index is positive and ptr sized 163 subptr(index, hdr_size_in_bytes); 164 jcc(Assembler::zero, done); 165 // initialize topmost word, divide index by 2, check if odd and test if zero 166 // note: for the remaining code to work, index must be a multiple of BytesPerWord 167#ifdef ASSERT 168 { Label L; 169 testptr(index, BytesPerWord - 1); 170 jcc(Assembler::zero, L); 171 stop("index is not a multiple of BytesPerWord"); 172 bind(L); 173 } 174#endif 175 xorptr(t1, t1); // use _zero reg to clear memory (shorter code) 176 if (UseIncDec) { 177 shrptr(index, 3); // divide by 8/16 and set carry flag if bit 2 was set 178 } else { 179 shrptr(index, 2); // use 2 instructions to avoid partial flag stall 180 shrptr(index, 1); 181 } 182#ifndef _LP64 183 // index could have been not a multiple of 8 (i.e., bit 2 was set) 184 { Label even; 185 // note: if index was a multiple of 8, than it cannot 186 // be 0 now otherwise it must have been 0 before 187 // => if it is even, we don't need to check for 0 again 188 jcc(Assembler::carryClear, even); 189 // clear topmost word (no jump needed if conditional assignment would work here) 190 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 0*BytesPerWord), t1); 191 // index could be 0 now, need to check again 192 jcc(Assembler::zero, done); 193 bind(even); 194 } 195#endif // !_LP64 196 // initialize remaining object fields: rdx is a multiple of 2 now 197 { Label loop; 198 bind(loop); 199 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 1*BytesPerWord), t1); 200 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 2*BytesPerWord), t1);) 201 decrement(index); 202 jcc(Assembler::notZero, loop); 203 } 204 205 // done 206 bind(done); 207} 208 209 210void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) { 211 assert(obj == rax, "obj must be in rax, for cmpxchg"); 212 assert(obj != t1 && obj != t2 && t1 != t2, "registers must be different"); // XXX really? 213 assert(header_size >= 0 && object_size >= header_size, "illegal sizes"); 214 215 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case); 216 217 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2); 218} 219 220void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2) { 221 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, 222 "con_size_in_bytes is not multiple of alignment"); 223 const int hdr_size_in_bytes = instanceOopDesc::base_offset_in_bytes(); 224 225 initialize_header(obj, klass, noreg, t1, t2); 226 227 // clear rest of allocated space 228 const Register t1_zero = t1; 229 const Register index = t2; 230 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below) 231 if (var_size_in_bytes != noreg) { 232 mov(index, var_size_in_bytes); 233 initialize_body(obj, index, hdr_size_in_bytes, t1_zero); 234 } else if (con_size_in_bytes <= threshold) { 235 // use explicit null stores 236 // code size = 2 + 3*n bytes (n = number of fields to clear) 237 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code) 238 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord) 239 movptr(Address(obj, i), t1_zero); 240 } else if (con_size_in_bytes > hdr_size_in_bytes) { 241 // use loop to null out the fields 242 // code size = 16 bytes for even n (n = number of fields to clear) 243 // initialize last object field first if odd number of fields 244 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code) 245 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3); 246 // initialize last object field if constant size is odd 247 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0) 248 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero); 249 // initialize remaining object fields: rdx is a multiple of 2 250 { Label loop; 251 bind(loop); 252 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)), 253 t1_zero); 254 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)), 255 t1_zero);) 256 decrement(index); 257 jcc(Assembler::notZero, loop); 258 } 259 } 260 261 if (CURRENT_ENV->dtrace_alloc_probes()) { 262 assert(obj == rax, "must be"); 263 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id))); 264 } 265 266 verify_oop(obj); 267} 268 269void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) { 270 assert(obj == rax, "obj must be in rax, for cmpxchg"); 271 assert_different_registers(obj, len, t1, t2, klass); 272 273 // determine alignment mask 274 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work"); 275 276 // check for negative or excessive length 277 cmpptr(len, (int32_t)max_array_allocation_length); 278 jcc(Assembler::above, slow_case); 279 280 const Register arr_size = t2; // okay to be the same 281 // align object end 282 movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask); 283 lea(arr_size, Address(arr_size, len, f)); 284 andptr(arr_size, ~MinObjAlignmentInBytesMask); 285 286 try_allocate(obj, arr_size, 0, t1, t2, slow_case); 287 288 initialize_header(obj, klass, len, t1, t2); 289 290 // clear rest of allocated space 291 const Register len_zero = len; 292 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero); 293 294 if (CURRENT_ENV->dtrace_alloc_probes()) { 295 assert(obj == rax, "must be"); 296 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id))); 297 } 298 299 verify_oop(obj); 300} 301 302 303 304void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) { 305 verify_oop(receiver); 306 // explicit NULL check not needed since load from [klass_offset] causes a trap 307 // check against inline cache 308 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check"); 309 int start_offset = offset(); 310 cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes())); 311 // if icache check fails, then jump to runtime routine 312 // Note: RECEIVER must still contain the receiver! 313 jump_cc(Assembler::notEqual, 314 RuntimeAddress(SharedRuntime::get_ic_miss_stub())); 315 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9); 316 assert(offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry"); 317} 318 319 320void C1_MacroAssembler::build_frame(int frame_size_in_bytes) { 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(frame_size_in_bytes); 327 328 push(rbp); 329#ifdef TIERED 330 // c2 leaves fpu stack dirty. Clean it on entry 331 if (UseSSE < 2 ) { 332 empty_FPU_stack(); 333 } 334#endif // TIERED 335 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0 336} 337 338 339void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) { 340 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0 341 pop(rbp); 342} 343 344 345void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) { 346 if (C1Breakpoint) int3(); 347 inline_cache_check(receiver, ic_klass); 348} 349 350 351void C1_MacroAssembler::verified_entry() { 352 if (C1Breakpoint)int3(); 353 // build frame 354 verify_FPU(0, "method_entry"); 355} 356 357 358#ifndef PRODUCT 359 360void C1_MacroAssembler::verify_stack_oop(int stack_offset) { 361 if (!VerifyOops) return; 362 verify_oop_addr(Address(rsp, stack_offset)); 363} 364 365void C1_MacroAssembler::verify_not_null_oop(Register r) { 366 if (!VerifyOops) return; 367 Label not_null; 368 testptr(r, r); 369 jcc(Assembler::notZero, not_null); 370 stop("non-null oop required"); 371 bind(not_null); 372 verify_oop(r); 373} 374 375void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) { 376#ifdef ASSERT 377 if (inv_rax) movptr(rax, 0xDEAD); 378 if (inv_rbx) movptr(rbx, 0xDEAD); 379 if (inv_rcx) movptr(rcx, 0xDEAD); 380 if (inv_rdx) movptr(rdx, 0xDEAD); 381 if (inv_rsi) movptr(rsi, 0xDEAD); 382 if (inv_rdi) movptr(rdi, 0xDEAD); 383#endif 384} 385 386#endif // ifndef PRODUCT 387