1 /* 2 * Copyright 2003-2010 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, 20 * CA 95054 USA or visit www.sun.com if you need additional information or 21 * have any questions. 22 * 23 */ 24 25 #include "incls/_precompiled.incl" 26 #include "incls/_sharedRuntime_x86_64.cpp.incl" 27 28 DeoptimizationBlob *SharedRuntime::_deopt_blob; 29 #ifdef COMPILER2 30 UncommonTrapBlob *SharedRuntime::_uncommon_trap_blob; 31 ExceptionBlob *OptoRuntime::_exception_blob; 32 #endif // COMPILER2 33 34 SafepointBlob *SharedRuntime::_polling_page_safepoint_handler_blob; 35 SafepointBlob *SharedRuntime::_polling_page_return_handler_blob; 36 RuntimeStub* SharedRuntime::_wrong_method_blob; 37 RuntimeStub* SharedRuntime::_ic_miss_blob; 38 RuntimeStub* SharedRuntime::_resolve_opt_virtual_call_blob; 39 RuntimeStub* SharedRuntime::_resolve_virtual_call_blob; 40 RuntimeStub* SharedRuntime::_resolve_static_call_blob; 41 42 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size; 43 44 #define __ masm-> 45 46 class SimpleRuntimeFrame { 47 48 public: 49 50 // Most of the runtime stubs have this simple frame layout. 51 // This class exists to make the layout shared in one place. 52 // Offsets are for compiler stack slots, which are jints. 53 enum layout { 54 // The frame sender code expects that rbp will be in the "natural" place and 55 // will override any oopMap setting for it. We must therefore force the layout 56 // so that it agrees with the frame sender code. 57 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt, 58 rbp_off2, 59 return_off, return_off2, 60 framesize 61 }; 62 }; 63 64 class RegisterSaver { 65 // Capture info about frame layout. Layout offsets are in jint 66 // units because compiler frame slots are jints. 67 #define DEF_XMM_OFFS(regnum) xmm ## regnum ## _off = xmm_off + (regnum)*16/BytesPerInt, xmm ## regnum ## H_off 68 enum layout { 69 fpu_state_off = frame::arg_reg_save_area_bytes/BytesPerInt, // fxsave save area 70 xmm_off = fpu_state_off + 160/BytesPerInt, // offset in fxsave save area 71 DEF_XMM_OFFS(0), 72 DEF_XMM_OFFS(1), 73 DEF_XMM_OFFS(2), 74 DEF_XMM_OFFS(3), 75 DEF_XMM_OFFS(4), 76 DEF_XMM_OFFS(5), 77 DEF_XMM_OFFS(6), 78 DEF_XMM_OFFS(7), 79 DEF_XMM_OFFS(8), 80 DEF_XMM_OFFS(9), 81 DEF_XMM_OFFS(10), 82 DEF_XMM_OFFS(11), 83 DEF_XMM_OFFS(12), 84 DEF_XMM_OFFS(13), 85 DEF_XMM_OFFS(14), 86 DEF_XMM_OFFS(15), 87 fpu_state_end = fpu_state_off + ((FPUStateSizeInWords-1)*wordSize / BytesPerInt), 88 fpu_stateH_end, 89 r15_off, r15H_off, 90 r14_off, r14H_off, 91 r13_off, r13H_off, 92 r12_off, r12H_off, 93 r11_off, r11H_off, 94 r10_off, r10H_off, 95 r9_off, r9H_off, 96 r8_off, r8H_off, 97 rdi_off, rdiH_off, 98 rsi_off, rsiH_off, 99 ignore_off, ignoreH_off, // extra copy of rbp 100 rsp_off, rspH_off, 101 rbx_off, rbxH_off, 102 rdx_off, rdxH_off, 103 rcx_off, rcxH_off, 104 rax_off, raxH_off, 105 // 16-byte stack alignment fill word: see MacroAssembler::push/pop_IU_state 106 align_off, alignH_off, 107 flags_off, flagsH_off, 108 // The frame sender code expects that rbp will be in the "natural" place and 109 // will override any oopMap setting for it. We must therefore force the layout 110 // so that it agrees with the frame sender code. 111 rbp_off, rbpH_off, // copy of rbp we will restore 112 return_off, returnH_off, // slot for return address 113 reg_save_size // size in compiler stack slots 114 }; 115 116 public: 117 static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words); 118 static void restore_live_registers(MacroAssembler* masm); 119 120 // Offsets into the register save area 121 // Used by deoptimization when it is managing result register 122 // values on its own 123 124 static int rax_offset_in_bytes(void) { return BytesPerInt * rax_off; } 125 static int rdx_offset_in_bytes(void) { return BytesPerInt * rdx_off; } 126 static int rbx_offset_in_bytes(void) { return BytesPerInt * rbx_off; } 127 static int xmm0_offset_in_bytes(void) { return BytesPerInt * xmm0_off; } 128 static int return_offset_in_bytes(void) { return BytesPerInt * return_off; } 129 130 // During deoptimization only the result registers need to be restored, 131 // all the other values have already been extracted. 132 static void restore_result_registers(MacroAssembler* masm); 133 }; 134 135 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words) { 136 137 // Always make the frame size 16-byte aligned 138 int frame_size_in_bytes = round_to(additional_frame_words*wordSize + 139 reg_save_size*BytesPerInt, 16); 140 // OopMap frame size is in compiler stack slots (jint's) not bytes or words 141 int frame_size_in_slots = frame_size_in_bytes / BytesPerInt; 142 // The caller will allocate additional_frame_words 143 int additional_frame_slots = additional_frame_words*wordSize / BytesPerInt; 144 // CodeBlob frame size is in words. 145 int frame_size_in_words = frame_size_in_bytes / wordSize; 146 *total_frame_words = frame_size_in_words; 147 148 // Save registers, fpu state, and flags. 149 // We assume caller has already pushed the return address onto the 150 // stack, so rsp is 8-byte aligned here. 151 // We push rpb twice in this sequence because we want the real rbp 152 // to be under the return like a normal enter. 153 154 __ enter(); // rsp becomes 16-byte aligned here 155 __ push_CPU_state(); // Push a multiple of 16 bytes 156 if (frame::arg_reg_save_area_bytes != 0) { 157 // Allocate argument register save area 158 __ subptr(rsp, frame::arg_reg_save_area_bytes); 159 } 160 161 // Set an oopmap for the call site. This oopmap will map all 162 // oop-registers and debug-info registers as callee-saved. This 163 // will allow deoptimization at this safepoint to find all possible 164 // debug-info recordings, as well as let GC find all oops. 165 166 OopMapSet *oop_maps = new OopMapSet(); 167 OopMap* map = new OopMap(frame_size_in_slots, 0); 168 map->set_callee_saved(VMRegImpl::stack2reg( rax_off + additional_frame_slots), rax->as_VMReg()); 169 map->set_callee_saved(VMRegImpl::stack2reg( rcx_off + additional_frame_slots), rcx->as_VMReg()); 170 map->set_callee_saved(VMRegImpl::stack2reg( rdx_off + additional_frame_slots), rdx->as_VMReg()); 171 map->set_callee_saved(VMRegImpl::stack2reg( rbx_off + additional_frame_slots), rbx->as_VMReg()); 172 // rbp location is known implicitly by the frame sender code, needs no oopmap 173 // and the location where rbp was saved by is ignored 174 map->set_callee_saved(VMRegImpl::stack2reg( rsi_off + additional_frame_slots), rsi->as_VMReg()); 175 map->set_callee_saved(VMRegImpl::stack2reg( rdi_off + additional_frame_slots), rdi->as_VMReg()); 176 map->set_callee_saved(VMRegImpl::stack2reg( r8_off + additional_frame_slots), r8->as_VMReg()); 177 map->set_callee_saved(VMRegImpl::stack2reg( r9_off + additional_frame_slots), r9->as_VMReg()); 178 map->set_callee_saved(VMRegImpl::stack2reg( r10_off + additional_frame_slots), r10->as_VMReg()); 179 map->set_callee_saved(VMRegImpl::stack2reg( r11_off + additional_frame_slots), r11->as_VMReg()); 180 map->set_callee_saved(VMRegImpl::stack2reg( r12_off + additional_frame_slots), r12->as_VMReg()); 181 map->set_callee_saved(VMRegImpl::stack2reg( r13_off + additional_frame_slots), r13->as_VMReg()); 182 map->set_callee_saved(VMRegImpl::stack2reg( r14_off + additional_frame_slots), r14->as_VMReg()); 183 map->set_callee_saved(VMRegImpl::stack2reg( r15_off + additional_frame_slots), r15->as_VMReg()); 184 map->set_callee_saved(VMRegImpl::stack2reg(xmm0_off + additional_frame_slots), xmm0->as_VMReg()); 185 map->set_callee_saved(VMRegImpl::stack2reg(xmm1_off + additional_frame_slots), xmm1->as_VMReg()); 186 map->set_callee_saved(VMRegImpl::stack2reg(xmm2_off + additional_frame_slots), xmm2->as_VMReg()); 187 map->set_callee_saved(VMRegImpl::stack2reg(xmm3_off + additional_frame_slots), xmm3->as_VMReg()); 188 map->set_callee_saved(VMRegImpl::stack2reg(xmm4_off + additional_frame_slots), xmm4->as_VMReg()); 189 map->set_callee_saved(VMRegImpl::stack2reg(xmm5_off + additional_frame_slots), xmm5->as_VMReg()); 190 map->set_callee_saved(VMRegImpl::stack2reg(xmm6_off + additional_frame_slots), xmm6->as_VMReg()); 191 map->set_callee_saved(VMRegImpl::stack2reg(xmm7_off + additional_frame_slots), xmm7->as_VMReg()); 192 map->set_callee_saved(VMRegImpl::stack2reg(xmm8_off + additional_frame_slots), xmm8->as_VMReg()); 193 map->set_callee_saved(VMRegImpl::stack2reg(xmm9_off + additional_frame_slots), xmm9->as_VMReg()); 194 map->set_callee_saved(VMRegImpl::stack2reg(xmm10_off + additional_frame_slots), xmm10->as_VMReg()); 195 map->set_callee_saved(VMRegImpl::stack2reg(xmm11_off + additional_frame_slots), xmm11->as_VMReg()); 196 map->set_callee_saved(VMRegImpl::stack2reg(xmm12_off + additional_frame_slots), xmm12->as_VMReg()); 197 map->set_callee_saved(VMRegImpl::stack2reg(xmm13_off + additional_frame_slots), xmm13->as_VMReg()); 198 map->set_callee_saved(VMRegImpl::stack2reg(xmm14_off + additional_frame_slots), xmm14->as_VMReg()); 199 map->set_callee_saved(VMRegImpl::stack2reg(xmm15_off + additional_frame_slots), xmm15->as_VMReg()); 200 201 // %%% These should all be a waste but we'll keep things as they were for now 202 if (true) { 203 map->set_callee_saved(VMRegImpl::stack2reg( raxH_off + additional_frame_slots), 204 rax->as_VMReg()->next()); 205 map->set_callee_saved(VMRegImpl::stack2reg( rcxH_off + additional_frame_slots), 206 rcx->as_VMReg()->next()); 207 map->set_callee_saved(VMRegImpl::stack2reg( rdxH_off + additional_frame_slots), 208 rdx->as_VMReg()->next()); 209 map->set_callee_saved(VMRegImpl::stack2reg( rbxH_off + additional_frame_slots), 210 rbx->as_VMReg()->next()); 211 // rbp location is known implicitly by the frame sender code, needs no oopmap 212 map->set_callee_saved(VMRegImpl::stack2reg( rsiH_off + additional_frame_slots), 213 rsi->as_VMReg()->next()); 214 map->set_callee_saved(VMRegImpl::stack2reg( rdiH_off + additional_frame_slots), 215 rdi->as_VMReg()->next()); 216 map->set_callee_saved(VMRegImpl::stack2reg( r8H_off + additional_frame_slots), 217 r8->as_VMReg()->next()); 218 map->set_callee_saved(VMRegImpl::stack2reg( r9H_off + additional_frame_slots), 219 r9->as_VMReg()->next()); 220 map->set_callee_saved(VMRegImpl::stack2reg( r10H_off + additional_frame_slots), 221 r10->as_VMReg()->next()); 222 map->set_callee_saved(VMRegImpl::stack2reg( r11H_off + additional_frame_slots), 223 r11->as_VMReg()->next()); 224 map->set_callee_saved(VMRegImpl::stack2reg( r12H_off + additional_frame_slots), 225 r12->as_VMReg()->next()); 226 map->set_callee_saved(VMRegImpl::stack2reg( r13H_off + additional_frame_slots), 227 r13->as_VMReg()->next()); 228 map->set_callee_saved(VMRegImpl::stack2reg( r14H_off + additional_frame_slots), 229 r14->as_VMReg()->next()); 230 map->set_callee_saved(VMRegImpl::stack2reg( r15H_off + additional_frame_slots), 231 r15->as_VMReg()->next()); 232 map->set_callee_saved(VMRegImpl::stack2reg(xmm0H_off + additional_frame_slots), 233 xmm0->as_VMReg()->next()); 234 map->set_callee_saved(VMRegImpl::stack2reg(xmm1H_off + additional_frame_slots), 235 xmm1->as_VMReg()->next()); 236 map->set_callee_saved(VMRegImpl::stack2reg(xmm2H_off + additional_frame_slots), 237 xmm2->as_VMReg()->next()); 238 map->set_callee_saved(VMRegImpl::stack2reg(xmm3H_off + additional_frame_slots), 239 xmm3->as_VMReg()->next()); 240 map->set_callee_saved(VMRegImpl::stack2reg(xmm4H_off + additional_frame_slots), 241 xmm4->as_VMReg()->next()); 242 map->set_callee_saved(VMRegImpl::stack2reg(xmm5H_off + additional_frame_slots), 243 xmm5->as_VMReg()->next()); 244 map->set_callee_saved(VMRegImpl::stack2reg(xmm6H_off + additional_frame_slots), 245 xmm6->as_VMReg()->next()); 246 map->set_callee_saved(VMRegImpl::stack2reg(xmm7H_off + additional_frame_slots), 247 xmm7->as_VMReg()->next()); 248 map->set_callee_saved(VMRegImpl::stack2reg(xmm8H_off + additional_frame_slots), 249 xmm8->as_VMReg()->next()); 250 map->set_callee_saved(VMRegImpl::stack2reg(xmm9H_off + additional_frame_slots), 251 xmm9->as_VMReg()->next()); 252 map->set_callee_saved(VMRegImpl::stack2reg(xmm10H_off + additional_frame_slots), 253 xmm10->as_VMReg()->next()); 254 map->set_callee_saved(VMRegImpl::stack2reg(xmm11H_off + additional_frame_slots), 255 xmm11->as_VMReg()->next()); 256 map->set_callee_saved(VMRegImpl::stack2reg(xmm12H_off + additional_frame_slots), 257 xmm12->as_VMReg()->next()); 258 map->set_callee_saved(VMRegImpl::stack2reg(xmm13H_off + additional_frame_slots), 259 xmm13->as_VMReg()->next()); 260 map->set_callee_saved(VMRegImpl::stack2reg(xmm14H_off + additional_frame_slots), 261 xmm14->as_VMReg()->next()); 262 map->set_callee_saved(VMRegImpl::stack2reg(xmm15H_off + additional_frame_slots), 263 xmm15->as_VMReg()->next()); 264 } 265 266 return map; 267 } 268 269 void RegisterSaver::restore_live_registers(MacroAssembler* masm) { 270 if (frame::arg_reg_save_area_bytes != 0) { 271 // Pop arg register save area 272 __ addptr(rsp, frame::arg_reg_save_area_bytes); 273 } 274 // Recover CPU state 275 __ pop_CPU_state(); 276 // Get the rbp described implicitly by the calling convention (no oopMap) 277 __ pop(rbp); 278 } 279 280 void RegisterSaver::restore_result_registers(MacroAssembler* masm) { 281 282 // Just restore result register. Only used by deoptimization. By 283 // now any callee save register that needs to be restored to a c2 284 // caller of the deoptee has been extracted into the vframeArray 285 // and will be stuffed into the c2i adapter we create for later 286 // restoration so only result registers need to be restored here. 287 288 // Restore fp result register 289 __ movdbl(xmm0, Address(rsp, xmm0_offset_in_bytes())); 290 // Restore integer result register 291 __ movptr(rax, Address(rsp, rax_offset_in_bytes())); 292 __ movptr(rdx, Address(rsp, rdx_offset_in_bytes())); 293 294 // Pop all of the register save are off the stack except the return address 295 __ addptr(rsp, return_offset_in_bytes()); 296 } 297 298 // The java_calling_convention describes stack locations as ideal slots on 299 // a frame with no abi restrictions. Since we must observe abi restrictions 300 // (like the placement of the register window) the slots must be biased by 301 // the following value. 302 static int reg2offset_in(VMReg r) { 303 // Account for saved rbp and return address 304 // This should really be in_preserve_stack_slots 305 return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size; 306 } 307 308 static int reg2offset_out(VMReg r) { 309 return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size; 310 } 311 312 // --------------------------------------------------------------------------- 313 // Read the array of BasicTypes from a signature, and compute where the 314 // arguments should go. Values in the VMRegPair regs array refer to 4-byte 315 // quantities. Values less than VMRegImpl::stack0 are registers, those above 316 // refer to 4-byte stack slots. All stack slots are based off of the stack pointer 317 // as framesizes are fixed. 318 // VMRegImpl::stack0 refers to the first slot 0(sp). 319 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher. Register 320 // up to RegisterImpl::number_of_registers) are the 64-bit 321 // integer registers. 322 323 // Note: the INPUTS in sig_bt are in units of Java argument words, which are 324 // either 32-bit or 64-bit depending on the build. The OUTPUTS are in 32-bit 325 // units regardless of build. Of course for i486 there is no 64 bit build 326 327 // The Java calling convention is a "shifted" version of the C ABI. 328 // By skipping the first C ABI register we can call non-static jni methods 329 // with small numbers of arguments without having to shuffle the arguments 330 // at all. Since we control the java ABI we ought to at least get some 331 // advantage out of it. 332 333 int SharedRuntime::java_calling_convention(const BasicType *sig_bt, 334 VMRegPair *regs, 335 int total_args_passed, 336 int is_outgoing) { 337 338 // Create the mapping between argument positions and 339 // registers. 340 static const Register INT_ArgReg[Argument::n_int_register_parameters_j] = { 341 j_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4, j_rarg5 342 }; 343 static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_j] = { 344 j_farg0, j_farg1, j_farg2, j_farg3, 345 j_farg4, j_farg5, j_farg6, j_farg7 346 }; 347 348 349 uint int_args = 0; 350 uint fp_args = 0; 351 uint stk_args = 0; // inc by 2 each time 352 353 for (int i = 0; i < total_args_passed; i++) { 354 switch (sig_bt[i]) { 355 case T_BOOLEAN: 356 case T_CHAR: 357 case T_BYTE: 358 case T_SHORT: 359 case T_INT: 360 if (int_args < Argument::n_int_register_parameters_j) { 361 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg()); 362 } else { 363 regs[i].set1(VMRegImpl::stack2reg(stk_args)); 364 stk_args += 2; 365 } 366 break; 367 case T_VOID: 368 // halves of T_LONG or T_DOUBLE 369 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half"); 370 regs[i].set_bad(); 371 break; 372 case T_LONG: 373 assert(sig_bt[i + 1] == T_VOID, "expecting half"); 374 // fall through 375 case T_OBJECT: 376 case T_ARRAY: 377 case T_ADDRESS: 378 if (int_args < Argument::n_int_register_parameters_j) { 379 regs[i].set2(INT_ArgReg[int_args++]->as_VMReg()); 380 } else { 381 regs[i].set2(VMRegImpl::stack2reg(stk_args)); 382 stk_args += 2; 383 } 384 break; 385 case T_FLOAT: 386 if (fp_args < Argument::n_float_register_parameters_j) { 387 regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg()); 388 } else { 389 regs[i].set1(VMRegImpl::stack2reg(stk_args)); 390 stk_args += 2; 391 } 392 break; 393 case T_DOUBLE: 394 assert(sig_bt[i + 1] == T_VOID, "expecting half"); 395 if (fp_args < Argument::n_float_register_parameters_j) { 396 regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg()); 397 } else { 398 regs[i].set2(VMRegImpl::stack2reg(stk_args)); 399 stk_args += 2; 400 } 401 break; 402 default: 403 ShouldNotReachHere(); 404 break; 405 } 406 } 407 408 return round_to(stk_args, 2); 409 } 410 411 // Patch the callers callsite with entry to compiled code if it exists. 412 static void patch_callers_callsite(MacroAssembler *masm) { 413 Label L; 414 __ verify_oop(rbx); 415 __ cmpptr(Address(rbx, in_bytes(methodOopDesc::code_offset())), (int32_t)NULL_WORD); 416 __ jcc(Assembler::equal, L); 417 418 // Save the current stack pointer 419 __ mov(r13, rsp); 420 // Schedule the branch target address early. 421 // Call into the VM to patch the caller, then jump to compiled callee 422 // rax isn't live so capture return address while we easily can 423 __ movptr(rax, Address(rsp, 0)); 424 425 // align stack so push_CPU_state doesn't fault 426 __ andptr(rsp, -(StackAlignmentInBytes)); 427 __ push_CPU_state(); 428 429 430 __ verify_oop(rbx); 431 // VM needs caller's callsite 432 // VM needs target method 433 // This needs to be a long call since we will relocate this adapter to 434 // the codeBuffer and it may not reach 435 436 // Allocate argument register save area 437 if (frame::arg_reg_save_area_bytes != 0) { 438 __ subptr(rsp, frame::arg_reg_save_area_bytes); 439 } 440 __ mov(c_rarg0, rbx); 441 __ mov(c_rarg1, rax); 442 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite))); 443 444 // De-allocate argument register save area 445 if (frame::arg_reg_save_area_bytes != 0) { 446 __ addptr(rsp, frame::arg_reg_save_area_bytes); 447 } 448 449 __ pop_CPU_state(); 450 // restore sp 451 __ mov(rsp, r13); 452 __ bind(L); 453 } 454 455 // Helper function to put tags in interpreter stack. 456 static void tag_stack(MacroAssembler *masm, const BasicType sig, int st_off) { 457 if (TaggedStackInterpreter) { 458 int tag_offset = st_off + Interpreter::expr_tag_offset_in_bytes(0); 459 if (sig == T_OBJECT || sig == T_ARRAY) { 460 __ movptr(Address(rsp, tag_offset), (int32_t) frame::TagReference); 461 } else if (sig == T_LONG || sig == T_DOUBLE) { 462 int next_tag_offset = st_off + Interpreter::expr_tag_offset_in_bytes(1); 463 __ movptr(Address(rsp, next_tag_offset), (int32_t) frame::TagValue); 464 __ movptr(Address(rsp, tag_offset), (int32_t) frame::TagValue); 465 } else { 466 __ movptr(Address(rsp, tag_offset), (int32_t) frame::TagValue); 467 } 468 } 469 } 470 471 472 static void gen_c2i_adapter(MacroAssembler *masm, 473 int total_args_passed, 474 int comp_args_on_stack, 475 const BasicType *sig_bt, 476 const VMRegPair *regs, 477 Label& skip_fixup) { 478 // Before we get into the guts of the C2I adapter, see if we should be here 479 // at all. We've come from compiled code and are attempting to jump to the 480 // interpreter, which means the caller made a static call to get here 481 // (vcalls always get a compiled target if there is one). Check for a 482 // compiled target. If there is one, we need to patch the caller's call. 483 patch_callers_callsite(masm); 484 485 __ bind(skip_fixup); 486 487 // Since all args are passed on the stack, total_args_passed * 488 // Interpreter::stackElementSize is the space we need. Plus 1 because 489 // we also account for the return address location since 490 // we store it first rather than hold it in rax across all the shuffling 491 492 int extraspace = (total_args_passed * Interpreter::stackElementSize()) + wordSize; 493 494 // stack is aligned, keep it that way 495 extraspace = round_to(extraspace, 2*wordSize); 496 497 // Get return address 498 __ pop(rax); 499 500 // set senderSP value 501 __ mov(r13, rsp); 502 503 __ subptr(rsp, extraspace); 504 505 // Store the return address in the expected location 506 __ movptr(Address(rsp, 0), rax); 507 508 // Now write the args into the outgoing interpreter space 509 for (int i = 0; i < total_args_passed; i++) { 510 if (sig_bt[i] == T_VOID) { 511 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half"); 512 continue; 513 } 514 515 // offset to start parameters 516 int st_off = (total_args_passed - i) * Interpreter::stackElementSize() + 517 Interpreter::value_offset_in_bytes(); 518 int next_off = st_off - Interpreter::stackElementSize(); 519 520 // Say 4 args: 521 // i st_off 522 // 0 32 T_LONG 523 // 1 24 T_VOID 524 // 2 16 T_OBJECT 525 // 3 8 T_BOOL 526 // - 0 return address 527 // 528 // However to make thing extra confusing. Because we can fit a long/double in 529 // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter 530 // leaves one slot empty and only stores to a single slot. In this case the 531 // slot that is occupied is the T_VOID slot. See I said it was confusing. 532 533 VMReg r_1 = regs[i].first(); 534 VMReg r_2 = regs[i].second(); 535 if (!r_1->is_valid()) { 536 assert(!r_2->is_valid(), ""); 537 continue; 538 } 539 if (r_1->is_stack()) { 540 // memory to memory use rax 541 int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace; 542 if (!r_2->is_valid()) { 543 // sign extend?? 544 __ movl(rax, Address(rsp, ld_off)); 545 __ movptr(Address(rsp, st_off), rax); 546 tag_stack(masm, sig_bt[i], st_off); 547 548 } else { 549 550 __ movq(rax, Address(rsp, ld_off)); 551 552 // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG 553 // T_DOUBLE and T_LONG use two slots in the interpreter 554 if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) { 555 // ld_off == LSW, ld_off+wordSize == MSW 556 // st_off == MSW, next_off == LSW 557 __ movq(Address(rsp, next_off), rax); 558 #ifdef ASSERT 559 // Overwrite the unused slot with known junk 560 __ mov64(rax, CONST64(0xdeadffffdeadaaaa)); 561 __ movptr(Address(rsp, st_off), rax); 562 #endif /* ASSERT */ 563 tag_stack(masm, sig_bt[i], next_off); 564 } else { 565 __ movq(Address(rsp, st_off), rax); 566 tag_stack(masm, sig_bt[i], st_off); 567 } 568 } 569 } else if (r_1->is_Register()) { 570 Register r = r_1->as_Register(); 571 if (!r_2->is_valid()) { 572 // must be only an int (or less ) so move only 32bits to slot 573 // why not sign extend?? 574 __ movl(Address(rsp, st_off), r); 575 tag_stack(masm, sig_bt[i], st_off); 576 } else { 577 // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG 578 // T_DOUBLE and T_LONG use two slots in the interpreter 579 if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) { 580 // long/double in gpr 581 #ifdef ASSERT 582 // Overwrite the unused slot with known junk 583 __ mov64(rax, CONST64(0xdeadffffdeadaaab)); 584 __ movptr(Address(rsp, st_off), rax); 585 #endif /* ASSERT */ 586 __ movq(Address(rsp, next_off), r); 587 tag_stack(masm, sig_bt[i], next_off); 588 } else { 589 __ movptr(Address(rsp, st_off), r); 590 tag_stack(masm, sig_bt[i], st_off); 591 } 592 } 593 } else { 594 assert(r_1->is_XMMRegister(), ""); 595 if (!r_2->is_valid()) { 596 // only a float use just part of the slot 597 __ movflt(Address(rsp, st_off), r_1->as_XMMRegister()); 598 tag_stack(masm, sig_bt[i], st_off); 599 } else { 600 #ifdef ASSERT 601 // Overwrite the unused slot with known junk 602 __ mov64(rax, CONST64(0xdeadffffdeadaaac)); 603 __ movptr(Address(rsp, st_off), rax); 604 #endif /* ASSERT */ 605 __ movdbl(Address(rsp, next_off), r_1->as_XMMRegister()); 606 tag_stack(masm, sig_bt[i], next_off); 607 } 608 } 609 } 610 611 // Schedule the branch target address early. 612 __ movptr(rcx, Address(rbx, in_bytes(methodOopDesc::interpreter_entry_offset()))); 613 __ jmp(rcx); 614 } 615 616 static void gen_i2c_adapter(MacroAssembler *masm, 617 int total_args_passed, 618 int comp_args_on_stack, 619 const BasicType *sig_bt, 620 const VMRegPair *regs) { 621 622 // 623 // We will only enter here from an interpreted frame and never from after 624 // passing thru a c2i. Azul allowed this but we do not. If we lose the 625 // race and use a c2i we will remain interpreted for the race loser(s). 626 // This removes all sorts of headaches on the x86 side and also eliminates 627 // the possibility of having c2i -> i2c -> c2i -> ... endless transitions. 628 629 630 // Note: r13 contains the senderSP on entry. We must preserve it since 631 // we may do a i2c -> c2i transition if we lose a race where compiled 632 // code goes non-entrant while we get args ready. 633 // In addition we use r13 to locate all the interpreter args as 634 // we must align the stack to 16 bytes on an i2c entry else we 635 // lose alignment we expect in all compiled code and register 636 // save code can segv when fxsave instructions find improperly 637 // aligned stack pointer. 638 639 __ movptr(rax, Address(rsp, 0)); 640 641 // Must preserve original SP for loading incoming arguments because 642 // we need to align the outgoing SP for compiled code. 643 __ movptr(r11, rsp); 644 645 // Cut-out for having no stack args. Since up to 2 int/oop args are passed 646 // in registers, we will occasionally have no stack args. 647 int comp_words_on_stack = 0; 648 if (comp_args_on_stack) { 649 // Sig words on the stack are greater-than VMRegImpl::stack0. Those in 650 // registers are below. By subtracting stack0, we either get a negative 651 // number (all values in registers) or the maximum stack slot accessed. 652 653 // Convert 4-byte c2 stack slots to words. 654 comp_words_on_stack = round_to(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize)>>LogBytesPerWord; 655 // Round up to miminum stack alignment, in wordSize 656 comp_words_on_stack = round_to(comp_words_on_stack, 2); 657 __ subptr(rsp, comp_words_on_stack * wordSize); 658 } 659 660 661 // Ensure compiled code always sees stack at proper alignment 662 __ andptr(rsp, -16); 663 664 // push the return address and misalign the stack that youngest frame always sees 665 // as far as the placement of the call instruction 666 __ push(rax); 667 668 // Put saved SP in another register 669 const Register saved_sp = rax; 670 __ movptr(saved_sp, r11); 671 672 // Will jump to the compiled code just as if compiled code was doing it. 673 // Pre-load the register-jump target early, to schedule it better. 674 __ movptr(r11, Address(rbx, in_bytes(methodOopDesc::from_compiled_offset()))); 675 676 // Now generate the shuffle code. Pick up all register args and move the 677 // rest through the floating point stack top. 678 for (int i = 0; i < total_args_passed; i++) { 679 if (sig_bt[i] == T_VOID) { 680 // Longs and doubles are passed in native word order, but misaligned 681 // in the 32-bit build. 682 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half"); 683 continue; 684 } 685 686 // Pick up 0, 1 or 2 words from SP+offset. 687 688 assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(), 689 "scrambled load targets?"); 690 // Load in argument order going down. 691 int ld_off = (total_args_passed - i)*Interpreter::stackElementSize() + Interpreter::value_offset_in_bytes(); 692 // Point to interpreter value (vs. tag) 693 int next_off = ld_off - Interpreter::stackElementSize(); 694 // 695 // 696 // 697 VMReg r_1 = regs[i].first(); 698 VMReg r_2 = regs[i].second(); 699 if (!r_1->is_valid()) { 700 assert(!r_2->is_valid(), ""); 701 continue; 702 } 703 if (r_1->is_stack()) { 704 // Convert stack slot to an SP offset (+ wordSize to account for return address ) 705 int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size + wordSize; 706 707 // We can use r13 as a temp here because compiled code doesn't need r13 as an input 708 // and if we end up going thru a c2i because of a miss a reasonable value of r13 709 // will be generated. 710 if (!r_2->is_valid()) { 711 // sign extend??? 712 __ movl(r13, Address(saved_sp, ld_off)); 713 __ movptr(Address(rsp, st_off), r13); 714 } else { 715 // 716 // We are using two optoregs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE 717 // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case 718 // So we must adjust where to pick up the data to match the interpreter. 719 // 720 // Interpreter local[n] == MSW, local[n+1] == LSW however locals 721 // are accessed as negative so LSW is at LOW address 722 723 // ld_off is MSW so get LSW 724 const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)? 725 next_off : ld_off; 726 __ movq(r13, Address(saved_sp, offset)); 727 // st_off is LSW (i.e. reg.first()) 728 __ movq(Address(rsp, st_off), r13); 729 } 730 } else if (r_1->is_Register()) { // Register argument 731 Register r = r_1->as_Register(); 732 assert(r != rax, "must be different"); 733 if (r_2->is_valid()) { 734 // 735 // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE 736 // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case 737 // So we must adjust where to pick up the data to match the interpreter. 738 739 const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)? 740 next_off : ld_off; 741 742 // this can be a misaligned move 743 __ movq(r, Address(saved_sp, offset)); 744 } else { 745 // sign extend and use a full word? 746 __ movl(r, Address(saved_sp, ld_off)); 747 } 748 } else { 749 if (!r_2->is_valid()) { 750 __ movflt(r_1->as_XMMRegister(), Address(saved_sp, ld_off)); 751 } else { 752 __ movdbl(r_1->as_XMMRegister(), Address(saved_sp, next_off)); 753 } 754 } 755 } 756 757 // 6243940 We might end up in handle_wrong_method if 758 // the callee is deoptimized as we race thru here. If that 759 // happens we don't want to take a safepoint because the 760 // caller frame will look interpreted and arguments are now 761 // "compiled" so it is much better to make this transition 762 // invisible to the stack walking code. Unfortunately if 763 // we try and find the callee by normal means a safepoint 764 // is possible. So we stash the desired callee in the thread 765 // and the vm will find there should this case occur. 766 767 __ movptr(Address(r15_thread, JavaThread::callee_target_offset()), rbx); 768 769 // put methodOop where a c2i would expect should we end up there 770 // only needed becaus eof c2 resolve stubs return methodOop as a result in 771 // rax 772 __ mov(rax, rbx); 773 __ jmp(r11); 774 } 775 776 // --------------------------------------------------------------- 777 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm, 778 int total_args_passed, 779 int comp_args_on_stack, 780 const BasicType *sig_bt, 781 const VMRegPair *regs, 782 AdapterFingerPrint* fingerprint) { 783 address i2c_entry = __ pc(); 784 785 gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs); 786 787 // ------------------------------------------------------------------------- 788 // Generate a C2I adapter. On entry we know rbx holds the methodOop during calls 789 // to the interpreter. The args start out packed in the compiled layout. They 790 // need to be unpacked into the interpreter layout. This will almost always 791 // require some stack space. We grow the current (compiled) stack, then repack 792 // the args. We finally end in a jump to the generic interpreter entry point. 793 // On exit from the interpreter, the interpreter will restore our SP (lest the 794 // compiled code, which relys solely on SP and not RBP, get sick). 795 796 address c2i_unverified_entry = __ pc(); 797 Label skip_fixup; 798 Label ok; 799 800 Register holder = rax; 801 Register receiver = j_rarg0; 802 Register temp = rbx; 803 804 { 805 __ verify_oop(holder); 806 __ load_klass(temp, receiver); 807 __ verify_oop(temp); 808 809 __ cmpptr(temp, Address(holder, compiledICHolderOopDesc::holder_klass_offset())); 810 __ movptr(rbx, Address(holder, compiledICHolderOopDesc::holder_method_offset())); 811 __ jcc(Assembler::equal, ok); 812 __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); 813 814 __ bind(ok); 815 // Method might have been compiled since the call site was patched to 816 // interpreted if that is the case treat it as a miss so we can get 817 // the call site corrected. 818 __ cmpptr(Address(rbx, in_bytes(methodOopDesc::code_offset())), (int32_t)NULL_WORD); 819 __ jcc(Assembler::equal, skip_fixup); 820 __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); 821 } 822 823 address c2i_entry = __ pc(); 824 825 gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup); 826 827 __ flush(); 828 return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry); 829 } 830 831 int SharedRuntime::c_calling_convention(const BasicType *sig_bt, 832 VMRegPair *regs, 833 int total_args_passed) { 834 // We return the amount of VMRegImpl stack slots we need to reserve for all 835 // the arguments NOT counting out_preserve_stack_slots. 836 837 // NOTE: These arrays will have to change when c1 is ported 838 #ifdef _WIN64 839 static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = { 840 c_rarg0, c_rarg1, c_rarg2, c_rarg3 841 }; 842 static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_c] = { 843 c_farg0, c_farg1, c_farg2, c_farg3 844 }; 845 #else 846 static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = { 847 c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, c_rarg5 848 }; 849 static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_c] = { 850 c_farg0, c_farg1, c_farg2, c_farg3, 851 c_farg4, c_farg5, c_farg6, c_farg7 852 }; 853 #endif // _WIN64 854 855 856 uint int_args = 0; 857 uint fp_args = 0; 858 uint stk_args = 0; // inc by 2 each time 859 860 for (int i = 0; i < total_args_passed; i++) { 861 switch (sig_bt[i]) { 862 case T_BOOLEAN: 863 case T_CHAR: 864 case T_BYTE: 865 case T_SHORT: 866 case T_INT: 867 if (int_args < Argument::n_int_register_parameters_c) { 868 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg()); 869 #ifdef _WIN64 870 fp_args++; 871 // Allocate slots for callee to stuff register args the stack. 872 stk_args += 2; 873 #endif 874 } else { 875 regs[i].set1(VMRegImpl::stack2reg(stk_args)); 876 stk_args += 2; 877 } 878 break; 879 case T_LONG: 880 assert(sig_bt[i + 1] == T_VOID, "expecting half"); 881 // fall through 882 case T_OBJECT: 883 case T_ARRAY: 884 case T_ADDRESS: 885 if (int_args < Argument::n_int_register_parameters_c) { 886 regs[i].set2(INT_ArgReg[int_args++]->as_VMReg()); 887 #ifdef _WIN64 888 fp_args++; 889 stk_args += 2; 890 #endif 891 } else { 892 regs[i].set2(VMRegImpl::stack2reg(stk_args)); 893 stk_args += 2; 894 } 895 break; 896 case T_FLOAT: 897 if (fp_args < Argument::n_float_register_parameters_c) { 898 regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg()); 899 #ifdef _WIN64 900 int_args++; 901 // Allocate slots for callee to stuff register args the stack. 902 stk_args += 2; 903 #endif 904 } else { 905 regs[i].set1(VMRegImpl::stack2reg(stk_args)); 906 stk_args += 2; 907 } 908 break; 909 case T_DOUBLE: 910 assert(sig_bt[i + 1] == T_VOID, "expecting half"); 911 if (fp_args < Argument::n_float_register_parameters_c) { 912 regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg()); 913 #ifdef _WIN64 914 int_args++; 915 // Allocate slots for callee to stuff register args the stack. 916 stk_args += 2; 917 #endif 918 } else { 919 regs[i].set2(VMRegImpl::stack2reg(stk_args)); 920 stk_args += 2; 921 } 922 break; 923 case T_VOID: // Halves of longs and doubles 924 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half"); 925 regs[i].set_bad(); 926 break; 927 default: 928 ShouldNotReachHere(); 929 break; 930 } 931 } 932 #ifdef _WIN64 933 // windows abi requires that we always allocate enough stack space 934 // for 4 64bit registers to be stored down. 935 if (stk_args < 8) { 936 stk_args = 8; 937 } 938 #endif // _WIN64 939 940 return stk_args; 941 } 942 943 // On 64 bit we will store integer like items to the stack as 944 // 64 bits items (sparc abi) even though java would only store 945 // 32bits for a parameter. On 32bit it will simply be 32 bits 946 // So this routine will do 32->32 on 32bit and 32->64 on 64bit 947 static void move32_64(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { 948 if (src.first()->is_stack()) { 949 if (dst.first()->is_stack()) { 950 // stack to stack 951 __ movslq(rax, Address(rbp, reg2offset_in(src.first()))); 952 __ movq(Address(rsp, reg2offset_out(dst.first())), rax); 953 } else { 954 // stack to reg 955 __ movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()))); 956 } 957 } else if (dst.first()->is_stack()) { 958 // reg to stack 959 // Do we really have to sign extend??? 960 // __ movslq(src.first()->as_Register(), src.first()->as_Register()); 961 __ movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register()); 962 } else { 963 // Do we really have to sign extend??? 964 // __ movslq(dst.first()->as_Register(), src.first()->as_Register()); 965 if (dst.first() != src.first()) { 966 __ movq(dst.first()->as_Register(), src.first()->as_Register()); 967 } 968 } 969 } 970 971 972 // An oop arg. Must pass a handle not the oop itself 973 static void object_move(MacroAssembler* masm, 974 OopMap* map, 975 int oop_handle_offset, 976 int framesize_in_slots, 977 VMRegPair src, 978 VMRegPair dst, 979 bool is_receiver, 980 int* receiver_offset) { 981 982 // must pass a handle. First figure out the location we use as a handle 983 984 Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register(); 985 986 // See if oop is NULL if it is we need no handle 987 988 if (src.first()->is_stack()) { 989 990 // Oop is already on the stack as an argument 991 int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots(); 992 map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots)); 993 if (is_receiver) { 994 *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size; 995 } 996 997 __ cmpptr(Address(rbp, reg2offset_in(src.first())), (int32_t)NULL_WORD); 998 __ lea(rHandle, Address(rbp, reg2offset_in(src.first()))); 999 // conditionally move a NULL 1000 __ cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first()))); 1001 } else { 1002 1003 // Oop is in an a register we must store it to the space we reserve 1004 // on the stack for oop_handles and pass a handle if oop is non-NULL 1005 1006 const Register rOop = src.first()->as_Register(); 1007 int oop_slot; 1008 if (rOop == j_rarg0) 1009 oop_slot = 0; 1010 else if (rOop == j_rarg1) 1011 oop_slot = 1; 1012 else if (rOop == j_rarg2) 1013 oop_slot = 2; 1014 else if (rOop == j_rarg3) 1015 oop_slot = 3; 1016 else if (rOop == j_rarg4) 1017 oop_slot = 4; 1018 else { 1019 assert(rOop == j_rarg5, "wrong register"); 1020 oop_slot = 5; 1021 } 1022 1023 oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset; 1024 int offset = oop_slot*VMRegImpl::stack_slot_size; 1025 1026 map->set_oop(VMRegImpl::stack2reg(oop_slot)); 1027 // Store oop in handle area, may be NULL 1028 __ movptr(Address(rsp, offset), rOop); 1029 if (is_receiver) { 1030 *receiver_offset = offset; 1031 } 1032 1033 __ cmpptr(rOop, (int32_t)NULL_WORD); 1034 __ lea(rHandle, Address(rsp, offset)); 1035 // conditionally move a NULL from the handle area where it was just stored 1036 __ cmovptr(Assembler::equal, rHandle, Address(rsp, offset)); 1037 } 1038 1039 // If arg is on the stack then place it otherwise it is already in correct reg. 1040 if (dst.first()->is_stack()) { 1041 __ movptr(Address(rsp, reg2offset_out(dst.first())), rHandle); 1042 } 1043 } 1044 1045 // A float arg may have to do float reg int reg conversion 1046 static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { 1047 assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move"); 1048 1049 // The calling conventions assures us that each VMregpair is either 1050 // all really one physical register or adjacent stack slots. 1051 // This greatly simplifies the cases here compared to sparc. 1052 1053 if (src.first()->is_stack()) { 1054 if (dst.first()->is_stack()) { 1055 __ movl(rax, Address(rbp, reg2offset_in(src.first()))); 1056 __ movptr(Address(rsp, reg2offset_out(dst.first())), rax); 1057 } else { 1058 // stack to reg 1059 assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters"); 1060 __ movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()))); 1061 } 1062 } else if (dst.first()->is_stack()) { 1063 // reg to stack 1064 assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters"); 1065 __ movflt(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister()); 1066 } else { 1067 // reg to reg 1068 // In theory these overlap but the ordering is such that this is likely a nop 1069 if ( src.first() != dst.first()) { 1070 __ movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister()); 1071 } 1072 } 1073 } 1074 1075 // A long move 1076 static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { 1077 1078 // The calling conventions assures us that each VMregpair is either 1079 // all really one physical register or adjacent stack slots. 1080 // This greatly simplifies the cases here compared to sparc. 1081 1082 if (src.is_single_phys_reg() ) { 1083 if (dst.is_single_phys_reg()) { 1084 if (dst.first() != src.first()) { 1085 __ mov(dst.first()->as_Register(), src.first()->as_Register()); 1086 } 1087 } else { 1088 assert(dst.is_single_reg(), "not a stack pair"); 1089 __ movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register()); 1090 } 1091 } else if (dst.is_single_phys_reg()) { 1092 assert(src.is_single_reg(), "not a stack pair"); 1093 __ movq(dst.first()->as_Register(), Address(rbp, reg2offset_out(src.first()))); 1094 } else { 1095 assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs"); 1096 __ movq(rax, Address(rbp, reg2offset_in(src.first()))); 1097 __ movq(Address(rsp, reg2offset_out(dst.first())), rax); 1098 } 1099 } 1100 1101 // A double move 1102 static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { 1103 1104 // The calling conventions assures us that each VMregpair is either 1105 // all really one physical register or adjacent stack slots. 1106 // This greatly simplifies the cases here compared to sparc. 1107 1108 if (src.is_single_phys_reg() ) { 1109 if (dst.is_single_phys_reg()) { 1110 // In theory these overlap but the ordering is such that this is likely a nop 1111 if ( src.first() != dst.first()) { 1112 __ movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister()); 1113 } 1114 } else { 1115 assert(dst.is_single_reg(), "not a stack pair"); 1116 __ movdbl(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister()); 1117 } 1118 } else if (dst.is_single_phys_reg()) { 1119 assert(src.is_single_reg(), "not a stack pair"); 1120 __ movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_out(src.first()))); 1121 } else { 1122 assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs"); 1123 __ movq(rax, Address(rbp, reg2offset_in(src.first()))); 1124 __ movq(Address(rsp, reg2offset_out(dst.first())), rax); 1125 } 1126 } 1127 1128 1129 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { 1130 // We always ignore the frame_slots arg and just use the space just below frame pointer 1131 // which by this time is free to use 1132 switch (ret_type) { 1133 case T_FLOAT: 1134 __ movflt(Address(rbp, -wordSize), xmm0); 1135 break; 1136 case T_DOUBLE: 1137 __ movdbl(Address(rbp, -wordSize), xmm0); 1138 break; 1139 case T_VOID: break; 1140 default: { 1141 __ movptr(Address(rbp, -wordSize), rax); 1142 } 1143 } 1144 } 1145 1146 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { 1147 // We always ignore the frame_slots arg and just use the space just below frame pointer 1148 // which by this time is free to use 1149 switch (ret_type) { 1150 case T_FLOAT: 1151 __ movflt(xmm0, Address(rbp, -wordSize)); 1152 break; 1153 case T_DOUBLE: 1154 __ movdbl(xmm0, Address(rbp, -wordSize)); 1155 break; 1156 case T_VOID: break; 1157 default: { 1158 __ movptr(rax, Address(rbp, -wordSize)); 1159 } 1160 } 1161 } 1162 1163 static void save_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) { 1164 for ( int i = first_arg ; i < arg_count ; i++ ) { 1165 if (args[i].first()->is_Register()) { 1166 __ push(args[i].first()->as_Register()); 1167 } else if (args[i].first()->is_XMMRegister()) { 1168 __ subptr(rsp, 2*wordSize); 1169 __ movdbl(Address(rsp, 0), args[i].first()->as_XMMRegister()); 1170 } 1171 } 1172 } 1173 1174 static void restore_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) { 1175 for ( int i = arg_count - 1 ; i >= first_arg ; i-- ) { 1176 if (args[i].first()->is_Register()) { 1177 __ pop(args[i].first()->as_Register()); 1178 } else if (args[i].first()->is_XMMRegister()) { 1179 __ movdbl(args[i].first()->as_XMMRegister(), Address(rsp, 0)); 1180 __ addptr(rsp, 2*wordSize); 1181 } 1182 } 1183 } 1184 1185 // --------------------------------------------------------------------------- 1186 // Generate a native wrapper for a given method. The method takes arguments 1187 // in the Java compiled code convention, marshals them to the native 1188 // convention (handlizes oops, etc), transitions to native, makes the call, 1189 // returns to java state (possibly blocking), unhandlizes any result and 1190 // returns. 1191 nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler *masm, 1192 methodHandle method, 1193 int total_in_args, 1194 int comp_args_on_stack, 1195 BasicType *in_sig_bt, 1196 VMRegPair *in_regs, 1197 BasicType ret_type) { 1198 // Native nmethod wrappers never take possesion of the oop arguments. 1199 // So the caller will gc the arguments. The only thing we need an 1200 // oopMap for is if the call is static 1201 // 1202 // An OopMap for lock (and class if static) 1203 OopMapSet *oop_maps = new OopMapSet(); 1204 intptr_t start = (intptr_t)__ pc(); 1205 1206 // We have received a description of where all the java arg are located 1207 // on entry to the wrapper. We need to convert these args to where 1208 // the jni function will expect them. To figure out where they go 1209 // we convert the java signature to a C signature by inserting 1210 // the hidden arguments as arg[0] and possibly arg[1] (static method) 1211 1212 int total_c_args = total_in_args + 1; 1213 if (method->is_static()) { 1214 total_c_args++; 1215 } 1216 1217 BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args); 1218 VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args); 1219 1220 int argc = 0; 1221 out_sig_bt[argc++] = T_ADDRESS; 1222 if (method->is_static()) { 1223 out_sig_bt[argc++] = T_OBJECT; 1224 } 1225 1226 for (int i = 0; i < total_in_args ; i++ ) { 1227 out_sig_bt[argc++] = in_sig_bt[i]; 1228 } 1229 1230 // Now figure out where the args must be stored and how much stack space 1231 // they require. 1232 // 1233 int out_arg_slots; 1234 out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args); 1235 1236 // Compute framesize for the wrapper. We need to handlize all oops in 1237 // incoming registers 1238 1239 // Calculate the total number of stack slots we will need. 1240 1241 // First count the abi requirement plus all of the outgoing args 1242 int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; 1243 1244 // Now the space for the inbound oop handle area 1245 1246 int oop_handle_offset = stack_slots; 1247 stack_slots += 6*VMRegImpl::slots_per_word; 1248 1249 // Now any space we need for handlizing a klass if static method 1250 1251 int oop_temp_slot_offset = 0; 1252 int klass_slot_offset = 0; 1253 int klass_offset = -1; 1254 int lock_slot_offset = 0; 1255 bool is_static = false; 1256 1257 if (method->is_static()) { 1258 klass_slot_offset = stack_slots; 1259 stack_slots += VMRegImpl::slots_per_word; 1260 klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size; 1261 is_static = true; 1262 } 1263 1264 // Plus a lock if needed 1265 1266 if (method->is_synchronized()) { 1267 lock_slot_offset = stack_slots; 1268 stack_slots += VMRegImpl::slots_per_word; 1269 } 1270 1271 // Now a place (+2) to save return values or temp during shuffling 1272 // + 4 for return address (which we own) and saved rbp 1273 stack_slots += 6; 1274 1275 // Ok The space we have allocated will look like: 1276 // 1277 // 1278 // FP-> | | 1279 // |---------------------| 1280 // | 2 slots for moves | 1281 // |---------------------| 1282 // | lock box (if sync) | 1283 // |---------------------| <- lock_slot_offset 1284 // | klass (if static) | 1285 // |---------------------| <- klass_slot_offset 1286 // | oopHandle area | 1287 // |---------------------| <- oop_handle_offset (6 java arg registers) 1288 // | outbound memory | 1289 // | based arguments | 1290 // | | 1291 // |---------------------| 1292 // | | 1293 // SP-> | out_preserved_slots | 1294 // 1295 // 1296 1297 1298 // Now compute actual number of stack words we need rounding to make 1299 // stack properly aligned. 1300 stack_slots = round_to(stack_slots, StackAlignmentInSlots); 1301 1302 int stack_size = stack_slots * VMRegImpl::stack_slot_size; 1303 1304 1305 // First thing make an ic check to see if we should even be here 1306 1307 // We are free to use all registers as temps without saving them and 1308 // restoring them except rbp. rbp is the only callee save register 1309 // as far as the interpreter and the compiler(s) are concerned. 1310 1311 1312 const Register ic_reg = rax; 1313 const Register receiver = j_rarg0; 1314 1315 Label ok; 1316 Label exception_pending; 1317 1318 assert_different_registers(ic_reg, receiver, rscratch1); 1319 __ verify_oop(receiver); 1320 __ load_klass(rscratch1, receiver); 1321 __ cmpq(ic_reg, rscratch1); 1322 __ jcc(Assembler::equal, ok); 1323 1324 __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); 1325 1326 __ bind(ok); 1327 1328 // Verified entry point must be aligned 1329 __ align(8); 1330 1331 int vep_offset = ((intptr_t)__ pc()) - start; 1332 1333 // The instruction at the verified entry point must be 5 bytes or longer 1334 // because it can be patched on the fly by make_non_entrant. The stack bang 1335 // instruction fits that requirement. 1336 1337 // Generate stack overflow check 1338 1339 if (UseStackBanging) { 1340 __ bang_stack_with_offset(StackShadowPages*os::vm_page_size()); 1341 } else { 1342 // need a 5 byte instruction to allow MT safe patching to non-entrant 1343 __ fat_nop(); 1344 } 1345 1346 // Generate a new frame for the wrapper. 1347 __ enter(); 1348 // -2 because return address is already present and so is saved rbp 1349 __ subptr(rsp, stack_size - 2*wordSize); 1350 1351 // Frame is now completed as far as size and linkage. 1352 1353 int frame_complete = ((intptr_t)__ pc()) - start; 1354 1355 #ifdef ASSERT 1356 { 1357 Label L; 1358 __ mov(rax, rsp); 1359 __ andptr(rax, -16); // must be 16 byte boundary (see amd64 ABI) 1360 __ cmpptr(rax, rsp); 1361 __ jcc(Assembler::equal, L); 1362 __ stop("improperly aligned stack"); 1363 __ bind(L); 1364 } 1365 #endif /* ASSERT */ 1366 1367 1368 // We use r14 as the oop handle for the receiver/klass 1369 // It is callee save so it survives the call to native 1370 1371 const Register oop_handle_reg = r14; 1372 1373 1374 1375 // 1376 // We immediately shuffle the arguments so that any vm call we have to 1377 // make from here on out (sync slow path, jvmti, etc.) we will have 1378 // captured the oops from our caller and have a valid oopMap for 1379 // them. 1380 1381 // ----------------- 1382 // The Grand Shuffle 1383 1384 // The Java calling convention is either equal (linux) or denser (win64) than the 1385 // c calling convention. However the because of the jni_env argument the c calling 1386 // convention always has at least one more (and two for static) arguments than Java. 1387 // Therefore if we move the args from java -> c backwards then we will never have 1388 // a register->register conflict and we don't have to build a dependency graph 1389 // and figure out how to break any cycles. 1390 // 1391 1392 // Record esp-based slot for receiver on stack for non-static methods 1393 int receiver_offset = -1; 1394 1395 // This is a trick. We double the stack slots so we can claim 1396 // the oops in the caller's frame. Since we are sure to have 1397 // more args than the caller doubling is enough to make 1398 // sure we can capture all the incoming oop args from the 1399 // caller. 1400 // 1401 OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/); 1402 1403 // Mark location of rbp (someday) 1404 // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, vmreg(rbp)); 1405 1406 // Use eax, ebx as temporaries during any memory-memory moves we have to do 1407 // All inbound args are referenced based on rbp and all outbound args via rsp. 1408 1409 1410 #ifdef ASSERT 1411 bool reg_destroyed[RegisterImpl::number_of_registers]; 1412 bool freg_destroyed[XMMRegisterImpl::number_of_registers]; 1413 for ( int r = 0 ; r < RegisterImpl::number_of_registers ; r++ ) { 1414 reg_destroyed[r] = false; 1415 } 1416 for ( int f = 0 ; f < XMMRegisterImpl::number_of_registers ; f++ ) { 1417 freg_destroyed[f] = false; 1418 } 1419 1420 #endif /* ASSERT */ 1421 1422 1423 int c_arg = total_c_args - 1; 1424 for ( int i = total_in_args - 1; i >= 0 ; i--, c_arg-- ) { 1425 #ifdef ASSERT 1426 if (in_regs[i].first()->is_Register()) { 1427 assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!"); 1428 } else if (in_regs[i].first()->is_XMMRegister()) { 1429 assert(!freg_destroyed[in_regs[i].first()->as_XMMRegister()->encoding()], "destroyed reg!"); 1430 } 1431 if (out_regs[c_arg].first()->is_Register()) { 1432 reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true; 1433 } else if (out_regs[c_arg].first()->is_XMMRegister()) { 1434 freg_destroyed[out_regs[c_arg].first()->as_XMMRegister()->encoding()] = true; 1435 } 1436 #endif /* ASSERT */ 1437 switch (in_sig_bt[i]) { 1438 case T_ARRAY: 1439 case T_OBJECT: 1440 object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg], 1441 ((i == 0) && (!is_static)), 1442 &receiver_offset); 1443 break; 1444 case T_VOID: 1445 break; 1446 1447 case T_FLOAT: 1448 float_move(masm, in_regs[i], out_regs[c_arg]); 1449 break; 1450 1451 case T_DOUBLE: 1452 assert( i + 1 < total_in_args && 1453 in_sig_bt[i + 1] == T_VOID && 1454 out_sig_bt[c_arg+1] == T_VOID, "bad arg list"); 1455 double_move(masm, in_regs[i], out_regs[c_arg]); 1456 break; 1457 1458 case T_LONG : 1459 long_move(masm, in_regs[i], out_regs[c_arg]); 1460 break; 1461 1462 case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); 1463 1464 default: 1465 move32_64(masm, in_regs[i], out_regs[c_arg]); 1466 } 1467 } 1468 1469 // point c_arg at the first arg that is already loaded in case we 1470 // need to spill before we call out 1471 c_arg++; 1472 1473 // Pre-load a static method's oop into r14. Used both by locking code and 1474 // the normal JNI call code. 1475 if (method->is_static()) { 1476 1477 // load oop into a register 1478 __ movoop(oop_handle_reg, JNIHandles::make_local(Klass::cast(method->method_holder())->java_mirror())); 1479 1480 // Now handlize the static class mirror it's known not-null. 1481 __ movptr(Address(rsp, klass_offset), oop_handle_reg); 1482 map->set_oop(VMRegImpl::stack2reg(klass_slot_offset)); 1483 1484 // Now get the handle 1485 __ lea(oop_handle_reg, Address(rsp, klass_offset)); 1486 // store the klass handle as second argument 1487 __ movptr(c_rarg1, oop_handle_reg); 1488 // and protect the arg if we must spill 1489 c_arg--; 1490 } 1491 1492 // Change state to native (we save the return address in the thread, since it might not 1493 // be pushed on the stack when we do a a stack traversal). It is enough that the pc() 1494 // points into the right code segment. It does not have to be the correct return pc. 1495 // We use the same pc/oopMap repeatedly when we call out 1496 1497 intptr_t the_pc = (intptr_t) __ pc(); 1498 oop_maps->add_gc_map(the_pc - start, map); 1499 1500 __ set_last_Java_frame(rsp, noreg, (address)the_pc); 1501 1502 1503 // We have all of the arguments setup at this point. We must not touch any register 1504 // argument registers at this point (what if we save/restore them there are no oop? 1505 1506 { 1507 SkipIfEqual skip(masm, &DTraceMethodProbes, false); 1508 // protect the args we've loaded 1509 save_args(masm, total_c_args, c_arg, out_regs); 1510 __ movoop(c_rarg1, JNIHandles::make_local(method())); 1511 __ call_VM_leaf( 1512 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 1513 r15_thread, c_rarg1); 1514 restore_args(masm, total_c_args, c_arg, out_regs); 1515 } 1516 1517 // RedefineClasses() tracing support for obsolete method entry 1518 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { 1519 // protect the args we've loaded 1520 save_args(masm, total_c_args, c_arg, out_regs); 1521 __ movoop(c_rarg1, JNIHandles::make_local(method())); 1522 __ call_VM_leaf( 1523 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 1524 r15_thread, c_rarg1); 1525 restore_args(masm, total_c_args, c_arg, out_regs); 1526 } 1527 1528 // Lock a synchronized method 1529 1530 // Register definitions used by locking and unlocking 1531 1532 const Register swap_reg = rax; // Must use rax for cmpxchg instruction 1533 const Register obj_reg = rbx; // Will contain the oop 1534 const Register lock_reg = r13; // Address of compiler lock object (BasicLock) 1535 const Register old_hdr = r13; // value of old header at unlock time 1536 1537 Label slow_path_lock; 1538 Label lock_done; 1539 1540 if (method->is_synchronized()) { 1541 1542 1543 const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes(); 1544 1545 // Get the handle (the 2nd argument) 1546 __ mov(oop_handle_reg, c_rarg1); 1547 1548 // Get address of the box 1549 1550 __ lea(lock_reg, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size)); 1551 1552 // Load the oop from the handle 1553 __ movptr(obj_reg, Address(oop_handle_reg, 0)); 1554 1555 if (UseBiasedLocking) { 1556 __ biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, lock_done, &slow_path_lock); 1557 } 1558 1559 // Load immediate 1 into swap_reg %rax 1560 __ movl(swap_reg, 1); 1561 1562 // Load (object->mark() | 1) into swap_reg %rax 1563 __ orptr(swap_reg, Address(obj_reg, 0)); 1564 1565 // Save (object->mark() | 1) into BasicLock's displaced header 1566 __ movptr(Address(lock_reg, mark_word_offset), swap_reg); 1567 1568 if (os::is_MP()) { 1569 __ lock(); 1570 } 1571 1572 // src -> dest iff dest == rax else rax <- dest 1573 __ cmpxchgptr(lock_reg, Address(obj_reg, 0)); 1574 __ jcc(Assembler::equal, lock_done); 1575 1576 // Hmm should this move to the slow path code area??? 1577 1578 // Test if the oopMark is an obvious stack pointer, i.e., 1579 // 1) (mark & 3) == 0, and 1580 // 2) rsp <= mark < mark + os::pagesize() 1581 // These 3 tests can be done by evaluating the following 1582 // expression: ((mark - rsp) & (3 - os::vm_page_size())), 1583 // assuming both stack pointer and pagesize have their 1584 // least significant 2 bits clear. 1585 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg 1586 1587 __ subptr(swap_reg, rsp); 1588 __ andptr(swap_reg, 3 - os::vm_page_size()); 1589 1590 // Save the test result, for recursive case, the result is zero 1591 __ movptr(Address(lock_reg, mark_word_offset), swap_reg); 1592 __ jcc(Assembler::notEqual, slow_path_lock); 1593 1594 // Slow path will re-enter here 1595 1596 __ bind(lock_done); 1597 } 1598 1599 1600 // Finally just about ready to make the JNI call 1601 1602 1603 // get JNIEnv* which is first argument to native 1604 1605 __ lea(c_rarg0, Address(r15_thread, in_bytes(JavaThread::jni_environment_offset()))); 1606 1607 // Now set thread in native 1608 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native); 1609 1610 __ call(RuntimeAddress(method->native_function())); 1611 1612 // Either restore the MXCSR register after returning from the JNI Call 1613 // or verify that it wasn't changed. 1614 if (RestoreMXCSROnJNICalls) { 1615 __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std())); 1616 1617 } 1618 else if (CheckJNICalls ) { 1619 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry()))); 1620 } 1621 1622 1623 // Unpack native results. 1624 switch (ret_type) { 1625 case T_BOOLEAN: __ c2bool(rax); break; 1626 case T_CHAR : __ movzwl(rax, rax); break; 1627 case T_BYTE : __ sign_extend_byte (rax); break; 1628 case T_SHORT : __ sign_extend_short(rax); break; 1629 case T_INT : /* nothing to do */ break; 1630 case T_DOUBLE : 1631 case T_FLOAT : 1632 // Result is in xmm0 we'll save as needed 1633 break; 1634 case T_ARRAY: // Really a handle 1635 case T_OBJECT: // Really a handle 1636 break; // can't de-handlize until after safepoint check 1637 case T_VOID: break; 1638 case T_LONG: break; 1639 default : ShouldNotReachHere(); 1640 } 1641 1642 // Switch thread to "native transition" state before reading the synchronization state. 1643 // This additional state is necessary because reading and testing the synchronization 1644 // state is not atomic w.r.t. GC, as this scenario demonstrates: 1645 // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted. 1646 // VM thread changes sync state to synchronizing and suspends threads for GC. 1647 // Thread A is resumed to finish this native method, but doesn't block here since it 1648 // didn't see any synchronization is progress, and escapes. 1649 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native_trans); 1650 1651 if(os::is_MP()) { 1652 if (UseMembar) { 1653 // Force this write out before the read below 1654 __ membar(Assembler::Membar_mask_bits( 1655 Assembler::LoadLoad | Assembler::LoadStore | 1656 Assembler::StoreLoad | Assembler::StoreStore)); 1657 } else { 1658 // Write serialization page so VM thread can do a pseudo remote membar. 1659 // We use the current thread pointer to calculate a thread specific 1660 // offset to write to within the page. This minimizes bus traffic 1661 // due to cache line collision. 1662 __ serialize_memory(r15_thread, rcx); 1663 } 1664 } 1665 1666 1667 // check for safepoint operation in progress and/or pending suspend requests 1668 { 1669 Label Continue; 1670 1671 __ cmp32(ExternalAddress((address)SafepointSynchronize::address_of_state()), 1672 SafepointSynchronize::_not_synchronized); 1673 1674 Label L; 1675 __ jcc(Assembler::notEqual, L); 1676 __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0); 1677 __ jcc(Assembler::equal, Continue); 1678 __ bind(L); 1679 1680 // Don't use call_VM as it will see a possible pending exception and forward it 1681 // and never return here preventing us from clearing _last_native_pc down below. 1682 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are 1683 // preserved and correspond to the bcp/locals pointers. So we do a runtime call 1684 // by hand. 1685 // 1686 save_native_result(masm, ret_type, stack_slots); 1687 __ mov(c_rarg0, r15_thread); 1688 __ mov(r12, rsp); // remember sp 1689 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 1690 __ andptr(rsp, -16); // align stack as required by ABI 1691 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans))); 1692 __ mov(rsp, r12); // restore sp 1693 __ reinit_heapbase(); 1694 // Restore any method result value 1695 restore_native_result(masm, ret_type, stack_slots); 1696 __ bind(Continue); 1697 } 1698 1699 // change thread state 1700 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java); 1701 1702 Label reguard; 1703 Label reguard_done; 1704 __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled); 1705 __ jcc(Assembler::equal, reguard); 1706 __ bind(reguard_done); 1707 1708 // native result if any is live 1709 1710 // Unlock 1711 Label unlock_done; 1712 Label slow_path_unlock; 1713 if (method->is_synchronized()) { 1714 1715 // Get locked oop from the handle we passed to jni 1716 __ movptr(obj_reg, Address(oop_handle_reg, 0)); 1717 1718 Label done; 1719 1720 if (UseBiasedLocking) { 1721 __ biased_locking_exit(obj_reg, old_hdr, done); 1722 } 1723 1724 // Simple recursive lock? 1725 1726 __ cmpptr(Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size), (int32_t)NULL_WORD); 1727 __ jcc(Assembler::equal, done); 1728 1729 // Must save rax if if it is live now because cmpxchg must use it 1730 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) { 1731 save_native_result(masm, ret_type, stack_slots); 1732 } 1733 1734 1735 // get address of the stack lock 1736 __ lea(rax, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size)); 1737 // get old displaced header 1738 __ movptr(old_hdr, Address(rax, 0)); 1739 1740 // Atomic swap old header if oop still contains the stack lock 1741 if (os::is_MP()) { 1742 __ lock(); 1743 } 1744 __ cmpxchgptr(old_hdr, Address(obj_reg, 0)); 1745 __ jcc(Assembler::notEqual, slow_path_unlock); 1746 1747 // slow path re-enters here 1748 __ bind(unlock_done); 1749 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) { 1750 restore_native_result(masm, ret_type, stack_slots); 1751 } 1752 1753 __ bind(done); 1754 1755 } 1756 { 1757 SkipIfEqual skip(masm, &DTraceMethodProbes, false); 1758 save_native_result(masm, ret_type, stack_slots); 1759 __ movoop(c_rarg1, JNIHandles::make_local(method())); 1760 __ call_VM_leaf( 1761 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 1762 r15_thread, c_rarg1); 1763 restore_native_result(masm, ret_type, stack_slots); 1764 } 1765 1766 __ reset_last_Java_frame(false, true); 1767 1768 // Unpack oop result 1769 if (ret_type == T_OBJECT || ret_type == T_ARRAY) { 1770 Label L; 1771 __ testptr(rax, rax); 1772 __ jcc(Assembler::zero, L); 1773 __ movptr(rax, Address(rax, 0)); 1774 __ bind(L); 1775 __ verify_oop(rax); 1776 } 1777 1778 // reset handle block 1779 __ movptr(rcx, Address(r15_thread, JavaThread::active_handles_offset())); 1780 __ movptr(Address(rcx, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD); 1781 1782 // pop our frame 1783 1784 __ leave(); 1785 1786 // Any exception pending? 1787 __ cmpptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD); 1788 __ jcc(Assembler::notEqual, exception_pending); 1789 1790 // Return 1791 1792 __ ret(0); 1793 1794 // Unexpected paths are out of line and go here 1795 1796 // forward the exception 1797 __ bind(exception_pending); 1798 1799 // and forward the exception 1800 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 1801 1802 1803 // Slow path locking & unlocking 1804 if (method->is_synchronized()) { 1805 1806 // BEGIN Slow path lock 1807 __ bind(slow_path_lock); 1808 1809 // has last_Java_frame setup. No exceptions so do vanilla call not call_VM 1810 // args are (oop obj, BasicLock* lock, JavaThread* thread) 1811 1812 // protect the args we've loaded 1813 save_args(masm, total_c_args, c_arg, out_regs); 1814 1815 __ mov(c_rarg0, obj_reg); 1816 __ mov(c_rarg1, lock_reg); 1817 __ mov(c_rarg2, r15_thread); 1818 1819 // Not a leaf but we have last_Java_frame setup as we want 1820 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), 3); 1821 restore_args(masm, total_c_args, c_arg, out_regs); 1822 1823 #ifdef ASSERT 1824 { Label L; 1825 __ cmpptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD); 1826 __ jcc(Assembler::equal, L); 1827 __ stop("no pending exception allowed on exit from monitorenter"); 1828 __ bind(L); 1829 } 1830 #endif 1831 __ jmp(lock_done); 1832 1833 // END Slow path lock 1834 1835 // BEGIN Slow path unlock 1836 __ bind(slow_path_unlock); 1837 1838 // If we haven't already saved the native result we must save it now as xmm registers 1839 // are still exposed. 1840 1841 if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) { 1842 save_native_result(masm, ret_type, stack_slots); 1843 } 1844 1845 __ lea(c_rarg1, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size)); 1846 1847 __ mov(c_rarg0, obj_reg); 1848 __ mov(r12, rsp); // remember sp 1849 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 1850 __ andptr(rsp, -16); // align stack as required by ABI 1851 1852 // Save pending exception around call to VM (which contains an EXCEPTION_MARK) 1853 // NOTE that obj_reg == rbx currently 1854 __ movptr(rbx, Address(r15_thread, in_bytes(Thread::pending_exception_offset()))); 1855 __ movptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD); 1856 1857 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C))); 1858 __ mov(rsp, r12); // restore sp 1859 __ reinit_heapbase(); 1860 #ifdef ASSERT 1861 { 1862 Label L; 1863 __ cmpptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int)NULL_WORD); 1864 __ jcc(Assembler::equal, L); 1865 __ stop("no pending exception allowed on exit complete_monitor_unlocking_C"); 1866 __ bind(L); 1867 } 1868 #endif /* ASSERT */ 1869 1870 __ movptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), rbx); 1871 1872 if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) { 1873 restore_native_result(masm, ret_type, stack_slots); 1874 } 1875 __ jmp(unlock_done); 1876 1877 // END Slow path unlock 1878 1879 } // synchronized 1880 1881 // SLOW PATH Reguard the stack if needed 1882 1883 __ bind(reguard); 1884 save_native_result(masm, ret_type, stack_slots); 1885 __ mov(r12, rsp); // remember sp 1886 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 1887 __ andptr(rsp, -16); // align stack as required by ABI 1888 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); 1889 __ mov(rsp, r12); // restore sp 1890 __ reinit_heapbase(); 1891 restore_native_result(masm, ret_type, stack_slots); 1892 // and continue 1893 __ jmp(reguard_done); 1894 1895 1896 1897 __ flush(); 1898 1899 nmethod *nm = nmethod::new_native_nmethod(method, 1900 masm->code(), 1901 vep_offset, 1902 frame_complete, 1903 stack_slots / VMRegImpl::slots_per_word, 1904 (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)), 1905 in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size), 1906 oop_maps); 1907 return nm; 1908 1909 } 1910 1911 #ifdef HAVE_DTRACE_H 1912 // --------------------------------------------------------------------------- 1913 // Generate a dtrace nmethod for a given signature. The method takes arguments 1914 // in the Java compiled code convention, marshals them to the native 1915 // abi and then leaves nops at the position you would expect to call a native 1916 // function. When the probe is enabled the nops are replaced with a trap 1917 // instruction that dtrace inserts and the trace will cause a notification 1918 // to dtrace. 1919 // 1920 // The probes are only able to take primitive types and java/lang/String as 1921 // arguments. No other java types are allowed. Strings are converted to utf8 1922 // strings so that from dtrace point of view java strings are converted to C 1923 // strings. There is an arbitrary fixed limit on the total space that a method 1924 // can use for converting the strings. (256 chars per string in the signature). 1925 // So any java string larger then this is truncated. 1926 1927 static int fp_offset[ConcreteRegisterImpl::number_of_registers] = { 0 }; 1928 static bool offsets_initialized = false; 1929 1930 1931 nmethod *SharedRuntime::generate_dtrace_nmethod(MacroAssembler *masm, 1932 methodHandle method) { 1933 1934 1935 // generate_dtrace_nmethod is guarded by a mutex so we are sure to 1936 // be single threaded in this method. 1937 assert(AdapterHandlerLibrary_lock->owned_by_self(), "must be"); 1938 1939 if (!offsets_initialized) { 1940 fp_offset[c_rarg0->as_VMReg()->value()] = -1 * wordSize; 1941 fp_offset[c_rarg1->as_VMReg()->value()] = -2 * wordSize; 1942 fp_offset[c_rarg2->as_VMReg()->value()] = -3 * wordSize; 1943 fp_offset[c_rarg3->as_VMReg()->value()] = -4 * wordSize; 1944 fp_offset[c_rarg4->as_VMReg()->value()] = -5 * wordSize; 1945 fp_offset[c_rarg5->as_VMReg()->value()] = -6 * wordSize; 1946 1947 fp_offset[c_farg0->as_VMReg()->value()] = -7 * wordSize; 1948 fp_offset[c_farg1->as_VMReg()->value()] = -8 * wordSize; 1949 fp_offset[c_farg2->as_VMReg()->value()] = -9 * wordSize; 1950 fp_offset[c_farg3->as_VMReg()->value()] = -10 * wordSize; 1951 fp_offset[c_farg4->as_VMReg()->value()] = -11 * wordSize; 1952 fp_offset[c_farg5->as_VMReg()->value()] = -12 * wordSize; 1953 fp_offset[c_farg6->as_VMReg()->value()] = -13 * wordSize; 1954 fp_offset[c_farg7->as_VMReg()->value()] = -14 * wordSize; 1955 1956 offsets_initialized = true; 1957 } 1958 // Fill in the signature array, for the calling-convention call. 1959 int total_args_passed = method->size_of_parameters(); 1960 1961 BasicType* in_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed); 1962 VMRegPair *in_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed); 1963 1964 // The signature we are going to use for the trap that dtrace will see 1965 // java/lang/String is converted. We drop "this" and any other object 1966 // is converted to NULL. (A one-slot java/lang/Long object reference 1967 // is converted to a two-slot long, which is why we double the allocation). 1968 BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed * 2); 1969 VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed * 2); 1970 1971 int i=0; 1972 int total_strings = 0; 1973 int first_arg_to_pass = 0; 1974 int total_c_args = 0; 1975 1976 // Skip the receiver as dtrace doesn't want to see it 1977 if( !method->is_static() ) { 1978 in_sig_bt[i++] = T_OBJECT; 1979 first_arg_to_pass = 1; 1980 } 1981 1982 // We need to convert the java args to where a native (non-jni) function 1983 // would expect them. To figure out where they go we convert the java 1984 // signature to a C signature. 1985 1986 SignatureStream ss(method->signature()); 1987 for ( ; !ss.at_return_type(); ss.next()) { 1988 BasicType bt = ss.type(); 1989 in_sig_bt[i++] = bt; // Collect remaining bits of signature 1990 out_sig_bt[total_c_args++] = bt; 1991 if( bt == T_OBJECT) { 1992 symbolOop s = ss.as_symbol_or_null(); 1993 if (s == vmSymbols::java_lang_String()) { 1994 total_strings++; 1995 out_sig_bt[total_c_args-1] = T_ADDRESS; 1996 } else if (s == vmSymbols::java_lang_Boolean() || 1997 s == vmSymbols::java_lang_Character() || 1998 s == vmSymbols::java_lang_Byte() || 1999 s == vmSymbols::java_lang_Short() || 2000 s == vmSymbols::java_lang_Integer() || 2001 s == vmSymbols::java_lang_Float()) { 2002 out_sig_bt[total_c_args-1] = T_INT; 2003 } else if (s == vmSymbols::java_lang_Long() || 2004 s == vmSymbols::java_lang_Double()) { 2005 out_sig_bt[total_c_args-1] = T_LONG; 2006 out_sig_bt[total_c_args++] = T_VOID; 2007 } 2008 } else if ( bt == T_LONG || bt == T_DOUBLE ) { 2009 in_sig_bt[i++] = T_VOID; // Longs & doubles take 2 Java slots 2010 // We convert double to long 2011 out_sig_bt[total_c_args-1] = T_LONG; 2012 out_sig_bt[total_c_args++] = T_VOID; 2013 } else if ( bt == T_FLOAT) { 2014 // We convert float to int 2015 out_sig_bt[total_c_args-1] = T_INT; 2016 } 2017 } 2018 2019 assert(i==total_args_passed, "validly parsed signature"); 2020 2021 // Now get the compiled-Java layout as input arguments 2022 int comp_args_on_stack; 2023 comp_args_on_stack = SharedRuntime::java_calling_convention( 2024 in_sig_bt, in_regs, total_args_passed, false); 2025 2026 // Now figure out where the args must be stored and how much stack space 2027 // they require (neglecting out_preserve_stack_slots but space for storing 2028 // the 1st six register arguments). It's weird see int_stk_helper. 2029 2030 int out_arg_slots; 2031 out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args); 2032 2033 // Calculate the total number of stack slots we will need. 2034 2035 // First count the abi requirement plus all of the outgoing args 2036 int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; 2037 2038 // Now space for the string(s) we must convert 2039 int* string_locs = NEW_RESOURCE_ARRAY(int, total_strings + 1); 2040 for (i = 0; i < total_strings ; i++) { 2041 string_locs[i] = stack_slots; 2042 stack_slots += max_dtrace_string_size / VMRegImpl::stack_slot_size; 2043 } 2044 2045 // Plus the temps we might need to juggle register args 2046 // regs take two slots each 2047 stack_slots += (Argument::n_int_register_parameters_c + 2048 Argument::n_float_register_parameters_c) * 2; 2049 2050 2051 // + 4 for return address (which we own) and saved rbp, 2052 2053 stack_slots += 4; 2054 2055 // Ok The space we have allocated will look like: 2056 // 2057 // 2058 // FP-> | | 2059 // |---------------------| 2060 // | string[n] | 2061 // |---------------------| <- string_locs[n] 2062 // | string[n-1] | 2063 // |---------------------| <- string_locs[n-1] 2064 // | ... | 2065 // | ... | 2066 // |---------------------| <- string_locs[1] 2067 // | string[0] | 2068 // |---------------------| <- string_locs[0] 2069 // | outbound memory | 2070 // | based arguments | 2071 // | | 2072 // |---------------------| 2073 // | | 2074 // SP-> | out_preserved_slots | 2075 // 2076 // 2077 2078 // Now compute actual number of stack words we need rounding to make 2079 // stack properly aligned. 2080 stack_slots = round_to(stack_slots, 4 * VMRegImpl::slots_per_word); 2081 2082 int stack_size = stack_slots * VMRegImpl::stack_slot_size; 2083 2084 intptr_t start = (intptr_t)__ pc(); 2085 2086 // First thing make an ic check to see if we should even be here 2087 2088 // We are free to use all registers as temps without saving them and 2089 // restoring them except rbp. rbp, is the only callee save register 2090 // as far as the interpreter and the compiler(s) are concerned. 2091 2092 const Register ic_reg = rax; 2093 const Register receiver = rcx; 2094 Label hit; 2095 Label exception_pending; 2096 2097 2098 __ verify_oop(receiver); 2099 __ cmpl(ic_reg, Address(receiver, oopDesc::klass_offset_in_bytes())); 2100 __ jcc(Assembler::equal, hit); 2101 2102 __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); 2103 2104 // verified entry must be aligned for code patching. 2105 // and the first 5 bytes must be in the same cache line 2106 // if we align at 8 then we will be sure 5 bytes are in the same line 2107 __ align(8); 2108 2109 __ bind(hit); 2110 2111 int vep_offset = ((intptr_t)__ pc()) - start; 2112 2113 2114 // The instruction at the verified entry point must be 5 bytes or longer 2115 // because it can be patched on the fly by make_non_entrant. The stack bang 2116 // instruction fits that requirement. 2117 2118 // Generate stack overflow check 2119 2120 if (UseStackBanging) { 2121 if (stack_size <= StackShadowPages*os::vm_page_size()) { 2122 __ bang_stack_with_offset(StackShadowPages*os::vm_page_size()); 2123 } else { 2124 __ movl(rax, stack_size); 2125 __ bang_stack_size(rax, rbx); 2126 } 2127 } else { 2128 // need a 5 byte instruction to allow MT safe patching to non-entrant 2129 __ fat_nop(); 2130 } 2131 2132 assert(((uintptr_t)__ pc() - start - vep_offset) >= 5, 2133 "valid size for make_non_entrant"); 2134 2135 // Generate a new frame for the wrapper. 2136 __ enter(); 2137 2138 // -4 because return address is already present and so is saved rbp, 2139 if (stack_size - 2*wordSize != 0) { 2140 __ subq(rsp, stack_size - 2*wordSize); 2141 } 2142 2143 // Frame is now completed as far a size and linkage. 2144 2145 int frame_complete = ((intptr_t)__ pc()) - start; 2146 2147 int c_arg, j_arg; 2148 2149 // State of input register args 2150 2151 bool live[ConcreteRegisterImpl::number_of_registers]; 2152 2153 live[j_rarg0->as_VMReg()->value()] = false; 2154 live[j_rarg1->as_VMReg()->value()] = false; 2155 live[j_rarg2->as_VMReg()->value()] = false; 2156 live[j_rarg3->as_VMReg()->value()] = false; 2157 live[j_rarg4->as_VMReg()->value()] = false; 2158 live[j_rarg5->as_VMReg()->value()] = false; 2159 2160 live[j_farg0->as_VMReg()->value()] = false; 2161 live[j_farg1->as_VMReg()->value()] = false; 2162 live[j_farg2->as_VMReg()->value()] = false; 2163 live[j_farg3->as_VMReg()->value()] = false; 2164 live[j_farg4->as_VMReg()->value()] = false; 2165 live[j_farg5->as_VMReg()->value()] = false; 2166 live[j_farg6->as_VMReg()->value()] = false; 2167 live[j_farg7->as_VMReg()->value()] = false; 2168 2169 2170 bool rax_is_zero = false; 2171 2172 // All args (except strings) destined for the stack are moved first 2173 for (j_arg = first_arg_to_pass, c_arg = 0 ; 2174 j_arg < total_args_passed ; j_arg++, c_arg++ ) { 2175 VMRegPair src = in_regs[j_arg]; 2176 VMRegPair dst = out_regs[c_arg]; 2177 2178 // Get the real reg value or a dummy (rsp) 2179 2180 int src_reg = src.first()->is_reg() ? 2181 src.first()->value() : 2182 rsp->as_VMReg()->value(); 2183 2184 bool useless = in_sig_bt[j_arg] == T_ARRAY || 2185 (in_sig_bt[j_arg] == T_OBJECT && 2186 out_sig_bt[c_arg] != T_INT && 2187 out_sig_bt[c_arg] != T_ADDRESS && 2188 out_sig_bt[c_arg] != T_LONG); 2189 2190 live[src_reg] = !useless; 2191 2192 if (dst.first()->is_stack()) { 2193 2194 // Even though a string arg in a register is still live after this loop 2195 // after the string conversion loop (next) it will be dead so we take 2196 // advantage of that now for simpler code to manage live. 2197 2198 live[src_reg] = false; 2199 switch (in_sig_bt[j_arg]) { 2200 2201 case T_ARRAY: 2202 case T_OBJECT: 2203 { 2204 Address stack_dst(rsp, reg2offset_out(dst.first())); 2205 2206 if (out_sig_bt[c_arg] == T_INT || out_sig_bt[c_arg] == T_LONG) { 2207 // need to unbox a one-word value 2208 Register in_reg = rax; 2209 if ( src.first()->is_reg() ) { 2210 in_reg = src.first()->as_Register(); 2211 } else { 2212 __ movq(rax, Address(rbp, reg2offset_in(src.first()))); 2213 rax_is_zero = false; 2214 } 2215 Label skipUnbox; 2216 __ movptr(Address(rsp, reg2offset_out(dst.first())), 2217 (int32_t)NULL_WORD); 2218 __ testq(in_reg, in_reg); 2219 __ jcc(Assembler::zero, skipUnbox); 2220 2221 BasicType bt = out_sig_bt[c_arg]; 2222 int box_offset = java_lang_boxing_object::value_offset_in_bytes(bt); 2223 Address src1(in_reg, box_offset); 2224 if ( bt == T_LONG ) { 2225 __ movq(in_reg, src1); 2226 __ movq(stack_dst, in_reg); 2227 assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); 2228 ++c_arg; // skip over T_VOID to keep the loop indices in sync 2229 } else { 2230 __ movl(in_reg, src1); 2231 __ movl(stack_dst, in_reg); 2232 } 2233 2234 __ bind(skipUnbox); 2235 } else if (out_sig_bt[c_arg] != T_ADDRESS) { 2236 // Convert the arg to NULL 2237 if (!rax_is_zero) { 2238 __ xorq(rax, rax); 2239 rax_is_zero = true; 2240 } 2241 __ movq(stack_dst, rax); 2242 } 2243 } 2244 break; 2245 2246 case T_VOID: 2247 break; 2248 2249 case T_FLOAT: 2250 // This does the right thing since we know it is destined for the 2251 // stack 2252 float_move(masm, src, dst); 2253 break; 2254 2255 case T_DOUBLE: 2256 // This does the right thing since we know it is destined for the 2257 // stack 2258 double_move(masm, src, dst); 2259 break; 2260 2261 case T_LONG : 2262 long_move(masm, src, dst); 2263 break; 2264 2265 case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); 2266 2267 default: 2268 move32_64(masm, src, dst); 2269 } 2270 } 2271 2272 } 2273 2274 // If we have any strings we must store any register based arg to the stack 2275 // This includes any still live xmm registers too. 2276 2277 int sid = 0; 2278 2279 if (total_strings > 0 ) { 2280 for (j_arg = first_arg_to_pass, c_arg = 0 ; 2281 j_arg < total_args_passed ; j_arg++, c_arg++ ) { 2282 VMRegPair src = in_regs[j_arg]; 2283 VMRegPair dst = out_regs[c_arg]; 2284 2285 if (src.first()->is_reg()) { 2286 Address src_tmp(rbp, fp_offset[src.first()->value()]); 2287 2288 // string oops were left untouched by the previous loop even if the 2289 // eventual (converted) arg is destined for the stack so park them 2290 // away now (except for first) 2291 2292 if (out_sig_bt[c_arg] == T_ADDRESS) { 2293 Address utf8_addr = Address( 2294 rsp, string_locs[sid++] * VMRegImpl::stack_slot_size); 2295 if (sid != 1) { 2296 // The first string arg won't be killed until after the utf8 2297 // conversion 2298 __ movq(utf8_addr, src.first()->as_Register()); 2299 } 2300 } else if (dst.first()->is_reg()) { 2301 if (in_sig_bt[j_arg] == T_FLOAT || in_sig_bt[j_arg] == T_DOUBLE) { 2302 2303 // Convert the xmm register to an int and store it in the reserved 2304 // location for the eventual c register arg 2305 XMMRegister f = src.first()->as_XMMRegister(); 2306 if (in_sig_bt[j_arg] == T_FLOAT) { 2307 __ movflt(src_tmp, f); 2308 } else { 2309 __ movdbl(src_tmp, f); 2310 } 2311 } else { 2312 // If the arg is an oop type we don't support don't bother to store 2313 // it remember string was handled above. 2314 bool useless = in_sig_bt[j_arg] == T_ARRAY || 2315 (in_sig_bt[j_arg] == T_OBJECT && 2316 out_sig_bt[c_arg] != T_INT && 2317 out_sig_bt[c_arg] != T_LONG); 2318 2319 if (!useless) { 2320 __ movq(src_tmp, src.first()->as_Register()); 2321 } 2322 } 2323 } 2324 } 2325 if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) { 2326 assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); 2327 ++c_arg; // skip over T_VOID to keep the loop indices in sync 2328 } 2329 } 2330 2331 // Now that the volatile registers are safe, convert all the strings 2332 sid = 0; 2333 2334 for (j_arg = first_arg_to_pass, c_arg = 0 ; 2335 j_arg < total_args_passed ; j_arg++, c_arg++ ) { 2336 if (out_sig_bt[c_arg] == T_ADDRESS) { 2337 // It's a string 2338 Address utf8_addr = Address( 2339 rsp, string_locs[sid++] * VMRegImpl::stack_slot_size); 2340 // The first string we find might still be in the original java arg 2341 // register 2342 2343 VMReg src = in_regs[j_arg].first(); 2344 2345 // We will need to eventually save the final argument to the trap 2346 // in the von-volatile location dedicated to src. This is the offset 2347 // from fp we will use. 2348 int src_off = src->is_reg() ? 2349 fp_offset[src->value()] : reg2offset_in(src); 2350 2351 // This is where the argument will eventually reside 2352 VMRegPair dst = out_regs[c_arg]; 2353 2354 if (src->is_reg()) { 2355 if (sid == 1) { 2356 __ movq(c_rarg0, src->as_Register()); 2357 } else { 2358 __ movq(c_rarg0, utf8_addr); 2359 } 2360 } else { 2361 // arg is still in the original location 2362 __ movq(c_rarg0, Address(rbp, reg2offset_in(src))); 2363 } 2364 Label done, convert; 2365 2366 // see if the oop is NULL 2367 __ testq(c_rarg0, c_rarg0); 2368 __ jcc(Assembler::notEqual, convert); 2369 2370 if (dst.first()->is_reg()) { 2371 // Save the ptr to utf string in the origina src loc or the tmp 2372 // dedicated to it 2373 __ movq(Address(rbp, src_off), c_rarg0); 2374 } else { 2375 __ movq(Address(rsp, reg2offset_out(dst.first())), c_rarg0); 2376 } 2377 __ jmp(done); 2378 2379 __ bind(convert); 2380 2381 __ lea(c_rarg1, utf8_addr); 2382 if (dst.first()->is_reg()) { 2383 __ movq(Address(rbp, src_off), c_rarg1); 2384 } else { 2385 __ movq(Address(rsp, reg2offset_out(dst.first())), c_rarg1); 2386 } 2387 // And do the conversion 2388 __ call(RuntimeAddress( 2389 CAST_FROM_FN_PTR(address, SharedRuntime::get_utf))); 2390 2391 __ bind(done); 2392 } 2393 if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) { 2394 assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); 2395 ++c_arg; // skip over T_VOID to keep the loop indices in sync 2396 } 2397 } 2398 // The get_utf call killed all the c_arg registers 2399 live[c_rarg0->as_VMReg()->value()] = false; 2400 live[c_rarg1->as_VMReg()->value()] = false; 2401 live[c_rarg2->as_VMReg()->value()] = false; 2402 live[c_rarg3->as_VMReg()->value()] = false; 2403 live[c_rarg4->as_VMReg()->value()] = false; 2404 live[c_rarg5->as_VMReg()->value()] = false; 2405 2406 live[c_farg0->as_VMReg()->value()] = false; 2407 live[c_farg1->as_VMReg()->value()] = false; 2408 live[c_farg2->as_VMReg()->value()] = false; 2409 live[c_farg3->as_VMReg()->value()] = false; 2410 live[c_farg4->as_VMReg()->value()] = false; 2411 live[c_farg5->as_VMReg()->value()] = false; 2412 live[c_farg6->as_VMReg()->value()] = false; 2413 live[c_farg7->as_VMReg()->value()] = false; 2414 } 2415 2416 // Now we can finally move the register args to their desired locations 2417 2418 rax_is_zero = false; 2419 2420 for (j_arg = first_arg_to_pass, c_arg = 0 ; 2421 j_arg < total_args_passed ; j_arg++, c_arg++ ) { 2422 2423 VMRegPair src = in_regs[j_arg]; 2424 VMRegPair dst = out_regs[c_arg]; 2425 2426 // Only need to look for args destined for the interger registers (since we 2427 // convert float/double args to look like int/long outbound) 2428 if (dst.first()->is_reg()) { 2429 Register r = dst.first()->as_Register(); 2430 2431 // Check if the java arg is unsupported and thereofre useless 2432 bool useless = in_sig_bt[j_arg] == T_ARRAY || 2433 (in_sig_bt[j_arg] == T_OBJECT && 2434 out_sig_bt[c_arg] != T_INT && 2435 out_sig_bt[c_arg] != T_ADDRESS && 2436 out_sig_bt[c_arg] != T_LONG); 2437 2438 2439 // If we're going to kill an existing arg save it first 2440 if (live[dst.first()->value()]) { 2441 // you can't kill yourself 2442 if (src.first() != dst.first()) { 2443 __ movq(Address(rbp, fp_offset[dst.first()->value()]), r); 2444 } 2445 } 2446 if (src.first()->is_reg()) { 2447 if (live[src.first()->value()] ) { 2448 if (in_sig_bt[j_arg] == T_FLOAT) { 2449 __ movdl(r, src.first()->as_XMMRegister()); 2450 } else if (in_sig_bt[j_arg] == T_DOUBLE) { 2451 __ movdq(r, src.first()->as_XMMRegister()); 2452 } else if (r != src.first()->as_Register()) { 2453 if (!useless) { 2454 __ movq(r, src.first()->as_Register()); 2455 } 2456 } 2457 } else { 2458 // If the arg is an oop type we don't support don't bother to store 2459 // it 2460 if (!useless) { 2461 if (in_sig_bt[j_arg] == T_DOUBLE || 2462 in_sig_bt[j_arg] == T_LONG || 2463 in_sig_bt[j_arg] == T_OBJECT ) { 2464 __ movq(r, Address(rbp, fp_offset[src.first()->value()])); 2465 } else { 2466 __ movl(r, Address(rbp, fp_offset[src.first()->value()])); 2467 } 2468 } 2469 } 2470 live[src.first()->value()] = false; 2471 } else if (!useless) { 2472 // full sized move even for int should be ok 2473 __ movq(r, Address(rbp, reg2offset_in(src.first()))); 2474 } 2475 2476 // At this point r has the original java arg in the final location 2477 // (assuming it wasn't useless). If the java arg was an oop 2478 // we have a bit more to do 2479 2480 if (in_sig_bt[j_arg] == T_ARRAY || in_sig_bt[j_arg] == T_OBJECT ) { 2481 if (out_sig_bt[c_arg] == T_INT || out_sig_bt[c_arg] == T_LONG) { 2482 // need to unbox a one-word value 2483 Label skip; 2484 __ testq(r, r); 2485 __ jcc(Assembler::equal, skip); 2486 BasicType bt = out_sig_bt[c_arg]; 2487 int box_offset = java_lang_boxing_object::value_offset_in_bytes(bt); 2488 Address src1(r, box_offset); 2489 if ( bt == T_LONG ) { 2490 __ movq(r, src1); 2491 } else { 2492 __ movl(r, src1); 2493 } 2494 __ bind(skip); 2495 2496 } else if (out_sig_bt[c_arg] != T_ADDRESS) { 2497 // Convert the arg to NULL 2498 __ xorq(r, r); 2499 } 2500 } 2501 2502 // dst can longer be holding an input value 2503 live[dst.first()->value()] = false; 2504 } 2505 if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) { 2506 assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); 2507 ++c_arg; // skip over T_VOID to keep the loop indices in sync 2508 } 2509 } 2510 2511 2512 // Ok now we are done. Need to place the nop that dtrace wants in order to 2513 // patch in the trap 2514 int patch_offset = ((intptr_t)__ pc()) - start; 2515 2516 __ nop(); 2517 2518 2519 // Return 2520 2521 __ leave(); 2522 __ ret(0); 2523 2524 __ flush(); 2525 2526 nmethod *nm = nmethod::new_dtrace_nmethod( 2527 method, masm->code(), vep_offset, patch_offset, frame_complete, 2528 stack_slots / VMRegImpl::slots_per_word); 2529 return nm; 2530 2531 } 2532 2533 #endif // HAVE_DTRACE_H 2534 2535 // this function returns the adjust size (in number of words) to a c2i adapter 2536 // activation for use during deoptimization 2537 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals ) { 2538 return (callee_locals - callee_parameters) * Interpreter::stackElementWords(); 2539 } 2540 2541 2542 uint SharedRuntime::out_preserve_stack_slots() { 2543 return 0; 2544 } 2545 2546 2547 //------------------------------generate_deopt_blob---------------------------- 2548 void SharedRuntime::generate_deopt_blob() { 2549 // Allocate space for the code 2550 ResourceMark rm; 2551 // Setup code generation tools 2552 CodeBuffer buffer("deopt_blob", 2048, 1024); 2553 MacroAssembler* masm = new MacroAssembler(&buffer); 2554 int frame_size_in_words; 2555 OopMap* map = NULL; 2556 OopMapSet *oop_maps = new OopMapSet(); 2557 2558 // ------------- 2559 // This code enters when returning to a de-optimized nmethod. A return 2560 // address has been pushed on the the stack, and return values are in 2561 // registers. 2562 // If we are doing a normal deopt then we were called from the patched 2563 // nmethod from the point we returned to the nmethod. So the return 2564 // address on the stack is wrong by NativeCall::instruction_size 2565 // We will adjust the value so it looks like we have the original return 2566 // address on the stack (like when we eagerly deoptimized). 2567 // In the case of an exception pending when deoptimizing, we enter 2568 // with a return address on the stack that points after the call we patched 2569 // into the exception handler. We have the following register state from, 2570 // e.g., the forward exception stub (see stubGenerator_x86_64.cpp). 2571 // rax: exception oop 2572 // rbx: exception handler 2573 // rdx: throwing pc 2574 // So in this case we simply jam rdx into the useless return address and 2575 // the stack looks just like we want. 2576 // 2577 // At this point we need to de-opt. We save the argument return 2578 // registers. We call the first C routine, fetch_unroll_info(). This 2579 // routine captures the return values and returns a structure which 2580 // describes the current frame size and the sizes of all replacement frames. 2581 // The current frame is compiled code and may contain many inlined 2582 // functions, each with their own JVM state. We pop the current frame, then 2583 // push all the new frames. Then we call the C routine unpack_frames() to 2584 // populate these frames. Finally unpack_frames() returns us the new target 2585 // address. Notice that callee-save registers are BLOWN here; they have 2586 // already been captured in the vframeArray at the time the return PC was 2587 // patched. 2588 address start = __ pc(); 2589 Label cont; 2590 2591 // Prolog for non exception case! 2592 2593 // Save everything in sight. 2594 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words); 2595 2596 // Normal deoptimization. Save exec mode for unpack_frames. 2597 __ movl(r14, Deoptimization::Unpack_deopt); // callee-saved 2598 __ jmp(cont); 2599 2600 int reexecute_offset = __ pc() - start; 2601 2602 // Reexecute case 2603 // return address is the pc describes what bci to do re-execute at 2604 2605 // No need to update map as each call to save_live_registers will produce identical oopmap 2606 (void) RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words); 2607 2608 __ movl(r14, Deoptimization::Unpack_reexecute); // callee-saved 2609 __ jmp(cont); 2610 2611 int exception_offset = __ pc() - start; 2612 2613 // Prolog for exception case 2614 2615 // all registers are dead at this entry point, except for rax, and 2616 // rdx which contain the exception oop and exception pc 2617 // respectively. Set them in TLS and fall thru to the 2618 // unpack_with_exception_in_tls entry point. 2619 2620 __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), rdx); 2621 __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), rax); 2622 2623 int exception_in_tls_offset = __ pc() - start; 2624 2625 // new implementation because exception oop is now passed in JavaThread 2626 2627 // Prolog for exception case 2628 // All registers must be preserved because they might be used by LinearScan 2629 // Exceptiop oop and throwing PC are passed in JavaThread 2630 // tos: stack at point of call to method that threw the exception (i.e. only 2631 // args are on the stack, no return address) 2632 2633 // make room on stack for the return address 2634 // It will be patched later with the throwing pc. The correct value is not 2635 // available now because loading it from memory would destroy registers. 2636 __ push(0); 2637 2638 // Save everything in sight. 2639 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words); 2640 2641 // Now it is safe to overwrite any register 2642 2643 // Deopt during an exception. Save exec mode for unpack_frames. 2644 __ movl(r14, Deoptimization::Unpack_exception); // callee-saved 2645 2646 // load throwing pc from JavaThread and patch it as the return address 2647 // of the current frame. Then clear the field in JavaThread 2648 2649 __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset())); 2650 __ movptr(Address(rbp, wordSize), rdx); 2651 __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD); 2652 2653 #ifdef ASSERT 2654 // verify that there is really an exception oop in JavaThread 2655 __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset())); 2656 __ verify_oop(rax); 2657 2658 // verify that there is no pending exception 2659 Label no_pending_exception; 2660 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset())); 2661 __ testptr(rax, rax); 2662 __ jcc(Assembler::zero, no_pending_exception); 2663 __ stop("must not have pending exception here"); 2664 __ bind(no_pending_exception); 2665 #endif 2666 2667 __ bind(cont); 2668 2669 // Call C code. Need thread and this frame, but NOT official VM entry 2670 // crud. We cannot block on this call, no GC can happen. 2671 // 2672 // UnrollBlock* fetch_unroll_info(JavaThread* thread) 2673 2674 // fetch_unroll_info needs to call last_java_frame(). 2675 2676 __ set_last_Java_frame(noreg, noreg, NULL); 2677 #ifdef ASSERT 2678 { Label L; 2679 __ cmpptr(Address(r15_thread, 2680 JavaThread::last_Java_fp_offset()), 2681 (int32_t)0); 2682 __ jcc(Assembler::equal, L); 2683 __ stop("SharedRuntime::generate_deopt_blob: last_Java_fp not cleared"); 2684 __ bind(L); 2685 } 2686 #endif // ASSERT 2687 __ mov(c_rarg0, r15_thread); 2688 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info))); 2689 2690 // Need to have an oopmap that tells fetch_unroll_info where to 2691 // find any register it might need. 2692 oop_maps->add_gc_map(__ pc() - start, map); 2693 2694 __ reset_last_Java_frame(false, false); 2695 2696 // Load UnrollBlock* into rdi 2697 __ mov(rdi, rax); 2698 2699 Label noException; 2700 __ cmpl(r14, Deoptimization::Unpack_exception); // Was exception pending? 2701 __ jcc(Assembler::notEqual, noException); 2702 __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset())); 2703 // QQQ this is useless it was NULL above 2704 __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset())); 2705 __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), (int32_t)NULL_WORD); 2706 __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD); 2707 2708 __ verify_oop(rax); 2709 2710 // Overwrite the result registers with the exception results. 2711 __ movptr(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax); 2712 // I think this is useless 2713 __ movptr(Address(rsp, RegisterSaver::rdx_offset_in_bytes()), rdx); 2714 2715 __ bind(noException); 2716 2717 // Only register save data is on the stack. 2718 // Now restore the result registers. Everything else is either dead 2719 // or captured in the vframeArray. 2720 RegisterSaver::restore_result_registers(masm); 2721 2722 // All of the register save area has been popped of the stack. Only the 2723 // return address remains. 2724 2725 // Pop all the frames we must move/replace. 2726 // 2727 // Frame picture (youngest to oldest) 2728 // 1: self-frame (no frame link) 2729 // 2: deopting frame (no frame link) 2730 // 3: caller of deopting frame (could be compiled/interpreted). 2731 // 2732 // Note: by leaving the return address of self-frame on the stack 2733 // and using the size of frame 2 to adjust the stack 2734 // when we are done the return to frame 3 will still be on the stack. 2735 2736 // Pop deoptimized frame 2737 __ movl(rcx, Address(rdi, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes())); 2738 __ addptr(rsp, rcx); 2739 2740 // rsp should be pointing at the return address to the caller (3) 2741 2742 // Stack bang to make sure there's enough room for these interpreter frames. 2743 if (UseStackBanging) { 2744 __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes())); 2745 __ bang_stack_size(rbx, rcx); 2746 } 2747 2748 // Load address of array of frame pcs into rcx 2749 __ movptr(rcx, Address(rdi, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes())); 2750 2751 // Trash the old pc 2752 __ addptr(rsp, wordSize); 2753 2754 // Load address of array of frame sizes into rsi 2755 __ movptr(rsi, Address(rdi, Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes())); 2756 2757 // Load counter into rdx 2758 __ movl(rdx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes())); 2759 2760 // Pick up the initial fp we should save 2761 __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_fp_offset_in_bytes())); 2762 2763 // Now adjust the caller's stack to make up for the extra locals 2764 // but record the original sp so that we can save it in the skeletal interpreter 2765 // frame and the stack walking of interpreter_sender will get the unextended sp 2766 // value and not the "real" sp value. 2767 2768 const Register sender_sp = r8; 2769 2770 __ mov(sender_sp, rsp); 2771 __ movl(rbx, Address(rdi, 2772 Deoptimization::UnrollBlock:: 2773 caller_adjustment_offset_in_bytes())); 2774 __ subptr(rsp, rbx); 2775 2776 // Push interpreter frames in a loop 2777 Label loop; 2778 __ bind(loop); 2779 __ movptr(rbx, Address(rsi, 0)); // Load frame size 2780 #ifdef CC_INTERP 2781 __ subptr(rbx, 4*wordSize); // we'll push pc and ebp by hand and 2782 #ifdef ASSERT 2783 __ push(0xDEADDEAD); // Make a recognizable pattern 2784 __ push(0xDEADDEAD); 2785 #else /* ASSERT */ 2786 __ subptr(rsp, 2*wordSize); // skip the "static long no_param" 2787 #endif /* ASSERT */ 2788 #else 2789 __ subptr(rbx, 2*wordSize); // We'll push pc and ebp by hand 2790 #endif // CC_INTERP 2791 __ pushptr(Address(rcx, 0)); // Save return address 2792 __ enter(); // Save old & set new ebp 2793 __ subptr(rsp, rbx); // Prolog 2794 #ifdef CC_INTERP 2795 __ movptr(Address(rbp, 2796 -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))), 2797 sender_sp); // Make it walkable 2798 #else /* CC_INTERP */ 2799 // This value is corrected by layout_activation_impl 2800 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD ); 2801 __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), sender_sp); // Make it walkable 2802 #endif /* CC_INTERP */ 2803 __ mov(sender_sp, rsp); // Pass sender_sp to next frame 2804 __ addptr(rsi, wordSize); // Bump array pointer (sizes) 2805 __ addptr(rcx, wordSize); // Bump array pointer (pcs) 2806 __ decrementl(rdx); // Decrement counter 2807 __ jcc(Assembler::notZero, loop); 2808 __ pushptr(Address(rcx, 0)); // Save final return address 2809 2810 // Re-push self-frame 2811 __ enter(); // Save old & set new ebp 2812 2813 // Allocate a full sized register save area. 2814 // Return address and rbp are in place, so we allocate two less words. 2815 __ subptr(rsp, (frame_size_in_words - 2) * wordSize); 2816 2817 // Restore frame locals after moving the frame 2818 __ movdbl(Address(rsp, RegisterSaver::xmm0_offset_in_bytes()), xmm0); 2819 __ movptr(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax); 2820 2821 // Call C code. Need thread but NOT official VM entry 2822 // crud. We cannot block on this call, no GC can happen. Call should 2823 // restore return values to their stack-slots with the new SP. 2824 // 2825 // void Deoptimization::unpack_frames(JavaThread* thread, int exec_mode) 2826 2827 // Use rbp because the frames look interpreted now 2828 __ set_last_Java_frame(noreg, rbp, NULL); 2829 2830 __ mov(c_rarg0, r15_thread); 2831 __ movl(c_rarg1, r14); // second arg: exec_mode 2832 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames))); 2833 2834 // Set an oopmap for the call site 2835 oop_maps->add_gc_map(__ pc() - start, 2836 new OopMap( frame_size_in_words, 0 )); 2837 2838 __ reset_last_Java_frame(true, false); 2839 2840 // Collect return values 2841 __ movdbl(xmm0, Address(rsp, RegisterSaver::xmm0_offset_in_bytes())); 2842 __ movptr(rax, Address(rsp, RegisterSaver::rax_offset_in_bytes())); 2843 // I think this is useless (throwing pc?) 2844 __ movptr(rdx, Address(rsp, RegisterSaver::rdx_offset_in_bytes())); 2845 2846 // Pop self-frame. 2847 __ leave(); // Epilog 2848 2849 // Jump to interpreter 2850 __ ret(0); 2851 2852 // Make sure all code is generated 2853 masm->flush(); 2854 2855 _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words); 2856 _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset); 2857 } 2858 2859 #ifdef COMPILER2 2860 //------------------------------generate_uncommon_trap_blob-------------------- 2861 void SharedRuntime::generate_uncommon_trap_blob() { 2862 // Allocate space for the code 2863 ResourceMark rm; 2864 // Setup code generation tools 2865 CodeBuffer buffer("uncommon_trap_blob", 2048, 1024); 2866 MacroAssembler* masm = new MacroAssembler(&buffer); 2867 2868 assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned"); 2869 2870 address start = __ pc(); 2871 2872 // Push self-frame. We get here with a return address on the 2873 // stack, so rsp is 8-byte aligned until we allocate our frame. 2874 __ subptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Epilog! 2875 2876 // No callee saved registers. rbp is assumed implicitly saved 2877 __ movptr(Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt), rbp); 2878 2879 // compiler left unloaded_class_index in j_rarg0 move to where the 2880 // runtime expects it. 2881 __ movl(c_rarg1, j_rarg0); 2882 2883 __ set_last_Java_frame(noreg, noreg, NULL); 2884 2885 // Call C code. Need thread but NOT official VM entry 2886 // crud. We cannot block on this call, no GC can happen. Call should 2887 // capture callee-saved registers as well as return values. 2888 // Thread is in rdi already. 2889 // 2890 // UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index); 2891 2892 __ mov(c_rarg0, r15_thread); 2893 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap))); 2894 2895 // Set an oopmap for the call site 2896 OopMapSet* oop_maps = new OopMapSet(); 2897 OopMap* map = new OopMap(SimpleRuntimeFrame::framesize, 0); 2898 2899 // location of rbp is known implicitly by the frame sender code 2900 2901 oop_maps->add_gc_map(__ pc() - start, map); 2902 2903 __ reset_last_Java_frame(false, false); 2904 2905 // Load UnrollBlock* into rdi 2906 __ mov(rdi, rax); 2907 2908 // Pop all the frames we must move/replace. 2909 // 2910 // Frame picture (youngest to oldest) 2911 // 1: self-frame (no frame link) 2912 // 2: deopting frame (no frame link) 2913 // 3: caller of deopting frame (could be compiled/interpreted). 2914 2915 // Pop self-frame. We have no frame, and must rely only on rax and rsp. 2916 __ addptr(rsp, (SimpleRuntimeFrame::framesize - 2) << LogBytesPerInt); // Epilog! 2917 2918 // Pop deoptimized frame (int) 2919 __ movl(rcx, Address(rdi, 2920 Deoptimization::UnrollBlock:: 2921 size_of_deoptimized_frame_offset_in_bytes())); 2922 __ addptr(rsp, rcx); 2923 2924 // rsp should be pointing at the return address to the caller (3) 2925 2926 // Stack bang to make sure there's enough room for these interpreter frames. 2927 if (UseStackBanging) { 2928 __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes())); 2929 __ bang_stack_size(rbx, rcx); 2930 } 2931 2932 // Load address of array of frame pcs into rcx (address*) 2933 __ movptr(rcx, 2934 Address(rdi, 2935 Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes())); 2936 2937 // Trash the return pc 2938 __ addptr(rsp, wordSize); 2939 2940 // Load address of array of frame sizes into rsi (intptr_t*) 2941 __ movptr(rsi, Address(rdi, 2942 Deoptimization::UnrollBlock:: 2943 frame_sizes_offset_in_bytes())); 2944 2945 // Counter 2946 __ movl(rdx, Address(rdi, 2947 Deoptimization::UnrollBlock:: 2948 number_of_frames_offset_in_bytes())); // (int) 2949 2950 // Pick up the initial fp we should save 2951 __ movptr(rbp, 2952 Address(rdi, 2953 Deoptimization::UnrollBlock::initial_fp_offset_in_bytes())); 2954 2955 // Now adjust the caller's stack to make up for the extra locals but 2956 // record the original sp so that we can save it in the skeletal 2957 // interpreter frame and the stack walking of interpreter_sender 2958 // will get the unextended sp value and not the "real" sp value. 2959 2960 const Register sender_sp = r8; 2961 2962 __ mov(sender_sp, rsp); 2963 __ movl(rbx, Address(rdi, 2964 Deoptimization::UnrollBlock:: 2965 caller_adjustment_offset_in_bytes())); // (int) 2966 __ subptr(rsp, rbx); 2967 2968 // Push interpreter frames in a loop 2969 Label loop; 2970 __ bind(loop); 2971 __ movptr(rbx, Address(rsi, 0)); // Load frame size 2972 __ subptr(rbx, 2 * wordSize); // We'll push pc and rbp by hand 2973 __ pushptr(Address(rcx, 0)); // Save return address 2974 __ enter(); // Save old & set new rbp 2975 __ subptr(rsp, rbx); // Prolog 2976 #ifdef CC_INTERP 2977 __ movptr(Address(rbp, 2978 -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))), 2979 sender_sp); // Make it walkable 2980 #else // CC_INTERP 2981 __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), 2982 sender_sp); // Make it walkable 2983 // This value is corrected by layout_activation_impl 2984 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD ); 2985 #endif // CC_INTERP 2986 __ mov(sender_sp, rsp); // Pass sender_sp to next frame 2987 __ addptr(rsi, wordSize); // Bump array pointer (sizes) 2988 __ addptr(rcx, wordSize); // Bump array pointer (pcs) 2989 __ decrementl(rdx); // Decrement counter 2990 __ jcc(Assembler::notZero, loop); 2991 __ pushptr(Address(rcx, 0)); // Save final return address 2992 2993 // Re-push self-frame 2994 __ enter(); // Save old & set new rbp 2995 __ subptr(rsp, (SimpleRuntimeFrame::framesize - 4) << LogBytesPerInt); 2996 // Prolog 2997 2998 // Use rbp because the frames look interpreted now 2999 __ set_last_Java_frame(noreg, rbp, NULL); 3000 3001 // Call C code. Need thread but NOT official VM entry 3002 // crud. We cannot block on this call, no GC can happen. Call should 3003 // restore return values to their stack-slots with the new SP. 3004 // Thread is in rdi already. 3005 // 3006 // BasicType unpack_frames(JavaThread* thread, int exec_mode); 3007 3008 __ mov(c_rarg0, r15_thread); 3009 __ movl(c_rarg1, Deoptimization::Unpack_uncommon_trap); 3010 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames))); 3011 3012 // Set an oopmap for the call site 3013 oop_maps->add_gc_map(__ pc() - start, new OopMap(SimpleRuntimeFrame::framesize, 0)); 3014 3015 __ reset_last_Java_frame(true, false); 3016 3017 // Pop self-frame. 3018 __ leave(); // Epilog 3019 3020 // Jump to interpreter 3021 __ ret(0); 3022 3023 // Make sure all code is generated 3024 masm->flush(); 3025 3026 _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, oop_maps, 3027 SimpleRuntimeFrame::framesize >> 1); 3028 } 3029 #endif // COMPILER2 3030 3031 3032 //------------------------------generate_handler_blob------ 3033 // 3034 // Generate a special Compile2Runtime blob that saves all registers, 3035 // and setup oopmap. 3036 // 3037 static SafepointBlob* generate_handler_blob(address call_ptr, bool cause_return) { 3038 assert(StubRoutines::forward_exception_entry() != NULL, 3039 "must be generated before"); 3040 3041 ResourceMark rm; 3042 OopMapSet *oop_maps = new OopMapSet(); 3043 OopMap* map; 3044 3045 // Allocate space for the code. Setup code generation tools. 3046 CodeBuffer buffer("handler_blob", 2048, 1024); 3047 MacroAssembler* masm = new MacroAssembler(&buffer); 3048 3049 address start = __ pc(); 3050 address call_pc = NULL; 3051 int frame_size_in_words; 3052 3053 // Make room for return address (or push it again) 3054 if (!cause_return) { 3055 __ push(rbx); 3056 } 3057 3058 // Save registers, fpu state, and flags 3059 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words); 3060 3061 // The following is basically a call_VM. However, we need the precise 3062 // address of the call in order to generate an oopmap. Hence, we do all the 3063 // work outselves. 3064 3065 __ set_last_Java_frame(noreg, noreg, NULL); 3066 3067 // The return address must always be correct so that frame constructor never 3068 // sees an invalid pc. 3069 3070 if (!cause_return) { 3071 // overwrite the dummy value we pushed on entry 3072 __ movptr(c_rarg0, Address(r15_thread, JavaThread::saved_exception_pc_offset())); 3073 __ movptr(Address(rbp, wordSize), c_rarg0); 3074 } 3075 3076 // Do the call 3077 __ mov(c_rarg0, r15_thread); 3078 __ call(RuntimeAddress(call_ptr)); 3079 3080 // Set an oopmap for the call site. This oopmap will map all 3081 // oop-registers and debug-info registers as callee-saved. This 3082 // will allow deoptimization at this safepoint to find all possible 3083 // debug-info recordings, as well as let GC find all oops. 3084 3085 oop_maps->add_gc_map( __ pc() - start, map); 3086 3087 Label noException; 3088 3089 __ reset_last_Java_frame(false, false); 3090 3091 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 3092 __ jcc(Assembler::equal, noException); 3093 3094 // Exception pending 3095 3096 RegisterSaver::restore_live_registers(masm); 3097 3098 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 3099 3100 // No exception case 3101 __ bind(noException); 3102 3103 // Normal exit, restore registers and exit. 3104 RegisterSaver::restore_live_registers(masm); 3105 3106 __ ret(0); 3107 3108 // Make sure all code is generated 3109 masm->flush(); 3110 3111 // Fill-out other meta info 3112 return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words); 3113 } 3114 3115 // 3116 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss 3117 // 3118 // Generate a stub that calls into vm to find out the proper destination 3119 // of a java call. All the argument registers are live at this point 3120 // but since this is generic code we don't know what they are and the caller 3121 // must do any gc of the args. 3122 // 3123 static RuntimeStub* generate_resolve_blob(address destination, const char* name) { 3124 assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before"); 3125 3126 // allocate space for the code 3127 ResourceMark rm; 3128 3129 CodeBuffer buffer(name, 1000, 512); 3130 MacroAssembler* masm = new MacroAssembler(&buffer); 3131 3132 int frame_size_in_words; 3133 3134 OopMapSet *oop_maps = new OopMapSet(); 3135 OopMap* map = NULL; 3136 3137 int start = __ offset(); 3138 3139 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words); 3140 3141 int frame_complete = __ offset(); 3142 3143 __ set_last_Java_frame(noreg, noreg, NULL); 3144 3145 __ mov(c_rarg0, r15_thread); 3146 3147 __ call(RuntimeAddress(destination)); 3148 3149 3150 // Set an oopmap for the call site. 3151 // We need this not only for callee-saved registers, but also for volatile 3152 // registers that the compiler might be keeping live across a safepoint. 3153 3154 oop_maps->add_gc_map( __ offset() - start, map); 3155 3156 // rax contains the address we are going to jump to assuming no exception got installed 3157 3158 // clear last_Java_sp 3159 __ reset_last_Java_frame(false, false); 3160 // check for pending exceptions 3161 Label pending; 3162 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 3163 __ jcc(Assembler::notEqual, pending); 3164 3165 // get the returned methodOop 3166 __ movptr(rbx, Address(r15_thread, JavaThread::vm_result_offset())); 3167 __ movptr(Address(rsp, RegisterSaver::rbx_offset_in_bytes()), rbx); 3168 3169 __ movptr(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax); 3170 3171 RegisterSaver::restore_live_registers(masm); 3172 3173 // We are back the the original state on entry and ready to go. 3174 3175 __ jmp(rax); 3176 3177 // Pending exception after the safepoint 3178 3179 __ bind(pending); 3180 3181 RegisterSaver::restore_live_registers(masm); 3182 3183 // exception pending => remove activation and forward to exception handler 3184 3185 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int)NULL_WORD); 3186 3187 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset())); 3188 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 3189 3190 // ------------- 3191 // make sure all code is generated 3192 masm->flush(); 3193 3194 // return the blob 3195 // frame_size_words or bytes?? 3196 return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, true); 3197 } 3198 3199 3200 void SharedRuntime::generate_stubs() { 3201 3202 _wrong_method_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method), 3203 "wrong_method_stub"); 3204 _ic_miss_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method_ic_miss), 3205 "ic_miss_stub"); 3206 _resolve_opt_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_opt_virtual_call_C), 3207 "resolve_opt_virtual_call"); 3208 3209 _resolve_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_virtual_call_C), 3210 "resolve_virtual_call"); 3211 3212 _resolve_static_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_static_call_C), 3213 "resolve_static_call"); 3214 _polling_page_safepoint_handler_blob = 3215 generate_handler_blob(CAST_FROM_FN_PTR(address, 3216 SafepointSynchronize::handle_polling_page_exception), false); 3217 3218 _polling_page_return_handler_blob = 3219 generate_handler_blob(CAST_FROM_FN_PTR(address, 3220 SafepointSynchronize::handle_polling_page_exception), true); 3221 3222 generate_deopt_blob(); 3223 3224 #ifdef COMPILER2 3225 generate_uncommon_trap_blob(); 3226 #endif // COMPILER2 3227 } 3228 3229 3230 #ifdef COMPILER2 3231 // This is here instead of runtime_x86_64.cpp because it uses SimpleRuntimeFrame 3232 // 3233 //------------------------------generate_exception_blob--------------------------- 3234 // creates exception blob at the end 3235 // Using exception blob, this code is jumped from a compiled method. 3236 // (see emit_exception_handler in x86_64.ad file) 3237 // 3238 // Given an exception pc at a call we call into the runtime for the 3239 // handler in this method. This handler might merely restore state 3240 // (i.e. callee save registers) unwind the frame and jump to the 3241 // exception handler for the nmethod if there is no Java level handler 3242 // for the nmethod. 3243 // 3244 // This code is entered with a jmp. 3245 // 3246 // Arguments: 3247 // rax: exception oop 3248 // rdx: exception pc 3249 // 3250 // Results: 3251 // rax: exception oop 3252 // rdx: exception pc in caller or ??? 3253 // destination: exception handler of caller 3254 // 3255 // Note: the exception pc MUST be at a call (precise debug information) 3256 // Registers rax, rdx, rcx, rsi, rdi, r8-r11 are not callee saved. 3257 // 3258 3259 void OptoRuntime::generate_exception_blob() { 3260 assert(!OptoRuntime::is_callee_saved_register(RDX_num), ""); 3261 assert(!OptoRuntime::is_callee_saved_register(RAX_num), ""); 3262 assert(!OptoRuntime::is_callee_saved_register(RCX_num), ""); 3263 3264 assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned"); 3265 3266 // Allocate space for the code 3267 ResourceMark rm; 3268 // Setup code generation tools 3269 CodeBuffer buffer("exception_blob", 2048, 1024); 3270 MacroAssembler* masm = new MacroAssembler(&buffer); 3271 3272 3273 address start = __ pc(); 3274 3275 // Exception pc is 'return address' for stack walker 3276 __ push(rdx); 3277 __ subptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Prolog 3278 3279 // Save callee-saved registers. See x86_64.ad. 3280 3281 // rbp is an implicitly saved callee saved register (i.e. the calling 3282 // convention will save restore it in prolog/epilog) Other than that 3283 // there are no callee save registers now that adapter frames are gone. 3284 3285 __ movptr(Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt), rbp); 3286 3287 // Store exception in Thread object. We cannot pass any arguments to the 3288 // handle_exception call, since we do not want to make any assumption 3289 // about the size of the frame where the exception happened in. 3290 // c_rarg0 is either rdi (Linux) or rcx (Windows). 3291 __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()),rax); 3292 __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), rdx); 3293 3294 // This call does all the hard work. It checks if an exception handler 3295 // exists in the method. 3296 // If so, it returns the handler address. 3297 // If not, it prepares for stack-unwinding, restoring the callee-save 3298 // registers of the frame being removed. 3299 // 3300 // address OptoRuntime::handle_exception_C(JavaThread* thread) 3301 3302 __ set_last_Java_frame(noreg, noreg, NULL); 3303 __ mov(c_rarg0, r15_thread); 3304 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, OptoRuntime::handle_exception_C))); 3305 3306 // Set an oopmap for the call site. This oopmap will only be used if we 3307 // are unwinding the stack. Hence, all locations will be dead. 3308 // Callee-saved registers will be the same as the frame above (i.e., 3309 // handle_exception_stub), since they were restored when we got the 3310 // exception. 3311 3312 OopMapSet* oop_maps = new OopMapSet(); 3313 3314 oop_maps->add_gc_map( __ pc()-start, new OopMap(SimpleRuntimeFrame::framesize, 0)); 3315 3316 __ reset_last_Java_frame(false, false); 3317 3318 // Restore callee-saved registers 3319 3320 // rbp is an implicitly saved callee saved register (i.e. the calling 3321 // convention will save restore it in prolog/epilog) Other than that 3322 // there are no callee save registers no that adapter frames are gone. 3323 3324 __ movptr(rbp, Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt)); 3325 3326 __ addptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Epilog 3327 __ pop(rdx); // No need for exception pc anymore 3328 3329 // rax: exception handler 3330 3331 // Restore SP from BP if the exception PC is a MethodHandle call. 3332 __ cmpl(Address(r15_thread, JavaThread::is_method_handle_exception_offset()), 0); 3333 __ cmovptr(Assembler::notEqual, rsp, rbp); 3334 3335 // We have a handler in rax (could be deopt blob). 3336 __ mov(r8, rax); 3337 3338 // Get the exception oop 3339 __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset())); 3340 // Get the exception pc in case we are deoptimized 3341 __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset())); 3342 #ifdef ASSERT 3343 __ movptr(Address(r15_thread, JavaThread::exception_handler_pc_offset()), (int)NULL_WORD); 3344 __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int)NULL_WORD); 3345 #endif 3346 // Clear the exception oop so GC no longer processes it as a root. 3347 __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), (int)NULL_WORD); 3348 3349 // rax: exception oop 3350 // r8: exception handler 3351 // rdx: exception pc 3352 // Jump to handler 3353 3354 __ jmp(r8); 3355 3356 // Make sure all code is generated 3357 masm->flush(); 3358 3359 // Set exception blob 3360 _exception_blob = ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1); 3361 } 3362 #endif // COMPILER2