1 /* 2 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "incls/_precompiled.incl" 26 #include "incls/_sharedRuntime_x86_32.cpp.incl" 27 28 #define __ masm-> 29 #ifdef COMPILER2 30 UncommonTrapBlob *SharedRuntime::_uncommon_trap_blob; 31 #endif // COMPILER2 32 33 DeoptimizationBlob *SharedRuntime::_deopt_blob; 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 class RegisterSaver { 45 enum { FPU_regs_live = 8 /*for the FPU stack*/+8/*eight more for XMM registers*/ }; 46 // Capture info about frame layout 47 enum layout { 48 fpu_state_off = 0, 49 fpu_state_end = fpu_state_off+FPUStateSizeInWords-1, 50 st0_off, st0H_off, 51 st1_off, st1H_off, 52 st2_off, st2H_off, 53 st3_off, st3H_off, 54 st4_off, st4H_off, 55 st5_off, st5H_off, 56 st6_off, st6H_off, 57 st7_off, st7H_off, 58 59 xmm0_off, xmm0H_off, 60 xmm1_off, xmm1H_off, 61 xmm2_off, xmm2H_off, 62 xmm3_off, xmm3H_off, 63 xmm4_off, xmm4H_off, 64 xmm5_off, xmm5H_off, 65 xmm6_off, xmm6H_off, 66 xmm7_off, xmm7H_off, 67 flags_off, 68 rdi_off, 69 rsi_off, 70 ignore_off, // extra copy of rbp, 71 rsp_off, 72 rbx_off, 73 rdx_off, 74 rcx_off, 75 rax_off, 76 // The frame sender code expects that rbp will be in the "natural" place and 77 // will override any oopMap setting for it. We must therefore force the layout 78 // so that it agrees with the frame sender code. 79 rbp_off, 80 return_off, // slot for return address 81 reg_save_size }; 82 83 84 public: 85 86 static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, 87 int* total_frame_words, bool verify_fpu = true); 88 static void restore_live_registers(MacroAssembler* masm); 89 90 static int rax_offset() { return rax_off; } 91 static int rbx_offset() { return rbx_off; } 92 93 // Offsets into the register save area 94 // Used by deoptimization when it is managing result register 95 // values on its own 96 97 static int raxOffset(void) { return rax_off; } 98 static int rdxOffset(void) { return rdx_off; } 99 static int rbxOffset(void) { return rbx_off; } 100 static int xmm0Offset(void) { return xmm0_off; } 101 // This really returns a slot in the fp save area, which one is not important 102 static int fpResultOffset(void) { return st0_off; } 103 104 // During deoptimization only the result register need to be restored 105 // all the other values have already been extracted. 106 107 static void restore_result_registers(MacroAssembler* masm); 108 109 }; 110 111 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, 112 int* total_frame_words, bool verify_fpu) { 113 114 int frame_size_in_bytes = (reg_save_size + additional_frame_words) * wordSize; 115 int frame_words = frame_size_in_bytes / wordSize; 116 *total_frame_words = frame_words; 117 118 assert(FPUStateSizeInWords == 27, "update stack layout"); 119 120 // save registers, fpu state, and flags 121 // We assume caller has already has return address slot on the stack 122 // We push epb twice in this sequence because we want the real rbp, 123 // to be under the return like a normal enter and we want to use pusha 124 // We push by hand instead of pusing push 125 __ enter(); 126 __ pusha(); 127 __ pushf(); 128 __ subptr(rsp,FPU_regs_live*sizeof(jdouble)); // Push FPU registers space 129 __ push_FPU_state(); // Save FPU state & init 130 131 if (verify_fpu) { 132 // Some stubs may have non standard FPU control word settings so 133 // only check and reset the value when it required to be the 134 // standard value. The safepoint blob in particular can be used 135 // in methods which are using the 24 bit control word for 136 // optimized float math. 137 138 #ifdef ASSERT 139 // Make sure the control word has the expected value 140 Label ok; 141 __ cmpw(Address(rsp, 0), StubRoutines::fpu_cntrl_wrd_std()); 142 __ jccb(Assembler::equal, ok); 143 __ stop("corrupted control word detected"); 144 __ bind(ok); 145 #endif 146 147 // Reset the control word to guard against exceptions being unmasked 148 // since fstp_d can cause FPU stack underflow exceptions. Write it 149 // into the on stack copy and then reload that to make sure that the 150 // current and future values are correct. 151 __ movw(Address(rsp, 0), StubRoutines::fpu_cntrl_wrd_std()); 152 } 153 154 __ frstor(Address(rsp, 0)); 155 if (!verify_fpu) { 156 // Set the control word so that exceptions are masked for the 157 // following code. 158 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); 159 } 160 161 // Save the FPU registers in de-opt-able form 162 163 __ fstp_d(Address(rsp, st0_off*wordSize)); // st(0) 164 __ fstp_d(Address(rsp, st1_off*wordSize)); // st(1) 165 __ fstp_d(Address(rsp, st2_off*wordSize)); // st(2) 166 __ fstp_d(Address(rsp, st3_off*wordSize)); // st(3) 167 __ fstp_d(Address(rsp, st4_off*wordSize)); // st(4) 168 __ fstp_d(Address(rsp, st5_off*wordSize)); // st(5) 169 __ fstp_d(Address(rsp, st6_off*wordSize)); // st(6) 170 __ fstp_d(Address(rsp, st7_off*wordSize)); // st(7) 171 172 if( UseSSE == 1 ) { // Save the XMM state 173 __ movflt(Address(rsp,xmm0_off*wordSize),xmm0); 174 __ movflt(Address(rsp,xmm1_off*wordSize),xmm1); 175 __ movflt(Address(rsp,xmm2_off*wordSize),xmm2); 176 __ movflt(Address(rsp,xmm3_off*wordSize),xmm3); 177 __ movflt(Address(rsp,xmm4_off*wordSize),xmm4); 178 __ movflt(Address(rsp,xmm5_off*wordSize),xmm5); 179 __ movflt(Address(rsp,xmm6_off*wordSize),xmm6); 180 __ movflt(Address(rsp,xmm7_off*wordSize),xmm7); 181 } else if( UseSSE >= 2 ) { 182 __ movdbl(Address(rsp,xmm0_off*wordSize),xmm0); 183 __ movdbl(Address(rsp,xmm1_off*wordSize),xmm1); 184 __ movdbl(Address(rsp,xmm2_off*wordSize),xmm2); 185 __ movdbl(Address(rsp,xmm3_off*wordSize),xmm3); 186 __ movdbl(Address(rsp,xmm4_off*wordSize),xmm4); 187 __ movdbl(Address(rsp,xmm5_off*wordSize),xmm5); 188 __ movdbl(Address(rsp,xmm6_off*wordSize),xmm6); 189 __ movdbl(Address(rsp,xmm7_off*wordSize),xmm7); 190 } 191 192 // Set an oopmap for the call site. This oopmap will map all 193 // oop-registers and debug-info registers as callee-saved. This 194 // will allow deoptimization at this safepoint to find all possible 195 // debug-info recordings, as well as let GC find all oops. 196 197 OopMapSet *oop_maps = new OopMapSet(); 198 OopMap* map = new OopMap( frame_words, 0 ); 199 200 #define STACK_OFFSET(x) VMRegImpl::stack2reg((x) + additional_frame_words) 201 202 map->set_callee_saved(STACK_OFFSET( rax_off), rax->as_VMReg()); 203 map->set_callee_saved(STACK_OFFSET( rcx_off), rcx->as_VMReg()); 204 map->set_callee_saved(STACK_OFFSET( rdx_off), rdx->as_VMReg()); 205 map->set_callee_saved(STACK_OFFSET( rbx_off), rbx->as_VMReg()); 206 // rbp, location is known implicitly, no oopMap 207 map->set_callee_saved(STACK_OFFSET( rsi_off), rsi->as_VMReg()); 208 map->set_callee_saved(STACK_OFFSET( rdi_off), rdi->as_VMReg()); 209 map->set_callee_saved(STACK_OFFSET(st0_off), as_FloatRegister(0)->as_VMReg()); 210 map->set_callee_saved(STACK_OFFSET(st1_off), as_FloatRegister(1)->as_VMReg()); 211 map->set_callee_saved(STACK_OFFSET(st2_off), as_FloatRegister(2)->as_VMReg()); 212 map->set_callee_saved(STACK_OFFSET(st3_off), as_FloatRegister(3)->as_VMReg()); 213 map->set_callee_saved(STACK_OFFSET(st4_off), as_FloatRegister(4)->as_VMReg()); 214 map->set_callee_saved(STACK_OFFSET(st5_off), as_FloatRegister(5)->as_VMReg()); 215 map->set_callee_saved(STACK_OFFSET(st6_off), as_FloatRegister(6)->as_VMReg()); 216 map->set_callee_saved(STACK_OFFSET(st7_off), as_FloatRegister(7)->as_VMReg()); 217 map->set_callee_saved(STACK_OFFSET(xmm0_off), xmm0->as_VMReg()); 218 map->set_callee_saved(STACK_OFFSET(xmm1_off), xmm1->as_VMReg()); 219 map->set_callee_saved(STACK_OFFSET(xmm2_off), xmm2->as_VMReg()); 220 map->set_callee_saved(STACK_OFFSET(xmm3_off), xmm3->as_VMReg()); 221 map->set_callee_saved(STACK_OFFSET(xmm4_off), xmm4->as_VMReg()); 222 map->set_callee_saved(STACK_OFFSET(xmm5_off), xmm5->as_VMReg()); 223 map->set_callee_saved(STACK_OFFSET(xmm6_off), xmm6->as_VMReg()); 224 map->set_callee_saved(STACK_OFFSET(xmm7_off), xmm7->as_VMReg()); 225 // %%% This is really a waste but we'll keep things as they were for now 226 if (true) { 227 #define NEXTREG(x) (x)->as_VMReg()->next() 228 map->set_callee_saved(STACK_OFFSET(st0H_off), NEXTREG(as_FloatRegister(0))); 229 map->set_callee_saved(STACK_OFFSET(st1H_off), NEXTREG(as_FloatRegister(1))); 230 map->set_callee_saved(STACK_OFFSET(st2H_off), NEXTREG(as_FloatRegister(2))); 231 map->set_callee_saved(STACK_OFFSET(st3H_off), NEXTREG(as_FloatRegister(3))); 232 map->set_callee_saved(STACK_OFFSET(st4H_off), NEXTREG(as_FloatRegister(4))); 233 map->set_callee_saved(STACK_OFFSET(st5H_off), NEXTREG(as_FloatRegister(5))); 234 map->set_callee_saved(STACK_OFFSET(st6H_off), NEXTREG(as_FloatRegister(6))); 235 map->set_callee_saved(STACK_OFFSET(st7H_off), NEXTREG(as_FloatRegister(7))); 236 map->set_callee_saved(STACK_OFFSET(xmm0H_off), NEXTREG(xmm0)); 237 map->set_callee_saved(STACK_OFFSET(xmm1H_off), NEXTREG(xmm1)); 238 map->set_callee_saved(STACK_OFFSET(xmm2H_off), NEXTREG(xmm2)); 239 map->set_callee_saved(STACK_OFFSET(xmm3H_off), NEXTREG(xmm3)); 240 map->set_callee_saved(STACK_OFFSET(xmm4H_off), NEXTREG(xmm4)); 241 map->set_callee_saved(STACK_OFFSET(xmm5H_off), NEXTREG(xmm5)); 242 map->set_callee_saved(STACK_OFFSET(xmm6H_off), NEXTREG(xmm6)); 243 map->set_callee_saved(STACK_OFFSET(xmm7H_off), NEXTREG(xmm7)); 244 #undef NEXTREG 245 #undef STACK_OFFSET 246 } 247 248 return map; 249 250 } 251 252 void RegisterSaver::restore_live_registers(MacroAssembler* masm) { 253 254 // Recover XMM & FPU state 255 if( UseSSE == 1 ) { 256 __ movflt(xmm0,Address(rsp,xmm0_off*wordSize)); 257 __ movflt(xmm1,Address(rsp,xmm1_off*wordSize)); 258 __ movflt(xmm2,Address(rsp,xmm2_off*wordSize)); 259 __ movflt(xmm3,Address(rsp,xmm3_off*wordSize)); 260 __ movflt(xmm4,Address(rsp,xmm4_off*wordSize)); 261 __ movflt(xmm5,Address(rsp,xmm5_off*wordSize)); 262 __ movflt(xmm6,Address(rsp,xmm6_off*wordSize)); 263 __ movflt(xmm7,Address(rsp,xmm7_off*wordSize)); 264 } else if( UseSSE >= 2 ) { 265 __ movdbl(xmm0,Address(rsp,xmm0_off*wordSize)); 266 __ movdbl(xmm1,Address(rsp,xmm1_off*wordSize)); 267 __ movdbl(xmm2,Address(rsp,xmm2_off*wordSize)); 268 __ movdbl(xmm3,Address(rsp,xmm3_off*wordSize)); 269 __ movdbl(xmm4,Address(rsp,xmm4_off*wordSize)); 270 __ movdbl(xmm5,Address(rsp,xmm5_off*wordSize)); 271 __ movdbl(xmm6,Address(rsp,xmm6_off*wordSize)); 272 __ movdbl(xmm7,Address(rsp,xmm7_off*wordSize)); 273 } 274 __ pop_FPU_state(); 275 __ addptr(rsp, FPU_regs_live*sizeof(jdouble)); // Pop FPU registers 276 277 __ popf(); 278 __ popa(); 279 // Get the rbp, described implicitly by the frame sender code (no oopMap) 280 __ pop(rbp); 281 282 } 283 284 void RegisterSaver::restore_result_registers(MacroAssembler* masm) { 285 286 // Just restore result register. Only used by deoptimization. By 287 // now any callee save register that needs to be restore to a c2 288 // caller of the deoptee has been extracted into the vframeArray 289 // and will be stuffed into the c2i adapter we create for later 290 // restoration so only result registers need to be restored here. 291 // 292 293 __ frstor(Address(rsp, 0)); // Restore fpu state 294 295 // Recover XMM & FPU state 296 if( UseSSE == 1 ) { 297 __ movflt(xmm0, Address(rsp, xmm0_off*wordSize)); 298 } else if( UseSSE >= 2 ) { 299 __ movdbl(xmm0, Address(rsp, xmm0_off*wordSize)); 300 } 301 __ movptr(rax, Address(rsp, rax_off*wordSize)); 302 __ movptr(rdx, Address(rsp, rdx_off*wordSize)); 303 // Pop all of the register save are off the stack except the return address 304 __ addptr(rsp, return_off * wordSize); 305 } 306 307 // The java_calling_convention describes stack locations as ideal slots on 308 // a frame with no abi restrictions. Since we must observe abi restrictions 309 // (like the placement of the register window) the slots must be biased by 310 // the following value. 311 static int reg2offset_in(VMReg r) { 312 // Account for saved rbp, and return address 313 // This should really be in_preserve_stack_slots 314 return (r->reg2stack() + 2) * VMRegImpl::stack_slot_size; 315 } 316 317 static int reg2offset_out(VMReg r) { 318 return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size; 319 } 320 321 // --------------------------------------------------------------------------- 322 // Read the array of BasicTypes from a signature, and compute where the 323 // arguments should go. Values in the VMRegPair regs array refer to 4-byte 324 // quantities. Values less than SharedInfo::stack0 are registers, those above 325 // refer to 4-byte stack slots. All stack slots are based off of the stack pointer 326 // as framesizes are fixed. 327 // VMRegImpl::stack0 refers to the first slot 0(sp). 328 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher. Register 329 // up to RegisterImpl::number_of_registers) are the 32-bit 330 // integer registers. 331 332 // Pass first two oop/int args in registers ECX and EDX. 333 // Pass first two float/double args in registers XMM0 and XMM1. 334 // Doubles have precedence, so if you pass a mix of floats and doubles 335 // the doubles will grab the registers before the floats will. 336 337 // Note: the INPUTS in sig_bt are in units of Java argument words, which are 338 // either 32-bit or 64-bit depending on the build. The OUTPUTS are in 32-bit 339 // units regardless of build. Of course for i486 there is no 64 bit build 340 341 342 // --------------------------------------------------------------------------- 343 // The compiled Java calling convention. 344 // Pass first two oop/int args in registers ECX and EDX. 345 // Pass first two float/double args in registers XMM0 and XMM1. 346 // Doubles have precedence, so if you pass a mix of floats and doubles 347 // the doubles will grab the registers before the floats will. 348 int SharedRuntime::java_calling_convention(const BasicType *sig_bt, 349 VMRegPair *regs, 350 int total_args_passed, 351 int is_outgoing) { 352 uint stack = 0; // Starting stack position for args on stack 353 354 355 // Pass first two oop/int args in registers ECX and EDX. 356 uint reg_arg0 = 9999; 357 uint reg_arg1 = 9999; 358 359 // Pass first two float/double args in registers XMM0 and XMM1. 360 // Doubles have precedence, so if you pass a mix of floats and doubles 361 // the doubles will grab the registers before the floats will. 362 // CNC - TURNED OFF FOR non-SSE. 363 // On Intel we have to round all doubles (and most floats) at 364 // call sites by storing to the stack in any case. 365 // UseSSE=0 ==> Don't Use ==> 9999+0 366 // UseSSE=1 ==> Floats only ==> 9999+1 367 // UseSSE>=2 ==> Floats or doubles ==> 9999+2 368 enum { fltarg_dontuse = 9999+0, fltarg_float_only = 9999+1, fltarg_flt_dbl = 9999+2 }; 369 uint fargs = (UseSSE>=2) ? 2 : UseSSE; 370 uint freg_arg0 = 9999+fargs; 371 uint freg_arg1 = 9999+fargs; 372 373 // Pass doubles & longs aligned on the stack. First count stack slots for doubles 374 int i; 375 for( i = 0; i < total_args_passed; i++) { 376 if( sig_bt[i] == T_DOUBLE ) { 377 // first 2 doubles go in registers 378 if( freg_arg0 == fltarg_flt_dbl ) freg_arg0 = i; 379 else if( freg_arg1 == fltarg_flt_dbl ) freg_arg1 = i; 380 else // Else double is passed low on the stack to be aligned. 381 stack += 2; 382 } else if( sig_bt[i] == T_LONG ) { 383 stack += 2; 384 } 385 } 386 int dstack = 0; // Separate counter for placing doubles 387 388 // Now pick where all else goes. 389 for( i = 0; i < total_args_passed; i++) { 390 // From the type and the argument number (count) compute the location 391 switch( sig_bt[i] ) { 392 case T_SHORT: 393 case T_CHAR: 394 case T_BYTE: 395 case T_BOOLEAN: 396 case T_INT: 397 case T_ARRAY: 398 case T_OBJECT: 399 case T_ADDRESS: 400 if( reg_arg0 == 9999 ) { 401 reg_arg0 = i; 402 regs[i].set1(rcx->as_VMReg()); 403 } else if( reg_arg1 == 9999 ) { 404 reg_arg1 = i; 405 regs[i].set1(rdx->as_VMReg()); 406 } else { 407 regs[i].set1(VMRegImpl::stack2reg(stack++)); 408 } 409 break; 410 case T_FLOAT: 411 if( freg_arg0 == fltarg_flt_dbl || freg_arg0 == fltarg_float_only ) { 412 freg_arg0 = i; 413 regs[i].set1(xmm0->as_VMReg()); 414 } else if( freg_arg1 == fltarg_flt_dbl || freg_arg1 == fltarg_float_only ) { 415 freg_arg1 = i; 416 regs[i].set1(xmm1->as_VMReg()); 417 } else { 418 regs[i].set1(VMRegImpl::stack2reg(stack++)); 419 } 420 break; 421 case T_LONG: 422 assert(sig_bt[i+1] == T_VOID, "missing Half" ); 423 regs[i].set2(VMRegImpl::stack2reg(dstack)); 424 dstack += 2; 425 break; 426 case T_DOUBLE: 427 assert(sig_bt[i+1] == T_VOID, "missing Half" ); 428 if( freg_arg0 == (uint)i ) { 429 regs[i].set2(xmm0->as_VMReg()); 430 } else if( freg_arg1 == (uint)i ) { 431 regs[i].set2(xmm1->as_VMReg()); 432 } else { 433 regs[i].set2(VMRegImpl::stack2reg(dstack)); 434 dstack += 2; 435 } 436 break; 437 case T_VOID: regs[i].set_bad(); break; 438 break; 439 default: 440 ShouldNotReachHere(); 441 break; 442 } 443 } 444 445 // return value can be odd number of VMRegImpl stack slots make multiple of 2 446 return round_to(stack, 2); 447 } 448 449 // Patch the callers callsite with entry to compiled code if it exists. 450 static void patch_callers_callsite(MacroAssembler *masm) { 451 Label L; 452 __ verify_oop(rbx); 453 __ cmpptr(Address(rbx, in_bytes(methodOopDesc::code_offset())), (int32_t)NULL_WORD); 454 __ jcc(Assembler::equal, L); 455 // Schedule the branch target address early. 456 // Call into the VM to patch the caller, then jump to compiled callee 457 // rax, isn't live so capture return address while we easily can 458 __ movptr(rax, Address(rsp, 0)); 459 __ pusha(); 460 __ pushf(); 461 462 if (UseSSE == 1) { 463 __ subptr(rsp, 2*wordSize); 464 __ movflt(Address(rsp, 0), xmm0); 465 __ movflt(Address(rsp, wordSize), xmm1); 466 } 467 if (UseSSE >= 2) { 468 __ subptr(rsp, 4*wordSize); 469 __ movdbl(Address(rsp, 0), xmm0); 470 __ movdbl(Address(rsp, 2*wordSize), xmm1); 471 } 472 #ifdef COMPILER2 473 // C2 may leave the stack dirty if not in SSE2+ mode 474 if (UseSSE >= 2) { 475 __ verify_FPU(0, "c2i transition should have clean FPU stack"); 476 } else { 477 __ empty_FPU_stack(); 478 } 479 #endif /* COMPILER2 */ 480 481 // VM needs caller's callsite 482 __ push(rax); 483 // VM needs target method 484 __ push(rbx); 485 __ verify_oop(rbx); 486 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite))); 487 __ addptr(rsp, 2*wordSize); 488 489 if (UseSSE == 1) { 490 __ movflt(xmm0, Address(rsp, 0)); 491 __ movflt(xmm1, Address(rsp, wordSize)); 492 __ addptr(rsp, 2*wordSize); 493 } 494 if (UseSSE >= 2) { 495 __ movdbl(xmm0, Address(rsp, 0)); 496 __ movdbl(xmm1, Address(rsp, 2*wordSize)); 497 __ addptr(rsp, 4*wordSize); 498 } 499 500 __ popf(); 501 __ popa(); 502 __ bind(L); 503 } 504 505 506 static void move_c2i_double(MacroAssembler *masm, XMMRegister r, int st_off) { 507 int next_off = st_off - Interpreter::stackElementSize; 508 __ movdbl(Address(rsp, next_off), r); 509 } 510 511 static void gen_c2i_adapter(MacroAssembler *masm, 512 int total_args_passed, 513 int comp_args_on_stack, 514 const BasicType *sig_bt, 515 const VMRegPair *regs, 516 Label& skip_fixup) { 517 // Before we get into the guts of the C2I adapter, see if we should be here 518 // at all. We've come from compiled code and are attempting to jump to the 519 // interpreter, which means the caller made a static call to get here 520 // (vcalls always get a compiled target if there is one). Check for a 521 // compiled target. If there is one, we need to patch the caller's call. 522 patch_callers_callsite(masm); 523 524 __ bind(skip_fixup); 525 526 #ifdef COMPILER2 527 // C2 may leave the stack dirty if not in SSE2+ mode 528 if (UseSSE >= 2) { 529 __ verify_FPU(0, "c2i transition should have clean FPU stack"); 530 } else { 531 __ empty_FPU_stack(); 532 } 533 #endif /* COMPILER2 */ 534 535 // Since all args are passed on the stack, total_args_passed * interpreter_ 536 // stack_element_size is the 537 // space we need. 538 int extraspace = total_args_passed * Interpreter::stackElementSize; 539 540 // Get return address 541 __ pop(rax); 542 543 // set senderSP value 544 __ movptr(rsi, rsp); 545 546 __ subptr(rsp, extraspace); 547 548 // Now write the args into the outgoing interpreter space 549 for (int i = 0; i < total_args_passed; i++) { 550 if (sig_bt[i] == T_VOID) { 551 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half"); 552 continue; 553 } 554 555 // st_off points to lowest address on stack. 556 int st_off = ((total_args_passed - 1) - i) * Interpreter::stackElementSize; 557 int next_off = st_off - Interpreter::stackElementSize; 558 559 // Say 4 args: 560 // i st_off 561 // 0 12 T_LONG 562 // 1 8 T_VOID 563 // 2 4 T_OBJECT 564 // 3 0 T_BOOL 565 VMReg r_1 = regs[i].first(); 566 VMReg r_2 = regs[i].second(); 567 if (!r_1->is_valid()) { 568 assert(!r_2->is_valid(), ""); 569 continue; 570 } 571 572 if (r_1->is_stack()) { 573 // memory to memory use fpu stack top 574 int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace; 575 576 if (!r_2->is_valid()) { 577 __ movl(rdi, Address(rsp, ld_off)); 578 __ movptr(Address(rsp, st_off), rdi); 579 } else { 580 581 // ld_off == LSW, ld_off+VMRegImpl::stack_slot_size == MSW 582 // st_off == MSW, st_off-wordSize == LSW 583 584 __ movptr(rdi, Address(rsp, ld_off)); 585 __ movptr(Address(rsp, next_off), rdi); 586 #ifndef _LP64 587 __ movptr(rdi, Address(rsp, ld_off + wordSize)); 588 __ movptr(Address(rsp, st_off), rdi); 589 #else 590 #ifdef ASSERT 591 // Overwrite the unused slot with known junk 592 __ mov64(rax, CONST64(0xdeadffffdeadaaaa)); 593 __ movptr(Address(rsp, st_off), rax); 594 #endif /* ASSERT */ 595 #endif // _LP64 596 } 597 } else if (r_1->is_Register()) { 598 Register r = r_1->as_Register(); 599 if (!r_2->is_valid()) { 600 __ movl(Address(rsp, st_off), r); 601 } else { 602 // long/double in gpr 603 NOT_LP64(ShouldNotReachHere()); 604 // Two VMRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG 605 // T_DOUBLE and T_LONG use two slots in the interpreter 606 if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) { 607 // long/double in gpr 608 #ifdef ASSERT 609 // Overwrite the unused slot with known junk 610 LP64_ONLY(__ mov64(rax, CONST64(0xdeadffffdeadaaab))); 611 __ movptr(Address(rsp, st_off), rax); 612 #endif /* ASSERT */ 613 __ movptr(Address(rsp, next_off), r); 614 } else { 615 __ movptr(Address(rsp, st_off), r); 616 } 617 } 618 } else { 619 assert(r_1->is_XMMRegister(), ""); 620 if (!r_2->is_valid()) { 621 __ movflt(Address(rsp, st_off), r_1->as_XMMRegister()); 622 } else { 623 assert(sig_bt[i] == T_DOUBLE || sig_bt[i] == T_LONG, "wrong type"); 624 move_c2i_double(masm, r_1->as_XMMRegister(), st_off); 625 } 626 } 627 } 628 629 // Schedule the branch target address early. 630 __ movptr(rcx, Address(rbx, in_bytes(methodOopDesc::interpreter_entry_offset()))); 631 // And repush original return address 632 __ push(rax); 633 __ jmp(rcx); 634 } 635 636 637 static void move_i2c_double(MacroAssembler *masm, XMMRegister r, Register saved_sp, int ld_off) { 638 int next_val_off = ld_off - Interpreter::stackElementSize; 639 __ movdbl(r, Address(saved_sp, next_val_off)); 640 } 641 642 static void gen_i2c_adapter(MacroAssembler *masm, 643 int total_args_passed, 644 int comp_args_on_stack, 645 const BasicType *sig_bt, 646 const VMRegPair *regs) { 647 // we're being called from the interpreter but need to find the 648 // compiled return entry point. The return address on the stack 649 // should point at it and we just need to pull the old value out. 650 // load up the pointer to the compiled return entry point and 651 // rewrite our return pc. The code is arranged like so: 652 // 653 // .word Interpreter::return_sentinel 654 // .word address_of_compiled_return_point 655 // return_entry_point: blah_blah_blah 656 // 657 // So we can find the appropriate return point by loading up the word 658 // just prior to the current return address we have on the stack. 659 // 660 // We will only enter here from an interpreted frame and never from after 661 // passing thru a c2i. Azul allowed this but we do not. If we lose the 662 // race and use a c2i we will remain interpreted for the race loser(s). 663 // This removes all sorts of headaches on the x86 side and also eliminates 664 // the possibility of having c2i -> i2c -> c2i -> ... endless transitions. 665 666 667 // Note: rsi contains the senderSP on entry. We must preserve it since 668 // we may do a i2c -> c2i transition if we lose a race where compiled 669 // code goes non-entrant while we get args ready. 670 671 // Pick up the return address 672 __ movptr(rax, Address(rsp, 0)); 673 674 // If UseSSE >= 2 then no cleanup is needed on the return to the 675 // interpreter so skip fixing up the return entry point unless 676 // VerifyFPU is enabled. 677 if (UseSSE < 2 || VerifyFPU) { 678 Label skip, chk_int; 679 // If we were called from the call stub we need to do a little bit different 680 // cleanup than if the interpreter returned to the call stub. 681 682 ExternalAddress stub_return_address(StubRoutines::_call_stub_return_address); 683 __ cmpptr(rax, stub_return_address.addr()); 684 __ jcc(Assembler::notEqual, chk_int); 685 assert(StubRoutines::x86::get_call_stub_compiled_return() != NULL, "must be set"); 686 __ lea(rax, ExternalAddress(StubRoutines::x86::get_call_stub_compiled_return())); 687 __ jmp(skip); 688 689 // It must be the interpreter since we never get here via a c2i (unlike Azul) 690 691 __ bind(chk_int); 692 #ifdef ASSERT 693 { 694 Label ok; 695 __ cmpl(Address(rax, -2*wordSize), Interpreter::return_sentinel); 696 __ jcc(Assembler::equal, ok); 697 __ int3(); 698 __ bind(ok); 699 } 700 #endif // ASSERT 701 __ movptr(rax, Address(rax, -wordSize)); 702 __ bind(skip); 703 } 704 705 // rax, now contains the compiled return entry point which will do an 706 // cleanup needed for the return from compiled to interpreted. 707 708 // Must preserve original SP for loading incoming arguments because 709 // we need to align the outgoing SP for compiled code. 710 __ movptr(rdi, rsp); 711 712 // Cut-out for having no stack args. Since up to 2 int/oop args are passed 713 // in registers, we will occasionally have no stack args. 714 int comp_words_on_stack = 0; 715 if (comp_args_on_stack) { 716 // Sig words on the stack are greater-than VMRegImpl::stack0. Those in 717 // registers are below. By subtracting stack0, we either get a negative 718 // number (all values in registers) or the maximum stack slot accessed. 719 // int comp_args_on_stack = VMRegImpl::reg2stack(max_arg); 720 // Convert 4-byte stack slots to words. 721 comp_words_on_stack = round_to(comp_args_on_stack*4, wordSize)>>LogBytesPerWord; 722 // Round up to miminum stack alignment, in wordSize 723 comp_words_on_stack = round_to(comp_words_on_stack, 2); 724 __ subptr(rsp, comp_words_on_stack * wordSize); 725 } 726 727 // Align the outgoing SP 728 __ andptr(rsp, -(StackAlignmentInBytes)); 729 730 // push the return address on the stack (note that pushing, rather 731 // than storing it, yields the correct frame alignment for the callee) 732 __ push(rax); 733 734 // Put saved SP in another register 735 const Register saved_sp = rax; 736 __ movptr(saved_sp, rdi); 737 738 739 // Will jump to the compiled code just as if compiled code was doing it. 740 // Pre-load the register-jump target early, to schedule it better. 741 __ movptr(rdi, Address(rbx, in_bytes(methodOopDesc::from_compiled_offset()))); 742 743 // Now generate the shuffle code. Pick up all register args and move the 744 // rest through the floating point stack top. 745 for (int i = 0; i < total_args_passed; i++) { 746 if (sig_bt[i] == T_VOID) { 747 // Longs and doubles are passed in native word order, but misaligned 748 // in the 32-bit build. 749 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half"); 750 continue; 751 } 752 753 // Pick up 0, 1 or 2 words from SP+offset. 754 755 assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(), 756 "scrambled load targets?"); 757 // Load in argument order going down. 758 int ld_off = (total_args_passed - i) * Interpreter::stackElementSize; 759 // Point to interpreter value (vs. tag) 760 int next_off = ld_off - Interpreter::stackElementSize; 761 // 762 // 763 // 764 VMReg r_1 = regs[i].first(); 765 VMReg r_2 = regs[i].second(); 766 if (!r_1->is_valid()) { 767 assert(!r_2->is_valid(), ""); 768 continue; 769 } 770 if (r_1->is_stack()) { 771 // Convert stack slot to an SP offset (+ wordSize to account for return address ) 772 int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size + wordSize; 773 774 // We can use rsi as a temp here because compiled code doesn't need rsi as an input 775 // and if we end up going thru a c2i because of a miss a reasonable value of rsi 776 // we be generated. 777 if (!r_2->is_valid()) { 778 // __ fld_s(Address(saved_sp, ld_off)); 779 // __ fstp_s(Address(rsp, st_off)); 780 __ movl(rsi, Address(saved_sp, ld_off)); 781 __ movptr(Address(rsp, st_off), rsi); 782 } else { 783 // Interpreter local[n] == MSW, local[n+1] == LSW however locals 784 // are accessed as negative so LSW is at LOW address 785 786 // ld_off is MSW so get LSW 787 // st_off is LSW (i.e. reg.first()) 788 // __ fld_d(Address(saved_sp, next_off)); 789 // __ fstp_d(Address(rsp, st_off)); 790 // 791 // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE 792 // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case 793 // So we must adjust where to pick up the data to match the interpreter. 794 // 795 // Interpreter local[n] == MSW, local[n+1] == LSW however locals 796 // are accessed as negative so LSW is at LOW address 797 798 // ld_off is MSW so get LSW 799 const int offset = (NOT_LP64(true ||) sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)? 800 next_off : ld_off; 801 __ movptr(rsi, Address(saved_sp, offset)); 802 __ movptr(Address(rsp, st_off), rsi); 803 #ifndef _LP64 804 __ movptr(rsi, Address(saved_sp, ld_off)); 805 __ movptr(Address(rsp, st_off + wordSize), rsi); 806 #endif // _LP64 807 } 808 } else if (r_1->is_Register()) { // Register argument 809 Register r = r_1->as_Register(); 810 assert(r != rax, "must be different"); 811 if (r_2->is_valid()) { 812 // 813 // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE 814 // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case 815 // So we must adjust where to pick up the data to match the interpreter. 816 817 const int offset = (NOT_LP64(true ||) sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)? 818 next_off : ld_off; 819 820 // this can be a misaligned move 821 __ movptr(r, Address(saved_sp, offset)); 822 #ifndef _LP64 823 assert(r_2->as_Register() != rax, "need another temporary register"); 824 // Remember r_1 is low address (and LSB on x86) 825 // So r_2 gets loaded from high address regardless of the platform 826 __ movptr(r_2->as_Register(), Address(saved_sp, ld_off)); 827 #endif // _LP64 828 } else { 829 __ movl(r, Address(saved_sp, ld_off)); 830 } 831 } else { 832 assert(r_1->is_XMMRegister(), ""); 833 if (!r_2->is_valid()) { 834 __ movflt(r_1->as_XMMRegister(), Address(saved_sp, ld_off)); 835 } else { 836 move_i2c_double(masm, r_1->as_XMMRegister(), saved_sp, ld_off); 837 } 838 } 839 } 840 841 // 6243940 We might end up in handle_wrong_method if 842 // the callee is deoptimized as we race thru here. If that 843 // happens we don't want to take a safepoint because the 844 // caller frame will look interpreted and arguments are now 845 // "compiled" so it is much better to make this transition 846 // invisible to the stack walking code. Unfortunately if 847 // we try and find the callee by normal means a safepoint 848 // is possible. So we stash the desired callee in the thread 849 // and the vm will find there should this case occur. 850 851 __ get_thread(rax); 852 __ movptr(Address(rax, JavaThread::callee_target_offset()), rbx); 853 854 // move methodOop to rax, in case we end up in an c2i adapter. 855 // the c2i adapters expect methodOop in rax, (c2) because c2's 856 // resolve stubs return the result (the method) in rax,. 857 // I'd love to fix this. 858 __ mov(rax, rbx); 859 860 __ jmp(rdi); 861 } 862 863 // --------------------------------------------------------------- 864 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm, 865 int total_args_passed, 866 int comp_args_on_stack, 867 const BasicType *sig_bt, 868 const VMRegPair *regs, 869 AdapterFingerPrint* fingerprint) { 870 address i2c_entry = __ pc(); 871 872 gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs); 873 874 // ------------------------------------------------------------------------- 875 // Generate a C2I adapter. On entry we know rbx, holds the methodOop during calls 876 // to the interpreter. The args start out packed in the compiled layout. They 877 // need to be unpacked into the interpreter layout. This will almost always 878 // require some stack space. We grow the current (compiled) stack, then repack 879 // the args. We finally end in a jump to the generic interpreter entry point. 880 // On exit from the interpreter, the interpreter will restore our SP (lest the 881 // compiled code, which relys solely on SP and not EBP, get sick). 882 883 address c2i_unverified_entry = __ pc(); 884 Label skip_fixup; 885 886 Register holder = rax; 887 Register receiver = rcx; 888 Register temp = rbx; 889 890 { 891 892 Label missed; 893 894 __ verify_oop(holder); 895 __ movptr(temp, Address(receiver, oopDesc::klass_offset_in_bytes())); 896 __ verify_oop(temp); 897 898 __ cmpptr(temp, Address(holder, compiledICHolderOopDesc::holder_klass_offset())); 899 __ movptr(rbx, Address(holder, compiledICHolderOopDesc::holder_method_offset())); 900 __ jcc(Assembler::notEqual, missed); 901 // Method might have been compiled since the call site was patched to 902 // interpreted if that is the case treat it as a miss so we can get 903 // the call site corrected. 904 __ cmpptr(Address(rbx, in_bytes(methodOopDesc::code_offset())), (int32_t)NULL_WORD); 905 __ jcc(Assembler::equal, skip_fixup); 906 907 __ bind(missed); 908 __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); 909 } 910 911 address c2i_entry = __ pc(); 912 913 gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup); 914 915 __ flush(); 916 return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry); 917 } 918 919 int SharedRuntime::c_calling_convention(const BasicType *sig_bt, 920 VMRegPair *regs, 921 int total_args_passed) { 922 // We return the amount of VMRegImpl stack slots we need to reserve for all 923 // the arguments NOT counting out_preserve_stack_slots. 924 925 uint stack = 0; // All arguments on stack 926 927 for( int i = 0; i < total_args_passed; i++) { 928 // From the type and the argument number (count) compute the location 929 switch( sig_bt[i] ) { 930 case T_BOOLEAN: 931 case T_CHAR: 932 case T_FLOAT: 933 case T_BYTE: 934 case T_SHORT: 935 case T_INT: 936 case T_OBJECT: 937 case T_ARRAY: 938 case T_ADDRESS: 939 regs[i].set1(VMRegImpl::stack2reg(stack++)); 940 break; 941 case T_LONG: 942 case T_DOUBLE: // The stack numbering is reversed from Java 943 // Since C arguments do not get reversed, the ordering for 944 // doubles on the stack must be opposite the Java convention 945 assert(sig_bt[i+1] == T_VOID, "missing Half" ); 946 regs[i].set2(VMRegImpl::stack2reg(stack)); 947 stack += 2; 948 break; 949 case T_VOID: regs[i].set_bad(); break; 950 default: 951 ShouldNotReachHere(); 952 break; 953 } 954 } 955 return stack; 956 } 957 958 // A simple move of integer like type 959 static void simple_move32(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { 960 if (src.first()->is_stack()) { 961 if (dst.first()->is_stack()) { 962 // stack to stack 963 // __ ld(FP, reg2offset(src.first()) + STACK_BIAS, L5); 964 // __ st(L5, SP, reg2offset(dst.first()) + STACK_BIAS); 965 __ movl2ptr(rax, Address(rbp, reg2offset_in(src.first()))); 966 __ movptr(Address(rsp, reg2offset_out(dst.first())), rax); 967 } else { 968 // stack to reg 969 __ movl2ptr(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()))); 970 } 971 } else if (dst.first()->is_stack()) { 972 // reg to stack 973 // no need to sign extend on 64bit 974 __ movptr(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register()); 975 } else { 976 if (dst.first() != src.first()) { 977 __ mov(dst.first()->as_Register(), src.first()->as_Register()); 978 } 979 } 980 } 981 982 // An oop arg. Must pass a handle not the oop itself 983 static void object_move(MacroAssembler* masm, 984 OopMap* map, 985 int oop_handle_offset, 986 int framesize_in_slots, 987 VMRegPair src, 988 VMRegPair dst, 989 bool is_receiver, 990 int* receiver_offset) { 991 992 // Because of the calling conventions we know that src can be a 993 // register or a stack location. dst can only be a stack location. 994 995 assert(dst.first()->is_stack(), "must be stack"); 996 // must pass a handle. First figure out the location we use as a handle 997 998 if (src.first()->is_stack()) { 999 // Oop is already on the stack as an argument 1000 Register rHandle = rax; 1001 Label nil; 1002 __ xorptr(rHandle, rHandle); 1003 __ cmpptr(Address(rbp, reg2offset_in(src.first())), (int32_t)NULL_WORD); 1004 __ jcc(Assembler::equal, nil); 1005 __ lea(rHandle, Address(rbp, reg2offset_in(src.first()))); 1006 __ bind(nil); 1007 __ movptr(Address(rsp, reg2offset_out(dst.first())), rHandle); 1008 1009 int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots(); 1010 map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots)); 1011 if (is_receiver) { 1012 *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size; 1013 } 1014 } else { 1015 // Oop is in an a register we must store it to the space we reserve 1016 // on the stack for oop_handles 1017 const Register rOop = src.first()->as_Register(); 1018 const Register rHandle = rax; 1019 int oop_slot = (rOop == rcx ? 0 : 1) * VMRegImpl::slots_per_word + oop_handle_offset; 1020 int offset = oop_slot*VMRegImpl::stack_slot_size; 1021 Label skip; 1022 __ movptr(Address(rsp, offset), rOop); 1023 map->set_oop(VMRegImpl::stack2reg(oop_slot)); 1024 __ xorptr(rHandle, rHandle); 1025 __ cmpptr(rOop, (int32_t)NULL_WORD); 1026 __ jcc(Assembler::equal, skip); 1027 __ lea(rHandle, Address(rsp, offset)); 1028 __ bind(skip); 1029 // Store the handle parameter 1030 __ movptr(Address(rsp, reg2offset_out(dst.first())), rHandle); 1031 if (is_receiver) { 1032 *receiver_offset = offset; 1033 } 1034 } 1035 } 1036 1037 // A float arg may have to do float reg int reg conversion 1038 static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { 1039 assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move"); 1040 1041 // Because of the calling convention we know that src is either a stack location 1042 // or an xmm register. dst can only be a stack location. 1043 1044 assert(dst.first()->is_stack() && ( src.first()->is_stack() || src.first()->is_XMMRegister()), "bad parameters"); 1045 1046 if (src.first()->is_stack()) { 1047 __ movl(rax, Address(rbp, reg2offset_in(src.first()))); 1048 __ movptr(Address(rsp, reg2offset_out(dst.first())), rax); 1049 } else { 1050 // reg to stack 1051 __ movflt(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister()); 1052 } 1053 } 1054 1055 // A long move 1056 static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { 1057 1058 // The only legal possibility for a long_move VMRegPair is: 1059 // 1: two stack slots (possibly unaligned) 1060 // as neither the java or C calling convention will use registers 1061 // for longs. 1062 1063 if (src.first()->is_stack() && dst.first()->is_stack()) { 1064 assert(src.second()->is_stack() && dst.second()->is_stack(), "must be all stack"); 1065 __ movptr(rax, Address(rbp, reg2offset_in(src.first()))); 1066 NOT_LP64(__ movptr(rbx, Address(rbp, reg2offset_in(src.second())))); 1067 __ movptr(Address(rsp, reg2offset_out(dst.first())), rax); 1068 NOT_LP64(__ movptr(Address(rsp, reg2offset_out(dst.second())), rbx)); 1069 } else { 1070 ShouldNotReachHere(); 1071 } 1072 } 1073 1074 // A double move 1075 static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { 1076 1077 // The only legal possibilities for a double_move VMRegPair are: 1078 // The painful thing here is that like long_move a VMRegPair might be 1079 1080 // Because of the calling convention we know that src is either 1081 // 1: a single physical register (xmm registers only) 1082 // 2: two stack slots (possibly unaligned) 1083 // dst can only be a pair of stack slots. 1084 1085 assert(dst.first()->is_stack() && (src.first()->is_XMMRegister() || src.first()->is_stack()), "bad args"); 1086 1087 if (src.first()->is_stack()) { 1088 // source is all stack 1089 __ movptr(rax, Address(rbp, reg2offset_in(src.first()))); 1090 NOT_LP64(__ movptr(rbx, Address(rbp, reg2offset_in(src.second())))); 1091 __ movptr(Address(rsp, reg2offset_out(dst.first())), rax); 1092 NOT_LP64(__ movptr(Address(rsp, reg2offset_out(dst.second())), rbx)); 1093 } else { 1094 // reg to stack 1095 // No worries about stack alignment 1096 __ movdbl(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister()); 1097 } 1098 } 1099 1100 1101 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { 1102 // We always ignore the frame_slots arg and just use the space just below frame pointer 1103 // which by this time is free to use 1104 switch (ret_type) { 1105 case T_FLOAT: 1106 __ fstp_s(Address(rbp, -wordSize)); 1107 break; 1108 case T_DOUBLE: 1109 __ fstp_d(Address(rbp, -2*wordSize)); 1110 break; 1111 case T_VOID: break; 1112 case T_LONG: 1113 __ movptr(Address(rbp, -wordSize), rax); 1114 NOT_LP64(__ movptr(Address(rbp, -2*wordSize), rdx)); 1115 break; 1116 default: { 1117 __ movptr(Address(rbp, -wordSize), rax); 1118 } 1119 } 1120 } 1121 1122 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { 1123 // We always ignore the frame_slots arg and just use the space just below frame pointer 1124 // which by this time is free to use 1125 switch (ret_type) { 1126 case T_FLOAT: 1127 __ fld_s(Address(rbp, -wordSize)); 1128 break; 1129 case T_DOUBLE: 1130 __ fld_d(Address(rbp, -2*wordSize)); 1131 break; 1132 case T_LONG: 1133 __ movptr(rax, Address(rbp, -wordSize)); 1134 NOT_LP64(__ movptr(rdx, Address(rbp, -2*wordSize))); 1135 break; 1136 case T_VOID: break; 1137 default: { 1138 __ movptr(rax, Address(rbp, -wordSize)); 1139 } 1140 } 1141 } 1142 1143 // --------------------------------------------------------------------------- 1144 // Generate a native wrapper for a given method. The method takes arguments 1145 // in the Java compiled code convention, marshals them to the native 1146 // convention (handlizes oops, etc), transitions to native, makes the call, 1147 // returns to java state (possibly blocking), unhandlizes any result and 1148 // returns. 1149 nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler *masm, 1150 methodHandle method, 1151 int total_in_args, 1152 int comp_args_on_stack, 1153 BasicType *in_sig_bt, 1154 VMRegPair *in_regs, 1155 BasicType ret_type) { 1156 1157 // An OopMap for lock (and class if static) 1158 OopMapSet *oop_maps = new OopMapSet(); 1159 1160 // We have received a description of where all the java arg are located 1161 // on entry to the wrapper. We need to convert these args to where 1162 // the jni function will expect them. To figure out where they go 1163 // we convert the java signature to a C signature by inserting 1164 // the hidden arguments as arg[0] and possibly arg[1] (static method) 1165 1166 int total_c_args = total_in_args + 1; 1167 if (method->is_static()) { 1168 total_c_args++; 1169 } 1170 1171 BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args); 1172 VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args); 1173 1174 int argc = 0; 1175 out_sig_bt[argc++] = T_ADDRESS; 1176 if (method->is_static()) { 1177 out_sig_bt[argc++] = T_OBJECT; 1178 } 1179 1180 int i; 1181 for (i = 0; i < total_in_args ; i++ ) { 1182 out_sig_bt[argc++] = in_sig_bt[i]; 1183 } 1184 1185 1186 // Now figure out where the args must be stored and how much stack space 1187 // they require (neglecting out_preserve_stack_slots but space for storing 1188 // the 1st six register arguments). It's weird see int_stk_helper. 1189 // 1190 int out_arg_slots; 1191 out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args); 1192 1193 // Compute framesize for the wrapper. We need to handlize all oops in 1194 // registers a max of 2 on x86. 1195 1196 // Calculate the total number of stack slots we will need. 1197 1198 // First count the abi requirement plus all of the outgoing args 1199 int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; 1200 1201 // Now the space for the inbound oop handle area 1202 1203 int oop_handle_offset = stack_slots; 1204 stack_slots += 2*VMRegImpl::slots_per_word; 1205 1206 // Now any space we need for handlizing a klass if static method 1207 1208 int klass_slot_offset = 0; 1209 int klass_offset = -1; 1210 int lock_slot_offset = 0; 1211 bool is_static = false; 1212 int oop_temp_slot_offset = 0; 1213 1214 if (method->is_static()) { 1215 klass_slot_offset = stack_slots; 1216 stack_slots += VMRegImpl::slots_per_word; 1217 klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size; 1218 is_static = true; 1219 } 1220 1221 // Plus a lock if needed 1222 1223 if (method->is_synchronized()) { 1224 lock_slot_offset = stack_slots; 1225 stack_slots += VMRegImpl::slots_per_word; 1226 } 1227 1228 // Now a place (+2) to save return values or temp during shuffling 1229 // + 2 for return address (which we own) and saved rbp, 1230 stack_slots += 4; 1231 1232 // Ok The space we have allocated will look like: 1233 // 1234 // 1235 // FP-> | | 1236 // |---------------------| 1237 // | 2 slots for moves | 1238 // |---------------------| 1239 // | lock box (if sync) | 1240 // |---------------------| <- lock_slot_offset (-lock_slot_rbp_offset) 1241 // | klass (if static) | 1242 // |---------------------| <- klass_slot_offset 1243 // | oopHandle area | 1244 // |---------------------| <- oop_handle_offset (a max of 2 registers) 1245 // | outbound memory | 1246 // | based arguments | 1247 // | | 1248 // |---------------------| 1249 // | | 1250 // SP-> | out_preserved_slots | 1251 // 1252 // 1253 // **************************************************************************** 1254 // WARNING - on Windows Java Natives use pascal calling convention and pop the 1255 // arguments off of the stack after the jni call. Before the call we can use 1256 // instructions that are SP relative. After the jni call we switch to FP 1257 // relative instructions instead of re-adjusting the stack on windows. 1258 // **************************************************************************** 1259 1260 1261 // Now compute actual number of stack words we need rounding to make 1262 // stack properly aligned. 1263 stack_slots = round_to(stack_slots, StackAlignmentInSlots); 1264 1265 int stack_size = stack_slots * VMRegImpl::stack_slot_size; 1266 1267 intptr_t start = (intptr_t)__ pc(); 1268 1269 // First thing make an ic check to see if we should even be here 1270 1271 // We are free to use all registers as temps without saving them and 1272 // restoring them except rbp,. rbp, is the only callee save register 1273 // as far as the interpreter and the compiler(s) are concerned. 1274 1275 1276 const Register ic_reg = rax; 1277 const Register receiver = rcx; 1278 Label hit; 1279 Label exception_pending; 1280 1281 1282 __ verify_oop(receiver); 1283 __ cmpptr(ic_reg, Address(receiver, oopDesc::klass_offset_in_bytes())); 1284 __ jcc(Assembler::equal, hit); 1285 1286 __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); 1287 1288 // verified entry must be aligned for code patching. 1289 // and the first 5 bytes must be in the same cache line 1290 // if we align at 8 then we will be sure 5 bytes are in the same line 1291 __ align(8); 1292 1293 __ bind(hit); 1294 1295 int vep_offset = ((intptr_t)__ pc()) - start; 1296 1297 #ifdef COMPILER1 1298 if (InlineObjectHash && method->intrinsic_id() == vmIntrinsics::_hashCode) { 1299 // Object.hashCode can pull the hashCode from the header word 1300 // instead of doing a full VM transition once it's been computed. 1301 // Since hashCode is usually polymorphic at call sites we can't do 1302 // this optimization at the call site without a lot of work. 1303 Label slowCase; 1304 Register receiver = rcx; 1305 Register result = rax; 1306 __ movptr(result, Address(receiver, oopDesc::mark_offset_in_bytes())); 1307 1308 // check if locked 1309 __ testptr(result, markOopDesc::unlocked_value); 1310 __ jcc (Assembler::zero, slowCase); 1311 1312 if (UseBiasedLocking) { 1313 // Check if biased and fall through to runtime if so 1314 __ testptr(result, markOopDesc::biased_lock_bit_in_place); 1315 __ jcc (Assembler::notZero, slowCase); 1316 } 1317 1318 // get hash 1319 __ andptr(result, markOopDesc::hash_mask_in_place); 1320 // test if hashCode exists 1321 __ jcc (Assembler::zero, slowCase); 1322 __ shrptr(result, markOopDesc::hash_shift); 1323 __ ret(0); 1324 __ bind (slowCase); 1325 } 1326 #endif // COMPILER1 1327 1328 // The instruction at the verified entry point must be 5 bytes or longer 1329 // because it can be patched on the fly by make_non_entrant. The stack bang 1330 // instruction fits that requirement. 1331 1332 // Generate stack overflow check 1333 1334 if (UseStackBanging) { 1335 __ bang_stack_with_offset(StackShadowPages*os::vm_page_size()); 1336 } else { 1337 // need a 5 byte instruction to allow MT safe patching to non-entrant 1338 __ fat_nop(); 1339 } 1340 1341 // Generate a new frame for the wrapper. 1342 __ enter(); 1343 // -2 because return address is already present and so is saved rbp, 1344 __ subptr(rsp, stack_size - 2*wordSize); 1345 1346 // Frame is now completed as far a size and linkage. 1347 1348 int frame_complete = ((intptr_t)__ pc()) - start; 1349 1350 // Calculate the difference between rsp and rbp,. We need to know it 1351 // after the native call because on windows Java Natives will pop 1352 // the arguments and it is painful to do rsp relative addressing 1353 // in a platform independent way. So after the call we switch to 1354 // rbp, relative addressing. 1355 1356 int fp_adjustment = stack_size - 2*wordSize; 1357 1358 #ifdef COMPILER2 1359 // C2 may leave the stack dirty if not in SSE2+ mode 1360 if (UseSSE >= 2) { 1361 __ verify_FPU(0, "c2i transition should have clean FPU stack"); 1362 } else { 1363 __ empty_FPU_stack(); 1364 } 1365 #endif /* COMPILER2 */ 1366 1367 // Compute the rbp, offset for any slots used after the jni call 1368 1369 int lock_slot_rbp_offset = (lock_slot_offset*VMRegImpl::stack_slot_size) - fp_adjustment; 1370 int oop_temp_slot_rbp_offset = (oop_temp_slot_offset*VMRegImpl::stack_slot_size) - fp_adjustment; 1371 1372 // We use rdi as a thread pointer because it is callee save and 1373 // if we load it once it is usable thru the entire wrapper 1374 const Register thread = rdi; 1375 1376 // We use rsi as the oop handle for the receiver/klass 1377 // It is callee save so it survives the call to native 1378 1379 const Register oop_handle_reg = rsi; 1380 1381 __ get_thread(thread); 1382 1383 1384 // 1385 // We immediately shuffle the arguments so that any vm call we have to 1386 // make from here on out (sync slow path, jvmti, etc.) we will have 1387 // captured the oops from our caller and have a valid oopMap for 1388 // them. 1389 1390 // ----------------- 1391 // The Grand Shuffle 1392 // 1393 // Natives require 1 or 2 extra arguments over the normal ones: the JNIEnv* 1394 // and, if static, the class mirror instead of a receiver. This pretty much 1395 // guarantees that register layout will not match (and x86 doesn't use reg 1396 // parms though amd does). Since the native abi doesn't use register args 1397 // and the java conventions does we don't have to worry about collisions. 1398 // All of our moved are reg->stack or stack->stack. 1399 // We ignore the extra arguments during the shuffle and handle them at the 1400 // last moment. The shuffle is described by the two calling convention 1401 // vectors we have in our possession. We simply walk the java vector to 1402 // get the source locations and the c vector to get the destinations. 1403 1404 int c_arg = method->is_static() ? 2 : 1 ; 1405 1406 // Record rsp-based slot for receiver on stack for non-static methods 1407 int receiver_offset = -1; 1408 1409 // This is a trick. We double the stack slots so we can claim 1410 // the oops in the caller's frame. Since we are sure to have 1411 // more args than the caller doubling is enough to make 1412 // sure we can capture all the incoming oop args from the 1413 // caller. 1414 // 1415 OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/); 1416 1417 // Mark location of rbp, 1418 // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, rbp->as_VMReg()); 1419 1420 // We know that we only have args in at most two integer registers (rcx, rdx). So rax, rbx 1421 // Are free to temporaries if we have to do stack to steck moves. 1422 // All inbound args are referenced based on rbp, and all outbound args via rsp. 1423 1424 for (i = 0; i < total_in_args ; i++, c_arg++ ) { 1425 switch (in_sig_bt[i]) { 1426 case T_ARRAY: 1427 case T_OBJECT: 1428 object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg], 1429 ((i == 0) && (!is_static)), 1430 &receiver_offset); 1431 break; 1432 case T_VOID: 1433 break; 1434 1435 case T_FLOAT: 1436 float_move(masm, in_regs[i], out_regs[c_arg]); 1437 break; 1438 1439 case T_DOUBLE: 1440 assert( i + 1 < total_in_args && 1441 in_sig_bt[i + 1] == T_VOID && 1442 out_sig_bt[c_arg+1] == T_VOID, "bad arg list"); 1443 double_move(masm, in_regs[i], out_regs[c_arg]); 1444 break; 1445 1446 case T_LONG : 1447 long_move(masm, in_regs[i], out_regs[c_arg]); 1448 break; 1449 1450 case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); 1451 1452 default: 1453 simple_move32(masm, in_regs[i], out_regs[c_arg]); 1454 } 1455 } 1456 1457 // Pre-load a static method's oop into rsi. Used both by locking code and 1458 // the normal JNI call code. 1459 if (method->is_static()) { 1460 1461 // load opp into a register 1462 __ movoop(oop_handle_reg, JNIHandles::make_local(Klass::cast(method->method_holder())->java_mirror())); 1463 1464 // Now handlize the static class mirror it's known not-null. 1465 __ movptr(Address(rsp, klass_offset), oop_handle_reg); 1466 map->set_oop(VMRegImpl::stack2reg(klass_slot_offset)); 1467 1468 // Now get the handle 1469 __ lea(oop_handle_reg, Address(rsp, klass_offset)); 1470 // store the klass handle as second argument 1471 __ movptr(Address(rsp, wordSize), oop_handle_reg); 1472 } 1473 1474 // Change state to native (we save the return address in the thread, since it might not 1475 // be pushed on the stack when we do a a stack traversal). It is enough that the pc() 1476 // points into the right code segment. It does not have to be the correct return pc. 1477 // We use the same pc/oopMap repeatedly when we call out 1478 1479 intptr_t the_pc = (intptr_t) __ pc(); 1480 oop_maps->add_gc_map(the_pc - start, map); 1481 1482 __ set_last_Java_frame(thread, rsp, noreg, (address)the_pc); 1483 1484 1485 // We have all of the arguments setup at this point. We must not touch any register 1486 // argument registers at this point (what if we save/restore them there are no oop? 1487 1488 { 1489 SkipIfEqual skip_if(masm, &DTraceMethodProbes, 0); 1490 __ movoop(rax, JNIHandles::make_local(method())); 1491 __ call_VM_leaf( 1492 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 1493 thread, rax); 1494 } 1495 1496 // RedefineClasses() tracing support for obsolete method entry 1497 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { 1498 __ movoop(rax, JNIHandles::make_local(method())); 1499 __ call_VM_leaf( 1500 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 1501 thread, rax); 1502 } 1503 1504 // These are register definitions we need for locking/unlocking 1505 const Register swap_reg = rax; // Must use rax, for cmpxchg instruction 1506 const Register obj_reg = rcx; // Will contain the oop 1507 const Register lock_reg = rdx; // Address of compiler lock object (BasicLock) 1508 1509 Label slow_path_lock; 1510 Label lock_done; 1511 1512 // Lock a synchronized method 1513 if (method->is_synchronized()) { 1514 1515 1516 const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes(); 1517 1518 // Get the handle (the 2nd argument) 1519 __ movptr(oop_handle_reg, Address(rsp, wordSize)); 1520 1521 // Get address of the box 1522 1523 __ lea(lock_reg, Address(rbp, lock_slot_rbp_offset)); 1524 1525 // Load the oop from the handle 1526 __ movptr(obj_reg, Address(oop_handle_reg, 0)); 1527 1528 if (UseBiasedLocking) { 1529 // Note that oop_handle_reg is trashed during this call 1530 __ biased_locking_enter(lock_reg, obj_reg, swap_reg, oop_handle_reg, false, lock_done, &slow_path_lock); 1531 } 1532 1533 // Load immediate 1 into swap_reg %rax, 1534 __ movptr(swap_reg, 1); 1535 1536 // Load (object->mark() | 1) into swap_reg %rax, 1537 __ orptr(swap_reg, Address(obj_reg, 0)); 1538 1539 // Save (object->mark() | 1) into BasicLock's displaced header 1540 __ movptr(Address(lock_reg, mark_word_offset), swap_reg); 1541 1542 if (os::is_MP()) { 1543 __ lock(); 1544 } 1545 1546 // src -> dest iff dest == rax, else rax, <- dest 1547 // *obj_reg = lock_reg iff *obj_reg == rax, else rax, = *(obj_reg) 1548 __ cmpxchgptr(lock_reg, Address(obj_reg, 0)); 1549 __ jcc(Assembler::equal, lock_done); 1550 1551 // Test if the oopMark is an obvious stack pointer, i.e., 1552 // 1) (mark & 3) == 0, and 1553 // 2) rsp <= mark < mark + os::pagesize() 1554 // These 3 tests can be done by evaluating the following 1555 // expression: ((mark - rsp) & (3 - os::vm_page_size())), 1556 // assuming both stack pointer and pagesize have their 1557 // least significant 2 bits clear. 1558 // NOTE: the oopMark is in swap_reg %rax, as the result of cmpxchg 1559 1560 __ subptr(swap_reg, rsp); 1561 __ andptr(swap_reg, 3 - os::vm_page_size()); 1562 1563 // Save the test result, for recursive case, the result is zero 1564 __ movptr(Address(lock_reg, mark_word_offset), swap_reg); 1565 __ jcc(Assembler::notEqual, slow_path_lock); 1566 // Slow path will re-enter here 1567 __ bind(lock_done); 1568 1569 if (UseBiasedLocking) { 1570 // Re-fetch oop_handle_reg as we trashed it above 1571 __ movptr(oop_handle_reg, Address(rsp, wordSize)); 1572 } 1573 } 1574 1575 1576 // Finally just about ready to make the JNI call 1577 1578 1579 // get JNIEnv* which is first argument to native 1580 1581 __ lea(rdx, Address(thread, in_bytes(JavaThread::jni_environment_offset()))); 1582 __ movptr(Address(rsp, 0), rdx); 1583 1584 // Now set thread in native 1585 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native); 1586 1587 __ call(RuntimeAddress(method->native_function())); 1588 1589 // WARNING - on Windows Java Natives use pascal calling convention and pop the 1590 // arguments off of the stack. We could just re-adjust the stack pointer here 1591 // and continue to do SP relative addressing but we instead switch to FP 1592 // relative addressing. 1593 1594 // Unpack native results. 1595 switch (ret_type) { 1596 case T_BOOLEAN: __ c2bool(rax); break; 1597 case T_CHAR : __ andptr(rax, 0xFFFF); break; 1598 case T_BYTE : __ sign_extend_byte (rax); break; 1599 case T_SHORT : __ sign_extend_short(rax); break; 1600 case T_INT : /* nothing to do */ break; 1601 case T_DOUBLE : 1602 case T_FLOAT : 1603 // Result is in st0 we'll save as needed 1604 break; 1605 case T_ARRAY: // Really a handle 1606 case T_OBJECT: // Really a handle 1607 break; // can't de-handlize until after safepoint check 1608 case T_VOID: break; 1609 case T_LONG: break; 1610 default : ShouldNotReachHere(); 1611 } 1612 1613 // Switch thread to "native transition" state before reading the synchronization state. 1614 // This additional state is necessary because reading and testing the synchronization 1615 // state is not atomic w.r.t. GC, as this scenario demonstrates: 1616 // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted. 1617 // VM thread changes sync state to synchronizing and suspends threads for GC. 1618 // Thread A is resumed to finish this native method, but doesn't block here since it 1619 // didn't see any synchronization is progress, and escapes. 1620 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans); 1621 1622 if(os::is_MP()) { 1623 if (UseMembar) { 1624 // Force this write out before the read below 1625 __ membar(Assembler::Membar_mask_bits( 1626 Assembler::LoadLoad | Assembler::LoadStore | 1627 Assembler::StoreLoad | Assembler::StoreStore)); 1628 } else { 1629 // Write serialization page so VM thread can do a pseudo remote membar. 1630 // We use the current thread pointer to calculate a thread specific 1631 // offset to write to within the page. This minimizes bus traffic 1632 // due to cache line collision. 1633 __ serialize_memory(thread, rcx); 1634 } 1635 } 1636 1637 if (AlwaysRestoreFPU) { 1638 // Make sure the control word is correct. 1639 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); 1640 } 1641 1642 // check for safepoint operation in progress and/or pending suspend requests 1643 { Label Continue; 1644 1645 __ cmp32(ExternalAddress((address)SafepointSynchronize::address_of_state()), 1646 SafepointSynchronize::_not_synchronized); 1647 1648 Label L; 1649 __ jcc(Assembler::notEqual, L); 1650 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0); 1651 __ jcc(Assembler::equal, Continue); 1652 __ bind(L); 1653 1654 // Don't use call_VM as it will see a possible pending exception and forward it 1655 // and never return here preventing us from clearing _last_native_pc down below. 1656 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are 1657 // preserved and correspond to the bcp/locals pointers. So we do a runtime call 1658 // by hand. 1659 // 1660 save_native_result(masm, ret_type, stack_slots); 1661 __ push(thread); 1662 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, 1663 JavaThread::check_special_condition_for_native_trans))); 1664 __ increment(rsp, wordSize); 1665 // Restore any method result value 1666 restore_native_result(masm, ret_type, stack_slots); 1667 1668 __ bind(Continue); 1669 } 1670 1671 // change thread state 1672 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java); 1673 1674 Label reguard; 1675 Label reguard_done; 1676 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled); 1677 __ jcc(Assembler::equal, reguard); 1678 1679 // slow path reguard re-enters here 1680 __ bind(reguard_done); 1681 1682 // Handle possible exception (will unlock if necessary) 1683 1684 // native result if any is live 1685 1686 // Unlock 1687 Label slow_path_unlock; 1688 Label unlock_done; 1689 if (method->is_synchronized()) { 1690 1691 Label done; 1692 1693 // Get locked oop from the handle we passed to jni 1694 __ movptr(obj_reg, Address(oop_handle_reg, 0)); 1695 1696 if (UseBiasedLocking) { 1697 __ biased_locking_exit(obj_reg, rbx, done); 1698 } 1699 1700 // Simple recursive lock? 1701 1702 __ cmpptr(Address(rbp, lock_slot_rbp_offset), (int32_t)NULL_WORD); 1703 __ jcc(Assembler::equal, done); 1704 1705 // Must save rax, if if it is live now because cmpxchg must use it 1706 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) { 1707 save_native_result(masm, ret_type, stack_slots); 1708 } 1709 1710 // get old displaced header 1711 __ movptr(rbx, Address(rbp, lock_slot_rbp_offset)); 1712 1713 // get address of the stack lock 1714 __ lea(rax, Address(rbp, lock_slot_rbp_offset)); 1715 1716 // Atomic swap old header if oop still contains the stack lock 1717 if (os::is_MP()) { 1718 __ lock(); 1719 } 1720 1721 // src -> dest iff dest == rax, else rax, <- dest 1722 // *obj_reg = rbx, iff *obj_reg == rax, else rax, = *(obj_reg) 1723 __ cmpxchgptr(rbx, Address(obj_reg, 0)); 1724 __ jcc(Assembler::notEqual, slow_path_unlock); 1725 1726 // slow path re-enters here 1727 __ bind(unlock_done); 1728 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) { 1729 restore_native_result(masm, ret_type, stack_slots); 1730 } 1731 1732 __ bind(done); 1733 1734 } 1735 1736 { 1737 SkipIfEqual skip_if(masm, &DTraceMethodProbes, 0); 1738 // Tell dtrace about this method exit 1739 save_native_result(masm, ret_type, stack_slots); 1740 __ movoop(rax, JNIHandles::make_local(method())); 1741 __ call_VM_leaf( 1742 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 1743 thread, rax); 1744 restore_native_result(masm, ret_type, stack_slots); 1745 } 1746 1747 // We can finally stop using that last_Java_frame we setup ages ago 1748 1749 __ reset_last_Java_frame(thread, false, true); 1750 1751 // Unpack oop result 1752 if (ret_type == T_OBJECT || ret_type == T_ARRAY) { 1753 Label L; 1754 __ cmpptr(rax, (int32_t)NULL_WORD); 1755 __ jcc(Assembler::equal, L); 1756 __ movptr(rax, Address(rax, 0)); 1757 __ bind(L); 1758 __ verify_oop(rax); 1759 } 1760 1761 // reset handle block 1762 __ movptr(rcx, Address(thread, JavaThread::active_handles_offset())); 1763 1764 __ movptr(Address(rcx, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD); 1765 1766 // Any exception pending? 1767 __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD); 1768 __ jcc(Assembler::notEqual, exception_pending); 1769 1770 1771 // no exception, we're almost done 1772 1773 // check that only result value is on FPU stack 1774 __ verify_FPU(ret_type == T_FLOAT || ret_type == T_DOUBLE ? 1 : 0, "native_wrapper normal exit"); 1775 1776 // Fixup floating pointer results so that result looks like a return from a compiled method 1777 if (ret_type == T_FLOAT) { 1778 if (UseSSE >= 1) { 1779 // Pop st0 and store as float and reload into xmm register 1780 __ fstp_s(Address(rbp, -4)); 1781 __ movflt(xmm0, Address(rbp, -4)); 1782 } 1783 } else if (ret_type == T_DOUBLE) { 1784 if (UseSSE >= 2) { 1785 // Pop st0 and store as double and reload into xmm register 1786 __ fstp_d(Address(rbp, -8)); 1787 __ movdbl(xmm0, Address(rbp, -8)); 1788 } 1789 } 1790 1791 // Return 1792 1793 __ leave(); 1794 __ ret(0); 1795 1796 // Unexpected paths are out of line and go here 1797 1798 // Slow path locking & unlocking 1799 if (method->is_synchronized()) { 1800 1801 // BEGIN Slow path lock 1802 1803 __ bind(slow_path_lock); 1804 1805 // has last_Java_frame setup. No exceptions so do vanilla call not call_VM 1806 // args are (oop obj, BasicLock* lock, JavaThread* thread) 1807 __ push(thread); 1808 __ push(lock_reg); 1809 __ push(obj_reg); 1810 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C))); 1811 __ addptr(rsp, 3*wordSize); 1812 1813 #ifdef ASSERT 1814 { Label L; 1815 __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), (int)NULL_WORD); 1816 __ jcc(Assembler::equal, L); 1817 __ stop("no pending exception allowed on exit from monitorenter"); 1818 __ bind(L); 1819 } 1820 #endif 1821 __ jmp(lock_done); 1822 1823 // END Slow path lock 1824 1825 // BEGIN Slow path unlock 1826 __ bind(slow_path_unlock); 1827 1828 // Slow path unlock 1829 1830 if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) { 1831 save_native_result(masm, ret_type, stack_slots); 1832 } 1833 // Save pending exception around call to VM (which contains an EXCEPTION_MARK) 1834 1835 __ pushptr(Address(thread, in_bytes(Thread::pending_exception_offset()))); 1836 __ movptr(Address(thread, in_bytes(Thread::pending_exception_offset())), NULL_WORD); 1837 1838 1839 // should be a peal 1840 // +wordSize because of the push above 1841 __ lea(rax, Address(rbp, lock_slot_rbp_offset)); 1842 __ push(rax); 1843 1844 __ push(obj_reg); 1845 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C))); 1846 __ addptr(rsp, 2*wordSize); 1847 #ifdef ASSERT 1848 { 1849 Label L; 1850 __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD); 1851 __ jcc(Assembler::equal, L); 1852 __ stop("no pending exception allowed on exit complete_monitor_unlocking_C"); 1853 __ bind(L); 1854 } 1855 #endif /* ASSERT */ 1856 1857 __ popptr(Address(thread, in_bytes(Thread::pending_exception_offset()))); 1858 1859 if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) { 1860 restore_native_result(masm, ret_type, stack_slots); 1861 } 1862 __ jmp(unlock_done); 1863 // END Slow path unlock 1864 1865 } 1866 1867 // SLOW PATH Reguard the stack if needed 1868 1869 __ bind(reguard); 1870 save_native_result(masm, ret_type, stack_slots); 1871 { 1872 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); 1873 } 1874 restore_native_result(masm, ret_type, stack_slots); 1875 __ jmp(reguard_done); 1876 1877 1878 // BEGIN EXCEPTION PROCESSING 1879 1880 // Forward the exception 1881 __ bind(exception_pending); 1882 1883 // remove possible return value from FPU register stack 1884 __ empty_FPU_stack(); 1885 1886 // pop our frame 1887 __ leave(); 1888 // and forward the exception 1889 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 1890 1891 __ flush(); 1892 1893 nmethod *nm = nmethod::new_native_nmethod(method, 1894 masm->code(), 1895 vep_offset, 1896 frame_complete, 1897 stack_slots / VMRegImpl::slots_per_word, 1898 (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)), 1899 in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size), 1900 oop_maps); 1901 return nm; 1902 1903 } 1904 1905 #ifdef HAVE_DTRACE_H 1906 // --------------------------------------------------------------------------- 1907 // Generate a dtrace nmethod for a given signature. The method takes arguments 1908 // in the Java compiled code convention, marshals them to the native 1909 // abi and then leaves nops at the position you would expect to call a native 1910 // function. When the probe is enabled the nops are replaced with a trap 1911 // instruction that dtrace inserts and the trace will cause a notification 1912 // to dtrace. 1913 // 1914 // The probes are only able to take primitive types and java/lang/String as 1915 // arguments. No other java types are allowed. Strings are converted to utf8 1916 // strings so that from dtrace point of view java strings are converted to C 1917 // strings. There is an arbitrary fixed limit on the total space that a method 1918 // can use for converting the strings. (256 chars per string in the signature). 1919 // So any java string larger then this is truncated. 1920 1921 nmethod *SharedRuntime::generate_dtrace_nmethod( 1922 MacroAssembler *masm, methodHandle method) { 1923 1924 // generate_dtrace_nmethod is guarded by a mutex so we are sure to 1925 // be single threaded in this method. 1926 assert(AdapterHandlerLibrary_lock->owned_by_self(), "must be"); 1927 1928 // Fill in the signature array, for the calling-convention call. 1929 int total_args_passed = method->size_of_parameters(); 1930 1931 BasicType* in_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed); 1932 VMRegPair *in_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed); 1933 1934 // The signature we are going to use for the trap that dtrace will see 1935 // java/lang/String is converted. We drop "this" and any other object 1936 // is converted to NULL. (A one-slot java/lang/Long object reference 1937 // is converted to a two-slot long, which is why we double the allocation). 1938 BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed * 2); 1939 VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed * 2); 1940 1941 int i=0; 1942 int total_strings = 0; 1943 int first_arg_to_pass = 0; 1944 int total_c_args = 0; 1945 1946 if( !method->is_static() ) { // Pass in receiver first 1947 in_sig_bt[i++] = T_OBJECT; 1948 first_arg_to_pass = 1; 1949 } 1950 1951 // We need to convert the java args to where a native (non-jni) function 1952 // would expect them. To figure out where they go we convert the java 1953 // signature to a C signature. 1954 1955 SignatureStream ss(method->signature()); 1956 for ( ; !ss.at_return_type(); ss.next()) { 1957 BasicType bt = ss.type(); 1958 in_sig_bt[i++] = bt; // Collect remaining bits of signature 1959 out_sig_bt[total_c_args++] = bt; 1960 if( bt == T_OBJECT) { 1961 symbolOop s = ss.as_symbol_or_null(); 1962 if (s == vmSymbols::java_lang_String()) { 1963 total_strings++; 1964 out_sig_bt[total_c_args-1] = T_ADDRESS; 1965 } else if (s == vmSymbols::java_lang_Boolean() || 1966 s == vmSymbols::java_lang_Character() || 1967 s == vmSymbols::java_lang_Byte() || 1968 s == vmSymbols::java_lang_Short() || 1969 s == vmSymbols::java_lang_Integer() || 1970 s == vmSymbols::java_lang_Float()) { 1971 out_sig_bt[total_c_args-1] = T_INT; 1972 } else if (s == vmSymbols::java_lang_Long() || 1973 s == vmSymbols::java_lang_Double()) { 1974 out_sig_bt[total_c_args-1] = T_LONG; 1975 out_sig_bt[total_c_args++] = T_VOID; 1976 } 1977 } else if ( bt == T_LONG || bt == T_DOUBLE ) { 1978 in_sig_bt[i++] = T_VOID; // Longs & doubles take 2 Java slots 1979 out_sig_bt[total_c_args++] = T_VOID; 1980 } 1981 } 1982 1983 assert(i==total_args_passed, "validly parsed signature"); 1984 1985 // Now get the compiled-Java layout as input arguments 1986 int comp_args_on_stack; 1987 comp_args_on_stack = SharedRuntime::java_calling_convention( 1988 in_sig_bt, in_regs, total_args_passed, false); 1989 1990 // Now figure out where the args must be stored and how much stack space 1991 // they require (neglecting out_preserve_stack_slots). 1992 1993 int out_arg_slots; 1994 out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args); 1995 1996 // Calculate the total number of stack slots we will need. 1997 1998 // First count the abi requirement plus all of the outgoing args 1999 int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; 2000 2001 // Now space for the string(s) we must convert 2002 2003 int* string_locs = NEW_RESOURCE_ARRAY(int, total_strings + 1); 2004 for (i = 0; i < total_strings ; i++) { 2005 string_locs[i] = stack_slots; 2006 stack_slots += max_dtrace_string_size / VMRegImpl::stack_slot_size; 2007 } 2008 2009 // + 2 for return address (which we own) and saved rbp, 2010 2011 stack_slots += 2; 2012 2013 // Ok The space we have allocated will look like: 2014 // 2015 // 2016 // FP-> | | 2017 // |---------------------| 2018 // | string[n] | 2019 // |---------------------| <- string_locs[n] 2020 // | string[n-1] | 2021 // |---------------------| <- string_locs[n-1] 2022 // | ... | 2023 // | ... | 2024 // |---------------------| <- string_locs[1] 2025 // | string[0] | 2026 // |---------------------| <- string_locs[0] 2027 // | outbound memory | 2028 // | based arguments | 2029 // | | 2030 // |---------------------| 2031 // | | 2032 // SP-> | out_preserved_slots | 2033 // 2034 // 2035 2036 // Now compute actual number of stack words we need rounding to make 2037 // stack properly aligned. 2038 stack_slots = round_to(stack_slots, 2 * VMRegImpl::slots_per_word); 2039 2040 int stack_size = stack_slots * VMRegImpl::stack_slot_size; 2041 2042 intptr_t start = (intptr_t)__ pc(); 2043 2044 // First thing make an ic check to see if we should even be here 2045 2046 // We are free to use all registers as temps without saving them and 2047 // restoring them except rbp. rbp, is the only callee save register 2048 // as far as the interpreter and the compiler(s) are concerned. 2049 2050 const Register ic_reg = rax; 2051 const Register receiver = rcx; 2052 Label hit; 2053 Label exception_pending; 2054 2055 2056 __ verify_oop(receiver); 2057 __ cmpl(ic_reg, Address(receiver, oopDesc::klass_offset_in_bytes())); 2058 __ jcc(Assembler::equal, hit); 2059 2060 __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); 2061 2062 // verified entry must be aligned for code patching. 2063 // and the first 5 bytes must be in the same cache line 2064 // if we align at 8 then we will be sure 5 bytes are in the same line 2065 __ align(8); 2066 2067 __ bind(hit); 2068 2069 int vep_offset = ((intptr_t)__ pc()) - start; 2070 2071 2072 // The instruction at the verified entry point must be 5 bytes or longer 2073 // because it can be patched on the fly by make_non_entrant. The stack bang 2074 // instruction fits that requirement. 2075 2076 // Generate stack overflow check 2077 2078 2079 if (UseStackBanging) { 2080 if (stack_size <= StackShadowPages*os::vm_page_size()) { 2081 __ bang_stack_with_offset(StackShadowPages*os::vm_page_size()); 2082 } else { 2083 __ movl(rax, stack_size); 2084 __ bang_stack_size(rax, rbx); 2085 } 2086 } else { 2087 // need a 5 byte instruction to allow MT safe patching to non-entrant 2088 __ fat_nop(); 2089 } 2090 2091 assert(((int)__ pc() - start - vep_offset) >= 5, 2092 "valid size for make_non_entrant"); 2093 2094 // Generate a new frame for the wrapper. 2095 __ enter(); 2096 2097 // -2 because return address is already present and so is saved rbp, 2098 if (stack_size - 2*wordSize != 0) { 2099 __ subl(rsp, stack_size - 2*wordSize); 2100 } 2101 2102 // Frame is now completed as far a size and linkage. 2103 2104 int frame_complete = ((intptr_t)__ pc()) - start; 2105 2106 // First thing we do store all the args as if we are doing the call. 2107 // Since the C calling convention is stack based that ensures that 2108 // all the Java register args are stored before we need to convert any 2109 // string we might have. 2110 2111 int sid = 0; 2112 int c_arg, j_arg; 2113 int string_reg = 0; 2114 2115 for (j_arg = first_arg_to_pass, c_arg = 0 ; 2116 j_arg < total_args_passed ; j_arg++, c_arg++ ) { 2117 2118 VMRegPair src = in_regs[j_arg]; 2119 VMRegPair dst = out_regs[c_arg]; 2120 assert(dst.first()->is_stack() || in_sig_bt[j_arg] == T_VOID, 2121 "stack based abi assumed"); 2122 2123 switch (in_sig_bt[j_arg]) { 2124 2125 case T_ARRAY: 2126 case T_OBJECT: 2127 if (out_sig_bt[c_arg] == T_ADDRESS) { 2128 // Any register based arg for a java string after the first 2129 // will be destroyed by the call to get_utf so we store 2130 // the original value in the location the utf string address 2131 // will eventually be stored. 2132 if (src.first()->is_reg()) { 2133 if (string_reg++ != 0) { 2134 simple_move32(masm, src, dst); 2135 } 2136 } 2137 } else if (out_sig_bt[c_arg] == T_INT || out_sig_bt[c_arg] == T_LONG) { 2138 // need to unbox a one-word value 2139 Register in_reg = rax; 2140 if ( src.first()->is_reg() ) { 2141 in_reg = src.first()->as_Register(); 2142 } else { 2143 simple_move32(masm, src, in_reg->as_VMReg()); 2144 } 2145 Label skipUnbox; 2146 __ movl(Address(rsp, reg2offset_out(dst.first())), NULL_WORD); 2147 if ( out_sig_bt[c_arg] == T_LONG ) { 2148 __ movl(Address(rsp, reg2offset_out(dst.second())), NULL_WORD); 2149 } 2150 __ testl(in_reg, in_reg); 2151 __ jcc(Assembler::zero, skipUnbox); 2152 assert(dst.first()->is_stack() && 2153 (!dst.second()->is_valid() || dst.second()->is_stack()), 2154 "value(s) must go into stack slots"); 2155 2156 BasicType bt = out_sig_bt[c_arg]; 2157 int box_offset = java_lang_boxing_object::value_offset_in_bytes(bt); 2158 if ( bt == T_LONG ) { 2159 __ movl(rbx, Address(in_reg, 2160 box_offset + VMRegImpl::stack_slot_size)); 2161 __ movl(Address(rsp, reg2offset_out(dst.second())), rbx); 2162 } 2163 __ movl(in_reg, Address(in_reg, box_offset)); 2164 __ movl(Address(rsp, reg2offset_out(dst.first())), in_reg); 2165 __ bind(skipUnbox); 2166 } else { 2167 // Convert the arg to NULL 2168 __ movl(Address(rsp, reg2offset_out(dst.first())), NULL_WORD); 2169 } 2170 if (out_sig_bt[c_arg] == T_LONG) { 2171 assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); 2172 ++c_arg; // Move over the T_VOID To keep the loop indices in sync 2173 } 2174 break; 2175 2176 case T_VOID: 2177 break; 2178 2179 case T_FLOAT: 2180 float_move(masm, src, dst); 2181 break; 2182 2183 case T_DOUBLE: 2184 assert( j_arg + 1 < total_args_passed && 2185 in_sig_bt[j_arg + 1] == T_VOID, "bad arg list"); 2186 double_move(masm, src, dst); 2187 break; 2188 2189 case T_LONG : 2190 long_move(masm, src, dst); 2191 break; 2192 2193 case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); 2194 2195 default: 2196 simple_move32(masm, src, dst); 2197 } 2198 } 2199 2200 // Now we must convert any string we have to utf8 2201 // 2202 2203 for (sid = 0, j_arg = first_arg_to_pass, c_arg = 0 ; 2204 sid < total_strings ; j_arg++, c_arg++ ) { 2205 2206 if (out_sig_bt[c_arg] == T_ADDRESS) { 2207 2208 Address utf8_addr = Address( 2209 rsp, string_locs[sid++] * VMRegImpl::stack_slot_size); 2210 __ leal(rax, utf8_addr); 2211 2212 // The first string we find might still be in the original java arg 2213 // register 2214 VMReg orig_loc = in_regs[j_arg].first(); 2215 Register string_oop; 2216 2217 // This is where the argument will eventually reside 2218 Address dest = Address(rsp, reg2offset_out(out_regs[c_arg].first())); 2219 2220 if (sid == 1 && orig_loc->is_reg()) { 2221 string_oop = orig_loc->as_Register(); 2222 assert(string_oop != rax, "smashed arg"); 2223 } else { 2224 2225 if (orig_loc->is_reg()) { 2226 // Get the copy of the jls object 2227 __ movl(rcx, dest); 2228 } else { 2229 // arg is still in the original location 2230 __ movl(rcx, Address(rbp, reg2offset_in(orig_loc))); 2231 } 2232 string_oop = rcx; 2233 2234 } 2235 Label nullString; 2236 __ movl(dest, NULL_WORD); 2237 __ testl(string_oop, string_oop); 2238 __ jcc(Assembler::zero, nullString); 2239 2240 // Now we can store the address of the utf string as the argument 2241 __ movl(dest, rax); 2242 2243 // And do the conversion 2244 __ call_VM_leaf(CAST_FROM_FN_PTR( 2245 address, SharedRuntime::get_utf), string_oop, rax); 2246 __ bind(nullString); 2247 } 2248 2249 if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) { 2250 assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); 2251 ++c_arg; // Move over the T_VOID To keep the loop indices in sync 2252 } 2253 } 2254 2255 2256 // Ok now we are done. Need to place the nop that dtrace wants in order to 2257 // patch in the trap 2258 2259 int patch_offset = ((intptr_t)__ pc()) - start; 2260 2261 __ nop(); 2262 2263 2264 // Return 2265 2266 __ leave(); 2267 __ ret(0); 2268 2269 __ flush(); 2270 2271 nmethod *nm = nmethod::new_dtrace_nmethod( 2272 method, masm->code(), vep_offset, patch_offset, frame_complete, 2273 stack_slots / VMRegImpl::slots_per_word); 2274 return nm; 2275 2276 } 2277 2278 #endif // HAVE_DTRACE_H 2279 2280 // this function returns the adjust size (in number of words) to a c2i adapter 2281 // activation for use during deoptimization 2282 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals ) { 2283 return (callee_locals - callee_parameters) * Interpreter::stackElementWords; 2284 } 2285 2286 2287 uint SharedRuntime::out_preserve_stack_slots() { 2288 return 0; 2289 } 2290 2291 2292 //------------------------------generate_deopt_blob---------------------------- 2293 void SharedRuntime::generate_deopt_blob() { 2294 // allocate space for the code 2295 ResourceMark rm; 2296 // setup code generation tools 2297 CodeBuffer buffer("deopt_blob", 1024, 1024); 2298 MacroAssembler* masm = new MacroAssembler(&buffer); 2299 int frame_size_in_words; 2300 OopMap* map = NULL; 2301 // Account for the extra args we place on the stack 2302 // by the time we call fetch_unroll_info 2303 const int additional_words = 2; // deopt kind, thread 2304 2305 OopMapSet *oop_maps = new OopMapSet(); 2306 2307 // ------------- 2308 // This code enters when returning to a de-optimized nmethod. A return 2309 // address has been pushed on the the stack, and return values are in 2310 // registers. 2311 // If we are doing a normal deopt then we were called from the patched 2312 // nmethod from the point we returned to the nmethod. So the return 2313 // address on the stack is wrong by NativeCall::instruction_size 2314 // We will adjust the value to it looks like we have the original return 2315 // address on the stack (like when we eagerly deoptimized). 2316 // In the case of an exception pending with deoptimized then we enter 2317 // with a return address on the stack that points after the call we patched 2318 // into the exception handler. We have the following register state: 2319 // rax,: exception 2320 // rbx,: exception handler 2321 // rdx: throwing pc 2322 // So in this case we simply jam rdx into the useless return address and 2323 // the stack looks just like we want. 2324 // 2325 // At this point we need to de-opt. We save the argument return 2326 // registers. We call the first C routine, fetch_unroll_info(). This 2327 // routine captures the return values and returns a structure which 2328 // describes the current frame size and the sizes of all replacement frames. 2329 // The current frame is compiled code and may contain many inlined 2330 // functions, each with their own JVM state. We pop the current frame, then 2331 // push all the new frames. Then we call the C routine unpack_frames() to 2332 // populate these frames. Finally unpack_frames() returns us the new target 2333 // address. Notice that callee-save registers are BLOWN here; they have 2334 // already been captured in the vframeArray at the time the return PC was 2335 // patched. 2336 address start = __ pc(); 2337 Label cont; 2338 2339 // Prolog for non exception case! 2340 2341 // Save everything in sight. 2342 2343 map = RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false); 2344 // Normal deoptimization 2345 __ push(Deoptimization::Unpack_deopt); 2346 __ jmp(cont); 2347 2348 int reexecute_offset = __ pc() - start; 2349 2350 // Reexecute case 2351 // return address is the pc describes what bci to do re-execute at 2352 2353 // No need to update map as each call to save_live_registers will produce identical oopmap 2354 (void) RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false); 2355 2356 __ push(Deoptimization::Unpack_reexecute); 2357 __ jmp(cont); 2358 2359 int exception_offset = __ pc() - start; 2360 2361 // Prolog for exception case 2362 2363 // all registers are dead at this entry point, except for rax, and 2364 // rdx which contain the exception oop and exception pc 2365 // respectively. Set them in TLS and fall thru to the 2366 // unpack_with_exception_in_tls entry point. 2367 2368 __ get_thread(rdi); 2369 __ movptr(Address(rdi, JavaThread::exception_pc_offset()), rdx); 2370 __ movptr(Address(rdi, JavaThread::exception_oop_offset()), rax); 2371 2372 int exception_in_tls_offset = __ pc() - start; 2373 2374 // new implementation because exception oop is now passed in JavaThread 2375 2376 // Prolog for exception case 2377 // All registers must be preserved because they might be used by LinearScan 2378 // Exceptiop oop and throwing PC are passed in JavaThread 2379 // tos: stack at point of call to method that threw the exception (i.e. only 2380 // args are on the stack, no return address) 2381 2382 // make room on stack for the return address 2383 // It will be patched later with the throwing pc. The correct value is not 2384 // available now because loading it from memory would destroy registers. 2385 __ push(0); 2386 2387 // Save everything in sight. 2388 2389 // No need to update map as each call to save_live_registers will produce identical oopmap 2390 (void) RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false); 2391 2392 // Now it is safe to overwrite any register 2393 2394 // store the correct deoptimization type 2395 __ push(Deoptimization::Unpack_exception); 2396 2397 // load throwing pc from JavaThread and patch it as the return address 2398 // of the current frame. Then clear the field in JavaThread 2399 __ get_thread(rdi); 2400 __ movptr(rdx, Address(rdi, JavaThread::exception_pc_offset())); 2401 __ movptr(Address(rbp, wordSize), rdx); 2402 __ movptr(Address(rdi, JavaThread::exception_pc_offset()), NULL_WORD); 2403 2404 #ifdef ASSERT 2405 // verify that there is really an exception oop in JavaThread 2406 __ movptr(rax, Address(rdi, JavaThread::exception_oop_offset())); 2407 __ verify_oop(rax); 2408 2409 // verify that there is no pending exception 2410 Label no_pending_exception; 2411 __ movptr(rax, Address(rdi, Thread::pending_exception_offset())); 2412 __ testptr(rax, rax); 2413 __ jcc(Assembler::zero, no_pending_exception); 2414 __ stop("must not have pending exception here"); 2415 __ bind(no_pending_exception); 2416 #endif 2417 2418 __ bind(cont); 2419 2420 // Compiled code leaves the floating point stack dirty, empty it. 2421 __ empty_FPU_stack(); 2422 2423 2424 // Call C code. Need thread and this frame, but NOT official VM entry 2425 // crud. We cannot block on this call, no GC can happen. 2426 __ get_thread(rcx); 2427 __ push(rcx); 2428 // fetch_unroll_info needs to call last_java_frame() 2429 __ set_last_Java_frame(rcx, noreg, noreg, NULL); 2430 2431 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info))); 2432 2433 // Need to have an oopmap that tells fetch_unroll_info where to 2434 // find any register it might need. 2435 2436 oop_maps->add_gc_map( __ pc()-start, map); 2437 2438 // Discard arg to fetch_unroll_info 2439 __ pop(rcx); 2440 2441 __ get_thread(rcx); 2442 __ reset_last_Java_frame(rcx, false, false); 2443 2444 // Load UnrollBlock into EDI 2445 __ mov(rdi, rax); 2446 2447 // Move the unpack kind to a safe place in the UnrollBlock because 2448 // we are very short of registers 2449 2450 Address unpack_kind(rdi, Deoptimization::UnrollBlock::unpack_kind_offset_in_bytes()); 2451 // retrieve the deopt kind from where we left it. 2452 __ pop(rax); 2453 __ movl(unpack_kind, rax); // save the unpack_kind value 2454 2455 Label noException; 2456 __ cmpl(rax, Deoptimization::Unpack_exception); // Was exception pending? 2457 __ jcc(Assembler::notEqual, noException); 2458 __ movptr(rax, Address(rcx, JavaThread::exception_oop_offset())); 2459 __ movptr(rdx, Address(rcx, JavaThread::exception_pc_offset())); 2460 __ movptr(Address(rcx, JavaThread::exception_oop_offset()), NULL_WORD); 2461 __ movptr(Address(rcx, JavaThread::exception_pc_offset()), NULL_WORD); 2462 2463 __ verify_oop(rax); 2464 2465 // Overwrite the result registers with the exception results. 2466 __ movptr(Address(rsp, RegisterSaver::raxOffset()*wordSize), rax); 2467 __ movptr(Address(rsp, RegisterSaver::rdxOffset()*wordSize), rdx); 2468 2469 __ bind(noException); 2470 2471 // Stack is back to only having register save data on the stack. 2472 // Now restore the result registers. Everything else is either dead or captured 2473 // in the vframeArray. 2474 2475 RegisterSaver::restore_result_registers(masm); 2476 2477 // Non standard control word may be leaked out through a safepoint blob, and we can 2478 // deopt at a poll point with the non standard control word. However, we should make 2479 // sure the control word is correct after restore_result_registers. 2480 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); 2481 2482 // All of the register save area has been popped of the stack. Only the 2483 // return address remains. 2484 2485 // Pop all the frames we must move/replace. 2486 // 2487 // Frame picture (youngest to oldest) 2488 // 1: self-frame (no frame link) 2489 // 2: deopting frame (no frame link) 2490 // 3: caller of deopting frame (could be compiled/interpreted). 2491 // 2492 // Note: by leaving the return address of self-frame on the stack 2493 // and using the size of frame 2 to adjust the stack 2494 // when we are done the return to frame 3 will still be on the stack. 2495 2496 // Pop deoptimized frame 2497 __ addptr(rsp, Address(rdi,Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes())); 2498 2499 // sp should be pointing at the return address to the caller (3) 2500 2501 // Stack bang to make sure there's enough room for these interpreter frames. 2502 if (UseStackBanging) { 2503 __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes())); 2504 __ bang_stack_size(rbx, rcx); 2505 } 2506 2507 // Load array of frame pcs into ECX 2508 __ movptr(rcx,Address(rdi,Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes())); 2509 2510 __ pop(rsi); // trash the old pc 2511 2512 // Load array of frame sizes into ESI 2513 __ movptr(rsi,Address(rdi,Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes())); 2514 2515 Address counter(rdi, Deoptimization::UnrollBlock::counter_temp_offset_in_bytes()); 2516 2517 __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes())); 2518 __ movl(counter, rbx); 2519 2520 // Pick up the initial fp we should save 2521 __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_fp_offset_in_bytes())); 2522 2523 // Now adjust the caller's stack to make up for the extra locals 2524 // but record the original sp so that we can save it in the skeletal interpreter 2525 // frame and the stack walking of interpreter_sender will get the unextended sp 2526 // value and not the "real" sp value. 2527 2528 Address sp_temp(rdi, Deoptimization::UnrollBlock::sender_sp_temp_offset_in_bytes()); 2529 __ movptr(sp_temp, rsp); 2530 __ movl2ptr(rbx, Address(rdi, Deoptimization::UnrollBlock::caller_adjustment_offset_in_bytes())); 2531 __ subptr(rsp, rbx); 2532 2533 // Push interpreter frames in a loop 2534 Label loop; 2535 __ bind(loop); 2536 __ movptr(rbx, Address(rsi, 0)); // Load frame size 2537 #ifdef CC_INTERP 2538 __ subptr(rbx, 4*wordSize); // we'll push pc and ebp by hand and 2539 #ifdef ASSERT 2540 __ push(0xDEADDEAD); // Make a recognizable pattern 2541 __ push(0xDEADDEAD); 2542 #else /* ASSERT */ 2543 __ subptr(rsp, 2*wordSize); // skip the "static long no_param" 2544 #endif /* ASSERT */ 2545 #else /* CC_INTERP */ 2546 __ subptr(rbx, 2*wordSize); // we'll push pc and rbp, by hand 2547 #endif /* CC_INTERP */ 2548 __ pushptr(Address(rcx, 0)); // save return address 2549 __ enter(); // save old & set new rbp, 2550 __ subptr(rsp, rbx); // Prolog! 2551 __ movptr(rbx, sp_temp); // sender's sp 2552 #ifdef CC_INTERP 2553 __ movptr(Address(rbp, 2554 -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))), 2555 rbx); // Make it walkable 2556 #else /* CC_INTERP */ 2557 // This value is corrected by layout_activation_impl 2558 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 2559 __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), rbx); // Make it walkable 2560 #endif /* CC_INTERP */ 2561 __ movptr(sp_temp, rsp); // pass to next frame 2562 __ addptr(rsi, wordSize); // Bump array pointer (sizes) 2563 __ addptr(rcx, wordSize); // Bump array pointer (pcs) 2564 __ decrementl(counter); // decrement counter 2565 __ jcc(Assembler::notZero, loop); 2566 __ pushptr(Address(rcx, 0)); // save final return address 2567 2568 // Re-push self-frame 2569 __ enter(); // save old & set new rbp, 2570 2571 // Return address and rbp, are in place 2572 // We'll push additional args later. Just allocate a full sized 2573 // register save area 2574 __ subptr(rsp, (frame_size_in_words-additional_words - 2) * wordSize); 2575 2576 // Restore frame locals after moving the frame 2577 __ movptr(Address(rsp, RegisterSaver::raxOffset()*wordSize), rax); 2578 __ movptr(Address(rsp, RegisterSaver::rdxOffset()*wordSize), rdx); 2579 __ fstp_d(Address(rsp, RegisterSaver::fpResultOffset()*wordSize)); // Pop float stack and store in local 2580 if( UseSSE>=2 ) __ movdbl(Address(rsp, RegisterSaver::xmm0Offset()*wordSize), xmm0); 2581 if( UseSSE==1 ) __ movflt(Address(rsp, RegisterSaver::xmm0Offset()*wordSize), xmm0); 2582 2583 // Set up the args to unpack_frame 2584 2585 __ pushl(unpack_kind); // get the unpack_kind value 2586 __ get_thread(rcx); 2587 __ push(rcx); 2588 2589 // set last_Java_sp, last_Java_fp 2590 __ set_last_Java_frame(rcx, noreg, rbp, NULL); 2591 2592 // Call C code. Need thread but NOT official VM entry 2593 // crud. We cannot block on this call, no GC can happen. Call should 2594 // restore return values to their stack-slots with the new SP. 2595 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames))); 2596 // Set an oopmap for the call site 2597 oop_maps->add_gc_map( __ pc()-start, new OopMap( frame_size_in_words, 0 )); 2598 2599 // rax, contains the return result type 2600 __ push(rax); 2601 2602 __ get_thread(rcx); 2603 __ reset_last_Java_frame(rcx, false, false); 2604 2605 // Collect return values 2606 __ movptr(rax,Address(rsp, (RegisterSaver::raxOffset() + additional_words + 1)*wordSize)); 2607 __ movptr(rdx,Address(rsp, (RegisterSaver::rdxOffset() + additional_words + 1)*wordSize)); 2608 2609 // Clear floating point stack before returning to interpreter 2610 __ empty_FPU_stack(); 2611 2612 // Check if we should push the float or double return value. 2613 Label results_done, yes_double_value; 2614 __ cmpl(Address(rsp, 0), T_DOUBLE); 2615 __ jcc (Assembler::zero, yes_double_value); 2616 __ cmpl(Address(rsp, 0), T_FLOAT); 2617 __ jcc (Assembler::notZero, results_done); 2618 2619 // return float value as expected by interpreter 2620 if( UseSSE>=1 ) __ movflt(xmm0, Address(rsp, (RegisterSaver::xmm0Offset() + additional_words + 1)*wordSize)); 2621 else __ fld_d(Address(rsp, (RegisterSaver::fpResultOffset() + additional_words + 1)*wordSize)); 2622 __ jmp(results_done); 2623 2624 // return double value as expected by interpreter 2625 __ bind(yes_double_value); 2626 if( UseSSE>=2 ) __ movdbl(xmm0, Address(rsp, (RegisterSaver::xmm0Offset() + additional_words + 1)*wordSize)); 2627 else __ fld_d(Address(rsp, (RegisterSaver::fpResultOffset() + additional_words + 1)*wordSize)); 2628 2629 __ bind(results_done); 2630 2631 // Pop self-frame. 2632 __ leave(); // Epilog! 2633 2634 // Jump to interpreter 2635 __ ret(0); 2636 2637 // ------------- 2638 // make sure all code is generated 2639 masm->flush(); 2640 2641 _deopt_blob = DeoptimizationBlob::create( &buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words); 2642 _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset); 2643 } 2644 2645 2646 #ifdef COMPILER2 2647 //------------------------------generate_uncommon_trap_blob-------------------- 2648 void SharedRuntime::generate_uncommon_trap_blob() { 2649 // allocate space for the code 2650 ResourceMark rm; 2651 // setup code generation tools 2652 CodeBuffer buffer("uncommon_trap_blob", 512, 512); 2653 MacroAssembler* masm = new MacroAssembler(&buffer); 2654 2655 enum frame_layout { 2656 arg0_off, // thread sp + 0 // Arg location for 2657 arg1_off, // unloaded_class_index sp + 1 // calling C 2658 // The frame sender code expects that rbp will be in the "natural" place and 2659 // will override any oopMap setting for it. We must therefore force the layout 2660 // so that it agrees with the frame sender code. 2661 rbp_off, // callee saved register sp + 2 2662 return_off, // slot for return address sp + 3 2663 framesize 2664 }; 2665 2666 address start = __ pc(); 2667 // Push self-frame. 2668 __ subptr(rsp, return_off*wordSize); // Epilog! 2669 2670 // rbp, is an implicitly saved callee saved register (i.e. the calling 2671 // convention will save restore it in prolog/epilog) Other than that 2672 // there are no callee save registers no that adapter frames are gone. 2673 __ movptr(Address(rsp, rbp_off*wordSize), rbp); 2674 2675 // Clear the floating point exception stack 2676 __ empty_FPU_stack(); 2677 2678 // set last_Java_sp 2679 __ get_thread(rdx); 2680 __ set_last_Java_frame(rdx, noreg, noreg, NULL); 2681 2682 // Call C code. Need thread but NOT official VM entry 2683 // crud. We cannot block on this call, no GC can happen. Call should 2684 // capture callee-saved registers as well as return values. 2685 __ movptr(Address(rsp, arg0_off*wordSize), rdx); 2686 // argument already in ECX 2687 __ movl(Address(rsp, arg1_off*wordSize),rcx); 2688 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap))); 2689 2690 // Set an oopmap for the call site 2691 OopMapSet *oop_maps = new OopMapSet(); 2692 OopMap* map = new OopMap( framesize, 0 ); 2693 // No oopMap for rbp, it is known implicitly 2694 2695 oop_maps->add_gc_map( __ pc()-start, map); 2696 2697 __ get_thread(rcx); 2698 2699 __ reset_last_Java_frame(rcx, false, false); 2700 2701 // Load UnrollBlock into EDI 2702 __ movptr(rdi, rax); 2703 2704 // Pop all the frames we must move/replace. 2705 // 2706 // Frame picture (youngest to oldest) 2707 // 1: self-frame (no frame link) 2708 // 2: deopting frame (no frame link) 2709 // 3: caller of deopting frame (could be compiled/interpreted). 2710 2711 // Pop self-frame. We have no frame, and must rely only on EAX and ESP. 2712 __ addptr(rsp,(framesize-1)*wordSize); // Epilog! 2713 2714 // Pop deoptimized frame 2715 __ movl2ptr(rcx, Address(rdi,Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes())); 2716 __ addptr(rsp, rcx); 2717 2718 // sp should be pointing at the return address to the caller (3) 2719 2720 // Stack bang to make sure there's enough room for these interpreter frames. 2721 if (UseStackBanging) { 2722 __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes())); 2723 __ bang_stack_size(rbx, rcx); 2724 } 2725 2726 2727 // Load array of frame pcs into ECX 2728 __ movl(rcx,Address(rdi,Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes())); 2729 2730 __ pop(rsi); // trash the pc 2731 2732 // Load array of frame sizes into ESI 2733 __ movptr(rsi,Address(rdi,Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes())); 2734 2735 Address counter(rdi, Deoptimization::UnrollBlock::counter_temp_offset_in_bytes()); 2736 2737 __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes())); 2738 __ movl(counter, rbx); 2739 2740 // Pick up the initial fp we should save 2741 __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_fp_offset_in_bytes())); 2742 2743 // Now adjust the caller's stack to make up for the extra locals 2744 // but record the original sp so that we can save it in the skeletal interpreter 2745 // frame and the stack walking of interpreter_sender will get the unextended sp 2746 // value and not the "real" sp value. 2747 2748 Address sp_temp(rdi, Deoptimization::UnrollBlock::sender_sp_temp_offset_in_bytes()); 2749 __ movptr(sp_temp, rsp); 2750 __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::caller_adjustment_offset_in_bytes())); 2751 __ subptr(rsp, rbx); 2752 2753 // Push interpreter frames in a loop 2754 Label loop; 2755 __ bind(loop); 2756 __ movptr(rbx, Address(rsi, 0)); // Load frame size 2757 #ifdef CC_INTERP 2758 __ subptr(rbx, 4*wordSize); // we'll push pc and ebp by hand and 2759 #ifdef ASSERT 2760 __ push(0xDEADDEAD); // Make a recognizable pattern 2761 __ push(0xDEADDEAD); // (parm to RecursiveInterpreter...) 2762 #else /* ASSERT */ 2763 __ subptr(rsp, 2*wordSize); // skip the "static long no_param" 2764 #endif /* ASSERT */ 2765 #else /* CC_INTERP */ 2766 __ subptr(rbx, 2*wordSize); // we'll push pc and rbp, by hand 2767 #endif /* CC_INTERP */ 2768 __ pushptr(Address(rcx, 0)); // save return address 2769 __ enter(); // save old & set new rbp, 2770 __ subptr(rsp, rbx); // Prolog! 2771 __ movptr(rbx, sp_temp); // sender's sp 2772 #ifdef CC_INTERP 2773 __ movptr(Address(rbp, 2774 -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))), 2775 rbx); // Make it walkable 2776 #else /* CC_INTERP */ 2777 // This value is corrected by layout_activation_impl 2778 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD ); 2779 __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), rbx); // Make it walkable 2780 #endif /* CC_INTERP */ 2781 __ movptr(sp_temp, rsp); // pass to next frame 2782 __ addptr(rsi, wordSize); // Bump array pointer (sizes) 2783 __ addptr(rcx, wordSize); // Bump array pointer (pcs) 2784 __ decrementl(counter); // decrement counter 2785 __ jcc(Assembler::notZero, loop); 2786 __ pushptr(Address(rcx, 0)); // save final return address 2787 2788 // Re-push self-frame 2789 __ enter(); // save old & set new rbp, 2790 __ subptr(rsp, (framesize-2) * wordSize); // Prolog! 2791 2792 2793 // set last_Java_sp, last_Java_fp 2794 __ get_thread(rdi); 2795 __ set_last_Java_frame(rdi, noreg, rbp, NULL); 2796 2797 // Call C code. Need thread but NOT official VM entry 2798 // crud. We cannot block on this call, no GC can happen. Call should 2799 // restore return values to their stack-slots with the new SP. 2800 __ movptr(Address(rsp,arg0_off*wordSize),rdi); 2801 __ movl(Address(rsp,arg1_off*wordSize), Deoptimization::Unpack_uncommon_trap); 2802 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames))); 2803 // Set an oopmap for the call site 2804 oop_maps->add_gc_map( __ pc()-start, new OopMap( framesize, 0 ) ); 2805 2806 __ get_thread(rdi); 2807 __ reset_last_Java_frame(rdi, true, false); 2808 2809 // Pop self-frame. 2810 __ leave(); // Epilog! 2811 2812 // Jump to interpreter 2813 __ ret(0); 2814 2815 // ------------- 2816 // make sure all code is generated 2817 masm->flush(); 2818 2819 _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, oop_maps, framesize); 2820 } 2821 #endif // COMPILER2 2822 2823 //------------------------------generate_handler_blob------ 2824 // 2825 // Generate a special Compile2Runtime blob that saves all registers, 2826 // setup oopmap, and calls safepoint code to stop the compiled code for 2827 // a safepoint. 2828 // 2829 static SafepointBlob* generate_handler_blob(address call_ptr, bool cause_return) { 2830 2831 // Account for thread arg in our frame 2832 const int additional_words = 1; 2833 int frame_size_in_words; 2834 2835 assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before"); 2836 2837 ResourceMark rm; 2838 OopMapSet *oop_maps = new OopMapSet(); 2839 OopMap* map; 2840 2841 // allocate space for the code 2842 // setup code generation tools 2843 CodeBuffer buffer("handler_blob", 1024, 512); 2844 MacroAssembler* masm = new MacroAssembler(&buffer); 2845 2846 const Register java_thread = rdi; // callee-saved for VC++ 2847 address start = __ pc(); 2848 address call_pc = NULL; 2849 2850 // If cause_return is true we are at a poll_return and there is 2851 // the return address on the stack to the caller on the nmethod 2852 // that is safepoint. We can leave this return on the stack and 2853 // effectively complete the return and safepoint in the caller. 2854 // Otherwise we push space for a return address that the safepoint 2855 // handler will install later to make the stack walking sensible. 2856 if( !cause_return ) 2857 __ push(rbx); // Make room for return address (or push it again) 2858 2859 map = RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false); 2860 2861 // The following is basically a call_VM. However, we need the precise 2862 // address of the call in order to generate an oopmap. Hence, we do all the 2863 // work ourselves. 2864 2865 // Push thread argument and setup last_Java_sp 2866 __ get_thread(java_thread); 2867 __ push(java_thread); 2868 __ set_last_Java_frame(java_thread, noreg, noreg, NULL); 2869 2870 // if this was not a poll_return then we need to correct the return address now. 2871 if( !cause_return ) { 2872 __ movptr(rax, Address(java_thread, JavaThread::saved_exception_pc_offset())); 2873 __ movptr(Address(rbp, wordSize), rax); 2874 } 2875 2876 // do the call 2877 __ call(RuntimeAddress(call_ptr)); 2878 2879 // Set an oopmap for the call site. This oopmap will map all 2880 // oop-registers and debug-info registers as callee-saved. This 2881 // will allow deoptimization at this safepoint to find all possible 2882 // debug-info recordings, as well as let GC find all oops. 2883 2884 oop_maps->add_gc_map( __ pc() - start, map); 2885 2886 // Discard arg 2887 __ pop(rcx); 2888 2889 Label noException; 2890 2891 // Clear last_Java_sp again 2892 __ get_thread(java_thread); 2893 __ reset_last_Java_frame(java_thread, false, false); 2894 2895 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 2896 __ jcc(Assembler::equal, noException); 2897 2898 // Exception pending 2899 2900 RegisterSaver::restore_live_registers(masm); 2901 2902 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 2903 2904 __ bind(noException); 2905 2906 // Normal exit, register restoring and exit 2907 RegisterSaver::restore_live_registers(masm); 2908 2909 __ ret(0); 2910 2911 // make sure all code is generated 2912 masm->flush(); 2913 2914 // Fill-out other meta info 2915 return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words); 2916 } 2917 2918 // 2919 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss 2920 // 2921 // Generate a stub that calls into vm to find out the proper destination 2922 // of a java call. All the argument registers are live at this point 2923 // but since this is generic code we don't know what they are and the caller 2924 // must do any gc of the args. 2925 // 2926 static RuntimeStub* generate_resolve_blob(address destination, const char* name) { 2927 assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before"); 2928 2929 // allocate space for the code 2930 ResourceMark rm; 2931 2932 CodeBuffer buffer(name, 1000, 512); 2933 MacroAssembler* masm = new MacroAssembler(&buffer); 2934 2935 int frame_size_words; 2936 enum frame_layout { 2937 thread_off, 2938 extra_words }; 2939 2940 OopMapSet *oop_maps = new OopMapSet(); 2941 OopMap* map = NULL; 2942 2943 int start = __ offset(); 2944 2945 map = RegisterSaver::save_live_registers(masm, extra_words, &frame_size_words); 2946 2947 int frame_complete = __ offset(); 2948 2949 const Register thread = rdi; 2950 __ get_thread(rdi); 2951 2952 __ push(thread); 2953 __ set_last_Java_frame(thread, noreg, rbp, NULL); 2954 2955 __ call(RuntimeAddress(destination)); 2956 2957 2958 // Set an oopmap for the call site. 2959 // We need this not only for callee-saved registers, but also for volatile 2960 // registers that the compiler might be keeping live across a safepoint. 2961 2962 oop_maps->add_gc_map( __ offset() - start, map); 2963 2964 // rax, contains the address we are going to jump to assuming no exception got installed 2965 2966 __ addptr(rsp, wordSize); 2967 2968 // clear last_Java_sp 2969 __ reset_last_Java_frame(thread, true, false); 2970 // check for pending exceptions 2971 Label pending; 2972 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 2973 __ jcc(Assembler::notEqual, pending); 2974 2975 // get the returned methodOop 2976 __ movptr(rbx, Address(thread, JavaThread::vm_result_offset())); 2977 __ movptr(Address(rsp, RegisterSaver::rbx_offset() * wordSize), rbx); 2978 2979 __ movptr(Address(rsp, RegisterSaver::rax_offset() * wordSize), rax); 2980 2981 RegisterSaver::restore_live_registers(masm); 2982 2983 // We are back the the original state on entry and ready to go. 2984 2985 __ jmp(rax); 2986 2987 // Pending exception after the safepoint 2988 2989 __ bind(pending); 2990 2991 RegisterSaver::restore_live_registers(masm); 2992 2993 // exception pending => remove activation and forward to exception handler 2994 2995 __ get_thread(thread); 2996 __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD); 2997 __ movptr(rax, Address(thread, Thread::pending_exception_offset())); 2998 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 2999 3000 // ------------- 3001 // make sure all code is generated 3002 masm->flush(); 3003 3004 // return the blob 3005 // frame_size_words or bytes?? 3006 return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_words, oop_maps, true); 3007 } 3008 3009 void SharedRuntime::generate_stubs() { 3010 3011 _wrong_method_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method), 3012 "wrong_method_stub"); 3013 3014 _ic_miss_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method_ic_miss), 3015 "ic_miss_stub"); 3016 3017 _resolve_opt_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_opt_virtual_call_C), 3018 "resolve_opt_virtual_call"); 3019 3020 _resolve_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_virtual_call_C), 3021 "resolve_virtual_call"); 3022 3023 _resolve_static_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_static_call_C), 3024 "resolve_static_call"); 3025 3026 _polling_page_safepoint_handler_blob = 3027 generate_handler_blob(CAST_FROM_FN_PTR(address, 3028 SafepointSynchronize::handle_polling_page_exception), false); 3029 3030 _polling_page_return_handler_blob = 3031 generate_handler_blob(CAST_FROM_FN_PTR(address, 3032 SafepointSynchronize::handle_polling_page_exception), true); 3033 3034 generate_deopt_blob(); 3035 #ifdef COMPILER2 3036 generate_uncommon_trap_blob(); 3037 #endif // COMPILER2 3038 }