1 /* 2 * Copyright (c) 1999, 2023, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "asm/assembler.hpp" 27 #include "c1/c1_Defs.hpp" 28 #include "c1/c1_FrameMap.hpp" 29 #include "c1/c1_MacroAssembler.hpp" 30 #include "c1/c1_Runtime1.hpp" 31 #include "ci/ciUtilities.hpp" 32 #include "compiler/compilerDefinitions.inline.hpp" 33 #include "compiler/oopMap.hpp" 34 #include "gc/shared/cardTable.hpp" 35 #include "gc/shared/cardTableBarrierSet.hpp" 36 #include "gc/shared/collectedHeap.hpp" 37 #include "gc/shared/tlab_globals.hpp" 38 #include "interpreter/interpreter.hpp" 39 #include "memory/universe.hpp" 40 #include "nativeInst_x86.hpp" 41 #include "oops/oop.inline.hpp" 42 #include "prims/jvmtiExport.hpp" 43 #include "register_x86.hpp" 44 #include "runtime/sharedRuntime.hpp" 45 #include "runtime/signature.hpp" 46 #include "runtime/stubRoutines.hpp" 47 #include "runtime/vframeArray.hpp" 48 #include "utilities/macros.hpp" 49 #include "vmreg_x86.inline.hpp" 50 51 // Implementation of StubAssembler 52 53 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, int args_size) { 54 // setup registers 55 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread); // is callee-saved register (Visual C++ calling conventions) 56 assert(!(oop_result1->is_valid() || metadata_result->is_valid()) || oop_result1 != metadata_result, "registers must be different"); 57 assert(oop_result1 != thread && metadata_result != thread, "registers must be different"); 58 assert(args_size >= 0, "illegal args_size"); 59 bool align_stack = false; 60 #ifdef _LP64 61 // At a method handle call, the stack may not be properly aligned 62 // when returning with an exception. 63 align_stack = (stub_id() == Runtime1::handle_exception_from_callee_id); 64 #endif 65 66 #ifdef _LP64 67 mov(c_rarg0, thread); 68 set_num_rt_args(0); // Nothing on stack 69 #else 70 set_num_rt_args(1 + args_size); 71 72 // push java thread (becomes first argument of C function) 73 get_thread(thread); 74 push(thread); 75 #endif // _LP64 76 77 int call_offset = -1; 78 if (!align_stack) { 79 set_last_Java_frame(thread, noreg, rbp, nullptr, rscratch1); 80 } else { 81 address the_pc = pc(); 82 call_offset = offset(); 83 set_last_Java_frame(thread, noreg, rbp, the_pc, rscratch1); 84 andptr(rsp, -(StackAlignmentInBytes)); // Align stack 85 } 86 87 // do the call 88 call(RuntimeAddress(entry)); 89 if (!align_stack) { 90 call_offset = offset(); 91 } 92 // verify callee-saved register 93 #ifdef ASSERT 94 guarantee(thread != rax, "change this code"); 95 push(rax); 96 { Label L; 97 get_thread(rax); 98 cmpptr(thread, rax); 99 jcc(Assembler::equal, L); 100 int3(); 101 stop("StubAssembler::call_RT: rdi not callee saved?"); 102 bind(L); 103 } 104 pop(rax); 105 #endif 106 reset_last_Java_frame(thread, true); 107 108 // discard thread and arguments 109 NOT_LP64(addptr(rsp, num_rt_args()*BytesPerWord)); 110 111 // check for pending exceptions 112 { Label L; 113 cmpptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD); 114 jcc(Assembler::equal, L); 115 // exception pending => remove activation and forward to exception handler 116 movptr(rax, Address(thread, Thread::pending_exception_offset())); 117 // make sure that the vm_results are cleared 118 if (oop_result1->is_valid()) { 119 movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD); 120 } 121 if (metadata_result->is_valid()) { 122 movptr(Address(thread, JavaThread::vm_result_2_offset()), NULL_WORD); 123 } 124 if (frame_size() == no_frame_size) { 125 leave(); 126 jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 127 } else if (_stub_id == Runtime1::forward_exception_id) { 128 should_not_reach_here(); 129 } else { 130 jump(RuntimeAddress(Runtime1::entry_for(Runtime1::forward_exception_id))); 131 } 132 bind(L); 133 } 134 // get oop results if there are any and reset the values in the thread 135 if (oop_result1->is_valid()) { 136 get_vm_result(oop_result1, thread); 137 } 138 if (metadata_result->is_valid()) { 139 get_vm_result_2(metadata_result, thread); 140 } 141 142 assert(call_offset >= 0, "Should be set"); 143 return call_offset; 144 } 145 146 147 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1) { 148 #ifdef _LP64 149 mov(c_rarg1, arg1); 150 #else 151 push(arg1); 152 #endif // _LP64 153 return call_RT(oop_result1, metadata_result, entry, 1); 154 } 155 156 157 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2) { 158 #ifdef _LP64 159 if (c_rarg1 == arg2) { 160 if (c_rarg2 == arg1) { 161 xchgq(arg1, arg2); 162 } else { 163 mov(c_rarg2, arg2); 164 mov(c_rarg1, arg1); 165 } 166 } else { 167 mov(c_rarg1, arg1); 168 mov(c_rarg2, arg2); 169 } 170 #else 171 push(arg2); 172 push(arg1); 173 #endif // _LP64 174 return call_RT(oop_result1, metadata_result, entry, 2); 175 } 176 177 178 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2, Register arg3) { 179 #ifdef _LP64 180 // if there is any conflict use the stack 181 if (arg1 == c_rarg2 || arg1 == c_rarg3 || 182 arg2 == c_rarg1 || arg2 == c_rarg3 || 183 arg3 == c_rarg1 || arg3 == c_rarg2) { 184 push(arg3); 185 push(arg2); 186 push(arg1); 187 pop(c_rarg1); 188 pop(c_rarg2); 189 pop(c_rarg3); 190 } else { 191 mov(c_rarg1, arg1); 192 mov(c_rarg2, arg2); 193 mov(c_rarg3, arg3); 194 } 195 #else 196 push(arg3); 197 push(arg2); 198 push(arg1); 199 #endif // _LP64 200 return call_RT(oop_result1, metadata_result, entry, 3); 201 } 202 203 204 // Implementation of StubFrame 205 206 class StubFrame: public StackObj { 207 private: 208 StubAssembler* _sasm; 209 bool _use_pop_on_epilog; 210 211 public: 212 StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments, bool use_pop_on_epilog = false); 213 void load_argument(int offset_in_words, Register reg); 214 215 ~StubFrame(); 216 }; 217 218 void StubAssembler::prologue(const char* name, bool must_gc_arguments) { 219 set_info(name, must_gc_arguments); 220 enter(); 221 } 222 223 void StubAssembler::epilogue(bool use_pop) { 224 use_pop ? pop(rbp) : leave(); 225 ret(0); 226 } 227 228 #define __ _sasm-> 229 230 StubFrame::StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments, bool use_pop_on_epilog) { 231 _sasm = sasm; 232 _use_pop_on_epilog = use_pop_on_epilog; 233 __ prologue(name, must_gc_arguments); 234 } 235 236 // load parameters that were stored with LIR_Assembler::store_parameter 237 // Note: offsets for store_parameter and load_argument must match 238 void StubFrame::load_argument(int offset_in_words, Register reg) { 239 __ load_parameter(offset_in_words, reg); 240 } 241 242 243 StubFrame::~StubFrame() { 244 __ epilogue(_use_pop_on_epilog); 245 } 246 247 #undef __ 248 249 250 // Implementation of Runtime1 251 252 const int float_regs_as_doubles_size_in_slots = pd_nof_fpu_regs_frame_map * 2; 253 const int xmm_regs_as_doubles_size_in_slots = FrameMap::nof_xmm_regs * 2; 254 255 // Stack layout for saving/restoring all the registers needed during a runtime 256 // call (this includes deoptimization) 257 // Note: note that users of this frame may well have arguments to some runtime 258 // while these values are on the stack. These positions neglect those arguments 259 // but the code in save_live_registers will take the argument count into 260 // account. 261 // 262 #ifdef _LP64 263 #define SLOT2(x) x, 264 #define SLOT_PER_WORD 2 265 #else 266 #define SLOT2(x) 267 #define SLOT_PER_WORD 1 268 #endif // _LP64 269 270 enum reg_save_layout { 271 // 64bit needs to keep stack 16 byte aligned. So we add some alignment dummies to make that 272 // happen and will assert if the stack size we create is misaligned 273 #ifdef _LP64 274 align_dummy_0, align_dummy_1, 275 #endif // _LP64 276 #ifdef _WIN64 277 // Windows always allocates space for it's argument registers (see 278 // frame::arg_reg_save_area_bytes). 279 arg_reg_save_1, arg_reg_save_1H, // 0, 4 280 arg_reg_save_2, arg_reg_save_2H, // 8, 12 281 arg_reg_save_3, arg_reg_save_3H, // 16, 20 282 arg_reg_save_4, arg_reg_save_4H, // 24, 28 283 #endif // _WIN64 284 xmm_regs_as_doubles_off, // 32 285 float_regs_as_doubles_off = xmm_regs_as_doubles_off + xmm_regs_as_doubles_size_in_slots, // 160 286 fpu_state_off = float_regs_as_doubles_off + float_regs_as_doubles_size_in_slots, // 224 287 // fpu_state_end_off is exclusive 288 fpu_state_end_off = fpu_state_off + (FPUStateSizeInWords / SLOT_PER_WORD), // 352 289 marker = fpu_state_end_off, SLOT2(markerH) // 352, 356 290 extra_space_offset, // 360 291 #ifdef _LP64 292 r15_off = extra_space_offset, r15H_off, // 360, 364 293 r14_off, r14H_off, // 368, 372 294 r13_off, r13H_off, // 376, 380 295 r12_off, r12H_off, // 384, 388 296 r11_off, r11H_off, // 392, 396 297 r10_off, r10H_off, // 400, 404 298 r9_off, r9H_off, // 408, 412 299 r8_off, r8H_off, // 416, 420 300 rdi_off, rdiH_off, // 424, 428 301 #else 302 rdi_off = extra_space_offset, 303 #endif // _LP64 304 rsi_off, SLOT2(rsiH_off) // 432, 436 305 rbp_off, SLOT2(rbpH_off) // 440, 444 306 rsp_off, SLOT2(rspH_off) // 448, 452 307 rbx_off, SLOT2(rbxH_off) // 456, 460 308 rdx_off, SLOT2(rdxH_off) // 464, 468 309 rcx_off, SLOT2(rcxH_off) // 472, 476 310 rax_off, SLOT2(raxH_off) // 480, 484 311 saved_rbp_off, SLOT2(saved_rbpH_off) // 488, 492 312 return_off, SLOT2(returnH_off) // 496, 500 313 reg_save_frame_size // As noted: neglects any parameters to runtime // 504 314 }; 315 316 // Save off registers which might be killed by calls into the runtime. 317 // Tries to smart of about FP registers. In particular we separate 318 // saving and describing the FPU registers for deoptimization since we 319 // have to save the FPU registers twice if we describe them and on P4 320 // saving FPU registers which don't contain anything appears 321 // expensive. The deopt blob is the only thing which needs to 322 // describe FPU registers. In all other cases it should be sufficient 323 // to simply save their current value. 324 // 325 static OopMap* generate_oop_map(StubAssembler* sasm, int num_rt_args, 326 bool save_fpu_registers = true) { 327 328 // In 64bit all the args are in regs so there are no additional stack slots 329 LP64_ONLY(num_rt_args = 0); 330 LP64_ONLY(assert((reg_save_frame_size * VMRegImpl::stack_slot_size) % 16 == 0, "must be 16 byte aligned");) 331 int frame_size_in_slots = reg_save_frame_size + num_rt_args; // args + thread 332 sasm->set_frame_size(frame_size_in_slots / VMRegImpl::slots_per_word); 333 334 // record saved value locations in an OopMap 335 // locations are offsets from sp after runtime call; num_rt_args is number of arguments in call, including thread 336 OopMap* map = new OopMap(frame_size_in_slots, 0); 337 map->set_callee_saved(VMRegImpl::stack2reg(rax_off + num_rt_args), rax->as_VMReg()); 338 map->set_callee_saved(VMRegImpl::stack2reg(rcx_off + num_rt_args), rcx->as_VMReg()); 339 map->set_callee_saved(VMRegImpl::stack2reg(rdx_off + num_rt_args), rdx->as_VMReg()); 340 map->set_callee_saved(VMRegImpl::stack2reg(rbx_off + num_rt_args), rbx->as_VMReg()); 341 map->set_callee_saved(VMRegImpl::stack2reg(rsi_off + num_rt_args), rsi->as_VMReg()); 342 map->set_callee_saved(VMRegImpl::stack2reg(rdi_off + num_rt_args), rdi->as_VMReg()); 343 #ifdef _LP64 344 map->set_callee_saved(VMRegImpl::stack2reg(r8_off + num_rt_args), r8->as_VMReg()); 345 map->set_callee_saved(VMRegImpl::stack2reg(r9_off + num_rt_args), r9->as_VMReg()); 346 map->set_callee_saved(VMRegImpl::stack2reg(r10_off + num_rt_args), r10->as_VMReg()); 347 map->set_callee_saved(VMRegImpl::stack2reg(r11_off + num_rt_args), r11->as_VMReg()); 348 map->set_callee_saved(VMRegImpl::stack2reg(r12_off + num_rt_args), r12->as_VMReg()); 349 map->set_callee_saved(VMRegImpl::stack2reg(r13_off + num_rt_args), r13->as_VMReg()); 350 map->set_callee_saved(VMRegImpl::stack2reg(r14_off + num_rt_args), r14->as_VMReg()); 351 map->set_callee_saved(VMRegImpl::stack2reg(r15_off + num_rt_args), r15->as_VMReg()); 352 353 // This is stupid but needed. 354 map->set_callee_saved(VMRegImpl::stack2reg(raxH_off + num_rt_args), rax->as_VMReg()->next()); 355 map->set_callee_saved(VMRegImpl::stack2reg(rcxH_off + num_rt_args), rcx->as_VMReg()->next()); 356 map->set_callee_saved(VMRegImpl::stack2reg(rdxH_off + num_rt_args), rdx->as_VMReg()->next()); 357 map->set_callee_saved(VMRegImpl::stack2reg(rbxH_off + num_rt_args), rbx->as_VMReg()->next()); 358 map->set_callee_saved(VMRegImpl::stack2reg(rsiH_off + num_rt_args), rsi->as_VMReg()->next()); 359 map->set_callee_saved(VMRegImpl::stack2reg(rdiH_off + num_rt_args), rdi->as_VMReg()->next()); 360 361 map->set_callee_saved(VMRegImpl::stack2reg(r8H_off + num_rt_args), r8->as_VMReg()->next()); 362 map->set_callee_saved(VMRegImpl::stack2reg(r9H_off + num_rt_args), r9->as_VMReg()->next()); 363 map->set_callee_saved(VMRegImpl::stack2reg(r10H_off + num_rt_args), r10->as_VMReg()->next()); 364 map->set_callee_saved(VMRegImpl::stack2reg(r11H_off + num_rt_args), r11->as_VMReg()->next()); 365 map->set_callee_saved(VMRegImpl::stack2reg(r12H_off + num_rt_args), r12->as_VMReg()->next()); 366 map->set_callee_saved(VMRegImpl::stack2reg(r13H_off + num_rt_args), r13->as_VMReg()->next()); 367 map->set_callee_saved(VMRegImpl::stack2reg(r14H_off + num_rt_args), r14->as_VMReg()->next()); 368 map->set_callee_saved(VMRegImpl::stack2reg(r15H_off + num_rt_args), r15->as_VMReg()->next()); 369 #endif // _LP64 370 371 int xmm_bypass_limit = FrameMap::get_num_caller_save_xmms(); 372 373 if (save_fpu_registers) { 374 #ifndef _LP64 375 if (UseSSE < 2) { 376 int fpu_off = float_regs_as_doubles_off; 377 for (int n = 0; n < FrameMap::nof_fpu_regs; n++) { 378 VMReg fpu_name_0 = FrameMap::fpu_regname(n); 379 map->set_callee_saved(VMRegImpl::stack2reg(fpu_off + num_rt_args), fpu_name_0); 380 // %%% This is really a waste but we'll keep things as they were for now 381 if (true) { 382 map->set_callee_saved(VMRegImpl::stack2reg(fpu_off + 1 + num_rt_args), fpu_name_0->next()); 383 } 384 fpu_off += 2; 385 } 386 assert(fpu_off == fpu_state_off, "incorrect number of fpu stack slots"); 387 388 if (UseSSE == 1) { 389 int xmm_off = xmm_regs_as_doubles_off; 390 for (int n = 0; n < FrameMap::nof_fpu_regs; n++) { 391 VMReg xmm_name_0 = as_XMMRegister(n)->as_VMReg(); 392 map->set_callee_saved(VMRegImpl::stack2reg(xmm_off + num_rt_args), xmm_name_0); 393 xmm_off += 2; 394 } 395 assert(xmm_off == float_regs_as_doubles_off, "incorrect number of xmm registers"); 396 } 397 } 398 #endif // !LP64 399 400 if (UseSSE >= 2) { 401 int xmm_off = xmm_regs_as_doubles_off; 402 for (int n = 0; n < FrameMap::nof_xmm_regs; n++) { 403 if (n < xmm_bypass_limit) { 404 VMReg xmm_name_0 = as_XMMRegister(n)->as_VMReg(); 405 map->set_callee_saved(VMRegImpl::stack2reg(xmm_off + num_rt_args), xmm_name_0); 406 // %%% This is really a waste but we'll keep things as they were for now 407 if (true) { 408 map->set_callee_saved(VMRegImpl::stack2reg(xmm_off + 1 + num_rt_args), xmm_name_0->next()); 409 } 410 } 411 xmm_off += 2; 412 } 413 assert(xmm_off == float_regs_as_doubles_off, "incorrect number of xmm registers"); 414 } 415 } 416 417 return map; 418 } 419 420 #define __ this-> 421 422 void C1_MacroAssembler::save_live_registers_no_oop_map(bool save_fpu_registers) { 423 __ block_comment("save_live_registers"); 424 425 // Push CPU state in multiple of 16 bytes 426 #ifdef _LP64 427 __ save_legacy_gprs(); 428 #else 429 __ pusha(); 430 #endif 431 432 // assert(float_regs_as_doubles_off % 2 == 0, "misaligned offset"); 433 // assert(xmm_regs_as_doubles_off % 2 == 0, "misaligned offset"); 434 435 __ subptr(rsp, extra_space_offset * VMRegImpl::stack_slot_size); 436 437 #ifdef ASSERT 438 __ movptr(Address(rsp, marker * VMRegImpl::stack_slot_size), (int32_t)0xfeedbeef); 439 #endif 440 441 if (save_fpu_registers) { 442 #ifndef _LP64 443 if (UseSSE < 2) { 444 // save FPU stack 445 __ fnsave(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size)); 446 __ fwait(); 447 448 #ifdef ASSERT 449 Label ok; 450 __ cmpw(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size), StubRoutines::x86::fpu_cntrl_wrd_std()); 451 __ jccb(Assembler::equal, ok); 452 __ stop("corrupted control word detected"); 453 __ bind(ok); 454 #endif 455 456 // Reset the control word to guard against exceptions being unmasked 457 // since fstp_d can cause FPU stack underflow exceptions. Write it 458 // into the on stack copy and then reload that to make sure that the 459 // current and future values are correct. 460 __ movw(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size), StubRoutines::x86::fpu_cntrl_wrd_std()); 461 __ frstor(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size)); 462 463 // Save the FPU registers in de-opt-able form 464 int offset = 0; 465 for (int n = 0; n < FrameMap::nof_fpu_regs; n++) { 466 __ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + offset)); 467 offset += 8; 468 } 469 470 if (UseSSE == 1) { 471 // save XMM registers as float because double not supported without SSE2(num MMX == num fpu) 472 int offset = 0; 473 for (int n = 0; n < FrameMap::nof_fpu_regs; n++) { 474 XMMRegister xmm_name = as_XMMRegister(n); 475 __ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + offset), xmm_name); 476 offset += 8; 477 } 478 } 479 } 480 #endif // !_LP64 481 482 if (UseSSE >= 2) { 483 // save XMM registers 484 // XMM registers can contain float or double values, but this is not known here, 485 // so always save them as doubles. 486 // note that float values are _not_ converted automatically, so for float values 487 // the second word contains only garbage data. 488 int xmm_bypass_limit = FrameMap::get_num_caller_save_xmms(); 489 int offset = 0; 490 for (int n = 0; n < xmm_bypass_limit; n++) { 491 XMMRegister xmm_name = as_XMMRegister(n); 492 __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + offset), xmm_name); 493 offset += 8; 494 } 495 } 496 } 497 498 // FPU stack must be empty now 499 NOT_LP64( __ verify_FPU(0, "save_live_registers"); ) 500 } 501 502 #undef __ 503 #define __ sasm-> 504 505 static void restore_fpu(C1_MacroAssembler* sasm, bool restore_fpu_registers) { 506 #ifdef _LP64 507 if (restore_fpu_registers) { 508 // restore XMM registers 509 int xmm_bypass_limit = FrameMap::get_num_caller_save_xmms(); 510 int offset = 0; 511 for (int n = 0; n < xmm_bypass_limit; n++) { 512 XMMRegister xmm_name = as_XMMRegister(n); 513 __ movdbl(xmm_name, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + offset)); 514 offset += 8; 515 } 516 } 517 #else 518 if (restore_fpu_registers) { 519 if (UseSSE >= 2) { 520 // restore XMM registers 521 int xmm_bypass_limit = FrameMap::nof_xmm_regs; 522 int offset = 0; 523 for (int n = 0; n < xmm_bypass_limit; n++) { 524 XMMRegister xmm_name = as_XMMRegister(n); 525 __ movdbl(xmm_name, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + offset)); 526 offset += 8; 527 } 528 } else if (UseSSE == 1) { 529 // restore XMM registers(num MMX == num fpu) 530 int offset = 0; 531 for (int n = 0; n < FrameMap::nof_fpu_regs; n++) { 532 XMMRegister xmm_name = as_XMMRegister(n); 533 __ movflt(xmm_name, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + offset)); 534 offset += 8; 535 } 536 } 537 538 if (UseSSE < 2) { 539 __ frstor(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size)); 540 } else { 541 // check that FPU stack is really empty 542 __ verify_FPU(0, "restore_live_registers"); 543 } 544 } else { 545 // check that FPU stack is really empty 546 __ verify_FPU(0, "restore_live_registers"); 547 } 548 #endif // _LP64 549 550 #ifdef ASSERT 551 { 552 Label ok; 553 __ cmpptr(Address(rsp, marker * VMRegImpl::stack_slot_size), (int32_t)0xfeedbeef); 554 __ jcc(Assembler::equal, ok); 555 __ stop("bad offsets in frame"); 556 __ bind(ok); 557 } 558 #endif // ASSERT 559 560 __ addptr(rsp, extra_space_offset * VMRegImpl::stack_slot_size); 561 } 562 563 #undef __ 564 #define __ this-> 565 566 void C1_MacroAssembler::restore_live_registers(bool restore_fpu_registers) { 567 __ block_comment("restore_live_registers"); 568 569 restore_fpu(this, restore_fpu_registers); 570 #ifdef _LP64 571 __ restore_legacy_gprs(); 572 #else 573 __ popa(); 574 #endif 575 576 } 577 578 579 void C1_MacroAssembler::restore_live_registers_except_rax(bool restore_fpu_registers) { 580 __ block_comment("restore_live_registers_except_rax"); 581 582 restore_fpu(this, restore_fpu_registers); 583 584 #ifdef _LP64 585 __ movptr(r15, Address(rsp, 0)); 586 __ movptr(r14, Address(rsp, wordSize)); 587 __ movptr(r13, Address(rsp, 2 * wordSize)); 588 __ movptr(r12, Address(rsp, 3 * wordSize)); 589 __ movptr(r11, Address(rsp, 4 * wordSize)); 590 __ movptr(r10, Address(rsp, 5 * wordSize)); 591 __ movptr(r9, Address(rsp, 6 * wordSize)); 592 __ movptr(r8, Address(rsp, 7 * wordSize)); 593 __ movptr(rdi, Address(rsp, 8 * wordSize)); 594 __ movptr(rsi, Address(rsp, 9 * wordSize)); 595 __ movptr(rbp, Address(rsp, 10 * wordSize)); 596 // skip rsp 597 __ movptr(rbx, Address(rsp, 12 * wordSize)); 598 __ movptr(rdx, Address(rsp, 13 * wordSize)); 599 __ movptr(rcx, Address(rsp, 14 * wordSize)); 600 601 __ addptr(rsp, 16 * wordSize); 602 #else 603 604 __ pop(rdi); 605 __ pop(rsi); 606 __ pop(rbp); 607 __ pop(rbx); // skip this value 608 __ pop(rbx); 609 __ pop(rdx); 610 __ pop(rcx); 611 __ addptr(rsp, BytesPerWord); 612 #endif // _LP64 613 } 614 615 #undef __ 616 #define __ sasm-> 617 618 static OopMap* save_live_registers(StubAssembler* sasm, int num_rt_args, 619 bool save_fpu_registers = true) { 620 __ save_live_registers_no_oop_map(save_fpu_registers); 621 return generate_oop_map(sasm, num_rt_args, save_fpu_registers); 622 } 623 624 static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) { 625 __ restore_live_registers(restore_fpu_registers); 626 } 627 628 static void restore_live_registers_except_rax(StubAssembler* sasm, bool restore_fpu_registers = true) { 629 sasm->restore_live_registers_except_rax(restore_fpu_registers); 630 } 631 632 633 void Runtime1::initialize_pd() { 634 // nothing to do 635 } 636 637 uint Runtime1::runtime_blob_current_thread_offset(frame f) { 638 return r15_off / 2; 639 } 640 641 // Target: the entry point of the method that creates and posts the exception oop. 642 // has_argument: true if the exception needs arguments (passed on the stack because 643 // registers must be preserved). 644 OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) { 645 // Preserve all registers. 646 int num_rt_args = has_argument ? (2 + 1) : 1; 647 OopMap* oop_map = save_live_registers(sasm, num_rt_args); 648 649 // Now all registers are saved and can be used freely. 650 // Verify that no old value is used accidentally. 651 __ invalidate_registers(true, true, true, true, true, true); 652 653 // Registers used by this stub. 654 const Register temp_reg = rbx; 655 656 // Load arguments for exception that are passed as arguments into the stub. 657 if (has_argument) { 658 #ifdef _LP64 659 __ movptr(c_rarg1, Address(rbp, 2*BytesPerWord)); 660 __ movptr(c_rarg2, Address(rbp, 3*BytesPerWord)); 661 #else 662 __ movptr(temp_reg, Address(rbp, 3*BytesPerWord)); 663 __ push(temp_reg); 664 __ movptr(temp_reg, Address(rbp, 2*BytesPerWord)); 665 __ push(temp_reg); 666 #endif // _LP64 667 } 668 int call_offset = __ call_RT(noreg, noreg, target, num_rt_args - 1); 669 670 OopMapSet* oop_maps = new OopMapSet(); 671 oop_maps->add_gc_map(call_offset, oop_map); 672 673 __ stop("should not reach here"); 674 675 return oop_maps; 676 } 677 678 679 OopMapSet* Runtime1::generate_handle_exception(StubID id, StubAssembler *sasm) { 680 __ block_comment("generate_handle_exception"); 681 682 // incoming parameters 683 const Register exception_oop = rax; 684 const Register exception_pc = rdx; 685 // other registers used in this stub 686 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread); 687 688 // Save registers, if required. 689 OopMapSet* oop_maps = new OopMapSet(); 690 OopMap* oop_map = nullptr; 691 switch (id) { 692 case forward_exception_id: 693 // We're handling an exception in the context of a compiled frame. 694 // The registers have been saved in the standard places. Perform 695 // an exception lookup in the caller and dispatch to the handler 696 // if found. Otherwise unwind and dispatch to the callers 697 // exception handler. 698 oop_map = generate_oop_map(sasm, 1 /*thread*/); 699 700 // load and clear pending exception oop into RAX 701 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset())); 702 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD); 703 704 // load issuing PC (the return address for this stub) into rdx 705 __ movptr(exception_pc, Address(rbp, 1*BytesPerWord)); 706 707 // make sure that the vm_results are cleared (may be unnecessary) 708 __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD); 709 __ movptr(Address(thread, JavaThread::vm_result_2_offset()), NULL_WORD); 710 break; 711 case handle_exception_nofpu_id: 712 case handle_exception_id: 713 // At this point all registers MAY be live. 714 oop_map = save_live_registers(sasm, 1 /*thread*/, id != handle_exception_nofpu_id); 715 break; 716 case handle_exception_from_callee_id: { 717 // At this point all registers except exception oop (RAX) and 718 // exception pc (RDX) are dead. 719 const int frame_size = 2 /*BP, return address*/ NOT_LP64(+ 1 /*thread*/) WIN64_ONLY(+ frame::arg_reg_save_area_bytes / BytesPerWord); 720 oop_map = new OopMap(frame_size * VMRegImpl::slots_per_word, 0); 721 sasm->set_frame_size(frame_size); 722 WIN64_ONLY(__ subq(rsp, frame::arg_reg_save_area_bytes)); 723 break; 724 } 725 default: ShouldNotReachHere(); 726 } 727 728 #if !defined(_LP64) && defined(COMPILER2) 729 if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) { 730 // C2 can leave the fpu stack dirty 731 __ empty_FPU_stack(); 732 } 733 #endif // !_LP64 && COMPILER2 734 735 // verify that only rax, and rdx is valid at this time 736 __ invalidate_registers(false, true, true, false, true, true); 737 // verify that rax, contains a valid exception 738 __ verify_not_null_oop(exception_oop); 739 740 // load address of JavaThread object for thread-local data 741 NOT_LP64(__ get_thread(thread);) 742 743 #ifdef ASSERT 744 // check that fields in JavaThread for exception oop and issuing pc are 745 // empty before writing to them 746 Label oop_empty; 747 __ cmpptr(Address(thread, JavaThread::exception_oop_offset()), NULL_WORD); 748 __ jcc(Assembler::equal, oop_empty); 749 __ stop("exception oop already set"); 750 __ bind(oop_empty); 751 752 Label pc_empty; 753 __ cmpptr(Address(thread, JavaThread::exception_pc_offset()), 0); 754 __ jcc(Assembler::equal, pc_empty); 755 __ stop("exception pc already set"); 756 __ bind(pc_empty); 757 #endif 758 759 // save exception oop and issuing pc into JavaThread 760 // (exception handler will load it from here) 761 __ movptr(Address(thread, JavaThread::exception_oop_offset()), exception_oop); 762 __ movptr(Address(thread, JavaThread::exception_pc_offset()), exception_pc); 763 764 // patch throwing pc into return address (has bci & oop map) 765 __ movptr(Address(rbp, 1*BytesPerWord), exception_pc); 766 767 // compute the exception handler. 768 // the exception oop and the throwing pc are read from the fields in JavaThread 769 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc)); 770 oop_maps->add_gc_map(call_offset, oop_map); 771 772 // rax: handler address 773 // will be the deopt blob if nmethod was deoptimized while we looked up 774 // handler regardless of whether handler existed in the nmethod. 775 776 // only rax, is valid at this time, all other registers have been destroyed by the runtime call 777 __ invalidate_registers(false, true, true, true, true, true); 778 779 // patch the return address, this stub will directly return to the exception handler 780 __ movptr(Address(rbp, 1*BytesPerWord), rax); 781 782 switch (id) { 783 case forward_exception_id: 784 case handle_exception_nofpu_id: 785 case handle_exception_id: 786 // Restore the registers that were saved at the beginning. 787 restore_live_registers(sasm, id != handle_exception_nofpu_id); 788 break; 789 case handle_exception_from_callee_id: 790 // WIN64_ONLY: No need to add frame::arg_reg_save_area_bytes to SP 791 // since we do a leave anyway. 792 793 // Pop the return address. 794 __ leave(); 795 __ pop(rcx); 796 __ jmp(rcx); // jump to exception handler 797 break; 798 default: ShouldNotReachHere(); 799 } 800 801 return oop_maps; 802 } 803 804 805 void Runtime1::generate_unwind_exception(StubAssembler *sasm) { 806 // incoming parameters 807 const Register exception_oop = rax; 808 // callee-saved copy of exception_oop during runtime call 809 const Register exception_oop_callee_saved = NOT_LP64(rsi) LP64_ONLY(r14); 810 // other registers used in this stub 811 const Register exception_pc = rdx; 812 const Register handler_addr = rbx; 813 const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread); 814 815 if (AbortVMOnException) { 816 __ enter(); 817 save_live_registers(sasm, 2); 818 __ call_VM_leaf(CAST_FROM_FN_PTR(address, check_abort_on_vm_exception), rax); 819 restore_live_registers(sasm); 820 __ leave(); 821 } 822 823 // verify that only rax, is valid at this time 824 __ invalidate_registers(false, true, true, true, true, true); 825 826 #ifdef ASSERT 827 // check that fields in JavaThread for exception oop and issuing pc are empty 828 NOT_LP64(__ get_thread(thread);) 829 Label oop_empty; 830 __ cmpptr(Address(thread, JavaThread::exception_oop_offset()), 0); 831 __ jcc(Assembler::equal, oop_empty); 832 __ stop("exception oop must be empty"); 833 __ bind(oop_empty); 834 835 Label pc_empty; 836 __ cmpptr(Address(thread, JavaThread::exception_pc_offset()), 0); 837 __ jcc(Assembler::equal, pc_empty); 838 __ stop("exception pc must be empty"); 839 __ bind(pc_empty); 840 #endif 841 842 // clear the FPU stack in case any FPU results are left behind 843 NOT_LP64( __ empty_FPU_stack(); ) 844 845 // save exception_oop in callee-saved register to preserve it during runtime calls 846 __ verify_not_null_oop(exception_oop); 847 __ movptr(exception_oop_callee_saved, exception_oop); 848 849 NOT_LP64(__ get_thread(thread);) 850 // Get return address (is on top of stack after leave). 851 __ movptr(exception_pc, Address(rsp, 0)); 852 853 // search the exception handler address of the caller (using the return address) 854 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc); 855 // rax: exception handler address of the caller 856 857 // Only RAX and RSI are valid at this time, all other registers have been destroyed by the call. 858 __ invalidate_registers(false, true, true, true, false, true); 859 860 // move result of call into correct register 861 __ movptr(handler_addr, rax); 862 863 // Restore exception oop to RAX (required convention of exception handler). 864 __ movptr(exception_oop, exception_oop_callee_saved); 865 866 // verify that there is really a valid exception in rax 867 __ verify_not_null_oop(exception_oop); 868 869 // get throwing pc (= return address). 870 // rdx has been destroyed by the call, so it must be set again 871 // the pop is also necessary to simulate the effect of a ret(0) 872 __ pop(exception_pc); 873 874 // continue at exception handler (return address removed) 875 // note: do *not* remove arguments when unwinding the 876 // activation since the caller assumes having 877 // all arguments on the stack when entering the 878 // runtime to determine the exception handler 879 // (GC happens at call site with arguments!) 880 // rax: exception oop 881 // rdx: throwing pc 882 // rbx: exception handler 883 __ jmp(handler_addr); 884 } 885 886 887 OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) { 888 // use the maximum number of runtime-arguments here because it is difficult to 889 // distinguish each RT-Call. 890 // Note: This number affects also the RT-Call in generate_handle_exception because 891 // the oop-map is shared for all calls. 892 const int num_rt_args = 2; // thread + dummy 893 894 DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob(); 895 assert(deopt_blob != nullptr, "deoptimization blob must have been created"); 896 897 OopMap* oop_map = save_live_registers(sasm, num_rt_args); 898 899 #ifdef _LP64 900 const Register thread = r15_thread; 901 // No need to worry about dummy 902 __ mov(c_rarg0, thread); 903 #else 904 __ push(rax); // push dummy 905 906 const Register thread = rdi; // is callee-saved register (Visual C++ calling conventions) 907 // push java thread (becomes first argument of C function) 908 __ get_thread(thread); 909 __ push(thread); 910 #endif // _LP64 911 __ set_last_Java_frame(thread, noreg, rbp, nullptr, rscratch1); 912 // do the call 913 __ call(RuntimeAddress(target)); 914 OopMapSet* oop_maps = new OopMapSet(); 915 oop_maps->add_gc_map(__ offset(), oop_map); 916 // verify callee-saved register 917 #ifdef ASSERT 918 guarantee(thread != rax, "change this code"); 919 __ push(rax); 920 { Label L; 921 __ get_thread(rax); 922 __ cmpptr(thread, rax); 923 __ jcc(Assembler::equal, L); 924 __ stop("StubAssembler::call_RT: rdi/r15 not callee saved?"); 925 __ bind(L); 926 } 927 __ pop(rax); 928 #endif 929 __ reset_last_Java_frame(thread, true); 930 #ifndef _LP64 931 __ pop(rcx); // discard thread arg 932 __ pop(rcx); // discard dummy 933 #endif // _LP64 934 935 // check for pending exceptions 936 { Label L; 937 __ cmpptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD); 938 __ jcc(Assembler::equal, L); 939 // exception pending => remove activation and forward to exception handler 940 941 __ testptr(rax, rax); // have we deoptimized? 942 __ jump_cc(Assembler::equal, 943 RuntimeAddress(Runtime1::entry_for(Runtime1::forward_exception_id))); 944 945 // the deopt blob expects exceptions in the special fields of 946 // JavaThread, so copy and clear pending exception. 947 948 // load and clear pending exception 949 __ movptr(rax, Address(thread, Thread::pending_exception_offset())); 950 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD); 951 952 // check that there is really a valid exception 953 __ verify_not_null_oop(rax); 954 955 // load throwing pc: this is the return address of the stub 956 __ movptr(rdx, Address(rsp, return_off * VMRegImpl::stack_slot_size)); 957 958 #ifdef ASSERT 959 // check that fields in JavaThread for exception oop and issuing pc are empty 960 Label oop_empty; 961 __ cmpptr(Address(thread, JavaThread::exception_oop_offset()), NULL_WORD); 962 __ jcc(Assembler::equal, oop_empty); 963 __ stop("exception oop must be empty"); 964 __ bind(oop_empty); 965 966 Label pc_empty; 967 __ cmpptr(Address(thread, JavaThread::exception_pc_offset()), NULL_WORD); 968 __ jcc(Assembler::equal, pc_empty); 969 __ stop("exception pc must be empty"); 970 __ bind(pc_empty); 971 #endif 972 973 // store exception oop and throwing pc to JavaThread 974 __ movptr(Address(thread, JavaThread::exception_oop_offset()), rax); 975 __ movptr(Address(thread, JavaThread::exception_pc_offset()), rdx); 976 977 restore_live_registers(sasm); 978 979 __ leave(); 980 __ addptr(rsp, BytesPerWord); // remove return address from stack 981 982 // Forward the exception directly to deopt blob. We can blow no 983 // registers and must leave throwing pc on the stack. A patch may 984 // have values live in registers so the entry point with the 985 // exception in tls. 986 __ jump(RuntimeAddress(deopt_blob->unpack_with_exception_in_tls())); 987 988 __ bind(L); 989 } 990 991 992 // Runtime will return true if the nmethod has been deoptimized during 993 // the patching process. In that case we must do a deopt reexecute instead. 994 995 Label cont; 996 997 __ testptr(rax, rax); // have we deoptimized? 998 __ jcc(Assembler::equal, cont); // no 999 1000 // Will reexecute. Proper return address is already on the stack we just restore 1001 // registers, pop all of our frame but the return address and jump to the deopt blob 1002 restore_live_registers(sasm); 1003 __ leave(); 1004 __ jump(RuntimeAddress(deopt_blob->unpack_with_reexecution())); 1005 1006 __ bind(cont); 1007 restore_live_registers(sasm); 1008 __ leave(); 1009 __ ret(0); 1010 1011 return oop_maps; 1012 } 1013 1014 1015 OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) { 1016 1017 // for better readability 1018 const bool must_gc_arguments = true; 1019 const bool dont_gc_arguments = false; 1020 1021 // default value; overwritten for some optimized stubs that are called from methods that do not use the fpu 1022 bool save_fpu_registers = true; 1023 1024 // stub code & info for the different stubs 1025 OopMapSet* oop_maps = nullptr; 1026 switch (id) { 1027 case forward_exception_id: 1028 { 1029 oop_maps = generate_handle_exception(id, sasm); 1030 __ leave(); 1031 __ ret(0); 1032 } 1033 break; 1034 1035 case new_instance_id: 1036 case fast_new_instance_id: 1037 case fast_new_instance_init_check_id: 1038 { 1039 Register klass = rdx; // Incoming 1040 Register obj = rax; // Result 1041 1042 if (id == new_instance_id) { 1043 __ set_info("new_instance", dont_gc_arguments); 1044 } else if (id == fast_new_instance_id) { 1045 __ set_info("fast new_instance", dont_gc_arguments); 1046 } else { 1047 assert(id == fast_new_instance_init_check_id, "bad StubID"); 1048 __ set_info("fast new_instance init check", dont_gc_arguments); 1049 } 1050 1051 __ enter(); 1052 OopMap* map = save_live_registers(sasm, 2); 1053 int call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_instance), klass); 1054 oop_maps = new OopMapSet(); 1055 oop_maps->add_gc_map(call_offset, map); 1056 restore_live_registers_except_rax(sasm); 1057 __ verify_oop(obj); 1058 __ leave(); 1059 __ ret(0); 1060 1061 // rax,: new instance 1062 } 1063 1064 break; 1065 1066 case counter_overflow_id: 1067 { 1068 Register bci = rax, method = rbx; 1069 __ enter(); 1070 OopMap* map = save_live_registers(sasm, 3); 1071 // Retrieve bci 1072 __ movl(bci, Address(rbp, 2*BytesPerWord)); 1073 // And a pointer to the Method* 1074 __ movptr(method, Address(rbp, 3*BytesPerWord)); 1075 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, counter_overflow), bci, method); 1076 oop_maps = new OopMapSet(); 1077 oop_maps->add_gc_map(call_offset, map); 1078 restore_live_registers(sasm); 1079 __ leave(); 1080 __ ret(0); 1081 } 1082 break; 1083 1084 case new_type_array_id: 1085 case new_object_array_id: 1086 { 1087 Register length = rbx; // Incoming 1088 Register klass = rdx; // Incoming 1089 Register obj = rax; // Result 1090 1091 if (id == new_type_array_id) { 1092 __ set_info("new_type_array", dont_gc_arguments); 1093 } else { 1094 __ set_info("new_object_array", dont_gc_arguments); 1095 } 1096 1097 #ifdef ASSERT 1098 // assert object type is really an array of the proper kind 1099 { 1100 Label ok; 1101 Register t0 = obj; 1102 __ movl(t0, Address(klass, Klass::layout_helper_offset())); 1103 __ sarl(t0, Klass::_lh_array_tag_shift); 1104 int tag = ((id == new_type_array_id) 1105 ? Klass::_lh_array_tag_type_value 1106 : Klass::_lh_array_tag_obj_value); 1107 __ cmpl(t0, tag); 1108 __ jcc(Assembler::equal, ok); 1109 __ stop("assert(is an array klass)"); 1110 __ should_not_reach_here(); 1111 __ bind(ok); 1112 } 1113 #endif // ASSERT 1114 1115 __ enter(); 1116 OopMap* map = save_live_registers(sasm, 3); 1117 int call_offset; 1118 if (id == new_type_array_id) { 1119 call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_type_array), klass, length); 1120 } else { 1121 call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_object_array), klass, length); 1122 } 1123 1124 oop_maps = new OopMapSet(); 1125 oop_maps->add_gc_map(call_offset, map); 1126 restore_live_registers_except_rax(sasm); 1127 1128 __ verify_oop(obj); 1129 __ leave(); 1130 __ ret(0); 1131 1132 // rax,: new array 1133 } 1134 break; 1135 1136 case new_multi_array_id: 1137 { StubFrame f(sasm, "new_multi_array", dont_gc_arguments); 1138 // rax,: klass 1139 // rbx,: rank 1140 // rcx: address of 1st dimension 1141 OopMap* map = save_live_registers(sasm, 4); 1142 int call_offset = __ call_RT(rax, noreg, CAST_FROM_FN_PTR(address, new_multi_array), rax, rbx, rcx); 1143 1144 oop_maps = new OopMapSet(); 1145 oop_maps->add_gc_map(call_offset, map); 1146 restore_live_registers_except_rax(sasm); 1147 1148 // rax,: new multi array 1149 __ verify_oop(rax); 1150 } 1151 break; 1152 1153 case register_finalizer_id: 1154 { 1155 __ set_info("register_finalizer", dont_gc_arguments); 1156 1157 // This is called via call_runtime so the arguments 1158 // will be place in C abi locations 1159 1160 #ifdef _LP64 1161 __ verify_oop(c_rarg0); 1162 __ mov(rax, c_rarg0); 1163 #else 1164 // The object is passed on the stack and we haven't pushed a 1165 // frame yet so it's one work away from top of stack. 1166 __ movptr(rax, Address(rsp, 1 * BytesPerWord)); 1167 __ verify_oop(rax); 1168 #endif // _LP64 1169 1170 // load the klass and check the has finalizer flag 1171 Label register_finalizer; 1172 Register t = rsi; 1173 __ load_klass(t, rax, rscratch1); 1174 __ movl(t, Address(t, Klass::access_flags_offset())); 1175 __ testl(t, JVM_ACC_HAS_FINALIZER); 1176 __ jcc(Assembler::notZero, register_finalizer); 1177 __ ret(0); 1178 1179 __ bind(register_finalizer); 1180 __ enter(); 1181 OopMap* oop_map = save_live_registers(sasm, 2 /*num_rt_args */); 1182 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), rax); 1183 oop_maps = new OopMapSet(); 1184 oop_maps->add_gc_map(call_offset, oop_map); 1185 1186 // Now restore all the live registers 1187 restore_live_registers(sasm); 1188 1189 __ leave(); 1190 __ ret(0); 1191 } 1192 break; 1193 1194 case throw_range_check_failed_id: 1195 { StubFrame f(sasm, "range_check_failed", dont_gc_arguments); 1196 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true); 1197 } 1198 break; 1199 1200 case throw_index_exception_id: 1201 { StubFrame f(sasm, "index_range_check_failed", dont_gc_arguments); 1202 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true); 1203 } 1204 break; 1205 1206 case throw_div0_exception_id: 1207 { StubFrame f(sasm, "throw_div0_exception", dont_gc_arguments); 1208 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false); 1209 } 1210 break; 1211 1212 case throw_null_pointer_exception_id: 1213 { StubFrame f(sasm, "throw_null_pointer_exception", dont_gc_arguments); 1214 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false); 1215 } 1216 break; 1217 1218 case handle_exception_nofpu_id: 1219 case handle_exception_id: 1220 { StubFrame f(sasm, "handle_exception", dont_gc_arguments); 1221 oop_maps = generate_handle_exception(id, sasm); 1222 } 1223 break; 1224 1225 case handle_exception_from_callee_id: 1226 { StubFrame f(sasm, "handle_exception_from_callee", dont_gc_arguments); 1227 oop_maps = generate_handle_exception(id, sasm); 1228 } 1229 break; 1230 1231 case unwind_exception_id: 1232 { __ set_info("unwind_exception", dont_gc_arguments); 1233 // note: no stubframe since we are about to leave the current 1234 // activation and we are calling a leaf VM function only. 1235 generate_unwind_exception(sasm); 1236 } 1237 break; 1238 1239 case throw_array_store_exception_id: 1240 { StubFrame f(sasm, "throw_array_store_exception", dont_gc_arguments); 1241 // tos + 0: link 1242 // + 1: return address 1243 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), true); 1244 } 1245 break; 1246 1247 case throw_class_cast_exception_id: 1248 { StubFrame f(sasm, "throw_class_cast_exception", dont_gc_arguments); 1249 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true); 1250 } 1251 break; 1252 1253 case throw_incompatible_class_change_error_id: 1254 { StubFrame f(sasm, "throw_incompatible_class_cast_exception", dont_gc_arguments); 1255 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false); 1256 } 1257 break; 1258 1259 case slow_subtype_check_id: 1260 { 1261 // Typical calling sequence: 1262 // __ push(klass_RInfo); // object klass or other subclass 1263 // __ push(sup_k_RInfo); // array element klass or other superclass 1264 // __ call(slow_subtype_check); 1265 // Note that the subclass is pushed first, and is therefore deepest. 1266 // Previous versions of this code reversed the names 'sub' and 'super'. 1267 // This was operationally harmless but made the code unreadable. 1268 enum layout { 1269 rax_off, SLOT2(raxH_off) 1270 rcx_off, SLOT2(rcxH_off) 1271 rsi_off, SLOT2(rsiH_off) 1272 rdi_off, SLOT2(rdiH_off) 1273 // saved_rbp_off, SLOT2(saved_rbpH_off) 1274 return_off, SLOT2(returnH_off) 1275 sup_k_off, SLOT2(sup_kH_off) 1276 klass_off, SLOT2(superH_off) 1277 framesize, 1278 result_off = klass_off // deepest argument is also the return value 1279 }; 1280 1281 __ set_info("slow_subtype_check", dont_gc_arguments); 1282 __ push(rdi); 1283 __ push(rsi); 1284 __ push(rcx); 1285 __ push(rax); 1286 1287 // This is called by pushing args and not with C abi 1288 __ movptr(rsi, Address(rsp, (klass_off) * VMRegImpl::stack_slot_size)); // subclass 1289 __ movptr(rax, Address(rsp, (sup_k_off) * VMRegImpl::stack_slot_size)); // superclass 1290 1291 Label miss; 1292 __ check_klass_subtype_slow_path(rsi, rax, rcx, rdi, nullptr, &miss); 1293 1294 // fallthrough on success: 1295 __ movptr(Address(rsp, (result_off) * VMRegImpl::stack_slot_size), 1); // result 1296 __ pop(rax); 1297 __ pop(rcx); 1298 __ pop(rsi); 1299 __ pop(rdi); 1300 __ ret(0); 1301 1302 __ bind(miss); 1303 __ movptr(Address(rsp, (result_off) * VMRegImpl::stack_slot_size), NULL_WORD); // result 1304 __ pop(rax); 1305 __ pop(rcx); 1306 __ pop(rsi); 1307 __ pop(rdi); 1308 __ ret(0); 1309 } 1310 break; 1311 1312 case monitorenter_nofpu_id: 1313 save_fpu_registers = false; 1314 // fall through 1315 case monitorenter_id: 1316 { 1317 StubFrame f(sasm, "monitorenter", dont_gc_arguments, true /* use_pop_on_epilog */); 1318 OopMap* map = save_live_registers(sasm, 3, save_fpu_registers); 1319 1320 // Called with store_parameter and not C abi 1321 1322 f.load_argument(1, rax); // rax,: object 1323 f.load_argument(0, rbx); // rbx,: lock address 1324 1325 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), rax, rbx); 1326 1327 oop_maps = new OopMapSet(); 1328 oop_maps->add_gc_map(call_offset, map); 1329 restore_live_registers(sasm, save_fpu_registers); 1330 } 1331 break; 1332 1333 case monitorexit_nofpu_id: 1334 save_fpu_registers = false; 1335 // fall through 1336 case monitorexit_id: 1337 { 1338 StubFrame f(sasm, "monitorexit", dont_gc_arguments); 1339 OopMap* map = save_live_registers(sasm, 2, save_fpu_registers); 1340 1341 // Called with store_parameter and not C abi 1342 1343 f.load_argument(0, rax); // rax,: lock address 1344 1345 // note: really a leaf routine but must setup last java sp 1346 // => use call_RT for now (speed can be improved by 1347 // doing last java sp setup manually) 1348 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), rax); 1349 1350 oop_maps = new OopMapSet(); 1351 oop_maps->add_gc_map(call_offset, map); 1352 restore_live_registers(sasm, save_fpu_registers); 1353 } 1354 break; 1355 1356 case deoptimize_id: 1357 { 1358 StubFrame f(sasm, "deoptimize", dont_gc_arguments); 1359 const int num_rt_args = 2; // thread, trap_request 1360 OopMap* oop_map = save_live_registers(sasm, num_rt_args); 1361 f.load_argument(0, rax); 1362 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, deoptimize), rax); 1363 oop_maps = new OopMapSet(); 1364 oop_maps->add_gc_map(call_offset, oop_map); 1365 restore_live_registers(sasm); 1366 DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob(); 1367 assert(deopt_blob != nullptr, "deoptimization blob must have been created"); 1368 __ leave(); 1369 __ jump(RuntimeAddress(deopt_blob->unpack_with_reexecution())); 1370 } 1371 break; 1372 1373 case access_field_patching_id: 1374 { StubFrame f(sasm, "access_field_patching", dont_gc_arguments); 1375 // we should set up register map 1376 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching)); 1377 } 1378 break; 1379 1380 case load_klass_patching_id: 1381 { StubFrame f(sasm, "load_klass_patching", dont_gc_arguments); 1382 // we should set up register map 1383 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching)); 1384 } 1385 break; 1386 1387 case load_mirror_patching_id: 1388 { StubFrame f(sasm, "load_mirror_patching", dont_gc_arguments); 1389 // we should set up register map 1390 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_mirror_patching)); 1391 } 1392 break; 1393 1394 case load_appendix_patching_id: 1395 { StubFrame f(sasm, "load_appendix_patching", dont_gc_arguments); 1396 // we should set up register map 1397 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_appendix_patching)); 1398 } 1399 break; 1400 1401 case dtrace_object_alloc_id: 1402 { // rax,: object 1403 StubFrame f(sasm, "dtrace_object_alloc", dont_gc_arguments); 1404 // we can't gc here so skip the oopmap but make sure that all 1405 // the live registers get saved. 1406 save_live_registers(sasm, 1); 1407 1408 __ NOT_LP64(push(rax)) LP64_ONLY(mov(c_rarg0, rax)); 1409 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)))); 1410 NOT_LP64(__ pop(rax)); 1411 1412 restore_live_registers(sasm); 1413 } 1414 break; 1415 1416 case fpu2long_stub_id: 1417 { 1418 #ifdef _LP64 1419 Label done; 1420 __ cvttsd2siq(rax, Address(rsp, wordSize)); 1421 __ cmp64(rax, ExternalAddress((address) StubRoutines::x86::double_sign_flip())); 1422 __ jccb(Assembler::notEqual, done); 1423 __ movq(rax, Address(rsp, wordSize)); 1424 __ subptr(rsp, 8); 1425 __ movq(Address(rsp, 0), rax); 1426 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup()))); 1427 __ pop(rax); 1428 __ bind(done); 1429 __ ret(0); 1430 #else 1431 // rax, and rdx are destroyed, but should be free since the result is returned there 1432 // preserve rsi,ecx 1433 __ push(rsi); 1434 __ push(rcx); 1435 1436 // check for NaN 1437 Label return0, do_return, return_min_jlong, do_convert; 1438 1439 Address value_high_word(rsp, wordSize + 4); 1440 Address value_low_word(rsp, wordSize); 1441 Address result_high_word(rsp, 3*wordSize + 4); 1442 Address result_low_word(rsp, 3*wordSize); 1443 1444 __ subptr(rsp, 32); // more than enough on 32bit 1445 __ fst_d(value_low_word); 1446 __ movl(rax, value_high_word); 1447 __ andl(rax, 0x7ff00000); 1448 __ cmpl(rax, 0x7ff00000); 1449 __ jcc(Assembler::notEqual, do_convert); 1450 __ movl(rax, value_high_word); 1451 __ andl(rax, 0xfffff); 1452 __ orl(rax, value_low_word); 1453 __ jcc(Assembler::notZero, return0); 1454 1455 __ bind(do_convert); 1456 __ fnstcw(Address(rsp, 0)); 1457 __ movzwl(rax, Address(rsp, 0)); 1458 __ orl(rax, 0xc00); 1459 __ movw(Address(rsp, 2), rax); 1460 __ fldcw(Address(rsp, 2)); 1461 __ fwait(); 1462 __ fistp_d(result_low_word); 1463 __ fldcw(Address(rsp, 0)); 1464 __ fwait(); 1465 // This gets the entire long in rax on 64bit 1466 __ movptr(rax, result_low_word); 1467 // testing of high bits 1468 __ movl(rdx, result_high_word); 1469 __ mov(rcx, rax); 1470 // What the heck is the point of the next instruction??? 1471 __ xorl(rcx, 0x0); 1472 __ movl(rsi, 0x80000000); 1473 __ xorl(rsi, rdx); 1474 __ orl(rcx, rsi); 1475 __ jcc(Assembler::notEqual, do_return); 1476 __ fldz(); 1477 __ fcomp_d(value_low_word); 1478 __ fnstsw_ax(); 1479 __ sahf(); 1480 __ jcc(Assembler::above, return_min_jlong); 1481 // return max_jlong 1482 __ movl(rdx, 0x7fffffff); 1483 __ movl(rax, 0xffffffff); 1484 __ jmp(do_return); 1485 1486 __ bind(return_min_jlong); 1487 __ movl(rdx, 0x80000000); 1488 __ xorl(rax, rax); 1489 __ jmp(do_return); 1490 1491 __ bind(return0); 1492 __ fpop(); 1493 __ xorptr(rdx,rdx); 1494 __ xorptr(rax,rax); 1495 1496 __ bind(do_return); 1497 __ addptr(rsp, 32); 1498 __ pop(rcx); 1499 __ pop(rsi); 1500 __ ret(0); 1501 #endif // _LP64 1502 } 1503 break; 1504 1505 case predicate_failed_trap_id: 1506 { 1507 StubFrame f(sasm, "predicate_failed_trap", dont_gc_arguments); 1508 1509 OopMap* map = save_live_registers(sasm, 1); 1510 1511 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap)); 1512 oop_maps = new OopMapSet(); 1513 oop_maps->add_gc_map(call_offset, map); 1514 restore_live_registers(sasm); 1515 __ leave(); 1516 DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob(); 1517 assert(deopt_blob != nullptr, "deoptimization blob must have been created"); 1518 1519 __ jump(RuntimeAddress(deopt_blob->unpack_with_reexecution())); 1520 } 1521 break; 1522 1523 default: 1524 { StubFrame f(sasm, "unimplemented entry", dont_gc_arguments); 1525 __ movptr(rax, (int)id); 1526 __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), rax); 1527 __ should_not_reach_here(); 1528 } 1529 break; 1530 } 1531 return oop_maps; 1532 } 1533 1534 #undef __ 1535 1536 const char *Runtime1::pd_name_for_address(address entry) { 1537 return "<unknown function>"; 1538 } --- EOF ---